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]; 245 246 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 247 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 248 249 extern atomic_long_t tcp_memory_allocated; 250 extern struct percpu_counter tcp_sockets_allocated; 251 extern unsigned long tcp_memory_pressure; 252 253 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 254 static inline bool tcp_under_memory_pressure(const struct sock *sk) 255 { 256 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 257 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 258 return true; 259 260 return tcp_memory_pressure; 261 } 262 /* 263 * The next routines deal with comparing 32 bit unsigned ints 264 * and worry about wraparound (automatic with unsigned arithmetic). 265 */ 266 267 static inline bool before(__u32 seq1, __u32 seq2) 268 { 269 return (__s32)(seq1-seq2) < 0; 270 } 271 #define after(seq2, seq1) before(seq1, seq2) 272 273 /* is s2<=s1<=s3 ? */ 274 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 275 { 276 return seq3 - seq2 >= seq1 - seq2; 277 } 278 279 static inline bool tcp_out_of_memory(struct sock *sk) 280 { 281 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 282 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 283 return true; 284 return false; 285 } 286 287 void sk_forced_mem_schedule(struct sock *sk, int size); 288 289 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 290 { 291 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 292 int orphans = percpu_counter_read_positive(ocp); 293 294 if (orphans << shift > sysctl_tcp_max_orphans) { 295 orphans = percpu_counter_sum_positive(ocp); 296 if (orphans << shift > sysctl_tcp_max_orphans) 297 return true; 298 } 299 return false; 300 } 301 302 bool tcp_check_oom(struct sock *sk, int shift); 303 304 305 extern struct proto tcp_prot; 306 307 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 308 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 309 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 310 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 311 312 void tcp_tasklet_init(void); 313 314 void tcp_v4_err(struct sk_buff *skb, u32); 315 316 void tcp_shutdown(struct sock *sk, int how); 317 318 int tcp_v4_early_demux(struct sk_buff *skb); 319 int tcp_v4_rcv(struct sk_buff *skb); 320 321 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 322 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 323 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 324 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 325 int flags); 326 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 327 size_t size, int flags); 328 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 329 size_t size, int flags); 330 void tcp_release_cb(struct sock *sk); 331 void tcp_wfree(struct sk_buff *skb); 332 void tcp_write_timer_handler(struct sock *sk); 333 void tcp_delack_timer_handler(struct sock *sk); 334 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 335 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 336 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 337 const struct tcphdr *th); 338 void tcp_rcv_space_adjust(struct sock *sk); 339 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 340 void tcp_twsk_destructor(struct sock *sk); 341 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 342 struct pipe_inode_info *pipe, size_t len, 343 unsigned int flags); 344 345 static inline void tcp_dec_quickack_mode(struct sock *sk, 346 const unsigned int pkts) 347 { 348 struct inet_connection_sock *icsk = inet_csk(sk); 349 350 if (icsk->icsk_ack.quick) { 351 if (pkts >= icsk->icsk_ack.quick) { 352 icsk->icsk_ack.quick = 0; 353 /* Leaving quickack mode we deflate ATO. */ 354 icsk->icsk_ack.ato = TCP_ATO_MIN; 355 } else 356 icsk->icsk_ack.quick -= pkts; 357 } 358 } 359 360 #define TCP_ECN_OK 1 361 #define TCP_ECN_QUEUE_CWR 2 362 #define TCP_ECN_DEMAND_CWR 4 363 #define TCP_ECN_SEEN 8 364 365 enum tcp_tw_status { 366 TCP_TW_SUCCESS = 0, 367 TCP_TW_RST = 1, 368 TCP_TW_ACK = 2, 369 TCP_TW_SYN = 3 370 }; 371 372 373 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 374 struct sk_buff *skb, 375 const struct tcphdr *th); 376 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 377 struct request_sock *req, bool fastopen, 378 bool *lost_race); 379 int tcp_child_process(struct sock *parent, struct sock *child, 380 struct sk_buff *skb); 381 void tcp_enter_loss(struct sock *sk); 382 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 383 void tcp_clear_retrans(struct tcp_sock *tp); 384 void tcp_update_metrics(struct sock *sk); 385 void tcp_init_metrics(struct sock *sk); 386 void tcp_metrics_init(void); 387 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 388 void tcp_close(struct sock *sk, long timeout); 389 void tcp_init_sock(struct sock *sk); 390 void tcp_init_transfer(struct sock *sk, int bpf_op); 391 __poll_t tcp_poll(struct file *file, struct socket *sock, 392 struct poll_table_struct *wait); 393 int tcp_getsockopt(struct sock *sk, int level, int optname, 394 char __user *optval, int __user *optlen); 395 int tcp_setsockopt(struct sock *sk, int level, int optname, 396 char __user *optval, unsigned int optlen); 397 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, int __user *optlen); 399 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 400 char __user *optval, unsigned int optlen); 401 void tcp_set_keepalive(struct sock *sk, int val); 402 void tcp_syn_ack_timeout(const struct request_sock *req); 403 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 404 int flags, int *addr_len); 405 int tcp_set_rcvlowat(struct sock *sk, int val); 406 void tcp_data_ready(struct sock *sk); 407 int tcp_mmap(struct file *file, struct socket *sock, 408 struct vm_area_struct *vma); 409 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 410 struct tcp_options_received *opt_rx, 411 int estab, struct tcp_fastopen_cookie *foc); 412 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 413 414 /* 415 * TCP v4 functions exported for the inet6 API 416 */ 417 418 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 419 void tcp_v4_mtu_reduced(struct sock *sk); 420 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 421 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 422 struct sock *tcp_create_openreq_child(const struct sock *sk, 423 struct request_sock *req, 424 struct sk_buff *skb); 425 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 426 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 427 struct request_sock *req, 428 struct dst_entry *dst, 429 struct request_sock *req_unhash, 430 bool *own_req); 431 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 432 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 433 int tcp_connect(struct sock *sk); 434 enum tcp_synack_type { 435 TCP_SYNACK_NORMAL, 436 TCP_SYNACK_FASTOPEN, 437 TCP_SYNACK_COOKIE, 438 }; 439 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 440 struct request_sock *req, 441 struct tcp_fastopen_cookie *foc, 442 enum tcp_synack_type synack_type); 443 int tcp_disconnect(struct sock *sk, int flags); 444 445 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 446 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 447 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 448 449 /* From syncookies.c */ 450 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 451 struct request_sock *req, 452 struct dst_entry *dst, u32 tsoff); 453 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 454 u32 cookie); 455 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 456 #ifdef CONFIG_SYN_COOKIES 457 458 /* Syncookies use a monotonic timer which increments every 60 seconds. 