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/crypto.h> 31 #include <linux/cryptohash.h> 32 #include <linux/kref.h> 33 #include <linux/ktime.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 1024 69 70 /* probing interval, default to 10 minutes as per RFC4821 */ 71 #define TCP_PROBE_INTERVAL 600 72 73 /* Specify interval when tcp mtu probing will stop */ 74 #define TCP_PROBE_THRESHOLD 8 75 76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 77 #define TCP_FASTRETRANS_THRESH 3 78 79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80 #define TCP_MAX_QUICKACKS 16U 81 82 /* urg_data states */ 83 #define TCP_URG_VALID 0x0100 84 #define TCP_URG_NOTYET 0x0200 85 #define TCP_URG_READ 0x0400 86 87 #define TCP_RETR1 3 /* 88 * This is how many retries it does before it 89 * tries to figure out if the gateway is 90 * down. Minimal RFC value is 3; it corresponds 91 * to ~3sec-8min depending on RTO. 92 */ 93 94 #define TCP_RETR2 15 /* 95 * This should take at least 96 * 90 minutes to time out. 97 * RFC1122 says that the limit is 100 sec. 98 * 15 is ~13-30min depending on RTO. 99 */ 100 101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 102 * when active opening a connection. 103 * RFC1122 says the minimum retry MUST 104 * be at least 180secs. Nevertheless 105 * this value is corresponding to 106 * 63secs of retransmission with the 107 * current initial RTO. 108 */ 109 110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 111 * when passive opening a connection. 112 * This is corresponding to 31secs of 113 * retransmission with the current 114 * initial RTO. 115 */ 116 117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 118 * state, about 60 seconds */ 119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 120 /* BSD style FIN_WAIT2 deadlock breaker. 121 * It used to be 3min, new value is 60sec, 122 * to combine FIN-WAIT-2 timeout with 123 * TIME-WAIT timer. 124 */ 125 126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 127 #if HZ >= 100 128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 129 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 130 #else 131 #define TCP_DELACK_MIN 4U 132 #define TCP_ATO_MIN 4U 133 #endif 134 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 135 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 138 * used as a fallback RTO for the 139 * initial data transmission if no 140 * valid RTT sample has been acquired, 141 * most likely due to retrans in 3WHS. 142 */ 143 144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 145 * for local resources. 146 */ 147 148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 150 #define TCP_KEEPALIVE_INTVL (75*HZ) 151 152 #define MAX_TCP_KEEPIDLE 32767 153 #define MAX_TCP_KEEPINTVL 32767 154 #define MAX_TCP_KEEPCNT 127 155 #define MAX_TCP_SYNCNT 127 156 157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 158 159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 161 * after this time. It should be equal 162 * (or greater than) TCP_TIMEWAIT_LEN 163 * to provide reliability equal to one 164 * provided by timewait state. 165 */ 166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 167 * timestamps. It must be less than 168 * minimal timewait lifetime. 169 */ 170 /* 171 * TCP option 172 */ 173 174 #define TCPOPT_NOP 1 /* Padding */ 175 #define TCPOPT_EOL 0 /* End of options */ 176 #define TCPOPT_MSS 2 /* Segment size negotiating */ 177 #define TCPOPT_WINDOW 3 /* Window scaling */ 178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 179 #define TCPOPT_SACK 5 /* SACK Block */ 180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 183 #define TCPOPT_EXP 254 /* Experimental */ 184 /* Magic number to be after the option value for sharing TCP 185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 186 */ 187 #define TCPOPT_FASTOPEN_MAGIC 0xF989 188 189 /* 190 * TCP option lengths 191 */ 192 193 #define TCPOLEN_MSS 4 194 #define TCPOLEN_WINDOW 3 195 #define TCPOLEN_SACK_PERM 2 196 #define TCPOLEN_TIMESTAMP 10 197 #define TCPOLEN_MD5SIG 18 198 #define TCPOLEN_FASTOPEN_BASE 2 199 #define TCPOLEN_EXP_FASTOPEN_BASE 4 200 201 /* But this is what stacks really send out. */ 202 #define TCPOLEN_TSTAMP_ALIGNED 12 203 #define TCPOLEN_WSCALE_ALIGNED 4 204 #define TCPOLEN_SACKPERM_ALIGNED 4 205 #define TCPOLEN_SACK_BASE 2 206 #define TCPOLEN_SACK_BASE_ALIGNED 4 207 #define TCPOLEN_SACK_PERBLOCK 8 208 #define TCPOLEN_MD5SIG_ALIGNED 20 209 #define TCPOLEN_MSS_ALIGNED 4 210 211 /* Flags in tp->nonagle */ 212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 213 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 215 216 /* TCP thin-stream limits */ 217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 218 219 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 220 #define TCP_INIT_CWND 10 221 222 /* Bit Flags for sysctl_tcp_fastopen */ 223 #define TFO_CLIENT_ENABLE 1 224 #define TFO_SERVER_ENABLE 2 225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 226 227 /* Accept SYN data w/o any cookie option */ 228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 229 230 /* Force enable TFO on all listeners, i.e., not requiring the 231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 232 */ 233 #define TFO_SERVER_WO_SOCKOPT1 0x400 234 #define TFO_SERVER_WO_SOCKOPT2 0x800 235 236 extern struct inet_timewait_death_row tcp_death_row; 237 238 /* sysctl variables for tcp */ 239 extern int sysctl_tcp_timestamps; 240 extern int sysctl_tcp_window_scaling; 241 extern int sysctl_tcp_sack; 242 extern int sysctl_tcp_fin_timeout; 243 extern int sysctl_tcp_keepalive_time; 244 extern int sysctl_tcp_keepalive_probes; 245 extern int sysctl_tcp_keepalive_intvl; 246 extern int sysctl_tcp_syn_retries; 247 extern int sysctl_tcp_synack_retries; 248 extern int sysctl_tcp_retries1; 249 extern int sysctl_tcp_retries2; 250 extern int sysctl_tcp_orphan_retries; 251 extern int sysctl_tcp_syncookies; 252 extern int sysctl_tcp_fastopen; 253 extern int sysctl_tcp_retrans_collapse; 254 extern int sysctl_tcp_stdurg; 255 extern int sysctl_tcp_rfc1337; 256 extern int sysctl_tcp_abort_on_overflow; 257 extern int sysctl_tcp_max_orphans; 258 extern int sysctl_tcp_fack; 259 extern int sysctl_tcp_reordering; 260 extern int sysctl_tcp_max_reordering; 261 extern int sysctl_tcp_dsack; 262 extern long sysctl_tcp_mem[3]; 263 extern int sysctl_tcp_wmem[3]; 264 extern int sysctl_tcp_rmem[3]; 265 extern int sysctl_tcp_app_win; 266 extern int sysctl_tcp_adv_win_scale; 267 extern int sysctl_tcp_tw_reuse; 268 extern int sysctl_tcp_frto; 269 extern int sysctl_tcp_low_latency; 270 extern int sysctl_tcp_nometrics_save; 271 extern int sysctl_tcp_moderate_rcvbuf; 272 extern int sysctl_tcp_tso_win_divisor; 273 extern int sysctl_tcp_workaround_signed_windows; 274 extern int sysctl_tcp_slow_start_after_idle; 275 extern int sysctl_tcp_thin_linear_timeouts; 276 extern int sysctl_tcp_thin_dupack; 277 extern int sysctl_tcp_early_retrans; 278 extern int sysctl_tcp_limit_output_bytes; 279 extern int sysctl_tcp_challenge_ack_limit; 280 extern unsigned int sysctl_tcp_notsent_lowat; 281 extern int sysctl_tcp_min_tso_segs; 282 extern int sysctl_tcp_autocorking; 283 extern int sysctl_tcp_invalid_ratelimit; 284 extern int sysctl_tcp_pacing_ss_ratio; 285 extern int sysctl_tcp_pacing_ca_ratio; 286 287 extern atomic_long_t tcp_memory_allocated; 288 extern struct percpu_counter tcp_sockets_allocated; 289 extern int tcp_memory_pressure; 290 291 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 292 static inline bool tcp_under_memory_pressure(const struct sock *sk) 293 { 294 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 295 return !!sk->sk_cgrp->memory_pressure; 296 297 return tcp_memory_pressure; 298 } 299 /* 300 * The next routines deal with comparing 32 bit unsigned ints 301 * and worry about wraparound (automatic with unsigned arithmetic). 