1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define FASTRETRANS_DEBUG 1 22 23 #include <linux/list.h> 24 #include <linux/tcp.h> 25 #include <linux/bug.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/cryptohash.h> 31 #include <linux/kref.h> 32 #include <linux/ktime.h> 33 34 #include <net/inet_connection_sock.h> 35 #include <net/inet_timewait_sock.h> 36 #include <net/inet_hashtables.h> 37 #include <net/checksum.h> 38 #include <net/request_sock.h> 39 #include <net/sock.h> 40 #include <net/snmp.h> 41 #include <net/ip.h> 42 #include <net/tcp_states.h> 43 #include <net/inet_ecn.h> 44 #include <net/dst.h> 45 46 #include <linux/seq_file.h> 47 #include <linux/memcontrol.h> 48 49 #include <linux/bpf.h> 50 #include <linux/filter.h> 51 #include <linux/bpf-cgroup.h> 52 53 extern struct inet_hashinfo tcp_hashinfo; 54 55 extern struct percpu_counter tcp_orphan_count; 56 void tcp_time_wait(struct sock *sk, int state, int timeo); 57 58 #define MAX_TCP_HEADER (128 + MAX_HEADER) 59 #define MAX_TCP_OPTION_SPACE 40 60 61 /* 62 * Never offer a window over 32767 without using window scaling. Some 63 * poor stacks do signed 16bit maths! 64 */ 65 #define MAX_TCP_WINDOW 32767U 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The least MTU to use for probing */ 71 #define TCP_BASE_MSS 1024 72 73 /* probing interval, default to 10 minutes as per RFC4821 */ 74 #define TCP_PROBE_INTERVAL 600 75 76 /* Specify interval when tcp mtu probing will stop */ 77 #define TCP_PROBE_THRESHOLD 8 78 79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 80 #define TCP_FASTRETRANS_THRESH 3 81 82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 83 #define TCP_MAX_QUICKACKS 16U 84 85 /* Maximal number of window scale according to RFC1323 */ 86 #define TCP_MAX_WSCALE 14U 87 88 /* urg_data states */ 89 #define TCP_URG_VALID 0x0100 90 #define TCP_URG_NOTYET 0x0200 91 #define TCP_URG_READ 0x0400 92 93 #define TCP_RETR1 3 /* 94 * This is how many retries it does before it 95 * tries to figure out if the gateway is 96 * down. Minimal RFC value is 3; it corresponds 97 * to ~3sec-8min depending on RTO. 98 */ 99 100 #define TCP_RETR2 15 /* 101 * This should take at least 102 * 90 minutes to time out. 103 * RFC1122 says that the limit is 100 sec. 104 * 15 is ~13-30min depending on RTO. 105 */ 106 107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 108 * when active opening a connection. 109 * RFC1122 says the minimum retry MUST 110 * be at least 180secs. Nevertheless 111 * this value is corresponding to 112 * 63secs of retransmission with the 113 * current initial RTO. 114 */ 115 116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 117 * when passive opening a connection. 118 * This is corresponding to 31secs of 119 * retransmission with the current 120 * initial RTO. 121 */ 122 123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 124 * state, about 60 seconds */ 125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 126 /* BSD style FIN_WAIT2 deadlock breaker. 127 * It used to be 3min, new value is 60sec, 128 * to combine FIN-WAIT-2 timeout with 129 * TIME-WAIT timer. 130 */ 131 132 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 133 #if HZ >= 100 134 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 135 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 136 #else 137 #define TCP_DELACK_MIN 4U 138 #define TCP_ATO_MIN 4U 139 #endif 140 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 141 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 142 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 143 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 144 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 145 * used as a fallback RTO for the 146 * initial data transmission if no 147 * valid RTT sample has been acquired, 148 * most likely due to retrans in 3WHS. 149 */ 150 151 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 152 * for local resources. 153 */ 154 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 155 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 156 #define TCP_KEEPALIVE_INTVL (75*HZ) 157 158 #define MAX_TCP_KEEPIDLE 32767 159 #define MAX_TCP_KEEPINTVL 32767 160 #define MAX_TCP_KEEPCNT 127 161 #define MAX_TCP_SYNCNT 127 162 163 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 164 165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 167 * after this time. It should be equal 168 * (or greater than) TCP_TIMEWAIT_LEN 169 * to provide reliability equal to one 170 * provided by timewait state. 171 */ 172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 173 * timestamps. It must be less than 174 * minimal timewait lifetime. 175 */ 176 /* 177 * TCP option 178 */ 179 180 #define TCPOPT_NOP 1 /* Padding */ 181 #define TCPOPT_EOL 0 /* End of options */ 182 #define TCPOPT_MSS 2 /* Segment size negotiating */ 183 #define TCPOPT_WINDOW 3 /* Window scaling */ 184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 185 #define TCPOPT_SACK 5 /* SACK Block */ 186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 189 #define TCPOPT_EXP 254 /* Experimental */ 190 /* Magic number to be after the option value for sharing TCP 191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 192 */ 193 #define TCPOPT_FASTOPEN_MAGIC 0xF989 194 195 /* 196 * TCP option lengths 197 */ 198 199 #define TCPOLEN_MSS 4 200 #define TCPOLEN_WINDOW 3 201 #define TCPOLEN_SACK_PERM 2 202 #define TCPOLEN_TIMESTAMP 10 203 #define TCPOLEN_MD5SIG 18 204 #define TCPOLEN_FASTOPEN_BASE 2 205 #define TCPOLEN_EXP_FASTOPEN_BASE 4 206 207 /* But this is what stacks really send out. */ 208 #define TCPOLEN_TSTAMP_ALIGNED 12 209 #define TCPOLEN_WSCALE_ALIGNED 4 210 #define TCPOLEN_SACKPERM_ALIGNED 4 211 #define TCPOLEN_SACK_BASE 2 212 #define TCPOLEN_SACK_BASE_ALIGNED 4 213 #define TCPOLEN_SACK_PERBLOCK 8 214 #define TCPOLEN_MD5SIG_ALIGNED 20 215 #define TCPOLEN_MSS_ALIGNED 4 216 217 /* Flags in tp->nonagle */ 218 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 219 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 220 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 221 222 /* TCP thin-stream limits */ 223 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 224 225 /* TCP initial congestion window as per rfc6928 */ 226 #define TCP_INIT_CWND 10 227 228 /* Bit Flags for sysctl_tcp_fastopen */ 229 #define TFO_CLIENT_ENABLE 1 230 #define TFO_SERVER_ENABLE 2 231 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 232 233 /* Accept SYN data w/o any cookie option */ 234 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 235 236 /* Force enable TFO on all listeners, i.e., not requiring the 237 * TCP_FASTOPEN socket option. 238 */ 239 #define TFO_SERVER_WO_SOCKOPT1 0x400 240 241 242 /* sysctl variables for tcp */ 243 extern int sysctl_tcp_fastopen; 244 extern int sysctl_tcp_retrans_collapse; 245 extern int sysctl_tcp_stdurg; 246 extern int sysctl_tcp_rfc1337; 247 extern int sysctl_tcp_abort_on_overflow; 248 extern int sysctl_tcp_max_orphans; 249 extern int sysctl_tcp_fack; 250 extern int sysctl_tcp_reordering; 251 extern int sysctl_tcp_max_reordering; 252 extern int sysctl_tcp_dsack; 253 extern long sysctl_tcp_mem[3]; 254 extern int sysctl_tcp_wmem[3]; 255 extern int sysctl_tcp_rmem[3]; 256 extern int sysctl_tcp_app_win; 257 extern int sysctl_tcp_adv_win_scale; 258 extern int sysctl_tcp_frto; 259 extern int sysctl_tcp_nometrics_save; 260 extern int sysctl_tcp_moderate_rcvbuf; 261 extern int sysctl_tcp_tso_win_divisor; 262 extern int sysctl_tcp_workaround_signed_windows; 263 extern int sysctl_tcp_slow_start_after_idle; 264 extern int sysctl_tcp_thin_linear_timeouts; 265 extern int sysctl_tcp_thin_dupack; 266 extern int sysctl_tcp_early_retrans; 267 extern int sysctl_tcp_recovery; 268 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 269 270 extern int sysctl_tcp_limit_output_bytes; 271 extern int sysctl_tcp_challenge_ack_limit; 272 extern int sysctl_tcp_min_tso_segs; 273 extern int sysctl_tcp_min_rtt_wlen; 274 extern int sysctl_tcp_autocorking; 275 extern int sysctl_tcp_invalid_ratelimit; 276 extern int sysctl_tcp_pacing_ss_ratio; 277 extern int sysctl_tcp_pacing_ca_ratio; 278 279 extern atomic_long_t tcp_memory_allocated; 280 extern struct percpu_counter tcp_sockets_allocated; 281 extern unsigned long tcp_memory_pressure; 282 283 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 284 static inline bool tcp_under_memory_pressure(const struct sock *sk) 285 { 286 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 287 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 288 return true; 289 290 return tcp_memory_pressure; 291 } 292 /* 293 * The next routines deal with comparing 32 bit unsigned ints 294 * and worry about wraparound (automatic with unsigned arithmetic). 