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_retrans_collapse; 244 extern int sysctl_tcp_stdurg; 245 extern int sysctl_tcp_rfc1337; 246 extern int sysctl_tcp_abort_on_overflow; 247 extern int sysctl_tcp_max_orphans; 248 extern int sysctl_tcp_fack; 249 extern int sysctl_tcp_reordering; 250 extern int sysctl_tcp_max_reordering; 251 extern int sysctl_tcp_dsack; 252 extern long sysctl_tcp_mem[3]; 253 extern int sysctl_tcp_wmem[3]; 254 extern int sysctl_tcp_rmem[3]; 255 extern int sysctl_tcp_app_win; 256 extern int sysctl_tcp_adv_win_scale; 257 extern int sysctl_tcp_frto; 258 extern int sysctl_tcp_nometrics_save; 259 extern int sysctl_tcp_moderate_rcvbuf; 260 extern int sysctl_tcp_tso_win_divisor; 261 extern int sysctl_tcp_workaround_signed_windows; 262 extern int sysctl_tcp_slow_start_after_idle; 263 extern int sysctl_tcp_thin_linear_timeouts; 264 extern int sysctl_tcp_thin_dupack; 265 extern int sysctl_tcp_early_retrans; 266 extern int sysctl_tcp_recovery; 267 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 268 269 extern int sysctl_tcp_limit_output_bytes; 270 extern int sysctl_tcp_challenge_ack_limit; 271 extern int sysctl_tcp_min_tso_segs; 272 extern int sysctl_tcp_min_rtt_wlen; 273 extern int sysctl_tcp_autocorking; 274 extern int sysctl_tcp_invalid_ratelimit; 275 extern int sysctl_tcp_pacing_ss_ratio; 276 extern int sysctl_tcp_pacing_ca_ratio; 277 278 extern atomic_long_t tcp_memory_allocated; 279 extern struct percpu_counter tcp_sockets_allocated; 280 extern unsigned long tcp_memory_pressure; 281 282 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 283 static inline bool tcp_under_memory_pressure(const struct sock *sk) 284 { 285 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 286 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 287 return true; 288 289 return tcp_memory_pressure; 290 } 291 /* 292 * The next routines deal with comparing 32 bit unsigned ints 293 * and worry about wraparound (automatic with unsigned arithmetic). 294 */ 295 296 static inline bool before(__u32 seq1, __u32 seq2) 297 { 298 return (__s32)(seq1-seq2) < 0; 299 } 300 #define after(seq2, seq1) before(seq1, seq2) 301 302 /* is s2<=s1<=s3 ? */ 303 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 304 { 305 return seq3 - seq2 >= seq1 - seq2; 306 } 307 308 static inline bool tcp_out_of_memory(struct sock *sk) 309 { 310 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 311 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 312 return true; 313 return false; 314 } 315 316 void sk_forced_mem_schedule(struct sock *sk, int size); 317 318 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 319 { 320 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 321 int orphans = percpu_counter_read_positive(ocp); 322 323 if (orphans << shift > sysctl_tcp_max_orphans) { 324 orphans = percpu_counter_sum_positive(ocp); 325 if (orphans << shift > sysctl_tcp_max_orphans) 326 return true; 327 } 328 return false; 329 } 330 331 bool tcp_check_oom(struct sock *sk, int shift); 332 333 334 extern struct proto tcp_prot; 335 336 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 337 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 338 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 339 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 340 341 void tcp_tasklet_init(void); 342 343 void tcp_v4_err(struct sk_buff *skb, u32); 344 345 void tcp_shutdown(struct sock *sk, int how); 346 347 int tcp_v4_early_demux(struct sk_buff *skb); 348 int tcp_v4_rcv(struct sk_buff *skb); 349 350 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 351 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 352 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 353 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 354 int flags); 355 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 356 size_t size, 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); 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 void tcp_init_transfer(struct sock *sk, int bpf_op); 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 __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 has_rxtstamp:1, /* SKB has a RX timestamp */ 816 unused:5; 817 __u32 ack_seq; /* Sequence number ACK'd */ 818 union { 819 struct { 820 /* There is space for up to 24 bytes */ 821 __u32 in_flight:30,/* Bytes in flight at transmit */ 822 is_app_limited:1, /* cwnd not fully used? */ 823 unused:1; 824 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 825 __u32 delivered; 826 /* start of send pipeline phase */ 827 u64 first_tx_mstamp; 828 /* when we reached the "delivered" count */ 829 u64 delivered_mstamp; 830 } tx; /* only used for outgoing skbs */ 831 union { 832 struct inet_skb_parm h4; 833 #if IS_ENABLED(CONFIG_IPV6) 834 struct inet6_skb_parm h6; 835 #endif 836 } header; /* For incoming skbs */ 837 }; 838 }; 839 840 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 841 842 843 #if IS_ENABLED(CONFIG_IPV6) 844 /* This is the variant of inet6_iif() that must be used by TCP, 845 * as TCP moves IP6CB into a different location in skb->cb[] 846 */ 847 static inline int tcp_v6_iif(const struct sk_buff *skb) 848 { 849 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 850 851 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 852 } 853 854 /* TCP_SKB_CB reference means this can not be used from early demux */ 855 static inline int tcp_v6_sdif(const struct sk_buff *skb) 856 { 857 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 858 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 859 return TCP_SKB_CB(skb)->header.h6.iif; 860 #endif 861 return 0; 862 } 863 #endif 864 865 /* TCP_SKB_CB reference means this can not be used from early demux */ 866 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 867 { 868 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 869 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 870 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 871 return true; 872 #endif 873 return false; 874 } 875 876 /* TCP_SKB_CB reference means this can not be used from early demux */ 877 static inline int tcp_v4_sdif(struct sk_buff *skb) 878 { 879 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 880 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 881 return TCP_SKB_CB(skb)->header.h4.iif; 882 #endif 883 return 0; 884 } 885 886 /* Due to TSO, an SKB can be composed of multiple actual 887 * packets. To keep these tracked properly, we use this. 888 */ 889 static inline int tcp_skb_pcount(const struct sk_buff *skb) 890 { 891 return TCP_SKB_CB(skb)->tcp_gso_segs; 892 } 893 894 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 895 { 896 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 897 } 898 899 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 900 { 901 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 902 } 903 904 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 905 static inline int tcp_skb_mss(const struct sk_buff *skb) 906 { 907 return TCP_SKB_CB(skb)->tcp_gso_size; 908 } 909 910 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 911 { 912 return likely(!TCP_SKB_CB(skb)->eor); 913 } 914 915 /* Events passed to congestion control interface */ 916 enum tcp_ca_event { 917 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 918 CA_EVENT_CWND_RESTART, /* congestion window restart */ 919 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 920 CA_EVENT_LOSS, /* loss timeout */ 921 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 922 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 923 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 924 CA_EVENT_NON_DELAYED_ACK, 925 }; 926 927 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 928 enum tcp_ca_ack_event_flags { 929 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 930 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 931 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 932 }; 933 934 /* 935 * Interface for adding new TCP congestion control handlers 936 */ 937 #define TCP_CA_NAME_MAX 16 938 #define TCP_CA_MAX 128 939 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 940 941 #define TCP_CA_UNSPEC 0 942 943 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 944 #define TCP_CONG_NON_RESTRICTED 0x1 945 /* Requires ECN/ECT set on all packets */ 946 #define TCP_CONG_NEEDS_ECN 0x2 947 948 union tcp_cc_info; 949 950 struct ack_sample { 951 u32 pkts_acked; 952 s32 rtt_us; 953 u32 in_flight; 954 }; 955 956 /* A rate sample measures the number of (original/retransmitted) data 957 * packets delivered "delivered" over an interval of time "interval_us". 958 * The tcp_rate.c code fills in the rate sample, and congestion 959 * control modules that define a cong_control function to run at the end 960 * of ACK processing can optionally chose to consult this sample when 961 * setting cwnd and pacing rate. 962 * A sample is invalid if "delivered" or "interval_us" is negative. 963 */ 964 struct rate_sample { 965 u64 prior_mstamp; /* starting timestamp for interval */ 966 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 967 s32 delivered; /* number of packets delivered over interval */ 968 long interval_us; /* time for tp->delivered to incr "delivered" */ 969 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 970 int losses; /* number of packets marked lost upon ACK */ 971 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 972 u32 prior_in_flight; /* in flight before this ACK */ 973 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 974 bool is_retrans; /* is sample from retransmission? */ 975 }; 976 977 struct tcp_congestion_ops { 978 struct list_head list; 979 u32 key; 980 u32 flags; 981 982 /* initialize private data (optional) */ 983 void (*init)(struct sock *sk); 984 /* cleanup private data (optional) */ 985 void (*release)(struct sock *sk); 986 987 /* return slow start threshold (required) */ 988 u32 (*ssthresh)(struct sock *sk); 989 /* do new cwnd calculation (required) */ 990 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 991 /* call before changing ca_state (optional) */ 992 void (*set_state)(struct sock *sk, u8 new_state); 993 /* call when cwnd event occurs (optional) */ 994 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 995 /* call when ack arrives (optional) */ 996 void (*in_ack_event)(struct sock *sk, u32 flags); 997 /* new value of cwnd after loss (required) */ 998 u32 (*undo_cwnd)(struct sock *sk); 999 /* hook for packet ack accounting (optional) */ 1000 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1001 /* suggest number of segments for each skb to transmit (optional) */ 1002 u32 (*tso_segs_goal)(struct sock *sk); 1003 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1004 u32 (*sndbuf_expand)(struct sock *sk); 1005 /* call when packets are delivered to update cwnd and pacing rate, 1006 * after all the ca_state processing. (optional) 1007 */ 1008 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1009 /* get info for inet_diag (optional) */ 1010 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1011 union tcp_cc_info *info); 1012 1013 char name[TCP_CA_NAME_MAX]; 1014 struct module *owner; 1015 }; 1016 1017 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1018 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1019 1020 void