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 TCP_DEBUG 1 22 #define FASTRETRANS_DEBUG 1 23 24 #include <linux/list.h> 25 #include <linux/tcp.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 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 extern struct percpu_counter tcp_orphan_count; 52 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54 #define MAX_TCP_HEADER (128 + MAX_HEADER) 55 #define MAX_TCP_OPTION_SPACE 40 56 57 /* 58 * Never offer a window over 32767 without using window scaling. Some 59 * poor stacks do signed 16bit maths! 60 */ 61 #define MAX_TCP_WINDOW 32767U 62 63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 64 #define TCP_MIN_MSS 88U 65 66 /* The least MTU to use for probing */ 67 #define TCP_BASE_MSS 512 68 69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 70 #define TCP_FASTRETRANS_THRESH 3 71 72 /* Maximal reordering. */ 73 #define TCP_MAX_REORDERING 127 74 75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 76 #define TCP_MAX_QUICKACKS 16U 77 78 /* urg_data states */ 79 #define TCP_URG_VALID 0x0100 80 #define TCP_URG_NOTYET 0x0200 81 #define TCP_URG_READ 0x0400 82 83 #define TCP_RETR1 3 /* 84 * This is how many retries it does before it 85 * tries to figure out if the gateway is 86 * down. Minimal RFC value is 3; it corresponds 87 * to ~3sec-8min depending on RTO. 88 */ 89 90 #define TCP_RETR2 15 /* 91 * This should take at least 92 * 90 minutes to time out. 93 * RFC1122 says that the limit is 100 sec. 94 * 15 is ~13-30min depending on RTO. 95 */ 96 97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 98 * connection: ~180sec is RFC minimum */ 99 100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 101 * connection: ~180sec is RFC minimum */ 102 103 104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 105 * socket. 7 is ~50sec-16min. 106 */ 107 108 109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 110 * state, about 60 seconds */ 111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 112 /* BSD style FIN_WAIT2 deadlock breaker. 113 * It used to be 3min, new value is 60sec, 114 * to combine FIN-WAIT-2 timeout with 115 * TIME-WAIT timer. 116 */ 117 118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 119 #if HZ >= 100 120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 121 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 122 #else 123 #define TCP_DELACK_MIN 4U 124 #define TCP_ATO_MIN 4U 125 #endif 126 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 127 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 129 130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 131 * for local resources. 132 */ 133 134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 136 #define TCP_KEEPALIVE_INTVL (75*HZ) 137 138 #define MAX_TCP_KEEPIDLE 32767 139 #define MAX_TCP_KEEPINTVL 32767 140 #define MAX_TCP_KEEPCNT 127 141 #define MAX_TCP_SYNCNT 127 142 143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 144 145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 147 * after this time. It should be equal 148 * (or greater than) TCP_TIMEWAIT_LEN 149 * to provide reliability equal to one 150 * provided by timewait state. 151 */ 152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 153 * timestamps. It must be less than 154 * minimal timewait lifetime. 155 */ 156 /* 157 * TCP option 158 */ 159 160 #define TCPOPT_NOP 1 /* Padding */ 161 #define TCPOPT_EOL 0 /* End of options */ 162 #define TCPOPT_MSS 2 /* Segment size negotiating */ 163 #define TCPOPT_WINDOW 3 /* Window scaling */ 164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 165 #define TCPOPT_SACK 5 /* SACK Block */ 166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 169 170 /* 171 * TCP option lengths 172 */ 173 174 #define TCPOLEN_MSS 4 175 #define TCPOLEN_WINDOW 3 176 #define TCPOLEN_SACK_PERM 2 177 #define TCPOLEN_TIMESTAMP 10 178 #define TCPOLEN_MD5SIG 18 179 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 180 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 181 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 182 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 183 184 /* But this is what stacks really send out. */ 185 #define TCPOLEN_TSTAMP_ALIGNED 12 186 #define TCPOLEN_WSCALE_ALIGNED 4 187 #define TCPOLEN_SACKPERM_ALIGNED 4 188 #define TCPOLEN_SACK_BASE 2 189 #define TCPOLEN_SACK_BASE_ALIGNED 4 190 #define TCPOLEN_SACK_PERBLOCK 8 191 #define TCPOLEN_MD5SIG_ALIGNED 20 192 #define TCPOLEN_MSS_ALIGNED 4 193 194 /* Flags in tp->nonagle */ 195 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 196 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 198 199 /* TCP thin-stream limits */ 200 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 201 202 extern struct inet_timewait_death_row tcp_death_row; 203 204 /* sysctl variables for tcp */ 205 extern int sysctl_tcp_timestamps; 206 extern int sysctl_tcp_window_scaling; 207 extern int sysctl_tcp_sack; 208 extern int sysctl_tcp_fin_timeout; 209 extern int sysctl_tcp_keepalive_time; 210 extern int sysctl_tcp_keepalive_probes; 211 extern int sysctl_tcp_keepalive_intvl; 212 extern int sysctl_tcp_syn_retries; 213 extern int sysctl_tcp_synack_retries; 214 extern int sysctl_tcp_retries1; 215 extern int sysctl_tcp_retries2; 216 extern int sysctl_tcp_orphan_retries; 217 extern int sysctl_tcp_syncookies; 218 extern int sysctl_tcp_retrans_collapse; 219 extern int sysctl_tcp_stdurg; 220 extern int sysctl_tcp_rfc1337; 221 extern int sysctl_tcp_abort_on_overflow; 222 extern int sysctl_tcp_max_orphans; 223 extern int sysctl_tcp_fack; 224 extern int sysctl_tcp_reordering; 225 extern int sysctl_tcp_ecn; 226 extern int sysctl_tcp_dsack; 227 extern int sysctl_tcp_mem[3]; 228 extern int sysctl_tcp_wmem[3]; 229 extern int sysctl_tcp_rmem[3]; 230 extern int sysctl_tcp_app_win; 231 extern int sysctl_tcp_adv_win_scale; 232 extern int sysctl_tcp_tw_reuse; 233 