295 */ 296 297 static inline bool before(__u32 seq1, __u32 seq2) 298 { 299 return (__s32)(seq1-seq2) < 0; 300 } 301 #define after(seq2, seq1) before(seq1, seq2) 302 303 /* is s2<=s1<=s3 ? */ 304 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 305 { 306 return seq3 - seq2 >= seq1 - seq2; 307 } 308 309 static inline bool tcp_out_of_memory(struct sock *sk) 310 { 311 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 312 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 313 return true; 314 return false; 315 } 316 317 void sk_forced_mem_schedule(struct sock *sk, int size); 318 319 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 320 { 321 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 322 int orphans = percpu_counter_read_positive(ocp); 323 324 if (orphans << shift > sysctl_tcp_max_orphans) { 325 orphans = percpu_counter_sum_positive(ocp); 326 if (orphans << shift > sysctl_tcp_max_orphans) 327 return true; 328 } 329 return false; 330 } 331 332 bool tcp_check_oom(struct sock *sk, int shift); 333 334 335 extern struct proto tcp_prot; 336 337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 338 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 340 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 341 342 void tcp_tasklet_init(void); 343 344 void tcp_v4_err(struct sk_buff *skb, u32); 345 346 void tcp_shutdown(struct sock *sk, int how); 347 348 int tcp_v4_early_demux(struct sk_buff *skb); 349 int tcp_v4_rcv(struct sk_buff *skb); 350 351 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 352 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 353 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 354 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 355 int flags); 356 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 357 size_t size, int flags); 358 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 359 size_t size, int flags); 360 void tcp_release_cb(struct sock *sk); 361 void tcp_wfree(struct sk_buff *skb); 362 void tcp_write_timer_handler(struct sock *sk); 363 void tcp_delack_timer_handler(struct sock *sk); 364 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 365 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 366 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 367 const struct tcphdr *th); 368 void tcp_rcv_space_adjust(struct sock *sk); 369 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 370 void tcp_twsk_destructor(struct sock *sk); 371 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 372 struct pipe_inode_info *pipe, size_t len, 373 unsigned int flags); 374 375 static inline void tcp_dec_quickack_mode(struct sock *sk, 376 const unsigned int pkts) 377 { 378 struct inet_connection_sock *icsk = inet_csk(sk); 379 380 if (icsk->icsk_ack.quick) { 381 if (pkts >= icsk->icsk_ack.quick) { 382 icsk->icsk_ack.quick = 0; 383 /* Leaving quickack mode we deflate ATO. */ 384 icsk->icsk_ack.ato = TCP_ATO_MIN; 385 } else 386 icsk->icsk_ack.quick -= pkts; 387 } 388 } 389 390 #define TCP_ECN_OK 1 391 #define TCP_ECN_QUEUE_CWR 2 392 #define TCP_ECN_DEMAND_CWR 4 393 #define TCP_ECN_SEEN 8 394 395 enum tcp_tw_status { 396 TCP_TW_SUCCESS = 0, 397 TCP_TW_RST = 1, 398 TCP_TW_ACK = 2, 399 TCP_TW_SYN = 3 400 }; 401 402 403 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 404 struct sk_buff *skb, 405 const struct tcphdr *th); 406 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 407 struct request_sock *req, bool fastopen); 408 int tcp_child_process(struct sock *parent, struct sock *child, 409 struct sk_buff *skb); 410 void tcp_enter_loss(struct sock *sk); 411 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 412 void tcp_clear_retrans(struct tcp_sock *tp); 413 void tcp_update_metrics(struct sock *sk); 414 void tcp_init_metrics(struct sock *sk); 415 void tcp_metrics_init(void); 416 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 417 void tcp_disable_fack(struct tcp_sock *tp); 418 void tcp_close(struct sock *sk, long timeout); 419 void tcp_init_sock(struct sock *sk); 420 unsigned int tcp_poll(struct file *file, struct socket *sock, 421 struct poll_table_struct *wait); 422 int tcp_getsockopt(struct sock *sk, int level, int optname, 423 char __user *optval, int __user *optlen); 424 int tcp_setsockopt(struct sock *sk, int level, int optname, 425 char __user *optval, unsigned int optlen); 426 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 427 char __user *optval, int __user *optlen); 428 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 429 char __user *optval, unsigned int optlen); 430 void tcp_set_keepalive(struct sock *sk, int val); 431 void tcp_syn_ack_timeout(const struct request_sock *req); 432 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 433 int flags, int *addr_len); 434 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 435 struct tcp_options_received *opt_rx, 436 int estab, struct tcp_fastopen_cookie *foc); 437 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 438 439 /* 440 * TCP v4 functions exported for the inet6 API 441 */ 442 443 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 444 void tcp_v4_mtu_reduced(struct sock *sk); 445 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 446 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 447 struct sock *tcp_create_openreq_child(const struct sock *sk, 448 struct request_sock *req, 449 struct sk_buff *skb); 450 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 451 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 452 struct request_sock *req, 453 struct dst_entry *dst, 454 struct request_sock *req_unhash, 455 bool *own_req); 456 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 457 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 458 int tcp_connect(struct sock *sk); 459 enum tcp_synack_type { 460 TCP_SYNACK_NORMAL, 461 TCP_SYNACK_FASTOPEN, 462 TCP_SYNACK_COOKIE, 463 }; 464 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 465 struct request_sock *req, 466 struct tcp_fastopen_cookie *foc, 467 enum tcp_synack_type synack_type); 468 int tcp_disconnect(struct sock *sk, int flags); 469 470 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 471 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 472 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 473 474 /* From syncookies.c */ 475 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 476 struct request_sock *req, 477 struct dst_entry *dst, u32 tsoff); 478 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 479 u32 cookie); 480 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 481 #ifdef CONFIG_SYN_COOKIES 482 483 /* Syncookies use a monotonic timer which increments every 60 seconds. 484 * This counter is used both as a hash input and partially encoded into 485 * the cookie value. A cookie is only validated further if the delta 486 * between the current counter value and the encoded one is less than this, 487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 488 * the counter advances immediately after a cookie is generated). 489 */ 490 #define MAX_SYNCOOKIE_AGE 2 491 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 492 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 493 494 /* syncookies: remember time of last synqueue overflow 495 * But do not dirty this field too often (once per second is enough) 496 * It is racy as we do not hold a lock, but race is very minor. 