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/dmaengine.h> 31 #include <linux/crypto.h> 32 #include <linux/cryptohash.h> 33 #include <linux/kref.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 extern 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 /* Offer an initial receive window of 10 mss. */ 65 #define TCP_DEFAULT_INIT_RCVWND 10 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 512 72 73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 74 #define TCP_FASTRETRANS_THRESH 3 75 76 /* Maximal reordering. */ 77 #define TCP_MAX_REORDERING 127 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 5 /* number of times to retry active opening a 102 * connection: ~180sec is RFC minimum */ 103 104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 105 * connection: ~180sec is RFC minimum */ 106 107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 108 * state, about 60 seconds */ 109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 110 /* BSD style FIN_WAIT2 deadlock breaker. 111 * It used to be 3min, new value is 60sec, 112 * to combine FIN-WAIT-2 timeout with 113 * TIME-WAIT timer. 114 */ 115 116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 117 #if HZ >= 100 118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 119 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 120 #else 121 #define TCP_DELACK_MIN 4U 122 #define TCP_ATO_MIN 4U 123 #endif 124 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 125 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 128 * used as a fallback RTO for the 129 * initial data transmission if no 130 * valid RTT sample has been acquired, 131 * most likely due to retrans in 3WHS. 132 */ 133 134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 135 * for local resources. 136 */ 137 138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 140 #define TCP_KEEPALIVE_INTVL (75*HZ) 141 142 #define MAX_TCP_KEEPIDLE 32767 143 #define MAX_TCP_KEEPINTVL 32767 144 #define MAX_TCP_KEEPCNT 127 145 #define MAX_TCP_SYNCNT 127 146 147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 148 149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 151 * after this time. It should be equal 152 * (or greater than) TCP_TIMEWAIT_LEN 153 * to provide reliability equal to one 154 * provided by timewait state. 155 */ 156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 157 * timestamps. It must be less than 158 * minimal timewait lifetime. 159 */ 160 /* 161 * TCP option 162 */ 163 164 #define TCPOPT_NOP 1 /* Padding */ 165 #define TCPOPT_EOL 0 /* End of options */ 166 #define TCPOPT_MSS 2 /* Segment size negotiating */ 167 #define TCPOPT_WINDOW 3 /* Window scaling */ 168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 169 #define TCPOPT_SACK 5 /* SACK Block */ 170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 173 #define TCPOPT_EXP 254 /* Experimental */ 174 /* Magic number to be after the option value for sharing TCP 175 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 176 */ 177 #define TCPOPT_FASTOPEN_MAGIC 0xF989 178 179 /* 180 * TCP option lengths 181 */ 182 183 #define TCPOLEN_MSS 4 184 #define TCPOLEN_WINDOW 3 185 #define TCPOLEN_SACK_PERM 2 186 #define TCPOLEN_TIMESTAMP 10 187 #define TCPOLEN_MD5SIG 18 188 #define TCPOLEN_EXP_FASTOPEN_BASE 4 189 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 190 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 191 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 192 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 193 194 /* But this is what stacks really send out. */ 195 #define TCPOLEN_TSTAMP_ALIGNED 12 196 #define TCPOLEN_WSCALE_ALIGNED 4 197 #define TCPOLEN_SACKPERM_ALIGNED 4 198 #define TCPOLEN_SACK_BASE 2 199 #define TCPOLEN_SACK_BASE_ALIGNED 4 200 #define TCPOLEN_SACK_PERBLOCK 8 201 #define TCPOLEN_MD5SIG_ALIGNED 20 202 #define TCPOLEN_MSS_ALIGNED 4 203 204 /* Flags in tp->nonagle */ 205 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 206 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 207 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 208 209 /* TCP thin-stream limits */ 210 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 211 212 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 213 #define TCP_INIT_CWND 10 214 215 /* Bit Flags for sysctl_tcp_fastopen */ 216 #define TFO_CLIENT_ENABLE 1 217 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 218 219 extern struct inet_timewait_death_row tcp_death_row; 220 221 /* sysctl variables for tcp */ 222 extern int sysctl_tcp_timestamps; 223 extern int sysctl_tcp_window_scaling; 224 extern int sysctl_tcp_sack; 225 extern int sysctl_tcp_fin_timeout; 226 extern int sysctl_tcp_keepalive_time; 227 extern int sysctl_tcp_keepalive_probes; 228 extern int sysctl_tcp_keepalive_intvl; 229 extern int sysctl_tcp_syn_retries; 230 extern int sysctl_tcp_synack_retries; 231 extern int sysctl_tcp_retries1; 232 extern int sysctl_tcp_retries2; 233 extern int sysctl_tcp_orphan_retries; 234 extern int sysctl_tcp_syncookies; 235 extern int sysctl_tcp_fastopen; 236 extern int sysctl_tcp_retrans_collapse; 237 extern int sysctl_tcp_stdurg; 238 extern int sysctl_tcp_rfc1337; 239 extern int sysctl_tcp_abort_on_overflow; 240 extern int sysctl_tcp_max_orphans; 241 extern int sysctl_tcp_fack; 242 extern int sysctl_tcp_reordering; 243 extern int sysctl_tcp_ecn; 244 extern int sysctl_tcp_dsack; 245 extern int sysctl_tcp_wmem[3]; 246 extern int sysctl_tcp_rmem[3]; 247 extern int sysctl_tcp_app_win; 248 extern int sysctl_tcp_adv_win_scale; 249 extern int sysctl_tcp_tw_reuse; 250 extern int sysctl_tcp_frto; 251 extern int sysctl_tcp_frto_response; 252 extern int sysctl_tcp_low_latency; 253 extern int sysctl_tcp_dma_copybreak; 254 extern int sysctl_tcp_nometrics_save; 255 extern int sysctl_tcp_moderate_rcvbuf; 256 extern int sysctl_tcp_tso_win_divisor; 257 extern int sysctl_tcp_abc; 258 extern int sysctl_tcp_mtu_probing; 259 extern int sysctl_tcp_base_mss; 260 extern int sysctl_tcp_workaround_signed_windows; 261 extern int sysctl_tcp_slow_start_after_idle; 262 extern int sysctl_tcp_max_ssthresh; 263 extern int sysctl_tcp_cookie_size; 264 extern int sysctl_tcp_thin_linear_timeouts; 265 extern int sysctl_tcp_thin_dupack; 266 extern int sysctl_tcp_early_retrans; 267 extern int sysctl_tcp_limit_output_bytes; 268 extern int sysctl_tcp_challenge_ack_limit; 269 270 extern atomic_long_t tcp_memory_allocated; 271 extern struct percpu_counter tcp_sockets_allocated; 272 extern int tcp_memory_pressure; 273 274 /* 275 * The next routines deal with comparing 32 bit unsigned ints 276 * and worry about wraparound (automatic with unsigned arithmetic). 