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 #define TCP_ECN_OK 1 350 #define TCP_ECN_QUEUE_CWR 2 351 #define TCP_ECN_DEMAND_CWR 4 352 353 static __inline__ void 354 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 355 { 356 if (sysctl_tcp_ecn && th->ece && th->cwr) 357 inet_rsk(req)->ecn_ok = 1; 358 } 359 360 enum tcp_tw_status { 361 TCP_TW_SUCCESS = 0, 362 TCP_TW_RST = 1, 363 TCP_TW_ACK = 2, 364 TCP_TW_SYN = 3 365 }; 366 367 368 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 369 struct sk_buff *skb, 370 const struct tcphdr *th); 371 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 372 struct request_sock *req, 373 struct request_sock **prev); 374 extern int tcp_child_process(struct sock *parent, struct sock *child, 375 struct sk_buff *skb); 376 extern int tcp_use_frto(struct sock *sk); 377 extern void tcp_enter_frto(struct sock *sk); 378 extern void tcp_enter_loss(struct sock *sk, int how); 379 extern void tcp_clear_retrans(struct tcp_sock *tp); 380 extern void tcp_update_metrics(struct sock *sk); 381 extern void tcp_close(struct sock *sk, long timeout); 382 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 383 struct poll_table_struct *wait); 384 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 385 char __user *optval, int __user *optlen); 386 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 387 char __user *optval, unsigned int optlen); 388 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 389 char __user *optval, int __user *optlen); 390 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 391 char __user *optval, unsigned int optlen); 392 extern void tcp_set_keepalive(struct sock *sk, int val); 393 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 394 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 395 size_t len, int nonblock, int flags, int *addr_len); 396 extern void tcp_parse_options(struct sk_buff *skb, 397 struct tcp_options_received *opt_rx, u8 **hvpp, 398 int estab); 399 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 400 401 /* 402 * TCP v4 functions exported for the inet6 API 403 */ 404 405 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 406 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 407 extern struct sock * tcp_create_openreq_child(struct sock *sk, 408 struct request_sock *req, 409 struct sk_buff *skb); 410 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 411 struct request_sock *req, 412 struct dst_entry *dst); 413 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 414 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 415 int addr_len); 416 extern int tcp_connect(struct sock *sk); 417 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 418 struct request_sock *req, 419 struct request_values *rvp); 420 extern int tcp_disconnect(struct sock *sk, int flags); 421 422 423 /* From syncookies.c */ 424 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 425 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 426 struct ip_options *opt); 427 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 428 __u16 *mss); 429 430 extern __u32 cookie_init_timestamp(struct request_sock *req); 431 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 432 433 /* From net/ipv6/syncookies.c */ 434 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 435 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 436 __u16 *mss); 437 438 /* tcp_output.c */ 439 440 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 441 int nonagle); 442 extern int tcp_may_send_now(struct sock *sk); 443 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 444 extern void tcp_retransmit_timer(struct sock *sk); 445 extern void tcp_xmit_retransmit_queue(struct sock *); 446 extern void tcp_simple_retransmit(struct sock *); 447 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 448 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 449 450 extern void tcp_send_probe0(struct sock *); 451 extern void tcp_send_partial(struct sock *); 452 extern int tcp_write_wakeup(struct sock *); 453 extern void tcp_send_fin(struct sock *sk); 454 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 455 extern int tcp_send_synack(struct sock *); 456 extern void tcp_push_one(struct sock *, unsigned int mss_now); 457 extern void tcp_send_ack(struct sock *sk); 458 extern void tcp_send_delayed_ack(struct sock *sk); 459 460 /* tcp_input.c */ 461 extern void tcp_cwnd_application_limited(struct sock *sk); 462 463 /* tcp_timer.c */ 464 extern void tcp_init_xmit_timers(struct sock *); 465 static inline void tcp_clear_xmit_timers(struct sock *sk) 466 { 467 inet_csk_clear_xmit_timers(sk); 468 } 469 470 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 471 extern unsigned int tcp_current_mss(struct sock *sk); 472 473 /* Bound MSS / TSO packet size with the half of the window */ 474 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 475 { 476 int cutoff; 477 478 /* When peer uses tiny windows, there is no use in packetizing 479 * to sub-MSS pieces for the sake of SWS or making sure there 480 * are enough packets in the pipe for fast recovery. 