497 */ 498 static inline void tcp_synq_overflow(const struct sock *sk) 499 { 500 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 501 unsigned long now = jiffies; 502 503 if (time_after(now, last_overflow + HZ)) 504 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 505 } 506 507 /* syncookies: no recent synqueue overflow on this listening socket? */ 508 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 509 { 510 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 511 512 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 513 } 514 515 static inline u32 tcp_cookie_time(void) 516 { 517 u64 val = get_jiffies_64(); 518 519 do_div(val, TCP_SYNCOOKIE_PERIOD); 520 return val; 521 } 522 523 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 524 u16 *mssp); 525 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 526 u64 cookie_init_timestamp(struct request_sock *req); 527 bool cookie_timestamp_decode(const struct net *net, 528 struct tcp_options_received *opt); 529 bool cookie_ecn_ok(const struct tcp_options_received *opt, 530 const struct net *net, const struct dst_entry *dst); 531 532 /* From net/ipv6/syncookies.c */ 533 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 534 u32 cookie); 535 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 536 537 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 538 const struct tcphdr *th, u16 *mssp); 539 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 540 #endif 541 /* tcp_output.c */ 542 543 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 544 int min_tso_segs); 545 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 546 int nonagle); 547 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 548 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 549 void tcp_retransmit_timer(struct sock *sk); 550 void tcp_xmit_retransmit_queue(struct sock *); 551 void tcp_simple_retransmit(struct sock *); 552 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 553 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 554 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 555 556 void tcp_send_probe0(struct sock *); 557 void tcp_send_partial(struct sock *); 558 int tcp_write_wakeup(struct sock *, int mib); 559 void tcp_send_fin(struct sock *sk); 560 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 561 int tcp_send_synack(struct sock *); 562 void tcp_push_one(struct sock *, unsigned int mss_now); 563 void tcp_send_ack(struct sock *sk); 564 void tcp_send_delayed_ack(struct sock *sk); 565 void tcp_send_loss_probe(struct sock *sk); 566 bool tcp_schedule_loss_probe(struct sock *sk); 567 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 568 const struct sk_buff *next_skb); 569 570 /* tcp_input.c */ 571 void tcp_rearm_rto(struct sock *sk); 572 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 573 void tcp_reset(struct sock *sk); 574 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 575 void tcp_fin(struct sock *sk); 576 577 /* tcp_timer.c */ 578 void tcp_init_xmit_timers(struct sock *); 579 static inline void tcp_clear_xmit_timers(struct sock *sk) 580 { 581 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 582 inet_csk_clear_xmit_timers(sk); 583 } 584 585 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 586 unsigned int tcp_current_mss(struct sock *sk); 587 588 /* Bound MSS / TSO packet size with the half of the window */ 589 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 590 { 591 int cutoff; 592 593 /* When peer uses tiny windows, there is no use in packetizing 594 * to sub-MSS pieces for the sake of SWS or making sure there 595 * are enough packets in the pipe for fast recovery. 596 * 597 * On the other hand, for extremely large MSS devices, handling 598 * smaller than MSS windows in this way does make sense. 599 */ 600 if (tp->max_window > TCP_MSS_DEFAULT) 601 cutoff = (tp->max_window >> 1); 602 else 603 cutoff = tp->max_window; 604 605 if (cutoff && pktsize > cutoff) 606 return max_t(int, cutoff, 68U - tp->tcp_header_len); 607 else 608 return pktsize; 609 } 610 611 /* tcp.c */ 612 void tcp_get_info(struct sock *, struct tcp_info *); 613 614 /* Read 'sendfile()'-style from a TCP socket */ 615 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 616 sk_read_actor_t recv_actor); 617 618 void tcp_initialize_rcv_mss(struct sock *sk); 619 620 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 621 int tcp_mss_to_mtu(struct sock *sk, int mss); 622 void tcp_mtup_init(struct sock *sk); 623 void tcp_init_buffer_space(struct sock *sk); 624 625 static inline void tcp_bound_rto(const struct sock *sk) 626 { 627 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 628 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 629 } 630 631 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 632 { 633 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 634 } 635 636 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 637 { 638 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 639 ntohl(TCP_FLAG_ACK) | 640 snd_wnd); 641 } 642 643 static inline void tcp_fast_path_on(struct tcp_sock *tp) 644 { 645 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 646 } 647 648 static inline void tcp_fast_path_check(struct sock *sk) 649 { 650 struct tcp_sock *tp = tcp_sk(sk); 651 652 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 653 tp->rcv_wnd && 654 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 655 !tp->urg_data) 656 tcp_fast_path_on(tp); 657 } 658 659 /* Compute the actual rto_min value */ 660 static inline u32 tcp_rto_min(struct sock *sk) 661 { 662 const struct dst_entry *dst = __sk_dst_get(sk); 663 u32 rto_min = TCP_RTO_MIN; 664 665 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 666 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 667 return rto_min; 668 } 669 670 static inline u32 tcp_rto_min_us(struct sock *sk) 671 { 672 return jiffies_to_usecs(tcp_rto_min(sk)); 673 } 674 675 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 676 { 677 return dst_metric_locked(dst, RTAX_CC_ALGO); 678 } 679 680 /* Minimum RTT in usec. ~0 means not available. */ 681 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 682 { 683 return minmax_get(&tp->rtt_min); 684 } 685 686 /* Compute the actual receive window we are currently advertising. 687 * Rcv_nxt can be after the window if our peer push more data 688 * than the offered window. 689 */ 690 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 691 { 692 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 693 694 if (win < 0) 695 win = 0; 696 return (u32) win; 697 } 698 699 /* Choose a new window, without checks for shrinking, and without 700 * scaling applied to the result. The caller does these things 701 * if necessary. This is a "raw" window selection. 702 */ 703 u32 __tcp_select_window(struct sock *sk); 704 705 void tcp_send_window_probe(struct sock *sk); 706 707 /* TCP uses 32bit jiffies to save some space. 708 * Note that this is different from tcp_time_stamp, which 709 * historically has been the same until linux-4.13. 710 */ 711 #define tcp_jiffies32 ((u32)jiffies) 712 713 /* 714 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 715 * It is no longer tied to jiffies, but to 1 ms clock. 716 * Note: double check if you want to use tcp_jiffies32 instead of this. 717 */ 718 #define TCP_TS_HZ 1000 719 720 static inline u64 tcp_clock_ns(void) 721 { 722 return local_clock(); 723 } 724 725 static inline u64 tcp_clock_us(void) 726 { 727 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 728 } 729 730 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 731 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 732 { 733 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 734 } 735 736 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 737 static inline u32 tcp_time_stamp_raw(void) 738 { 739 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 740 } 741 742 743 /* Refresh 1us clock of a TCP socket, 744 * ensuring monotically increasing values. 745 */ 746 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 747 { 748 u64 val = tcp_clock_us(); 749 750 if (val > tp->tcp_mstamp) 751 tp->tcp_mstamp = val; 752 } 753 754 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 755 { 756 return max_t(s64, t1 - t0, 0); 757 } 758 759 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 760 { 761 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 762 } 763 764 765 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 766 767 #define TCPHDR_FIN 0x01 768 #define TCPHDR_SYN 0x02 769 #define TCPHDR_RST 0x04 770 #define TCPHDR_PSH 0x08 771 #define TCPHDR_ACK 0x10 772 #define TCPHDR_URG 0x20 773 #define TCPHDR_ECE 0x40 774 #define TCPHDR_CWR 0x80 775 776 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 777 778 /* This is what the send packet queuing engine uses to pass 779 * TCP per-packet control information to the transmission code. 780 * We also store the host-order sequence numbers in here too. 781 * This is 44 bytes if IPV6 is enabled. 