extern int sysctl_tcp_frto; 234 extern int sysctl_tcp_frto_response; 235 extern int sysctl_tcp_low_latency; 236 extern int sysctl_tcp_dma_copybreak; 237 extern int sysctl_tcp_nometrics_save; 238 extern int sysctl_tcp_moderate_rcvbuf; 239 extern int sysctl_tcp_tso_win_divisor; 240 extern int sysctl_tcp_abc; 241 extern int sysctl_tcp_mtu_probing; 242 extern int sysctl_tcp_base_mss; 243 extern int sysctl_tcp_workaround_signed_windows; 244 extern int sysctl_tcp_slow_start_after_idle; 245 extern int sysctl_tcp_max_ssthresh; 246 extern int sysctl_tcp_cookie_size; 247 extern int sysctl_tcp_thin_linear_timeouts; 248 extern int sysctl_tcp_thin_dupack; 249 250 extern atomic_t tcp_memory_allocated; 251 extern struct percpu_counter tcp_sockets_allocated; 252 extern int tcp_memory_pressure; 253 254 /* 255 * The next routines deal with comparing 32 bit unsigned ints 256 * and worry about wraparound (automatic with unsigned arithmetic). 257 */ 258 259 static inline int before(__u32 seq1, __u32 seq2) 260 { 261 return (__s32)(seq1-seq2) < 0; 262 } 263 #define after(seq2, seq1) before(seq1, seq2) 264 265 /* is s2<=s1<=s3 ? */ 266 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 267 { 268 return seq3 - seq2 >= seq1 - seq2; 269 } 270 271 static inline int tcp_too_many_orphans(struct sock *sk, int num) 272 { 273 return (num > sysctl_tcp_max_orphans) || 274 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 275 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]); 276 } 277 278 /* syncookies: remember time of last synqueue overflow */ 279 static inline void tcp_synq_overflow(struct sock *sk) 280 { 281 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 282 } 283 284 /* syncookies: no recent synqueue overflow on this listening socket? */ 285 static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 286 { 287 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 288 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT); 289 } 290 291 extern struct proto tcp_prot; 292 293 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 294 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 295 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 296 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 297 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 298 299 extern void tcp_v4_err(struct sk_buff *skb, u32); 300 301 extern void tcp_shutdown (struct sock *sk, int how); 302 303 extern int tcp_v4_rcv(struct sk_buff *skb); 304 305 extern int tcp_v4_remember_stamp(struct sock *sk); 306 307 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 308 309 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock, 310 struct msghdr *msg, size_t size); 311 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 312 313 extern int tcp_ioctl(struct sock *sk, 314 int cmd, 315 unsigned long arg); 316 317 extern int tcp_rcv_state_process(struct sock *sk, 318 struct sk_buff *skb, 319 struct tcphdr *th, 320 unsigned len); 321 322 extern int tcp_rcv_established(struct sock *sk, 323 struct sk_buff *skb, 324 struct tcphdr *th, 325 unsigned len); 326 327 extern void tcp_rcv_space_adjust(struct sock *sk); 328 329 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 330 331 extern int tcp_twsk_unique(struct sock *sk, 332 struct sock *sktw, void *twp); 333 334 extern void tcp_twsk_destructor(struct sock *sk); 335 336 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 337 struct pipe_inode_info *pipe, size_t len, unsigned int flags); 338 339 static inline void tcp_dec_quickack_mode(struct sock *sk, 340 const unsigned int pkts) 341 { 342 struct inet_connection_sock *icsk = inet_csk(sk); 343 344 if (icsk->icsk_ack.quick) { 345 if (pkts >= icsk->icsk_ack.quick) { 346 icsk->icsk_ack.quick = 0; 347 /* Leaving quickack mode we deflate ATO. */ 348 icsk->icsk_ack.ato = TCP_ATO_MIN; 349 } else 350 icsk->icsk_ack.quick -= pkts; 351 } 352 } 353 354 extern void tcp_enter_quickack_mode(struct sock *sk); 355 356 #define TCP_ECN_OK 1 357 #define TCP_ECN_QUEUE_CWR 2 358 #define TCP_ECN_DEMAND_CWR 4 359 360 static __inline__ void 361 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 362 { 363 if (sysctl_tcp_ecn && th->ece && th->cwr) 364 inet_rsk(req)->ecn_ok = 1; 365 } 366 367 enum tcp_tw_status { 368 TCP_TW_SUCCESS = 0, 369 TCP_TW_RST = 1, 370 TCP_TW_ACK = 2, 371 TCP_TW_SYN = 3 372 }; 373 374 375 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 376 struct sk_buff *skb, 377 const struct tcphdr *th); 378 379 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 380 struct request_sock *req, 381 struct request_sock **prev); 382 extern int tcp_child_process(struct sock *parent, 383 struct sock *child, 384 struct sk_buff *skb); 385 extern int tcp_use_frto(struct sock *sk); 386 extern void tcp_enter_frto(struct sock *sk); 387 extern void tcp_enter_loss(struct sock *sk, int how); 388 extern void tcp_clear_retrans(struct tcp_sock *tp); 389 extern void tcp_update_metrics(struct sock *sk); 390 391 extern void tcp_close(struct sock *sk, 392 long timeout); 393 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 394 395 extern int tcp_getsockopt(struct sock *sk, int level, 396 int optname, 397 char __user *optval, 398 int __user *optlen); 399 extern int tcp_setsockopt(struct sock *sk, int level, 400 int optname, char __user *optval, 401 unsigned int optlen); 402 extern int compat_tcp_getsockopt(struct sock *sk, 403 int level, int optname, 404 char __user *optval, int __user *optlen); 405 extern int compat_tcp_setsockopt(struct sock *sk, 406 int level, int optname, 407 char __user *optval, unsigned int optlen); 408 extern void tcp_set_keepalive(struct sock *sk, int val); 409 extern void tcp_syn_ack_timeout(struct sock *sk, 410 struct request_sock *req); 411 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 412 struct msghdr *msg, 413 size_t len, int nonblock, 414 int flags, int *addr_len); 415 416 extern void