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 enum tcp_queue { 555 TCP_FRAG_IN_WRITE_QUEUE, 556 TCP_FRAG_IN_RTX_QUEUE, 557 }; 558 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 559 struct sk_buff *skb, u32 len, 560 unsigned int mss_now, gfp_t gfp); 561 562 void tcp_send_probe0(struct sock *); 563 void tcp_send_partial(struct sock *); 564 int tcp_write_wakeup(struct sock *, int mib); 565 void tcp_send_fin(struct sock *sk); 566 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 567 int tcp_send_synack(struct sock *); 568 void tcp_push_one(struct sock *, unsigned int mss_now); 569 void tcp_send_ack(struct sock *sk); 570 void tcp_send_delayed_ack(struct sock *sk); 571 void tcp_send_loss_probe(struct sock *sk); 572 bool tcp_schedule_loss_probe(struct sock *sk); 573 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 574 const struct sk_buff *next_skb); 575 576 /* tcp_input.c */ 577 void tcp_rearm_rto(struct sock *sk); 578 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 579 void tcp_reset(struct sock *sk); 580 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 581 void tcp_fin(struct sock *sk); 582 583 /* tcp_timer.c */ 584 void tcp_init_xmit_timers(struct sock *); 585 static inline void tcp_clear_xmit_timers(struct sock *sk) 586 { 587 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 588 inet_csk_clear_xmit_timers(sk); 589 } 590 591 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 592 unsigned int tcp_current_mss(struct sock *sk); 593 594 /* Bound MSS / TSO packet size with the half of the window */ 595 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 596 { 597 int cutoff; 598 599 /* When peer uses tiny windows, there is no use in packetizing 600 * to sub-MSS pieces for the sake of SWS or making sure there 601 * are enough packets in the pipe for fast recovery. 602 * 603 * On the other hand, for extremely large MSS devices, handling 604 * smaller than MSS windows in this way does make sense. 605 */ 606 if (tp->max_window > TCP_MSS_DEFAULT) 607 cutoff = (tp->max_window >> 1); 608 else 609 cutoff = tp->max_window; 610 611 if (cutoff && pktsize > cutoff) 612 return max_t(int, cutoff, 68U - tp->tcp_header_len); 613 else 614 return pktsize; 615 } 616 617 /* tcp.c */ 618 void tcp_get_info(struct sock *, struct tcp_info *); 619 620 /* Read 'sendfile()'-style from a TCP socket */ 621 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 622 sk_read_actor_t recv_actor); 623 624 void tcp_initialize_rcv_mss(struct sock *sk); 625 626 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 627 int tcp_mss_to_mtu(struct sock *sk, int mss); 628 void tcp_mtup_init(struct sock *sk); 629 void tcp_init_buffer_space(struct sock *sk); 630 631 static inline void tcp_bound_rto(const struct sock *sk) 632 { 633 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 634 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 635 } 636 637 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 638 { 639 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 640 } 641 642 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 643 { 644 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 645 ntohl(TCP_FLAG_ACK) | 646 snd_wnd); 647 } 648 649 static inline void tcp_fast_path_on(struct tcp_sock *tp) 650 { 651 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 652 } 653 654 static inline void tcp_fast_path_check(struct sock *sk) 655 { 656 struct tcp_sock *tp = tcp_sk(sk); 657 658 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 659 tp->rcv_wnd && 660 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 661 !tp->urg_data) 662 tcp_fast_path_on(tp); 663 } 664 665 /* Compute the actual rto_min value */ 666 static inline u32 tcp_rto_min(struct sock *sk) 667 { 668 const struct dst_entry *dst = __sk_dst_get(sk); 669 u32 rto_min = TCP_RTO_MIN; 670 671 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 672 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 673 return rto_min; 674 } 675 676 static inline u32 tcp_rto_min_us(struct sock *sk) 677 { 678 return jiffies_to_usecs(tcp_rto_min(sk)); 679 } 680 681 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 682 { 683 return dst_metric_locked(dst, RTAX_CC_ALGO); 684 } 685 686 /* Minimum RTT in usec. ~0 means not available. */ 687 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 688 { 689 return minmax_get(&tp->rtt_min); 690 } 691 692 /* Compute the actual receive window we are currently advertising. 693 * Rcv_nxt can be after the window if our peer push more data 694 * than the offered window. 695 */ 696 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 697 { 698 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 699 700 if (win < 0) 701 win = 0; 702 return (u32) win; 703 } 704 705 /* Choose a new window, without checks for shrinking, and without 706 * scaling applied to the result. The caller does these things 707 * if necessary. This is a "raw" window selection. 708 */ 709 u32 __tcp_select_window(struct sock *sk); 710 711 void tcp_send_window_probe(struct sock *sk); 712 713 /* TCP uses 32bit jiffies to save some space. 714 * Note that this is different from tcp_time_stamp, which 715 * historically has been the same until linux-4.13. 716 */ 717 #define tcp_jiffies32 ((u32)jiffies) 718 719 /* 720 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 721 * It is no longer tied to jiffies, but to 1 ms clock. 722 * Note: double check if you want to use tcp_jiffies32 instead of this. 723 */ 724 #define TCP_TS_HZ 1000 725 726 static inline u64 tcp_clock_ns(void) 727 { 728 return local_clock(); 729 } 730 731 static inline u64 tcp_clock_us(void) 732 { 733 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 734 } 735 736 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 737 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 738 { 739 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 740 } 741 742 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 743 static inline u32 tcp_time_stamp_raw(void) 744 { 745 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 746 } 747 748 749 /* Refresh 1us clock of a TCP socket, 750 * ensuring monotically increasing values. 