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 void 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 unsigned int tcp_poll(struct file *file, struct socket *sock, 420 struct poll_table_struct *wait); 421 int tcp_getsockopt(struct sock *sk, int level, int optname, 422 char __user *optval, int __user *optlen); 423 int tcp_setsockopt(struct sock *sk, int level, int optname, 424 char __user *optval, unsigned int optlen); 425 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 426 char __user *optval, int __user *optlen); 427 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 428 char __user *optval, unsigned int optlen); 429 void tcp_set_keepalive(struct sock *sk, int val); 430 void tcp_syn_ack_timeout(const struct request_sock *req); 431 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 432 int flags, int *addr_len); 433 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 434 struct tcp_options_received *opt_rx, 435 int estab, struct tcp_fastopen_cookie *foc); 436 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 437 438 /* 439 * TCP v4 functions exported for the inet6 API 440 */ 441 442 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 443 void tcp_v4_mtu_reduced(struct sock *sk); 444 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 445 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 446 struct sock *tcp_create_openreq_child(const struct sock *sk, 447 struct request_sock *req, 448 struct sk_buff *skb); 449 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 450 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 451 struct request_sock *req, 452 struct dst_entry *dst, 453 struct request_sock *req_unhash, 454 bool *own_req); 455 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 456 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 457 int tcp_connect(struct sock *sk); 458 enum tcp_synack_type { 459 TCP_SYNACK_NORMAL, 460 TCP_SYNACK_FASTOPEN, 461 TCP_SYNACK_COOKIE, 462 }; 463 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 464 struct request_sock *req, 465 struct tcp_fastopen_cookie *foc, 466 enum tcp_synack_type synack_type); 467 int tcp_disconnect(struct sock *sk, int flags); 468 469 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 470 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 471 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 472 473 /* From syncookies.c */ 474 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 475 struct request_sock *req, 476 struct dst_entry *dst, u32 tsoff); 477 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 478 u32 cookie); 479 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 480 #ifdef CONFIG_SYN_COOKIES 481 482 /* Syncookies use a monotonic timer which increments every 60 seconds. 483 * This counter is used both as a hash input and partially encoded into 484 * the cookie value. A cookie is only validated further if the delta 485 * between the current counter value and the encoded one is less than this, 486 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 487 * the counter advances immediately after a cookie is generated). 488 */ 489 #define MAX_SYNCOOKIE_AGE 2 490 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 491 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 492 493 /* syncookies: remember time of last synqueue overflow 494 * But do not dirty this field too often (once per second is enough) 495 * It is racy as we do not hold a lock, but race is very minor. 496 */ 497 static inline void tcp_synq_overflow(const struct sock *sk) 498 { 499 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 500 unsigned long now = jiffies; 501 502 if (time_after(now, last_overflow + HZ)) 503 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 504 } 505 506 /* syncookies: no recent synqueue overflow on this listening socket? */ 507 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 508 { 509 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 510 511 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 512 } 513 514 static inline u32 tcp_cookie_time(void) 515 { 516 u64 val = get_jiffies_64(); 517 518 do_div(val, TCP_SYNCOOKIE_PERIOD); 519 return val; 520 } 521 522 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 523 u16 *mssp); 524 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 525 u64 cookie_init_timestamp(struct request_sock *req); 526 bool cookie_timestamp_decode(const struct net *net, 527 struct tcp_options_received *opt); 528 bool cookie_ecn_ok(const struct tcp_options_received *opt, 529 const struct net *net, const struct dst_entry *dst); 530 531 /* From net/ipv6/syncookies.c */ 532 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 533 u32 cookie); 534 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 535 536 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 537 const struct tcphdr *th, u16 *mssp); 538 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 539 #endif 540 /* tcp_output.c */ 541 542 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 543 int min_tso_segs); 544 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 545 int nonagle); 546 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 547 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 548 void tcp_retransmit_timer(struct sock *sk); 549 void tcp_xmit_retransmit_queue(struct sock *); 550 void tcp_simple_retransmit(struct sock *); 551 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 552 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 553 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 554 555 void tcp_send_probe0(struct sock *); 556 void tcp_send_partial(struct sock *); 557 int tcp_write_wakeup(struct sock *, int mib); 558 void tcp_send_fin(struct sock *sk); 559 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 560 int tcp_send_synack(struct sock *); 561 void tcp_push_one(struct sock *, unsigned int mss_now); 562 void tcp_send_ack(struct sock *sk); 563 void tcp_send_delayed_ack(struct sock *sk); 564 void tcp_send_loss_probe(struct sock *sk); 565 bool tcp_schedule_loss_probe(struct sock *sk); 566 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 567 const struct sk_buff *next_skb); 568 569 /* tcp_input.c */ 570 void tcp_rearm_rto(struct sock *sk); 571 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 572 void tcp_reset(struct sock *sk); 573 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 574 void tcp_fin(struct sock *sk); 575 576 /* tcp_timer.c */ 577 void tcp_init_xmit_timers(struct sock *); 578 static inline void tcp_clear_xmit_timers(struct sock *sk) 579 { 580 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 581 inet_csk_clear_xmit_timers(sk); 582 } 583 584 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 585 unsigned int tcp_current_mss(struct sock *sk); 586 587 /* Bound MSS / TSO packet size with the half of the window */ 588 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 589 { 590 int cutoff; 591 592 /* When peer uses tiny windows, there is no use in packetizing 593 * to sub-MSS pieces for the sake of SWS or making sure there 594 * are enough packets in the pipe for fast recovery. 