tcp_parse_options(struct sk_buff *skb, 417 struct tcp_options_received *opt_rx, 418 u8 **hvpp, 419 int estab); 420 421 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 422 423 /* 424 * TCP v4 functions exported for the inet6 API 425 */ 426 427 extern void tcp_v4_send_check(struct sock *sk, 428 struct sk_buff *skb); 429 430 extern int tcp_v4_conn_request(struct sock *sk, 431 struct sk_buff *skb); 432 433 extern struct sock * tcp_create_openreq_child(struct sock *sk, 434 struct request_sock *req, 435 struct sk_buff *skb); 436 437 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 438 struct sk_buff *skb, 439 struct request_sock *req, 440 struct dst_entry *dst); 441 442 extern int tcp_v4_do_rcv(struct sock *sk, 443 struct sk_buff *skb); 444 445 extern int tcp_v4_connect(struct sock *sk, 446 struct sockaddr *uaddr, 447 int addr_len); 448 449 extern int tcp_connect(struct sock *sk); 450 451 extern struct sk_buff * tcp_make_synack(struct sock *sk, 452 struct dst_entry *dst, 453 struct request_sock *req, 454 struct request_values *rvp); 455 456 extern int tcp_disconnect(struct sock *sk, int flags); 457 458 459 /* From syncookies.c */ 460 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 461 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 462 struct ip_options *opt); 463 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 464 __u16 *mss); 465 466 extern __u32 cookie_init_timestamp(struct request_sock *req); 467 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt); 468 469 /* From net/ipv6/syncookies.c */ 470 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 471 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 472 __u16 *mss); 473 474 /* tcp_output.c */ 475 476 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 477 int nonagle); 478 extern int tcp_may_send_now(struct sock *sk); 479 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 480 extern void tcp_retransmit_timer(struct sock *sk); 481 extern void tcp_xmit_retransmit_queue(struct sock *); 482 extern void tcp_simple_retransmit(struct sock *); 483 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 484 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 485 486 extern void tcp_send_probe0(struct sock *); 487 extern void tcp_send_partial(struct sock *); 488 extern int tcp_write_wakeup(struct sock *); 489 extern void tcp_send_fin(struct sock *sk); 490 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 491 extern int tcp_send_synack(struct sock *); 492 extern void tcp_push_one(struct sock *, unsigned int mss_now); 493 extern void tcp_send_ack(struct sock *sk); 494 extern void tcp_send_delayed_ack(struct sock *sk); 495 496 /* tcp_input.c */ 497 extern void tcp_cwnd_application_limited(struct sock *sk); 498 499 /* tcp_timer.c */ 500 extern void tcp_init_xmit_timers(struct sock *); 501 static inline void tcp_clear_xmit_timers(struct sock *sk) 502 { 503 inet_csk_clear_xmit_timers(sk); 504 } 505 506 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 507 extern unsigned int tcp_current_mss(struct sock *sk); 508 509 /* Bound MSS / TSO packet size with the half of the window */ 510 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 511 { 512 if (tp->max_window && pktsize > (tp->max_window >> 1)) 513 return max(tp->max_window >> 1, 68U - tp->tcp_header_len); 514 else 515 return pktsize; 516 } 517 518 /* tcp.c */ 519 extern void tcp_get_info(struct sock *, struct tcp_info *); 520 521 /* Read 'sendfile()'-style from a TCP socket */ 522 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 523 unsigned int, size_t); 524 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 525 sk_read_actor_t recv_actor); 526 527 extern void tcp_initialize_rcv_mss(struct sock *sk); 528 529 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 530 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 531 extern void tcp_mtup_init(struct sock *sk); 532 533 static inline void tcp_bound_rto(const struct sock *sk) 534 { 535 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 536 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 537 } 538 539 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 540 { 541 return (tp->srtt >> 3) + tp->rttvar; 542 } 543 544 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 545 { 546 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 547 ntohl(TCP_FLAG_ACK) | 548 snd_wnd); 549 } 550 551 static inline void tcp_fast_path_on(struct tcp_sock *tp) 552 { 553 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 554 } 555 556 static inline void tcp_fast_path_check(struct sock *sk) 557 { 558 struct tcp_sock *tp = tcp_sk(sk); 559 560 if (skb_queue_empty(&tp->out_of_order_queue) && 561 tp->rcv_wnd && 562 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 563 !tp->urg_data) 564 tcp_fast_path_on(tp); 565 } 566 567 /* Compute the actual rto_min value */ 568 static inline u32 tcp_rto_min(struct sock *sk) 569 { 570 struct dst_entry *dst = __sk_dst_get(sk); 571 u32 rto_min = TCP_RTO_MIN; 572 573 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 574 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 575 return rto_min; 576 } 577 578 /* Compute the actual receive window we are currently advertising. 579 * Rcv_nxt can be after the window if our peer push more data 580 * than the offered window. 581 */ 582 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 583 { 584 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 585 586 if (win < 0) 587 win = 0; 588 return (u32) win; 589 } 590 591 /* Choose a new window, without checks for shrinking, and without 592 * scaling applied to the result. The caller does these things 593 * if necessary. This is a "raw" window selection. 594 */ 595 extern u32 __tcp_select_window(struct sock *sk); 596 597 /* TCP timestamps are only 32-bits, this causes a slight 598 * complication on 64-bit systems since we store a snapshot 599 * of jiffies in the buffer control blocks below. We decided 600 * to use only the low 32-bits of jiffies and hide the ugly 601 * casts with the following macro. 