782 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 783 */ 784 struct tcp_skb_cb { 785 __u32 seq; /* Starting sequence number */ 786 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 787 union { 788 /* Note : tcp_tw_isn is used in input path only 789 * (isn chosen by tcp_timewait_state_process()) 790 * 791 * tcp_gso_segs/size are used in write queue only, 792 * cf tcp_skb_pcount()/tcp_skb_mss() 793 */ 794 __u32 tcp_tw_isn; 795 struct { 796 u16 tcp_gso_segs; 797 u16 tcp_gso_size; 798 }; 799 800 /* Used to stash the receive timestamp while this skb is in the 801 * out of order queue, as skb->tstamp is overwritten by the 802 * rbnode. 803 */ 804 ktime_t swtstamp; 805 }; 806 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 807 808 __u8 sacked; /* State flags for SACK/FACK. */ 809 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 810 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 811 #define TCPCB_LOST 0x04 /* SKB is lost */ 812 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 813 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 814 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 815 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 816 TCPCB_REPAIRED) 817 818 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 819 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 820 eor:1, /* Is skb MSG_EOR marked? */ 821 has_rxtstamp:1, /* SKB has a RX timestamp */ 822 unused:5; 823 __u32 ack_seq; /* Sequence number ACK'd */ 824 union { 825 struct { 826 /* There is space for up to 24 bytes */ 827 __u32 in_flight:30,/* Bytes in flight at transmit */ 828 is_app_limited:1, /* cwnd not fully used? */ 829 unused:1; 830 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 831 __u32 delivered; 832 /* start of send pipeline phase */ 833 u64 first_tx_mstamp; 834 /* when we reached the "delivered" count */ 835 u64 delivered_mstamp; 836 } tx; /* only used for outgoing skbs */ 837 union { 838 struct inet_skb_parm h4; 839 #if IS_ENABLED(CONFIG_IPV6) 840 struct inet6_skb_parm h6; 841 #endif 842 } header; /* For incoming skbs */ 843 struct { 844 __u32 key; 845 __u32 flags; 846 struct bpf_map *map; 847 } bpf; 848 }; 849 }; 850 851 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 852 853 854 #if IS_ENABLED(CONFIG_IPV6) 855 /* This is the variant of inet6_iif() that must be used by TCP, 856 * as TCP moves IP6CB into a different location in skb->cb[] 857 */ 858 static inline int tcp_v6_iif(const struct sk_buff *skb) 859 { 860 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 861 862 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 863 } 864 865 /* TCP_SKB_CB reference means this can not be used from early demux */ 866 static inline int tcp_v6_sdif(const struct sk_buff *skb) 867 { 868 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 869 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 870 return TCP_SKB_CB(skb)->header.h6.iif; 871 #endif 872 return 0; 873 } 874 #endif 875 876 /* TCP_SKB_CB reference means this can not be used from early demux */ 877 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 878 { 879 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 880 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 881 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 882 return true; 883 #endif 884 return false; 885 } 886 887 /* TCP_SKB_CB reference means this can not be used from early demux */ 888 static inline int tcp_v4_sdif(struct sk_buff *skb) 889 { 890 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 891 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 892 return TCP_SKB_CB(skb)->header.h4.iif; 893 #endif 894 return 0; 895 } 896 897 /* Due to TSO, an SKB can be composed of multiple actual 898 * packets. To keep these tracked properly, we use this. 899 */ 900 static inline int tcp_skb_pcount(const struct sk_buff *skb) 901 { 902 return TCP_SKB_CB(skb)->tcp_gso_segs; 903 } 904 905 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 906 { 907 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 908 } 909 910 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 911 { 912 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 913 } 914 915 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 916 static inline int tcp_skb_mss(const struct sk_buff *skb) 917 { 918 return TCP_SKB_CB(skb)->tcp_gso_size; 919 } 920 921 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 922 { 923 return likely(!TCP_SKB_CB(skb)->eor); 924 } 925 926 /* Events passed to congestion control interface */ 927 enum tcp_ca_event { 928 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 929 CA_EVENT_CWND_RESTART, /* congestion window restart */ 930 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 931 CA_EVENT_LOSS, /* loss timeout */ 932 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 933 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 934 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 935 CA_EVENT_NON_DELAYED_ACK, 936 }; 937 938 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 939 enum tcp_ca_ack_event_flags { 940 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 941 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 942 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 943 }; 944 945 /* 946 * Interface for adding new TCP congestion control handlers 947 */ 948 #define TCP_CA_NAME_MAX 16 949 #define TCP_CA_MAX 128 950 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 951 952 #define TCP_CA_UNSPEC 0 953 954 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 955 #define TCP_CONG_NON_RESTRICTED 0x1 956 /* Requires ECN/ECT set on all packets */ 957 #define TCP_CONG_NEEDS_ECN 0x2 958 959 union tcp_cc_info; 960 961 struct ack_sample { 962 u32 pkts_acked; 963 s32 rtt_us; 964 u32 in_flight; 965 }; 966 967 /* A rate sample measures the number of (original/retransmitted) data 968 * packets delivered "delivered" over an interval of time "interval_us". 969 * The tcp_rate.c code fills in the rate sample, and congestion 970 * control modules that define a cong_control function to run at the end 971 * of ACK processing can optionally chose to consult this sample when 972 * setting cwnd and pacing rate. 973 * A sample is invalid if "delivered" or "interval_us" is negative. 974 */ 975 struct rate_sample { 976 u64 prior_mstamp; /* starting timestamp for interval */ 977 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 978 s32 delivered; /* number of packets delivered over interval */ 979 long interval_us; /* time for tp->delivered to incr "delivered" */ 980 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 981 int losses; /* number of packets marked lost upon ACK */ 982 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 983 u32 prior_in_flight; /* in flight before this ACK */ 984 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 985 bool is_retrans; /* is sample from retransmission? */ 986 }; 987 988 struct tcp_congestion_ops { 989 struct list_head list; 990 u32 key; 991 u32 flags; 992 993 /* initialize private data (optional) */ 994 void (*init)(struct sock *sk); 995 /* cleanup private data (optional) */ 996 void (*release)(struct sock *sk); 997 998 /* return slow start threshold (required) */ 999 u32 (*ssthresh)(struct sock *sk); 1000 /* do new cwnd calculation (required) */ 1001 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1002 /* call before changing ca_state (optional) */ 1003 void (*set_state)(struct sock *sk, u8 new_state); 1004 /* call when cwnd event occurs (optional) */ 1005 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1006 /* call when ack arrives (optional) */ 1007 void (*in_ack_event)(struct sock *sk, u32 flags); 1008 /* new value of cwnd after loss (required) */ 1009 u32 (*undo_cwnd)(struct sock *sk); 1010 /* hook for packet ack accounting (optional) */ 1011 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1012 /* suggest number of segments for each skb to transmit (optional) */ 1013 u32 (*tso_segs_goal)(struct sock *sk); 1014 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1015 u32 (*sndbuf_expand)(struct sock *sk); 1016 /* call when packets are delivered to update cwnd and pacing rate, 1017 * after all the ca_state processing. (optional) 1018 */ 1019 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1020 /* get info for inet_diag (optional) */ 1021 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1022 