1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define TCP_DEBUG 1 22 #define FASTRETRANS_DEBUG 1 23 24 #include <linux/list.h> 25 #include <linux/tcp.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/dmaengine.h> 31 #include <linux/crypto.h> 32 #include <linux/cryptohash.h> 33 #include <linux/kref.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 extern struct percpu_counter tcp_orphan_count; 52 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54 #define MAX_TCP_HEADER (128 + MAX_HEADER) 55 #define MAX_TCP_OPTION_SPACE 40 56 57 /* 58 * Never offer a window over 32767 without using window scaling. Some 59 * poor stacks do signed 16bit maths! 60 */ 61 #define MAX_TCP_WINDOW 32767U 62 63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 64 #define TCP_MIN_MSS 88U 65 66 /* The least MTU to use for probing */ 67 #define TCP_BASE_MSS 512 68 69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 70 #define TCP_FASTRETRANS_THRESH 3 71 72 /* Maximal reordering. */ 73 #define TCP_MAX_REORDERING 127 74 75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 76 #define TCP_MAX_QUICKACKS 16U 77 78 /* urg_data states */ 79 #define TCP_URG_VALID 0x0100 80 #define TCP_URG_NOTYET 0x0200 81 #define TCP_URG_READ 0x0400 82 83 #define TCP_RETR1 3 /* 84 * This is how many retries it does before it 85 * tries to figure out if the gateway is 86 * down. Minimal RFC value is 3; it corresponds 87 * to ~3sec-8min depending on RTO. 88 */ 89 90 #define TCP_RETR2 15 /* 91 * This should take at least 92 * 90 minutes to time out. 93 * RFC1122 says that the limit is 100 sec. 94 * 15 is ~13-30min depending on RTO. 95 */ 96 97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 98 * connection: ~180sec is RFC minimum */ 99 100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 101 * connection: ~180sec is RFC minimum */ 102 103 104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 105 * socket. 7 is ~50sec-16min. 106 */ 107 108 109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 110 * state, about 60 seconds */ 111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 112 /* BSD style FIN_WAIT2 deadlock breaker. 113 * It used to be 3min, new value is 60sec, 114 * to combine FIN-WAIT-2 timeout with 115 * TIME-WAIT timer. 116 */ 117 118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 119 #if HZ >= 100 120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 121 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 122 #else 123 #define TCP_DELACK_MIN 4U 124 #define TCP_ATO_MIN 4U 125 #endif 126 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 127 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 129 130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 131 * for local resources. 132 */ 133 134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 136 #define TCP_KEEPALIVE_INTVL (75*HZ) 137 138 #define MAX_TCP_KEEPIDLE 32767 139 #define MAX_TCP_KEEPINTVL 32767 140 #define MAX_TCP_KEEPCNT 127 141 #define MAX_TCP_SYNCNT 127 142 143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 144 145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 147 * after this time. It should be equal 148 * (or greater than) TCP_TIMEWAIT_LEN 149 * to provide reliability equal to one 150 * provided by timewait state. 151 */ 152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 153 * timestamps. It must be less than 154 * minimal timewait lifetime. 155 */ 156 /* 157 * TCP option 158 */ 159 160 #define TCPOPT_NOP 1 /* Padding */ 161 #define TCPOPT_EOL 0 /* End of options */ 162 #define TCPOPT_MSS 2 /* Segment size negotiating */ 163 #define TCPOPT_WINDOW 3 /* Window scaling */ 164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 165 #define TCPOPT_SACK 5 /* SACK Block */ 166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 169 170 /* 171 * TCP option lengths 172 */ 173 174 #define TCPOLEN_MSS 4 175 #define TCPOLEN_WINDOW 3 176 #define TCPOLEN_SACK_PERM 2 177 #define TCPOLEN_TIMESTAMP 10 178 #define TCPOLEN_MD5SIG 18 179 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 180 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 181 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 182 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 183 184 /* But this is what stacks really send out. */ 185 #define TCPOLEN_TSTAMP_ALIGNED 12 186 #define TCPOLEN_WSCALE_ALIGNED 4 187 #define TCPOLEN_SACKPERM_ALIGNED 4 188 #define TCPOLEN_SACK_BASE 2 189 #define TCPOLEN_SACK_BASE_ALIGNED 4 190 #define TCPOLEN_SACK_PERBLOCK 8 191 #define TCPOLEN_MD5SIG_ALIGNED 20 192 #define