459 * This counter is used both as a hash input and partially encoded into 460 * the cookie value. A cookie is only validated further if the delta 461 * between the current counter value and the encoded one is less than this, 462 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 463 * the counter advances immediately after a cookie is generated). 464 */ 465 #define MAX_SYNCOOKIE_AGE 2 466 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 467 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 468 469 /* syncookies: remember time of last synqueue overflow 470 * But do not dirty this field too often (once per second is enough) 471 * It is racy as we do not hold a lock, but race is very minor. 472 */ 473 static inline void tcp_synq_overflow(const struct sock *sk) 474 { 475 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 476 unsigned long now = jiffies; 477 478 if (time_after(now, last_overflow + HZ)) 479 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 480 } 481 482 /* syncookies: no recent synqueue overflow on this listening socket? */ 483 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 484 { 485 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 486 487 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 488 } 489 490 static inline u32 tcp_cookie_time(void) 491 { 492 u64 val = get_jiffies_64(); 493 494 do_div(val, TCP_SYNCOOKIE_PERIOD); 495 return val; 496 } 497 498 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 499 u16 *mssp); 500 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 501 u64 cookie_init_timestamp(struct request_sock *req); 502 bool cookie_timestamp_decode(const struct net *net, 503 struct tcp_options_received *opt); 504 bool cookie_ecn_ok(const struct tcp_options_received *opt, 505 const struct net *net, const struct dst_entry *dst); 506 507 /* From net/ipv6/syncookies.c */ 508 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 509 u32 cookie); 510 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 511 512 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 513 const struct tcphdr *th, u16 *mssp); 514 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 515 #endif 516 /* tcp_output.c */ 517 518 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, bool advancing_rto); 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 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 561 sock_put(sk); 562 563 inet_csk_clear_xmit_timers(sk); 564 } 565 566 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 567 unsigned int tcp_current_mss(struct sock *sk); 568 569 /* Bound MSS / TSO packet size with the half of the window */ 570 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 571 { 572 int cutoff; 573 574 /* When peer uses tiny windows, there is no use in packetizing 575 * to sub-MSS pieces for the sake of SWS or making sure there 576 * are enough packets in the pipe for fast recovery. 577 * 578 * On the other hand, for extremely large MSS devices, handling 579 * smaller than MSS windows in this way does make sense. 580 */ 581 if (tp->max_window > TCP_MSS_DEFAULT) 582 cutoff = (tp->max_window >> 1); 583 else 584 cutoff = tp->max_window; 585 586 if (cutoff && pktsize > cutoff) 587 return max_t(int, cutoff, 68U - tp->tcp_header_len); 588 else 589 return pktsize; 590 } 591 592 /* tcp.c */ 593 void tcp_get_info(struct sock *, struct tcp_info *); 594 595 /* Read 'sendfile()'-style from a TCP socket */ 596 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 597 sk_read_actor_t recv_actor); 598 599 void tcp_initialize_rcv_mss(struct sock *sk); 600 601 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 602 int tcp_mss_to_mtu(struct sock *sk, int mss); 603 void tcp_mtup_init(struct sock *sk); 604 void tcp_init_buffer_space(struct sock *sk); 605 606 static inline void tcp_bound_rto(const struct sock *sk) 607 { 608 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 609 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 610 } 611 612 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 613 { 614 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 615 } 616 617 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 618 { 619 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 620 ntohl(TCP_FLAG_ACK) | 621 snd_wnd); 622 } 623 624 static inline void tcp_fast_path_on(struct tcp_sock *tp) 625 { 626 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 627 } 628 629 static inline void tcp_fast_path_check(struct sock *sk) 630 { 631 struct tcp_sock *tp = tcp_sk(sk); 632 633 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 634 tp->rcv_wnd && 635 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 636 !tp->urg_data) 637 tcp_fast_path_on(tp); 638 } 639 640 /* Compute the actual rto_min value */ 641 static inline u32 tcp_rto_min(struct sock *sk) 642 { 643 const struct dst_entry *dst = __sk_dst_get(sk); 644 u32 rto_min = TCP_RTO_MIN; 645 646 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 647 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 648 return rto_min; 649 } 650 651 static inline u32 tcp_rto_min_us(struct sock *sk) 652 { 653 return jiffies_to_usecs(tcp_rto_min(sk)); 654 } 655 656 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 657 { 658 return dst_metric_locked(dst, RTAX_CC_ALGO); 659 } 660 661 /* Minimum RTT in usec. ~0 means not available. */ 662 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 663 { 664 return minmax_get(&tp->rtt_min); 665 } 666 667 /* Compute the actual receive window we are currently advertising. 668 * Rcv_nxt can be after the window if our peer push more data 669 * than the offered window. 670 */ 671 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 672 { 673 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 674 675 if (win < 0) 676 win = 0; 677 return (u32) win; 678 } 679 680 /* Choose a new window, without checks for shrinking, and without 681 * scaling applied to the result. The caller does these things 682 * if necessary. This is a "raw" window selection. 