277 */ 278 279 static inline bool before(__u32 seq1, __u32 seq2) 280 { 281 return (__s32)(seq1-seq2) < 0; 282 } 283 #define after(seq2, seq1) before(seq1, seq2) 284 285 /* is s2<=s1<=s3 ? */ 286 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 287 { 288 return seq3 - seq2 >= seq1 - seq2; 289 } 290 291 static inline bool tcp_out_of_memory(struct sock *sk) 292 { 293 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 294 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 295 return true; 296 return false; 297 } 298 299 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 300 { 301 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 302 int orphans = percpu_counter_read_positive(ocp); 303 304 if (orphans << shift > sysctl_tcp_max_orphans) { 305 orphans = percpu_counter_sum_positive(ocp); 306 if (orphans << shift > sysctl_tcp_max_orphans) 307 return true; 308 } 309 return false; 310 } 311 312 extern bool tcp_check_oom(struct sock *sk, int shift); 313 314 /* syncookies: remember time of last synqueue overflow */ 315 static inline void tcp_synq_overflow(struct sock *sk) 316 { 317 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 318 } 319 320 /* syncookies: no recent synqueue overflow on this listening socket? */ 321 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 322 { 323 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 324 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 325 } 326 327 extern struct proto tcp_prot; 328 329 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 330 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 331 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 332 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 333 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 334 335 extern void tcp_init_mem(struct net *net); 336 337 extern void tcp_tasklet_init(void); 338 339 extern void tcp_v4_err(struct sk_buff *skb, u32); 340 341 extern void tcp_shutdown (struct sock *sk, int how); 342 343 extern void tcp_v4_early_demux(struct sk_buff *skb); 344 extern int tcp_v4_rcv(struct sk_buff *skb); 345 346 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk); 347 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 348 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 349 size_t size); 350 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 351 size_t size, int flags); 352 extern void tcp_release_cb(struct sock *sk); 353 extern void tcp_write_timer_handler(struct sock *sk); 354 extern void tcp_delack_timer_handler(struct sock *sk); 355 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 356 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 357 const struct tcphdr *th, unsigned int len); 358 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 359 const struct tcphdr *th, unsigned int len); 360 extern void tcp_rcv_space_adjust(struct sock *sk); 361 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 362 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 363 extern void tcp_twsk_destructor(struct sock *sk); 364 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 365 struct pipe_inode_info *pipe, size_t len, 366 unsigned int flags); 367 368 static inline void tcp_dec_quickack_mode(struct sock *sk, 369 const unsigned int pkts) 370 { 371 struct inet_connection_sock *icsk = inet_csk(sk); 372 373 if (icsk->icsk_ack.quick) { 374 if (pkts >= icsk->icsk_ack.quick) { 375 icsk->icsk_ack.quick = 0; 376 /* Leaving quickack mode we deflate ATO. */ 377 icsk->icsk_ack.ato = TCP_ATO_MIN; 378 } else 379 icsk->icsk_ack.quick -= pkts; 380 } 381 } 382 383 #define TCP_ECN_OK 1 384 #define TCP_ECN_QUEUE_CWR 2 385 #define TCP_ECN_DEMAND_CWR 4 386 #define TCP_ECN_SEEN 8 387 388 enum tcp_tw_status { 389 TCP_TW_SUCCESS = 0, 390 TCP_TW_RST = 1, 391 TCP_TW_ACK = 2, 392 TCP_TW_SYN = 3 393 }; 394 395 396 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 397 struct sk_buff *skb, 398 const struct tcphdr *th); 399 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 400 struct request_sock *req, 401 struct request_sock **prev); 402 extern int tcp_child_process(struct sock *parent, struct sock *child, 403 struct sk_buff *skb); 404 extern bool tcp_use_frto(struct sock *sk); 405 extern void tcp_enter_frto(struct sock *sk); 406 extern void tcp_enter_loss(struct sock *sk, int how); 407 extern void tcp_clear_retrans(struct tcp_sock *tp); 408 extern void tcp_update_metrics(struct sock *sk); 409 extern void tcp_init_metrics(struct sock *sk); 410 extern void tcp_metrics_init(void); 411 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check); 412 extern bool tcp_remember_stamp(struct sock *sk); 413 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 414 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 415 struct tcp_fastopen_cookie *cookie, 416 int *syn_loss, unsigned long *last_syn_loss); 417 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 418 struct tcp_fastopen_cookie *cookie, 419 bool syn_lost); 420 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 421 extern void tcp_disable_fack(struct tcp_sock *tp); 422 extern void tcp_close(struct sock *sk, long timeout); 423 extern void tcp_init_sock(struct sock *sk); 424 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 425 struct poll_table_struct *wait); 426 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 427 char __user *optval, int __user *optlen); 428 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 429 char __user *optval, unsigned int optlen); 430 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 431 char __user *optval, int __user *optlen); 432 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 433 char __user *optval, unsigned int optlen); 434 extern void tcp_set_keepalive(struct sock *sk, int val); 435 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 436 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 437 size_t len, int nonblock, int flags, int *addr_len); 438 extern void tcp_parse_options(const struct sk_buff *skb, 439 struct tcp_options_received *opt_rx, const u8 **hvpp, 440 int estab, struct tcp_fastopen_cookie *foc); 441 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 