TCPOLEN_MSS_ALIGNED 4 193 194 /* Flags in tp->nonagle */ 195 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 196 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 198 199 /* TCP thin-stream limits */ 200 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 201 202 extern struct inet_timewait_death_row tcp_death_row; 203 204 /* sysctl variables for tcp */ 205 extern int sysctl_tcp_timestamps; 206 extern int sysctl_tcp_window_scaling; 207 extern int sysctl_tcp_sack; 208 extern int sysctl_tcp_fin_timeout; 209 extern int sysctl_tcp_keepalive_time; 210 extern int sysctl_tcp_keepalive_probes; 211 extern int sysctl_tcp_keepalive_intvl; 212 extern int sysctl_tcp_syn_retries; 213 extern int sysctl_tcp_synack_retries; 214 extern int sysctl_tcp_retries1; 215 extern int sysctl_tcp_retries2; 216 extern int sysctl_tcp_orphan_retries; 217 extern int sysctl_tcp_syncookies; 218 extern int sysctl_tcp_retrans_collapse; 219 extern int sysctl_tcp_stdurg; 220 extern int sysctl_tcp_rfc1337; 221 extern int sysctl_tcp_abort_on_overflow; 222 extern int sysctl_tcp_max_orphans; 223 extern int sysctl_tcp_fack; 224 extern int sysctl_tcp_reordering; 225 extern int sysctl_tcp_ecn; 226 extern int sysctl_tcp_dsack; 227 extern int sysctl_tcp_mem[3]; 228 extern int sysctl_tcp_wmem[3]; 229 extern int sysctl_tcp_rmem[3]; 230 extern int sysctl_tcp_app_win; 231 extern int sysctl_tcp_adv_win_scale; 232 extern int sysctl_tcp_tw_reuse; 233 extern int sysctl_tcp_frto; 234 extern int sysctl_tcp_frto_response; 235 extern int sysctl_tcp_low_latency; 236 extern int sysctl_tcp_dma_copybreak; 237 extern int sysctl_tcp_nometrics_save; 238 extern int sysctl_tcp_moderate_rcvbuf; 239 extern int sysctl_tcp_tso_win_divisor; 240 extern int sysctl_tcp_abc; 241 extern int sysctl_tcp_mtu_probing; 242 extern int sysctl_tcp_base_mss; 243 extern int sysctl_tcp_workaround_signed_windows; 244 extern int sysctl_tcp_slow_start_after_idle; 245 extern int sysctl_tcp_max_ssthresh; 246 extern int sysctl_tcp_cookie_size; 247 extern int sysctl_tcp_thin_linear_timeouts; 248 extern int sysctl_tcp_thin_dupack; 249 250 extern atomic_t tcp_memory_allocated; 251 extern struct percpu_counter tcp_sockets_allocated; 252 extern int tcp_memory_pressure; 253 254 /* 255 * The next routines deal with comparing 32 bit unsigned ints 256 * and worry about wraparound (automatic with unsigned arithmetic). 257 */ 258 259 static inline int before(__u32 seq1, __u32 seq2) 260 { 261 return (__s32)(seq1-seq2) < 0; 262 } 263 #define after(seq2, seq1) before(seq1, seq2) 264 265 /* is s2<=s1<=s3 ? */ 266 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 267 { 268 return seq3 - seq2 >= seq1 - seq2; 269 } 270 271 static inline int tcp_too_many_orphans(struct sock *sk, int num) 272 { 273 return (num > sysctl_tcp_max_orphans) || 274 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 275 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]); 276 } 277 278 /* syncookies: remember time of last synqueue overflow */ 279 static inline void tcp_synq_overflow(struct sock *sk) 280 { 281 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 282 } 283 284 /* syncookies: no recent synqueue overflow on this listening socket? */ 285 static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 286 { 287 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 288 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT); 289 } 290 291 extern struct proto tcp_prot; 292 293 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 294 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 295 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 296 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 297 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 298 299 extern void tcp_v4_err(struct sk_buff *skb, u32); 300 301 extern void tcp_shutdown (struct sock *sk, int how); 302 303 extern int tcp_v4_rcv(struct sk_buff *skb); 304 305 extern int tcp_v4_remember_stamp(struct sock *sk); 306 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 307 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 308 size_t size); 309 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 310 size_t size, int flags); 311 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 312 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 313 struct tcphdr *th, unsigned len); 314 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 315 struct tcphdr *th, unsigned len); 316 extern void tcp_rcv_space_adjust(struct sock *sk); 317 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 318 