683 */ 684 u32 __tcp_select_window(struct sock *sk); 685 686 void tcp_send_window_probe(struct sock *sk); 687 688 /* TCP uses 32bit jiffies to save some space. 689 * Note that this is different from tcp_time_stamp, which 690 * historically has been the same until linux-4.13. 691 */ 692 #define tcp_jiffies32 ((u32)jiffies) 693 694 /* 695 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 696 * It is no longer tied to jiffies, but to 1 ms clock. 697 * Note: double check if you want to use tcp_jiffies32 instead of this. 698 */ 699 #define TCP_TS_HZ 1000 700 701 static inline u64 tcp_clock_ns(void) 702 { 703 return local_clock(); 704 } 705 706 static inline u64 tcp_clock_us(void) 707 { 708 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 709 } 710 711 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 712 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 713 { 714 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 715 } 716 717 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 718 static inline u32 tcp_time_stamp_raw(void) 719 { 720 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 721 } 722 723 724 /* Refresh 1us clock of a TCP socket, 725 * ensuring monotically increasing values. 726 */ 727 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 728 { 729 u64 val = tcp_clock_us(); 730 731 if (val > tp->tcp_mstamp) 732 tp->tcp_mstamp = val; 733 } 734 735 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 736 { 737 return max_t(s64, t1 - t0, 0); 738 } 739 740 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 741 { 742 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 743 } 744 745 746 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 747 748 #define TCPHDR_FIN 0x01 749 #define TCPHDR_SYN 0x02 750 #define TCPHDR_RST 0x04 751 #define TCPHDR_PSH 0x08 752 #define TCPHDR_ACK 0x10 753 #define TCPHDR_URG 0x20 754 #define TCPHDR_ECE 0x40 755 #define TCPHDR_CWR 0x80 756 757 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 758 759 /* This is what the send packet queuing engine uses to pass 760 * TCP per-packet control information to the transmission code. 761 * We also store the host-order sequence numbers in here too. 762 * This is 44 bytes if IPV6 is enabled. 763 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 764 */ 765 struct tcp_skb_cb { 766 __u32 seq; /* Starting sequence number */ 767 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 768 union { 769 /* Note : tcp_tw_isn is used in input path only 770 * (isn chosen by tcp_timewait_state_process()) 771 * 772 * tcp_gso_segs/size are used in write queue only, 773 * cf tcp_skb_pcount()/tcp_skb_mss() 774 */ 775 __u32 tcp_tw_isn; 776 struct { 777 u16 tcp_gso_segs; 778 u16 tcp_gso_size; 779 }; 780 }; 781 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 782 783 __u8 sacked; /* State flags for SACK. */ 784 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 785 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 786 #define TCPCB_LOST 0x04 /* SKB is lost */ 787 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 788 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 789 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 790 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 791 TCPCB_REPAIRED) 792 793 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 794 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 795 eor:1, /* Is skb MSG_EOR marked? */ 796 has_rxtstamp:1, /* SKB has a RX timestamp */ 797 unused:5; 798 __u32 ack_seq; /* Sequence number ACK'd */ 799 union { 800 struct { 801 /* There is space for up to 24 bytes */ 802 __u32 in_flight:30,/* Bytes in flight at transmit */ 803 is_app_limited:1, /* cwnd not fully used? */ 804 unused:1; 805 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 806 __u32 delivered; 807 /* start of send pipeline phase */ 808 u64 first_tx_mstamp; 809 /* when we reached the "delivered" count */ 810 u64 delivered_mstamp; 811 } tx; /* only used for outgoing skbs */ 812 union { 813 struct inet_skb_parm h4; 814 #if IS_ENABLED(CONFIG_IPV6) 815 struct inet6_skb_parm h6; 816 #endif 817 } header; /* For incoming skbs */ 818 struct { 819 __u32 flags; 820 struct sock *sk_redir; 821 void *data_end; 822 } bpf; 823 }; 824 }; 825 826 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 827 828 829 #if IS_ENABLED(CONFIG_IPV6) 830 /* This is the variant of inet6_iif() that must be used by TCP, 831 * as TCP moves IP6CB into a different location in skb->cb[] 832 */ 833 static inline int tcp_v6_iif(const struct sk_buff *skb) 834 { 835 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 836 837 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 838 } 839 840 /* TCP_SKB_CB reference means this can not be used from early demux */ 841 static inline int tcp_v6_sdif(const struct sk_buff *skb) 842 { 843 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 844 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 845 return TCP_SKB_CB(skb)->header.h6.iif; 846 #endif 847 return 0; 848 } 849 #endif 850 851 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(IPCB(skb)->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 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 961 }; 962 963 struct tcp_congestion_ops { 964 struct list_head list; 965 u32 key; 966 u32 flags; 967 968 /* initialize private data (optional) */ 969 void (*init)(struct sock *sk); 970 /* cleanup private data (optional) */ 971 void (*release)(struct sock *sk); 972 973 /* return slow start threshold (required) */ 974 u32 (*ssthresh)(struct sock *sk); 975 /* do new cwnd calculation (required) */ 976 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 977 /* call before changing ca_state (optional) */ 978 void (*set_state)(struct sock *sk, u8 new_state); 979 /* call when cwnd event occurs (optional) */ 980 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 981 /* call when ack arrives (optional) */ 982 void (*in_ack_event)(struct sock *sk, u32 flags); 983 /* new value of cwnd after loss (required) */ 984 u32 (*undo_cwnd)(struct sock *sk); 985 /* hook for packet ack accounting (optional) */ 986 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 987 /* override sysctl_tcp_min_tso_segs */ 988 u32 (*min_tso_segs)(struct sock *sk); 989 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 990 u32 (*sndbuf_expand)(struct sock *sk); 991 /* call when packets are delivered to update cwnd and pacing rate, 992 * after all the ca_state processing. (optional) 993 */ 994 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 995 /* get info for inet_diag (optional) */ 996 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 