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 bool tcp_too_many_orphans(struct sock *sk, int shift) 272 { 273 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 274 int orphans = percpu_counter_read_positive(ocp); 275 276 if (orphans << shift > sysctl_tcp_max_orphans) { 277 orphans = percpu_counter_sum_positive(ocp); 278 if (orphans << shift > sysctl_tcp_max_orphans) 279 return true; 280 } 281 282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 283 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]) 284 return true; 285 return false; 286 } 287 288 /* syncookies: remember time of last synqueue overflow */ 289 static inline void tcp_synq_overflow(struct sock *sk) 290 { 291 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 292 } 293 294 /* syncookies: no recent synqueue overflow on this listening socket? */ 295 static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 296 { 297 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 298 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT); 299 } 300 301 extern struct proto tcp_prot; 302 303 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 304 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 305 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 306 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 307 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 308 309 extern void tcp_v4_err(struct sk_buff *skb, u32); 310 311 extern void tcp_shutdown (struct sock *sk, int how); 312 313 extern int tcp_v4_rcv(struct sk_buff *skb); 314 315 extern int tcp_v4_remember_stamp(struct sock *sk); 316 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 317 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 318 size_t size); 319 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 320 size_t size, int flags); 321 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 322 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 323 struct tcphdr *th, unsigned len); 324 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 325 struct tcphdr *th, unsigned len); 326 extern void tcp_rcv_space_adjust(struct sock *sk); 327 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 328 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 329 extern void tcp_twsk_destructor(struct sock *sk); 330 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 331 struct pipe_inode_info *pipe, size_t len, 332 unsigned int flags); 333 334 static inline void tcp_dec_quickack_mode(struct sock *sk, 335 const unsigned int pkts) 336 { 337 struct inet_connection_sock *icsk = inet_csk(sk); 338 339 if (icsk->icsk_ack.quick) { 340 if (pkts >= icsk->icsk_ack.quick) { 341 icsk->icsk_ack.quick = 0; 342 /* Leaving quickack mode we deflate ATO. */ 343 icsk->icsk_ack.ato = TCP_ATO_MIN; 344 } else 345 icsk->icsk_ack.quick -= pkts; 346 } 347 } 348 349 extern void tcp_enter_quickack_mode(struct sock *sk); 350 351 #define TCP_ECN_OK 1 352 #define TCP_ECN_QUEUE_CWR 2 353 #define TCP_ECN_DEMAND_CWR 4 354 355 static __inline__ void 356 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 357 { 358 if (sysctl_tcp_ecn && th->ece && th->cwr) 359 inet_rsk(req)->ecn_ok = 1; 360 } 361 362 enum tcp_tw_status { 363 TCP_TW_SUCCESS = 0, 364 TCP_TW_RST = 1, 365 TCP_TW_ACK = 2, 366 TCP_TW_SYN = 3 367 }; 368 369 370 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 371 struct sk_buff *skb, 372 const struct tcphdr *th); 373 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 374 struct request_sock *req, 375 struct request_sock **prev); 376 extern int tcp_child_process(struct sock *parent, struct sock *child, 377 struct sk_buff *skb); 378 extern int tcp_use_frto(struct sock *sk); 379 extern void tcp_enter_frto(struct sock *sk); 380 extern void tcp_enter_loss(struct sock *sk, int how); 381 extern void tcp_clear_retrans(struct tcp_sock *tp); 382 extern void tcp_update_metrics(struct sock *sk); 383 extern void tcp_close(struct sock *sk, long timeout); 384 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 385 struct poll_table_struct *wait); 386 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 387 char __user *optval, int __user *optlen); 388 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 389 char __user *optval, unsigned int optlen); 390 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 391 char __user *optval, int __user *optlen); 392 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 393 char __user *optval, unsigned int optlen); 394 extern void tcp_set_keepalive(struct sock *sk, int val); 395 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 396 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 397 size_t len, int nonblock, int flags, int *addr_len); 398 extern void tcp_parse_options(struct sk_buff *skb, 399 struct tcp_options_received *opt_rx, u8 **hvpp, 400 int estab); 401 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 402 403 /* 404 * TCP v4 functions exported for the inet6 API 405 */ 406 407 