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 319 extern void tcp_twsk_destructor(struct sock *sk); 320 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 321 struct pipe_inode_info *pipe, size_t len, 322 unsigned int flags); 323 324 static inline void tcp_dec_quickack_mode(struct sock *sk, 325 const unsigned int pkts) 326 { 327 struct inet_connection_sock *icsk = inet_csk(sk); 328 329 if (icsk->icsk_ack.quick) { 330 if (pkts >= icsk->icsk_ack.quick) { 331 icsk->icsk_ack.quick = 0; 332 /* Leaving quickack mode we deflate ATO. */ 333 icsk->icsk_ack.ato = TCP_ATO_MIN; 334 } else 335 icsk->icsk_ack.quick -= pkts; 336 } 337 } 338 339 extern void tcp_enter_quickack_mode(struct sock *sk); 340 341 #define TCP_ECN_OK 1 342 #define TCP_ECN_QUEUE_CWR 2 343 #define TCP_ECN_DEMAND_CWR 4 344 345 static __inline__ void 346 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 347 { 348 if (sysctl_tcp_ecn && th->ece && th->cwr) 349 inet_rsk(req)->ecn_ok = 1; 350 } 351 352 enum tcp_tw_status { 353 TCP_TW_SUCCESS = 0, 354 TCP_TW_RST = 1, 355 TCP_TW_ACK = 2, 356 TCP_TW_SYN = 3 357 }; 358 359 360 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 361 struct sk_buff *skb, 362 const struct tcphdr *th); 363 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 364 struct request_sock *req, 365 struct request_sock **prev); 366 extern int tcp_child_process(struct sock *parent, struct sock *child, 367 struct sk_buff *skb); 368 extern int tcp_use_frto(struct sock *sk); 369 extern void tcp_enter_frto(struct sock *sk); 370 extern void tcp_enter_loss(struct sock *sk, int how); 371 extern void tcp_clear_retrans(struct tcp_sock *tp); 372 extern void tcp_update_metrics(struct sock *sk); 373 extern void tcp_close(struct sock *sk, long timeout); 374 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 375 struct poll_table_struct *wait); 376 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 377 char __user *optval, int __user *optlen); 378 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 379 char __user *optval, unsigned int optlen); 380 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 381 char __user *optval, int __user *optlen); 382 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 383 char __user *optval, unsigned int optlen); 384 extern void tcp_set_keepalive(struct sock *sk, int val); 385 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 386 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 387 size_t len, int nonblock, int flags, int *addr_len); 388 extern void tcp_parse_options(struct sk_buff *skb, 389 struct tcp_options_received *opt_rx, u8 **hvpp, 390 int estab); 391 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 392 393 /* 394 * TCP v4 functions exported for the inet6 API 395 */ 396 397 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 398 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 399 extern struct sock * tcp_create_openreq_child(struct sock *sk, 400 struct request_sock *req, 401 struct sk_buff *skb); 402 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 403 struct request_sock *req, 404 struct dst_entry *dst); 405 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 406 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 407 int addr_len); 408 extern int tcp_connect(struct sock *sk); 409 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 410 struct request_sock *req, 411 struct request_values *rvp); 412 extern int tcp_disconnect(struct sock *sk, int flags); 413 414 415 /* From syncookies.c */ 416 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 417 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 418 struct ip_options *opt); 419 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 420 __u16 *mss); 421 422 extern __u32 cookie_init_timestamp(struct request_sock *req); 423 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 424 425 /* From net/ipv6/syncookies.c */ 426 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 427 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 428 __u16 *mss); 429 430 /* tcp_output.c */ 431 432 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 433 int nonagle); 434 extern int tcp_may_send_now(struct sock *sk); 435 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 436 extern void tcp_retransmit_timer(struct sock *sk); 437 extern void tcp_xmit_retransmit_queue(struct sock *); 438 extern void tcp_simple_retransmit(struct sock *); 439 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 440 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 441 442 extern void tcp_send_probe0(struct sock *); 