997 union tcp_cc_info *info); 998 999 char name[TCP_CA_NAME_MAX]; 1000 struct module *owner; 1001 }; 1002 1003 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1004 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1005 1006 void tcp_assign_congestion_control(struct sock *sk); 1007 void tcp_init_congestion_control(struct sock *sk); 1008 void tcp_cleanup_congestion_control(struct sock *sk); 1009 int tcp_set_default_congestion_control(struct net *net, const char *name); 1010 void tcp_get_default_congestion_control(struct net *net, char *name); 1011 void tcp_get_available_congestion_control(char *buf, size_t len); 1012 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1013 int tcp_set_allowed_congestion_control(char *allowed); 1014 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1015 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1016 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1017 1018 u32 tcp_reno_ssthresh(struct sock *sk); 1019 u32 tcp_reno_undo_cwnd(struct sock *sk); 1020 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1021 extern struct tcp_congestion_ops tcp_reno; 1022 1023 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1024 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1025 #ifdef CONFIG_INET 1026 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1027 #else 1028 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1029 { 1030 return NULL; 1031 } 1032 #endif 1033 1034 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1035 { 1036 const struct inet_connection_sock *icsk = inet_csk(sk); 1037 1038 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1039 } 1040 1041 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1042 { 1043 struct inet_connection_sock *icsk = inet_csk(sk); 1044 1045 if (icsk->icsk_ca_ops->set_state) 1046 icsk->icsk_ca_ops->set_state(sk, ca_state); 1047 icsk->icsk_ca_state = ca_state; 1048 } 1049 1050 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1051 { 1052 const struct inet_connection_sock *icsk = inet_csk(sk); 1053 1054 if (icsk->icsk_ca_ops->cwnd_event) 1055 icsk->icsk_ca_ops->cwnd_event(sk, event); 1056 } 1057 1058 /* From tcp_rate.c */ 1059 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1060 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1061 struct rate_sample *rs); 1062 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1063 bool is_sack_reneg, struct rate_sample *rs); 1064 void tcp_rate_check_app_limited(struct sock *sk); 1065 1066 /* These functions determine how the current flow behaves in respect of SACK 1067 * handling. SACK is negotiated with the peer, and therefore it can vary 1068 * between different flows. 1069 * 1070 * tcp_is_sack - SACK enabled 1071 * tcp_is_reno - No SACK 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 unsigned int tcp_left_out(const struct tcp_sock *tp) 1084 { 1085 return tp->sacked_out + tp->lost_out; 1086 } 1087 1088 /* This determines how many packets are "in the network" to the best 1089 * of our knowledge. In many cases it is conservative, but where 1090 * detailed information is available from the receiver (via SACK 1091 * blocks etc.) we can make more aggressive calculations. 1092 * 1093 * Use this for decisions involving congestion control, use just 1094 * tp->packets_out to determine if the send queue is empty or not. 1095 * 1096 * Read this equation as: 1097 * 1098 * "Packets sent once on transmission queue" MINUS 1099 * "Packets left network, but not honestly ACKed yet" PLUS 1100 * "Packets fast retransmitted" 1101 */ 1102 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1103 { 1104 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1105 } 1106 1107 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1108 1109 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1110 { 1111 return tp->snd_cwnd < tp->snd_ssthresh; 1112 } 1113 1114 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1115 { 1116 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1117 } 1118 1119 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1120 { 1121 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1122 (1 << inet_csk(sk)->icsk_ca_state); 1123 } 1124 1125 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1126 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1127 * ssthresh. 1128 */ 1129 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1130 { 1131 const struct tcp_sock *tp = tcp_sk(sk); 1132 1133 if (tcp_in_cwnd_reduction(sk)) 1134 return tp->snd_ssthresh; 1135 else 1136 return max(tp->snd_ssthresh, 1137 ((tp->snd_cwnd >> 1) + 1138 (tp->snd_cwnd >> 2))); 1139 } 1140 1141 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1142 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1143 1144 void tcp_enter_cwr(struct sock *sk); 1145 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1146 1147 /* The maximum number of MSS of available cwnd for which TSO defers 1148 * sending if not using sysctl_tcp_tso_win_divisor. 1149 */ 1150 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1151 { 1152 return 3; 1153 } 1154 1155 /* Returns end sequence number of the receiver's advertised window */ 1156 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1157 { 1158 return tp->snd_una + tp->snd_wnd; 1159 } 1160 1161 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1162 * flexible approach. The RFC suggests cwnd should not be raised unless 1163 * it was fully used previously. And that's exactly what we do in 1164 * congestion avoidance mode. But in slow start we allow cwnd to grow 1165 * as long as the application has used half the cwnd. 1166 * Example : 1167 * cwnd is 10 (IW10), but application sends 9 frames. 1168 * We allow cwnd to reach 18 when all frames are ACKed. 1169 * This check is safe because it's as aggressive as slow start which already 1170 * risks 100% overshoot. The advantage is that we discourage application to 1171 * either send more filler packets or data to artificially blow up the cwnd 1172 * usage, and allow application-limited process to probe bw more aggressively. 1173 */ 1174 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1175 { 1176 const struct tcp_sock *tp = tcp_sk(sk); 1177 1178 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1179 if (tcp_in_slow_start(tp)) 1180 return tp->snd_cwnd < 2 * tp->max_packets_out; 1181 1182 return tp->is_cwnd_limited; 1183 } 1184 1185 /* Something is really bad, we could not queue an additional packet, 1186 * because qdisc is full or receiver sent a 0 window. 