602 */ 603 #define tcp_time_stamp ((__u32)(jiffies)) 604 605 /* This is what the send packet queuing engine uses to pass 606 * TCP per-packet control information to the transmission 607 * code. We also store the host-order sequence numbers in 608 * here too. This is 36 bytes on 32-bit architectures, 609 * 40 bytes on 64-bit machines, if this grows please adjust 610 * skbuff.h:skbuff->cb[xxx] size appropriately. 611 */ 612 struct tcp_skb_cb { 613 union { 614 struct inet_skb_parm h4; 615 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 616 struct inet6_skb_parm h6; 617 #endif 618 } header; /* For incoming frames */ 619 __u32 seq; /* Starting sequence number */ 620 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 621 __u32 when; /* used to compute rtt's */ 622 __u8 flags; /* TCP header flags. */ 623 624 /* NOTE: These must match up to the flags byte in a 625 * real TCP header. 626 */ 627 #define TCPCB_FLAG_FIN 0x01 628 #define TCPCB_FLAG_SYN 0x02 629 #define TCPCB_FLAG_RST 0x04 630 #define TCPCB_FLAG_PSH 0x08 631 #define TCPCB_FLAG_ACK 0x10 632 #define TCPCB_FLAG_URG 0x20 633 #define TCPCB_FLAG_ECE 0x40 634 #define TCPCB_FLAG_CWR 0x80 635 636 __u8 sacked; /* State flags for SACK/FACK. */ 637 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 638 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 639 #define TCPCB_LOST 0x04 /* SKB is lost */ 640 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 641 642 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 643 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 644 645 __u32 ack_seq; /* Sequence number ACK'd */ 646 }; 647 648 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 649 650 /* Due to TSO, an SKB can be composed of multiple actual 651 * packets. To keep these tracked properly, we use this. 652 */ 653 static inline int tcp_skb_pcount(const struct sk_buff *skb) 654 { 655 return skb_shinfo(skb)->gso_segs; 656 } 657 658 /* This is valid iff tcp_skb_pcount() > 1. */ 659 static inline int tcp_skb_mss(const struct sk_buff *skb) 660 { 661 return skb_shinfo(skb)->gso_size; 662 } 663 664 /* Events passed to congestion control interface */ 665 enum tcp_ca_event { 666 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 667 CA_EVENT_CWND_RESTART, /* congestion window restart */ 668 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 669 CA_EVENT_FRTO, /* fast recovery timeout */ 670 CA_EVENT_LOSS, /* loss timeout */ 671 CA_EVENT_FAST_ACK, /* in sequence ack */ 672 CA_EVENT_SLOW_ACK, /* other ack */ 673 }; 674 675 /* 676 * Interface for adding new TCP congestion control handlers 677 */ 678 #define TCP_CA_NAME_MAX 16 679 #define TCP_CA_MAX 128 680 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 681 682 #define TCP_CONG_NON_RESTRICTED 0x1 683 #define TCP_CONG_RTT_STAMP 0x2 684 685 struct tcp_congestion_ops { 686 struct list_head list; 687 unsigned long flags; 688 689 /* initialize private data (optional) */ 690 void (*init)(struct sock *sk); 691 /* cleanup private data (optional) */ 692 void (*release)(struct sock *sk); 693 694 /* return slow start threshold (required) */ 695 u32 (*ssthresh)(struct sock *sk); 696 /* lower bound for congestion window (optional) */ 697 u32 (*min_cwnd)(const struct sock *sk); 698 /* do new cwnd calculation (required) */ 699 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 700 /* call before changing ca_state (optional) */ 701 void (*set_state)(struct sock *sk, u8 new_state); 702 /* call when cwnd event occurs (optional) */ 703 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 704 /* new value of cwnd after loss (optional) */ 705 u32 (*undo_cwnd)(struct sock *sk); 706 /* hook for packet ack accounting (optional) */ 707 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 708 /* get info for inet_diag (optional) */ 709 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 710 711 char name[TCP_CA_NAME_MAX]; 712 struct module *owner; 713 }; 714 715 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 716 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 717 718 extern void tcp_init_congestion_control(struct sock *sk); 719 extern void tcp_cleanup_congestion_control(struct sock *sk); 720 extern int tcp_set_default_congestion_control(const char *name); 721 extern void tcp_get_default_congestion_control(char *name); 722 extern void tcp_get_available_congestion_control(char *buf, size_t len); 723 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 724 extern int tcp_set_allowed_congestion_control(char *allowed); 725 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 726 extern void tcp_slow_start(struct tcp_sock *tp); 727 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 728 729 extern struct tcp_congestion_ops tcp_init_congestion_ops; 730 extern u32 tcp_reno_ssthresh(struct sock *sk); 731 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 732 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 733 extern struct tcp_congestion_ops tcp_reno; 734 735 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 736 { 737 struct inet_connection_sock *icsk = inet_csk(sk); 738 739 if (icsk->icsk_ca_ops->set_state) 740 icsk->icsk_ca_ops->set_state(sk, ca_state); 741 icsk->icsk_ca_state = ca_state; 742 } 743 744 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 745 { 746 const struct inet_connection_sock *icsk = inet_csk(sk); 747 748 if (icsk->icsk_ca_ops->cwnd_event) 749 icsk->icsk_ca_ops->cwnd_event(sk, event); 750 } 751 752 /* These functions determine how the current flow behaves in respect of SACK 753 * handling. SACK is negotiated with the peer, and therefore it can vary 754 * between different flows. 