481 * 482 * On the other hand, for extremely large MSS devices, handling 483 * smaller than MSS windows in this way does make sense. 484 */ 485 if (tp->max_window >= 512) 486 cutoff = (tp->max_window >> 1); 487 else 488 cutoff = tp->max_window; 489 490 if (cutoff && pktsize > cutoff) 491 return max_t(int, cutoff, 68U - tp->tcp_header_len); 492 else 493 return pktsize; 494 } 495 496 /* tcp.c */ 497 extern void tcp_get_info(struct sock *, struct tcp_info *); 498 499 /* Read 'sendfile()'-style from a TCP socket */ 500 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 501 unsigned int, size_t); 502 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 503 sk_read_actor_t recv_actor); 504 505 extern void tcp_initialize_rcv_mss(struct sock *sk); 506 507 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 508 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 509 extern void tcp_mtup_init(struct sock *sk); 510 511 static inline void tcp_bound_rto(const struct sock *sk) 512 { 513 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 514 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 515 } 516 517 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 518 { 519 return (tp->srtt >> 3) + tp->rttvar; 520 } 521 522 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 523 { 524 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 525 ntohl(TCP_FLAG_ACK) | 526 snd_wnd); 527 } 528 529 static inline void tcp_fast_path_on(struct tcp_sock *tp) 530 { 531 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 532 } 533 534 static inline void tcp_fast_path_check(struct sock *sk) 535 { 536 struct tcp_sock *tp = tcp_sk(sk); 537 538 if (skb_queue_empty(&tp->out_of_order_queue) && 539 tp->rcv_wnd && 540 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 541 !tp->urg_data) 542 tcp_fast_path_on(tp); 543 } 544 545 /* Compute the actual rto_min value */ 546 static inline u32 tcp_rto_min(struct sock *sk) 547 { 548 struct dst_entry *dst = __sk_dst_get(sk); 549 u32 rto_min = TCP_RTO_MIN; 550 551 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 552 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 553 return rto_min; 554 } 555 556 /* Compute the actual receive window we are currently advertising. 557 * Rcv_nxt can be after the window if our peer push more data 558 * than the offered window. 559 */ 560 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 561 { 562 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 563 564 if (win < 0) 565 win = 0; 566 return (u32) win; 567 } 568 569 /* Choose a new window, without checks for shrinking, and without 570 * scaling applied to the result. The caller does these things 571 * if necessary. This is a "raw" window selection. 572 */ 573 extern u32 __tcp_select_window(struct sock *sk); 574 575 /* TCP timestamps are only 32-bits, this causes a slight 576 * complication on 64-bit systems since we store a snapshot 577 * of jiffies in the buffer control blocks below. We decided 578 * to use only the low 32-bits of jiffies and hide the ugly 579 * casts with the following macro. 580 */ 581 #define tcp_time_stamp ((__u32)(jiffies)) 582 583 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 584 585 #define TCPHDR_FIN 0x01 586 #define TCPHDR_SYN 0x02 587 #define TCPHDR_RST 0x04 588 #define TCPHDR_PSH 0x08 589 #define TCPHDR_ACK 0x10 590 #define TCPHDR_URG 0x20 591 #define TCPHDR_ECE 0x40 592 #define TCPHDR_CWR 0x80 593 594 /* This is what the send packet queuing engine uses to pass 595 * TCP per-packet control information to the transmission code. 596 * We also store the host-order sequence numbers in here too. 597 * This is 44 bytes if IPV6 is enabled. 