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