tcp_assign_congestion_control(struct sock *sk); 1021 void tcp_init_congestion_control(struct sock *sk); 1022 void tcp_cleanup_congestion_control(struct sock *sk); 1023 int tcp_set_default_congestion_control(const char *name); 1024 void tcp_get_default_congestion_control(char *name); 1025 void tcp_get_available_congestion_control(char *buf, size_t len); 1026 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1027 int tcp_set_allowed_congestion_control(char *allowed); 1028 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1029 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1030 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1031 1032 u32 tcp_reno_ssthresh(struct sock *sk); 1033 u32 tcp_reno_undo_cwnd(struct sock *sk); 1034 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1035 extern struct tcp_congestion_ops tcp_reno; 1036 1037 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1038 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1039 #ifdef CONFIG_INET 1040 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1041 #else 1042 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1043 { 1044 return NULL; 1045 } 1046 #endif 1047 1048 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1049 { 1050 const struct inet_connection_sock *icsk = inet_csk(sk); 1051 1052 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1053 } 1054 1055 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1056 { 1057 struct inet_connection_sock *icsk = inet_csk(sk); 1058 1059 if (icsk->icsk_ca_ops->set_state) 1060 icsk->icsk_ca_ops->set_state(sk, ca_state); 1061 icsk->icsk_ca_state = ca_state; 1062 } 1063 1064 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1065 { 1066 const struct inet_connection_sock *icsk = inet_csk(sk); 1067 1068 if (icsk->icsk_ca_ops->cwnd_event) 1069 icsk->icsk_ca_ops->cwnd_event(sk, event); 1070 } 1071 1072 /* From tcp_rate.c */ 1073 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1074 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1075 struct rate_sample *rs); 1076 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1077 struct rate_sample *rs); 1078 void tcp_rate_check_app_limited(struct sock *sk); 1079 1080 /* These functions determine how the current flow behaves in respect of SACK 1081 * handling. SACK is negotiated with the peer, and therefore it can vary 1082 * between different flows. 1083 * 1084 * tcp_is_sack - SACK enabled 1085 * tcp_is_reno - No SACK 1086 * tcp_is_fack - FACK enabled, implies SACK enabled 1087 */ 1088 static inline int tcp_is_sack(const struct tcp_sock *tp) 1089 { 1090 return tp->rx_opt.sack_ok; 1091 } 1092 1093 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1094 { 1095 return !tcp_is_sack(tp); 1096 } 1097 1098 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1099 { 1100 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1101 } 1102 1103 static inline void tcp_enable_fack(struct tcp_sock *tp) 1104 { 1105 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1106 } 1107 1108 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1109 { 1110 return tp->sacked_out + tp->lost_out; 1111 } 1112 1113 /* This determines how many packets are "in the network" to the best 1114 * of our knowledge. In many cases it is conservative, but where 1115 * detailed information is available from the receiver (via SACK 1116 * blocks etc.) we can make more aggressive calculations. 1117 * 1118 * Use this for decisions involving congestion control, use just 1119 * tp->packets_out to determine if the send queue is empty or not. 1120 * 1121 * Read this equation as: 1122 * 1123 * "Packets sent once on transmission queue" MINUS 1124 * "Packets left network, but not honestly ACKed yet" PLUS 1125 * "Packets fast retransmitted" 1126 */ 1127 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1128 { 1129 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1130 } 1131 1132 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1133 1134 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1135 { 1136 return tp->snd_cwnd < tp->snd_ssthresh; 1137 } 1138 1139 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1140 { 1141 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1142 } 1143 1144 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1145 { 1146 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1147 (1 << inet_csk(sk)->icsk_ca_state); 1148 } 1149 1150 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1151 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1152 * ssthresh. 1153 */ 1154 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1155 { 1156 const struct tcp_sock *tp = tcp_sk(sk); 1157 1158 if (tcp_in_cwnd_reduction(sk)) 1159 return tp->snd_ssthresh; 1160 else 1161 return max(tp->snd_ssthresh, 1162 ((tp->snd_cwnd >> 1) + 1163 (tp->snd_cwnd >> 2))); 1164 } 1165 1166 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1167 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1168 1169 void tcp_enter_cwr(struct sock *sk); 1170 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1171 1172 /* The maximum number of MSS of available cwnd for which TSO defers 1173 * sending if not using sysctl_tcp_tso_win_divisor. 1174 */ 1175 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1176 { 1177 return 3; 1178 } 1179 1180 /* Returns end sequence number of the receiver's advertised window */ 1181 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1182 { 1183 return tp->snd_una + tp->snd_wnd; 1184 } 1185 1186 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1187 * flexible approach. The RFC suggests cwnd should not be raised unless 1188 * it was fully used previously. And that's exactly what we do in 1189 * congestion avoidance mode. But in slow start we allow cwnd to grow 1190 * as long as the application has used half the cwnd. 1191 * Example : 1192 * cwnd is 10 (IW10), but application sends 9 frames. 