598 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 599 */ 600 struct tcp_skb_cb { 601 union { 602 struct inet_skb_parm h4; 603 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 604 struct inet6_skb_parm h6; 605 #endif 606 } header; /* For incoming frames */ 607 __u32 seq; /* Starting sequence number */ 608 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 609 __u32 when; /* used to compute rtt's */ 610 __u8 flags; /* TCP header flags. */ 611 __u8 sacked; /* State flags for SACK/FACK. */ 612 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 613 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 614 #define TCPCB_LOST 0x04 /* SKB is lost */ 615 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 616 617 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 618 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 619 620 __u32 ack_seq; /* Sequence number ACK'd */ 621 }; 622 623 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 624 625 /* Due to TSO, an SKB can be composed of multiple actual 626 * packets. To keep these tracked properly, we use this. 627 */ 628 static inline int tcp_skb_pcount(const struct sk_buff *skb) 629 { 630 return skb_shinfo(skb)->gso_segs; 631 } 632 633 /* This is valid iff tcp_skb_pcount() > 1. */ 634 static inline int tcp_skb_mss(const struct sk_buff *skb) 635 { 636 return skb_shinfo(skb)->gso_size; 637 } 638 639 /* Events passed to congestion control interface */ 640 enum tcp_ca_event { 641 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 642 CA_EVENT_CWND_RESTART, /* congestion window restart */ 643 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 644 CA_EVENT_FRTO, /* fast recovery timeout */ 645 CA_EVENT_LOSS, /* loss timeout */ 646 CA_EVENT_FAST_ACK, /* in sequence ack */ 647 CA_EVENT_SLOW_ACK, /* other ack */ 648 }; 649 650 /* 651 * Interface for adding new TCP congestion control handlers 652 */ 653 #define TCP_CA_NAME_MAX 16 654 #define TCP_CA_MAX 128 655 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 656 657 #define TCP_CONG_NON_RESTRICTED 0x1 658 #define TCP_CONG_RTT_STAMP 0x2 659 660 struct tcp_congestion_ops { 661 struct list_head list; 662 unsigned long flags; 663 664 /* initialize private data (optional) */ 665 void (*init)(struct sock *sk); 666 /* cleanup private data (optional) */ 667 void (*release)(struct sock *sk); 668 669 /* return slow start threshold (required) */ 670 u32 (*ssthresh)(struct sock *sk); 671 /* lower bound for congestion window (optional) */ 672 u32 (*min_cwnd)(const struct sock *sk); 673 /* do new cwnd calculation (required) */ 674 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 675 /* call before changing ca_state (optional) */ 676 void (*set_state)(struct sock *sk, u8 new_state); 677 /* call when cwnd event occurs (optional) */ 678 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 679 /* new value of cwnd after loss (optional) */ 680 u32 (*undo_cwnd)(struct sock *sk); 681 /* hook for packet ack accounting (optional) */ 682 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 683 /* get info for inet_diag (optional) */ 684 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 685 686 char name[TCP_CA_NAME_MAX]; 687 struct module *owner; 688 }; 689 690 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 691 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 692 693 extern void tcp_init_congestion_control(struct sock *sk); 694 extern void tcp_cleanup_congestion_control(struct sock *sk); 695 extern int tcp_set_default_congestion_control(const char *name); 696 extern void tcp_get_default_congestion_control(char *name); 697 extern void tcp_get_available_congestion_control(char *buf, size_t len); 698 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 699 extern int tcp_set_allowed_congestion_control(char *allowed); 700 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 701 extern void tcp_slow_start(struct tcp_sock *tp); 702 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 703 704 extern struct tcp_congestion_ops tcp_init_congestion_ops; 705 extern u32 tcp_reno_ssthresh(struct sock *sk); 706 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 707 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 708 extern struct tcp_congestion_ops tcp_reno; 709 710 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 711 { 712 struct inet_connection_sock *icsk = inet_csk(sk); 713 714 if (icsk->icsk_ca_ops->set_state) 715 icsk->icsk_ca_ops->set_state(sk, ca_state); 716 icsk->icsk_ca_state = ca_state; 717 } 718 719 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 720 { 721 const struct inet_connection_sock *icsk = inet_csk(sk); 722 723 if (icsk->icsk_ca_ops->cwnd_event) 724 icsk->icsk_ca_ops->cwnd_event(sk, event); 725 } 726 727 /* These functions determine how the current flow behaves in respect of SACK 728 * handling. SACK is negotiated with the peer, and therefore it can vary 729 * between different flows. 