442 443 /* 444 * TCP v4 functions exported for the inet6 API 445 */ 446 447 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 448 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 449 extern struct sock * tcp_create_openreq_child(struct sock *sk, 450 struct request_sock *req, 451 struct sk_buff *skb); 452 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 453 struct request_sock *req, 454 struct dst_entry *dst); 455 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 456 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 457 int addr_len); 458 extern int tcp_connect(struct sock *sk); 459 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 460 struct request_sock *req, 461 struct request_values *rvp); 462 extern int tcp_disconnect(struct sock *sk, int flags); 463 464 void tcp_connect_init(struct sock *sk); 465 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 466 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 467 468 /* From syncookies.c */ 469 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 470 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 471 struct ip_options *opt); 472 #ifdef CONFIG_SYN_COOKIES 473 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 474 __u16 *mss); 475 #else 476 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 477 struct sk_buff *skb, 478 __u16 *mss) 479 { 480 return 0; 481 } 482 #endif 483 484 extern __u32 cookie_init_timestamp(struct request_sock *req); 485 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 486 487 /* From net/ipv6/syncookies.c */ 488 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 489 #ifdef CONFIG_SYN_COOKIES 490 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 491 __u16 *mss); 492 #else 493 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 494 struct sk_buff *skb, 495 __u16 *mss) 496 { 497 return 0; 498 } 499 #endif 500 /* tcp_output.c */ 501 502 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 503 int nonagle); 504 extern bool tcp_may_send_now(struct sock *sk); 505 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 506 extern void tcp_retransmit_timer(struct sock *sk); 507 extern void tcp_xmit_retransmit_queue(struct sock *); 508 extern void tcp_simple_retransmit(struct sock *); 509 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 510 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 511 512 extern void tcp_send_probe0(struct sock *); 513 extern void tcp_send_partial(struct sock *); 514 extern int tcp_write_wakeup(struct sock *); 515 extern void tcp_send_fin(struct sock *sk); 516 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 517 extern int tcp_send_synack(struct sock *); 518 extern bool tcp_syn_flood_action(struct sock *sk, 519 const struct sk_buff *skb, 520 const char *proto); 521 extern void tcp_push_one(struct sock *, unsigned int mss_now); 522 extern void tcp_send_ack(struct sock *sk); 523 extern void tcp_send_delayed_ack(struct sock *sk); 524 525 /* tcp_input.c */ 526 extern void tcp_cwnd_application_limited(struct sock *sk); 527 extern void tcp_resume_early_retransmit(struct sock *sk); 528 extern void tcp_rearm_rto(struct sock *sk); 529 530 /* tcp_timer.c */ 531 extern void tcp_init_xmit_timers(struct sock *); 532 static inline void tcp_clear_xmit_timers(struct sock *sk) 533 { 534 inet_csk_clear_xmit_timers(sk); 535 } 536 537 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 538 extern unsigned int tcp_current_mss(struct sock *sk); 539 540 /* Bound MSS / TSO packet size with the half of the window */ 541 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 542 { 543 int cutoff; 544 545 /* When peer uses tiny windows, there is no use in packetizing 546 * to sub-MSS pieces for the sake of SWS or making sure there 547 * are enough packets in the pipe for fast recovery. 548 * 549 * On the other hand, for extremely large MSS devices, handling 550 * smaller than MSS windows in this way does make sense. 551 */ 552 if (tp->max_window >= 512) 553 cutoff = (tp->max_window >> 1); 554 else 555 cutoff = tp->max_window; 556 557 if (cutoff && pktsize > cutoff) 558 return max_t(int, cutoff, 68U - tp->tcp_header_len); 559 else 560 return pktsize; 561 } 562 563 /* tcp.c */ 564 extern void tcp_get_info(const struct sock *, struct tcp_info *); 565 566 /* Read 'sendfile()'-style from a TCP socket */ 567 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 568 unsigned int, size_t); 569 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 570 sk_read_actor_t recv_actor); 571 572 extern void tcp_initialize_rcv_mss(struct sock *sk); 573 574 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 575 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 576 extern void tcp_mtup_init(struct sock *sk); 577 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt); 578 579 static inline void tcp_bound_rto(const struct sock *sk) 580 { 581 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 582 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 583 } 584 585 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 586 { 587 return (tp->srtt >> 3) + tp->rttvar; 588 } 589 590 extern void tcp_set_rto(struct sock *sk); 591 592 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 593 { 594 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 595 ntohl(TCP_FLAG_ACK) | 596 snd_wnd); 597 } 598 599 static inline void tcp_fast_path_on(struct tcp_sock *tp) 600 { 601 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 602 } 603 604 static inline void tcp_fast_path_check(struct sock *sk) 605 { 606 struct tcp_sock *tp = tcp_sk(sk); 607 608 if (skb_queue_empty(&tp->out_of_order_queue) && 609 tp->rcv_wnd && 610 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 611 !tp->urg_data) 612 tcp_fast_path_on(tp); 613 } 614 615 /* Compute the actual rto_min value */ 616 static inline u32 tcp_rto_min(struct sock *sk) 617 { 618 const struct dst_entry *dst = __sk_dst_get(sk); 619 u32 rto_min = TCP_RTO_MIN; 620 621 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 622 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 623 return rto_min; 624 } 625 626 /* Compute the actual receive window we are currently advertising. 627 * Rcv_nxt can be after the window if our peer push more data 628 * than the offered window. 