595 * 596 * On the other hand, for extremely large MSS devices, handling 597 * smaller than MSS windows in this way does make sense. 598 */ 599 if (tp->max_window > TCP_MSS_DEFAULT) 600 cutoff = (tp->max_window >> 1); 601 else 602 cutoff = tp->max_window; 603 604 if (cutoff && pktsize > cutoff) 605 return max_t(int, cutoff, 68U - tp->tcp_header_len); 606 else 607 return pktsize; 608 } 609 610 /* tcp.c */ 611 void tcp_get_info(struct sock *, struct tcp_info *); 612 613 /* Read 'sendfile()'-style from a TCP socket */ 614 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 615 sk_read_actor_t recv_actor); 616 617 void tcp_initialize_rcv_mss(struct sock *sk); 618 619 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 620 int tcp_mss_to_mtu(struct sock *sk, int mss); 621 void tcp_mtup_init(struct sock *sk); 622 void tcp_init_buffer_space(struct sock *sk); 623 624 static inline void tcp_bound_rto(const struct sock *sk) 625 { 626 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 627 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 628 } 629 630 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 631 { 632 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 633 } 634 635 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 636 { 637 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 638 ntohl(TCP_FLAG_ACK) | 639 snd_wnd); 640 } 641 642 static inline void tcp_fast_path_on(struct tcp_sock *tp) 643 { 644 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 645 } 646 647 static inline void tcp_fast_path_check(struct sock *sk) 648 { 649 struct tcp_sock *tp = tcp_sk(sk); 650 651 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 652 tp->rcv_wnd && 653 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 654 !tp->urg_data) 655 tcp_fast_path_on(tp); 656 } 657 658 /* Compute the actual rto_min value */ 659 static inline u32 tcp_rto_min(struct sock *sk) 660 { 661 const struct dst_entry *dst = __sk_dst_get(sk); 662 u32 rto_min = TCP_RTO_MIN; 663 664 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 665 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 666 return rto_min; 667 } 668 669 static inline u32 tcp_rto_min_us(struct sock *sk) 670 { 671 return jiffies_to_usecs(tcp_rto_min(sk)); 672 } 673 674 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 675 { 676 return dst_metric_locked(dst, RTAX_CC_ALGO); 677 } 678 679 /* Minimum RTT in usec. ~0 means not available. */ 680 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 681 { 682 return minmax_get(&tp->rtt_min); 683 } 684 685 /* Compute the actual receive window we are currently advertising. 686 * Rcv_nxt can be after the window if our peer push more data 687 * than the offered window. 688 */ 689 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 690 { 691 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 692 693 if (win < 0) 694 win = 0; 695 return (u32) win; 696 } 697 698 /* Choose a new window, without checks for shrinking, and without 699 * scaling applied to the result. The caller does these things 700 * if necessary. This is a "raw" window selection. 701 */ 702 u32 __tcp_select_window(struct sock *sk); 703 704 void tcp_send_window_probe(struct sock *sk); 705 706 /* TCP uses 32bit jiffies to save some space. 707 * Note that this is different from tcp_time_stamp, which 708 * historically has been the same until linux-4.13. 709 */ 710 #define tcp_jiffies32 ((u32)jiffies) 711 712 /* 713 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 714 * It is no longer tied to jiffies, but to 1 ms clock. 715 * Note: double check if you want to use tcp_jiffies32 instead of this. 716 */ 717 #define TCP_TS_HZ 1000 718 719 static inline u64 tcp_clock_ns(void) 720 { 721 return local_clock(); 722 } 723 724 static inline u64 tcp_clock_us(void) 725 { 726 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 727 } 728 729 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 730 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 731 { 732 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 733 } 734 735 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 736 static inline u32 tcp_time_stamp_raw(void) 737 { 738 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 739 } 740 741 742 /* Refresh 1us clock of a TCP socket, 743 * ensuring monotically increasing values. 744 */ 745 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 746 { 747 u64 val = tcp_clock_us(); 748 749 if (val > tp->tcp_mstamp) 750 tp->tcp_mstamp = val; 751 } 752 753 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 754 { 755 return max_t(s64, t1 - t0, 0); 756 } 757 758 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 759 { 760 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 761 } 762 763 764 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 765 766 #define TCPHDR_FIN 0x01 767 #define TCPHDR_SYN 0x02 768 #define TCPHDR_RST 0x04 769 #define TCPHDR_PSH 0x08 770 #define TCPHDR_ACK 0x10 771 #define TCPHDR_URG 0x20 772 #define TCPHDR_ECE 0x40 773 #define TCPHDR_CWR 0x80 774 775 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 776 777 /* This is what the send packet queuing engine uses to pass 778 * TCP per-packet control information to the transmission code. 779 * We also store the host-order sequence numbers in here too. 780 * This is 44 bytes if IPV6 is enabled. 781 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 782 */ 783 struct tcp_skb_cb { 784 __u32 seq; /* Starting sequence number */ 785 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 786 union { 787 /* Note : tcp_tw_isn is used in input path only 788 * (isn chosen by tcp_timewait_state_process()) 789 * 790 * tcp_gso_segs/size are used in write queue only, 791 * cf tcp_skb_pcount()/tcp_skb_mss() 792 */ 793 __u32 tcp_tw_isn; 794 struct { 795 u16 tcp_gso_segs; 796 u16 tcp_gso_size; 797 }; 798 }; 799 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 800 801 __u8 sacked; /* State flags for SACK/FACK. */ 802 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 803 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 804 #define TCPCB_LOST 0x04 /* SKB is lost */ 805 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 806 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 807 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 808 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 809 TCPCB_REPAIRED) 810 811 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 