730 * 731 * tcp_is_sack - SACK enabled 732 * tcp_is_reno - No SACK 733 * tcp_is_fack - FACK enabled, implies SACK enabled 734 */ 735 static inline int tcp_is_sack(const struct tcp_sock *tp) 736 { 737 return tp->rx_opt.sack_ok; 738 } 739 740 static inline int tcp_is_reno(const struct tcp_sock *tp) 741 { 742 return !tcp_is_sack(tp); 743 } 744 745 static inline int tcp_is_fack(const struct tcp_sock *tp) 746 { 747 return tp->rx_opt.sack_ok & 2; 748 } 749 750 static inline void tcp_enable_fack(struct tcp_sock *tp) 751 { 752 tp->rx_opt.sack_ok |= 2; 753 } 754 755 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 756 { 757 return tp->sacked_out + tp->lost_out; 758 } 759 760 /* This determines how many packets are "in the network" to the best 761 * of our knowledge. In many cases it is conservative, but where 762 * detailed information is available from the receiver (via SACK 763 * blocks etc.) we can make more aggressive calculations. 764 * 765 * Use this for decisions involving congestion control, use just 766 * tp->packets_out to determine if the send queue is empty or not. 767 * 768 * Read this equation as: 769 * 770 * "Packets sent once on transmission queue" MINUS 771 * "Packets left network, but not honestly ACKed yet" PLUS 772 * "Packets fast retransmitted" 773 */ 774 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 775 { 776 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 777 } 778 779 #define TCP_INFINITE_SSTHRESH 0x7fffffff 780 781 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 782 { 783 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 784 } 785 786 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 787 * The exception is rate halving phase, when cwnd is decreasing towards 788 * ssthresh. 789 */ 790 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 791 { 792 const struct tcp_sock *tp = tcp_sk(sk); 793 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 794 return tp->snd_ssthresh; 795 else 796 return max(tp->snd_ssthresh, 797 ((tp->snd_cwnd >> 1) + 798 (tp->snd_cwnd >> 2))); 799 } 800 801 /* Use define here intentionally to get WARN_ON location shown at the caller */ 802 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 803 804 /* 805 * Convert RFC 3390 larger initial window into an equivalent number of packets. 806 * This is based on the numbers specified in RFC 5681, 3.1. 807 */ 808 static inline u32 rfc3390_bytes_to_packets(const u32 smss) 809 { 810 return smss <= 1095 ? 4 : (smss > 2190 ? 2 : 3); 811 } 812 813 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 814 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 815 816 /* Slow start with delack produces 3 packets of burst, so that 817 * it is safe "de facto". This will be the default - same as 818 * the default reordering threshold - but if reordering increases, 819 * we must be able to allow cwnd to burst at least this much in order 820 * to not pull it back when holes are filled. 821 */ 822 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 823 { 824 return tp->reordering; 825 } 826 827 /* Returns end sequence number of the receiver's advertised window */ 828 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 829 { 830 return tp->snd_una + tp->snd_wnd; 831 } 832 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 833 834 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 835 const struct sk_buff *skb) 836 { 837 if (skb->len < mss) 838 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 839 } 840 841 static inline void tcp_check_probe_timer(struct sock *sk) 842 { 843 struct tcp_sock *tp = tcp_sk(sk); 844 const struct inet_connection_sock *icsk = inet_csk(sk); 845 846 if (!tp->packets_out && !icsk->icsk_pending) 847 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 848 icsk->icsk_rto, TCP_RTO_MAX); 849 } 850 851 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 852 { 853 tp->snd_wl1 = seq; 854 } 855 856 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 857 { 858 tp->snd_wl1 = seq; 859 } 860 861 /* 862 * Calculate(/check) TCP checksum 863 */ 864 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 865 __be32 daddr, __wsum base) 866 { 867 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 868 } 869 870 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 871 { 872 return __skb_checksum_complete(skb); 873 } 874 875 static inline int tcp_checksum_complete(struct sk_buff *skb) 876 { 877 return !skb_csum_unnecessary(skb) && 878 __tcp_checksum_complete(skb); 879 } 880 881 /* Prequeue for VJ style copy to user, combined with checksumming. */ 882 883 static inline void tcp_prequeue_init(struct tcp_sock *tp) 884 { 885 tp->ucopy.task = NULL; 886 tp->ucopy.len = 0; 887 tp->ucopy.memory = 0; 888 skb_queue_head_init(&tp->ucopy.prequeue); 889 #ifdef CONFIG_NET_DMA 890 tp->ucopy.dma_chan = NULL; 891 tp->ucopy.wakeup = 0; 892 tp->ucopy.pinned_list = NULL; 893 tp->ucopy.dma_cookie = 0; 894 #endif 895 } 896 897 /* Packet is added to VJ-style prequeue for processing in process 898 * context, if a reader task is waiting. Apparently, this exciting 899 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 900 * failed somewhere. Latency? Burstiness? Well, at least now we will 901 * see, why it failed. 