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 408 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 409 extern struct sock * tcp_create_openreq_child(struct sock *sk, 410 struct request_sock *req, 411 struct sk_buff *skb); 412 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 413 struct request_sock *req, 414 struct dst_entry *dst); 415 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 416 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 417 int addr_len); 418 extern int tcp_connect(struct sock *sk); 419 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 420 struct request_sock *req, 421 struct request_values *rvp); 422 extern int tcp_disconnect(struct sock *sk, int flags); 423 424 425 /* From syncookies.c */ 426 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 427 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 428 struct ip_options *opt); 429 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 430 __u16 *mss); 431 432 extern __u32 cookie_init_timestamp(struct request_sock *req); 433 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 434 435 /* From net/ipv6/syncookies.c */ 436 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 437 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 438 __u16 *mss); 439 440 /* tcp_output.c */ 441 442 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 443 int nonagle); 444 extern int tcp_may_send_now(struct sock *sk); 445 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 446 extern void tcp_retransmit_timer(struct sock *sk); 447 extern void tcp_xmit_retransmit_queue(struct sock *); 448 extern void tcp_simple_retransmit(struct sock *); 449 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 450 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 451 452 extern void tcp_send_probe0(struct sock *); 453 extern void tcp_send_partial(struct sock *); 454 extern int tcp_write_wakeup(struct sock *); 455 extern void tcp_send_fin(struct sock *sk); 456 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 457 extern int tcp_send_synack(struct sock *); 458 extern void tcp_push_one(struct sock *, unsigned int mss_now); 459 extern void tcp_send_ack(struct sock *sk); 460 extern void tcp_send_delayed_ack(struct sock *sk); 461 462 /* tcp_input.c */ 463 extern void tcp_cwnd_application_limited(struct sock *sk); 464 465 /* tcp_timer.c */ 466 extern void tcp_init_xmit_timers(struct sock *); 467 static inline void tcp_clear_xmit_timers(struct sock *sk) 468 { 469 inet_csk_clear_xmit_timers(sk); 470 } 471 472 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 473 extern unsigned int tcp_current_mss(struct sock *sk); 474 475 /* Bound MSS / TSO packet size with the half of the window */ 476 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 477 { 478 if (tp->max_window && pktsize > (tp->max_window >> 1)) 479 return max(tp->max_window >> 1, 68U - tp->tcp_header_len); 480 else 481 return pktsize; 482 } 483 484 /* tcp.c */ 485 extern void tcp_get_info(struct sock *, struct tcp_info *); 486 487 /* Read 'sendfile()'-style from a TCP socket */ 488 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 489 unsigned int, size_t); 490 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 491 sk_read_actor_t recv_actor); 492 493 extern void tcp_initialize_rcv_mss(struct sock *sk); 494 495 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 496 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 497 extern void tcp_mtup_init(struct sock *sk); 498 499 static inline void tcp_bound_rto(const struct sock *sk) 500 { 501 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 502 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 503 } 504 505 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 506 { 507 return (tp->srtt >> 3) + tp->rttvar; 508 } 509 510 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 511 { 512 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 513 ntohl(TCP_FLAG_ACK) | 514 snd_wnd); 515 } 516 517 static inline void tcp_fast_path_on(struct tcp_sock *tp) 518 { 519 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 520 } 521 522 static inline void tcp_fast_path_check(struct sock *sk) 523 { 524 struct tcp_sock *tp = tcp_sk(sk); 525 526 if (skb_queue_empty(&tp->out_of_order_queue) && 527 tp->rcv_wnd && 528 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 529 !tp->urg_data) 530 tcp_fast_path_on(tp); 531 } 532 533 /* Compute the actual rto_min value */ 534 static inline u32 tcp_rto_min(struct sock *sk) 535 { 536 struct dst_entry *dst = __sk_dst_get(sk); 537 u32 rto_min = TCP_RTO_MIN; 538 539 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 540 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 541 return rto_min; 542 } 543 544 /* Compute the actual receive window we are currently advertising. 545 * Rcv_nxt can be after the window if our peer push more data 546 * than the offered window. 