812 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 813 eor:1, /* Is skb MSG_EOR marked? */ 814 has_rxtstamp:1, /* SKB has a RX timestamp */ 815 unused:5; 816 __u32 ack_seq; /* Sequence number ACK'd */ 817 union { 818 struct { 819 /* There is space for up to 24 bytes */ 820 __u32 in_flight:30,/* Bytes in flight at transmit */ 821 is_app_limited:1, /* cwnd not fully used? */ 822 unused:1; 823 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 824 __u32 delivered; 825 /* start of send pipeline phase */ 826 u64 first_tx_mstamp; 827 /* when we reached the "delivered" count */ 828 u64 delivered_mstamp; 829 } tx; /* only used for outgoing skbs */ 830 union { 831 struct inet_skb_parm h4; 832 #if IS_ENABLED(CONFIG_IPV6) 833 struct inet6_skb_parm h6; 834 #endif 835 } header; /* For incoming skbs */ 836 }; 837 }; 838 839 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 840 841 842 #if IS_ENABLED(CONFIG_IPV6) 843 /* This is the variant of inet6_iif() that must be used by TCP, 844 * as TCP moves IP6CB into a different location in skb->cb[] 845 */ 846 static inline int tcp_v6_iif(const struct sk_buff *skb) 847 { 848 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 849 850 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 851 } 852 853 /* TCP_SKB_CB reference means this can not be used from early demux */ 854 static inline int tcp_v6_sdif(const struct sk_buff *skb) 855 { 856 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 857 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 858 return TCP_SKB_CB(skb)->header.h6.iif; 859 #endif 860 return 0; 861 } 862 #endif 863 864 /* TCP_SKB_CB reference means this can not be used from early demux */ 865 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 866 { 867 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 868 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 869 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 870 return true; 871 #endif 872 return false; 873 } 874 875 /* TCP_SKB_CB reference means this can not be used from early demux */ 876 static inline int tcp_v4_sdif(struct sk_buff *skb) 877 { 878 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 879 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 880 return TCP_SKB_CB(skb)->header.h4.iif; 881 #endif 882 return 0; 883 } 884 885 /* Due to TSO, an SKB can be composed of multiple actual 886 * packets. To keep these tracked properly, we use this. 887 */ 888 static inline int tcp_skb_pcount(const struct sk_buff *skb) 889 { 890 return TCP_SKB_CB(skb)->tcp_gso_segs; 891 } 892 893 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 894 { 895 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 896 } 897 898 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 899 { 900 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 901 } 902 903 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 904 static inline int tcp_skb_mss(const struct sk_buff *skb) 905 { 906 return TCP_SKB_CB(skb)->tcp_gso_size; 907 } 908 909 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 910 { 911 return likely(!TCP_SKB_CB(skb)->eor); 912 } 913 914 /* Events passed to congestion control interface */ 915 enum tcp_ca_event { 916 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 917 CA_EVENT_CWND_RESTART, /* congestion window restart */ 918 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 919 CA_EVENT_LOSS, /* loss timeout */ 920 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 921 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 922 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 923 CA_EVENT_NON_DELAYED_ACK, 924 }; 925 926 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 927 enum tcp_ca_ack_event_flags { 928 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 929 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 930 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 931 }; 932 933 /* 934 * Interface for adding new TCP congestion control handlers 935 */ 936 #define TCP_CA_NAME_MAX 16 937 #define TCP_CA_MAX 128 938 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 939 940 #define TCP_CA_UNSPEC 0 941 942 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 943 #define TCP_CONG_NON_RESTRICTED 0x1 944 /* Requires ECN/ECT set on all packets */ 945 #define TCP_CONG_NEEDS_ECN 0x2 946 947 union tcp_cc_info; 948 949 struct ack_sample { 950 u32 pkts_acked; 951 s32 rtt_us; 952 u32 in_flight; 953 }; 954 955 /* A rate sample measures the number of (original/retransmitted) data 956 * packets delivered "delivered" over an interval of time "interval_us". 957 * The tcp_rate.c code fills in the rate sample, and congestion 958 * control modules that define a cong_control function to run at the end 959 * of ACK processing can optionally chose to consult this sample when 960 * setting cwnd and pacing rate. 961 * A sample is invalid if "delivered" or "interval_us" is negative. 962 */ 963 struct rate_sample { 964 u64 prior_mstamp; /* starting timestamp for interval */ 965 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 966 s32 delivered; /* number of packets delivered over interval */ 967 long interval_us; /* time for tp->delivered to incr "delivered" */ 968 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 969 int losses; /* number of packets marked lost upon ACK */ 970 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 971 u32 prior_in_flight; /* in flight before this ACK */ 972 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 973 bool is_retrans; /* is sample from retransmission? */ 974 }; 975 976 struct tcp_congestion_ops { 977 struct list_head list; 978 u32 key; 979 u32 flags; 980 981 /* initialize private data (optional) */ 982 void (*init)(struct sock *sk); 983 /* cleanup private data (optional) */ 984 void (*release)(struct sock *sk); 985 986 /* return slow start threshold (required) */ 987 u32 (*ssthresh)(struct sock *sk); 988 /* do new cwnd calculation (required) */ 989 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 990 /* call before changing ca_state (optional) */ 991 void (*set_state)(struct sock *sk, u8 new_state); 992 /* call when cwnd event occurs (optional) */ 993 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 994 /* call when ack arrives (optional) */ 995 void (*in_ack_event)(struct sock *sk, u32 flags); 996 /* new value of cwnd after loss (required) */ 997 u32 (*undo_cwnd)(struct sock *sk); 998 /* hook for packet ack accounting (optional) */ 999 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1000 /* suggest number of segments for each skb to transmit (optional) */ 1001 u32 (*tso_segs_goal)(struct sock *sk); 1002 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1003 u32 (*sndbuf_expand)(struct sock *sk); 1004 /* call when packets are delivered to update cwnd and pacing rate, 1005 * after all the ca_state processing. (optional) 1006 */ 1007 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1008 /* get info for inet_diag (optional) */ 1009 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1010 union tcp_cc_info *info); 1011 1012 char name[TCP_CA_NAME_MAX]; 1013 struct module *owner; 1014 }; 1015 1016 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1017 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1018 1019 void tcp_assign_congestion_control(struct sock *sk); 1020 void tcp_init_congestion_control(struct sock *sk); 1021 void tcp_cleanup_congestion_control(struct sock *sk); 1022 int tcp_set_default_congestion_control(const char *name); 1023 void tcp_get_default_congestion_control(char *name); 1024 void tcp_get_available_congestion_control(char *buf, size_t len); 1025 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1026 int tcp_set_allowed_congestion_control(char *allowed); 1027 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1028 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1029 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1030 1031 u32 tcp_reno_ssthresh(struct sock *sk); 1032 u32 tcp_reno_undo_cwnd(struct sock *sk); 1033 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1034 extern struct tcp_congestion_ops tcp_reno; 1035 1036 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1037 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1038 #ifdef CONFIG_INET 1039 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1040 #else 1041 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1042 { 1043 return NULL; 1044 } 1045 #endif 1046 1047 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1048 { 1049 const struct inet_connection_sock *icsk = inet_csk(sk); 1050 1051 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1052 } 1053 1054 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1055 { 1056 struct inet_connection_sock *icsk = inet_csk(sk); 1057 1058 if (icsk->icsk_ca_ops->set_state) 1059 icsk->icsk_ca_ops->set_state(sk, ca_state); 1060 icsk->icsk_ca_state = ca_state; 1061 } 1062 1063 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1064 { 1065 const struct inet_connection_sock *icsk = inet_csk(sk); 1066 1067 if (icsk->icsk_ca_ops->cwnd_event) 1068 icsk->icsk_ca_ops->cwnd_event(sk, event); 1069 } 1070 1071 /* From tcp_rate.c */ 1072 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1073 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1074 struct rate_sample *rs); 1075 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1076 struct rate_sample *rs); 1077 void tcp_rate_check_app_limited(struct sock *sk); 1078 1079 /* These functions determine how the current flow behaves in respect of SACK 1080 * handling. SACK is negotiated with the peer, and therefore it can vary 1081 * between different flows. 1082 * 1083 * tcp_is_sack - SACK enabled 1084 * tcp_is_reno - No SACK 1085 * tcp_is_fack - FACK enabled, implies SACK enabled 1086 */ 1087 static inline int tcp_is_sack(const struct tcp_sock *tp) 1088 { 1089 return tp->rx_opt.sack_ok; 1090 } 1091 1092 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1093 { 1094 return !tcp_is_sack(tp); 1095 } 1096 1097 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1098 { 1099 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1100 } 1101 1102 static inline void tcp_enable_fack(struct tcp_sock *tp) 1103 { 1104 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1105 } 1106 1107 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1108 { 1109 return tp->sacked_out + tp->lost_out; 1110 } 1111 1112 /* This determines how many packets are "in the network" to the best 1113 * of our knowledge. In many cases it is conservative, but where 1114 * detailed information is available from the receiver (via SACK 1115 * blocks etc.) we can make more aggressive calculations. 1116 * 1117 * Use this for decisions involving congestion control, use just 1118 * tp->packets_out to determine if the send queue is empty or not. 1119 * 1120 * Read this equation as: 1121 * 1122 * "Packets sent once on transmission queue" MINUS 1123 * "Packets left network, but not honestly ACKed yet" PLUS 1124 * "Packets fast retransmitted" 1125 */ 1126 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1127 { 1128 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1129 } 1130 1131 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1132 1133 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1134 { 1135 return tp->snd_cwnd < tp->snd_ssthresh; 1136 } 1137 1138 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1139 { 1140 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1141 } 1142 1143 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1144 { 1145 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1146 (1 << inet_csk(sk)->icsk_ca_state); 1147 } 1148 1149 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1150 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1151 * ssthresh. 1152 */ 1153 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1154 { 1155 const struct tcp_sock *tp = tcp_sk(sk); 1156 1157 if (tcp_in_cwnd_reduction(sk)) 1158 return tp->snd_ssthresh; 1159 else 1160 return max(tp->snd_ssthresh, 1161 ((tp->snd_cwnd >> 1) + 1162 (tp->snd_cwnd >> 2))); 1163 } 1164 1165 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1166 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1167 1168 void tcp_enter_cwr(struct sock *sk); 1169 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1170 1171 /* The maximum number of MSS of available cwnd for which TSO defers 1172 * sending if not using sysctl_tcp_tso_win_divisor. 1173 */ 1174 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1175 { 1176 return 3; 1177 } 1178 1179 /* Returns end sequence number of the receiver's advertised window */ 1180 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1181 { 1182 return tp->snd_una + tp->snd_wnd; 1183 } 1184 1185 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1186 * flexible approach. The RFC suggests cwnd should not be raised unless 1187 * it was fully used previously. And that's exactly what we do in 1188 * congestion avoidance mode. But in slow start we allow cwnd to grow 1189 * as long as the application has used half the cwnd. 1190 * Example : 1191 * cwnd is 10 (IW10), but application sends 9 frames. 1192 * We allow cwnd to reach 18 when all frames are ACKed. 1193 * This check is safe because it's as aggressive as slow start which already 1194 * risks 100% overshoot. The advantage is that we discourage application to 1195 * either send more filler packets or data to artificially blow up the cwnd 1196 * usage, and allow application-limited process to probe bw more aggressively. 