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) 297 298 extern void tcp_v4_err(struct sk_buff *skb, u32); 299 300 extern void tcp_shutdown (struct sock *sk, int how); 301 302 extern int tcp_v4_rcv(struct sk_buff *skb); 303 304 extern int tcp_v4_remember_stamp(struct sock *sk); 305 306 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 307 308 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock, 309 struct msghdr *msg, size_t size); 310 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 311 312 extern int tcp_ioctl(struct sock *sk, 313 int cmd, 314 unsigned long arg); 315 316 extern int tcp_rcv_state_process(struct sock *sk, 317 struct sk_buff *skb, 318 struct tcphdr *th, 319 unsigned len); 320 321 extern int tcp_rcv_established(struct sock *sk, 322 struct sk_buff *skb, 323 struct tcphdr *th, 324 unsigned len); 325 326 extern void tcp_rcv_space_adjust(struct sock *sk); 327 328 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 329 330 extern int tcp_twsk_unique(struct sock *sk, 331 struct sock *sktw, void *twp); 332 333 extern void tcp_twsk_destructor(struct sock *sk); 334 335 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 336 struct pipe_inode_info *pipe, size_t len, unsigned int flags); 337 338 static inline void tcp_dec_quickack_mode(struct sock *sk, 339 const unsigned int pkts) 340 { 341 struct inet_connection_sock *icsk = inet_csk(sk); 342 343 if (icsk->icsk_ack.quick) { 344 if (pkts >= icsk->icsk_ack.quick) { 345 icsk->icsk_ack.quick = 0; 346 /* Leaving quickack mode we deflate ATO. */ 347 icsk->icsk_ack.ato = TCP_ATO_MIN; 348 } else 349 icsk->icsk_ack.quick -= pkts; 350 } 351 } 352 353 extern void tcp_enter_quickack_mode(struct sock *sk); 354 355 #define TCP_ECN_OK 1 356 #define TCP_ECN_QUEUE_CWR 2 357 #define TCP_ECN_DEMAND_CWR 4 358 359 static __inline__ void 360 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 361 { 362 if (sysctl_tcp_ecn && th->ece && th->cwr) 363 inet_rsk(req)->ecn_ok = 1; 364 } 365 366 enum tcp_tw_status { 367 TCP_TW_SUCCESS = 0, 368 TCP_TW_RST = 1, 369 TCP_TW_ACK = 2, 370 TCP_TW_SYN = 3 371 }; 372 373 374 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 375 struct sk_buff *skb, 376 const struct tcphdr *th); 377 378 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 379 struct request_sock *req, 380 struct request_sock **prev); 381 extern int tcp_child_process(struct sock *parent, 382 struct sock *child, 383 struct sk_buff *skb); 384 extern int tcp_use_frto(struct sock *sk); 385 extern void tcp_enter_frto(struct sock *sk); 386 extern void tcp_enter_loss(struct sock *sk, int how); 387 extern void tcp_clear_retrans(struct tcp_sock *tp); 388 extern void tcp_update_metrics(struct sock *sk); 389 390 extern void tcp_close(struct sock *sk, 391 long timeout); 392 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 393 394 extern int tcp_getsockopt(struct sock *sk, int level, 395 int optname, 396 char __user *optval, 397 int __user *optlen); 398 extern int tcp_setsockopt(struct sock *sk, int level, 399 int optname, char __user *optval, 400 unsigned int optlen); 401 extern int compat_tcp_getsockopt(struct sock *sk, 402 int level, int optname, 403 char __user *optval, int __user *optlen); 404 extern int compat_tcp_setsockopt(struct sock *sk, 405 int level, int optname, 406 char __user *optval, unsigned int optlen); 407 extern void tcp_set_keepalive(struct sock *sk, int val); 408 extern void tcp_syn_ack_timeout(struct sock *sk, 409 struct request_sock *req); 410 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 411 struct msghdr *msg, 412 size_t len, int nonblock, 413 int flags, int *addr_len); 414 415 extern void tcp_parse_options(struct sk_buff *skb, 416 struct tcp_options_received *opt_rx, 417 u8 **hvpp, 418 int estab); 419 420 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 421 422 /* 423 * TCP v4 functions exported for the inet6 API 424 */ 425 426 extern void tcp_v4_send_check(struct sock *sk, int len, 427 struct sk_buff *skb); 428 429 extern int tcp_v4_conn_request(struct sock *sk, 430 struct sk_buff *skb); 431 432 extern struct sock * tcp_create_openreq_child(struct sock *sk, 433 struct request_sock *req, 434 struct sk_buff *skb); 435 436 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 437 struct sk_buff *skb, 438 struct request_sock *req, 439 struct dst_entry *dst); 440 441 extern int tcp_v4_do_rcv(struct sock *sk, 442 struct sk_buff *skb); 443 444 extern int tcp_v4_connect(struct sock *sk, 445 struct sockaddr *uaddr, 446 int addr_len); 447 448 extern int tcp_connect(struct sock *sk); 449 450 extern struct sk_buff * tcp_make_synack(struct sock *sk, 451 struct dst_entry *dst, 452 struct request_sock *req, 453 struct request_values *rvp); 454 455 extern int tcp_disconnect(struct sock *sk, int flags); 456 457 458 /* From syncookies.c */ 459 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 460 