302 */ 303 304 static inline bool before(__u32 seq1, __u32 seq2) 305 { 306 return (__s32)(seq1-seq2) < 0; 307 } 308 #define after(seq2, seq1) before(seq1, seq2) 309 310 /* is s2<=s1<=s3 ? */ 311 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 312 { 313 return seq3 - seq2 >= seq1 - seq2; 314 } 315 316 static inline bool tcp_out_of_memory(struct sock *sk) 317 { 318 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 319 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 320 return true; 321 return false; 322 } 323 324 void sk_forced_mem_schedule(struct sock *sk, int size); 325 326 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 327 { 328 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 329 int orphans = percpu_counter_read_positive(ocp); 330 331 if (orphans << shift > sysctl_tcp_max_orphans) { 332 orphans = percpu_counter_sum_positive(ocp); 333 if (orphans << shift > sysctl_tcp_max_orphans) 334 return true; 335 } 336 return false; 337 } 338 339 bool tcp_check_oom(struct sock *sk, int shift); 340 341 342 extern struct proto tcp_prot; 343 344 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 345 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 346 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 347 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 348 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 349 350 void tcp_tasklet_init(void); 351 352 void tcp_v4_err(struct sk_buff *skb, u32); 353 354 void tcp_shutdown(struct sock *sk, int how); 355 356 void tcp_v4_early_demux(struct sk_buff *skb); 357 int tcp_v4_rcv(struct sk_buff *skb); 358 359 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 360 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 361 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 362 int flags); 363 void tcp_release_cb(struct sock *sk); 364 void tcp_wfree(struct sk_buff *skb); 365 void tcp_write_timer_handler(struct sock *sk); 366 void tcp_delack_timer_handler(struct sock *sk); 367 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 368 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 369 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 370 const struct tcphdr *th, unsigned int len); 371 void tcp_rcv_space_adjust(struct sock *sk); 372 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 373 void tcp_twsk_destructor(struct sock *sk); 374 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 375 struct pipe_inode_info *pipe, size_t len, 376 unsigned int flags); 377 378 static inline void tcp_dec_quickack_mode(struct sock *sk, 379 const unsigned int pkts) 380 { 381 struct inet_connection_sock *icsk = inet_csk(sk); 382 383 if (icsk->icsk_ack.quick) { 384 if (pkts >= icsk->icsk_ack.quick) { 385 icsk->icsk_ack.quick = 0; 386 /* Leaving quickack mode we deflate ATO. */ 387 icsk->icsk_ack.ato = TCP_ATO_MIN; 388 } else 389 icsk->icsk_ack.quick -= pkts; 390 } 391 } 392 393 #define TCP_ECN_OK 1 394 #define TCP_ECN_QUEUE_CWR 2 395 #define TCP_ECN_DEMAND_CWR 4 396 #define TCP_ECN_SEEN 8 397 398 enum tcp_tw_status { 399 TCP_TW_SUCCESS = 0, 400 TCP_TW_RST = 1, 401 TCP_TW_ACK = 2, 402 TCP_TW_SYN = 3 403 }; 404 405 406 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 407 struct sk_buff *skb, 408 const struct tcphdr *th); 409 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 410 struct request_sock *req, bool fastopen); 411 int tcp_child_process(struct sock *parent, struct sock *child, 412 struct sk_buff *skb); 413 void tcp_enter_loss(struct sock *sk); 414 void tcp_clear_retrans(struct tcp_sock *tp); 415 void tcp_update_metrics(struct sock *sk); 416 void tcp_init_metrics(struct sock *sk); 417 void tcp_metrics_init(void); 418 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 419 bool paws_check, bool timestamps); 420 bool tcp_remember_stamp(struct sock *sk); 421 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 422 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 423 void tcp_disable_fack(struct tcp_sock *tp); 424 void tcp_close(struct sock *sk, long timeout); 425 void tcp_init_sock(struct sock *sk); 426 unsigned int tcp_poll(struct file *file, struct socket *sock, 427 struct poll_table_struct *wait); 428 int tcp_getsockopt(struct sock *sk, int level, int optname, 429 char __user *optval, int __user *optlen); 430 int tcp_setsockopt(struct sock *sk, int level, int optname, 431 char __user *optval, unsigned int optlen); 432 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 433 char __user *optval, int __user *optlen); 434 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 435 char __user *optval, unsigned int optlen); 436 void tcp_set_keepalive(struct sock *sk, int val); 437 void tcp_syn_ack_timeout(const struct request_sock *req); 438 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 439 int flags, int *addr_len); 440 void tcp_parse_options(const struct sk_buff *skb, 441 struct tcp_options_received *opt_rx, 442 int estab, struct tcp_fastopen_cookie *foc); 443 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 444 445 /* 446 * TCP v4 functions exported for the inet6 API 447 */ 448 449 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 450 void tcp_v4_mtu_reduced(struct sock *sk); 451 void tcp_req_err(struct sock *sk, u32 seq); 452 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 453 struct sock *tcp_create_openreq_child(const struct sock *sk, 454 struct request_sock *req, 455 struct sk_buff *skb); 456 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 457 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 458 struct request_sock *req, 459 struct dst_entry *dst); 460 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 461 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 462 int tcp_connect(struct sock *sk); 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 int tcp_disconnect(struct sock *sk, int flags); 467 468 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 469 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 470 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 471 472 /* From syncookies.c */ 473 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 474 struct request_sock *req, 475 struct dst_entry *dst); 476 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 477 u32 cookie); 478 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 479 #ifdef CONFIG_SYN_COOKIES 480 481 /* Syncookies use a monotonic timer which increments every 60 seconds. 