1197 */ 1198 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1199 { 1200 const struct tcp_sock *tp = tcp_sk(sk); 1201 1202 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1203 if (tcp_in_slow_start(tp)) 1204 return tp->snd_cwnd < 2 * tp->max_packets_out; 1205 1206 return tp->is_cwnd_limited; 1207 } 1208 1209 /* Something is really bad, we could not queue an additional packet, 1210 * because qdisc is full or receiver sent a 0 window. 1211 * We do not want to add fuel to the fire, or abort too early, 1212 * so make sure the timer we arm now is at least 200ms in the future, 1213 * regardless of current icsk_rto value (as it could be ~2ms) 1214 */ 1215 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1216 { 1217 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1218 } 1219 1220 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1221 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1222 unsigned long max_when) 1223 { 1224 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1225 1226 return (unsigned long)min_t(u64, when, max_when); 1227 } 1228 1229 static inline void tcp_check_probe_timer(struct sock *sk) 1230 { 1231 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1232 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1233 tcp_probe0_base(sk), TCP_RTO_MAX); 1234 } 1235 1236 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1237 { 1238 tp->snd_wl1 = seq; 1239 } 1240 1241 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1242 { 1243 tp->snd_wl1 = seq; 1244 } 1245 1246 /* 1247 * Calculate(/check) TCP checksum 1248 */ 1249 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1250 __be32 daddr, __wsum base) 1251 { 1252 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1253 } 1254 1255 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1256 { 1257 return __skb_checksum_complete(skb); 1258 } 1259 1260 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1261 { 1262 return !skb_csum_unnecessary(skb) && 1263 __tcp_checksum_complete(skb); 1264 } 1265 1266 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1267 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1268 1269 #undef STATE_TRACE 1270 1271 #ifdef STATE_TRACE 1272 static const char *statename[]={ 1273 "Unused","Established","Syn Sent","Syn Recv", 1274 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1275 "Close Wait","Last ACK","Listen","Closing" 1276 }; 1277 #endif 1278 void tcp_set_state(struct sock *sk, int state); 1279 1280 void tcp_done(struct sock *sk); 1281 1282 int tcp_abort(struct sock *sk, int err); 1283 1284 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1285 { 1286 rx_opt->dsack = 0; 1287 rx_opt->num_sacks = 0; 1288 } 1289 1290 u32 tcp_default_init_rwnd(u32 mss); 1291 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1292 1293 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1294 { 1295 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1296 struct tcp_sock *tp = tcp_sk(sk); 1297 s32 delta; 1298 1299 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out || 1300 ca_ops->cong_control) 1301 return; 1302 delta = tcp_jiffies32 - tp->lsndtime; 1303 if (delta > inet_csk(sk)->icsk_rto) 1304 tcp_cwnd_restart(sk, delta); 1305 } 1306 1307 /* Determine a window scaling and initial window to offer. */ 1308 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1309 __u32 *window_clamp, int wscale_ok, 1310 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1311 1312 static inline int tcp_win_from_space(int space) 1313 { 1314 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale; 1315 1316 return tcp_adv_win_scale <= 0 ? 1317 (space>>(-tcp_adv_win_scale)) : 1318 space - (space>>tcp_adv_win_scale); 1319 } 1320 1321 /* Note: caller must be prepared to deal with negative returns */ 1322 static inline int tcp_space(const struct sock *sk) 1323 { 1324 return tcp_win_from_space(sk->sk_rcvbuf - 1325 atomic_read(&sk->sk_rmem_alloc)); 1326 } 1327 1328 static inline int tcp_full_space(const struct sock *sk) 1329 { 1330 return tcp_win_from_space(sk->sk_rcvbuf); 1331 } 1332 1333 extern void tcp_openreq_init_rwin(struct request_sock *req, 1334 const struct sock *sk_listener, 1335 const struct dst_entry *dst); 1336 1337 void tcp_enter_memory_pressure(struct sock *sk); 1338 void tcp_leave_memory_pressure(struct sock *sk); 1339 1340 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1341 { 1342 struct net *net = sock_net((struct sock *)tp); 1343 1344 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1345 } 1346 1347 static inline int keepalive_time_when(const struct tcp_sock *tp) 1348 { 1349 struct net *net = sock_net((struct sock *)tp); 1350 1351 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1352 } 1353 1354 static inline int keepalive_probes(const struct tcp_sock *tp) 1355 { 1356 struct net *net = sock_net((struct sock *)tp); 1357 1358 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1359 } 1360 1361 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1362 { 1363 const struct inet_connection_sock *icsk = &tp->inet_conn; 1364 1365 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1366 tcp_jiffies32 - tp->rcv_tstamp); 1367 } 1368 1369 static inline int tcp_fin_time(const struct sock *sk) 1370 { 1371 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1372 const int rto = inet_csk(sk)->icsk_rto; 1373 1374 if (fin_timeout < (rto << 2) - (rto >> 1)) 1375 fin_timeout = (rto << 2) - (rto >> 1); 1376 1377 return fin_timeout; 1378 } 1379 1380 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1381 int paws_win) 1382 { 1383 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1384 return true; 1385 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1386 return true; 1387 /* 1388 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1389 * then following tcp messages have valid values. Ignore 0 value, 1390 * or else 'negative' tsval might forbid us to accept their packets. 1391 */ 1392 if (!rx_opt->ts_recent) 1393 return true; 1394 return false; 1395 } 1396 1397 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1398 int rst) 1399 { 1400 if (tcp_paws_check(rx_opt, 0)) 1401 return false; 1402 1403 /* RST segments are not recommended to carry timestamp, 1404 and, if they do, it is recommended to ignore PAWS because 1405 "their cleanup function should take precedence over timestamps." 1406 Certainly, it is mistake. It is necessary to understand the reasons 1407 of this constraint to relax it: if peer reboots, clock may go 1408 out-of-sync and half-open connections will not be reset. 