8)8) --ANK 902 * 903 * NOTE: is this not too big to inline? 904 */ 905 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 906 { 907 struct tcp_sock *tp = tcp_sk(sk); 908 909 if (sysctl_tcp_low_latency || !tp->ucopy.task) 910 return 0; 911 912 __skb_queue_tail(&tp->ucopy.prequeue, skb); 913 tp->ucopy.memory += skb->truesize; 914 if (tp->ucopy.memory > sk->sk_rcvbuf) { 915 struct sk_buff *skb1; 916 917 BUG_ON(sock_owned_by_user(sk)); 918 919 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 920 sk_backlog_rcv(sk, skb1); 921 NET_INC_STATS_BH(sock_net(sk), 922 LINUX_MIB_TCPPREQUEUEDROPPED); 923 } 924 925 tp->ucopy.memory = 0; 926 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 927 wake_up_interruptible_sync_poll(sk_sleep(sk), 928 POLLIN | POLLRDNORM | POLLRDBAND); 929 if (!inet_csk_ack_scheduled(sk)) 930 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 931 (3 * tcp_rto_min(sk)) / 4, 932 TCP_RTO_MAX); 933 } 934 return 1; 935 } 936 937 938 #undef STATE_TRACE 939 940 #ifdef STATE_TRACE 941 static const char *statename[]={ 942 "Unused","Established","Syn Sent","Syn Recv", 943 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 944 "Close Wait","Last ACK","Listen","Closing" 945 }; 946 #endif 947 extern void tcp_set_state(struct sock *sk, int state); 948 949 extern void tcp_done(struct sock *sk); 950 951 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 952 { 953 rx_opt->dsack = 0; 954 rx_opt->num_sacks = 0; 955 } 956 957 /* Determine a window scaling and initial window to offer. */ 958 extern void tcp_select_initial_window(int __space, __u32 mss, 959 __u32 *rcv_wnd, __u32 *window_clamp, 960 int wscale_ok, __u8 *rcv_wscale, 961 __u32 init_rcv_wnd); 962 963 static inline int tcp_win_from_space(int space) 964 { 965 return sysctl_tcp_adv_win_scale<=0 ? 966 (space>>(-sysctl_tcp_adv_win_scale)) : 967 space - (space>>sysctl_tcp_adv_win_scale); 968 } 969 970 /* Note: caller must be prepared to deal with negative returns */ 971 static inline int tcp_space(const struct sock *sk) 972 { 973 return tcp_win_from_space(sk->sk_rcvbuf - 974 atomic_read(&sk->sk_rmem_alloc)); 975 } 976 977 static inline int tcp_full_space(const struct sock *sk) 978 { 979 return tcp_win_from_space(sk->sk_rcvbuf); 980 } 981 982 static inline void tcp_openreq_init(struct request_sock *req, 983 struct tcp_options_received *rx_opt, 984 struct sk_buff *skb) 985 { 986 struct inet_request_sock *ireq = inet_rsk(req); 987 988 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 989 req->cookie_ts = 0; 990 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 991 req->mss = rx_opt->mss_clamp; 992 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 993 ireq->tstamp_ok = rx_opt->tstamp_ok; 994 ireq->sack_ok = rx_opt->sack_ok; 995 ireq->snd_wscale = rx_opt->snd_wscale; 996 ireq->wscale_ok = rx_opt->wscale_ok; 997 ireq->acked = 0; 998 ireq->ecn_ok = 0; 999 ireq->rmt_port = tcp_hdr(skb)->source; 1000 ireq->loc_port = tcp_hdr(skb)->dest; 1001 } 1002 1003 extern void tcp_enter_memory_pressure(struct sock *sk); 1004 1005 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1006 { 1007 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1008 } 1009 1010 static inline int keepalive_time_when(const struct tcp_sock *tp) 1011 { 1012 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1013 } 1014 1015 static inline int keepalive_probes(const struct tcp_sock *tp) 1016 { 1017 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1018 } 1019 1020 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1021 { 1022 const struct inet_connection_sock *icsk = &tp->inet_conn; 1023 1024 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1025 tcp_time_stamp - tp->rcv_tstamp); 1026 } 1027 1028 static inline int tcp_fin_time(const struct sock *sk) 1029 { 1030 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1031 const int rto = inet_csk(sk)->icsk_rto; 1032 1033 if (fin_timeout < (rto << 2) - (rto >> 1)) 1034 fin_timeout = (rto << 2) - (rto >> 1); 1035 1036 return fin_timeout; 1037 } 1038 1039 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1040 int paws_win) 1041 { 1042 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1043 return 1; 1044 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1045 return 1; 1046 1047 return 0; 1048 } 1049 1050 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1051 int rst) 1052 { 1053 if (tcp_paws_check(rx_opt, 0)) 1054 return 0; 1055 1056 /* RST segments are not recommended to carry timestamp, 1057 and, if they do, it is recommended to ignore PAWS because 1058 "their cleanup function should take precedence over timestamps." 