1193 * We allow cwnd to reach 18 when all frames are ACKed. 1194 * This check is safe because it's as aggressive as slow start which already 1195 * risks 100% overshoot. The advantage is that we discourage application to 1196 * either send more filler packets or data to artificially blow up the cwnd 1197 * usage, and allow application-limited process to probe bw more aggressively. 1198 */ 1199 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1200 { 1201 const struct tcp_sock *tp = tcp_sk(sk); 1202 1203 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1204 if (tcp_in_slow_start(tp)) 1205 return tp->snd_cwnd < 2 * tp->max_packets_out; 1206 1207 return tp->is_cwnd_limited; 1208 } 1209 1210 /* Something is really bad, we could not queue an additional packet, 1211 * because qdisc is full or receiver sent a 0 window. 1212 * We do not want to add fuel to the fire, or abort too early, 1213 * so make sure the timer we arm now is at least 200ms in the future, 1214 * regardless of current icsk_rto value (as it could be ~2ms) 1215 */ 1216 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1217 { 1218 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1219 } 1220 1221 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1222 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1223 unsigned long max_when) 1224 { 1225 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1226 1227 return (unsigned long)min_t(u64, when, max_when); 1228 } 1229 1230 static inline void tcp_check_probe_timer(struct sock *sk) 1231 { 1232 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1233 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1234 tcp_probe0_base(sk), TCP_RTO_MAX); 1235 } 1236 1237 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1238 { 1239 tp->snd_wl1 = seq; 1240 } 1241 1242 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1243 { 1244 tp->snd_wl1 = seq; 1245 } 1246 1247 /* 1248 * Calculate(/check) TCP checksum 1249 */ 1250 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1251 __be32 daddr, __wsum base) 1252 { 1253 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1254 } 1255 1256 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1257 { 1258 return __skb_checksum_complete(skb); 1259 } 1260 1261 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1262 { 1263 return !skb_csum_unnecessary(skb) && 1264 __tcp_checksum_complete(skb); 1265 } 1266 1267 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1268 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1269 1270 #undef STATE_TRACE 1271 1272 #ifdef STATE_TRACE 1273 static const char *statename[]={ 1274 "Unused","Established","Syn Sent","Syn Recv", 1275 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1276 "Close Wait","Last ACK","Listen","Closing" 1277 }; 1278 #endif 1279 void tcp_set_state(struct sock *sk, int state); 1280 1281 void tcp_done(struct sock *sk); 1282 1283 int tcp_abort(struct sock *sk, int err); 1284 1285 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1286 { 1287 rx_opt->dsack = 0; 1288 rx_opt->num_sacks = 0; 1289 } 1290 1291 u32 tcp_default_init_rwnd(u32 mss); 1292 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1293 1294 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1295 { 1296 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1297 struct tcp_sock *tp = tcp_sk(sk); 1298 s32 delta; 1299 1300 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out || 1301 ca_ops->cong_control) 1302 return; 1303 delta = tcp_jiffies32 - tp->lsndtime; 1304 if (delta > inet_csk(sk)->icsk_rto) 1305 tcp_cwnd_restart(sk, delta); 1306 } 1307 1308 /* Determine a window scaling and initial window to offer. */ 1309 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1310 __u32 *window_clamp, int wscale_ok, 1311 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1312 1313 static inline int tcp_win_from_space(int space) 1314 { 1315 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale; 1316 1317 return tcp_adv_win_scale <= 0 ? 1318 (space>>(-tcp_adv_win_scale)) : 1319 space - (space>>tcp_adv_win_scale); 1320 } 1321 1322 /* Note: caller must be prepared to deal with negative returns */ 1323 static inline int tcp_space(const struct sock *sk) 1324 { 1325 return tcp_win_from_space(sk->sk_rcvbuf - 1326 atomic_read(&sk->sk_rmem_alloc)); 1327 } 1328 1329 static inline int tcp_full_space(const struct sock *sk) 1330 { 1331 return tcp_win_from_space(sk->sk_rcvbuf); 1332 } 1333 1334 extern void tcp_openreq_init_rwin(struct request_sock *req, 1335 const struct sock *sk_listener, 1336 const struct dst_entry *dst); 1337 1338 void tcp_enter_memory_pressure(struct sock *sk); 1339 void tcp_leave_memory_pressure(struct sock *sk); 1340 1341 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1342 { 1343 struct net *net = sock_net((struct sock *)tp); 1344 1345 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1346 } 1347 1348 static inline int keepalive_time_when(const struct tcp_sock *tp) 1349 { 1350 struct net *net = sock_net((struct sock *)tp); 1351 1352 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1353 } 1354 1355 static inline int keepalive_probes(const struct tcp_sock *tp) 1356 { 1357 struct net *net = sock_net((struct sock *)tp); 1358 1359 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1360 } 1361 1362 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1363 { 1364 const struct inet_connection_sock *icsk = &tp->inet_conn; 1365 1366 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1367 tcp_jiffies32 - tp->rcv_tstamp); 1368 } 1369 1370 static inline int tcp_fin_time(const struct sock *sk) 1371 { 1372 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1373 const int rto = inet_csk(sk)->icsk_rto; 1374 1375 if (fin_timeout < (rto << 2) - (rto >> 1)) 1376 fin_timeout = (rto << 2) - (rto >> 1); 1377 1378 return fin_timeout; 1379 } 1380 1381 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1382 int paws_win) 1383 { 1384 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1385 return true; 1386 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1387 return true; 1388 /* 1389 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1390 * then following tcp messages have valid values. Ignore 0 value, 1391 * or else 'negative' tsval might forbid us to accept their packets. 