547 */ 548 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 549 { 550 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 551 552 if (win < 0) 553 win = 0; 554 return (u32) win; 555 } 556 557 /* Choose a new window, without checks for shrinking, and without 558 * scaling applied to the result. The caller does these things 559 * if necessary. This is a "raw" window selection. 560 */ 561 extern u32 __tcp_select_window(struct sock *sk); 562 563 /* TCP timestamps are only 32-bits, this causes a slight 564 * complication on 64-bit systems since we store a snapshot 565 * of jiffies in the buffer control blocks below. We decided 566 * to use only the low 32-bits of jiffies and hide the ugly 567 * casts with the following macro. 568 */ 569 #define tcp_time_stamp ((__u32)(jiffies)) 570 571 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 572 573 #define TCPHDR_FIN 0x01 574 #define TCPHDR_SYN 0x02 575 #define TCPHDR_RST 0x04 576 #define TCPHDR_PSH 0x08 577 #define TCPHDR_ACK 0x10 578 #define TCPHDR_URG 0x20 579 #define TCPHDR_ECE 0x40 580 #define TCPHDR_CWR 0x80 581 582 /* This is what the send packet queuing engine uses to pass 583 * TCP per-packet control information to the transmission code. 584 * We also store the host-order sequence numbers in here too. 585 * This is 44 bytes if IPV6 is enabled. 586 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 587 */ 588 struct tcp_skb_cb { 589 union { 590 struct inet_skb_parm h4; 591 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 592 struct inet6_skb_parm h6; 593 #endif 594 } header; /* For incoming frames */ 595 __u32 seq; /* Starting sequence number */ 596 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 597 __u32 when; /* used to compute rtt's */ 598 __u8 flags; /* TCP header flags. */ 599 __u8 sacked; /* State flags for SACK/FACK. */ 600 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 601 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 602 #define TCPCB_LOST 0x04 /* SKB is lost */ 603 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 604 605 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 606 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 607 608 __u32 ack_seq; /* Sequence number ACK'd */ 609 }; 610 611 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 612 613 /* Due to TSO, an SKB can be composed of multiple actual 614 * packets. To keep these tracked properly, we use this. 615 */ 616 static inline int tcp_skb_pcount(const struct sk_buff *skb) 617 { 618 return skb_shinfo(skb)->gso_segs; 619 } 620 621 /* This is valid iff tcp_skb_pcount() > 1. */ 622 static inline int tcp_skb_mss(const struct sk_buff *skb) 623 { 624 return skb_shinfo(skb)->gso_size; 625 } 626 627 /* Events passed to congestion control interface */ 628 enum tcp_ca_event { 629 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 630 CA_EVENT_CWND_RESTART, /* congestion window restart */ 631 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 632 CA_EVENT_FRTO, /* fast recovery timeout */ 633 CA_EVENT_LOSS, /* loss timeout */ 634 CA_EVENT_FAST_ACK, /* in sequence ack */ 635 CA_EVENT_SLOW_ACK, /* other ack */ 636 }; 637 638 /* 639 * Interface for adding new TCP congestion control handlers 640 */ 641 #define TCP_CA_NAME_MAX 16 642 #define TCP_CA_MAX 128 643 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 644 645 #define TCP_CONG_NON_RESTRICTED 0x1 646 #define TCP_CONG_RTT_STAMP 0x2 647 648 struct tcp_congestion_ops { 649 struct list_head list; 650 unsigned long flags; 651 652 /* initialize private data (optional) */ 653 void (*init)(struct sock *sk); 654 /* cleanup private data (optional) */ 655 void (*release)(struct sock *sk); 656 657 /* return slow start threshold (required) */ 658 u32 (*ssthresh)(struct sock *sk); 659 /* lower bound for congestion window (optional) */ 660 u32 (*min_cwnd)(const struct sock *sk); 661 /* do new cwnd calculation (required) */ 662 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 663 /* call before changing ca_state (optional) */ 664 void (*set_state)(struct sock *sk, u8 new_state); 665 /* call when cwnd event occurs (optional) */ 666 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 667 /* new value of cwnd after loss (optional) */ 668 u32 (*undo_cwnd)(struct sock *sk); 669 /* hook for packet ack accounting (optional) */ 670 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 671 /* get info for inet_diag (optional) */ 672 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 673 674 char name[TCP_CA_NAME_MAX]; 675 struct module *owner; 676 }; 677 678 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 679 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 680 681 extern void tcp_init_congestion_control(struct sock *sk); 682 extern void tcp_cleanup_congestion_control(struct sock *sk); 683 extern int tcp_set_default_congestion_control(const char *name); 684 extern void tcp_get_default_congestion_control(char *name); 685 extern void tcp_get_available_congestion_control(char *buf, size_t len); 686 