482 * This counter is used both as a hash input and partially encoded into 483 * the cookie value. A cookie is only validated further if the delta 484 * between the current counter value and the encoded one is less than this, 485 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 486 * the counter advances immediately after a cookie is generated). 487 */ 488 #define MAX_SYNCOOKIE_AGE 2 489 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 490 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 491 492 /* syncookies: remember time of last synqueue overflow 493 * But do not dirty this field too often (once per second is enough) 494 * It is racy as we do not hold a lock, but race is very minor. 495 */ 496 static inline void tcp_synq_overflow(const struct sock *sk) 497 { 498 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 499 unsigned long now = jiffies; 500 501 if (time_after(now, last_overflow + HZ)) 502 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 503 } 504 505 /* syncookies: no recent synqueue overflow on this listening socket? */ 506 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 507 { 508 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 509 510 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 511 } 512 513 static inline u32 tcp_cookie_time(void) 514 { 515 u64 val = get_jiffies_64(); 516 517 do_div(val, TCP_SYNCOOKIE_PERIOD); 518 return val; 519 } 520 521 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 522 u16 *mssp); 523 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 524 __u32 cookie_init_timestamp(struct request_sock *req); 525 bool cookie_timestamp_decode(struct tcp_options_received *opt); 526 bool cookie_ecn_ok(const struct tcp_options_received *opt, 527 const struct net *net, const struct dst_entry *dst); 528 529 /* From net/ipv6/syncookies.c */ 530 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 531 u32 cookie); 532 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 533 534 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 535 const struct tcphdr *th, u16 *mssp); 536 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 537 #endif 538 /* tcp_output.c */ 539 540 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 541 int nonagle); 542 bool tcp_may_send_now(struct sock *sk); 543 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 544 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 545 void tcp_retransmit_timer(struct sock *sk); 546 void tcp_xmit_retransmit_queue(struct sock *); 547 void tcp_simple_retransmit(struct sock *); 548 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 549 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 550 551 void tcp_send_probe0(struct sock *); 552 void tcp_send_partial(struct sock *); 553 int tcp_write_wakeup(struct sock *, int mib); 554 void tcp_send_fin(struct sock *sk); 555 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 556 int tcp_send_synack(struct sock *); 557 void tcp_push_one(struct sock *, unsigned int mss_now); 558 void tcp_send_ack(struct sock *sk); 559 void tcp_send_delayed_ack(struct sock *sk); 560 void tcp_send_loss_probe(struct sock *sk); 561 bool tcp_schedule_loss_probe(struct sock *sk); 562 563 /* tcp_input.c */ 564 void tcp_resume_early_retransmit(struct sock *sk); 565 void tcp_rearm_rto(struct sock *sk); 566 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 567 void tcp_reset(struct sock *sk); 568 569 /* tcp_timer.c */ 570 void tcp_init_xmit_timers(struct sock *); 571 static inline void tcp_clear_xmit_timers(struct sock *sk) 572 { 573 inet_csk_clear_xmit_timers(sk); 574 } 575 576 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 577 unsigned int tcp_current_mss(struct sock *sk); 578 579 /* Bound MSS / TSO packet size with the half of the window */ 580 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 581 { 582 int cutoff; 583 584 /* When peer uses tiny windows, there is no use in packetizing 585 * to sub-MSS pieces for the sake of SWS or making sure there 586 * are enough packets in the pipe for fast recovery. 587 * 588 * On the other hand, for extremely large MSS devices, handling 589 * smaller than MSS windows in this way does make sense. 590 */ 591 if (tp->max_window >= 512) 592 cutoff = (tp->max_window >> 1); 593 else 594 cutoff = tp->max_window; 595 596 if (cutoff && pktsize > cutoff) 597 return max_t(int, cutoff, 68U - tp->tcp_header_len); 598 else 599 return pktsize; 600 } 601 602 /* tcp.c */ 603 void tcp_get_info(struct sock *, struct tcp_info *); 604 605 /* Read 'sendfile()'-style from a TCP socket */ 606 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 607 unsigned int, size_t); 608 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 609 sk_read_actor_t recv_actor); 610 611 void tcp_initialize_rcv_mss(struct sock *sk); 612 613 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 614 int tcp_mss_to_mtu(struct sock *sk, int mss); 615 void tcp_mtup_init(struct sock *sk); 616 void tcp_init_buffer_space(struct sock *sk); 617 618 static inline void tcp_bound_rto(const struct sock *sk) 619 { 620 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 621 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 622 } 623 624 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 625 { 626 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 627 } 628 629 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 630 { 631 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 632 ntohl(TCP_FLAG_ACK) | 633 snd_wnd); 634 } 635 636 static inline void tcp_fast_path_on(struct tcp_sock *tp) 637 { 638 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 639 } 640 641 static inline void tcp_fast_path_check(struct sock *sk) 642 { 643 struct tcp_sock *tp = tcp_sk(sk); 644 645 if (skb_queue_empty(&tp->out_of_order_queue) && 646 tp->rcv_wnd && 647 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 648 !tp->urg_data) 649 tcp_fast_path_on(tp); 650 } 651 652 /* Compute the actual rto_min value */ 653 static inline u32 tcp_rto_min(struct sock *sk) 654 { 655 const struct dst_entry *dst = __sk_dst_get(sk); 656 u32 rto_min = TCP_RTO_MIN; 657 658 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 659 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 660 return rto_min; 661 } 662 663 static inline u32 tcp_rto_min_us(struct sock *sk) 664 { 665 return jiffies_to_usecs(tcp_rto_min(sk)); 666 } 667 668 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 669 { 670 return dst_metric_locked(dst, RTAX_CC_ALGO); 671 } 672 673 /* Compute the actual receive window we are currently advertising. 