1059 Certainly, it is mistake. It is necessary to understand the reasons 1060 of this constraint to relax it: if peer reboots, clock may go 1061 out-of-sync and half-open connections will not be reset. 1062 Actually, the problem would be not existing if all 1063 the implementations followed draft about maintaining clock 1064 via reboots. Linux-2.2 DOES NOT! 1065 1066 However, we can relax time bounds for RST segments to MSL. 1067 */ 1068 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1069 return 0; 1070 return 1; 1071 } 1072 1073 #define TCP_CHECK_TIMER(sk) do { } while (0) 1074 1075 static inline void tcp_mib_init(struct net *net) 1076 { 1077 /* See RFC 2012 */ 1078 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1079 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1080 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1081 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1082 } 1083 1084 /* from STCP */ 1085 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1086 { 1087 tp->lost_skb_hint = NULL; 1088 tp->scoreboard_skb_hint = NULL; 1089 } 1090 1091 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1092 { 1093 tcp_clear_retrans_hints_partial(tp); 1094 tp->retransmit_skb_hint = NULL; 1095 } 1096 1097 /* MD5 Signature */ 1098 struct crypto_hash; 1099 1100 /* - key database */ 1101 struct tcp_md5sig_key { 1102 u8 *key; 1103 u8 keylen; 1104 }; 1105 1106 struct tcp4_md5sig_key { 1107 struct tcp_md5sig_key base; 1108 __be32 addr; 1109 }; 1110 1111 struct tcp6_md5sig_key { 1112 struct tcp_md5sig_key base; 1113 #if 0 1114 u32 scope_id; /* XXX */ 1115 #endif 1116 struct in6_addr addr; 1117 }; 1118 1119 /* - sock block */ 1120 struct tcp_md5sig_info { 1121 struct tcp4_md5sig_key *keys4; 1122 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1123 struct tcp6_md5sig_key *keys6; 1124 u32 entries6; 1125 u32 alloced6; 1126 #endif 1127 u32 entries4; 1128 u32 alloced4; 1129 }; 1130 1131 /* - pseudo header */ 1132 struct tcp4_pseudohdr { 1133 __be32 saddr; 1134 __be32 daddr; 1135 __u8 pad; 1136 __u8 protocol; 1137 __be16 len; 1138 }; 1139 1140 struct tcp6_pseudohdr { 1141 struct in6_addr saddr; 1142 struct in6_addr daddr; 1143 __be32 len; 1144 __be32 protocol; /* including padding */ 1145 }; 1146 1147 union tcp_md5sum_block { 1148 struct tcp4_pseudohdr ip4; 1149 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1150 struct tcp6_pseudohdr ip6; 1151 #endif 1152 }; 1153 1154 /* - pool: digest algorithm, hash description and scratch buffer */ 1155 struct tcp_md5sig_pool { 1156 struct hash_desc md5_desc; 1157 union tcp_md5sum_block md5_blk; 1158 }; 1159 1160 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1161 1162 /* - functions */ 1163 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1164 struct sock *sk, struct request_sock *req, 1165 struct sk_buff *skb); 1166 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1167 struct sock *addr_sk); 1168 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1169 u8 newkeylen); 1170 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1171 1172 #ifdef CONFIG_TCP_MD5SIG 1173 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1174 &(struct tcp_md5sig_key) { \ 1175 .key = (twsk)->tw_md5_key, \ 1176 .keylen = (twsk)->tw_md5_keylen, \ 1177 } : NULL) 1178 #else 1179 #define tcp_twsk_md5_key(twsk) NULL 1180 #endif 1181 1182 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1183 extern void tcp_free_md5sig_pool(void); 1184 1185 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1186 extern void tcp_put_md5sig_pool(void); 1187 1188 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1189 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1190 unsigned header_len); 1191 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1192 struct tcp_md5sig_key *key); 1193 1194 /* write queue abstraction */ 1195 static inline void tcp_write_queue_purge(struct sock *sk) 1196 { 1197 struct sk_buff *skb; 1198 1199 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1200 sk_wmem_free_skb(sk, skb); 1201 sk_mem_reclaim(sk); 1202 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1203 } 1204 1205 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1206 { 1207 return skb_peek(&sk->sk_write_queue); 1208 } 1209 1210 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1211 { 1212 return skb_peek_tail(&sk->sk_write_queue); 1213 } 1214 1215 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1216 { 1217 return skb_queue_next(&sk->sk_write_queue, skb); 1218 } 1219 1220 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1221 { 1222 return skb_queue_prev(&sk->sk_write_queue, skb); 1223 } 1224 1225 #define