union tcp_cc_info *info); 1023 1024 char name[TCP_CA_NAME_MAX]; 1025 struct module *owner; 1026 }; 1027 1028 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1029 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1030 1031 void tcp_assign_congestion_control(struct sock *sk); 1032 void tcp_init_congestion_control(struct sock *sk); 1033 void tcp_cleanup_congestion_control(struct sock *sk); 1034 int tcp_set_default_congestion_control(const char *name); 1035 void tcp_get_default_congestion_control(char *name); 1036 void tcp_get_available_congestion_control(char *buf, size_t len); 1037 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1038 int tcp_set_allowed_congestion_control(char *allowed); 1039 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1040 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1041 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1042 1043 u32 tcp_reno_ssthresh(struct sock *sk); 1044 u32 tcp_reno_undo_cwnd(struct sock *sk); 1045 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1046 extern struct tcp_congestion_ops tcp_reno; 1047 1048 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1049 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1050 #ifdef CONFIG_INET 1051 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1052 #else 1053 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1054 { 1055 return NULL; 1056 } 1057 #endif 1058 1059 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1060 { 1061 const struct inet_connection_sock *icsk = inet_csk(sk); 1062 1063 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1064 } 1065 1066 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1067 { 1068 struct inet_connection_sock *icsk = inet_csk(sk); 1069 1070 if (icsk->icsk_ca_ops->set_state) 1071 icsk->icsk_ca_ops->set_state(sk, ca_state); 1072 icsk->icsk_ca_state = ca_state; 1073 } 1074 1075 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1076 { 1077 const struct inet_connection_sock *icsk = inet_csk(sk); 1078 1079 if (icsk->icsk_ca_ops->cwnd_event) 1080 icsk->icsk_ca_ops->cwnd_event(sk, event); 1081 } 1082 1083 /* From tcp_rate.c */ 1084 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1085 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1086 struct rate_sample *rs); 1087 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1088 struct rate_sample *rs); 1089 void tcp_rate_check_app_limited(struct sock *sk); 1090 1091 /* These functions determine how the current flow behaves in respect of SACK 1092 * handling. SACK is negotiated with the peer, and therefore it can vary 1093 * between different flows. 1094 * 1095 * tcp_is_sack - SACK enabled 1096 * tcp_is_reno - No SACK 1097 * tcp_is_fack - FACK enabled, implies SACK enabled 1098 */ 1099 static inline int tcp_is_sack(const struct tcp_sock *tp) 1100 { 1101 return tp->rx_opt.sack_ok; 1102 } 1103 1104 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1105 { 1106 return !tcp_is_sack(tp); 1107 } 1108 1109 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1110 { 1111 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1112 } 1113 1114 static inline void tcp_enable_fack(struct tcp_sock *tp) 1115 { 1116 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1117 } 1118 1119 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1120 { 1121 return tp->sacked_out + tp->lost_out; 1122 } 1123 1124 /* This determines how many packets are "in the network" to the best 1125 * of our knowledge. In many cases it is conservative, but where 1126 * detailed information is available from the receiver (via SACK 1127 * blocks etc.) we can make more aggressive calculations. 1128 * 1129 * Use this for decisions involving congestion control, use just 1130 * tp->packets_out to determine if the send queue is empty or not. 1131 * 1132 * Read this equation as: 1133 * 1134 * "Packets sent once on transmission queue" MINUS 1135 * "Packets left network, but not honestly ACKed yet" PLUS 1136 * "Packets fast retransmitted" 1137 */ 1138 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1139 { 1140 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1141 } 1142 1143 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1144 1145 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1146 { 1147 return tp->snd_cwnd < tp->snd_ssthresh; 1148 } 1149 1150 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1151 { 1152 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1153 } 1154 1155 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1156 { 1157 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1158 (1 << inet_csk(sk)->icsk_ca_state); 1159 } 1160 1161 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1162 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1163 * ssthresh. 1164 */ 1165 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1166 { 1167 const struct tcp_sock *tp = tcp_sk(sk); 1168 1169 if (tcp_in_cwnd_reduction(sk)) 1170 return tp->snd_ssthresh; 1171 else 1172 return max(tp->snd_ssthresh, 1173 ((tp->snd_cwnd >> 1) + 1174 (tp->snd_cwnd >> 2))); 1175 } 1176 1177 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1178 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1179 1180 void tcp_enter_cwr(struct sock *sk); 1181 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1182 1183 /* The maximum number of MSS of available cwnd for which TSO defers 1184 * sending if not using sysctl_tcp_tso_win_divisor. 1185 */ 1186 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1187 { 1188 return 3; 1189 } 1190 1191 /* Returns end sequence number of the receiver's advertised window */ 1192 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1193 { 1194 return tp->snd_una + tp->snd_wnd; 1195 } 1196 1197 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1198 * flexible approach. The RFC suggests cwnd should not be raised unless 1199 * it was fully used previously. And that's exactly what we do in 1200 * congestion avoidance mode. But in slow start we allow cwnd to grow 1201 * as long as the application has used half the cwnd. 1202 * Example : 1203 * cwnd is 10 (IW10), but application sends 9 frames. 1204 * We allow cwnd to reach 18 when all frames are ACKed. 1205 * This check is safe because it's as aggressive as slow start which already 1206 * risks 100% overshoot. The advantage is that we discourage application to 1207 * either send more filler packets or data to artificially blow up the cwnd 1208 * usage, and allow application-limited process to probe bw more aggressively. 1209 */ 1210 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1211 { 1212 const struct tcp_sock *tp = tcp_sk(sk); 1213 1214 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1215 if (tcp_in_slow_start(tp)) 1216 return tp->snd_cwnd < 2 * tp->max_packets_out; 1217 1218 return tp->is_cwnd_limited; 1219 } 1220 1221 /* Something is really bad, we could not queue an additional packet, 1222 * because qdisc is full or receiver sent a 0 window. 1223 * We do not want to add fuel to the fire, or abort too early, 1224 * so make sure the timer we arm now is at least 200ms in the future, 1225 * regardless of current icsk_rto value (as it could be ~2ms) 1226 */ 1227 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1228 { 1229 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1230 } 1231 1232 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1233 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1234 unsigned long max_when) 1235 { 1236 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1237 1238 return (unsigned long)min_t(u64, when, max_when); 1239 } 1240 1241 static inline void tcp_check_probe_timer(struct sock *sk) 1242 { 1243 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1244 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1245 tcp_probe0_base(sk), TCP_RTO_MAX); 1246 } 1247 1248 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1249 { 1250 tp->snd_wl1 = seq; 1251 } 1252 1253 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1254 { 1255 tp->snd_wl1 = seq; 1256 } 1257 1258 /* 1259 * Calculate(/check) TCP checksum 1260 */ 1261 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1262 __be32 daddr, __wsum base) 1263 { 1264 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1265 } 1266 1267 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1268 { 1269 return __skb_checksum_complete(skb); 1270 } 1271 1272 