674 * Rcv_nxt can be after the window if our peer push more data 675 * than the offered window. 676 */ 677 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 678 { 679 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 680 681 if (win < 0) 682 win = 0; 683 return (u32) win; 684 } 685 686 /* Choose a new window, without checks for shrinking, and without 687 * scaling applied to the result. The caller does these things 688 * if necessary. This is a "raw" window selection. 689 */ 690 u32 __tcp_select_window(struct sock *sk); 691 692 void tcp_send_window_probe(struct sock *sk); 693 694 /* TCP timestamps are only 32-bits, this causes a slight 695 * complication on 64-bit systems since we store a snapshot 696 * of jiffies in the buffer control blocks below. We decided 697 * to use only the low 32-bits of jiffies and hide the ugly 698 * casts with the following macro. 699 */ 700 #define tcp_time_stamp ((__u32)(jiffies)) 701 702 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 703 { 704 return skb->skb_mstamp.stamp_jiffies; 705 } 706 707 708 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 709 710 #define TCPHDR_FIN 0x01 711 #define TCPHDR_SYN 0x02 712 #define TCPHDR_RST 0x04 713 #define TCPHDR_PSH 0x08 714 #define TCPHDR_ACK 0x10 715 #define TCPHDR_URG 0x20 716 #define TCPHDR_ECE 0x40 717 #define TCPHDR_CWR 0x80 718 719 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 720 721 /* This is what the send packet queuing engine uses to pass 722 * TCP per-packet control information to the transmission code. 723 * We also store the host-order sequence numbers in here too. 724 * This is 44 bytes if IPV6 is enabled. 725 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 726 */ 727 struct tcp_skb_cb { 728 __u32 seq; /* Starting sequence number */ 729 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 730 union { 731 /* Note : tcp_tw_isn is used in input path only 732 * (isn chosen by tcp_timewait_state_process()) 733 * 734 * tcp_gso_segs/size are used in write queue only, 735 * cf tcp_skb_pcount()/tcp_skb_mss() 736 */ 737 __u32 tcp_tw_isn; 738 struct { 739 u16 tcp_gso_segs; 740 u16 tcp_gso_size; 741 }; 742 }; 743 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 744 745 __u8 sacked; /* State flags for SACK/FACK. */ 746 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 747 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 748 #define TCPCB_LOST 0x04 /* SKB is lost */ 749 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 750 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 751 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 752 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 753 TCPCB_REPAIRED) 754 755 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 756 /* 1 byte hole */ 757 __u32 ack_seq; /* Sequence number ACK'd */ 758 union { 759 struct inet_skb_parm h4; 760 #if IS_ENABLED(CONFIG_IPV6) 761 struct inet6_skb_parm h6; 762 #endif 763 } header; /* For incoming frames */ 764 }; 765 766 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 767 768 769 #if IS_ENABLED(CONFIG_IPV6) 770 /* This is the variant of inet6_iif() that must be used by TCP, 771 * as TCP moves IP6CB into a different location in skb->cb[] 772 */ 773 static inline int tcp_v6_iif(const struct sk_buff *skb) 774 { 775 return TCP_SKB_CB(skb)->header.h6.iif; 776 } 777 #endif 778 779 /* Due to TSO, an SKB can be composed of multiple actual 780 * packets. To keep these tracked properly, we use this. 781 */ 782 static inline int tcp_skb_pcount(const struct sk_buff *skb) 783 { 784 return TCP_SKB_CB(skb)->tcp_gso_segs; 785 } 786 787 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 788 { 789 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 790 } 791 792 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 793 { 794 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 795 } 796 797 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 798 static inline int tcp_skb_mss(const struct sk_buff *skb) 799 { 800 return TCP_SKB_CB(skb)->tcp_gso_size; 801 } 802 803 /* Events passed to congestion control interface */ 804 enum tcp_ca_event { 805 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 806 CA_EVENT_CWND_RESTART, /* congestion window restart */ 807 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 808 CA_EVENT_LOSS, /* loss timeout */ 809 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 810 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 811 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 812 CA_EVENT_NON_DELAYED_ACK, 813 }; 814 815 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 816 enum tcp_ca_ack_event_flags { 817 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 818 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 819 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 820 }; 821 822 /* 823 * Interface for adding new TCP congestion control handlers 824 */ 825 #define TCP_CA_NAME_MAX 16 826 #define TCP_CA_MAX 128 827 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 828 829 #define TCP_CA_UNSPEC 0 830 831 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 832 #define TCP_CONG_NON_RESTRICTED 0x1 833 /* Requires ECN/ECT set on all packets */ 834 #define TCP_CONG_NEEDS_ECN 0x2 835 836 union tcp_cc_info; 837 838 struct tcp_congestion_ops { 839 struct list_head list; 840 u32 key; 841 u32 flags; 842 843 /* initialize private data (optional) */ 844 void (*init)(struct sock *sk); 845 /* cleanup private data (optional) */ 846 void (*release)(struct sock *sk); 847 848 /* return slow start threshold (required) */ 849 u32 (*ssthresh)(struct sock *sk); 850 /* do new cwnd calculation (required) */ 851 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 852 /* call before changing ca_state (optional) */ 853 void (*set_state)(struct sock *sk, u8 new_state); 854 /* call when cwnd event occurs (optional) */ 855 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 856 /* call when ack arrives (optional) */ 857 void (*in_ack_event)(struct sock *sk, u32 flags); 858 /* new value of cwnd after loss (optional) */ 859 u32 (*undo_cwnd)(struct sock *sk); 860 /* hook for packet ack accounting (optional) */ 861 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 862 /* get info for inet_diag (optional) */ 863 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 864 union tcp_cc_info *info); 865 866 char name[TCP_CA_NAME_MAX]; 867 struct module *owner; 868 }; 869 870 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 871 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 872 873 void tcp_assign_congestion_control(struct sock *sk); 874 void tcp_init_congestion_control(struct sock *sk); 875 void tcp_cleanup_congestion_control(struct sock *sk); 876 int tcp_set_default_congestion_control(const