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 461 struct ip_options *opt); 462 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 463 __u16 *mss); 464 465 extern __u32 cookie_init_timestamp(struct request_sock *req); 466 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt); 467 468 /* From net/ipv6/syncookies.c */ 469 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 470 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 471 __u16 *mss); 472 473 /* tcp_output.c */ 474 475 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 476 int nonagle); 477 extern int tcp_may_send_now(struct sock *sk); 478 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 479 extern void tcp_retransmit_timer(struct sock *sk); 480 extern void tcp_xmit_retransmit_queue(struct sock *); 481 extern void tcp_simple_retransmit(struct sock *); 482 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 483 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 484 485 extern void tcp_send_probe0(struct sock *); 486 extern void tcp_send_partial(struct sock *); 487 extern int tcp_write_wakeup(struct sock *); 488 extern void tcp_send_fin(struct sock *sk); 489 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 490 extern int tcp_send_synack(struct sock *); 491 extern void tcp_push_one(struct sock *, unsigned int mss_now); 492 extern void tcp_send_ack(struct sock *sk); 493 extern void tcp_send_delayed_ack(struct sock *sk); 494 495 /* tcp_input.c */ 496 extern void tcp_cwnd_application_limited(struct sock *sk); 497 498 /* tcp_timer.c */ 499 extern void tcp_init_xmit_timers(struct sock *); 500 static inline void tcp_clear_xmit_timers(struct sock *sk) 501 { 502 inet_csk_clear_xmit_timers(sk); 503 } 504 505 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 506 extern unsigned int tcp_current_mss(struct sock *sk); 507 508 /* Bound MSS / TSO packet size with the half of the window */ 509 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 510 { 511 if (tp->max_window && pktsize > (tp->max_window >> 1)) 512 return max(tp->max_window >> 1, 68U - tp->tcp_header_len); 513 else 514 return pktsize; 515 } 516 517 /* tcp.c */ 518 extern void tcp_get_info(struct sock *, struct tcp_info *); 519 520 /* Read 'sendfile()'-style from a TCP socket */ 521 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 522 unsigned int, size_t); 523 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 524 sk_read_actor_t recv_actor); 525 526 extern void tcp_initialize_rcv_mss(struct sock *sk); 527 528 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 529 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 530 extern void tcp_mtup_init(struct sock *sk); 531 532 static inline void tcp_bound_rto(const struct sock *sk) 533 { 534 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 535 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 536 } 537 538 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 539 { 540 return (tp->srtt >> 3) + tp->rttvar; 541 } 542 543 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 544 { 545 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 546 ntohl(TCP_FLAG_ACK) | 547 snd_wnd); 548 } 549 550 static inline void tcp_fast_path_on(struct tcp_sock *tp) 551 { 552 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 553 } 554 555 static inline void tcp_fast_path_check(struct sock *sk) 556 { 557 struct tcp_sock *tp = tcp_sk(sk); 558 559 if (skb_queue_empty(&tp->out_of_order_queue) && 560 tp->rcv_wnd && 561 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 562 !tp->urg_data) 563 tcp_fast_path_on(tp); 564 } 565 566 /* Compute the actual rto_min value */ 567 static inline u32 tcp_rto_min(struct sock *sk) 568 { 569 struct dst_entry *dst = __sk_dst_get(sk); 570 u32 rto_min = TCP_RTO_MIN; 571 572 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 573 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 574 return rto_min; 575 } 576 577 /* Compute the actual receive window we are currently advertising. 578 * Rcv_nxt can be after the window if our peer push more data 579 * than the offered window. 580 */ 581 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 582 { 583 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 584 585 if (win < 0) 586 win = 0; 587 return (u32) win; 588 } 589 590 /* Choose a new window, without checks for shrinking, and without 591 * scaling applied to the result. The caller does these things 592 * if necessary. This is a "raw" window selection. 593 */ 594 extern u32 __tcp_select_window(struct sock *sk); 595 596 /* TCP timestamps are only 32-bits, this causes a slight 597 * complication on 64-bit systems since we store a snapshot 598 * of jiffies in the buffer control blocks below. We decided 599 * to use only the low 32-bits of jiffies and hide the ugly 600 * casts with the following macro. 