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1273 { 1274 return !skb_csum_unnecessary(skb) && 1275 __tcp_checksum_complete(skb); 1276 } 1277 1278 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1279 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1280 1281 #undef STATE_TRACE 1282 1283 #ifdef STATE_TRACE 1284 static const char *statename[]={ 1285 "Unused","Established","Syn Sent","Syn Recv", 1286 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1287 "Close Wait","Last ACK","Listen","Closing" 1288 }; 1289 #endif 1290 void tcp_set_state(struct sock *sk, int state); 1291 1292 void tcp_done(struct sock *sk); 1293 1294 int tcp_abort(struct sock *sk, int err); 1295 1296 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1297 { 1298 rx_opt->dsack = 0; 1299 rx_opt->num_sacks = 0; 1300 } 1301 1302 u32 tcp_default_init_rwnd(u32 mss); 1303 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1304 1305 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1306 { 1307 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1308 struct tcp_sock *tp = tcp_sk(sk); 1309 s32 delta; 1310 1311 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out || 1312 ca_ops->cong_control) 1313 return; 1314 delta = tcp_jiffies32 - tp->lsndtime; 1315 if (delta > inet_csk(sk)->icsk_rto) 1316 tcp_cwnd_restart(sk, delta); 1317 } 1318 1319 /* Determine a window scaling and initial window to offer. */ 1320 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1321 __u32 *window_clamp, int wscale_ok, 1322 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1323 1324 static inline int tcp_win_from_space(int space) 1325 { 1326 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale; 1327 1328 return tcp_adv_win_scale <= 0 ? 1329 (space>>(-tcp_adv_win_scale)) : 1330 space - (space>>tcp_adv_win_scale); 1331 } 1332 1333 /* Note: caller must be prepared to deal with negative returns */ 1334 static inline int tcp_space(const struct sock *sk) 1335 { 1336 return tcp_win_from_space(sk->sk_rcvbuf - 1337 atomic_read(&sk->sk_rmem_alloc)); 1338 } 1339 1340 static inline int tcp_full_space(const struct sock *sk) 1341 { 1342 return tcp_win_from_space(sk->sk_rcvbuf); 1343 } 1344 1345 extern void tcp_openreq_init_rwin(struct request_sock *req, 1346 const struct sock *sk_listener, 1347 const struct dst_entry *dst); 1348 1349 void tcp_enter_memory_pressure(struct sock *sk); 1350 void tcp_leave_memory_pressure(struct sock *sk); 1351 1352 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1353 { 1354 struct net *net = sock_net((struct sock *)tp); 1355 1356 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1357 } 1358 1359 static inline int keepalive_time_when(const struct tcp_sock *tp) 1360 { 1361 struct net *net = sock_net((struct sock *)tp); 1362 1363 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1364 } 1365 1366 static inline int keepalive_probes(const struct tcp_sock *tp) 1367 { 1368 struct net *net = sock_net((struct sock *)tp); 1369 1370 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1371 } 1372 1373 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1374 { 1375 const struct inet_connection_sock *icsk = &tp->inet_conn; 1376 1377 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1378 tcp_jiffies32 - tp->rcv_tstamp); 1379 } 1380 1381 static inline int tcp_fin_time(const struct sock *sk) 1382 { 1383 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1384 const int rto = inet_csk(sk)->icsk_rto; 1385 1386 if (fin_timeout < (rto << 2) - (rto >> 1)) 1387 fin_timeout = (rto << 2) - (rto >> 1); 1388 1389 return fin_timeout; 1390 } 1391 1392 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1393 int paws_win) 1394 { 1395 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1396 return true; 1397 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1398 return true; 1399 /* 1400 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1401 * then following tcp messages have valid values. Ignore 0 value, 1402 * or else 'negative' tsval might forbid us to accept their packets. 1403 */ 1404 if (!rx_opt->ts_recent) 1405 return true; 1406 return false; 1407 } 1408 1409 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1410 int rst) 1411 { 1412 if (tcp_paws_check(rx_opt, 0)) 1413 return false; 1414 1415 /* RST segments are not recommended to carry timestamp, 1416 and, if they do, it is recommended to ignore PAWS because 1417 "their cleanup function should take precedence over timestamps." 1418 Certainly, it is mistake. It is necessary to understand the reasons 1419 of this constraint to relax it: if peer reboots, clock may go 1420 out-of-sync and half-open connections will not be reset. 1421 Actually, the problem would be not existing if all 1422 the implementations followed draft about maintaining clock 1423 via reboots. Linux-2.2 DOES NOT! 1424 1425 However, we can relax time bounds for RST segments to MSL. 1426 */ 1427 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1428 return false; 1429 return true; 1430 } 1431 1432 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1433 int mib_idx, u32 *last_oow_ack_time); 1434 1435 static inline void tcp_mib_init(struct net *net) 1436 { 1437 /* See RFC 2012 */ 1438 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1439 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1440 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1441 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1442 } 1443 1444 /* from STCP */ 1445 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1446 { 1447 tp->lost_skb_hint = NULL; 1448 } 1449 1450 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1451 { 1452 tcp_clear_retrans_hints_partial(tp); 1453 tp->retransmit_skb_hint = NULL; 1454 } 1455 1456 union tcp_md5_addr { 1457 struct in_addr a4; 1458 #if IS_ENABLED(CONFIG_IPV6) 1459 struct in6_addr a6; 1460 #endif 1461 }; 1462 1463 /* - key database */ 1464 struct tcp_md5sig_key { 1465 struct hlist_node node; 1466 u8 keylen; 1467 u8 family; /* AF_INET or AF_INET6 */ 1468 union tcp_md5_addr addr; 1469 u8 prefixlen; 1470 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1471 struct rcu_head rcu; 1472 }; 1473 1474 /* - sock block */ 1475 struct tcp_md5sig_info { 1476 struct hlist_head head; 1477 struct rcu_head rcu; 1478 }; 1479 1480 /* - pseudo header */ 1481 struct tcp4_pseudohdr { 1482 __be32 saddr; 1483 __be32 daddr; 1484 __u8 pad; 1485 __u8 protocol; 1486 __be16 len; 1487 }; 1488 1489 struct tcp6_pseudohdr { 1490 struct in6_addr saddr; 1491 struct in6_addr daddr; 1492 __be32 len; 1493 __be32 protocol; /* including padding */ 1494 }; 1495 1496 union tcp_md5sum_block { 1497 struct tcp4_pseudohdr ip4; 1498 #if IS_ENABLED(CONFIG_IPV6) 1499 struct tcp6_pseudohdr ip6; 1500 #endif 1501 }; 1502 1503 /* - pool: digest algorithm, hash description and scratch buffer */ 1504 struct tcp_md5sig_pool { 1505 struct ahash_request *md5_req; 1506 void *scratch; 1507 }; 1508 1509 /* - functions */ 1510 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1511 const struct sock *sk, const struct sk_buff *skb); 1512 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1513 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1514 gfp_t gfp); 1515 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1516 int family, u8 prefixlen); 1517 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1518 const struct sock *addr_sk); 1519 1520 #ifdef CONFIG_TCP_MD5SIG 1521 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1522 const union tcp_md5_addr *addr, 1523 int family); 1524 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1525 #else 1526 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1527 const union tcp_md5_addr *addr, 1528 int family) 1529 { 1530 return NULL; 1531 } 1532 #define tcp_twsk_md5_key(twsk) NULL 1533 #endif 1534 1535 bool tcp_alloc_md5sig_pool(void); 1536 1537 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1538 static inline void tcp_put_md5sig_pool(void) 1539 { 1540 local_bh_enable(); 1541 } 1542 1543 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1544 unsigned int header_len); 1545 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1546 const struct tcp_md5sig_key *key); 1547 1548 /* From tcp_fastopen.c */ 1549 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1550 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1551 unsigned long *last_syn_loss); 1552 