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 687 extern int tcp_set_allowed_congestion_control(char *allowed); 688 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 689 extern void tcp_slow_start(struct tcp_sock *tp); 690 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 691 692 extern struct tcp_congestion_ops tcp_init_congestion_ops; 693 extern u32 tcp_reno_ssthresh(struct sock *sk); 694 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 695 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 696 extern struct tcp_congestion_ops tcp_reno; 697 698 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 699 { 700 struct inet_connection_sock *icsk = inet_csk(sk); 701 702 if (icsk->icsk_ca_ops->set_state) 703 icsk->icsk_ca_ops->set_state(sk, ca_state); 704 icsk->icsk_ca_state = ca_state; 705 } 706 707 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 708 { 709 const struct inet_connection_sock *icsk = inet_csk(sk); 710 711 if (icsk->icsk_ca_ops->cwnd_event) 712 icsk->icsk_ca_ops->cwnd_event(sk, event); 713 } 714 715 /* These functions determine how the current flow behaves in respect of SACK 716 * handling. SACK is negotiated with the peer, and therefore it can vary 717 * between different flows. 718 * 719 * tcp_is_sack - SACK enabled 720 * tcp_is_reno - No SACK 721 * tcp_is_fack - FACK enabled, implies SACK enabled 722 */ 723 static inline int tcp_is_sack(const struct tcp_sock *tp) 724 { 725 return tp->rx_opt.sack_ok; 726 } 727 728 static inline int tcp_is_reno(const struct tcp_sock *tp) 729 { 730 return !tcp_is_sack(tp); 731 } 732 733 static inline int tcp_is_fack(const struct tcp_sock *tp) 734 { 735 return tp->rx_opt.sack_ok & 2; 736 } 737 738 static inline void tcp_enable_fack(struct tcp_sock *tp) 739 { 740 tp->rx_opt.sack_ok |= 2; 741 } 742 743 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 744 { 745 return tp->sacked_out + tp->lost_out; 746 } 747 748 /* This determines how many packets are "in the network" to the best 749 * of our knowledge. In many cases it is conservative, but where 750 * detailed information is available from the receiver (via SACK 751 * blocks etc.) we can make more aggressive calculations. 752 * 753 * Use this for decisions involving congestion control, use just 754 * tp->packets_out to determine if the send queue is empty or not. 755 * 756 * Read this equation as: 757 * 758 * "Packets sent once on transmission queue" MINUS 759 * "Packets left network, but not honestly ACKed yet" PLUS 760 * "Packets fast retransmitted" 761 */ 762 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 763 { 764 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 765 } 766 767 #define TCP_INFINITE_SSTHRESH 0x7fffffff 768 769 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 770 { 771 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 772 } 773 774 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 775 * The exception is rate halving phase, when cwnd is decreasing towards 776 * ssthresh. 777 */ 778 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 779 { 780 const struct tcp_sock *tp = tcp_sk(sk); 781 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 782 return tp->snd_ssthresh; 783 else 784 return max(tp->snd_ssthresh, 785 ((tp->snd_cwnd >> 1) + 786 (tp->snd_cwnd >> 2))); 787 } 788 789 /* Use define here intentionally to get WARN_ON location shown at the caller */ 790 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 791 792 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 793 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 794 795 /* Slow start with delack produces 3 packets of burst, so that 796 * it is safe "de facto". This will be the default - same as 797 * the default reordering threshold - but if reordering increases, 798 * we must be able to allow cwnd to burst at least this much in order 799 * to not pull it back when holes are filled. 800 */ 801 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 802 { 803 return tp->reordering; 804 } 805 806 /* Returns end sequence number of the receiver's advertised window */ 807 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 808 { 809 return tp->snd_una + tp->snd_wnd; 810 } 811 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 812 813 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 814 const struct sk_buff *skb) 815 { 816 if (skb->len < mss) 817 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 818 } 819 820 static inline void tcp_check_probe_timer(struct sock *sk) 821 { 822 struct tcp_sock *tp = tcp_sk(sk); 823 const struct inet_connection_sock *icsk = inet_csk(sk); 824 825 if (!tp->packets_out && !icsk->icsk_pending) 826 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 827 icsk->icsk_rto, TCP_RTO_MAX); 828 } 829 830 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 831 { 832 tp->snd_wl1 = seq; 833 } 834 835 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 836 { 837 tp->snd_wl1 = seq; 838 } 839 840 /* 841 * Calculate(/check) TCP checksum 842 */ 843 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 844 __be32 daddr, __wsum base) 845 { 846 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 847 } 848 849 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 850 { 851 return __skb_checksum_complete(skb); 852 } 853 854 static inline int tcp_checksum_complete(struct sk_buff *skb) 855 { 856 return !skb_csum_unnecessary(skb) && 857 __tcp_checksum_complete(skb); 858 } 859 860 /* Prequeue for VJ style copy to user, combined with checksumming. */ 861 862 static inline void tcp_prequeue_init(struct tcp_sock *tp) 863 { 864 tp->ucopy.task = NULL; 865 tp->ucopy.len = 0; 866 tp->ucopy.memory = 0; 867 skb_queue_head_init(&tp->ucopy.prequeue); 868 #ifdef CONFIG_NET_DMA 869 tp->ucopy.dma_chan = NULL; 870 tp->ucopy.wakeup = 0; 871 tp->ucopy.pinned_list = NULL; 872 tp->ucopy.dma_cookie = 0; 873 #endif 874 } 875 876 /* Packet is added to VJ-style prequeue for processing in process 877 * context, if a reader task is waiting. Apparently, this exciting 878 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 879 * failed somewhere. Latency? Burstiness? Well, at least now we will 880 * see, why it failed. 8)8) --ANK 881 * 882 * NOTE: is this not too big to inline? 883 */ 884 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 885 { 886 struct tcp_sock *tp = tcp_sk(sk); 887 888 if (sysctl_tcp_low_latency || !tp->ucopy.task) 889 return 0; 890 891 __skb_queue_tail(&tp->ucopy.prequeue, skb); 892 tp->ucopy.memory += skb->truesize; 893 if (tp->ucopy.memory > sk->sk_rcvbuf) { 894 struct sk_buff *skb1; 895 896 BUG_ON(sock_owned_by_user(sk)); 897 898 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 899 sk_backlog_rcv(sk, skb1); 900 NET_INC_STATS_BH(sock_net(sk), 901 LINUX_MIB_TCPPREQUEUEDROPPED); 902 } 903 904 tp->ucopy.memory = 0; 905 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 906 wake_up_interruptible_sync_poll(sk_sleep(sk), 907 POLLIN | POLLRDNORM | POLLRDBAND); 908 if (!inet_csk_ack_scheduled(sk)) 909 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 910 (3 * tcp_rto_min(sk)) / 4, 911 TCP_RTO_MAX); 912 } 913 return 1; 914 } 915 916 917 #undef STATE_TRACE 918 919 #ifdef STATE_TRACE 920 static const char *statename[]={ 921 "Unused","Established","Syn Sent","Syn Recv", 922 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 923 "Close Wait","Last ACK","Listen","Closing" 924 }; 925 #endif 926 extern void tcp_set_state(struct sock *sk, int state); 927 928 extern void tcp_done(struct sock *sk); 929 930 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 931 { 932 rx_opt->dsack = 0; 933 rx_opt->num_sacks = 0; 934 } 935 936 /* Determine a window scaling and initial window to offer. */ 937 extern void tcp_select_initial_window(int __space, __u32 mss, 938 __u32 *rcv_wnd, __u32 *window_clamp, 939 int wscale_ok, __u8 *rcv_wscale, 940 __u32 init_rcv_wnd); 941 942 static inline int tcp_win_from_space(int space) 943 { 944 return sysctl_tcp_adv_win_scale<=0 ? 945 (space>>(-sysctl_tcp_adv_win_scale)) : 946 space - (space>>sysctl_tcp_adv_win_scale); 947 } 948 949 /* Note: caller must be prepared to deal with negative returns */ 950 static inline int tcp_space(const struct sock *sk) 951 { 952 return tcp_win_from_space(sk->sk_rcvbuf - 953 atomic_read(&sk->sk_rmem_alloc)); 954 } 955 956 static inline int tcp_full_space(const struct sock *sk) 957 { 958 return tcp_win_from_space(sk->sk_rcvbuf); 959 } 960 961 static inline void tcp_openreq_init(struct request_sock *req, 962 struct tcp_options_received *rx_opt, 963 struct sk_buff *skb) 964 { 965 struct inet_request_sock *ireq = inet_rsk(req); 966 967 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 968 req->cookie_ts = 0; 969 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 970 req->mss = rx_opt->mss_clamp; 971 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 972 ireq->tstamp_ok = rx_opt->tstamp_ok; 973 ireq->sack_ok = rx_opt->sack_ok; 974 ireq->snd_wscale = rx_opt->snd_wscale; 975 ireq->wscale_ok = rx_opt->wscale_ok; 976 ireq->acked = 0; 977 ireq->ecn_ok = 0; 978 ireq->rmt_port = tcp_hdr(skb)->source; 979 ireq->loc_port = tcp_hdr(skb)->dest; 980 } 981 982 extern void tcp_enter_memory_pressure(struct sock *sk); 983 984 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 985 { 986 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 987 } 988 989 static inline int keepalive_time_when(const struct tcp_sock *tp) 990 { 991 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 992 } 993 994 static inline int keepalive_probes(const struct tcp_sock *tp) 995 { 996 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 997 } 998 999 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1000 { 1001 const struct inet_connection_sock *icsk = &tp->inet_conn; 1002 1003 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1004 tcp_time_stamp - tp->rcv_tstamp); 1005 } 1006 1007 static inline int tcp_fin_time(const struct sock *sk) 1008 { 1009 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1010 const int rto = inet_csk(sk)->icsk_rto; 1011 1012 if (fin_timeout < (rto << 2) - (rto >> 1)) 1013 fin_timeout = (rto << 2) - (rto >> 1); 1014 1015 return fin_timeout; 1016 } 1017 1018 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1019 int paws_win) 1020 { 1021 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1022 return 1; 1023 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1024 return 1; 1025 1026 return 0; 1027 } 1028 1029 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1030 int rst) 1031 { 1032 if (tcp_paws_check(rx_opt, 0)) 1033 return 0; 1034 1035 /* RST segments are not recommended to carry timestamp, 1036 and, if they do, it is recommended to ignore PAWS because 1037 "their cleanup function should take precedence over timestamps." 1038 Certainly, it is mistake. It is necessary to understand the reasons 1039 of this constraint to relax it: if peer reboots, clock may go 1040 out-of-sync and half-open connections will not be reset. 1041 Actually, the problem would be not existing if all 1042 the implementations followed draft about maintaining clock 1043 via reboots. Linux-2.2 DOES NOT! 1044 1045 However, we can relax time bounds for RST segments to MSL. 1046 */ 1047 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1048 return 0; 1049 return 1; 1050 } 1051 1052 #define TCP_CHECK_TIMER(sk) do { } while (0) 1053 1054 static inline void tcp_mib_init(struct net *net) 1055 { 1056 /* See RFC 2012 */ 1057 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1058 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1059 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1060 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1061 } 1062 1063 /* from STCP */ 1064 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1065 { 1066 tp->lost_skb_hint = NULL; 1067 tp->scoreboard_skb_hint = NULL; 1068 } 1069 1070 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1071 { 1072 tcp_clear_retrans_hints_partial(tp); 1073 tp->retransmit_skb_hint = NULL; 1074 } 1075 1076 /* MD5 Signature */ 1077 struct crypto_hash; 1078 1079 /* - key database */ 1080 struct tcp_md5sig_key { 1081 u8 *key; 1082 u8 keylen; 1083 }; 1084 1085 struct tcp4_md5sig_key { 1086 struct tcp_md5sig_key base; 1087 __be32 addr; 1088 }; 1089 1090 struct tcp6_md5sig_key { 1091 struct tcp_md5sig_key base; 1092 #if 0 1093 u32 scope_id; /* XXX */ 1094 #endif 1095 struct in6_addr addr; 1096 }; 1097 1098 /* - sock block */ 1099 struct tcp_md5sig_info { 1100 struct tcp4_md5sig_key *keys4; 1101 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1102 struct tcp6_md5sig_key *keys6; 1103 u32 entries6; 1104 u32 alloced6; 1105 #endif 1106 u32 entries4; 1107 u32 alloced4; 1108 }; 1109 1110 /* - pseudo header */ 1111 struct tcp4_pseudohdr { 1112 __be32 saddr; 1113 __be32 daddr; 1114 __u8 pad; 1115 __u8 protocol; 1116 __be16 len; 1117 }; 1118 1119 struct tcp6_pseudohdr { 1120 struct in6_addr saddr; 1121 struct in6_addr daddr; 1122 __be32 len; 1123 __be32 protocol; /* including padding */ 1124 }; 1125 1126 union tcp_md5sum_block { 1127 struct tcp4_pseudohdr ip4; 1128 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1129 struct tcp6_pseudohdr ip6; 1130 #endif 1131 }; 1132 1133 /* - pool: digest algorithm, hash description and scratch buffer */ 1134 struct tcp_md5sig_pool { 1135 struct hash_desc md5_desc; 1136 union tcp_md5sum_block md5_blk; 1137 }; 1138 1139 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1140 1141 /* - functions */ 1142 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1143 struct sock *sk, struct request_sock *req, 1144 struct sk_buff *skb); 1145 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1146 struct sock *addr_sk); 1147 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1148 u8 newkeylen); 1149 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1150 1151 #ifdef CONFIG_TCP_MD5SIG 1152 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1153 &(struct tcp_md5sig_key) { \ 1154 .key = (twsk)->tw_md5_key, \ 1155 .keylen = (twsk)->tw_md5_keylen, \ 1156 } : NULL) 1157 #else 1158 #define tcp_twsk_md5_key(twsk) NULL 1159 #endif 1160 1161 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1162 extern void tcp_free_md5sig_pool(void); 1163 1164 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1165 extern void tcp_put_md5sig_pool(void); 1166 1167 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1168 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1169 unsigned header_len); 1170 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1171 struct tcp_md5sig_key *key); 1172 1173 /* write queue abstraction */ 1174 static inline void tcp_write_queue_purge(struct sock *sk) 1175 { 1176 struct sk_buff *skb; 1177 1178 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1179 sk_wmem_free_skb(sk, skb); 1180 sk_mem_reclaim(sk); 1181 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1182 } 