1187 * We do not want to add fuel to the fire, or abort too early, 1188 * so make sure the timer we arm now is at least 200ms in the future, 1189 * regardless of current icsk_rto value (as it could be ~2ms) 1190 */ 1191 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1192 { 1193 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1194 } 1195 1196 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1197 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1198 unsigned long max_when) 1199 { 1200 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1201 1202 return (unsigned long)min_t(u64, when, max_when); 1203 } 1204 1205 static inline void tcp_check_probe_timer(struct sock *sk) 1206 { 1207 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1208 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1209 tcp_probe0_base(sk), TCP_RTO_MAX); 1210 } 1211 1212 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1213 { 1214 tp->snd_wl1 = seq; 1215 } 1216 1217 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1218 { 1219 tp->snd_wl1 = seq; 1220 } 1221 1222 /* 1223 * Calculate(/check) TCP checksum 1224 */ 1225 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1226 __be32 daddr, __wsum base) 1227 { 1228 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1229 } 1230 1231 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1232 { 1233 return __skb_checksum_complete(skb); 1234 } 1235 1236 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1237 { 1238 return !skb_csum_unnecessary(skb) && 1239 __tcp_checksum_complete(skb); 1240 } 1241 1242 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1243 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1244 1245 #undef STATE_TRACE 1246 1247 #ifdef STATE_TRACE 1248 static const char *statename[]={ 1249 "Unused","Established","Syn Sent","Syn Recv", 1250 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1251 "Close Wait","Last ACK","Listen","Closing" 1252 }; 1253 #endif 1254 void tcp_set_state(struct sock *sk, int state); 1255 1256 void tcp_done(struct sock *sk); 1257 1258 int tcp_abort(struct sock *sk, int err); 1259 1260 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1261 { 1262 rx_opt->dsack = 0; 1263 rx_opt->num_sacks = 0; 1264 } 1265 1266 u32 tcp_default_init_rwnd(u32 mss); 1267 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1268 1269 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1270 { 1271 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1272 struct tcp_sock *tp = tcp_sk(sk); 1273 s32 delta; 1274 1275 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1276 ca_ops->cong_control) 1277 return; 1278 delta = tcp_jiffies32 - tp->lsndtime; 1279 if (delta > inet_csk(sk)->icsk_rto) 1280 tcp_cwnd_restart(sk, delta); 1281 } 1282 1283 /* Determine a window scaling and initial window to offer. */ 1284 void tcp_select_initial_window(const struct sock *sk, int __space, 1285 __u32 mss, __u32 *rcv_wnd, 1286 __u32 *window_clamp, int wscale_ok, 1287 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1288 1289 static inline int tcp_win_from_space(const struct sock *sk, int space) 1290 { 1291 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1292 1293 return tcp_adv_win_scale <= 0 ? 1294 (space>>(-tcp_adv_win_scale)) : 1295 space - (space>>tcp_adv_win_scale); 1296 } 1297 1298 /* Note: caller must be prepared to deal with negative returns */ 1299 static inline int tcp_space(const struct sock *sk) 1300 { 1301 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1302 atomic_read(&sk->sk_rmem_alloc)); 1303 } 1304 1305 static inline int tcp_full_space(const struct sock *sk) 1306 { 1307 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1308 } 1309 1310 extern void tcp_openreq_init_rwin(struct request_sock *req, 1311 const struct sock *sk_listener, 1312 const struct dst_entry *dst); 1313 1314 void tcp_enter_memory_pressure(struct sock *sk); 1315 void tcp_leave_memory_pressure(struct sock *sk); 1316 1317 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1318 { 1319 struct net *net = sock_net((struct sock *)tp); 1320 1321 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1322 } 1323 1324 static inline int keepalive_time_when(const struct tcp_sock *tp) 1325 { 1326 struct net *net = sock_net((struct sock *)tp); 1327 1328 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1329 } 1330 1331 static inline int keepalive_probes(const struct tcp_sock *tp) 1332 { 1333 struct net *net = sock_net((struct sock *)tp); 1334 1335 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1336 } 1337 1338 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1339 { 1340 const struct inet_connection_sock *icsk = &tp->inet_conn; 1341 1342 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1343 tcp_jiffies32 - tp->rcv_tstamp); 1344 } 1345 1346 static inline int tcp_fin_time(const struct sock *sk) 1347 { 1348 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1349 const int rto = inet_csk(sk)->icsk_rto; 1350 1351 if (fin_timeout < (rto << 2) - (rto >> 1)) 1352 fin_timeout = (rto << 2) - (rto >> 1); 1353 1354 return fin_timeout; 1355 } 1356 1357 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1358 int paws_win) 1359 { 1360 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1361 return true; 1362 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1363 return true; 1364 /* 1365 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1366 * then following tcp messages have valid values. Ignore 0 value, 1367 * or else 'negative' tsval might forbid us to accept their packets. 1368 */ 1369 if (!rx_opt->ts_recent) 1370 return true; 1371 return false; 1372 } 1373 1374 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1375 int rst) 1376 { 1377 if (tcp_paws_check(rx_opt, 0)) 1378 return false; 1379 1380 /* RST segments are not recommended to carry timestamp, 1381 and, if they do, it is recommended to ignore PAWS because 1382 "their cleanup function should take precedence over timestamps." 1383 Certainly, it is mistake. It is necessary to understand the reasons 1384 of this constraint to relax it: if peer reboots, clock may go 1385 out-of-sync and half-open connections will not be reset. 