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1553 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1554 u16 try_exp); 1555 struct tcp_fastopen_request { 1556 /* Fast Open cookie. Size 0 means a cookie request */ 1557 struct tcp_fastopen_cookie cookie; 1558 struct msghdr *data; /* data in MSG_FASTOPEN */ 1559 size_t size; 1560 int copied; /* queued in tcp_connect() */ 1561 }; 1562 void tcp_free_fastopen_req(struct tcp_sock *tp); 1563 1564 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1565 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1566 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1567 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1568 struct request_sock *req, 1569 struct tcp_fastopen_cookie *foc); 1570 void tcp_fastopen_init_key_once(bool publish); 1571 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1572 struct tcp_fastopen_cookie *cookie); 1573 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1574 #define TCP_FASTOPEN_KEY_LENGTH 16 1575 1576 /* Fastopen key context */ 1577 struct tcp_fastopen_context { 1578 struct crypto_cipher *tfm; 1579 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1580 struct rcu_head rcu; 1581 }; 1582 1583 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1584 void tcp_fastopen_active_disable(struct sock *sk); 1585 bool tcp_fastopen_active_should_disable(struct sock *sk); 1586 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1587 void tcp_fastopen_active_timeout_reset(void); 1588 1589 /* Latencies incurred by various limits for a sender. They are 1590 * chronograph-like stats that are mutually exclusive. 1591 */ 1592 enum tcp_chrono { 1593 TCP_CHRONO_UNSPEC, 1594 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1595 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1596 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1597 __TCP_CHRONO_MAX, 1598 }; 1599 1600 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1601 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1602 1603 /* write queue abstraction */ 1604 static inline void tcp_write_queue_purge(struct sock *sk) 1605 { 1606 struct sk_buff *skb; 1607 1608 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1609 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1610 sk_wmem_free_skb(sk, skb); 1611 sk_mem_reclaim(sk); 1612 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1613 } 1614 1615 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1616 { 1617 return skb_peek(&sk->sk_write_queue); 1618 } 1619 1620 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1621 { 1622 return skb_peek_tail(&sk->sk_write_queue); 1623 } 1624 1625 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1626 const struct sk_buff *skb) 1627 { 1628 return skb_queue_next(&sk->sk_write_queue, skb); 1629 } 1630 1631 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1632 const struct sk_buff *skb) 1633 { 1634 return skb_queue_prev(&sk->sk_write_queue, skb); 1635 } 1636 1637 #define tcp_for_write_queue(skb, sk) \ 1638 skb_queue_walk(&(sk)->sk_write_queue, skb) 1639 1640 #define tcp_for_write_queue_from(skb, sk) \ 1641 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1642 1643 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1644 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1645 1646 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1647 { 1648 return sk->sk_send_head; 1649 } 1650 1651 static inline bool tcp_skb_is_last(const struct sock *sk, 1652 const struct sk_buff *skb) 1653 { 1654 return skb_queue_is_last(&sk->sk_write_queue, skb); 1655 } 1656 1657 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1658 { 1659 if (tcp_skb_is_last(sk, skb)) 1660 sk->sk_send_head = NULL; 1661 else 1662 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1663 } 1664 1665 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1666 { 1667 if (sk->sk_send_head == skb_unlinked) { 1668 sk->sk_send_head = NULL; 1669 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1670 } 1671 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1672 tcp_sk(sk)->highest_sack = NULL; 1673 } 1674 1675 static inline void tcp_init_send_head(struct sock *sk) 1676 { 1677 sk->sk_send_head = NULL; 1678 } 1679 1680 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1681 { 1682 __skb_queue_tail(&sk->sk_write_queue, skb); 1683 } 1684 1685 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1686 { 1687 __tcp_add_write_queue_tail(sk, skb); 1688 1689 /* Queue it, remembering where we must start sending. */ 1690 if (sk->sk_send_head == NULL) { 1691 sk->sk_send_head = skb; 1692 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1693 1694 if (tcp_sk(sk)->highest_sack == NULL) 1695 tcp_sk(sk)->highest_sack = skb; 1696 } 1697 } 1698 1699 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1700 { 1701 __skb_queue_head(&sk->sk_write_queue, skb); 1702 } 1703 1704 /* Insert buff after skb on the write queue of sk. */ 1705 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1706 struct sk_buff *buff, 1707 struct sock *sk) 1708 { 1709 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1710 } 1711 1712 /* Insert new before skb on the write queue of sk. */ 1713 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1714 struct sk_buff *skb, 1715 struct sock *sk) 1716 { 1717 __skb_queue_before(&sk->sk_write_queue, skb, new); 1718 1719 if (sk->sk_send_head == skb) 1720 sk->sk_send_head = new; 1721 } 1722 1723 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1724 { 1725 __skb_unlink(skb, &sk->sk_write_queue); 1726 } 1727 1728 static inline bool tcp_write_queue_empty(struct sock *sk) 1729 { 1730 return skb_queue_empty(&sk->sk_write_queue); 1731 } 1732 1733 static inline void tcp_push_pending_frames(struct sock *sk) 1734 { 1735 if (tcp_send_head(sk)) { 1736 struct tcp_sock *tp = tcp_sk(sk); 1737 1738 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1739 } 1740 } 1741 1742 /* Start sequence of the skb just after the highest skb with SACKed 1743 * bit, valid only if sacked_out > 0 or when the caller has ensured 1744 * validity by itself. 1745 */ 1746 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1747 { 1748 if (!tp->sacked_out) 1749 return tp->snd_una; 1750 1751 if (tp->highest_sack == NULL) 1752 return tp->snd_nxt; 1753 1754 return TCP_SKB_CB(tp->highest_sack)->seq; 1755 } 1756 1757 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1758 { 1759 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1760 tcp_write_queue_next(sk, skb); 1761 } 1762 1763 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1764 { 1765 return tcp_sk(sk)->highest_sack; 1766 } 1767 1768 static inline void tcp_highest_sack_reset(struct sock *sk) 1769 { 1770 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1771 } 1772 1773 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1774 static inline void tcp_highest_sack_combine(struct sock *sk, 1775 struct sk_buff *old, 1776 struct sk_buff *new) 1777 { 1778 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1779 tcp_sk(sk)->highest_sack = new; 1780 } 1781 1782 /* This helper checks if socket has IP_TRANSPARENT set */ 1783 static inline bool inet_sk_transparent(const struct sock *sk) 1784 { 1785 switch (sk->sk_state) { 1786 case TCP_TIME_WAIT: 1787 return inet_twsk(sk)->tw_transparent; 1788 case TCP_NEW_SYN_RECV: 1789 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1790 } 1791 return inet_sk(sk)->transparent; 1792 } 1793 1794 /* Determines whether this is a thin stream (which may suffer from 1795 * increased latency). Used to trigger latency-reducing mechanisms. 