1409 Actually, the problem would be not existing if all 1410 the implementations followed draft about maintaining clock 1411 via reboots. Linux-2.2 DOES NOT! 1412 1413 However, we can relax time bounds for RST segments to MSL. 1414 */ 1415 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1416 return false; 1417 return true; 1418 } 1419 1420 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1421 int mib_idx, u32 *last_oow_ack_time); 1422 1423 static inline void tcp_mib_init(struct net *net) 1424 { 1425 /* See RFC 2012 */ 1426 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1427 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1428 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1429 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1430 } 1431 1432 /* from STCP */ 1433 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1434 { 1435 tp->lost_skb_hint = NULL; 1436 } 1437 1438 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1439 { 1440 tcp_clear_retrans_hints_partial(tp); 1441 tp->retransmit_skb_hint = NULL; 1442 } 1443 1444 union tcp_md5_addr { 1445 struct in_addr a4; 1446 #if IS_ENABLED(CONFIG_IPV6) 1447 struct in6_addr a6; 1448 #endif 1449 }; 1450 1451 /* - key database */ 1452 struct tcp_md5sig_key { 1453 struct hlist_node node; 1454 u8 keylen; 1455 u8 family; /* AF_INET or AF_INET6 */ 1456 union tcp_md5_addr addr; 1457 u8 prefixlen; 1458 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1459 struct rcu_head rcu; 1460 }; 1461 1462 /* - sock block */ 1463 struct tcp_md5sig_info { 1464 struct hlist_head head; 1465 struct rcu_head rcu; 1466 }; 1467 1468 /* - pseudo header */ 1469 struct tcp4_pseudohdr { 1470 __be32 saddr; 1471 __be32 daddr; 1472 __u8 pad; 1473 __u8 protocol; 1474 __be16 len; 1475 }; 1476 1477 struct tcp6_pseudohdr { 1478 struct in6_addr saddr; 1479 struct in6_addr daddr; 1480 __be32 len; 1481 __be32 protocol; /* including padding */ 1482 }; 1483 1484 union tcp_md5sum_block { 1485 struct tcp4_pseudohdr ip4; 1486 #if IS_ENABLED(CONFIG_IPV6) 1487 struct tcp6_pseudohdr ip6; 1488 #endif 1489 }; 1490 1491 /* - pool: digest algorithm, hash description and scratch buffer */ 1492 struct tcp_md5sig_pool { 1493 struct ahash_request *md5_req; 1494 void *scratch; 1495 }; 1496 1497 /* - functions */ 1498 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1499 const struct sock *sk, const struct sk_buff *skb); 1500 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1501 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1502 gfp_t gfp); 1503 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1504 int family, u8 prefixlen); 1505 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1506 const struct sock *addr_sk); 1507 1508 #ifdef CONFIG_TCP_MD5SIG 1509 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1510 const union tcp_md5_addr *addr, 1511 int family); 1512 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1513 #else 1514 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1515 const union tcp_md5_addr *addr, 1516 int family) 1517 { 1518 return NULL; 1519 } 1520 #define tcp_twsk_md5_key(twsk) NULL 1521 #endif 1522 1523 bool tcp_alloc_md5sig_pool(void); 1524 1525 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1526 static inline void tcp_put_md5sig_pool(void) 1527 { 1528 local_bh_enable(); 1529 } 1530 1531 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1532 unsigned int header_len); 1533 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1534 const struct tcp_md5sig_key *key); 1535 1536 /* From tcp_fastopen.c */ 1537 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1538 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1539 unsigned long *last_syn_loss); 1540 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1541 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1542 u16 try_exp); 1543 struct tcp_fastopen_request { 1544 /* Fast Open cookie. Size 0 means a cookie request */ 1545 struct tcp_fastopen_cookie cookie; 1546 struct msghdr *data; /* data in MSG_FASTOPEN */ 1547 size_t size; 1548 int copied; /* queued in tcp_connect() */ 1549 }; 1550 void tcp_free_fastopen_req(struct tcp_sock *tp); 1551 1552 void tcp_fastopen_ctx_destroy(struct net *net); 1553 int tcp_fastopen_reset_cipher(struct net *net, void *key, unsigned int len); 1554 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1555 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1556 struct request_sock *req, 1557 struct tcp_fastopen_cookie *foc); 1558 void tcp_fastopen_init_key_once(struct net *net); 1559 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1560 struct tcp_fastopen_cookie *cookie); 1561 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1562 #define TCP_FASTOPEN_KEY_LENGTH 16 1563 1564 /* Fastopen key context */ 1565 struct tcp_fastopen_context { 1566 struct crypto_cipher *tfm; 1567 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1568 struct rcu_head rcu; 1569 }; 1570 1571 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1572 void tcp_fastopen_active_disable(struct sock *sk); 1573 bool tcp_fastopen_active_should_disable(struct sock *sk); 1574 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1575 void tcp_fastopen_active_timeout_reset(void); 1576 1577 /* Latencies incurred by various limits for a sender. They are 1578 * chronograph-like stats that are mutually exclusive. 1579 */ 1580 enum tcp_chrono { 1581 TCP_CHRONO_UNSPEC, 1582 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1583 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1584 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1585 __TCP_CHRONO_MAX, 1586 }; 1587 1588 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1589 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1590 1591 /* write queue abstraction */ 1592 static inline void tcp_write_queue_purge(struct sock *sk) 1593 { 1594 struct sk_buff *skb; 1595 1596 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1597 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1598 sk_wmem_free_skb(sk, skb); 1599 sk_mem_reclaim(sk); 1600 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1601 } 1602 1603 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1604 { 1605 return skb_peek(&sk->sk_write_queue); 1606 } 1607 1608 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1609 { 1610 return skb_peek_tail(&sk->sk_write_queue); 1611 } 1612 1613 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1614 const struct sk_buff *skb) 1615 { 1616 return skb_queue_next(&sk->sk_write_queue, skb); 1617 } 1618 1619 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1620 const struct sk_buff *skb) 1621 { 1622 return skb_queue_prev(&sk->sk_write_queue, skb); 1623 } 1624 1625 #define tcp_for_write_queue(skb, sk) \ 1626 skb_queue_walk(&(sk)->sk_write_queue, skb) 1627 1628 #define tcp_for_write_queue_from(skb, sk) \ 1629 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1630 1631 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1632 