1386 Actually, the problem would be not existing if all 1387 the implementations followed draft about maintaining clock 1388 via reboots. Linux-2.2 DOES NOT! 1389 1390 However, we can relax time bounds for RST segments to MSL. 1391 */ 1392 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1393 return false; 1394 return true; 1395 } 1396 1397 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1398 int mib_idx, u32 *last_oow_ack_time); 1399 1400 static inline void tcp_mib_init(struct net *net) 1401 { 1402 /* See RFC 2012 */ 1403 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1404 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1405 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1406 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1407 } 1408 1409 /* from STCP */ 1410 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1411 { 1412 tp->lost_skb_hint = NULL; 1413 } 1414 1415 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1416 { 1417 tcp_clear_retrans_hints_partial(tp); 1418 tp->retransmit_skb_hint = NULL; 1419 } 1420 1421 union tcp_md5_addr { 1422 struct in_addr a4; 1423 #if IS_ENABLED(CONFIG_IPV6) 1424 struct in6_addr a6; 1425 #endif 1426 }; 1427 1428 /* - key database */ 1429 struct tcp_md5sig_key { 1430 struct hlist_node node; 1431 u8 keylen; 1432 u8 family; /* AF_INET or AF_INET6 */ 1433 union tcp_md5_addr addr; 1434 u8 prefixlen; 1435 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1436 struct rcu_head rcu; 1437 }; 1438 1439 /* - sock block */ 1440 struct tcp_md5sig_info { 1441 struct hlist_head head; 1442 struct rcu_head rcu; 1443 }; 1444 1445 /* - pseudo header */ 1446 struct tcp4_pseudohdr { 1447 __be32 saddr; 1448 __be32 daddr; 1449 __u8 pad; 1450 __u8 protocol; 1451 __be16 len; 1452 }; 1453 1454 struct tcp6_pseudohdr { 1455 struct in6_addr saddr; 1456 struct in6_addr daddr; 1457 __be32 len; 1458 __be32 protocol; /* including padding */ 1459 }; 1460 1461 union tcp_md5sum_block { 1462 struct tcp4_pseudohdr ip4; 1463 #if IS_ENABLED(CONFIG_IPV6) 1464 struct tcp6_pseudohdr ip6; 1465 #endif 1466 }; 1467 1468 /* - pool: digest algorithm, hash description and scratch buffer */ 1469 struct tcp_md5sig_pool { 1470 struct ahash_request *md5_req; 1471 void *scratch; 1472 }; 1473 1474 /* - functions */ 1475 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1476 const struct sock *sk, const struct sk_buff *skb); 1477 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1478 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1479 gfp_t gfp); 1480 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1481 int family, u8 prefixlen); 1482 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1483 const struct sock *addr_sk); 1484 1485 #ifdef CONFIG_TCP_MD5SIG 1486 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1487 const union tcp_md5_addr *addr, 1488 int family); 1489 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1490 #else 1491 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1492 const union tcp_md5_addr *addr, 1493 int family) 1494 { 1495 return NULL; 1496 } 1497 #define tcp_twsk_md5_key(twsk) NULL 1498 #endif 1499 1500 bool tcp_alloc_md5sig_pool(void); 1501 1502 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1503 static inline void tcp_put_md5sig_pool(void) 1504 { 1505 local_bh_enable(); 1506 } 1507 1508 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1509 unsigned int header_len); 1510 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1511 const struct tcp_md5sig_key *key); 1512 1513 /* From tcp_fastopen.c */ 1514 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1515 struct tcp_fastopen_cookie *cookie); 1516 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1517 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1518 u16 try_exp); 1519 struct tcp_fastopen_request { 1520 /* Fast Open cookie. Size 0 means a cookie request */ 1521 struct tcp_fastopen_cookie cookie; 1522 struct msghdr *data; /* data in MSG_FASTOPEN */ 1523 size_t size; 1524 int copied; /* queued in tcp_connect() */ 1525 }; 1526 void tcp_free_fastopen_req(struct tcp_sock *tp); 1527 void tcp_fastopen_destroy_cipher(struct sock *sk); 1528 void tcp_fastopen_ctx_destroy(struct net *net); 1529 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1530 void *key, unsigned int len); 1531 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1532 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1533 struct request_sock *req, 1534 struct tcp_fastopen_cookie *foc, 1535 const struct dst_entry *dst); 1536 void tcp_fastopen_init_key_once(struct net *net); 1537 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1538 struct tcp_fastopen_cookie *cookie); 1539 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1540 #define TCP_FASTOPEN_KEY_LENGTH 16 1541 1542 /* Fastopen key context */ 1543 struct tcp_fastopen_context { 1544 struct crypto_cipher *tfm; 1545 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1546 struct rcu_head rcu; 1547 }; 1548 1549 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1550 void tcp_fastopen_active_disable(struct sock *sk); 1551 bool tcp_fastopen_active_should_disable(struct sock *sk); 1552 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1553 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1554 1555 /* Latencies incurred by various limits for a sender. They are 1556 * chronograph-like stats that are mutually exclusive. 1557 */ 1558 enum tcp_chrono { 1559 TCP_CHRONO_UNSPEC, 1560 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1561 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1562 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1563 __TCP_CHRONO_MAX, 1564 }; 1565 1566 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1567 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1568 1569 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1570 * the same memory storage than skb->destructor/_skb_refdst 1571 */ 1572 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1573 { 1574 skb->destructor = NULL; 1575 skb->_skb_refdst = 0UL; 1576 } 1577 1578 #define tcp_skb_tsorted_save(skb) { \ 1579 unsigned long _save = skb->_skb_refdst; \ 1580 skb->_skb_refdst = 0UL; 1581 1582 #define tcp_skb_tsorted_restore(skb) \ 1583 skb->_skb_refdst = _save; \ 1584 } 1585 1586 void tcp_write_queue_purge(struct sock *sk); 1587 1588 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1589 { 1590 return skb_rb_first(&sk->tcp_rtx_queue); 1591 } 1592 1593 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1594 { 1595 return skb_peek(&sk->sk_write_queue); 1596 } 1597 1598 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1599 { 1600 return skb_peek_tail(&sk->sk_write_queue); 1601 } 1602 1603 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1604 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1605 1606 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1607 { 1608 return skb_peek(&sk->sk_write_queue); 1609 } 1610 1611 static inline bool tcp_skb_is_last(const struct sock *sk, 1612 const struct sk_buff *skb) 1613 { 1614 return skb_queue_is_last(&sk->sk_write_queue, skb); 1615 } 1616 1617 static inline bool tcp_write_queue_empty(const struct sock *sk) 1618 { 1619 return skb_queue_empty(&sk->sk_write_queue); 1620 } 1621 1622 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1623 { 1624 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1625 } 1626 1627 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1628 { 1629 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1630 } 1631 1632 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1633 { 1634 if (tcp_write_queue_empty(sk)) 1635 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1636 } 1637 