1392 */ 1393 if (!rx_opt->ts_recent) 1394 return true; 1395 return false; 1396 } 1397 1398 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1399 int rst) 1400 { 1401 if (tcp_paws_check(rx_opt, 0)) 1402 return false; 1403 1404 /* RST segments are not recommended to carry timestamp, 1405 and, if they do, it is recommended to ignore PAWS because 1406 "their cleanup function should take precedence over timestamps." 1407 Certainly, it is mistake. It is necessary to understand the reasons 1408 of this constraint to relax it: if peer reboots, clock may go 1409 out-of-sync and half-open connections will not be reset. 1410 Actually, the problem would be not existing if all 1411 the implementations followed draft about maintaining clock 1412 via reboots. Linux-2.2 DOES NOT! 1413 1414 However, we can relax time bounds for RST segments to MSL. 1415 */ 1416 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1417 return false; 1418 return true; 1419 } 1420 1421 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1422 int mib_idx, u32 *last_oow_ack_time); 1423 1424 static inline void tcp_mib_init(struct net *net) 1425 { 1426 /* See RFC 2012 */ 1427 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1428 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1429 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1430 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1431 } 1432 1433 /* from STCP */ 1434 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1435 { 1436 tp->lost_skb_hint = NULL; 1437 } 1438 1439 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1440 { 1441 tcp_clear_retrans_hints_partial(tp); 1442 tp->retransmit_skb_hint = NULL; 1443 } 1444 1445 union tcp_md5_addr { 1446 struct in_addr a4; 1447 #if IS_ENABLED(CONFIG_IPV6) 1448 struct in6_addr a6; 1449 #endif 1450 }; 1451 1452 /* - key database */ 1453 struct tcp_md5sig_key { 1454 struct hlist_node node; 1455 u8 keylen; 1456 u8 family; /* AF_INET or AF_INET6 */ 1457 union tcp_md5_addr addr; 1458 u8 prefixlen; 1459 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1460 struct rcu_head rcu; 1461 }; 1462 1463 /* - sock block */ 1464 struct tcp_md5sig_info { 1465 struct hlist_head head; 1466 struct rcu_head rcu; 1467 }; 1468 1469 /* - pseudo header */ 1470 struct tcp4_pseudohdr { 1471 __be32 saddr; 1472 __be32 daddr; 1473 __u8 pad; 1474 __u8 protocol; 1475 __be16 len; 1476 }; 1477 1478 struct tcp6_pseudohdr { 1479 struct in6_addr saddr; 1480 struct in6_addr daddr; 1481 __be32 len; 1482 __be32 protocol; /* including padding */ 1483 }; 1484 1485 union tcp_md5sum_block { 1486 struct tcp4_pseudohdr ip4; 1487 #if IS_ENABLED(CONFIG_IPV6) 1488 struct tcp6_pseudohdr ip6; 1489 #endif 1490 }; 1491 1492 /* - pool: digest algorithm, hash description and scratch buffer */ 1493 struct tcp_md5sig_pool { 1494 struct ahash_request *md5_req; 1495 void *scratch; 1496 }; 1497 1498 /* - functions */ 1499 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1500 const struct sock *sk, const struct sk_buff *skb); 1501 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1502 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1503 gfp_t gfp); 1504 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1505 int family, u8 prefixlen); 1506 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1507 const struct sock *addr_sk); 1508 1509 #ifdef CONFIG_TCP_MD5SIG 1510 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1511 const union tcp_md5_addr *addr, 1512 int family); 1513 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1514 #else 1515 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1516 const union tcp_md5_addr *addr, 1517 int family) 1518 { 1519 return NULL; 1520 } 1521 #define tcp_twsk_md5_key(twsk) NULL 1522 #endif 1523 1524 bool tcp_alloc_md5sig_pool(void); 1525 1526 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1527 static inline void tcp_put_md5sig_pool(void) 1528 { 1529 local_bh_enable(); 1530 } 1531 1532 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1533 unsigned int header_len); 1534 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1535 const struct tcp_md5sig_key *key); 1536 1537 /* From tcp_fastopen.c */ 1538 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1539 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1540 unsigned long *last_syn_loss); 1541 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1542 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1543 u16 try_exp); 1544 struct tcp_fastopen_request { 1545 /* Fast Open cookie. Size 0 means a cookie request */ 1546 struct tcp_fastopen_cookie cookie; 1547 struct msghdr *data; /* data in MSG_FASTOPEN */ 1548 size_t size; 1549 int copied; /* queued in tcp_connect() */ 1550 }; 1551 void tcp_free_fastopen_req(struct tcp_sock *tp); 1552 1553 void tcp_fastopen_ctx_destroy(struct