char *name); 877 void tcp_get_default_congestion_control(char *name); 878 void tcp_get_available_congestion_control(char *buf, size_t len); 879 void tcp_get_allowed_congestion_control(char *buf, size_t len); 880 int tcp_set_allowed_congestion_control(char *allowed); 881 int tcp_set_congestion_control(struct sock *sk, const char *name); 882 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 883 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 884 885 u32 tcp_reno_ssthresh(struct sock *sk); 886 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 887 extern struct tcp_congestion_ops tcp_reno; 888 889 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 890 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 891 #ifdef CONFIG_INET 892 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 893 #else 894 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 895 { 896 return NULL; 897 } 898 #endif 899 900 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 901 { 902 const struct inet_connection_sock *icsk = inet_csk(sk); 903 904 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 905 } 906 907 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 908 { 909 struct inet_connection_sock *icsk = inet_csk(sk); 910 911 if (icsk->icsk_ca_ops->set_state) 912 icsk->icsk_ca_ops->set_state(sk, ca_state); 913 icsk->icsk_ca_state = ca_state; 914 } 915 916 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 917 { 918 const struct inet_connection_sock *icsk = inet_csk(sk); 919 920 if (icsk->icsk_ca_ops->cwnd_event) 921 icsk->icsk_ca_ops->cwnd_event(sk, event); 922 } 923 924 /* These functions determine how the current flow behaves in respect of SACK 925 * handling. SACK is negotiated with the peer, and therefore it can vary 926 * between different flows. 927 * 928 * tcp_is_sack - SACK enabled 929 * tcp_is_reno - No SACK 930 * tcp_is_fack - FACK enabled, implies SACK enabled 931 */ 932 static inline int tcp_is_sack(const struct tcp_sock *tp) 933 { 934 return tp->rx_opt.sack_ok; 935 } 936 937 static inline bool tcp_is_reno(const struct tcp_sock *tp) 938 { 939 return !tcp_is_sack(tp); 940 } 941 942 static inline bool tcp_is_fack(const struct tcp_sock *tp) 943 { 944 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 945 } 946 947 static inline void tcp_enable_fack(struct tcp_sock *tp) 948 { 949 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 950 } 951 952 /* TCP early-retransmit (ER) is similar to but more conservative than 953 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 954 */ 955 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 956 { 957 tp->do_early_retrans = sysctl_tcp_early_retrans && 958 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 959 sysctl_tcp_reordering == 3; 960 } 961 962 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 963 { 964 tp->do_early_retrans = 0; 965 } 966 967 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 968 { 969 return tp->sacked_out + tp->lost_out; 970 } 971 972 /* This determines how many packets are "in the network" to the best 973 * of our knowledge. In many cases it is conservative, but where 974 * detailed information is available from the receiver (via SACK 975 * blocks etc.) we can make more aggressive calculations. 976 * 977 * Use this for decisions involving congestion control, use just 978 * tp->packets_out to determine if the send queue is empty or not. 979 * 980 * Read this equation as: 981 * 982 * "Packets sent once on transmission queue" MINUS 983 * "Packets left network, but not honestly ACKed yet" PLUS 984 * "Packets fast retransmitted" 985 */ 986 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 987 { 988 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 989 } 990 991 #define TCP_INFINITE_SSTHRESH 0x7fffffff 992 993 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 994 { 995 return tp->snd_cwnd < tp->snd_ssthresh; 996 } 997 998 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 999 { 1000 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1001 } 1002 1003 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1004 { 1005 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1006 (1 << inet_csk(sk)->icsk_ca_state); 1007 } 1008 1009 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1010 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1011 * ssthresh. 1012 */ 1013 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1014 { 1015 const struct tcp_sock *tp = tcp_sk(sk); 1016 1017 if (tcp_in_cwnd_reduction(sk)) 1018 return tp->snd_ssthresh; 1019 else 1020 return max(tp->snd_ssthresh, 1021 ((tp->snd_cwnd >> 1) + 1022 (tp->snd_cwnd >> 2))); 1023 } 1024 1025 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1026 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1027 1028 void tcp_enter_cwr(struct sock *sk); 1029 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1030 1031 /* The maximum number of MSS of available cwnd for which TSO defers 1032 * sending if not using sysctl_tcp_tso_win_divisor. 1033 */ 1034 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1035 { 1036 return 3; 1037 } 1038 1039 /* Slow start with delack produces 3 packets of burst, so that 1040 * it is safe "de facto". This will be the default - same as 1041 * the default reordering threshold - but if reordering increases, 1042 * we must be able to allow cwnd to burst at least this much in order 1043 * to not pull it back when holes are filled. 1044 */ 1045 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1046 { 1047 return tp->reordering; 1048 } 1049 1050 /* Returns end sequence number of the receiver's advertised window */ 1051 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1052 { 1053 return tp->snd_una + tp->snd_wnd; 1054 } 1055 1056 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1057 * flexible approach. The RFC suggests cwnd should not be raised unless 1058 * it was fully used previously. And that's exactly what we do in 1059 * congestion avoidance mode. But in slow start we allow cwnd to grow 1060 * as long as the application has used half the cwnd. 1061 * Example : 1062 * cwnd is 10 (IW10), but application sends 9 frames. 1063 * We allow cwnd to reach 18 when all frames are ACKed. 1064 * This check is safe because it's as aggressive as slow start which already 1065 * risks 100% overshoot. The advantage is that we discourage application to 1066 * either send more filler packets or data to artificially blow up the cwnd 1067 * usage, and allow application-limited process to probe bw more aggressively. 1068 */ 1069 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1070 { 1071 const struct tcp_sock *tp = tcp_sk(sk); 1072 1073 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1074 if (tcp_in_slow_start(tp)) 1075 return tp->snd_cwnd < 2 * tp->max_packets_out; 1076 1077 return tp->is_cwnd_limited; 1078 } 1079 1080 /* Something is really bad, we could not queue an additional packet, 1081 * because qdisc is full or receiver sent a 0 window. 