629 */ 630 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 631 { 632 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 633 634 if (win < 0) 635 win = 0; 636 return (u32) win; 637 } 638 639 /* Choose a new window, without checks for shrinking, and without 640 * scaling applied to the result. The caller does these things 641 * if necessary. This is a "raw" window selection. 642 */ 643 extern u32 __tcp_select_window(struct sock *sk); 644 645 void tcp_send_window_probe(struct sock *sk); 646 647 /* TCP timestamps are only 32-bits, this causes a slight 648 * complication on 64-bit systems since we store a snapshot 649 * of jiffies in the buffer control blocks below. We decided 650 * to use only the low 32-bits of jiffies and hide the ugly 651 * casts with the following macro. 652 */ 653 #define tcp_time_stamp ((__u32)(jiffies)) 654 655 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 656 657 #define TCPHDR_FIN 0x01 658 #define TCPHDR_SYN 0x02 659 #define TCPHDR_RST 0x04 660 #define TCPHDR_PSH 0x08 661 #define TCPHDR_ACK 0x10 662 #define TCPHDR_URG 0x20 663 #define TCPHDR_ECE 0x40 664 #define TCPHDR_CWR 0x80 665 666 /* This is what the send packet queuing engine uses to pass 667 * TCP per-packet control information to the transmission code. 668 * We also store the host-order sequence numbers in here too. 669 * This is 44 bytes if IPV6 is enabled. 670 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 671 */ 672 struct tcp_skb_cb { 673 union { 674 struct inet_skb_parm h4; 675 #if IS_ENABLED(CONFIG_IPV6) 676 struct inet6_skb_parm h6; 677 #endif 678 } header; /* For incoming frames */ 679 __u32 seq; /* Starting sequence number */ 680 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 681 __u32 when; /* used to compute rtt's */ 682 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 683 684 __u8 sacked; /* State flags for SACK/FACK. */ 685 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 686 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 687 #define TCPCB_LOST 0x04 /* SKB is lost */ 688 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 689 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 690 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 691 692 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 693 /* 1 byte hole */ 694 __u32 ack_seq; /* Sequence number ACK'd */ 695 }; 696 697 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 698 699 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 700 * 701 * If we receive a SYN packet with these bits set, it means a network is 702 * playing bad games with TOS bits. In order to avoid possible false congestion 703 * notifications, we disable TCP ECN negociation. 704 */ 705 static inline void 706 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb) 707 { 708 const struct tcphdr *th = tcp_hdr(skb); 709 710 if (sysctl_tcp_ecn && th->ece && th->cwr && 711 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 712 inet_rsk(req)->ecn_ok = 1; 713 } 714 715 /* Due to TSO, an SKB can be composed of multiple actual 716 * packets. To keep these tracked properly, we use this. 717 */ 718 static inline int tcp_skb_pcount(const struct sk_buff *skb) 719 { 720 return skb_shinfo(skb)->gso_segs; 721 } 722 723 /* This is valid iff tcp_skb_pcount() > 1. */ 724 static inline int tcp_skb_mss(const struct sk_buff *skb) 725 { 726 return skb_shinfo(skb)->gso_size; 727 } 728 729 /* Events passed to congestion control interface */ 730 enum tcp_ca_event { 731 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 732 CA_EVENT_CWND_RESTART, /* congestion window restart */ 733 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 734 CA_EVENT_FRTO, /* fast recovery timeout */ 735 CA_EVENT_LOSS, /* loss timeout */ 736 CA_EVENT_FAST_ACK, /* in sequence ack */ 737 CA_EVENT_SLOW_ACK, /* other ack */ 738 }; 739 740 /* 741 * Interface for adding new TCP congestion control handlers 742 */ 743 #define TCP_CA_NAME_MAX 16 744 #define TCP_CA_MAX 128 745 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 746 747 #define TCP_CONG_NON_RESTRICTED 0x1 748 #define TCP_CONG_RTT_STAMP 0x2 749 750 struct tcp_congestion_ops { 751 struct list_head list; 752 unsigned long flags; 753 754 /* initialize private data (optional) */ 755 void (*init)(struct sock *sk); 756 /* cleanup private data (optional) */ 757 void (*release)(struct sock *sk); 758 759 /* return slow start threshold (required) */ 760 u32 (*ssthresh)(struct sock *sk); 761 /* lower bound for congestion window (optional) */ 762 u32 (*min_cwnd)(const struct sock *sk); 763 /* do new cwnd calculation (required) */ 764 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 765 /* call before changing ca_state (optional) */ 766 void (*set_state)(struct sock *sk, u8 new_state); 767 /* call when cwnd event occurs (optional) */ 768 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 769 /* new value of cwnd after loss (optional) */ 770 u32 (*undo_cwnd)(struct sock *sk); 771 /* hook for packet ack accounting (optional) */ 772 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 773 /* get info for inet_diag (optional) */ 774 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 775 776 char name[TCP_CA_NAME_MAX]; 777 struct module *owner; 778 }; 779 780 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 781 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 782 783 extern void tcp_init_congestion_control(struct sock *sk); 784 extern void tcp_cleanup_congestion_control(struct sock *sk); 785 extern int tcp_set_default_congestion_control(const char *name); 786 extern void tcp_get_default_congestion_control(char *name); 787 extern void tcp_get_available_congestion_control(char *buf, size_t len); 788 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 789 extern int tcp_set_allowed_congestion_control(char *allowed); 790 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 791 extern void tcp_slow_start(struct tcp_sock *tp); 792 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 793 794 extern struct tcp_congestion_ops tcp_init_congestion_ops; 795 extern u32 tcp_reno_ssthresh(struct sock *sk); 796 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 797 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 798 extern struct tcp_congestion_ops tcp_reno; 799 800 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 801 { 802 struct inet_connection_sock *icsk = inet_csk(sk); 803 804 if (icsk->icsk_ca_ops->set_state) 805 icsk->icsk_ca_ops->set_state(sk, ca_state); 806 icsk->icsk_ca_state = ca_state; 807 } 808 809 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 810 { 811 const struct inet_connection_sock *icsk = inet_csk(sk); 812 813 if (icsk->icsk_ca_ops->cwnd_event) 814 icsk->icsk_ca_ops->cwnd_event(sk, event); 815 } 816 817 /* These functions determine how the current flow behaves in respect of SACK 818 * handling. SACK is negotiated with the peer, and therefore it can vary 819 * between different flows. 