601 */ 602 #define tcp_time_stamp ((__u32)(jiffies)) 603 604 /* This is what the send packet queuing engine uses to pass 605 * TCP per-packet control information to the transmission 606 * code. We also store the host-order sequence numbers in 607 * here too. This is 36 bytes on 32-bit architectures, 608 * 40 bytes on 64-bit machines, if this grows please adjust 609 * skbuff.h:skbuff->cb[xxx] size appropriately. 610 */ 611 struct tcp_skb_cb { 612 union { 613 struct inet_skb_parm h4; 614 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 615 struct inet6_skb_parm h6; 616 #endif 617 } header; /* For incoming frames */ 618 __u32 seq; /* Starting sequence number */ 619 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 620 __u32 when; /* used to compute rtt's */ 621 __u8 flags; /* TCP header flags. */ 622 623 /* NOTE: These must match up to the flags byte in a 624 * real TCP header. 625 */ 626 #define TCPCB_FLAG_FIN 0x01 627 #define TCPCB_FLAG_SYN 0x02 628 #define TCPCB_FLAG_RST 0x04 629 #define TCPCB_FLAG_PSH 0x08 630 #define TCPCB_FLAG_ACK 0x10 631 #define TCPCB_FLAG_URG 0x20 632 #define TCPCB_FLAG_ECE 0x40 633 #define TCPCB_FLAG_CWR 0x80 634 635 __u8 sacked; /* State flags for SACK/FACK. */ 636 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 637 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 638 #define TCPCB_LOST 0x04 /* SKB is lost */ 639 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 640 641 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 642 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 643 644 __u32 ack_seq; /* Sequence number ACK'd */ 645 }; 646 647 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 648 649 /* Due to TSO, an SKB can be composed of multiple actual 650 * packets. To keep these tracked properly, we use this. 651 */ 652 static inline int tcp_skb_pcount(const struct sk_buff *skb) 653 { 654 return skb_shinfo(skb)->gso_segs; 655 } 656 657 /* This is valid iff tcp_skb_pcount() > 1. */ 658 static inline int tcp_skb_mss(const struct sk_buff *skb) 659 { 660 return skb_shinfo(skb)->gso_size; 661 } 662 663 /* Events passed to congestion control interface */ 664 enum tcp_ca_event { 665 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 666 CA_EVENT_CWND_RESTART, /* congestion window restart */ 667 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 668 CA_EVENT_FRTO, /* fast recovery timeout */ 669 CA_EVENT_LOSS, /* loss timeout */ 670 CA_EVENT_FAST_ACK, /* in sequence ack */ 671 CA_EVENT_SLOW_ACK, /* other ack */ 672 }; 673 674 /* 675 * Interface for adding new TCP congestion control handlers 676 */ 677 #define TCP_CA_NAME_MAX 16 678 #define TCP_CA_MAX 128 679 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 680 681 #define TCP_CONG_NON_RESTRICTED 0x1 682 #define TCP_CONG_RTT_STAMP 0x2 683 684 struct tcp_congestion_ops { 685 struct list_head list; 686 unsigned long flags; 687 688 /* initialize private data (optional) */ 689 void (*init)(struct sock *sk); 690 /* cleanup private data (optional) */ 691 void (*release)(struct sock *sk); 692 693 /* return slow start threshold (required) */ 694 u32 (*ssthresh)(struct sock *sk); 695 /* lower bound for congestion window (optional) */ 696 u32 (*min_cwnd)(const struct sock *sk); 697 /* do new cwnd calculation (required) */ 698 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 699 /* call before changing ca_state (optional) */ 700 void (*set_state)(struct sock *sk, u8 new_state); 701 /* call when cwnd event occurs (optional) */ 702 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 703 /* new value of cwnd after loss (optional) */ 704 u32 (*undo_cwnd)(struct sock *sk); 705 /* hook for packet ack accounting (optional) */ 706 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 707 /* get info for inet_diag (optional) */ 708 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 709 710 char name[TCP_CA_NAME_MAX]; 711 struct module *owner; 712 }; 713 714 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 715 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 716 717 extern void tcp_init_congestion_control(struct sock *sk); 718 extern void tcp_cleanup_congestion_control(struct sock *sk); 719 extern int tcp_set_default_congestion_control(const char *name); 720 extern void tcp_get_default_congestion_control(char *name); 721 extern void tcp_get_available_congestion_control(char *buf, size_t len); 722 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 723 extern int tcp_set_allowed_congestion_control(char *allowed); 724 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 725 extern void tcp_slow_start(struct tcp_sock *tp); 726 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 727 728 extern struct tcp_congestion_ops tcp_init_congestion_ops; 729 extern u32 tcp_reno_ssthresh(struct sock *sk); 730 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 731 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 732 extern struct tcp_congestion_ops tcp_reno; 733 734 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 735 { 736 struct inet_connection_sock *icsk = inet_csk(sk); 737 738 if (icsk->icsk_ca_ops->set_state) 739 icsk->icsk_ca_ops->set_state(sk, ca_state); 740 icsk->icsk_ca_state = ca_state; 741 } 742 743 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 744 { 745 const struct inet_connection_sock *icsk = inet_csk(sk); 746 747 if (icsk->icsk_ca_ops->cwnd_event) 748 icsk->icsk_ca_ops->cwnd_event(sk, event); 749 } 750 751 /* These functions determine how the current flow behaves in respect of SACK 752 * handling. SACK is negotiated with the peer, and therefore it can vary 753 * between different flows. 