1638 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1639 { 1640 __skb_queue_tail(&sk->sk_write_queue, skb); 1641 } 1642 1643 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1644 { 1645 __tcp_add_write_queue_tail(sk, skb); 1646 1647 /* Queue it, remembering where we must start sending. */ 1648 if (sk->sk_write_queue.next == skb) 1649 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1650 } 1651 1652 /* Insert new before skb on the write queue of sk. */ 1653 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1654 struct sk_buff *skb, 1655 struct sock *sk) 1656 { 1657 __skb_queue_before(&sk->sk_write_queue, skb, new); 1658 } 1659 1660 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1661 { 1662 tcp_skb_tsorted_anchor_cleanup(skb); 1663 __skb_unlink(skb, &sk->sk_write_queue); 1664 } 1665 1666 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1667 1668 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1669 { 1670 tcp_skb_tsorted_anchor_cleanup(skb); 1671 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1672 } 1673 1674 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1675 { 1676 list_del(&skb->tcp_tsorted_anchor); 1677 tcp_rtx_queue_unlink(skb, sk); 1678 sk_wmem_free_skb(sk, skb); 1679 } 1680 1681 static inline void tcp_push_pending_frames(struct sock *sk) 1682 { 1683 if (tcp_send_head(sk)) { 1684 struct tcp_sock *tp = tcp_sk(sk); 1685 1686 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1687 } 1688 } 1689 1690 /* Start sequence of the skb just after the highest skb with SACKed 1691 * bit, valid only if sacked_out > 0 or when the caller has ensured 1692 * validity by itself. 1693 */ 1694 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1695 { 1696 if (!tp->sacked_out) 1697 return tp->snd_una; 1698 1699 if (tp->highest_sack == NULL) 1700 return tp->snd_nxt; 1701 1702 return TCP_SKB_CB(tp->highest_sack)->seq; 1703 } 1704 1705 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1706 { 1707 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1708 } 1709 1710 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1711 { 1712 return tcp_sk(sk)->highest_sack; 1713 } 1714 1715 static inline void tcp_highest_sack_reset(struct sock *sk) 1716 { 1717 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1718 } 1719 1720 /* Called when old skb is about to be deleted and replaced by new skb */ 1721 static inline void tcp_highest_sack_replace(struct sock *sk, 1722 struct sk_buff *old, 1723 struct sk_buff *new) 1724 { 1725 if (old == tcp_highest_sack(sk)) 1726 tcp_sk(sk)->highest_sack = new; 1727 } 1728 1729 /* This helper checks if socket has IP_TRANSPARENT set */ 1730 static inline bool inet_sk_transparent(const struct sock *sk) 1731 { 1732 switch (sk->sk_state) { 1733 case TCP_TIME_WAIT: 1734 return inet_twsk(sk)->tw_transparent; 1735 case TCP_NEW_SYN_RECV: 1736 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1737 } 1738 return inet_sk(sk)->transparent; 1739 } 1740 1741 /* Determines whether this is a thin stream (which may suffer from 1742 * increased latency). Used to trigger latency-reducing mechanisms. 1743 */ 1744 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1745 { 1746 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1747 } 1748 1749 /* /proc */ 1750 enum tcp_seq_states { 1751 TCP_SEQ_STATE_LISTENING, 1752 TCP_SEQ_STATE_ESTABLISHED, 1753 }; 1754 1755 int tcp_seq_open(struct inode *inode, struct file *file); 1756 1757 struct tcp_seq_afinfo { 1758 char *name; 1759 sa_family_t family; 1760 const struct file_operations *seq_fops; 1761 struct seq_operations seq_ops; 1762 }; 1763 1764 struct tcp_iter_state { 1765 struct seq_net_private p; 1766 sa_family_t family; 1767 enum tcp_seq_states state; 1768 struct sock *syn_wait_sk; 1769 int bucket, offset, sbucket, num; 1770 loff_t last_pos; 1771 }; 1772 1773 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1774 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1775 1776 extern struct request_sock_ops tcp_request_sock_ops; 1777 extern struct request_sock_ops tcp6_request_sock_ops; 1778 1779 void tcp_v4_destroy_sock(struct sock *sk); 1780 1781 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1782 netdev_features_t features); 1783 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1784 int tcp_gro_complete(struct sk_buff *skb); 1785 1786 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1787 1788 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1789 { 1790 struct net *net = sock_net((struct sock *)tp); 1791 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1792 } 1793 1794 static inline bool tcp_stream_memory_free(const struct sock *sk) 1795 { 1796 const struct tcp_sock *tp = tcp_sk(sk); 1797 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1798 1799 return notsent_bytes < tcp_notsent_lowat(tp); 1800 } 1801 1802 #ifdef CONFIG_PROC_FS 1803 int tcp4_proc_init(void); 1804 void tcp4_proc_exit(void); 1805 #endif 1806 1807 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1808 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1809 const struct tcp_request_sock_ops *af_ops, 1810 struct sock *sk, struct sk_buff *skb); 1811 1812 /* TCP af-specific functions */ 1813 struct tcp_sock_af_ops { 1814 #ifdef CONFIG_TCP_MD5SIG 1815 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1816 const struct sock *addr_sk); 1817 int (*calc_md5_hash)(char *location, 1818 const struct tcp_md5sig_key *md5, 1819 const struct sock *sk, 1820 const struct sk_buff *skb); 1821 int (*md5_parse)(struct sock *sk, 1822 int optname, 1823 char __user *optval, 1824 int optlen); 1825 #endif 1826 }; 1827 1828 struct tcp_request_sock_ops { 1829 u16 mss_clamp; 1830 #ifdef CONFIG_TCP_MD5SIG 1831 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1832 const struct sock *addr_sk); 1833 int (*calc_md5_hash) (char *location, 1834 const struct tcp_md5sig_key *md5, 1835 const struct sock *sk, 1836 const struct sk_buff *skb); 1837 #endif 1838 void (*init_req)(struct request_sock *req, 1839 const struct sock *sk_listener, 1840 struct sk_buff *skb); 1841 #ifdef CONFIG_SYN_COOKIES 1842 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1843 __u16 *mss); 1844 #endif 1845 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1846 const struct request_sock *req); 1847 u32 (*init_seq)(const struct sk_buff *skb); 1848 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1849 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1850 struct flowi *fl, struct request_sock *req, 1851 struct tcp_fastopen_cookie *foc, 1852 enum tcp_synack_type synack_type); 1853 }; 1854 1855 #ifdef CONFIG_SYN_COOKIES 1856 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1857 const struct sock *sk, struct sk_buff *skb, 1858 __u16 *mss) 1859 { 1860 tcp_synq_overflow(sk); 1861 