751 */ 752 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 753 { 754 u64 val = tcp_clock_us(); 755 756 if (val > tp->tcp_mstamp) 757 tp->tcp_mstamp = val; 758 } 759 760 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 761 { 762 return max_t(s64, t1 - t0, 0); 763 } 764 765 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 766 { 767 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 768 } 769 770 771 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 772 773 #define TCPHDR_FIN 0x01 774 #define TCPHDR_SYN 0x02 775 #define TCPHDR_RST 0x04 776 #define TCPHDR_PSH 0x08 777 #define TCPHDR_ACK 0x10 778 #define TCPHDR_URG 0x20 779 #define TCPHDR_ECE 0x40 780 #define TCPHDR_CWR 0x80 781 782 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 783 784 /* This is what the send packet queuing engine uses to pass 785 * TCP per-packet control information to the transmission code. 786 * We also store the host-order sequence numbers in here too. 787 * This is 44 bytes if IPV6 is enabled. 788 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 789 */ 790 struct tcp_skb_cb { 791 __u32 seq; /* Starting sequence number */ 792 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 793 union { 794 /* Note : tcp_tw_isn is used in input path only 795 * (isn chosen by tcp_timewait_state_process()) 796 * 797 * tcp_gso_segs/size are used in write queue only, 798 * cf tcp_skb_pcount()/tcp_skb_mss() 799 */ 800 __u32 tcp_tw_isn; 801 struct { 802 u16 tcp_gso_segs; 803 u16 tcp_gso_size; 804 }; 805 }; 806 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 807 808 __u8 sacked; /* State flags for SACK/FACK. */ 809 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 810 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 811 #define TCPCB_LOST 0x04 /* SKB is lost */ 812 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 813 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 814 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 815 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 816 TCPCB_REPAIRED) 817 818 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 819 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 820 eor:1, /* Is skb MSG_EOR marked? */ 821 has_rxtstamp:1, /* SKB has a RX timestamp */ 822 unused:5; 823 __u32 ack_seq; /* Sequence number ACK'd */ 824 union { 825 struct { 826 /* There is space for up to 24 bytes */ 827 __u32 in_flight:30,/* Bytes in flight at transmit */ 828 is_app_limited:1, /* cwnd not fully used? */ 829 unused:1; 830 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 831 __u32 delivered; 832 /* start of send pipeline phase */ 833 u64 first_tx_mstamp; 834 /* when we reached the "delivered" count */ 835 u64 delivered_mstamp; 836 } tx; /* only used for outgoing skbs */ 837 union { 838 struct inet_skb_parm h4; 839 #if IS_ENABLED(CONFIG_IPV6) 840 struct inet6_skb_parm h6; 841 #endif 842 } header; /* For incoming skbs */ 843 }; 844 }; 845 846 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 847 848 849 #if IS_ENABLED(CONFIG_IPV6) 850 /* This is the variant of inet6_iif() that must be used by TCP, 851 * as TCP moves IP6CB into a different location in skb->cb[] 852 */ 853 static inline int tcp_v6_iif(const struct sk_buff *skb) 854 { 855 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 856 857 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 858 } 859 860 /* TCP_SKB_CB reference means this can not be used from early demux */ 861 static inline int tcp_v6_sdif(const struct sk_buff *skb) 862 { 863 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 864 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 865 return TCP_SKB_CB(skb)->header.h6.iif; 866 #endif 867 return 0; 868 } 869 #endif 870 871 /* TCP_SKB_CB reference means this can not be used from early demux */ 872 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 873 { 874 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 875 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 876 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 877 return true; 878 #endif 879 return false; 880 } 881 882 /* TCP_SKB_CB reference means this can not be used from early demux */ 883 static inline int tcp_v4_sdif(struct sk_buff *skb) 884 { 885 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 886 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 887 return TCP_SKB_CB(skb)->header.h4.iif; 888 #endif 889 return 0; 890 } 891 892 /* Due to TSO, an SKB can be composed of multiple actual 893 * packets. To keep these tracked properly, we use this. 894 */ 895 static inline int tcp_skb_pcount(const struct sk_buff *skb) 896 { 897 return TCP_SKB_CB(skb)->tcp_gso_segs; 898 } 899 900 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 901 { 902 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 903 } 904 905 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 906 { 907 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 908 } 909 910 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 911 static inline int tcp_skb_mss(const struct sk_buff *skb) 912 { 913 return TCP_SKB_CB(skb)->tcp_gso_size; 914 } 915 916 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 917 { 918 return likely(!TCP_SKB_CB(skb)->eor); 919 } 920 921 /* Events passed to congestion control interface */ 922 enum tcp_ca_event { 923 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 924 CA_EVENT_CWND_RESTART, /* congestion window restart */ 925 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 926 CA_EVENT_LOSS, /* loss timeout */ 927 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 928 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 929 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 930 CA_EVENT_NON_DELAYED_ACK, 931 }; 932 933 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 934 enum tcp_ca_ack_event_flags { 935 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 936 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 937 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 938 }; 939 940 /* 941 * Interface for adding new TCP congestion control handlers 942 */ 943 #define TCP_CA_NAME_MAX 16 944 #define TCP_CA_MAX 128 945 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 946 947 #define TCP_CA_UNSPEC 0 948 949 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 950 #define TCP_CONG_NON_RESTRICTED 0x1 951 /* Requires ECN/ECT set on all packets */ 952 #define TCP_CONG_NEEDS_ECN 0x2 953 954 union tcp_cc_info; 955 956 struct ack_sample { 957 u32 pkts_acked; 958 s32 rtt_us; 959 u32 in_flight; 960 }; 961 962 /* A rate sample measures the number of (original/retransmitted) data 963 * packets delivered "delivered" over an interval of time "interval_us". 