tcp_for_write_queue(skb, sk) \ 1226 skb_queue_walk(&(sk)->sk_write_queue, skb) 1227 1228 #define tcp_for_write_queue_from(skb, sk) \ 1229 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1230 1231 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1232 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1233 1234 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1235 { 1236 return sk->sk_send_head; 1237 } 1238 1239 static inline bool tcp_skb_is_last(const struct sock *sk, 1240 const struct sk_buff *skb) 1241 { 1242 return skb_queue_is_last(&sk->sk_write_queue, skb); 1243 } 1244 1245 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1246 { 1247 if (tcp_skb_is_last(sk, skb)) 1248 sk->sk_send_head = NULL; 1249 else 1250 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1251 } 1252 1253 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1254 { 1255 if (sk->sk_send_head == skb_unlinked) 1256 sk->sk_send_head = NULL; 1257 } 1258 1259 static inline void tcp_init_send_head(struct sock *sk) 1260 { 1261 sk->sk_send_head = NULL; 1262 } 1263 1264 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1265 { 1266 __skb_queue_tail(&sk->sk_write_queue, skb); 1267 } 1268 1269 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1270 { 1271 __tcp_add_write_queue_tail(sk, skb); 1272 1273 /* Queue it, remembering where we must start sending. */ 1274 if (sk->sk_send_head == NULL) { 1275 sk->sk_send_head = skb; 1276 1277 if (tcp_sk(sk)->highest_sack == NULL) 1278 tcp_sk(sk)->highest_sack = skb; 1279 } 1280 } 1281 1282 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1283 { 1284 __skb_queue_head(&sk->sk_write_queue, skb); 1285 } 1286 1287 /* Insert buff after skb on the write queue of sk. */ 1288 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1289 struct sk_buff *buff, 1290 struct sock *sk) 1291 { 1292 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1293 } 1294 1295 /* Insert new before skb on the write queue of sk. */ 1296 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1297 struct sk_buff *skb, 1298 struct sock *sk) 1299 { 1300 __skb_queue_before(&sk->sk_write_queue, skb, new); 1301 1302 if (sk->sk_send_head == skb) 1303 sk->sk_send_head = new; 1304 } 1305 1306 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1307 { 1308 __skb_unlink(skb, &sk->sk_write_queue); 1309 } 1310 1311 static inline int tcp_write_queue_empty(struct sock *sk) 1312 { 1313 return skb_queue_empty(&sk->sk_write_queue); 1314 } 1315 1316 static inline void tcp_push_pending_frames(struct sock *sk) 1317 { 1318 if (tcp_send_head(sk)) { 1319 struct tcp_sock *tp = tcp_sk(sk); 1320 1321 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1322 } 1323 } 1324 1325 /* Start sequence of the highest skb with SACKed bit, valid only if 1326 * sacked > 0 or when the caller has ensured validity by itself. 1327 */ 1328 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1329 { 1330 if (!tp->sacked_out) 1331 return tp->snd_una; 1332 1333 if (tp->highest_sack == NULL) 1334 return tp->snd_nxt; 1335 1336 return TCP_SKB_CB(tp->highest_sack)->seq; 1337 } 1338 1339 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1340 { 1341 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1342 tcp_write_queue_next(sk, skb); 1343 } 1344 1345 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1346 { 1347 return tcp_sk(sk)->highest_sack; 1348 } 1349 1350 static inline void tcp_highest_sack_reset(struct sock *sk) 1351 { 1352 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1353 } 1354 1355 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1356 static inline void tcp_highest_sack_combine(struct sock *sk, 1357 struct sk_buff *old, 1358 struct sk_buff *new) 1359 { 1360 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1361 tcp_sk(sk)->highest_sack = new; 1362 } 1363 1364 /* Determines whether this is a thin stream (which may suffer from 1365 * increased latency). Used to trigger latency-reducing mechanisms. 1366 */ 1367 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1368 { 1369 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1370 } 1371 1372 /* /proc */ 1373 enum tcp_seq_states { 1374 TCP_SEQ_STATE_LISTENING, 1375 TCP_SEQ_STATE_OPENREQ, 1376 TCP_SEQ_STATE_ESTABLISHED, 1377 TCP_SEQ_STATE_TIME_WAIT, 1378 }; 1379 1380 struct tcp_seq_afinfo { 1381 char *name; 1382 sa_family_t family; 1383 struct file_operations seq_fops; 1384 struct seq_operations seq_ops; 1385 }; 1386 1387 struct tcp_iter_state { 1388 struct seq_net_private p; 1389 sa_family_t family; 1390 enum tcp_seq_states state; 1391 struct sock *syn_wait_sk; 1392 int bucket, offset, sbucket, num, uid; 1393 loff_t last_pos; 1394 }; 1395 1396 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1397 