1796 */ 1797 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1798 { 1799 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1800 } 1801 1802 /* /proc */ 1803 enum tcp_seq_states { 1804 TCP_SEQ_STATE_LISTENING, 1805 TCP_SEQ_STATE_ESTABLISHED, 1806 }; 1807 1808 int tcp_seq_open(struct inode *inode, struct file *file); 1809 1810 struct tcp_seq_afinfo { 1811 char *name; 1812 sa_family_t family; 1813 const struct file_operations *seq_fops; 1814 struct seq_operations seq_ops; 1815 }; 1816 1817 struct tcp_iter_state { 1818 struct seq_net_private p; 1819 sa_family_t family; 1820 enum tcp_seq_states state; 1821 struct sock *syn_wait_sk; 1822 int bucket, offset, sbucket, num; 1823 loff_t last_pos; 1824 }; 1825 1826 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1827 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1828 1829 extern struct request_sock_ops tcp_request_sock_ops; 1830 extern struct request_sock_ops tcp6_request_sock_ops; 1831 1832 void tcp_v4_destroy_sock(struct sock *sk); 1833 1834 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1835 netdev_features_t features); 1836 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1837 int tcp_gro_complete(struct sk_buff *skb); 1838 1839 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1840 1841 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1842 { 1843 struct net *net = sock_net((struct sock *)tp); 1844 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1845 } 1846 1847 static inline bool tcp_stream_memory_free(const struct sock *sk) 1848 { 1849 const struct tcp_sock *tp = tcp_sk(sk); 1850 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1851 1852 return notsent_bytes < tcp_notsent_lowat(tp); 1853 } 1854 1855 #ifdef CONFIG_PROC_FS 1856 int tcp4_proc_init(void); 1857 void tcp4_proc_exit(void); 1858 #endif 1859 1860 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1861 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1862 const struct tcp_request_sock_ops *af_ops, 1863 struct sock *sk, struct sk_buff *skb); 1864 1865 /* TCP af-specific functions */ 1866 struct tcp_sock_af_ops { 1867 #ifdef CONFIG_TCP_MD5SIG 1868 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1869 const struct sock *addr_sk); 1870 int (*calc_md5_hash)(char *location, 1871 const struct tcp_md5sig_key *md5, 1872 const struct sock *sk, 1873 const struct sk_buff *skb); 1874 int (*md5_parse)(struct sock *sk, 1875 int optname, 1876 char __user *optval, 1877 int optlen); 1878 #endif 1879 }; 1880 1881 struct tcp_request_sock_ops { 1882 u16 mss_clamp; 1883 #ifdef CONFIG_TCP_MD5SIG 1884 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1885 const struct sock *addr_sk); 1886 int (*calc_md5_hash) (char *location, 1887 const struct tcp_md5sig_key *md5, 1888 const struct sock *sk, 1889 const struct sk_buff *skb); 1890 #endif 1891 void (*init_req)(struct request_sock *req, 1892 const struct sock *sk_listener, 1893 struct sk_buff *skb); 1894 #ifdef CONFIG_SYN_COOKIES 1895 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1896 __u16 *mss); 1897 #endif 1898 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1899 const struct request_sock *req); 1900 u32 (*init_seq)(const struct sk_buff *skb); 1901 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1902 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1903 struct flowi *fl, struct request_sock *req, 1904 struct tcp_fastopen_cookie *foc, 1905 enum tcp_synack_type synack_type); 1906 }; 1907 1908 #ifdef CONFIG_SYN_COOKIES 1909 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1910 const struct sock *sk, struct sk_buff *skb, 1911 __u16 *mss) 1912 { 1913 tcp_synq_overflow(sk); 1914 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1915 return ops->cookie_init_seq(skb, mss); 1916 } 1917 #else 1918 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1919 const struct sock *sk, struct sk_buff *skb, 1920 __u16 *mss) 1921 { 1922 return 0; 1923 } 1924 #endif 1925 1926 int tcpv4_offload_init(void); 1927 1928 void tcp_v4_init(void); 1929 void tcp_init(void); 1930 1931 /* tcp_recovery.c */ 1932 extern void tcp_rack_mark_lost(struct sock *sk); 1933 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1934 u64 xmit_time); 1935 extern void tcp_rack_reo_timeout(struct sock *sk); 1936 1937 /* At how many usecs into the future should the RTO fire? */ 1938 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1939 { 1940 const struct sk_buff *skb = tcp_write_queue_head(sk); 1941 u32 rto = inet_csk(sk)->icsk_rto; 1942 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1943 1944 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1945 } 1946 1947 /* 1948 * Save and compile IPv4 options, return a pointer to it 1949 */ 1950 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1951 struct sk_buff *skb) 1952 { 1953 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1954 struct ip_options_rcu *dopt = NULL; 1955 1956 if (opt->optlen) { 1957 int opt_size = sizeof(*dopt) + opt->optlen; 1958 1959 dopt = kmalloc(opt_size, GFP_ATOMIC); 1960 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1961 kfree(dopt); 1962 dopt = NULL; 1963 } 1964 } 1965 return dopt; 1966 } 1967 1968 /* locally generated TCP pure ACKs have skb->truesize == 2 1969 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1970 * This is much faster than dissecting the packet to find out. 1971 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1972 */ 1973 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1974 { 1975 return skb->truesize == 2; 1976 } 1977 1978 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1979 { 1980 skb->truesize = 2; 1981 } 1982 1983 static inline int tcp_inq(struct sock *sk) 1984 { 1985 struct tcp_sock *tp = tcp_sk(sk); 1986 int answ; 1987 1988 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1989 answ = 0; 1990 } else if (sock_flag(sk, SOCK_URGINLINE) || 1991 !tp->urg_data || 1992 before(tp->urg_seq, tp->copied_seq) || 1993 !before(tp->urg_seq, tp->rcv_nxt)) { 1994 1995 answ = tp->rcv_nxt - tp->copied_seq; 1996 1997 /* Subtract 1, if FIN was received */ 1998 if (answ && sock_flag(sk, SOCK_DONE)) 1999 answ--; 2000 } else { 2001 answ = tp->urg_seq - tp->copied_seq; 2002 } 2003 2004 return answ; 2005 } 2006 2007 int tcp_peek_len(struct socket *sock); 2008 2009 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2010 { 2011 u16 segs_in; 2012 2013 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2014 tp->segs_in += segs_in; 2015 if (skb->len > tcp_hdrlen(skb)) 2016 tp->data_segs_in += segs_in; 2017 } 2018 2019 /* 2020 * TCP listen path runs lockless. 2021 * We forced "struct sock" to be const qualified to make sure 2022 * we don't modify one of its field by mistake. 2023 * Here, we increment sk_drops which is an atomic_t, so we can safely 2024 * make sock writable again. 2025 */ 2026 static inline void tcp_listendrop(const struct sock *sk) 2027 { 2028 atomic_inc(&((struct sock *)sk)->sk_drops); 2029 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2030 } 2031 2032 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2033 2034 /* 2035 * Interface for adding Upper Level Protocols over TCP 2036 */ 2037 2038 #define TCP_ULP_NAME_MAX 16 2039 #define TCP_ULP_MAX 128 2040 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2041 2042 struct tcp_ulp_ops { 2043 struct list_head list; 2044 2045 /* initialize ulp */ 2046 int (*init)(struct sock *sk); 2047 /* cleanup ulp */ 2048 void (*release)(struct sock *sk); 2049 2050 char name[TCP_ULP_NAME_MAX]; 2051 struct module *owner; 2052 }; 2053 int tcp_register_ulp(struct tcp_ulp_ops *type); 2054 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2055 int tcp_set_ulp(struct sock *sk, const char *name); 2056 void tcp_get_available_ulp(char *buf, size_t len); 2057 void tcp_cleanup_ulp(struct sock *sk); 2058 2059 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2060 * is < 0, then the BPF op failed (for example if the loaded BPF 2061 * program does not support the chosen operation or there is no BPF 2062 * program loaded). 2063 */ 2064 #ifdef CONFIG_BPF 2065 static inline int tcp_call_bpf(struct sock *sk, int op) 2066 { 2067 struct bpf_sock_ops_kern sock_ops; 2068 int ret; 2069 2070 if (sk_fullsock(sk)) 2071 sock_owned_by_me(sk); 2072 2073 memset(&sock_ops, 0, sizeof(sock_ops)); 2074 sock_ops.sk = sk; 2075 sock_ops.op = op; 2076 2077 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2078 if (ret == 0) 2079 ret = sock_ops.reply; 2080 else 2081 ret = -1; 2082 return ret; 2083 } 2084 #else 2085 static inline int tcp_call_bpf(struct sock *sk, int op) 2086 { 2087 return -EPERM; 2088 } 2089 #endif 2090 2091 static inline u32 tcp_timeout_init(struct sock *sk) 2092 { 2093 int timeout; 2094 2095 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2096 2097 if (timeout <= 0) 2098 timeout = TCP_TIMEOUT_INIT; 2099 return timeout; 2100 } 2101 2102 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2103 { 2104 int rwnd; 2105 2106 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2107 2108 if (rwnd < 0) 2109 rwnd = 0; 2110 return rwnd; 2111 } 2112 2113 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2114 { 2115 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2116 } 2117 #endif /* _TCP_H */ 2118