964 * The tcp_rate.c code fills in the rate sample, and congestion 965 * control modules that define a cong_control function to run at the end 966 * of ACK processing can optionally chose to consult this sample when 967 * setting cwnd and pacing rate. 968 * A sample is invalid if "delivered" or "interval_us" is negative. 969 */ 970 struct rate_sample { 971 u64 prior_mstamp; /* starting timestamp for interval */ 972 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 973 s32 delivered; /* number of packets delivered over interval */ 974 long interval_us; /* time for tp->delivered to incr "delivered" */ 975 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 976 int losses; /* number of packets marked lost upon ACK */ 977 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 978 u32 prior_in_flight; /* in flight before this ACK */ 979 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 980 bool is_retrans; /* is sample from retransmission? */ 981 }; 982 983 struct tcp_congestion_ops { 984 struct list_head list; 985 u32 key; 986 u32 flags; 987 988 /* initialize private data (optional) */ 989 void (*init)(struct sock *sk); 990 /* cleanup private data (optional) */ 991 void (*release)(struct sock *sk); 992 993 /* return slow start threshold (required) */ 994 u32 (*ssthresh)(struct sock *sk); 995 /* do new cwnd calculation (required) */ 996 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 997 /* call before changing ca_state (optional) */ 998 void (*set_state)(struct sock *sk, u8 new_state); 999 /* call when cwnd event occurs (optional) */ 1000 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1001 /* call when ack arrives (optional) */ 1002 void (*in_ack_event)(struct sock *sk, u32 flags); 1003 /* new value of cwnd after loss (required) */ 1004 u32 (*undo_cwnd)(struct sock *sk); 1005 /* hook for packet ack accounting (optional) */ 1006 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1007 /* suggest number of segments for each skb to transmit (optional) */ 1008 u32 (*tso_segs_goal)(struct sock *sk); 1009 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1010 u32 (*sndbuf_expand)(struct sock *sk); 1011 /* call when packets are delivered to update cwnd and pacing rate, 1012 * after all the ca_state processing. (optional) 1013 */ 1014 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1015 /* get info for inet_diag (optional) */ 1016 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1017 union tcp_cc_info *info); 1018 1019 char name[TCP_CA_NAME_MAX]; 1020 struct module *owner; 1021 }; 1022 1023 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1024 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1025 1026 void tcp_assign_congestion_control(struct sock *sk); 1027 void tcp_init_congestion_control(struct sock *sk); 1028 void tcp_cleanup_congestion_control(struct sock *sk); 1029 int tcp_set_default_congestion_control(const char *name); 1030 void tcp_get_default_congestion_control(char *name); 1031 void tcp_get_available_congestion_control(char *buf, size_t len); 1032 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1033 int tcp_set_allowed_congestion_control(char *allowed); 1034 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1035 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1036 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1037 1038 u32 tcp_reno_ssthresh(struct sock *sk); 1039 u32 tcp_reno_undo_cwnd(struct sock *sk); 1040 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1041 extern struct tcp_congestion_ops tcp_reno; 1042 1043 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1044 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1045 #ifdef CONFIG_INET 1046 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1047 #else 1048 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1049 { 1050 return NULL; 1051 } 1052 #endif 1053 1054 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1055 { 1056 const struct inet_connection_sock *icsk = inet_csk(sk); 1057 1058 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1059 } 1060 1061 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1062 { 1063 struct inet_connection_sock *icsk = inet_csk(sk); 1064 1065 if (icsk->icsk_ca_ops->set_state) 1066 icsk->icsk_ca_ops->set_state(sk, ca_state); 1067 icsk->icsk_ca_state = ca_state; 1068 } 1069 1070 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1071 { 1072 const struct inet_connection_sock *icsk = inet_csk(sk); 1073 1074 if (icsk->icsk_ca_ops->cwnd_event) 1075 icsk->icsk_ca_ops->cwnd_event(sk, event); 1076 } 1077 1078 /* From tcp_rate.c */ 1079 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1080 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1081 struct rate_sample *rs); 1082 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1083 struct rate_sample *rs); 1084 void tcp_rate_check_app_limited(struct sock *sk); 1085 1086 /* These functions determine how the current flow behaves in respect of SACK 1087 * handling. SACK is negotiated with the peer, and therefore it can vary 1088 * between different flows. 