1183 1184 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1185 { 1186 return skb_peek(&sk->sk_write_queue); 1187 } 1188 1189 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1190 { 1191 return skb_peek_tail(&sk->sk_write_queue); 1192 } 1193 1194 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1195 { 1196 return skb_queue_next(&sk->sk_write_queue, skb); 1197 } 1198 1199 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1200 { 1201 return skb_queue_prev(&sk->sk_write_queue, skb); 1202 } 1203 1204 #define tcp_for_write_queue(skb, sk) \ 1205 skb_queue_walk(&(sk)->sk_write_queue, skb) 1206 1207 #define tcp_for_write_queue_from(skb, sk) \ 1208 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1209 1210 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1211 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1212 1213 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1214 { 1215 return sk->sk_send_head; 1216 } 1217 1218 static inline bool tcp_skb_is_last(const struct sock *sk, 1219 const struct sk_buff *skb) 1220 { 1221 return skb_queue_is_last(&sk->sk_write_queue, skb); 1222 } 1223 1224 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1225 { 1226 if (tcp_skb_is_last(sk, skb)) 1227 sk->sk_send_head = NULL; 1228 else 1229 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1230 } 1231 1232 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1233 { 1234 if (sk->sk_send_head == skb_unlinked) 1235 sk->sk_send_head = NULL; 1236 } 1237 1238 static inline void tcp_init_send_head(struct sock *sk) 1239 { 1240 sk->sk_send_head = NULL; 1241 } 1242 1243 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1244 { 1245 __skb_queue_tail(&sk->sk_write_queue, skb); 1246 } 1247 1248 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1249 { 1250 __tcp_add_write_queue_tail(sk, skb); 1251 1252 /* Queue it, remembering where we must start sending. */ 1253 if (sk->sk_send_head == NULL) { 1254 sk->sk_send_head = skb; 1255 1256 if (tcp_sk(sk)->highest_sack == NULL) 1257 tcp_sk(sk)->highest_sack = skb; 1258 } 1259 } 1260 1261 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1262 { 1263 __skb_queue_head(&sk->sk_write_queue, skb); 1264 } 1265 1266 /* Insert buff after skb on the write queue of sk. */ 1267 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1268 struct sk_buff *buff, 1269 struct sock *sk) 1270 { 1271 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1272 } 1273 1274 /* Insert new before skb on the write queue of sk. */ 1275 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1276 struct sk_buff *skb, 1277 struct sock *sk) 1278 { 1279 __skb_queue_before(&sk->sk_write_queue, skb, new); 1280 1281 if (sk->sk_send_head == skb) 1282 sk->sk_send_head = new; 1283 } 1284 1285 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1286 { 1287 __skb_unlink(skb, &sk->sk_write_queue); 1288 } 1289 1290 static inline int tcp_write_queue_empty(struct sock *sk) 1291 { 1292 return skb_queue_empty(&sk->sk_write_queue); 1293 } 1294 1295 static inline void tcp_push_pending_frames(struct sock *sk) 1296 { 1297 if (tcp_send_head(sk)) { 1298 struct tcp_sock *tp = tcp_sk(sk); 1299 1300 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1301 } 1302 } 1303 1304 /* Start sequence of the highest skb with SACKed bit, valid only if 1305 * sacked > 0 or when the caller has ensured validity by itself. 1306 */ 1307 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1308 { 1309 if (!tp->sacked_out) 1310 return tp->snd_una; 1311 1312 if (tp->highest_sack == NULL) 1313 return tp->snd_nxt; 1314 1315 return TCP_SKB_CB(tp->highest_sack)->seq; 1316 } 1317 1318 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1319 { 1320 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1321 tcp_write_queue_next(sk, skb); 1322 } 1323 1324 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1325 { 1326 return tcp_sk(sk)->highest_sack; 1327 } 1328 1329 static inline void tcp_highest_sack_reset(struct sock *sk) 1330 { 1331 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1332 } 1333 1334 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1335 static inline void tcp_highest_sack_combine(struct sock *sk, 1336 struct sk_buff *old, 1337 struct sk_buff *new) 1338 { 1339 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1340 tcp_sk(sk)->highest_sack = new; 1341 } 1342 1343 /* Determines whether this is a thin stream (which may suffer from 1344 * increased latency). Used to trigger latency-reducing mechanisms. 