755 * 756 * tcp_is_sack - SACK enabled 757 * tcp_is_reno - No SACK 758 * tcp_is_fack - FACK enabled, implies SACK enabled 759 */ 760 static inline int tcp_is_sack(const struct tcp_sock *tp) 761 { 762 return tp->rx_opt.sack_ok; 763 } 764 765 static inline int tcp_is_reno(const struct tcp_sock *tp) 766 { 767 return !tcp_is_sack(tp); 768 } 769 770 static inline int tcp_is_fack(const struct tcp_sock *tp) 771 { 772 return tp->rx_opt.sack_ok & 2; 773 } 774 775 static inline void tcp_enable_fack(struct tcp_sock *tp) 776 { 777 tp->rx_opt.sack_ok |= 2; 778 } 779 780 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 781 { 782 return tp->sacked_out + tp->lost_out; 783 } 784 785 /* This determines how many packets are "in the network" to the best 786 * of our knowledge. In many cases it is conservative, but where 787 * detailed information is available from the receiver (via SACK 788 * blocks etc.) we can make more aggressive calculations. 789 * 790 * Use this for decisions involving congestion control, use just 791 * tp->packets_out to determine if the send queue is empty or not. 792 * 793 * Read this equation as: 794 * 795 * "Packets sent once on transmission queue" MINUS 796 * "Packets left network, but not honestly ACKed yet" PLUS 797 * "Packets fast retransmitted" 798 */ 799 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 800 { 801 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 802 } 803 804 #define TCP_INFINITE_SSTHRESH 0x7fffffff 805 806 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 807 { 808 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 809 } 810 811 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 812 * The exception is rate halving phase, when cwnd is decreasing towards 813 * ssthresh. 814 */ 815 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 816 { 817 const struct tcp_sock *tp = tcp_sk(sk); 818 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 819 return tp->snd_ssthresh; 820 else 821 return max(tp->snd_ssthresh, 822 ((tp->snd_cwnd >> 1) + 823 (tp->snd_cwnd >> 2))); 824 } 825 826 /* Use define here intentionally to get WARN_ON location shown at the caller */ 827 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 828 829 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 830 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 831 832 /* Slow start with delack produces 3 packets of burst, so that 833 * it is safe "de facto". This will be the default - same as 834 * the default reordering threshold - but if reordering increases, 835 * we must be able to allow cwnd to burst at least this much in order 836 * to not pull it back when holes are filled. 837 */ 838 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 839 { 840 return tp->reordering; 841 } 842 843 /* Returns end sequence number of the receiver's advertised window */ 844 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 845 { 846 return tp->snd_una + tp->snd_wnd; 847 } 848 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 849 850 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 851 const struct sk_buff *skb) 852 { 853 if (skb->len < mss) 854 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 855 } 856 857 static inline void tcp_check_probe_timer(struct sock *sk) 858 { 859 struct tcp_sock *tp = tcp_sk(sk); 860 const struct inet_connection_sock *icsk = inet_csk(sk); 861 862 if (!tp->packets_out && !icsk->icsk_pending) 863 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 864 icsk->icsk_rto, TCP_RTO_MAX); 865 } 866 867 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 868 { 869 tp->snd_wl1 = seq; 870 } 871 872 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 873 { 874 tp->snd_wl1 = seq; 875 } 876 877 /* 878 * Calculate(/check) TCP checksum 879 */ 880 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 881 __be32 daddr, __wsum base) 882 { 883 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 884 } 885 886 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 887 { 888 return __skb_checksum_complete(skb); 889 } 890 891 static inline int tcp_checksum_complete(struct sk_buff *skb) 892 { 893 return !skb_csum_unnecessary(skb) && 894 __tcp_checksum_complete(skb); 895 } 896 897 /* Prequeue for VJ style copy to user, combined with checksumming. */ 898 899 static inline void tcp_prequeue_init(struct tcp_sock *tp) 900 { 901 tp->ucopy.task = NULL; 902 tp->ucopy.len = 0; 903 tp->ucopy.memory = 0; 904 skb_queue_head_init(&tp->ucopy.prequeue); 905 #ifdef CONFIG_NET_DMA 906 tp->ucopy.dma_chan = NULL; 907 tp->ucopy.wakeup = 0; 908 tp->ucopy.pinned_list = NULL; 909 tp->ucopy.dma_cookie = 0; 910 #endif 911 } 912 913 /* Packet is added to VJ-style prequeue for processing in process 914 * context, if a reader task is waiting. Apparently, this exciting 915 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 916 * failed somewhere. Latency? Burstiness? Well, at least now we will 917 * see, why it failed. 8)8) --ANK 918 * 919 * NOTE: is this not too big to inline? 920 */ 921 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 922 { 923 struct tcp_sock *tp = tcp_sk(sk); 924 925 if (sysctl_tcp_low_latency || !tp->ucopy.task) 926 return 0; 927 928 __skb_queue_tail(&tp->ucopy.prequeue, skb); 929 tp->ucopy.memory += skb->truesize; 930 if (tp->ucopy.memory > sk->sk_rcvbuf) { 931 struct sk_buff *skb1; 932 933 BUG_ON(sock_owned_by_user(sk)); 934 935 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 936 sk_backlog_rcv(sk, skb1); 937 NET_INC_STATS_BH(sock_net(sk), 938 LINUX_MIB_TCPPREQUEUEDROPPED); 939 } 940 941 tp->ucopy.memory = 0; 942 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 943 wake_up_interruptible_sync_poll(sk_sleep(sk), 944 POLLIN | POLLRDNORM | POLLRDBAND); 945 if (!inet_csk_ack_scheduled(sk)) 946 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 947 (3 * tcp_rto_min(sk)) / 4, 948 TCP_RTO_MAX); 949 } 950 return 1; 951 } 952 953 954 #undef STATE_TRACE 955 956 #ifdef STATE_TRACE 957 static const char *statename[]={ 958 "Unused","Established","Syn Sent","Syn Recv", 959 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 960 "Close Wait","Last ACK","Listen","Closing" 961 }; 962 #endif 963 extern void tcp_set_state(struct sock *sk, int state); 964 965 extern void tcp_done(struct sock *sk); 966 967 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 968 { 969 rx_opt->dsack = 0; 970 rx_opt->num_sacks = 0; 971 } 972 973 /* Determine a window scaling and initial window to offer. */ 974 extern void tcp_select_initial_window(int __space, __u32 mss, 975 __u32 *rcv_wnd, __u32 *window_clamp, 976 int wscale_ok, __u8 *rcv_wscale, 977 __u32 init_rcv_wnd); 978 979 static inline int tcp_win_from_space(int space) 980 { 981 return sysctl_tcp_adv_win_scale<=0 ? 