754 * 755 * tcp_is_sack - SACK enabled 756 * tcp_is_reno - No SACK 757 * tcp_is_fack - FACK enabled, implies SACK enabled 758 */ 759 static inline int tcp_is_sack(const struct tcp_sock *tp) 760 { 761 return tp->rx_opt.sack_ok; 762 } 763 764 static inline int tcp_is_reno(const struct tcp_sock *tp) 765 { 766 return !tcp_is_sack(tp); 767 } 768 769 static inline int tcp_is_fack(const struct tcp_sock *tp) 770 { 771 return tp->rx_opt.sack_ok & 2; 772 } 773 774 static inline void tcp_enable_fack(struct tcp_sock *tp) 775 { 776 tp->rx_opt.sack_ok |= 2; 777 } 778 779 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 780 { 781 return tp->sacked_out + tp->lost_out; 782 } 783 784 /* This determines how many packets are "in the network" to the best 785 * of our knowledge. In many cases it is conservative, but where 786 * detailed information is available from the receiver (via SACK 787 * blocks etc.) we can make more aggressive calculations. 788 * 789 * Use this for decisions involving congestion control, use just 790 * tp->packets_out to determine if the send queue is empty or not. 791 * 792 * Read this equation as: 793 * 794 * "Packets sent once on transmission queue" MINUS 795 * "Packets left network, but not honestly ACKed yet" PLUS 796 * "Packets fast retransmitted" 797 */ 798 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 799 { 800 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 801 } 802 803 #define TCP_INFINITE_SSTHRESH 0x7fffffff 804 805 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 806 { 807 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 808 } 809 810 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 811 * The exception is rate halving phase, when cwnd is decreasing towards 812 * ssthresh. 813 */ 814 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 815 { 816 const struct tcp_sock *tp = tcp_sk(sk); 817 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 818 return tp->snd_ssthresh; 819 else 820 return max(tp->snd_ssthresh, 821 ((tp->snd_cwnd >> 1) + 822 (tp->snd_cwnd >> 2))); 823 } 824 825 /* Use define here intentionally to get WARN_ON location shown at the caller */ 826 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 827 828 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 829 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 830 831 /* Slow start with delack produces 3 packets of burst, so that 832 * it is safe "de facto". This will be the default - same as 833 * the default reordering threshold - but if reordering increases, 834 * we must be able to allow cwnd to burst at least this much in order 835 * to not pull it back when holes are filled. 836 */ 837 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 838 { 839 return tp->reordering; 840 } 841 842 /* Returns end sequence number of the receiver's advertised window */ 843 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 844 { 845 return tp->snd_una + tp->snd_wnd; 846 } 847 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 848 849 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 850 const struct sk_buff *skb) 851 { 852 if (skb->len < mss) 853 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 854 } 855 856 static inline void tcp_check_probe_timer(struct sock *sk) 857 { 858 struct tcp_sock *tp = tcp_sk(sk); 859 const struct inet_connection_sock *icsk = inet_csk(sk); 860 861 if (!tp->packets_out && !icsk->icsk_pending) 862 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 863 icsk->icsk_rto, TCP_RTO_MAX); 864 } 865 866 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 867 { 868 tp->snd_wl1 = seq; 869 } 870 871 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 872 { 873 tp->snd_wl1 = seq; 874 } 875 876 /* 877 * Calculate(/check) TCP checksum 878 */ 879 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 880 __be32 daddr, __wsum base) 881 { 882 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 883 } 884 885 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 886 { 887 return __skb_checksum_complete(skb); 888 } 889 890 static inline int tcp_checksum_complete(struct sk_buff *skb) 891 { 892 return !skb_csum_unnecessary(skb) && 893 __tcp_checksum_complete(skb); 894 } 895 896 /* Prequeue for VJ style copy to user, combined with checksumming. */ 897 898 static inline void tcp_prequeue_init(struct tcp_sock *tp) 899 { 900 tp->ucopy.task = NULL; 901 tp->ucopy.len = 0; 902 tp->ucopy.memory = 0; 903 skb_queue_head_init(&tp->ucopy.prequeue); 904 #ifdef CONFIG_NET_DMA 905 tp->ucopy.dma_chan = NULL; 906 tp->ucopy.wakeup = 0; 907 tp->ucopy.pinned_list = NULL; 908 tp->ucopy.dma_cookie = 0; 909 #endif 910 } 911 912 /* Packet is added to VJ-style prequeue for processing in process 913 * context, if a reader task is waiting. Apparently, this exciting 914 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 915 * failed somewhere. Latency? Burstiness? Well, at least now we will 916 * see, why it failed. 