1082 * We do not want to add fuel to the fire, or abort too early, 1083 * so make sure the timer we arm now is at least 200ms in the future, 1084 * regardless of current icsk_rto value (as it could be ~2ms) 1085 */ 1086 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1087 { 1088 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1089 } 1090 1091 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1092 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1093 unsigned long max_when) 1094 { 1095 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1096 1097 return (unsigned long)min_t(u64, when, max_when); 1098 } 1099 1100 static inline void tcp_check_probe_timer(struct sock *sk) 1101 { 1102 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1103 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1104 tcp_probe0_base(sk), TCP_RTO_MAX); 1105 } 1106 1107 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1108 { 1109 tp->snd_wl1 = seq; 1110 } 1111 1112 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1113 { 1114 tp->snd_wl1 = seq; 1115 } 1116 1117 /* 1118 * Calculate(/check) TCP checksum 1119 */ 1120 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1121 __be32 daddr, __wsum base) 1122 { 1123 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1124 } 1125 1126 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1127 { 1128 return __skb_checksum_complete(skb); 1129 } 1130 1131 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1132 { 1133 return !skb_csum_unnecessary(skb) && 1134 __tcp_checksum_complete(skb); 1135 } 1136 1137 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1138 1139 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1140 { 1141 tp->ucopy.task = NULL; 1142 tp->ucopy.len = 0; 1143 tp->ucopy.memory = 0; 1144 skb_queue_head_init(&tp->ucopy.prequeue); 1145 } 1146 1147 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1148 1149 #undef STATE_TRACE 1150 1151 #ifdef STATE_TRACE 1152 static const char *statename[]={ 1153 "Unused","Established","Syn Sent","Syn Recv", 1154 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1155 "Close Wait","Last ACK","Listen","Closing" 1156 }; 1157 #endif 1158 void tcp_set_state(struct sock *sk, int state); 1159 1160 void tcp_done(struct sock *sk); 1161 1162 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1163 { 1164 rx_opt->dsack = 0; 1165 rx_opt->num_sacks = 0; 1166 } 1167 1168 u32 tcp_default_init_rwnd(u32 mss); 1169 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1170 1171 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1172 { 1173 struct tcp_sock *tp = tcp_sk(sk); 1174 s32 delta; 1175 1176 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) 1177 return; 1178 delta = tcp_time_stamp - tp->lsndtime; 1179 if (delta > inet_csk(sk)->icsk_rto) 1180 tcp_cwnd_restart(sk, delta); 1181 } 1182 1183 /* Determine a window scaling and initial window to offer. */ 1184 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1185 __u32 *window_clamp, int wscale_ok, 1186 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1187 1188 static inline int tcp_win_from_space(int space) 1189 { 1190 return sysctl_tcp_adv_win_scale<=0 ? 1191 (space>>(-sysctl_tcp_adv_win_scale)) : 1192 space - (space>>sysctl_tcp_adv_win_scale); 1193 } 1194 1195 /* Note: caller must be prepared to deal with negative returns */ 1196 static inline int tcp_space(const struct sock *sk) 1197 { 1198 return tcp_win_from_space(sk->sk_rcvbuf - 1199 atomic_read(&sk->sk_rmem_alloc)); 1200 } 1201 1202 static inline int tcp_full_space(const struct sock *sk) 1203 { 1204 return tcp_win_from_space(sk->sk_rcvbuf); 1205 } 1206 1207 extern void tcp_openreq_init_rwin(struct request_sock *req, 1208 const struct sock *sk_listener, 1209 const struct dst_entry *dst); 1210 1211 void tcp_enter_memory_pressure(struct sock *sk); 1212 1213 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1214 { 1215 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1216 } 1217 1218 static inline int keepalive_time_when(const struct tcp_sock *tp) 1219 { 1220 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1221 } 1222 1223 static inline int keepalive_probes(const struct tcp_sock *tp) 1224 { 1225 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1226 } 1227 1228 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1229 { 1230 const struct inet_connection_sock *icsk = &tp->inet_conn; 1231 1232 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1233 tcp_time_stamp - tp->rcv_tstamp); 1234 } 1235 1236 static inline int tcp_fin_time(const struct sock *sk) 1237 { 1238 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1239 const int rto = inet_csk(sk)->icsk_rto; 1240 1241 if (fin_timeout < (rto << 2) - (rto >> 1)) 1242 fin_timeout = (rto << 2) - (rto >> 1); 1243 1244 return fin_timeout; 1245 } 1246 1247 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1248 int paws_win) 1249 { 1250 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1251 return true; 1252 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1253 return true; 1254 /* 1255 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1256 * then following tcp messages have valid values. Ignore 0 value, 1257 * or else 'negative' tsval might forbid us to accept their packets. 1258 */ 1259 if (!rx_opt->ts_recent) 1260 return true; 1261 return false; 1262 } 1263 1264 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1265 int rst) 1266 { 1267 if (tcp_paws_check(rx_opt, 0)) 1268 return false; 1269 1270 /* RST segments are not recommended to carry timestamp, 1271 and, if they do, it is recommended to ignore PAWS because 1272 "their cleanup function should take precedence over timestamps." 1273 Certainly, it is mistake. It is necessary to understand the reasons 1274 of this constraint to relax it: if peer reboots, clock may go 1275 out-of-sync and half-open connections will not be reset. 