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1398 1399 extern struct request_sock_ops tcp_request_sock_ops; 1400 extern struct request_sock_ops tcp6_request_sock_ops; 1401 1402 extern void tcp_v4_destroy_sock(struct sock *sk); 1403 1404 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1405 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1406 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1407 struct sk_buff *skb); 1408 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1409 struct sk_buff *skb); 1410 extern int tcp_gro_complete(struct sk_buff *skb); 1411 extern int tcp4_gro_complete(struct sk_buff *skb); 1412 1413 #ifdef CONFIG_PROC_FS 1414 extern int tcp4_proc_init(void); 1415 extern void tcp4_proc_exit(void); 1416 #endif 1417 1418 /* TCP af-specific functions */ 1419 struct tcp_sock_af_ops { 1420 #ifdef CONFIG_TCP_MD5SIG 1421 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1422 struct sock *addr_sk); 1423 int (*calc_md5_hash) (char *location, 1424 struct tcp_md5sig_key *md5, 1425 struct sock *sk, 1426 struct request_sock *req, 1427 struct sk_buff *skb); 1428 int (*md5_add) (struct sock *sk, 1429 struct sock *addr_sk, 1430 u8 *newkey, 1431 u8 len); 1432 int (*md5_parse) (struct sock *sk, 1433 char __user *optval, 1434 int optlen); 1435 #endif 1436 }; 1437 1438 struct tcp_request_sock_ops { 1439 #ifdef CONFIG_TCP_MD5SIG 1440 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1441 struct request_sock *req); 1442 int (*calc_md5_hash) (char *location, 1443 struct tcp_md5sig_key *md5, 1444 struct sock *sk, 1445 struct request_sock *req, 1446 struct sk_buff *skb); 1447 #endif 1448 }; 1449 1450 /* Using SHA1 for now, define some constants. 1451 */ 1452 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1453 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1454 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1455 1456 extern int tcp_cookie_generator(u32 *bakery); 1457 1458 /** 1459 * struct tcp_cookie_values - each socket needs extra space for the 1460 * cookies, together with (optional) space for any SYN data. 1461 * 1462 * A tcp_sock contains a pointer to the current value, and this is 1463 * cloned to the tcp_timewait_sock. 1464 * 1465 * @cookie_pair: variable data from the option exchange. 1466 * 1467 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1468 * indicates default (sysctl_tcp_cookie_size). 1469 * After cookie sent, remembers size of cookie. 1470 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1471 * 1472 * @s_data_desired: user specified tcpct_s_data_desired. When the 1473 * constant payload is specified (@s_data_constant), 1474 * holds its length instead. 1475 * Range 0 to TCP_MSS_DESIRED. 1476 * 1477 * @s_data_payload: constant data that is to be included in the 1478 * payload of SYN or SYNACK segments when the 1479 * cookie option is present. 1480 */ 1481 struct tcp_cookie_values { 1482 struct kref kref; 1483 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1484 u8 cookie_pair_size; 1485 u8 cookie_desired; 1486 u16 s_data_desired:11, 1487 s_data_constant:1, 1488 s_data_in:1, 1489 s_data_out:1, 1490 s_data_unused:2; 1491 u8 s_data_payload[0]; 1492 }; 1493 1494 static inline void tcp_cookie_values_release(struct kref *kref) 1495 { 1496 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1497 } 1498 1499 /* The length of constant payload data. Note that s_data_desired is 1500 * overloaded, depending on s_data_constant: either the length of constant 1501 * data (returned here) or the limit on variable data. 1502 */ 1503 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1504 { 1505 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1506 ? tp->cookie_values->s_data_desired 1507 : 0; 1508 } 1509 1510 /** 1511 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1512 * 1513 * As tcp_request_sock has already been extended in other places, the 1514 * only remaining method is to pass stack values along as function 1515 * parameters. These parameters are not needed after sending SYNACK. 1516 * 1517 * @cookie_bakery: cryptographic secret and message workspace. 1518 * 1519 * @cookie_plus: bytes in authenticator/cookie option, copied from 1520 * struct tcp_options_received (above). 1521 */ 1522 struct tcp_extend_values { 1523 struct request_values rv; 1524 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1525 u8 cookie_plus:6, 1526 cookie_out_never:1, 1527 cookie_in_always:1; 1528 }; 1529 1530 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1531 { 1532 return (struct tcp_extend_values *)rvp; 1533 } 1534 1535 extern void tcp_v4_init(void); 1536 extern void tcp_init(void); 1537 1538 #endif /* _TCP_H */ 1539