1089 * 1090 * tcp_is_sack - SACK enabled 1091 * tcp_is_reno - No SACK 1092 * tcp_is_fack - FACK enabled, implies SACK enabled 1093 */ 1094 static inline int tcp_is_sack(const struct tcp_sock *tp) 1095 { 1096 return tp->rx_opt.sack_ok; 1097 } 1098 1099 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1100 { 1101 return !tcp_is_sack(tp); 1102 } 1103 1104 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1105 { 1106 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1107 } 1108 1109 static inline void tcp_enable_fack(struct tcp_sock *tp) 1110 { 1111 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1112 } 1113 1114 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1115 { 1116 return tp->sacked_out + tp->lost_out; 1117 } 1118 1119 /* This determines how many packets are "in the network" to the best 1120 * of our knowledge. In many cases it is conservative, but where 1121 * detailed information is available from the receiver (via SACK 1122 * blocks etc.) we can make more aggressive calculations. 1123 * 1124 * Use this for decisions involving congestion control, use just 1125 * tp->packets_out to determine if the send queue is empty or not. 1126 * 1127 * Read this equation as: 1128 * 1129 * "Packets sent once on transmission queue" MINUS 1130 * "Packets left network, but not honestly ACKed yet" PLUS 1131 * "Packets fast retransmitted" 1132 */ 1133 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1134 { 1135 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1136 } 1137 1138 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1139 1140 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1141 { 1142 return tp->snd_cwnd < tp->snd_ssthresh; 1143 } 1144 1145 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1146 { 1147 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1148 } 1149 1150 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1151 { 1152 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1153 (1 << inet_csk(sk)->icsk_ca_state); 1154 } 1155 1156 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1157 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1158 * ssthresh. 1159 */ 1160 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1161 { 1162 const struct tcp_sock *tp = tcp_sk(sk); 1163 1164 if (tcp_in_cwnd_reduction(sk)) 1165 return tp->snd_ssthresh; 1166 else 1167 return max(tp->snd_ssthresh, 1168 ((tp->snd_cwnd >> 1) + 1169 (tp->snd_cwnd >> 2))); 1170 } 1171 1172 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1173 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1174 1175 void tcp_enter_cwr(struct sock *sk); 1176 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1177 1178 /* The maximum number of MSS of available cwnd for which TSO defers 1179 * sending if not using sysctl_tcp_tso_win_divisor. 1180 */ 1181 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1182 { 1183 return 3; 1184 } 1185 1186 /* Returns end sequence number of the receiver's advertised window */ 1187 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1188 { 1189 return tp->snd_una + tp->snd_wnd; 1190 } 1191 1192 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1193 * flexible approach. The RFC suggests cwnd should not be raised unless 1194 * it was fully used previously. And that's exactly what we do in 1195 * congestion avoidance mode. But in slow start we allow cwnd to grow 1196 * as long as the application has used half the cwnd. 1197 * Example : 1198 * cwnd is 10 (IW10), but application sends 9 frames. 1199 * We allow cwnd to reach 18 when all frames are ACKed. 1200 * This check is safe because it's as aggressive as slow start which already 1201 * risks 100% overshoot. The advantage is that we discourage application to 1202 * either send more filler packets or data to artificially blow up the cwnd 1203 * usage, and allow application-limited process to probe bw more aggressively. 1204 */ 1205 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1206 { 1207 const struct tcp_sock *tp = tcp_sk(sk); 1208 1209 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1210 if (tcp_in_slow_start(tp)) 1211 return tp->snd_cwnd < 2 * tp->max_packets_out; 1212 1213 return tp->is_cwnd_limited; 1214 } 1215 1216 /* Something is really bad, we could not queue an additional packet, 1217 * because qdisc is full or receiver sent a 0 window. 1218 * We do not want to add fuel to the fire, or abort too early, 1219 * so make sure the timer we arm now is at least 200ms in the future, 1220 * regardless of current icsk_rto value (as it could be ~2ms) 1221 */ 1222 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1223 { 1224 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1225 } 1226 1227 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1228 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1229 unsigned long max_when) 1230 { 1231 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1232 1233 return (unsigned long)min_t(u64, when, max_when); 1234 } 1235 1236 static inline void tcp_check_probe_timer(struct sock *sk) 1237 { 1238 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1239 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1240 tcp_probe0_base(sk), TCP_RTO_MAX); 1241 } 1242 1243 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1244 { 1245 tp->snd_wl1 = seq; 1246 } 1247 1248 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1249 { 1250 tp->snd_wl1 = seq; 1251 } 1252 1253 /* 1254 * Calculate(/check) TCP checksum 1255 */ 1256 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1257 __be32 daddr, __wsum base) 1258 { 1259 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1260 } 1261 1262 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1263 { 1264 return __skb_checksum_complete(skb); 1265 } 1266 1267 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1268 { 1269 return !skb_csum_unnecessary(skb) && 1270 __tcp_checksum_complete(skb); 1271 } 1272 1273 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1274 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1275 1276 #undef STATE_TRACE 1277 1278 #ifdef STATE_TRACE 1279 static const char *statename[]={ 1280 "Unused","Established","Syn Sent","Syn Recv", 1281 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1282 "Close Wait","Last ACK","Listen","Closing" 1283 }; 1284 #endif 1285 void tcp_set_state(struct sock *sk, int state); 1286 1287 void