1276 Actually, the problem would be not existing if all 1277 the implementations followed draft about maintaining clock 1278 via reboots. Linux-2.2 DOES NOT! 1279 1280 However, we can relax time bounds for RST segments to MSL. 1281 */ 1282 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1283 return false; 1284 return true; 1285 } 1286 1287 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1288 int mib_idx, u32 *last_oow_ack_time); 1289 1290 static inline void tcp_mib_init(struct net *net) 1291 { 1292 /* See RFC 2012 */ 1293 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1294 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1295 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1296 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1297 } 1298 1299 /* from STCP */ 1300 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1301 { 1302 tp->lost_skb_hint = NULL; 1303 } 1304 1305 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1306 { 1307 tcp_clear_retrans_hints_partial(tp); 1308 tp->retransmit_skb_hint = NULL; 1309 } 1310 1311 /* MD5 Signature */ 1312 struct crypto_hash; 1313 1314 union tcp_md5_addr { 1315 struct in_addr a4; 1316 #if IS_ENABLED(CONFIG_IPV6) 1317 struct in6_addr a6; 1318 #endif 1319 }; 1320 1321 /* - key database */ 1322 struct tcp_md5sig_key { 1323 struct hlist_node node; 1324 u8 keylen; 1325 u8 family; /* AF_INET or AF_INET6 */ 1326 union tcp_md5_addr addr; 1327 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1328 struct rcu_head rcu; 1329 }; 1330 1331 /* - sock block */ 1332 struct tcp_md5sig_info { 1333 struct hlist_head head; 1334 struct rcu_head rcu; 1335 }; 1336 1337 /* - pseudo header */ 1338 struct tcp4_pseudohdr { 1339 __be32 saddr; 1340 __be32 daddr; 1341 __u8 pad; 1342 __u8 protocol; 1343 __be16 len; 1344 }; 1345 1346 struct tcp6_pseudohdr { 1347 struct in6_addr saddr; 1348 struct in6_addr daddr; 1349 __be32 len; 1350 __be32 protocol; /* including padding */ 1351 }; 1352 1353 union tcp_md5sum_block { 1354 struct tcp4_pseudohdr ip4; 1355 #if IS_ENABLED(CONFIG_IPV6) 1356 struct tcp6_pseudohdr ip6; 1357 #endif 1358 }; 1359 1360 /* - pool: digest algorithm, hash description and scratch buffer */ 1361 struct tcp_md5sig_pool { 1362 struct hash_desc md5_desc; 1363 union tcp_md5sum_block md5_blk; 1364 }; 1365 1366 /* - functions */ 1367 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1368 const struct sock *sk, const struct sk_buff *skb); 1369 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1370 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1371 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1372 int family); 1373 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1374 const struct sock *addr_sk); 1375 1376 #ifdef CONFIG_TCP_MD5SIG 1377 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1378 const union tcp_md5_addr *addr, 1379 int family); 1380 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1381 #else 1382 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1383 const union tcp_md5_addr *addr, 1384 int family) 1385 { 1386 return NULL; 1387 } 1388 #define tcp_twsk_md5_key(twsk) NULL 1389 #endif 1390 1391 bool tcp_alloc_md5sig_pool(void); 1392 1393 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1394 static inline void tcp_put_md5sig_pool(void) 1395 { 1396 local_bh_enable(); 1397 } 1398 1399 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1400 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1401 unsigned int header_len); 1402 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1403 const struct tcp_md5sig_key *key); 1404 1405 /* From tcp_fastopen.c */ 1406 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1407 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1408 unsigned long *last_syn_loss); 1409 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1410 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1411 u16 try_exp); 1412 struct tcp_fastopen_request { 1413 /* Fast Open cookie. Size 0 means a cookie request */ 1414 struct tcp_fastopen_cookie cookie; 1415 struct msghdr *data; /* data in MSG_FASTOPEN */ 1416 size_t size; 1417 int copied; /* queued in tcp_connect() */ 1418 }; 1419 void tcp_free_fastopen_req(struct tcp_sock *tp); 1420 1421 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1422 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1423 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1424 struct request_sock *req, 1425 struct tcp_fastopen_cookie *foc, 1426 struct dst_entry *dst); 1427 void tcp_fastopen_init_key_once(bool publish); 1428 #define TCP_FASTOPEN_KEY_LENGTH 16 1429 1430 /* Fastopen key context */ 1431 struct tcp_fastopen_context { 1432 struct crypto_cipher *tfm; 1433 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1434 struct rcu_head rcu; 1435 }; 1436 1437 /* write queue abstraction */ 1438 static inline void tcp_write_queue_purge(struct sock *sk) 1439 { 1440 struct sk_buff *skb; 1441 1442 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1443 sk_wmem_free_skb(sk, skb); 1444 sk_mem_reclaim(sk); 1445 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1446 } 1447 1448 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1449 { 1450 return skb_peek(&sk->sk_write_queue); 1451 } 1452 1453 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1454 { 1455 return skb_peek_tail(&sk->sk_write_queue); 1456 } 1457 1458 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1459 const struct sk_buff *skb) 1460 { 1461 return skb_queue_next(&sk->sk_write_queue, skb); 1462 } 1463 1464 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1465 const struct sk_buff *skb) 1466 { 1467 return skb_queue_prev(&sk->sk_write_queue, skb); 1468 } 1469 1470 #define tcp_for_write_queue(skb, sk) \ 1471 skb_queue_walk(&(sk)->sk_write_queue, skb) 1472 1473 #define tcp_for_write_queue_from(skb, sk) \ 1474 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1475 1476 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1477 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1478 1479 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1480 { 1481 return sk->sk_send_head; 1482 } 1483 1484 static inline bool tcp_skb_is_last(const struct sock *sk, 1485 const struct sk_buff *skb) 1486 { 1487 return skb_queue_is_last(&sk->sk_write_queue, skb); 1488 } 1489 1490 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1491 { 1492 if (tcp_skb_is_last(sk, skb)) 1493 sk->sk_send_head = NULL; 1494 else 1495 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1496 } 1497 1498 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1499 { 1500 if (sk->sk_send_head == skb_unlinked) 1501 sk->sk_send_head = NULL; 1502 } 1503 1504 static inline void tcp_init_send_head(struct sock *sk) 1505 { 1506 sk->sk_send_head = NULL; 1507 } 1508 1509 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1510 { 1511 __skb_queue_tail(&sk->sk_write_queue, skb); 1512 } 1513 1514 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1515 { 1516 __tcp_add_write_queue_tail(sk, skb); 1517 1518 /* Queue it, remembering where we must start sending. */ 1519 if (sk->sk_send_head == NULL) { 1520 sk->sk_send_head = skb; 1521 1522 if (tcp_sk(sk)->highest_sack == NULL) 1523 tcp_sk(sk)->highest_sack = skb; 1524 } 1525 } 1526 1527 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1528 { 1529 __skb_queue_head(&sk->sk_write_queue, skb); 1530 } 1531 1532 /* Insert buff after skb on the write queue of sk. */ 1533 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1534 struct sk_buff *buff, 1535 struct sock *sk) 1536 { 1537 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1538 } 1539 1540 /* Insert new before skb on the write queue of sk. */ 1541 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1542 struct sk_buff *skb, 1543 struct sock *sk) 1544 { 1545 __skb_queue_before(&sk->sk_write_queue, skb, new); 1546 1547 if (sk->sk_send_head == skb) 1548 sk->sk_send_head = new; 1549 } 1550 1551 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1552 { 1553 __skb_unlink(skb, &sk->sk_write_queue); 1554 } 1555 1556 static inline bool tcp_write_queue_empty(struct sock *sk) 1557 { 1558 return skb_queue_empty(&sk->sk_write_queue); 1559 } 1560 1561 static inline void tcp_push_pending_frames(struct sock *sk) 1562 { 1563 if (tcp_send_head(sk)) { 1564 struct tcp_sock *tp = tcp_sk(sk); 1565 1566 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1567 } 1568 } 1569 1570 /* Start sequence of the skb just after the highest skb with SACKed 1571 * bit, valid only if sacked_out > 0 or when the caller has ensured 1572 * validity by itself. 