net *net); 1554 int tcp_fastopen_reset_cipher(struct net *net, void *key, unsigned int len); 1555 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1556 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1557 struct request_sock *req, 1558 struct tcp_fastopen_cookie *foc); 1559 void tcp_fastopen_init_key_once(struct net *net); 1560 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1561 struct tcp_fastopen_cookie *cookie); 1562 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1563 #define TCP_FASTOPEN_KEY_LENGTH 16 1564 1565 /* Fastopen key context */ 1566 struct tcp_fastopen_context { 1567 struct crypto_cipher *tfm; 1568 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1569 struct rcu_head rcu; 1570 }; 1571 1572 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1573 void tcp_fastopen_active_disable(struct sock *sk); 1574 bool tcp_fastopen_active_should_disable(struct sock *sk); 1575 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1576 void tcp_fastopen_active_timeout_reset(void); 1577 1578 /* Latencies incurred by various limits for a sender. They are 1579 * chronograph-like stats that are mutually exclusive. 1580 */ 1581 enum tcp_chrono { 1582 TCP_CHRONO_UNSPEC, 1583 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1584 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1585 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1586 __TCP_CHRONO_MAX, 1587 }; 1588 1589 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1590 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1591 1592 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1593 * the same memory storage than skb->destructor/_skb_refdst 1594 */ 1595 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1596 { 1597 skb->destructor = NULL; 1598 skb->_skb_refdst = 0UL; 1599 } 1600 1601 #define tcp_skb_tsorted_save(skb) { \ 1602 unsigned long _save = skb->_skb_refdst; \ 1603 skb->_skb_refdst = 0UL; 1604 1605 #define tcp_skb_tsorted_restore(skb) \ 1606 skb->_skb_refdst = _save; \ 1607 } 1608 1609 /* write queue abstraction */ 1610 static inline void tcp_write_queue_purge(struct sock *sk) 1611 { 1612 struct sk_buff *skb; 1613 1614 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1615 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { 1616 tcp_skb_tsorted_anchor_cleanup(skb); 1617 sk_wmem_free_skb(sk, skb); 1618 } 1619 INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue); 1620 sk_mem_reclaim(sk); 1621 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1622 } 1623 1624 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1625 { 1626 return skb_peek(&sk->sk_write_queue); 1627 } 1628 1629 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1630 { 1631 return skb_peek_tail(&sk->sk_write_queue); 1632 } 1633 1634 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1635 const struct sk_buff *skb) 1636 { 1637 return skb_queue_next(&sk->sk_write_queue, skb); 1638 } 1639 1640 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1641 const struct sk_buff *skb) 1642 { 1643 return skb_queue_prev(&sk->sk_write_queue, skb); 1644 } 1645 1646 #define tcp_for_write_queue(skb, sk) \ 1647 skb_queue_walk(&(sk)->sk_write_queue, skb) 1648 1649 #define tcp_for_write_queue_from(skb, sk) \ 1650 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1651 1652 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1653 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1654 1655 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1656 { 1657 return sk->sk_send_head; 1658 } 1659 1660 static inline bool tcp_skb_is_last(const struct sock *sk, 1661 const struct sk_buff *skb) 1662 { 1663 return skb_queue_is_last(&sk->sk_write_queue, skb); 1664 } 1665 1666 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1667 { 1668 if (tcp_skb_is_last(sk, skb)) 1669 sk->sk_send_head = NULL; 1670 else 1671 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1672 } 1673 1674 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1675 { 1676 if (sk->sk_send_head == skb_unlinked) { 1677 sk->sk_send_head = NULL; 1678 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1679 } 1680 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1681 tcp_sk(sk)->highest_sack = NULL; 1682 } 1683 1684 static inline void tcp_init_send_head(struct sock *sk) 1685 { 1686 sk->sk_send_head = NULL; 1687 } 1688 1689 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1690 { 1691 __skb_queue_tail(&sk->sk_write_queue, skb); 1692 } 1693 1694 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1695 { 1696 __tcp_add_write_queue_tail(sk, skb); 1697 1698 /* Queue it, remembering where we must start sending. */ 1699 if (sk->sk_send_head == NULL) { 1700 sk->sk_send_head = skb; 1701 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1702 1703 if (tcp_sk(sk)->highest_sack == NULL) 1704 tcp_sk(sk)->highest_sack = skb; 1705 } 1706 } 1707 1708 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1709 { 1710 __skb_queue_head(&sk->sk_write_queue, skb); 1711 } 1712 1713 /* Insert buff after skb on the write queue of sk. */ 1714 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1715 struct sk_buff *buff, 1716 struct sock *sk) 1717 { 1718 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1719 } 1720 1721 /* Insert new before skb on the write queue of sk. */ 1722 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1723 struct sk_buff *skb, 1724 struct sock *sk) 1725 { 1726 __skb_queue_before(&sk->sk_write_queue, skb, new); 1727 1728 if (sk->sk_send_head == skb) 1729 sk->sk_send_head = new; 1730 } 1731 1732 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1733 { 1734 list_del(&skb->tcp_tsorted_anchor); 1735 tcp_skb_tsorted_anchor_cleanup(skb); 1736 __skb_unlink(skb, &sk->sk_write_queue); 1737 } 1738 1739 static inline bool tcp_write_queue_empty(struct sock *sk) 1740 { 1741 return skb_queue_empty(&sk->sk_write_queue); 1742 } 1743 1744 static inline void tcp_push_pending_frames(struct sock *sk) 1745 { 1746 if (tcp_send_head(sk)) { 1747 struct tcp_sock *tp = tcp_sk(sk); 1748 1749 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1750 } 1751 } 1752 1753 /* Start sequence of the skb just after the highest skb with SACKed 1754 * bit, valid only if sacked_out > 0 or when the caller has ensured 1755 * validity by itself. 