1345 */ 1346 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1347 { 1348 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1349 } 1350 1351 /* /proc */ 1352 enum tcp_seq_states { 1353 TCP_SEQ_STATE_LISTENING, 1354 TCP_SEQ_STATE_OPENREQ, 1355 TCP_SEQ_STATE_ESTABLISHED, 1356 TCP_SEQ_STATE_TIME_WAIT, 1357 }; 1358 1359 struct tcp_seq_afinfo { 1360 char *name; 1361 sa_family_t family; 1362 struct file_operations seq_fops; 1363 struct seq_operations seq_ops; 1364 }; 1365 1366 struct tcp_iter_state { 1367 struct seq_net_private p; 1368 sa_family_t family; 1369 enum tcp_seq_states state; 1370 struct sock *syn_wait_sk; 1371 int bucket, offset, sbucket, num, uid; 1372 loff_t last_pos; 1373 }; 1374 1375 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1376 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1377 1378 extern struct request_sock_ops tcp_request_sock_ops; 1379 extern struct request_sock_ops tcp6_request_sock_ops; 1380 1381 extern void tcp_v4_destroy_sock(struct sock *sk); 1382 1383 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1384 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1385 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1386 struct sk_buff *skb); 1387 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1388 struct sk_buff *skb); 1389 extern int tcp_gro_complete(struct sk_buff *skb); 1390 extern int tcp4_gro_complete(struct sk_buff *skb); 1391 1392 #ifdef CONFIG_PROC_FS 1393 extern int tcp4_proc_init(void); 1394 extern void tcp4_proc_exit(void); 1395 #endif 1396 1397 /* TCP af-specific functions */ 1398 struct tcp_sock_af_ops { 1399 #ifdef CONFIG_TCP_MD5SIG 1400 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1401 struct sock *addr_sk); 1402 int (*calc_md5_hash) (char *location, 1403 struct tcp_md5sig_key *md5, 1404 struct sock *sk, 1405 struct request_sock *req, 1406 struct sk_buff *skb); 1407 int (*md5_add) (struct sock *sk, 1408 struct sock *addr_sk, 1409 u8 *newkey, 1410 u8 len); 1411 int (*md5_parse) (struct sock *sk, 1412 char __user *optval, 1413 int optlen); 1414 #endif 1415 }; 1416 1417 struct tcp_request_sock_ops { 1418 #ifdef CONFIG_TCP_MD5SIG 1419 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1420 struct request_sock *req); 1421 int (*calc_md5_hash) (char *location, 1422 struct tcp_md5sig_key *md5, 1423 struct sock *sk, 1424 struct request_sock *req, 1425 struct sk_buff *skb); 1426 #endif 1427 }; 1428 1429 /* Using SHA1 for now, define some constants. 1430 */ 1431 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1432 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1433 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1434 1435 extern int tcp_cookie_generator(u32 *bakery); 1436 1437 /** 1438 * struct tcp_cookie_values - each socket needs extra space for the 1439 * cookies, together with (optional) space for any SYN data. 1440 * 1441 * A tcp_sock contains a pointer to the current value, and this is 1442 * cloned to the tcp_timewait_sock. 1443 * 1444 * @cookie_pair: variable data from the option exchange. 1445 * 1446 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1447 * indicates default (sysctl_tcp_cookie_size). 1448 * After cookie sent, remembers size of cookie. 1449 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1450 * 1451 * @s_data_desired: user specified tcpct_s_data_desired. When the 1452 * constant payload is specified (@s_data_constant), 1453 * holds its length instead. 1454 * Range 0 to TCP_MSS_DESIRED. 1455 * 1456 * @s_data_payload: constant data that is to be included in the 1457 * payload of SYN or SYNACK segments when the 1458 * cookie option is present. 1459 */ 1460 struct tcp_cookie_values { 1461 struct kref kref; 1462 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1463 u8 cookie_pair_size; 1464 u8 cookie_desired; 1465 u16 s_data_desired:11, 1466 s_data_constant:1, 1467 s_data_in:1, 1468 s_data_out:1, 1469 s_data_unused:2; 1470 u8 s_data_payload[0]; 1471 }; 1472 1473 static inline void tcp_cookie_values_release(struct kref *kref) 1474 { 1475 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1476 } 1477 1478 /* The length of constant payload data. Note that s_data_desired is 1479 * overloaded, depending on s_data_constant: either the length of constant 1480 * data (returned here) or the limit on variable data. 1481 */ 1482 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1483 { 1484 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1485 ? tp->cookie_values->s_data_desired 1486 : 0; 1487 } 1488 1489 /** 1490 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1491 * 1492 * As tcp_request_sock has already been extended in other places, the 1493 * only remaining method is to pass stack values along as function 1494 * parameters. These parameters are not needed after sending SYNACK. 1495 * 1496 * @cookie_bakery: cryptographic secret and message workspace. 1497 * 1498 * @cookie_plus: bytes in authenticator/cookie option, copied from 1499 * struct tcp_options_received (above). 1500 */ 1501 struct tcp_extend_values { 1502 struct request_values rv; 1503 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1504 u8 cookie_plus:6, 1505 cookie_out_never:1, 1506 cookie_in_always:1; 1507 }; 1508 1509 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1510 { 1511 return (struct tcp_extend_values *)rvp; 1512 } 1513 1514 extern void tcp_v4_init(void); 1515 extern void tcp_init(void); 1516 1517 #endif /* _TCP_H */ 1518