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1862 return ops->cookie_init_seq(skb, mss); 1863 } 1864 #else 1865 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1866 const struct sock *sk, struct sk_buff *skb, 1867 __u16 *mss) 1868 { 1869 return 0; 1870 } 1871 #endif 1872 1873 int tcpv4_offload_init(void); 1874 1875 void tcp_v4_init(void); 1876 void tcp_init(void); 1877 1878 /* tcp_recovery.c */ 1879 extern void tcp_rack_mark_lost(struct sock *sk); 1880 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1881 u64 xmit_time); 1882 extern void tcp_rack_reo_timeout(struct sock *sk); 1883 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1884 1885 /* At how many usecs into the future should the RTO fire? */ 1886 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1887 { 1888 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1889 u32 rto = inet_csk(sk)->icsk_rto; 1890 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1891 1892 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1893 } 1894 1895 /* 1896 * Save and compile IPv4 options, return a pointer to it 1897 */ 1898 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1899 struct sk_buff *skb) 1900 { 1901 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1902 struct ip_options_rcu *dopt = NULL; 1903 1904 if (opt->optlen) { 1905 int opt_size = sizeof(*dopt) + opt->optlen; 1906 1907 dopt = kmalloc(opt_size, GFP_ATOMIC); 1908 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1909 kfree(dopt); 1910 dopt = NULL; 1911 } 1912 } 1913 return dopt; 1914 } 1915 1916 /* locally generated TCP pure ACKs have skb->truesize == 2 1917 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1918 * This is much faster than dissecting the packet to find out. 1919 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1920 */ 1921 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1922 { 1923 return skb->truesize == 2; 1924 } 1925 1926 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1927 { 1928 skb->truesize = 2; 1929 } 1930 1931 static inline int tcp_inq(struct sock *sk) 1932 { 1933 struct tcp_sock *tp = tcp_sk(sk); 1934 int answ; 1935 1936 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1937 answ = 0; 1938 } else if (sock_flag(sk, SOCK_URGINLINE) || 1939 !tp->urg_data || 1940 before(tp->urg_seq, tp->copied_seq) || 1941 !before(tp->urg_seq, tp->rcv_nxt)) { 1942 1943 answ = tp->rcv_nxt - tp->copied_seq; 1944 1945 /* Subtract 1, if FIN was received */ 1946 if (answ && sock_flag(sk, SOCK_DONE)) 1947 answ--; 1948 } else { 1949 answ = tp->urg_seq - tp->copied_seq; 1950 } 1951 1952 return answ; 1953 } 1954 1955 int tcp_peek_len(struct socket *sock); 1956 1957 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1958 { 1959 u16 segs_in; 1960 1961 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1962 tp->segs_in += segs_in; 1963 if (skb->len > tcp_hdrlen(skb)) 1964 tp->data_segs_in += segs_in; 1965 } 1966 1967 /* 1968 * TCP listen path runs lockless. 1969 * We forced "struct sock" to be const qualified to make sure 1970 * we don't modify one of its field by mistake. 1971 * Here, we increment sk_drops which is an atomic_t, so we can safely 1972 * make sock writable again. 1973 */ 1974 static inline void tcp_listendrop(const struct sock *sk) 1975 { 1976 atomic_inc(&((struct sock *)sk)->sk_drops); 1977 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1978 } 1979 1980 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 1981 1982 /* 1983 * Interface for adding Upper Level Protocols over TCP 1984 */ 1985 1986 #define TCP_ULP_NAME_MAX 16 1987 #define TCP_ULP_MAX 128 1988 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 1989 1990 enum { 1991 TCP_ULP_TLS, 1992 TCP_ULP_BPF, 1993 }; 1994 1995 struct tcp_ulp_ops { 1996 struct list_head list; 1997 1998 /* initialize ulp */ 1999 int (*init)(struct sock *sk); 2000 /* cleanup ulp */ 2001 void (*release)(struct sock *sk); 2002 2003 int uid; 2004 char name[TCP_ULP_NAME_MAX]; 2005 bool user_visible; 2006 struct module *owner; 2007 }; 2008 int tcp_register_ulp(struct tcp_ulp_ops *type); 2009 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2010 int tcp_set_ulp(struct sock *sk, const char *name); 2011 int tcp_set_ulp_id(struct sock *sk, const int ulp); 2012 void tcp_get_available_ulp(char *buf, size_t len); 2013 void tcp_cleanup_ulp(struct sock *sk); 2014 2015 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2016 * is < 0, then the BPF op failed (for example if the loaded BPF 2017 * program does not support the chosen operation or there is no BPF 2018 * program loaded). 2019 */ 2020 #ifdef CONFIG_BPF 2021 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2022 { 2023 struct bpf_sock_ops_kern sock_ops; 2024 int ret; 2025 2026 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2027 if (sk_fullsock(sk)) { 2028 sock_ops.is_fullsock = 1; 2029 sock_owned_by_me(sk); 2030 } 2031 2032 sock_ops.sk = sk; 2033 sock_ops.op = op; 2034 if (nargs > 0) 2035 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2036 2037 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2038 if (ret == 0) 2039 ret = sock_ops.reply; 2040 else 2041 ret = -1; 2042 return ret; 2043 } 2044 2045 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2046 { 2047 u32 args[2] = {arg1, arg2}; 2048 2049 return tcp_call_bpf(sk, op, 2, args); 2050 } 2051 2052 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2053 u32 arg3) 2054 { 2055 u32 args[3] = {arg1, arg2, arg3}; 2056 2057 return tcp_call_bpf(sk, op, 3, args); 2058 } 2059 2060 #else 2061 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2062 { 2063 return -EPERM; 2064 } 2065 2066 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2067 { 2068 return -EPERM; 2069 } 2070 2071 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2072 u32 arg3) 2073 { 2074 return -EPERM; 2075 } 2076 2077 #endif 2078 2079 static inline u32 tcp_timeout_init(struct sock *sk) 2080 { 2081 int timeout; 2082 2083 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2084 2085 if (timeout <= 0) 2086 timeout = TCP_TIMEOUT_INIT; 2087 return timeout; 2088 } 2089 2090 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2091 { 2092 int rwnd; 2093 2094 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2095 2096 if (rwnd < 0) 2097 rwnd = 0; 2098 return rwnd; 2099 } 2100 2101 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2102 { 2103 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2104 } 2105 2106 #if IS_ENABLED(CONFIG_SMC) 2107 extern struct static_key_false tcp_have_smc; 2108 #endif 2109 2110 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2111 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2112 void (*cad)(struct sock *sk, u32 ack_seq)); 2113 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2114 2115 #endif 2116 2117 #endif /* _TCP_H */ 2118