443 extern void tcp_send_partial(struct sock *); 444 extern int tcp_write_wakeup(struct sock *); 445 extern void tcp_send_fin(struct sock *sk); 446 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 447 extern int tcp_send_synack(struct sock *); 448 extern void tcp_push_one(struct sock *, unsigned int mss_now); 449 extern void tcp_send_ack(struct sock *sk); 450 extern void tcp_send_delayed_ack(struct sock *sk); 451 452 /* tcp_input.c */ 453 extern void tcp_cwnd_application_limited(struct sock *sk); 454 455 /* tcp_timer.c */ 456 extern void tcp_init_xmit_timers(struct sock *); 457 static inline void tcp_clear_xmit_timers(struct sock *sk) 458 { 459 inet_csk_clear_xmit_timers(sk); 460 } 461 462 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 463 extern unsigned int tcp_current_mss(struct sock *sk); 464 465 /* Bound MSS / TSO packet size with the half of the window */ 466 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 467 { 468 if (tp->max_window && pktsize > (tp->max_window >> 1)) 469 return max(tp->max_window >> 1, 68U - tp->tcp_header_len); 470 else 471 return pktsize; 472 } 473 474 /* tcp.c */ 475 extern void tcp_get_info(struct sock *, struct tcp_info *); 476 477 /* Read 'sendfile()'-style from a TCP socket */ 478 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 479 unsigned int, size_t); 480 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 481 sk_read_actor_t recv_actor); 482 483 extern void tcp_initialize_rcv_mss(struct sock *sk); 484 485 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 486 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 487 extern void tcp_mtup_init(struct sock *sk); 488 489 static inline void tcp_bound_rto(const struct sock *sk) 490 { 491 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 492 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 493 } 494 495 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 496 { 497 return (tp->srtt >> 3) + tp->rttvar; 498 } 499 500 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 501 { 502 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 503 ntohl(TCP_FLAG_ACK) | 504 snd_wnd); 505 } 506 507 static inline void tcp_fast_path_on(struct tcp_sock *tp) 508 { 509 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 510 } 511 512 static inline void tcp_fast_path_check(struct sock *sk) 513 { 514 struct tcp_sock *tp = tcp_sk(sk); 515 516 if (skb_queue_empty(&tp->out_of_order_queue) && 517 tp->rcv_wnd && 518 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 519 !tp->urg_data) 520 tcp_fast_path_on(tp); 521 } 522 523 /* Compute the actual rto_min value */ 524 static inline u32 tcp_rto_min(struct sock *sk) 525 { 526 struct dst_entry *dst = __sk_dst_get(sk); 527 u32 rto_min = TCP_RTO_MIN; 528 529 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 530 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 531 return rto_min; 532 } 533 534 /* Compute the actual receive window we are currently advertising. 535 * Rcv_nxt can be after the window if our peer push more data 536 * than the offered window. 537 */ 538 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 539 { 540 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 541 542 if (win < 0) 543 win = 0; 544 return (u32) win; 545 } 546 547 /* Choose a new window, without checks for shrinking, and without 548 * scaling applied to the result. The caller does these things 549 * if necessary. This is a "raw" window selection. 550 */ 551 extern u32 __tcp_select_window(struct sock *sk); 552 553 /* TCP timestamps are only 32-bits, this causes a slight 554 * complication on 64-bit systems since we store a snapshot 555 * of jiffies in the buffer control blocks below. We decided 556 * to use only the low 32-bits of jiffies and hide the ugly 557 * casts with the following macro. 558 */ 559 #define tcp_time_stamp ((__u32)(jiffies)) 560 561 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 562 563 #define TCPHDR_FIN 0x01 564 #define TCPHDR_SYN 0x02 565 #define TCPHDR_RST 0x04 566 #define TCPHDR_PSH 0x08 567 #define TCPHDR_ACK 0x10 568 #define TCPHDR_URG 0x20 569 #define TCPHDR_ECE 0x40 570 #define TCPHDR_CWR 0x80 571 572 /* This is what the send packet queuing engine uses to pass 573 * TCP per-packet control information to the transmission code. 574 * We also store the host-order sequence numbers in here too. 575 * This is 44 bytes if IPV6 is enabled. 