982 (space>>(-sysctl_tcp_adv_win_scale)) : 983 space - (space>>sysctl_tcp_adv_win_scale); 984 } 985 986 /* Note: caller must be prepared to deal with negative returns */ 987 static inline int tcp_space(const struct sock *sk) 988 { 989 return tcp_win_from_space(sk->sk_rcvbuf - 990 atomic_read(&sk->sk_rmem_alloc)); 991 } 992 993 static inline int tcp_full_space(const struct sock *sk) 994 { 995 return tcp_win_from_space(sk->sk_rcvbuf); 996 } 997 998 static inline void tcp_openreq_init(struct request_sock *req, 999 struct tcp_options_received *rx_opt, 1000 struct sk_buff *skb) 1001 { 1002 struct inet_request_sock *ireq = inet_rsk(req); 1003 1004 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1005 req->cookie_ts = 0; 1006 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1007 req->mss = rx_opt->mss_clamp; 1008 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1009 ireq->tstamp_ok = rx_opt->tstamp_ok; 1010 ireq->sack_ok = rx_opt->sack_ok; 1011 ireq->snd_wscale = rx_opt->snd_wscale; 1012 ireq->wscale_ok = rx_opt->wscale_ok; 1013 ireq->acked = 0; 1014 ireq->ecn_ok = 0; 1015 ireq->rmt_port = tcp_hdr(skb)->source; 1016 ireq->loc_port = tcp_hdr(skb)->dest; 1017 } 1018 1019 extern void tcp_enter_memory_pressure(struct sock *sk); 1020 1021 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1022 { 1023 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1024 } 1025 1026 static inline int keepalive_time_when(const struct tcp_sock *tp) 1027 { 1028 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1029 } 1030 1031 static inline int keepalive_probes(const struct tcp_sock *tp) 1032 { 1033 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1034 } 1035 1036 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1037 { 1038 const struct inet_connection_sock *icsk = &tp->inet_conn; 1039 1040 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1041 tcp_time_stamp - tp->rcv_tstamp); 1042 } 1043 1044 static inline int tcp_fin_time(const struct sock *sk) 1045 { 1046 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1047 const int rto = inet_csk(sk)->icsk_rto; 1048 1049 if (fin_timeout < (rto << 2) - (rto >> 1)) 1050 fin_timeout = (rto << 2) - (rto >> 1); 1051 1052 return fin_timeout; 1053 } 1054 1055 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1056 int paws_win) 1057 { 1058 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1059 return 1; 1060 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1061 return 1; 1062 1063 return 0; 1064 } 1065 1066 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1067 int rst) 1068 { 1069 if (tcp_paws_check(rx_opt, 0)) 1070 return 0; 1071 1072 /* RST segments are not recommended to carry timestamp, 1073 and, if they do, it is recommended to ignore PAWS because 1074 "their cleanup function should take precedence over timestamps." 1075 Certainly, it is mistake. It is necessary to understand the reasons 1076 of this constraint to relax it: if peer reboots, clock may go 1077 out-of-sync and half-open connections will not be reset. 1078 Actually, the problem would be not existing if all 1079 the implementations followed draft about maintaining clock 1080 via reboots. Linux-2.2 DOES NOT! 1081 1082 However, we can relax time bounds for RST segments to MSL. 1083 */ 1084 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1085 return 0; 1086 return 1; 1087 } 1088 1089 #define TCP_CHECK_TIMER(sk) do { } while (0) 1090 1091 static inline void tcp_mib_init(struct net *net) 1092 { 1093 /* See RFC 2012 */ 1094 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1095 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1096 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1097 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1098 } 1099 1100 /* from STCP */ 1101 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1102 { 1103 tp->lost_skb_hint = NULL; 1104 tp->scoreboard_skb_hint = NULL; 1105 } 1106 1107 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1108 { 1109 tcp_clear_retrans_hints_partial(tp); 1110 tp->retransmit_skb_hint = NULL; 1111 } 1112 1113 /* MD5 Signature */ 1114 struct crypto_hash; 1115 1116 /* - key database */ 1117 struct tcp_md5sig_key { 1118 u8 *key; 1119 u8 keylen; 1120 }; 1121 1122 struct tcp4_md5sig_key { 1123 struct tcp_md5sig_key base; 1124 __be32 addr; 1125 }; 1126 1127 struct tcp6_md5sig_key { 1128 struct tcp_md5sig_key base; 1129 #if 0 1130 u32 scope_id; /* XXX */ 1131 #endif 1132 struct in6_addr addr; 1133 }; 1134 1135 /* - sock block */ 1136 struct tcp_md5sig_info { 1137 struct tcp4_md5sig_key *keys4; 1138 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1139 struct tcp6_md5sig_key *keys6; 1140 u32 entries6; 1141 u32 alloced6; 1142 #endif 1143 u32 entries4; 1144 u32 alloced4; 1145 }; 1146 1147 /* - pseudo header */ 1148 struct tcp4_pseudohdr { 1149 __be32 saddr; 1150 __be32 daddr; 1151 __u8 pad; 1152 __u8 protocol; 1153 __be16 len; 1154 }; 1155 1156 struct tcp6_pseudohdr { 1157 struct in6_addr saddr; 1158 struct in6_addr daddr; 1159 __be32 len; 1160 __be32 protocol; /* including padding */ 1161 }; 1162 1163 union tcp_md5sum_block { 1164 struct tcp4_pseudohdr ip4; 