820 * 821 * tcp_is_sack - SACK enabled 822 * tcp_is_reno - No SACK 823 * tcp_is_fack - FACK enabled, implies SACK enabled 824 */ 825 static inline int tcp_is_sack(const struct tcp_sock *tp) 826 { 827 return tp->rx_opt.sack_ok; 828 } 829 830 static inline bool tcp_is_reno(const struct tcp_sock *tp) 831 { 832 return !tcp_is_sack(tp); 833 } 834 835 static inline bool tcp_is_fack(const struct tcp_sock *tp) 836 { 837 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 838 } 839 840 static inline void tcp_enable_fack(struct tcp_sock *tp) 841 { 842 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 843 } 844 845 /* TCP early-retransmit (ER) is similar to but more conservative than 846 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 847 */ 848 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 849 { 850 tp->do_early_retrans = sysctl_tcp_early_retrans && 851 !sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3; 852 tp->early_retrans_delayed = 0; 853 } 854 855 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 856 { 857 tp->do_early_retrans = 0; 858 } 859 860 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 861 { 862 return tp->sacked_out + tp->lost_out; 863 } 864 865 /* This determines how many packets are "in the network" to the best 866 * of our knowledge. In many cases it is conservative, but where 867 * detailed information is available from the receiver (via SACK 868 * blocks etc.) we can make more aggressive calculations. 869 * 870 * Use this for decisions involving congestion control, use just 871 * tp->packets_out to determine if the send queue is empty or not. 872 * 873 * Read this equation as: 874 * 875 * "Packets sent once on transmission queue" MINUS 876 * "Packets left network, but not honestly ACKed yet" PLUS 877 * "Packets fast retransmitted" 878 */ 879 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 880 { 881 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 882 } 883 884 #define TCP_INFINITE_SSTHRESH 0x7fffffff 885 886 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 887 { 888 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 889 } 890 891 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 892 * The exception is rate halving phase, when cwnd is decreasing towards 893 * ssthresh. 894 */ 895 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 896 { 897 const struct tcp_sock *tp = tcp_sk(sk); 898 899 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 900 return tp->snd_ssthresh; 901 else 902 return max(tp->snd_ssthresh, 903 ((tp->snd_cwnd >> 1) + 904 (tp->snd_cwnd >> 2))); 905 } 906 907 /* Use define here intentionally to get WARN_ON location shown at the caller */ 908 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 909 910 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 911 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 912 913 /* The maximum number of MSS of available cwnd for which TSO defers 914 * sending if not using sysctl_tcp_tso_win_divisor. 915 */ 916 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 917 { 918 return 3; 919 } 920 921 /* Slow start with delack produces 3 packets of burst, so that 922 * it is safe "de facto". This will be the default - same as 923 * the default reordering threshold - but if reordering increases, 924 * we must be able to allow cwnd to burst at least this much in order 925 * to not pull it back when holes are filled. 926 */ 927 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 928 { 929 return tp->reordering; 930 } 931 932 /* Returns end sequence number of the receiver's advertised window */ 933 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 934 { 935 return tp->snd_una + tp->snd_wnd; 936 } 937 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 938 939 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 940 const struct sk_buff *skb) 941 { 942 if (skb->len < mss) 943 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 944 } 945 946 static inline void tcp_check_probe_timer(struct sock *sk) 947 { 948 const struct tcp_sock *tp = tcp_sk(sk); 949 const struct inet_connection_sock *icsk = inet_csk(sk); 950 951 if (!tp->packets_out && !icsk->icsk_pending) 952 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 953 icsk->icsk_rto, TCP_RTO_MAX); 954 } 955 956 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 957 { 958 tp->snd_wl1 = seq; 959 } 960 961 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 962 { 963 tp->snd_wl1 = seq; 964 } 965 966 /* 967 * Calculate(/check) TCP checksum 968 */ 969 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 970 __be32 daddr, __wsum base) 971 { 972 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 973 } 974 975 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 976 { 977 return __skb_checksum_complete(skb); 978 } 979 980 static inline bool tcp_checksum_complete(struct sk_buff *skb) 981 { 982 return !skb_csum_unnecessary(skb) && 983 __tcp_checksum_complete(skb); 984 } 985 986 /* Prequeue for VJ style copy to user, combined with checksumming. */ 987 988 static inline void tcp_prequeue_init(struct tcp_sock *tp) 989 { 990 tp->ucopy.task = NULL; 991 tp->ucopy.len = 0; 992 tp->ucopy.memory = 0; 993 skb_queue_head_init(&tp->ucopy.prequeue); 994 #ifdef CONFIG_NET_DMA 995 tp->ucopy.dma_chan = NULL; 996 tp->ucopy.wakeup = 0; 997 tp->ucopy.pinned_list = NULL; 998 tp->ucopy.dma_cookie = 0; 999 #endif 1000 } 1001 1002 /* Packet is added to VJ-style prequeue for processing in process 1003 * context, if a reader task is waiting. Apparently, this exciting 1004 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 1005 * failed somewhere. Latency? Burstiness? Well, at least now we will 1006 * see, why it failed. 8)8) --ANK 1007 * 1008 * NOTE: is this not too big to inline? 