tcp_done(struct sock *sk); 1288 1289 int tcp_abort(struct sock *sk, int err); 1290 1291 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1292 { 1293 rx_opt->dsack = 0; 1294 rx_opt->num_sacks = 0; 1295 } 1296 1297 u32 tcp_default_init_rwnd(u32 mss); 1298 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1299 1300 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1301 { 1302 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1303 struct tcp_sock *tp = tcp_sk(sk); 1304 s32 delta; 1305 1306 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out || 1307 ca_ops->cong_control) 1308 return; 1309 delta = tcp_jiffies32 - tp->lsndtime; 1310 if (delta > inet_csk(sk)->icsk_rto) 1311 tcp_cwnd_restart(sk, delta); 1312 } 1313 1314 /* Determine a window scaling and initial window to offer. */ 1315 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1316 __u32 *window_clamp, int wscale_ok, 1317 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1318 1319 static inline int tcp_win_from_space(int space) 1320 { 1321 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale; 1322 1323 return tcp_adv_win_scale <= 0 ? 1324 (space>>(-tcp_adv_win_scale)) : 1325 space - (space>>tcp_adv_win_scale); 1326 } 1327 1328 /* Note: caller must be prepared to deal with negative returns */ 1329 static inline int tcp_space(const struct sock *sk) 1330 { 1331 return tcp_win_from_space(sk->sk_rcvbuf - 1332 atomic_read(&sk->sk_rmem_alloc)); 1333 } 1334 1335 static inline int tcp_full_space(const struct sock *sk) 1336 { 1337 return tcp_win_from_space(sk->sk_rcvbuf); 1338 } 1339 1340 extern void tcp_openreq_init_rwin(struct request_sock *req, 1341 const struct sock *sk_listener, 1342 const struct dst_entry *dst); 1343 1344 void tcp_enter_memory_pressure(struct sock *sk); 1345 void tcp_leave_memory_pressure(struct sock *sk); 1346 1347 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1348 { 1349 struct net *net = sock_net((struct sock *)tp); 1350 1351 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1352 } 1353 1354 static inline int keepalive_time_when(const struct tcp_sock *tp) 1355 { 1356 struct net *net = sock_net((struct sock *)tp); 1357 1358 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1359 } 1360 1361 static inline int keepalive_probes(const struct tcp_sock *tp) 1362 { 1363 struct net *net = sock_net((struct sock *)tp); 1364 1365 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1366 } 1367 1368 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1369 { 1370 const struct inet_connection_sock *icsk = &tp->inet_conn; 1371 1372 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1373 tcp_jiffies32 - tp->rcv_tstamp); 1374 } 1375 1376 static inline int tcp_fin_time(const struct sock *sk) 1377 { 1378 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1379 const int rto = inet_csk(sk)->icsk_rto; 1380 1381 if (fin_timeout < (rto << 2) - (rto >> 1)) 1382 fin_timeout = (rto << 2) - (rto >> 1); 1383 1384 return fin_timeout; 1385 } 1386 1387 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1388 int paws_win) 1389 { 1390 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1391 return true; 1392 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1393 return true; 1394 /* 1395 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1396 * then following tcp messages have valid values. Ignore 0 value, 1397 * or else 'negative' tsval might forbid us to accept their packets. 1398 */ 1399 if (!rx_opt->ts_recent) 1400 return true; 1401 return false; 1402 } 1403 1404 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1405 int rst) 1406 { 1407 if (tcp_paws_check(rx_opt, 0)) 1408 return false; 1409 1410 /* RST segments are not recommended to carry timestamp, 1411 and, if they do, it is recommended to ignore PAWS because 1412 "their cleanup function should take precedence over timestamps." 1413 Certainly, it is mistake. It is necessary to understand the reasons 1414 of this constraint to relax it: if peer reboots, clock may go 1415 out-of-sync and half-open connections will not be reset. 1416 Actually, the problem would be not existing if all 1417 the implementations followed draft about maintaining clock 1418 via reboots. Linux-2.2 DOES NOT! 1419 1420 However, we can relax time bounds for RST segments to MSL. 1421 */ 1422 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1423 return false; 1424 return true; 1425 } 1426 1427 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1428 int mib_idx, u32 *last_oow_ack_time); 1429 1430 static inline void tcp_mib_init(struct net *net) 1431 { 1432 /* See RFC 2012 */ 1433 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1434 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1435 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1436 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1437 } 1438 1439 /* from STCP */ 1440 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1441 { 1442 tp->lost_skb_hint = NULL; 1443 } 1444 1445 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1446 { 1447 tcp_clear_retrans_hints_partial(tp); 1448 tp->retransmit_skb_hint = NULL; 1449 } 1450 1451 union tcp_md5_addr { 1452 struct in_addr a4; 1453 #if IS_ENABLED(CONFIG_IPV6) 1454 struct in6_addr a6; 1455 #endif 1456 }; 1457 1458 /* - key database */ 1459 struct tcp_md5sig_key { 1460 struct hlist_node node; 1461 u8 keylen; 1462 u8 family; /* AF_INET or AF_INET6 */ 1463 union tcp_md5_addr addr; 1464 u8 prefixlen; 1465 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1466 struct rcu_head rcu; 1467 }; 1468 1469 /* - sock block */ 1470 struct tcp_md5sig_info { 1471 struct hlist_head head; 1472 struct rcu_head rcu; 1473 }; 1474 1475 /* - pseudo header */ 1476 struct tcp4_pseudohdr { 1477 __be32 saddr; 1478 __be32 daddr; 1479 __u8 pad; 1480 __u8 protocol; 1481 __be16 len; 1482 }; 1483 1484 struct tcp6_pseudohdr { 1485 struct in6_addr saddr; 1486 struct in6_addr daddr; 1487 __be32 len; 1488 __be32 protocol; /* including padding */ 1489 }; 1490 1491 union tcp_md5sum_block { 1492 struct tcp4_pseudohdr ip4; 1493 #if IS_ENABLED(CONFIG_IPV6) 1494 