8)8) --ANK 917 * 918 * NOTE: is this not too big to inline? 919 */ 920 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 921 { 922 struct tcp_sock *tp = tcp_sk(sk); 923 924 if (sysctl_tcp_low_latency || !tp->ucopy.task) 925 return 0; 926 927 __skb_queue_tail(&tp->ucopy.prequeue, skb); 928 tp->ucopy.memory += skb->truesize; 929 if (tp->ucopy.memory > sk->sk_rcvbuf) { 930 struct sk_buff *skb1; 931 932 BUG_ON(sock_owned_by_user(sk)); 933 934 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 935 sk_backlog_rcv(sk, skb1); 936 NET_INC_STATS_BH(sock_net(sk), 937 LINUX_MIB_TCPPREQUEUEDROPPED); 938 } 939 940 tp->ucopy.memory = 0; 941 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 942 wake_up_interruptible_sync_poll(sk->sk_sleep, 943 POLLIN | POLLRDNORM | POLLRDBAND); 944 if (!inet_csk_ack_scheduled(sk)) 945 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 946 (3 * tcp_rto_min(sk)) / 4, 947 TCP_RTO_MAX); 948 } 949 return 1; 950 } 951 952 953 #undef STATE_TRACE 954 955 #ifdef STATE_TRACE 956 static const char *statename[]={ 957 "Unused","Established","Syn Sent","Syn Recv", 958 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 959 "Close Wait","Last ACK","Listen","Closing" 960 }; 961 #endif 962 extern void tcp_set_state(struct sock *sk, int state); 963 964 extern void tcp_done(struct sock *sk); 965 966 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 967 { 968 rx_opt->dsack = 0; 969 rx_opt->num_sacks = 0; 970 } 971 972 /* Determine a window scaling and initial window to offer. */ 973 extern void tcp_select_initial_window(int __space, __u32 mss, 974 __u32 *rcv_wnd, __u32 *window_clamp, 975 int wscale_ok, __u8 *rcv_wscale, 976 __u32 init_rcv_wnd); 977 978 static inline int tcp_win_from_space(int space) 979 { 980 return sysctl_tcp_adv_win_scale<=0 ? 981 (space>>(-sysctl_tcp_adv_win_scale)) : 982 space - (space>>sysctl_tcp_adv_win_scale); 983 } 984 985 /* Note: caller must be prepared to deal with negative returns */ 986 static inline int tcp_space(const struct sock *sk) 987 { 988 return tcp_win_from_space(sk->sk_rcvbuf - 989 atomic_read(&sk->sk_rmem_alloc)); 990 } 991 992 static inline int tcp_full_space(const struct sock *sk) 993 { 994 return tcp_win_from_space(sk->sk_rcvbuf); 995 } 996 997 static inline void tcp_openreq_init(struct request_sock *req, 998 struct tcp_options_received *rx_opt, 999 struct sk_buff *skb) 1000 { 1001 struct inet_request_sock *ireq = inet_rsk(req); 1002 1003 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1004 req->cookie_ts = 0; 1005 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1006 req->mss = rx_opt->mss_clamp; 1007 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1008 ireq->tstamp_ok = rx_opt->tstamp_ok; 1009 ireq->sack_ok = rx_opt->sack_ok; 1010 ireq->snd_wscale = rx_opt->snd_wscale; 1011 ireq->wscale_ok = rx_opt->wscale_ok; 1012 ireq->acked = 0; 1013 ireq->ecn_ok = 0; 1014 ireq->rmt_port = tcp_hdr(skb)->source; 1015 ireq->loc_port = tcp_hdr(skb)->dest; 1016 } 1017 1018 extern void tcp_enter_memory_pressure(struct sock *sk); 1019 1020 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1021 { 1022 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1023 } 1024 1025 static inline int keepalive_time_when(const struct tcp_sock *tp) 1026 { 1027 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1028 } 1029 1030 static inline int keepalive_probes(const struct tcp_sock *tp) 1031 { 1032 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1033 } 1034 1035 static inline int tcp_fin_time(const struct sock *sk) 1036 { 1037 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1038 const int rto = inet_csk(sk)->icsk_rto; 1039 1040 if (fin_timeout < (rto << 2) - (rto >> 1)) 1041 fin_timeout = (rto << 2) - (rto >> 1); 1042 1043 return fin_timeout; 1044 } 1045 1046 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1047 int paws_win) 1048 { 1049 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1050 return 1; 1051 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1052 return 1; 1053 1054 return 0; 1055 } 1056 1057 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1058 int rst) 1059 { 1060 if (tcp_paws_check(rx_opt, 0)) 1061 return 0; 1062 1063 /* RST segments are not recommended to carry timestamp, 1064 and, if they do, it is recommended to ignore PAWS because 1065 "their cleanup function should take precedence over timestamps." 