1009 */ 1010 static inline bool tcp_prequeue(struct sock *sk, struct sk_buff *skb) 1011 { 1012 struct tcp_sock *tp = tcp_sk(sk); 1013 1014 if (sysctl_tcp_low_latency || !tp->ucopy.task) 1015 return false; 1016 1017 __skb_queue_tail(&tp->ucopy.prequeue, skb); 1018 tp->ucopy.memory += skb->truesize; 1019 if (tp->ucopy.memory > sk->sk_rcvbuf) { 1020 struct sk_buff *skb1; 1021 1022 BUG_ON(sock_owned_by_user(sk)); 1023 1024 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 1025 sk_backlog_rcv(sk, skb1); 1026 NET_INC_STATS_BH(sock_net(sk), 1027 LINUX_MIB_TCPPREQUEUEDROPPED); 1028 } 1029 1030 tp->ucopy.memory = 0; 1031 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 1032 wake_up_interruptible_sync_poll(sk_sleep(sk), 1033 POLLIN | POLLRDNORM | POLLRDBAND); 1034 if (!inet_csk_ack_scheduled(sk)) 1035 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 1036 (3 * tcp_rto_min(sk)) / 4, 1037 TCP_RTO_MAX); 1038 } 1039 return true; 1040 } 1041 1042 1043 #undef STATE_TRACE 1044 1045 #ifdef STATE_TRACE 1046 static const char *statename[]={ 1047 "Unused","Established","Syn Sent","Syn Recv", 1048 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1049 "Close Wait","Last ACK","Listen","Closing" 1050 }; 1051 #endif 1052 extern void tcp_set_state(struct sock *sk, int state); 1053 1054 extern void tcp_done(struct sock *sk); 1055 1056 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1057 { 1058 rx_opt->dsack = 0; 1059 rx_opt->num_sacks = 0; 1060 } 1061 1062 /* Determine a window scaling and initial window to offer. */ 1063 extern void tcp_select_initial_window(int __space, __u32 mss, 1064 __u32 *rcv_wnd, __u32 *window_clamp, 1065 int wscale_ok, __u8 *rcv_wscale, 1066 __u32 init_rcv_wnd); 1067 1068 static inline int tcp_win_from_space(int space) 1069 { 1070 return sysctl_tcp_adv_win_scale<=0 ? 1071 (space>>(-sysctl_tcp_adv_win_scale)) : 1072 space - (space>>sysctl_tcp_adv_win_scale); 1073 } 1074 1075 /* Note: caller must be prepared to deal with negative returns */ 1076 static inline int tcp_space(const struct sock *sk) 1077 { 1078 return tcp_win_from_space(sk->sk_rcvbuf - 1079 atomic_read(&sk->sk_rmem_alloc)); 1080 } 1081 1082 static inline int tcp_full_space(const struct sock *sk) 1083 { 1084 return tcp_win_from_space(sk->sk_rcvbuf); 1085 } 1086 1087 static inline void tcp_openreq_init(struct request_sock *req, 1088 struct tcp_options_received *rx_opt, 1089 struct sk_buff *skb) 1090 { 1091 struct inet_request_sock *ireq = inet_rsk(req); 1092 1093 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1094 req->cookie_ts = 0; 1095 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1096 req->mss = rx_opt->mss_clamp; 1097 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1098 ireq->tstamp_ok = rx_opt->tstamp_ok; 1099 ireq->sack_ok = rx_opt->sack_ok; 1100 ireq->snd_wscale = rx_opt->snd_wscale; 1101 ireq->wscale_ok = rx_opt->wscale_ok; 1102 ireq->acked = 0; 1103 ireq->ecn_ok = 0; 1104 ireq->rmt_port = tcp_hdr(skb)->source; 1105 ireq->loc_port = tcp_hdr(skb)->dest; 1106 } 1107 1108 extern void tcp_enter_memory_pressure(struct sock *sk); 1109 1110 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1111 { 1112 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1113 } 1114 1115 static inline int keepalive_time_when(const struct tcp_sock *tp) 1116 { 1117 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1118 } 1119 1120 static inline int keepalive_probes(const struct tcp_sock *tp) 1121 { 1122 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1123 } 1124 1125 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1126 { 1127 const struct inet_connection_sock *icsk = &tp->inet_conn; 1128 1129 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1130 tcp_time_stamp - tp->rcv_tstamp); 1131 } 1132 1133 static inline int tcp_fin_time(const struct sock *sk) 1134 { 1135 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1136 const int rto = inet_csk(sk)->icsk_rto; 1137 1138 if (fin_timeout < (rto << 2) - (rto >> 1)) 1139 fin_timeout = (rto << 2) - (rto >> 1); 1140 1141 return fin_timeout; 1142 } 1143 1144 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1145 int paws_win) 1146 { 1147 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1148 return true; 1149 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1150 return true; 1151 /* 1152 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1153 * then following tcp messages have valid values. Ignore 0 value, 1154 * or else 'negative' tsval might forbid us to accept their packets. 1155 */ 1156 if (!rx_opt->ts_recent) 1157 return true; 1158 return false; 1159 } 1160 1161 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1162 int rst) 1163 { 1164 if (tcp_paws_check(rx_opt, 0)) 1165 return false; 1166 1167 /* RST segments are not recommended to carry timestamp, 1168 and, if they do, it is recommended to ignore PAWS because 1169 "their cleanup function should take precedence over timestamps." 1170 Certainly, it is mistake. It is necessary to understand the reasons 1171 of this constraint to relax it: if peer reboots, clock may go 1172 out-of-sync and half-open connections will not be reset. 1173 Actually, the problem would be not existing if all 1174 the implementations followed draft about maintaining clock 1175 via reboots. Linux-2.2 DOES NOT! 1176 1177 However, we can relax time bounds for RST segments to MSL. 1178 */ 1179 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1180 return false; 1181 return true; 1182 } 1183 1184 static inline void tcp_mib_init(struct net *net) 1185 { 1186 /* See RFC 2012 */ 1187 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1188 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1189 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1190 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1191 } 1192 1193 /* from STCP */ 1194 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1195 { 1196 tp->lost_skb_hint = NULL; 1197 tp->scoreboard_skb_hint = NULL; 1198 } 1199 1200 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1201 { 1202 tcp_clear_retrans_hints_partial(tp); 1203 tp->retransmit_skb_hint = NULL; 1204 } 1205 1206 /* MD5 Signature */ 1207 struct crypto_hash; 1208 1209 union tcp_md5_addr { 1210 struct in_addr a4; 1211 #if IS_ENABLED(CONFIG_IPV6) 1212 struct in6_addr a6; 1213 #endif 1214 }; 1215 1216 /* - key database */ 1217 struct tcp_md5sig_key { 1218 struct hlist_node node; 1219 u8 keylen; 1220 u8 family; /* AF_INET or AF_INET6 */ 1221 union tcp_md5_addr addr; 1222 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1223 struct rcu_head rcu; 1224 }; 1225 1226 /* - sock block */ 