struct tcp6_pseudohdr ip6; 1495 #endif 1496 }; 1497 1498 /* - pool: digest algorithm, hash description and scratch buffer */ 1499 struct tcp_md5sig_pool { 1500 struct ahash_request *md5_req; 1501 void *scratch; 1502 }; 1503 1504 /* - functions */ 1505 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1506 const struct sock *sk, const struct sk_buff *skb); 1507 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1508 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1509 gfp_t gfp); 1510 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1511 int family, u8 prefixlen); 1512 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1513 const struct sock *addr_sk); 1514 1515 #ifdef CONFIG_TCP_MD5SIG 1516 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1517 const union tcp_md5_addr *addr, 1518 int family); 1519 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1520 #else 1521 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1522 const union tcp_md5_addr *addr, 1523 int family) 1524 { 1525 return NULL; 1526 } 1527 #define tcp_twsk_md5_key(twsk) NULL 1528 #endif 1529 1530 bool tcp_alloc_md5sig_pool(void); 1531 1532 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1533 static inline void tcp_put_md5sig_pool(void) 1534 { 1535 local_bh_enable(); 1536 } 1537 1538 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1539 unsigned int header_len); 1540 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1541 const struct tcp_md5sig_key *key); 1542 1543 /* From tcp_fastopen.c */ 1544 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1545 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1546 unsigned long *last_syn_loss); 1547 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1548 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1549 u16 try_exp); 1550 struct tcp_fastopen_request { 1551 /* Fast Open cookie. Size 0 means a cookie request */ 1552 struct tcp_fastopen_cookie cookie; 1553 struct msghdr *data; /* data in MSG_FASTOPEN */ 1554 size_t size; 1555 int copied; /* queued in tcp_connect() */ 1556 }; 1557 void tcp_free_fastopen_req(struct tcp_sock *tp); 1558 1559 void tcp_fastopen_ctx_destroy(struct net *net); 1560 int tcp_fastopen_reset_cipher(struct net *net, void *key, unsigned int len); 1561 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1562 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1563 struct request_sock *req, 1564 struct tcp_fastopen_cookie *foc); 1565 void tcp_fastopen_init_key_once(struct net *net); 1566 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1567 struct tcp_fastopen_cookie *cookie); 1568 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1569 #define TCP_FASTOPEN_KEY_LENGTH 16 1570 1571 /* Fastopen key context */ 1572 struct tcp_fastopen_context { 1573 struct crypto_cipher *tfm; 1574 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1575 struct rcu_head rcu; 1576 }; 1577 1578 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1579 void tcp_fastopen_active_disable(struct sock *sk); 1580 bool tcp_fastopen_active_should_disable(struct sock *sk); 1581 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1582 void tcp_fastopen_active_timeout_reset(void); 1583 1584 /* Latencies incurred by various limits for a sender. They are 1585 * chronograph-like stats that are mutually exclusive. 1586 */ 1587 enum tcp_chrono { 1588 TCP_CHRONO_UNSPEC, 1589 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1590 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1591 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1592 __TCP_CHRONO_MAX, 1593 }; 1594 1595 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1596 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1597 1598 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1599 * the same memory storage than skb->destructor/_skb_refdst 1600 */ 1601 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1602 { 1603 skb->destructor = NULL; 1604 skb->_skb_refdst = 0UL; 1605 } 1606 1607 #define tcp_skb_tsorted_save(skb) { \ 1608 unsigned long _save = skb->_skb_refdst; \ 1609 skb->_skb_refdst = 0UL; 1610 1611 #define tcp_skb_tsorted_restore(skb) \ 1612 skb->_skb_refdst = _save; \ 1613 } 1614 1615 void tcp_write_queue_purge(struct sock *sk); 1616 1617 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1618 { 1619 return skb_rb_first(&sk->tcp_rtx_queue); 1620 } 1621 1622 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1623 { 1624 return skb_peek(&sk->sk_write_queue); 1625 } 1626 1627 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1628 { 1629 return skb_peek_tail(&sk->sk_write_queue); 1630 } 1631 1632 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1633 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1634 1635 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1636 { 1637 return skb_peek(&sk->sk_write_queue); 1638 } 1639 1640 static inline bool tcp_skb_is_last(const struct sock *sk, 1641 const struct sk_buff *skb) 1642 { 1643 return skb_queue_is_last(&sk->sk_write_queue, skb); 1644 } 1645 1646 static inline bool tcp_write_queue_empty(const struct sock *sk) 1647 { 1648 return skb_queue_empty(&sk->sk_write_queue); 1649 } 1650 1651 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1652 { 1653 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1654 } 1655 1656 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1657 { 1658 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1659 } 1660 1661 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1662 { 1663 if (tcp_write_queue_empty(sk)) 1664 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1665 1666 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1667 tcp_sk(sk)->highest_sack = NULL; 1668 } 1669 1670 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1671 { 1672 __skb_queue_tail(&sk->sk_write_queue, skb); 1673 } 1674 1675 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1676 { 1677 __tcp_add_write_queue_tail(sk, skb); 1678 1679 /* Queue it, remembering where we must start sending. */ 1680 if (sk->sk_write_queue.next == skb) { 1681 