1066 Certainly, it is mistake. It is necessary to understand the reasons 1067 of this constraint to relax it: if peer reboots, clock may go 1068 out-of-sync and half-open connections will not be reset. 1069 Actually, the problem would be not existing if all 1070 the implementations followed draft about maintaining clock 1071 via reboots. Linux-2.2 DOES NOT! 1072 1073 However, we can relax time bounds for RST segments to MSL. 1074 */ 1075 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1076 return 0; 1077 return 1; 1078 } 1079 1080 #define TCP_CHECK_TIMER(sk) do { } while (0) 1081 1082 static inline void tcp_mib_init(struct net *net) 1083 { 1084 /* See RFC 2012 */ 1085 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1086 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1087 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1088 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1089 } 1090 1091 /* from STCP */ 1092 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1093 { 1094 tp->lost_skb_hint = NULL; 1095 tp->scoreboard_skb_hint = NULL; 1096 } 1097 1098 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1099 { 1100 tcp_clear_retrans_hints_partial(tp); 1101 tp->retransmit_skb_hint = NULL; 1102 } 1103 1104 /* MD5 Signature */ 1105 struct crypto_hash; 1106 1107 /* - key database */ 1108 struct tcp_md5sig_key { 1109 u8 *key; 1110 u8 keylen; 1111 }; 1112 1113 struct tcp4_md5sig_key { 1114 struct tcp_md5sig_key base; 1115 __be32 addr; 1116 }; 1117 1118 struct tcp6_md5sig_key { 1119 struct tcp_md5sig_key base; 1120 #if 0 1121 u32 scope_id; /* XXX */ 1122 #endif 1123 struct in6_addr addr; 1124 }; 1125 1126 /* - sock block */ 1127 struct tcp_md5sig_info { 1128 struct tcp4_md5sig_key *keys4; 1129 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1130 struct tcp6_md5sig_key *keys6; 1131 u32 entries6; 1132 u32 alloced6; 1133 #endif 1134 u32 entries4; 1135 u32 alloced4; 1136 }; 1137 1138 /* - pseudo header */ 1139 struct tcp4_pseudohdr { 1140 __be32 saddr; 1141 __be32 daddr; 1142 __u8 pad; 1143 __u8 protocol; 1144 __be16 len; 1145 }; 1146 1147 struct tcp6_pseudohdr { 1148 struct in6_addr saddr; 1149 struct in6_addr daddr; 1150 __be32 len; 1151 __be32 protocol; /* including padding */ 1152 }; 1153 1154 union tcp_md5sum_block { 1155 struct tcp4_pseudohdr ip4; 1156 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1157 struct tcp6_pseudohdr ip6; 1158 #endif 1159 }; 1160 1161 /* - pool: digest algorithm, hash description and scratch buffer */ 1162 struct tcp_md5sig_pool { 1163 struct hash_desc md5_desc; 1164 union tcp_md5sum_block md5_blk; 1165 }; 1166 1167 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1168 1169 /* - functions */ 1170 extern int tcp_v4_md5_hash_skb(char *md5_hash, 1171 struct tcp_md5sig_key *key, 1172 struct sock *sk, 1173 struct request_sock *req, 1174 struct sk_buff *skb); 1175 1176 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1177 struct sock *addr_sk); 1178 1179 extern int tcp_v4_md5_do_add(struct sock *sk, 1180 __be32 addr, 1181 u8 *newkey, 1182 u8 newkeylen); 1183 1184 extern int tcp_v4_md5_do_del(struct sock *sk, 1185 __be32 addr); 1186 1187 #ifdef CONFIG_TCP_MD5SIG 1188 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1189 &(struct tcp_md5sig_key) { \ 1190 .key = (twsk)->tw_md5_key, \ 1191 .keylen = (twsk)->tw_md5_keylen, \ 1192 } : NULL) 1193 #else 1194 #define tcp_twsk_md5_key(twsk) NULL 1195 #endif 1196 1197 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1198 extern void tcp_free_md5sig_pool(void); 1199 1200 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu); 1201 extern void __tcp_put_md5sig_pool(void); 1202 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1203 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1204 unsigned header_len); 1205 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1206 struct tcp_md5sig_key *key); 1207 1208 static inline 1209 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) 1210 { 1211 int cpu = get_cpu(); 1212 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu); 1213 if (!ret) 1214 put_cpu(); 1215 return ret; 1216 } 1217 1218 static inline void tcp_put_md5sig_pool(void) 1219 { 1220 __tcp_put_md5sig_pool(); 1221 put_cpu(); 1222 } 1223 1224 /* write queue abstraction */ 1225 static inline void tcp_write_queue_purge(struct sock *sk) 1226 { 1227 struct sk_buff *skb; 1228 1229 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1230 sk_wmem_free_skb(sk, skb); 1231 sk_mem_reclaim(sk); 1232 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1233 } 1234 1235 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1236 { 1237 return skb_peek(&sk->sk_write_queue); 