1227 struct tcp_md5sig_info { 1228 struct hlist_head head; 1229 struct rcu_head rcu; 1230 }; 1231 1232 /* - pseudo header */ 1233 struct tcp4_pseudohdr { 1234 __be32 saddr; 1235 __be32 daddr; 1236 __u8 pad; 1237 __u8 protocol; 1238 __be16 len; 1239 }; 1240 1241 struct tcp6_pseudohdr { 1242 struct in6_addr saddr; 1243 struct in6_addr daddr; 1244 __be32 len; 1245 __be32 protocol; /* including padding */ 1246 }; 1247 1248 union tcp_md5sum_block { 1249 struct tcp4_pseudohdr ip4; 1250 #if IS_ENABLED(CONFIG_IPV6) 1251 struct tcp6_pseudohdr ip6; 1252 #endif 1253 }; 1254 1255 /* - pool: digest algorithm, hash description and scratch buffer */ 1256 struct tcp_md5sig_pool { 1257 struct hash_desc md5_desc; 1258 union tcp_md5sum_block md5_blk; 1259 }; 1260 1261 /* - functions */ 1262 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1263 const struct sock *sk, 1264 const struct request_sock *req, 1265 const struct sk_buff *skb); 1266 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1267 int family, const u8 *newkey, 1268 u8 newkeylen, gfp_t gfp); 1269 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1270 int family); 1271 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1272 struct sock *addr_sk); 1273 1274 #ifdef CONFIG_TCP_MD5SIG 1275 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1276 const union tcp_md5_addr *addr, int family); 1277 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1278 #else 1279 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1280 const union tcp_md5_addr *addr, 1281 int family) 1282 { 1283 return NULL; 1284 } 1285 #define tcp_twsk_md5_key(twsk) NULL 1286 #endif 1287 1288 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *); 1289 extern void tcp_free_md5sig_pool(void); 1290 1291 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1292 extern void tcp_put_md5sig_pool(void); 1293 1294 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1295 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1296 unsigned int header_len); 1297 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1298 const struct tcp_md5sig_key *key); 1299 1300 struct tcp_fastopen_request { 1301 /* Fast Open cookie. Size 0 means a cookie request */ 1302 struct tcp_fastopen_cookie cookie; 1303 struct msghdr *data; /* data in MSG_FASTOPEN */ 1304 u16 copied; /* queued in tcp_connect() */ 1305 }; 1306 1307 void tcp_free_fastopen_req(struct tcp_sock *tp); 1308 1309 /* write queue abstraction */ 1310 static inline void tcp_write_queue_purge(struct sock *sk) 1311 { 1312 struct sk_buff *skb; 1313 1314 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1315 sk_wmem_free_skb(sk, skb); 1316 sk_mem_reclaim(sk); 1317 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1318 } 1319 1320 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1321 { 1322 return skb_peek(&sk->sk_write_queue); 1323 } 1324 1325 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1326 { 1327 return skb_peek_tail(&sk->sk_write_queue); 1328 } 1329 1330 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1331 const struct sk_buff *skb) 1332 { 1333 return skb_queue_next(&sk->sk_write_queue, skb); 1334 } 1335 1336 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1337 const struct sk_buff *skb) 1338 { 1339 return skb_queue_prev(&sk->sk_write_queue, skb); 1340 } 1341 1342 #define tcp_for_write_queue(skb, sk) \ 1343 skb_queue_walk(&(sk)->sk_write_queue, skb) 1344 1345 #define tcp_for_write_queue_from(skb, sk) \ 1346 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1347 1348 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1349 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1350 1351 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1352 { 1353 return sk->sk_send_head; 1354 } 1355 1356 static inline bool tcp_skb_is_last(const struct sock *sk, 1357 const struct sk_buff *skb) 1358 { 1359 return skb_queue_is_last(&sk->sk_write_queue, skb); 1360 } 1361 1362 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1363 { 1364 if (tcp_skb_is_last(sk, skb)) 1365 sk->sk_send_head = NULL; 1366 else 1367 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1368 } 1369 1370 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1371 { 1372 if (sk->sk_send_head == skb_unlinked) 1373 sk->sk_send_head = NULL; 1374 } 1375 1376 static inline void tcp_init_send_head(struct sock *sk) 1377 { 1378 sk->sk_send_head = NULL; 1379 } 1380 1381 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1382 { 1383 __skb_queue_tail(&sk->sk_write_queue, skb); 1384 } 1385 1386 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1387 { 1388 __tcp_add_write_queue_tail(sk, skb); 1389 1390 /* Queue it, remembering where we must start sending. */ 1391 if (sk->sk_send_head == NULL) { 1392 sk->sk_send_head = skb; 1393 1394 if (tcp_sk(sk)->highest_sack == NULL) 1395 tcp_sk(sk)->highest_sack = skb; 1396 } 1397 } 1398 1399 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1400 { 1401 __skb_queue_head(&sk->sk_write_queue, skb); 1402 } 1403 1404 /* Insert buff after skb on the write queue of sk. */ 1405 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1406 struct sk_buff *buff, 1407 struct sock *sk) 1408 { 1409 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1410 } 1411 1412 /* Insert new before skb on the write queue of sk. */ 1413 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1414 struct sk_buff *skb, 1415 struct sock *sk) 1416 { 1417 __skb_queue_before(&sk->sk_write_queue, skb, new); 1418 1419 if (sk->sk_send_head == skb) 1420 sk->sk_send_head = new; 1421 } 1422 1423 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1424 { 1425 __skb_unlink(skb, &sk->sk_write_queue); 1426 } 1427 1428 static inline bool tcp_write_queue_empty(struct sock *sk) 1429 { 1430 return skb_queue_empty(&sk->sk_write_queue); 1431 } 1432 1433 static inline void tcp_push_pending_frames(struct sock *sk) 1434 { 1435 if (tcp_send_head(sk)) { 1436 struct tcp_sock *tp = tcp_sk(sk); 1437 1438 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1439 } 1440 } 1441 1442 /* Start sequence of the skb just after the highest skb with SACKed 1443 * bit, valid only if sacked_out > 0 or when the caller has ensured 1444 * validity by itself. 