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1633 1634 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1635 { 1636 return sk->sk_send_head; 1637 } 1638 1639 static inline bool tcp_skb_is_last(const struct sock *sk, 1640 const struct sk_buff *skb) 1641 { 1642 return skb_queue_is_last(&sk->sk_write_queue, skb); 1643 } 1644 1645 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1646 { 1647 if (tcp_skb_is_last(sk, skb)) 1648 sk->sk_send_head = NULL; 1649 else 1650 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1651 } 1652 1653 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1654 { 1655 if (sk->sk_send_head == skb_unlinked) { 1656 sk->sk_send_head = NULL; 1657 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1658 } 1659 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1660 tcp_sk(sk)->highest_sack = NULL; 1661 } 1662 1663 static inline void tcp_init_send_head(struct sock *sk) 1664 { 1665 sk->sk_send_head = NULL; 1666 } 1667 1668 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1669 { 1670 __skb_queue_tail(&sk->sk_write_queue, skb); 1671 } 1672 1673 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1674 { 1675 __tcp_add_write_queue_tail(sk, skb); 1676 1677 /* Queue it, remembering where we must start sending. */ 1678 if (sk->sk_send_head == NULL) { 1679 sk->sk_send_head = skb; 1680 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1681 1682 if (tcp_sk(sk)->highest_sack == NULL) 1683 tcp_sk(sk)->highest_sack = skb; 1684 } 1685 } 1686 1687 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1688 { 1689 __skb_queue_head(&sk->sk_write_queue, skb); 1690 } 1691 1692 /* Insert buff after skb on the write queue of sk. */ 1693 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1694 struct sk_buff *buff, 1695 struct sock *sk) 1696 { 1697 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1698 } 1699 1700 /* Insert new before skb on the write queue of sk. */ 1701 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1702 struct sk_buff *skb, 1703 struct sock *sk) 1704 { 1705 __skb_queue_before(&sk->sk_write_queue, skb, new); 1706 1707 if (sk->sk_send_head == skb) 1708 sk->sk_send_head = new; 1709 } 1710 1711 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1712 { 1713 __skb_unlink(skb, &sk->sk_write_queue); 1714 } 1715 1716 static inline bool tcp_write_queue_empty(struct sock *sk) 1717 { 1718 return skb_queue_empty(&sk->sk_write_queue); 1719 } 1720 1721 static inline void tcp_push_pending_frames(struct sock *sk) 1722 { 1723 if (tcp_send_head(sk)) { 1724 struct tcp_sock *tp = tcp_sk(sk); 1725 1726 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1727 } 1728 } 1729 1730 /* Start sequence of the skb just after the highest skb with SACKed 1731 * bit, valid only if sacked_out > 0 or when the caller has ensured 1732 * validity by itself. 1733 */ 1734 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1735 { 1736 if (!tp->sacked_out) 1737 return tp->snd_una; 1738 1739 if (tp->highest_sack == NULL) 1740 return tp->snd_nxt; 1741 1742 return TCP_SKB_CB(tp->highest_sack)->seq; 1743 } 1744 1745 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1746 { 1747 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1748 tcp_write_queue_next(sk, skb); 1749 } 1750 1751 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1752 { 1753 return tcp_sk(sk)->highest_sack; 1754 } 1755 1756 static inline void tcp_highest_sack_reset(struct sock *sk) 1757 { 1758 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1759 } 1760 1761 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1762 static inline void tcp_highest_sack_combine(struct sock *sk, 1763 struct sk_buff *old, 1764 struct sk_buff *new) 1765 { 1766 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1767 tcp_sk(sk)->highest_sack = new; 1768 } 1769 1770 /* This helper checks if socket has IP_TRANSPARENT set */ 1771 static inline bool inet_sk_transparent(const struct sock *sk) 1772 { 1773 switch (sk->sk_state) { 1774 case TCP_TIME_WAIT: 1775 return inet_twsk(sk)->tw_transparent; 1776 case TCP_NEW_SYN_RECV: 1777 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1778 } 1779 return inet_sk(sk)->transparent; 1780 } 1781 1782 /* Determines whether this is a thin stream (which may suffer from 1783 * increased latency). Used to trigger latency-reducing mechanisms. 1784 */ 1785 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1786 { 1787 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1788 } 1789 1790 /* /proc */ 1791 enum tcp_seq_states { 1792 TCP_SEQ_STATE_LISTENING, 1793 TCP_SEQ_STATE_ESTABLISHED, 1794 }; 1795 1796 int tcp_seq_open(struct inode *inode, struct file *file); 1797 1798 struct tcp_seq_afinfo { 1799 char *name; 1800 sa_family_t family; 1801 const struct file_operations *seq_fops; 1802 struct seq_operations seq_ops; 1803 }; 1804 1805 struct tcp_iter_state { 1806 struct seq_net_private p; 1807 sa_family_t family; 1808 enum tcp_seq_states state; 1809 struct sock *syn_wait_sk; 1810 int bucket, offset, sbucket, num; 1811 loff_t last_pos; 1812 }; 1813 1814 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1815 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1816 1817 extern struct request_sock_ops tcp_request_sock_ops; 1818 extern struct request_sock_ops tcp6_request_sock_ops; 1819 1820 void tcp_v4_destroy_sock(struct sock *sk); 1821 1822 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1823 netdev_features_t features); 1824 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1825 int tcp_gro_complete(struct sk_buff *skb); 1826 1827 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1828 1829 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1830 { 1831 struct net *net = sock_net((struct sock *)tp); 1832 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1833 } 1834 1835 static inline bool tcp_stream_memory_free(const struct sock *sk) 1836 { 1837 const struct tcp_sock *tp = tcp_sk(sk); 1838 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1839 1840 return notsent_bytes < tcp_notsent_lowat(tp); 1841 } 1842 1843 #ifdef CONFIG_PROC_FS 1844 int tcp4_proc_init(void); 1845 void tcp4_proc_exit(void); 1846 #endif 1847 1848 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1849 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1850 const struct tcp_request_sock_ops *af_ops, 1851 struct sock *sk, struct sk_buff *skb); 1852 1853 /* TCP af-specific functions */ 1854 struct tcp_sock_af_ops { 1855 #ifdef CONFIG_TCP_MD5SIG 1856 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1857 const struct sock *addr_sk); 1858 int (*calc_md5_hash)(char *location, 1859 const struct tcp_md5sig_key *md5, 1860 const struct sock *sk, 1861 const struct sk_buff *skb); 1862 int (*md5_parse)(struct sock *sk, 1863 int optname, 1864 char __user *optval, 