1238 } 1239 1240 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1241 { 1242 return skb_peek_tail(&sk->sk_write_queue); 1243 } 1244 1245 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1246 { 1247 return skb_queue_next(&sk->sk_write_queue, skb); 1248 } 1249 1250 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1251 { 1252 return skb_queue_prev(&sk->sk_write_queue, skb); 1253 } 1254 1255 #define tcp_for_write_queue(skb, sk) \ 1256 skb_queue_walk(&(sk)->sk_write_queue, skb) 1257 1258 #define tcp_for_write_queue_from(skb, sk) \ 1259 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1260 1261 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1262 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1263 1264 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1265 { 1266 return sk->sk_send_head; 1267 } 1268 1269 static inline bool tcp_skb_is_last(const struct sock *sk, 1270 const struct sk_buff *skb) 1271 { 1272 return skb_queue_is_last(&sk->sk_write_queue, skb); 1273 } 1274 1275 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1276 { 1277 if (tcp_skb_is_last(sk, skb)) 1278 sk->sk_send_head = NULL; 1279 else 1280 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1281 } 1282 1283 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1284 { 1285 if (sk->sk_send_head == skb_unlinked) 1286 sk->sk_send_head = NULL; 1287 } 1288 1289 static inline void tcp_init_send_head(struct sock *sk) 1290 { 1291 sk->sk_send_head = NULL; 1292 } 1293 1294 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1295 { 1296 __skb_queue_tail(&sk->sk_write_queue, skb); 1297 } 1298 1299 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1300 { 1301 __tcp_add_write_queue_tail(sk, skb); 1302 1303 /* Queue it, remembering where we must start sending. */ 1304 if (sk->sk_send_head == NULL) { 1305 sk->sk_send_head = skb; 1306 1307 if (tcp_sk(sk)->highest_sack == NULL) 1308 tcp_sk(sk)->highest_sack = skb; 1309 } 1310 } 1311 1312 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1313 { 1314 __skb_queue_head(&sk->sk_write_queue, skb); 1315 } 1316 1317 /* Insert buff after skb on the write queue of sk. */ 1318 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1319 struct sk_buff *buff, 1320 struct sock *sk) 1321 { 1322 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1323 } 1324 1325 /* Insert new before skb on the write queue of sk. */ 1326 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1327 struct sk_buff *skb, 1328 struct sock *sk) 1329 { 1330 __skb_queue_before(&sk->sk_write_queue, skb, new); 1331 1332 if (sk->sk_send_head == skb) 1333 sk->sk_send_head = new; 1334 } 1335 1336 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1337 { 1338 __skb_unlink(skb, &sk->sk_write_queue); 1339 } 1340 1341 static inline int tcp_write_queue_empty(struct sock *sk) 1342 { 1343 return skb_queue_empty(&sk->sk_write_queue); 1344 } 1345 1346 static inline void tcp_push_pending_frames(struct sock *sk) 1347 { 1348 if (tcp_send_head(sk)) { 1349 struct tcp_sock *tp = tcp_sk(sk); 1350 1351 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1352 } 1353 } 1354 1355 /* Start sequence of the highest skb with SACKed bit, valid only if 1356 * sacked > 0 or when the caller has ensured validity by itself. 1357 */ 1358 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1359 { 1360 if (!tp->sacked_out) 1361 return tp->snd_una; 1362 1363 if (tp->highest_sack == NULL) 1364 return tp->snd_nxt; 1365 1366 return TCP_SKB_CB(tp->highest_sack)->seq; 1367 } 1368 1369 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1370 { 1371 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1372 tcp_write_queue_next(sk, skb); 1373 } 1374 1375 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1376 { 1377 return tcp_sk(sk)->highest_sack; 1378 } 1379 1380 static inline void tcp_highest_sack_reset(struct sock *sk) 1381 { 1382 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1383 } 1384 1385 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1386 static inline void tcp_highest_sack_combine(struct sock *sk, 1387 struct sk_buff *old, 1388 struct sk_buff *new) 1389 { 1390 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1391 tcp_sk(sk)->highest_sack = new; 1392 } 1393 1394 /* Determines whether this is a thin stream (which may suffer from 1395 * increased latency). Used to trigger latency-reducing mechanisms. 