1445 */ 1446 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1447 { 1448 if (!tp->sacked_out) 1449 return tp->snd_una; 1450 1451 if (tp->highest_sack == NULL) 1452 return tp->snd_nxt; 1453 1454 return TCP_SKB_CB(tp->highest_sack)->seq; 1455 } 1456 1457 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1458 { 1459 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1460 tcp_write_queue_next(sk, skb); 1461 } 1462 1463 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1464 { 1465 return tcp_sk(sk)->highest_sack; 1466 } 1467 1468 static inline void tcp_highest_sack_reset(struct sock *sk) 1469 { 1470 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1471 } 1472 1473 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1474 static inline void tcp_highest_sack_combine(struct sock *sk, 1475 struct sk_buff *old, 1476 struct sk_buff *new) 1477 { 1478 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1479 tcp_sk(sk)->highest_sack = new; 1480 } 1481 1482 /* Determines whether this is a thin stream (which may suffer from 1483 * increased latency). Used to trigger latency-reducing mechanisms. 1484 */ 1485 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1486 { 1487 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1488 } 1489 1490 /* /proc */ 1491 enum tcp_seq_states { 1492 TCP_SEQ_STATE_LISTENING, 1493 TCP_SEQ_STATE_OPENREQ, 1494 TCP_SEQ_STATE_ESTABLISHED, 1495 TCP_SEQ_STATE_TIME_WAIT, 1496 }; 1497 1498 int tcp_seq_open(struct inode *inode, struct file *file); 1499 1500 struct tcp_seq_afinfo { 1501 char *name; 1502 sa_family_t family; 1503 const struct file_operations *seq_fops; 1504 struct seq_operations seq_ops; 1505 }; 1506 1507 struct tcp_iter_state { 1508 struct seq_net_private p; 1509 sa_family_t family; 1510 enum tcp_seq_states state; 1511 struct sock *syn_wait_sk; 1512 int bucket, offset, sbucket, num, uid; 1513 loff_t last_pos; 1514 }; 1515 1516 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1517 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1518 1519 extern struct request_sock_ops tcp_request_sock_ops; 1520 extern struct request_sock_ops tcp6_request_sock_ops; 1521 1522 extern void tcp_v4_destroy_sock(struct sock *sk); 1523 1524 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1525 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, 1526 netdev_features_t features); 1527 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1528 struct sk_buff *skb); 1529 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1530 struct sk_buff *skb); 1531 extern int tcp_gro_complete(struct sk_buff *skb); 1532 extern int tcp4_gro_complete(struct sk_buff *skb); 1533 1534 #ifdef CONFIG_PROC_FS 1535 extern int tcp4_proc_init(void); 1536 extern void tcp4_proc_exit(void); 1537 #endif 1538 1539 /* TCP af-specific functions */ 1540 struct tcp_sock_af_ops { 1541 #ifdef CONFIG_TCP_MD5SIG 1542 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1543 struct sock *addr_sk); 1544 int (*calc_md5_hash) (char *location, 1545 struct tcp_md5sig_key *md5, 1546 const struct sock *sk, 1547 const struct request_sock *req, 1548 const struct sk_buff *skb); 1549 int (*md5_parse) (struct sock *sk, 1550 char __user *optval, 1551 int optlen); 1552 #endif 1553 }; 1554 1555 struct tcp_request_sock_ops { 1556 #ifdef CONFIG_TCP_MD5SIG 1557 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1558 struct request_sock *req); 1559 int (*calc_md5_hash) (char *location, 1560 struct tcp_md5sig_key *md5, 1561 const struct sock *sk, 1562 const struct request_sock *req, 1563 const struct sk_buff *skb); 1564 #endif 1565 }; 1566 1567 /* Using SHA1 for now, define some constants. 1568 */ 1569 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1570 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1571 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1572 1573 extern int tcp_cookie_generator(u32 *bakery); 1574 1575 /** 1576 * struct tcp_cookie_values - each socket needs extra space for the 1577 * cookies, together with (optional) space for any SYN data. 1578 * 1579 * A tcp_sock contains a pointer to the current value, and this is 1580 * cloned to the tcp_timewait_sock. 1581 * 1582 * @cookie_pair: variable data from the option exchange. 1583 * 1584 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1585 * indicates default (sysctl_tcp_cookie_size). 1586 * After cookie sent, remembers size of cookie. 1587 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1588 * 1589 * @s_data_desired: user specified tcpct_s_data_desired. When the 1590 * constant payload is specified (@s_data_constant), 1591 * holds its length instead. 1592 * Range 0 to TCP_MSS_DESIRED. 1593 * 1594 * @s_data_payload: constant data that is to be included in the 1595 * payload of SYN or SYNACK segments when the 1596 * cookie option is present. 1597 */ 1598 struct tcp_cookie_values { 1599 struct kref kref; 1600 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1601 u8 cookie_pair_size; 1602 u8 cookie_desired; 1603 u16 s_data_desired:11, 1604 s_data_constant:1, 1605 s_data_in:1, 1606 s_data_out:1, 1607 s_data_unused:2; 1608 u8 s_data_payload[0]; 1609 }; 1610 1611 static inline void tcp_cookie_values_release(struct kref *kref) 1612 { 1613 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1614 } 1615 1616 /* The length of constant payload data. Note that s_data_desired is 1617 * overloaded, depending on s_data_constant: either the length of constant 1618 * data (returned here) or the limit on variable data. 1619 */ 1620 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1621 { 1622 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1623 ? tp->cookie_values->s_data_desired 1624 : 0; 1625 } 1626 1627 /** 1628 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1629 * 1630 * As tcp_request_sock has already been extended in other places, the 1631 * only remaining method is to pass stack values along as function 1632 * parameters. These parameters are not needed after sending SYNACK. 1633 * 1634 * @cookie_bakery: cryptographic secret and message workspace. 1635 * 1636 * @cookie_plus: bytes in authenticator/cookie option, copied from 1637 * struct tcp_options_received (above). 1638 */ 1639 struct tcp_extend_values { 1640 struct request_values rv; 1641 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1642 u8 cookie_plus:6, 1643 cookie_out_never:1, 1644 cookie_in_always:1; 1645 }; 1646 1647 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1648 { 1649 return (struct tcp_extend_values *)rvp; 1650 } 1651 1652 extern void tcp_v4_init(void); 1653 extern void tcp_init(void); 1654 1655 #endif /* _TCP_H */ 1656