1865 int optlen); 1866 #endif 1867 }; 1868 1869 struct tcp_request_sock_ops { 1870 u16 mss_clamp; 1871 #ifdef CONFIG_TCP_MD5SIG 1872 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1873 const struct sock *addr_sk); 1874 int (*calc_md5_hash) (char *location, 1875 const struct tcp_md5sig_key *md5, 1876 const struct sock *sk, 1877 const struct sk_buff *skb); 1878 #endif 1879 void (*init_req)(struct request_sock *req, 1880 const struct sock *sk_listener, 1881 struct sk_buff *skb); 1882 #ifdef CONFIG_SYN_COOKIES 1883 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1884 __u16 *mss); 1885 #endif 1886 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1887 const struct request_sock *req); 1888 u32 (*init_seq)(const struct sk_buff *skb); 1889 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1890 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1891 struct flowi *fl, struct request_sock *req, 1892 struct tcp_fastopen_cookie *foc, 1893 enum tcp_synack_type synack_type); 1894 }; 1895 1896 #ifdef CONFIG_SYN_COOKIES 1897 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1898 const struct sock *sk, struct sk_buff *skb, 1899 __u16 *mss) 1900 { 1901 tcp_synq_overflow(sk); 1902 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1903 return ops->cookie_init_seq(skb, mss); 1904 } 1905 #else 1906 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1907 const struct sock *sk, struct sk_buff *skb, 1908 __u16 *mss) 1909 { 1910 return 0; 1911 } 1912 #endif 1913 1914 int tcpv4_offload_init(void); 1915 1916 void tcp_v4_init(void); 1917 void tcp_init(void); 1918 1919 /* tcp_recovery.c */ 1920 extern void tcp_rack_mark_lost(struct sock *sk); 1921 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1922 u64 xmit_time); 1923 extern void tcp_rack_reo_timeout(struct sock *sk); 1924 1925 /* At how many usecs into the future should the RTO fire? */ 1926 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1927 { 1928 const struct sk_buff *skb = tcp_write_queue_head(sk); 1929 u32 rto = inet_csk(sk)->icsk_rto; 1930 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1931 1932 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1933 } 1934 1935 /* 1936 * Save and compile IPv4 options, return a pointer to it 1937 */ 1938 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1939 struct sk_buff *skb) 1940 { 1941 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1942 struct ip_options_rcu *dopt = NULL; 1943 1944 if (opt->optlen) { 1945 int opt_size = sizeof(*dopt) + opt->optlen; 1946 1947 dopt = kmalloc(opt_size, GFP_ATOMIC); 1948 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1949 kfree(dopt); 1950 dopt = NULL; 1951 } 1952 } 1953 return dopt; 1954 } 1955 1956 /* locally generated TCP pure ACKs have skb->truesize == 2 1957 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1958 * This is much faster than dissecting the packet to find out. 1959 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1960 */ 1961 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1962 { 1963 return skb->truesize == 2; 1964 } 1965 1966 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1967 { 1968 skb->truesize = 2; 1969 } 1970 1971 static inline int tcp_inq(struct sock *sk) 1972 { 1973 struct tcp_sock *tp = tcp_sk(sk); 1974 int answ; 1975 1976 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1977 answ = 0; 1978 } else if (sock_flag(sk, SOCK_URGINLINE) || 1979 !tp->urg_data || 1980 before(tp->urg_seq, tp->copied_seq) || 1981 !before(tp->urg_seq, tp->rcv_nxt)) { 1982 1983 answ = tp->rcv_nxt - tp->copied_seq; 1984 1985 /* Subtract 1, if FIN was received */ 1986 if (answ && sock_flag(sk, SOCK_DONE)) 1987 answ--; 1988 } else { 1989 answ = tp->urg_seq - tp->copied_seq; 1990 } 1991 1992 return answ; 1993 } 1994 1995 int tcp_peek_len(struct socket *sock); 1996 1997 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1998 { 1999 u16 segs_in; 2000 2001 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2002 tp->segs_in += segs_in; 2003 if (skb->len > tcp_hdrlen(skb)) 2004 tp->data_segs_in += segs_in; 2005 } 2006 2007 /* 2008 * TCP listen path runs lockless. 2009 * We forced "struct sock" to be const qualified to make sure 2010 * we don't modify one of its field by mistake. 2011 * Here, we increment sk_drops which is an atomic_t, so we can safely 2012 * make sock writable again. 2013 */ 2014 static inline void tcp_listendrop(const struct sock *sk) 2015 { 2016 atomic_inc(&((struct sock *)sk)->sk_drops); 2017 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2018 } 2019 2020 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2021 2022 /* 2023 * Interface for adding Upper Level Protocols over TCP 2024 */ 2025 2026 #define TCP_ULP_NAME_MAX 16 2027 #define TCP_ULP_MAX 128 2028 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2029 2030 struct tcp_ulp_ops { 2031 struct list_head list; 2032 2033 /* initialize ulp */ 2034 int (*init)(struct sock *sk); 2035 /* cleanup ulp */ 2036 void (*release)(struct sock *sk); 2037 2038 char name[TCP_ULP_NAME_MAX]; 2039 struct module *owner; 2040 }; 2041 int tcp_register_ulp(struct tcp_ulp_ops *type); 2042 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2043 int tcp_set_ulp(struct sock *sk, const char *name); 2044 void tcp_get_available_ulp(char *buf, size_t len); 2045 void tcp_cleanup_ulp(struct sock *sk); 2046 2047 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2048 * is < 0, then the BPF op failed (for example if the loaded BPF 2049 * program does not support the chosen operation or there is no BPF 2050 * program loaded). 2051 */ 2052 #ifdef CONFIG_BPF 2053 static inline int tcp_call_bpf(struct sock *sk, int op) 2054 { 2055 struct bpf_sock_ops_kern sock_ops; 2056 int ret; 2057 2058 if (sk_fullsock(sk)) 2059 sock_owned_by_me(sk); 2060 2061 memset(&sock_ops, 0, sizeof(sock_ops)); 2062 sock_ops.sk = sk; 2063 sock_ops.op = op; 2064 2065 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2066 if (ret == 0) 2067 ret = sock_ops.reply; 2068 else 2069 ret = -1; 2070 return ret; 2071 } 2072 #else 2073 static inline int tcp_call_bpf(struct sock *sk, int op) 2074 { 2075 return -EPERM; 2076 } 2077 #endif 2078 2079 static inline u32 tcp_timeout_init(struct sock *sk) 2080 { 2081 int timeout; 2082 2083 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2084 2085 if (timeout <= 0) 2086 timeout = TCP_TIMEOUT_INIT; 2087 return timeout; 2088 } 2089 2090 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2091 { 2092 int rwnd; 2093 2094 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2095 2096 if (rwnd < 0) 2097 rwnd = 0; 2098 return rwnd; 2099 } 2100 2101 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2102 { 2103 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2104 } 2105 #endif /* _TCP_H */ 2106