1165 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1166 struct tcp6_pseudohdr ip6; 1167 #endif 1168 }; 1169 1170 /* - pool: digest algorithm, hash description and scratch buffer */ 1171 struct tcp_md5sig_pool { 1172 struct hash_desc md5_desc; 1173 union tcp_md5sum_block md5_blk; 1174 }; 1175 1176 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1177 1178 /* - functions */ 1179 extern int tcp_v4_md5_hash_skb(char *md5_hash, 1180 struct tcp_md5sig_key *key, 1181 struct sock *sk, 1182 struct request_sock *req, 1183 struct sk_buff *skb); 1184 1185 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1186 struct sock *addr_sk); 1187 1188 extern int tcp_v4_md5_do_add(struct sock *sk, 1189 __be32 addr, 1190 u8 *newkey, 1191 u8 newkeylen); 1192 1193 extern int tcp_v4_md5_do_del(struct sock *sk, 1194 __be32 addr); 1195 1196 #ifdef CONFIG_TCP_MD5SIG 1197 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1198 &(struct tcp_md5sig_key) { \ 1199 .key = (twsk)->tw_md5_key, \ 1200 .keylen = (twsk)->tw_md5_keylen, \ 1201 } : NULL) 1202 #else 1203 #define tcp_twsk_md5_key(twsk) NULL 1204 #endif 1205 1206 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1207 extern void tcp_free_md5sig_pool(void); 1208 1209 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1210 extern void tcp_put_md5sig_pool(void); 1211 1212 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1213 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1214 unsigned header_len); 1215 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1216 struct tcp_md5sig_key *key); 1217 1218 /* write queue abstraction */ 1219 static inline void tcp_write_queue_purge(struct sock *sk) 1220 { 1221 struct sk_buff *skb; 1222 1223 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1224 sk_wmem_free_skb(sk, skb); 1225 sk_mem_reclaim(sk); 1226 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1227 } 1228 1229 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1230 { 1231 return skb_peek(&sk->sk_write_queue); 1232 } 1233 1234 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1235 { 1236 return skb_peek_tail(&sk->sk_write_queue); 1237 } 1238 1239 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1240 { 1241 return skb_queue_next(&sk->sk_write_queue, skb); 1242 } 1243 1244 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1245 { 1246 return skb_queue_prev(&sk->sk_write_queue, skb); 1247 } 1248 1249 #define tcp_for_write_queue(skb, sk) \ 1250 skb_queue_walk(&(sk)->sk_write_queue, skb) 1251 1252 #define tcp_for_write_queue_from(skb, sk) \ 1253 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1254 1255 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1256 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1257 1258 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1259 { 1260 return sk->sk_send_head; 1261 } 1262 1263 static inline bool tcp_skb_is_last(const struct sock *sk, 1264 const struct sk_buff *skb) 1265 { 1266 return skb_queue_is_last(&sk->sk_write_queue, skb); 1267 } 1268 1269 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1270 { 1271 if (tcp_skb_is_last(sk, skb)) 1272 sk->sk_send_head = NULL; 1273 else 1274 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1275 } 1276 1277 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1278 { 1279 if (sk->sk_send_head == skb_unlinked) 1280 sk->sk_send_head = NULL; 1281 } 1282 1283 static inline void tcp_init_send_head(struct sock *sk) 1284 { 1285 sk->sk_send_head = NULL; 1286 } 1287 1288 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1289 { 1290 __skb_queue_tail(&sk->sk_write_queue, skb); 1291 } 1292 1293 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1294 { 1295 __tcp_add_write_queue_tail(sk, skb); 1296 1297 /* Queue it, remembering where we must start sending. */ 1298 if (sk->sk_send_head == NULL) { 1299 sk->sk_send_head = skb; 1300 1301 if (tcp_sk(sk)->highest_sack == NULL) 1302 tcp_sk(sk)->highest_sack = skb; 1303 } 1304 } 1305 1306 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1307 { 1308 __skb_queue_head(&sk->sk_write_queue, skb); 1309 } 1310 1311 /* Insert buff after skb on the write queue of sk. */ 1312 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1313 struct sk_buff *buff, 1314 struct sock *sk) 1315 { 1316 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1317 } 1318 1319 /* Insert new before skb on the write queue of sk. */ 1320 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1321 struct sk_buff *skb, 1322 struct sock *sk) 1323 { 1324 __skb_queue_before(&sk->sk_write_queue, skb, new); 1325 1326 if (sk->sk_send_head == skb) 1327 sk->sk_send_head = new; 1328 } 1329 1330 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1331 { 1332 __skb_unlink(skb, &sk->sk_write_queue); 1333 } 1334 1335 static inline int tcp_write_queue_empty(struct sock *sk) 1336 { 1337 return skb_queue_empty(&sk->sk_write_queue); 1338 } 1339 1340 static inline void tcp_push_pending_frames(struct sock *sk) 1341 { 1342 if (tcp_send_head(sk)) { 1343 struct tcp_sock *tp = tcp_sk(sk); 1344 1345 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1346 } 1347 } 1348 1349 /* Start sequence of the highest skb with SACKed bit, valid only if 1350 * sacked > 0 or when the caller has ensured validity by itself. 1351 */ 1352 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1353 { 1354 if (!tp->sacked_out) 1355 return tp->snd_una; 1356 1357 if (tp->highest_sack == NULL) 1358 return tp->snd_nxt; 1359 1360 return TCP_SKB_CB(tp->highest_sack)->seq; 1361 } 1362 1363 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1364 { 1365 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1366 tcp_write_queue_next(sk, skb); 1367 } 1368 1369 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1370 { 1371 return tcp_sk(sk)->highest_sack; 1372 } 1373 1374 static inline void tcp_highest_sack_reset(struct sock *sk) 1375 { 1376 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1377 } 1378 1379 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1380 static inline void tcp_highest_sack_combine(struct sock *sk, 1381 struct sk_buff *old, 1382 struct sk_buff *new) 1383 { 1384 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1385 tcp_sk(sk)->highest_sack = new; 1386 } 1387 1388 /* Determines whether this is a thin stream (which may suffer from 1389 * increased latency). Used to trigger latency-reducing mechanisms. 