1396 */ 1397 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1398 { 1399 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1400 } 1401 1402 /* /proc */ 1403 enum tcp_seq_states { 1404 TCP_SEQ_STATE_LISTENING, 1405 TCP_SEQ_STATE_OPENREQ, 1406 TCP_SEQ_STATE_ESTABLISHED, 1407 TCP_SEQ_STATE_TIME_WAIT, 1408 }; 1409 1410 struct tcp_seq_afinfo { 1411 char *name; 1412 sa_family_t family; 1413 struct file_operations seq_fops; 1414 struct seq_operations seq_ops; 1415 }; 1416 1417 struct tcp_iter_state { 1418 struct seq_net_private p; 1419 sa_family_t family; 1420 enum tcp_seq_states state; 1421 struct sock *syn_wait_sk; 1422 int bucket, sbucket, num, uid; 1423 }; 1424 1425 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1426 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1427 1428 extern struct request_sock_ops tcp_request_sock_ops; 1429 extern struct request_sock_ops tcp6_request_sock_ops; 1430 1431 extern void tcp_v4_destroy_sock(struct sock *sk); 1432 1433 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1434 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1435 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1436 struct sk_buff *skb); 1437 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1438 struct sk_buff *skb); 1439 extern int tcp_gro_complete(struct sk_buff *skb); 1440 extern int tcp4_gro_complete(struct sk_buff *skb); 1441 1442 #ifdef CONFIG_PROC_FS 1443 extern int tcp4_proc_init(void); 1444 extern void tcp4_proc_exit(void); 1445 #endif 1446 1447 /* TCP af-specific functions */ 1448 struct tcp_sock_af_ops { 1449 #ifdef CONFIG_TCP_MD5SIG 1450 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1451 struct sock *addr_sk); 1452 int (*calc_md5_hash) (char *location, 1453 struct tcp_md5sig_key *md5, 1454 struct sock *sk, 1455 struct request_sock *req, 1456 struct sk_buff *skb); 1457 int (*md5_add) (struct sock *sk, 1458 struct sock *addr_sk, 1459 u8 *newkey, 1460 u8 len); 1461 int (*md5_parse) (struct sock *sk, 1462 char __user *optval, 1463 int optlen); 1464 #endif 1465 }; 1466 1467 struct tcp_request_sock_ops { 1468 #ifdef CONFIG_TCP_MD5SIG 1469 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1470 struct request_sock *req); 1471 int (*calc_md5_hash) (char *location, 1472 struct tcp_md5sig_key *md5, 1473 struct sock *sk, 1474 struct request_sock *req, 1475 struct sk_buff *skb); 1476 #endif 1477 }; 1478 1479 /* Using SHA1 for now, define some constants. 1480 */ 1481 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1482 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1483 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1484 1485 extern int tcp_cookie_generator(u32 *bakery); 1486 1487 /** 1488 * struct tcp_cookie_values - each socket needs extra space for the 1489 * cookies, together with (optional) space for any SYN data. 1490 * 1491 * A tcp_sock contains a pointer to the current value, and this is 1492 * cloned to the tcp_timewait_sock. 1493 * 1494 * @cookie_pair: variable data from the option exchange. 1495 * 1496 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1497 * indicates default (sysctl_tcp_cookie_size). 1498 * After cookie sent, remembers size of cookie. 1499 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1500 * 1501 * @s_data_desired: user specified tcpct_s_data_desired. When the 1502 * constant payload is specified (@s_data_constant), 1503 * holds its length instead. 1504 * Range 0 to TCP_MSS_DESIRED. 1505 * 1506 * @s_data_payload: constant data that is to be included in the 1507 * payload of SYN or SYNACK segments when the 1508 * cookie option is present. 1509 */ 1510 struct tcp_cookie_values { 1511 struct kref kref; 1512 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1513 u8 cookie_pair_size; 1514 u8 cookie_desired; 1515 u16 s_data_desired:11, 1516 s_data_constant:1, 1517 s_data_in:1, 1518 s_data_out:1, 1519 s_data_unused:2; 1520 u8 s_data_payload[0]; 1521 }; 1522 1523 static inline void tcp_cookie_values_release(struct kref *kref) 1524 { 1525 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1526 } 1527 1528 /* The length of constant payload data. Note that s_data_desired is 1529 * overloaded, depending on s_data_constant: either the length of constant 1530 * data (returned here) or the limit on variable data. 1531 */ 1532 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1533 { 1534 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1535 ? tp->cookie_values->s_data_desired 1536 : 0; 1537 } 1538 1539 /** 1540 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1541 * 1542 * As tcp_request_sock has already been extended in other places, the 1543 * only remaining method is to pass stack values along as function 1544 * parameters. These parameters are not needed after sending SYNACK. 1545 * 1546 * @cookie_bakery: cryptographic secret and message workspace. 1547 * 1548 * @cookie_plus: bytes in authenticator/cookie option, copied from 1549 * struct tcp_options_received (above). 1550 */ 1551 struct tcp_extend_values { 1552 struct request_values rv; 1553 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1554 u8 cookie_plus:6, 1555 cookie_out_never:1, 1556 cookie_in_always:1; 1557 }; 1558 1559 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1560 { 1561 return (struct tcp_extend_values *)rvp; 1562 } 1563 1564 extern void tcp_v4_init(void); 1565 extern void tcp_init(void); 1566 1567 #endif /* _TCP_H */ 1568