1756 */ 1757 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1758 { 1759 if (!tp->sacked_out) 1760 return tp->snd_una; 1761 1762 if (tp->highest_sack == NULL) 1763 return tp->snd_nxt; 1764 1765 return TCP_SKB_CB(tp->highest_sack)->seq; 1766 } 1767 1768 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1769 { 1770 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1771 tcp_write_queue_next(sk, skb); 1772 } 1773 1774 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1775 { 1776 return tcp_sk(sk)->highest_sack; 1777 } 1778 1779 static inline void tcp_highest_sack_reset(struct sock *sk) 1780 { 1781 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1782 } 1783 1784 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1785 static inline void tcp_highest_sack_combine(struct sock *sk, 1786 struct sk_buff *old, 1787 struct sk_buff *new) 1788 { 1789 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1790 tcp_sk(sk)->highest_sack = new; 1791 } 1792 1793 /* This helper checks if socket has IP_TRANSPARENT set */ 1794 static inline bool inet_sk_transparent(const struct sock *sk) 1795 { 1796 switch (sk->sk_state) { 1797 case TCP_TIME_WAIT: 1798 return inet_twsk(sk)->tw_transparent; 1799 case TCP_NEW_SYN_RECV: 1800 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1801 } 1802 return inet_sk(sk)->transparent; 1803 } 1804 1805 /* Determines whether this is a thin stream (which may suffer from 1806 * increased latency). Used to trigger latency-reducing mechanisms. 1807 */ 1808 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1809 { 1810 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1811 } 1812 1813 /* /proc */ 1814 enum tcp_seq_states { 1815 TCP_SEQ_STATE_LISTENING, 1816 TCP_SEQ_STATE_ESTABLISHED, 1817 }; 1818 1819 int tcp_seq_open(struct inode *inode, struct file *file); 1820 1821 struct tcp_seq_afinfo { 1822 char *name; 1823 sa_family_t family; 1824 const struct file_operations *seq_fops; 1825 struct seq_operations seq_ops; 1826 }; 1827 1828 struct tcp_iter_state { 1829 struct seq_net_private p; 1830 sa_family_t family; 1831 enum tcp_seq_states state; 1832 struct sock *syn_wait_sk; 1833 int bucket, offset, sbucket, num; 1834 loff_t last_pos; 1835 }; 1836 1837 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1838 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1839 1840 extern struct request_sock_ops tcp_request_sock_ops; 1841 extern struct request_sock_ops tcp6_request_sock_ops; 1842 1843 void tcp_v4_destroy_sock(struct sock *sk); 1844 1845 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1846 netdev_features_t features); 1847 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1848 int tcp_gro_complete(struct sk_buff *skb); 1849 1850 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1851 1852 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1853 { 1854 struct net *net = sock_net((struct sock *)tp); 1855 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1856 } 1857 1858 static inline bool tcp_stream_memory_free(const struct sock *sk) 1859 { 1860 const struct tcp_sock *tp = tcp_sk(sk); 1861 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1862 1863 return notsent_bytes < tcp_notsent_lowat(tp); 1864 } 1865 1866 #ifdef CONFIG_PROC_FS 1867 int tcp4_proc_init(void); 1868 void tcp4_proc_exit(void); 1869 #endif 1870 1871 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1872 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1873 const struct tcp_request_sock_ops *af_ops, 1874 struct sock *sk, struct sk_buff *skb); 1875 1876 /* TCP af-specific functions */ 1877 struct tcp_sock_af_ops { 1878 #ifdef CONFIG_TCP_MD5SIG 1879 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1880 const struct sock *addr_sk); 1881 int (*calc_md5_hash)(char *location, 1882 const struct tcp_md5sig_key *md5, 1883 const struct sock *sk, 1884 const struct sk_buff *skb); 1885 int (*md5_parse)(struct sock *sk, 1886 int optname, 1887 char __user *optval, 1888 int optlen); 1889 #endif 1890 }; 1891 1892 struct tcp_request_sock_ops { 1893 u16 mss_clamp; 1894 #ifdef CONFIG_TCP_MD5SIG 1895 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1896 const struct sock *addr_sk); 1897 int (*calc_md5_hash) (char *location, 1898 const struct tcp_md5sig_key *md5, 1899 const struct sock *sk, 1900 const struct sk_buff *skb); 1901 #endif 1902 void (*init_req)(struct request_sock *req, 1903 const struct sock *sk_listener, 1904 struct sk_buff *skb); 1905 #ifdef CONFIG_SYN_COOKIES 1906 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1907 __u16 *mss); 1908 #endif 1909 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1910 const struct request_sock *req); 1911 u32 (*init_seq)(const struct sk_buff *skb); 1912 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1913 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1914 struct flowi *fl, struct request_sock *req, 1915 struct tcp_fastopen_cookie *foc, 1916 enum tcp_synack_type synack_type); 1917 }; 1918 1919 #ifdef CONFIG_SYN_COOKIES 1920 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1921 