1573 */ 1574 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1575 { 1576 if (!tp->sacked_out) 1577 return tp->snd_una; 1578 1579 if (tp->highest_sack == NULL) 1580 return tp->snd_nxt; 1581 1582 return TCP_SKB_CB(tp->highest_sack)->seq; 1583 } 1584 1585 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1586 { 1587 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1588 tcp_write_queue_next(sk, skb); 1589 } 1590 1591 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1592 { 1593 return tcp_sk(sk)->highest_sack; 1594 } 1595 1596 static inline void tcp_highest_sack_reset(struct sock *sk) 1597 { 1598 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1599 } 1600 1601 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1602 static inline void tcp_highest_sack_combine(struct sock *sk, 1603 struct sk_buff *old, 1604 struct sk_buff *new) 1605 { 1606 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1607 tcp_sk(sk)->highest_sack = new; 1608 } 1609 1610 /* Determines whether this is a thin stream (which may suffer from 1611 * increased latency). Used to trigger latency-reducing mechanisms. 1612 */ 1613 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1614 { 1615 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1616 } 1617 1618 /* /proc */ 1619 enum tcp_seq_states { 1620 TCP_SEQ_STATE_LISTENING, 1621 TCP_SEQ_STATE_OPENREQ, 1622 TCP_SEQ_STATE_ESTABLISHED, 1623 }; 1624 1625 int tcp_seq_open(struct inode *inode, struct file *file); 1626 1627 struct tcp_seq_afinfo { 1628 char *name; 1629 sa_family_t family; 1630 const struct file_operations *seq_fops; 1631 struct seq_operations seq_ops; 1632 }; 1633 1634 struct tcp_iter_state { 1635 struct seq_net_private p; 1636 sa_family_t family; 1637 enum tcp_seq_states state; 1638 struct sock *syn_wait_sk; 1639 int bucket, offset, sbucket, num; 1640 kuid_t uid; 1641 loff_t last_pos; 1642 }; 1643 1644 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1645 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1646 1647 extern struct request_sock_ops tcp_request_sock_ops; 1648 extern struct request_sock_ops tcp6_request_sock_ops; 1649 1650 void tcp_v4_destroy_sock(struct sock *sk); 1651 1652 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1653 netdev_features_t features); 1654 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1655 int tcp_gro_complete(struct sk_buff *skb); 1656 1657 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1658 1659 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1660 { 1661 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1662 } 1663 1664 static inline bool tcp_stream_memory_free(const struct sock *sk) 1665 { 1666 const struct tcp_sock *tp = tcp_sk(sk); 1667 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1668 1669 return notsent_bytes < tcp_notsent_lowat(tp); 1670 } 1671 1672 #ifdef CONFIG_PROC_FS 1673 int tcp4_proc_init(void); 1674 void tcp4_proc_exit(void); 1675 #endif 1676 1677 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1678 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1679 const struct tcp_request_sock_ops *af_ops, 1680 struct sock *sk, struct sk_buff *skb); 1681 1682 /* TCP af-specific functions */ 1683 struct tcp_sock_af_ops { 1684 #ifdef CONFIG_TCP_MD5SIG 1685 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1686 const struct sock *addr_sk); 1687 int (*calc_md5_hash)(char *location, 1688 const struct tcp_md5sig_key *md5, 1689 const struct sock *sk, 1690 const struct sk_buff *skb); 1691 int (*md5_parse)(struct sock *sk, 1692 char __user *optval, 1693 int optlen); 1694 #endif 1695 }; 1696 1697 struct tcp_request_sock_ops { 1698 u16 mss_clamp; 1699 #ifdef CONFIG_TCP_MD5SIG 1700 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1701 const struct sock *addr_sk); 1702 int (*calc_md5_hash) (char *location, 1703 const struct tcp_md5sig_key *md5, 1704 const struct sock *sk, 1705 const struct sk_buff *skb); 1706 #endif 1707 void (*init_req)(struct request_sock *req, 1708 const struct sock *sk_listener, 1709 struct sk_buff *skb); 1710 #ifdef CONFIG_SYN_COOKIES 1711 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1712 __u16 *mss); 1713 #endif 1714 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1715 const struct request_sock *req, 1716 bool *strict); 1717 __u32 (*init_seq)(const struct sk_buff *skb); 1718 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1719 struct flowi *fl, struct request_sock *req, 1720 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1721 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1722 const unsigned long timeout); 1723 }; 1724 1725 #ifdef CONFIG_SYN_COOKIES 1726 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1727 const struct sock *sk, struct sk_buff *skb, 1728 __u16 *mss) 1729 { 1730 tcp_synq_overflow(sk); 1731 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1732 return ops->cookie_init_seq(skb, mss); 1733 } 1734 #else 1735 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1736 const struct sock *sk, struct sk_buff *skb, 1737 __u16 *mss) 1738 { 1739 return 0; 1740 } 1741 #endif 1742 1743 int tcpv4_offload_init(void); 1744 1745 void tcp_v4_init(void); 1746 void tcp_init(void); 1747 1748 /* 1749 * Save and compile IPv4 options, return a pointer to it 1750 */ 1751 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1752 { 1753 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1754 struct ip_options_rcu *dopt = NULL; 1755 1756 if (opt->optlen) { 1757 int opt_size = sizeof(*dopt) + opt->optlen; 1758 1759 dopt = kmalloc(opt_size, GFP_ATOMIC); 1760 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1761 kfree(dopt); 1762 dopt = NULL; 1763 } 1764 } 1765 return dopt; 1766 } 1767 1768 /* locally generated TCP pure ACKs have skb->truesize == 2 1769 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1770 * This is much faster than dissecting the packet to find out. 1771 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1772 */ 1773 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1774 { 1775 return skb->truesize == 2; 1776 } 1777 1778 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1779 { 1780 skb->truesize = 2; 1781 } 1782 1783 #endif /* _TCP_H */ 1784