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1682 1683 if (tcp_sk(sk)->highest_sack == NULL) 1684 tcp_sk(sk)->highest_sack = skb; 1685 } 1686 } 1687 1688 /* Insert new before skb on the write queue of sk. */ 1689 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1690 struct sk_buff *skb, 1691 struct sock *sk) 1692 { 1693 __skb_queue_before(&sk->sk_write_queue, skb, new); 1694 } 1695 1696 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1697 { 1698 tcp_skb_tsorted_anchor_cleanup(skb); 1699 __skb_unlink(skb, &sk->sk_write_queue); 1700 } 1701 1702 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1703 1704 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1705 { 1706 tcp_skb_tsorted_anchor_cleanup(skb); 1707 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1708 } 1709 1710 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1711 { 1712 list_del(&skb->tcp_tsorted_anchor); 1713 tcp_rtx_queue_unlink(skb, sk); 1714 sk_wmem_free_skb(sk, skb); 1715 } 1716 1717 static inline void tcp_push_pending_frames(struct sock *sk) 1718 { 1719 if (tcp_send_head(sk)) { 1720 struct tcp_sock *tp = tcp_sk(sk); 1721 1722 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1723 } 1724 } 1725 1726 /* Start sequence of the skb just after the highest skb with SACKed 1727 * bit, valid only if sacked_out > 0 or when the caller has ensured 1728 * validity by itself. 1729 */ 1730 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1731 { 1732 if (!tp->sacked_out) 1733 return tp->snd_una; 1734 1735 if (tp->highest_sack == NULL) 1736 return tp->snd_nxt; 1737 1738 return TCP_SKB_CB(tp->highest_sack)->seq; 1739 } 1740 1741 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1742 { 1743 struct sk_buff *next = skb_rb_next(skb); 1744 1745 tcp_sk(sk)->highest_sack = next ?: tcp_send_head(sk); 1746 } 1747 1748 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1749 { 1750 return tcp_sk(sk)->highest_sack; 1751 } 1752 1753 static inline void tcp_highest_sack_reset(struct sock *sk) 1754 { 1755 struct sk_buff *skb = tcp_rtx_queue_head(sk); 1756 1757 tcp_sk(sk)->highest_sack = skb ?: tcp_send_head(sk); 1758 } 1759 1760 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1761 static inline void tcp_highest_sack_combine(struct sock *sk, 1762 struct sk_buff *old, 1763 struct sk_buff *new) 1764 { 1765 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1766 tcp_sk(sk)->highest_sack = new; 1767 } 1768 1769 /* This helper checks if socket has IP_TRANSPARENT set */ 1770 static inline bool inet_sk_transparent(const struct sock *sk) 1771 { 1772 switch (sk->sk_state) { 1773 case TCP_TIME_WAIT: 1774 return inet_twsk(sk)->tw_transparent; 1775 case TCP_NEW_SYN_RECV: 1776 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1777 } 1778 return inet_sk(sk)->transparent; 1779 } 1780 1781 /* Determines whether this is a thin stream (which may suffer from 1782 * increased latency). Used to trigger latency-reducing mechanisms. 1783 */ 1784 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1785 { 1786 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1787 } 1788 1789 /* /proc */ 1790 enum tcp_seq_states { 1791 TCP_SEQ_STATE_LISTENING, 1792 TCP_SEQ_STATE_ESTABLISHED, 1793 }; 1794 1795 int tcp_seq_open(struct inode *inode, struct file *file); 1796 1797 struct tcp_seq_afinfo { 1798 char *name; 1799 sa_family_t family; 1800 const struct file_operations *seq_fops; 1801 struct seq_operations seq_ops; 1802 }; 1803 1804 struct tcp_iter_state { 1805 struct seq_net_private p; 1806 sa_family_t family; 1807 enum tcp_seq_states state; 1808 struct sock *syn_wait_sk; 1809 int bucket, offset, sbucket, num; 1810 loff_t last_pos; 1811 }; 1812 1813 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1814 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1815 1816 extern struct request_sock_ops tcp_request_sock_ops; 1817 extern struct request_sock_ops tcp6_request_sock_ops; 1818 1819 void tcp_v4_destroy_sock(struct sock *sk); 1820 1821 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1822 netdev_features_t features); 1823 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1824 int tcp_gro_complete(struct sk_buff *skb); 1825 1826 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1827 1828 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1829 { 1830 struct net *net = sock_net((struct sock *)tp); 1831 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1832 } 1833 1834 static inline bool tcp_stream_memory_free(const struct sock *sk) 1835 { 1836 const struct tcp_sock *tp = tcp_sk(sk); 1837 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1838 1839 return notsent_bytes < tcp_notsent_lowat(tp); 1840 } 1841 1842 #ifdef CONFIG_PROC_FS 1843 int tcp4_proc_init(void); 1844 void tcp4_proc_exit(void); 1845 #endif 1846 1847 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1848 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1849 const struct tcp_request_sock_ops *af_ops, 1850 struct sock *sk, struct sk_buff *skb); 1851 1852 /* TCP af-specific functions */ 1853 struct tcp_sock_af_ops { 1854 #ifdef CONFIG_TCP_MD5SIG 1855 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1856 const struct sock *addr_sk); 1857 int (*calc_md5_hash)(char *location, 1858 const struct tcp_md5sig_key *md5, 1859 const struct sock *sk, 1860 const struct sk_buff *skb); 1861 int (*md5_parse)(struct sock *sk, 1862 int optname, 1863 char __user *optval, 1864 int optlen); 1865 #endif 1866 }; 1867 1868 struct tcp_request_sock_ops { 1869 u16 mss_clamp; 1870 #ifdef CONFIG_TCP_MD5SIG 1871 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1872 const struct sock *addr_sk); 1873 int (*calc_md5_hash) (char *location, 1874 const struct tcp_md5sig_key *md5, 1875 const struct sock *sk, 1876 const struct sk_buff *skb); 1877 #endif 1878 void (*init_req)(struct request_sock *req, 1879 const struct sock *sk_listener, 1880 struct sk_buff *skb); 1881 #ifdef CONFIG_SYN_COOKIES 1882 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1883 __u16 *mss); 1884 #endif 1885 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1886 const struct