const struct sock *sk, struct sk_buff *skb, 1922 __u16 *mss) 1923 { 1924 tcp_synq_overflow(sk); 1925 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1926 return ops->cookie_init_seq(skb, mss); 1927 } 1928 #else 1929 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1930 const struct sock *sk, struct sk_buff *skb, 1931 __u16 *mss) 1932 { 1933 return 0; 1934 } 1935 #endif 1936 1937 int tcpv4_offload_init(void); 1938 1939 void tcp_v4_init(void); 1940 void tcp_init(void); 1941 1942 /* tcp_recovery.c */ 1943 extern void tcp_rack_mark_lost(struct sock *sk); 1944 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1945 u64 xmit_time); 1946 extern void tcp_rack_reo_timeout(struct sock *sk); 1947 1948 /* At how many usecs into the future should the RTO fire? */ 1949 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1950 { 1951 const struct sk_buff *skb = tcp_write_queue_head(sk); 1952 u32 rto = inet_csk(sk)->icsk_rto; 1953 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1954 1955 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1956 } 1957 1958 /* 1959 * Save and compile IPv4 options, return a pointer to it 1960 */ 1961 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1962 struct sk_buff *skb) 1963 { 1964 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1965 struct ip_options_rcu *dopt = NULL; 1966 1967 if (opt->optlen) { 1968 int opt_size = sizeof(*dopt) + opt->optlen; 1969 1970 dopt = kmalloc(opt_size, GFP_ATOMIC); 1971 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1972 kfree(dopt); 1973 dopt = NULL; 1974 } 1975 } 1976 return dopt; 1977 } 1978 1979 /* locally generated TCP pure ACKs have skb->truesize == 2 1980 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1981 * This is much faster than dissecting the packet to find out. 1982 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1983 */ 1984 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1985 { 1986 return skb->truesize == 2; 1987 } 1988 1989 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1990 { 1991 skb->truesize = 2; 1992 } 1993 1994 static inline int tcp_inq(struct sock *sk) 1995 { 1996 struct tcp_sock *tp = tcp_sk(sk); 1997 int answ; 1998 1999 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2000 answ = 0; 2001 } else if (sock_flag(sk, SOCK_URGINLINE) || 2002 !tp->urg_data || 2003 before(tp->urg_seq, tp->copied_seq) || 2004 !before(tp->urg_seq, tp->rcv_nxt)) { 2005 2006 answ = tp->rcv_nxt - tp->copied_seq; 2007 2008 /* Subtract 1, if FIN was received */ 2009 if (answ && sock_flag(sk, SOCK_DONE)) 2010 answ--; 2011 } else { 2012 answ = tp->urg_seq - tp->copied_seq; 2013 } 2014 2015 return answ; 2016 } 2017 2018 int tcp_peek_len(struct socket *sock); 2019 2020 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2021 { 2022 u16 segs_in; 2023 2024 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2025 tp->segs_in += segs_in; 2026 if (skb->len > tcp_hdrlen(skb)) 2027 tp->data_segs_in += segs_in; 2028 } 2029 2030 /* 2031 * TCP listen path runs lockless. 2032 * We forced "struct sock" to be const qualified to make sure 2033 * we don't modify one of its field by mistake. 2034 * Here, we increment sk_drops which is an atomic_t, so we can safely 2035 * make sock writable again. 2036 */ 2037 static inline void tcp_listendrop(const struct sock *sk) 2038 { 2039 atomic_inc(&((struct sock *)sk)->sk_drops); 2040 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2041 } 2042 2043 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2044 2045 /* 2046 * Interface for adding Upper Level Protocols over TCP 2047 */ 2048 2049 #define TCP_ULP_NAME_MAX 16 2050 #define TCP_ULP_MAX 128 2051 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2052 2053 struct tcp_ulp_ops { 2054 struct list_head list; 2055 2056 /* initialize ulp */ 2057 int (*init)(struct sock *sk); 2058 /* cleanup ulp */ 2059 void (*release)(struct sock *sk); 2060 2061 char name[TCP_ULP_NAME_MAX]; 2062 struct module *owner; 2063 }; 2064 int tcp_register_ulp(struct tcp_ulp_ops *type); 2065 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2066 int tcp_set_ulp(struct sock *sk, const char *name); 2067 void tcp_get_available_ulp(char *buf, size_t len); 2068 void tcp_cleanup_ulp(struct sock *sk); 2069 2070 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2071 * is < 0, then the BPF op failed (for example if the loaded BPF 2072 * program does not support the chosen operation or there is no BPF 2073 * program loaded). 2074 */ 2075 #ifdef CONFIG_BPF 2076 static inline int tcp_call_bpf(struct sock *sk, int op) 2077 { 2078 struct bpf_sock_ops_kern sock_ops; 2079 int ret; 2080 2081 if (sk_fullsock(sk)) 2082 sock_owned_by_me(sk); 2083 2084 memset(&sock_ops, 0, sizeof(sock_ops)); 2085 sock_ops.sk = sk; 2086 sock_ops.op = op; 2087 2088 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2089 if (ret == 0) 2090 ret = sock_ops.reply; 2091 else 2092 ret = -1; 2093 return ret; 2094 } 2095 #else 2096 static inline int tcp_call_bpf(struct sock *sk, int op) 2097 { 2098 return -EPERM; 2099 } 2100 #endif 2101 2102 static inline u32 tcp_timeout_init(struct sock *sk) 2103 { 2104 int timeout; 2105 2106 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2107 2108 if (timeout <= 0) 2109 timeout = TCP_TIMEOUT_INIT; 2110 return timeout; 2111 } 2112 2113 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2114 { 2115 int rwnd; 2116 2117 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2118 2119 if (rwnd < 0) 2120 rwnd = 0; 2121 return rwnd; 2122 } 2123 2124 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2125 { 2126 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2127 } 2128 #endif /* _TCP_H */ 2129