1390 */ 1391 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1392 { 1393 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1394 } 1395 1396 /* /proc */ 1397 enum tcp_seq_states { 1398 TCP_SEQ_STATE_LISTENING, 1399 TCP_SEQ_STATE_OPENREQ, 1400 TCP_SEQ_STATE_ESTABLISHED, 1401 TCP_SEQ_STATE_TIME_WAIT, 1402 }; 1403 1404 struct tcp_seq_afinfo { 1405 char *name; 1406 sa_family_t family; 1407 struct file_operations seq_fops; 1408 struct seq_operations seq_ops; 1409 }; 1410 1411 struct tcp_iter_state { 1412 struct seq_net_private p; 1413 sa_family_t family; 1414 enum tcp_seq_states state; 1415 struct sock *syn_wait_sk; 1416 int bucket, sbucket, num, uid; 1417 }; 1418 1419 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1420 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1421 1422 extern struct request_sock_ops tcp_request_sock_ops; 1423 extern struct request_sock_ops tcp6_request_sock_ops; 1424 1425 extern void tcp_v4_destroy_sock(struct sock *sk); 1426 1427 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1428 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1429 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1430 struct sk_buff *skb); 1431 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1432 struct sk_buff *skb); 1433 extern int tcp_gro_complete(struct sk_buff *skb); 1434 extern int tcp4_gro_complete(struct sk_buff *skb); 1435 1436 #ifdef CONFIG_PROC_FS 1437 extern int tcp4_proc_init(void); 1438 extern void tcp4_proc_exit(void); 1439 #endif 1440 1441 /* TCP af-specific functions */ 1442 struct tcp_sock_af_ops { 1443 #ifdef CONFIG_TCP_MD5SIG 1444 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1445 struct sock *addr_sk); 1446 int (*calc_md5_hash) (char *location, 1447 struct tcp_md5sig_key *md5, 1448 struct sock *sk, 1449 struct request_sock *req, 1450 struct sk_buff *skb); 1451 int (*md5_add) (struct sock *sk, 1452 struct sock *addr_sk, 1453 u8 *newkey, 1454 u8 len); 1455 int (*md5_parse) (struct sock *sk, 1456 char __user *optval, 1457 int optlen); 1458 #endif 1459 }; 1460 1461 struct tcp_request_sock_ops { 1462 #ifdef CONFIG_TCP_MD5SIG 1463 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1464 struct request_sock *req); 1465 int (*calc_md5_hash) (char *location, 1466 struct tcp_md5sig_key *md5, 1467 struct sock *sk, 1468 struct request_sock *req, 1469 struct sk_buff *skb); 1470 #endif 1471 }; 1472 1473 /* Using SHA1 for now, define some constants. 1474 */ 1475 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1476 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1477 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1478 1479 extern int tcp_cookie_generator(u32 *bakery); 1480 1481 /** 1482 * struct tcp_cookie_values - each socket needs extra space for the 1483 * cookies, together with (optional) space for any SYN data. 1484 * 1485 * A tcp_sock contains a pointer to the current value, and this is 1486 * cloned to the tcp_timewait_sock. 1487 * 1488 * @cookie_pair: variable data from the option exchange. 1489 * 1490 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1491 * indicates default (sysctl_tcp_cookie_size). 1492 * After cookie sent, remembers size of cookie. 1493 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1494 * 1495 * @s_data_desired: user specified tcpct_s_data_desired. When the 1496 * constant payload is specified (@s_data_constant), 1497 * holds its length instead. 1498 * Range 0 to TCP_MSS_DESIRED. 1499 * 1500 * @s_data_payload: constant data that is to be included in the 1501 * payload of SYN or SYNACK segments when the 1502 * cookie option is present. 1503 */ 1504 struct tcp_cookie_values { 1505 struct kref kref; 1506 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1507 u8 cookie_pair_size; 1508 u8 cookie_desired; 1509 u16 s_data_desired:11, 1510 s_data_constant:1, 1511 s_data_in:1, 1512 s_data_out:1, 1513 s_data_unused:2; 1514 u8 s_data_payload[0]; 1515 }; 1516 1517 static inline void tcp_cookie_values_release(struct kref *kref) 1518 { 1519 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1520 } 1521 1522 /* The length of constant payload data. Note that s_data_desired is 1523 * overloaded, depending on s_data_constant: either the length of constant 1524 * data (returned here) or the limit on variable data. 1525 */ 1526 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1527 { 1528 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1529 ? tp->cookie_values->s_data_desired 1530 : 0; 1531 } 1532 1533 /** 1534 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1535 * 1536 * As tcp_request_sock has already been extended in other places, the 1537 * only remaining method is to pass stack values along as function 1538 * parameters. These parameters are not needed after sending SYNACK. 1539 * 1540 * @cookie_bakery: cryptographic secret and message workspace. 1541 * 1542 * @cookie_plus: bytes in authenticator/cookie option, copied from 1543 * struct tcp_options_received (above). 1544 */ 1545 struct tcp_extend_values { 1546 struct request_values rv; 1547 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1548 u8 cookie_plus:6, 1549 cookie_out_never:1, 1550 cookie_in_always:1; 1551 }; 1552 1553 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1554 { 1555 return (struct tcp_extend_values *)rvp; 1556 } 1557 1558 extern void tcp_v4_init(void); 1559 extern void tcp_init(void); 1560 1561 #endif /* _TCP_H */ 1562