request_sock *req); 1887 u32 (*init_seq)(const struct sk_buff *skb); 1888 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1889 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1890 struct flowi *fl, struct request_sock *req, 1891 struct tcp_fastopen_cookie *foc, 1892 enum tcp_synack_type synack_type); 1893 }; 1894 1895 #ifdef CONFIG_SYN_COOKIES 1896 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1897 const struct sock *sk, struct sk_buff *skb, 1898 __u16 *mss) 1899 { 1900 tcp_synq_overflow(sk); 1901 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1902 return ops->cookie_init_seq(skb, mss); 1903 } 1904 #else 1905 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1906 const struct sock *sk, struct sk_buff *skb, 1907 __u16 *mss) 1908 { 1909 return 0; 1910 } 1911 #endif 1912 1913 int tcpv4_offload_init(void); 1914 1915 void tcp_v4_init(void); 1916 void tcp_init(void); 1917 1918 /* tcp_recovery.c */ 1919 extern void tcp_rack_mark_lost(struct sock *sk); 1920 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1921 u64 xmit_time); 1922 extern void tcp_rack_reo_timeout(struct sock *sk); 1923 1924 /* At how many usecs into the future should the RTO fire? */ 1925 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1926 { 1927 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1928 u32 rto = inet_csk(sk)->icsk_rto; 1929 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1930 1931 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1932 } 1933 1934 /* 1935 * Save and compile IPv4 options, return a pointer to it 1936 */ 1937 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1938 struct sk_buff *skb) 1939 { 1940 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1941 struct ip_options_rcu *dopt = NULL; 1942 1943 if (opt->optlen) { 1944 int opt_size = sizeof(*dopt) + opt->optlen; 1945 1946 dopt = kmalloc(opt_size, GFP_ATOMIC); 1947 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1948 kfree(dopt); 1949 dopt = NULL; 1950 } 1951 } 1952 return dopt; 1953 } 1954 1955 /* locally generated TCP pure ACKs have skb->truesize == 2 1956 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1957 * This is much faster than dissecting the packet to find out. 1958 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1959 */ 1960 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1961 { 1962 return skb->truesize == 2; 1963 } 1964 1965 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1966 { 1967 skb->truesize = 2; 1968 } 1969 1970 static inline int tcp_inq(struct sock *sk) 1971 { 1972 struct tcp_sock *tp = tcp_sk(sk); 1973 int answ; 1974 1975 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1976 answ = 0; 1977 } else if (sock_flag(sk, SOCK_URGINLINE) || 1978 !tp->urg_data || 1979 before(tp->urg_seq, tp->copied_seq) || 1980 !before(tp->urg_seq, tp->rcv_nxt)) { 1981 1982 answ = tp->rcv_nxt - tp->copied_seq; 1983 1984 /* Subtract 1, if FIN was received */ 1985 if (answ && sock_flag(sk, SOCK_DONE)) 1986 answ--; 1987 } else { 1988 answ = tp->urg_seq - tp->copied_seq; 1989 } 1990 1991 return answ; 1992 } 1993 1994 int tcp_peek_len(struct socket *sock); 1995 1996 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1997 { 1998 u16 segs_in; 1999 2000 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2001 tp->segs_in += segs_in; 2002 if (skb->len > tcp_hdrlen(skb)) 2003 tp->data_segs_in += segs_in; 2004 } 2005 2006 /* 2007 * TCP listen path runs lockless. 2008 * We forced "struct sock" to be const qualified to make sure 2009 * we don't modify one of its field by mistake. 2010 * Here, we increment sk_drops which is an atomic_t, so we can safely 2011 * make sock writable again. 2012 */ 2013 static inline void tcp_listendrop(const struct sock *sk) 2014 { 2015 atomic_inc(&((struct sock *)sk)->sk_drops); 2016 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2017 } 2018 2019 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2020 2021 /* 2022 * Interface for adding Upper Level Protocols over TCP 2023 */ 2024 2025 #define TCP_ULP_NAME_MAX 16 2026 #define TCP_ULP_MAX 128 2027 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2028 2029 struct tcp_ulp_ops { 2030 struct list_head list; 2031 2032 /* initialize ulp */ 2033 int (*init)(struct sock *sk); 2034 /* cleanup ulp */ 2035 void (*release)(struct sock *sk); 2036 2037 char name[TCP_ULP_NAME_MAX]; 2038 struct module *owner; 2039 }; 2040 int tcp_register_ulp(struct tcp_ulp_ops *type); 2041 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2042 int tcp_set_ulp(struct sock *sk, const char *name); 2043 void tcp_get_available_ulp(char *buf, size_t len); 2044 void tcp_cleanup_ulp(struct sock *sk); 2045 2046 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2047 * is < 0, then the BPF op failed (for example if the loaded BPF 2048 * program does not support the chosen operation or there is no BPF 2049 * program loaded). 2050 */ 2051 #ifdef CONFIG_BPF 2052 static inline int tcp_call_bpf(struct sock *sk, int op) 2053 { 2054 struct bpf_sock_ops_kern sock_ops; 2055 int ret; 2056 2057 if (sk_fullsock(sk)) 2058 sock_owned_by_me(sk); 2059 2060 memset(&sock_ops, 0, sizeof(sock_ops)); 2061 sock_ops.sk = sk; 2062 sock_ops.op = op; 2063 2064 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2065 if (ret == 0) 2066 ret = sock_ops.reply; 2067 else 2068 ret = -1; 2069 return ret; 2070 } 2071 #else 2072 static inline int tcp_call_bpf(struct sock *sk, int op) 2073 { 2074 return -EPERM; 2075 } 2076 #endif 2077 2078 static inline u32 tcp_timeout_init(struct sock *sk) 2079 { 2080 int timeout; 2081 2082 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2083 2084 if (timeout <= 0) 2085 timeout = TCP_TIMEOUT_INIT; 2086 return timeout; 2087 } 2088 2089 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2090 { 2091 int rwnd; 2092 2093 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2094 2095 if (rwnd < 0) 2096 rwnd = 0; 2097 return rwnd; 2098 } 2099 2100 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2101 { 2102 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2103 } 2104 #endif /* _TCP_H */ 2105