576 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 577 */ 578 struct tcp_skb_cb { 579 union { 580 struct inet_skb_parm h4; 581 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 582 struct inet6_skb_parm h6; 583 #endif 584 } header; /* For incoming frames */ 585 __u32 seq; /* Starting sequence number */ 586 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 587 __u32 when; /* used to compute rtt's */ 588 __u8 flags; /* TCP header flags. */ 589 __u8 sacked; /* State flags for SACK/FACK. */ 590 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 591 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 592 #define TCPCB_LOST 0x04 /* SKB is lost */ 593 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 594 595 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 596 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 597 598 __u32 ack_seq; /* Sequence number ACK'd */ 599 }; 600 601 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 602 603 /* Due to TSO, an SKB can be composed of multiple actual 604 * packets. To keep these tracked properly, we use this. 605 */ 606 static inline int tcp_skb_pcount(const struct sk_buff *skb) 607 { 608 return skb_shinfo(skb)->gso_segs; 609 } 610 611 /* This is valid iff tcp_skb_pcount() > 1. */ 612 static inline int tcp_skb_mss(const struct sk_buff *skb) 613 { 614 return skb_shinfo(skb)->gso_size; 615 } 616 617 /* Events passed to congestion control interface */ 618 enum tcp_ca_event { 619 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 620 CA_EVENT_CWND_RESTART, /* congestion window restart */ 621 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 622 CA_EVENT_FRTO, /* fast recovery timeout */ 623 CA_EVENT_LOSS, /* loss timeout */ 624 CA_EVENT_FAST_ACK, /* in sequence ack */ 625 CA_EVENT_SLOW_ACK, /* other ack */ 626 }; 627 628 /* 629 * Interface for adding new TCP congestion control handlers 630 */ 631 #define TCP_CA_NAME_MAX 16 632 #define TCP_CA_MAX 128 633 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 634 635 #define TCP_CONG_NON_RESTRICTED 0x1 636 #define TCP_CONG_RTT_STAMP 0x2 637 638 struct tcp_congestion_ops { 639 struct list_head list; 640 unsigned long flags; 641 642 /* initialize private data (optional) */ 643 void (*init)(struct sock *sk); 644 /* cleanup private data (optional) */ 645 void (*release)(struct sock *sk); 646 647 /* return slow start threshold (required) */ 648 u32 (*ssthresh)(struct sock *sk); 649 /* lower bound for congestion window (optional) */ 650 u32 (*min_cwnd)(const struct sock *sk); 651 /* do new cwnd calculation (required) */ 652 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 653 /* call before changing ca_state (optional) */ 654 void (*set_state)(struct sock *sk, u8 new_state); 655 /* call when cwnd event occurs (optional) */ 656 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 657 /* new value of cwnd after loss (optional) */ 658 u32 (*undo_cwnd)(struct sock *sk); 659 /* hook for packet ack accounting (optional) */ 660 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 661 /* get info for inet_diag (optional) */ 662 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 663 664 char name[TCP_CA_NAME_MAX]; 665 struct module *owner; 666 }; 667 668 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 669 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 670 671 extern void tcp_init_congestion_control(struct sock *sk); 672 extern void tcp_cleanup_congestion_control(struct sock *sk); 673 extern int tcp_set_default_congestion_control(const char *name); 674 extern void tcp_get_default_congestion_control(char *name); 675 extern void tcp_get_available_congestion_control(char *buf, size_t len); 676 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 677 extern int tcp_set_allowed_congestion_control(char *allowed); 678 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 679 extern void tcp_slow_start(struct tcp_sock *tp); 680 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 681 682 extern struct tcp_congestion_ops tcp_init_congestion_ops; 683 extern u32 tcp_reno_ssthresh(struct sock *sk); 684 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 685 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 686 extern struct tcp_congestion_ops tcp_reno; 687 688 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 689 { 690 struct inet_connection_sock *icsk = inet_csk(sk); 691 692 if (icsk->icsk_ca_ops->set_state) 693 icsk->icsk_ca_ops->set_state(sk, ca_state); 694 icsk->icsk_ca_state = ca_state; 695 } 696 697 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 698 { 699 const struct inet_connection_sock *icsk = inet_csk(sk); 700 701 if (icsk->icsk_ca_ops->cwnd_event) 702 icsk->icsk_ca_ops->cwnd_event(sk, event); 703 } 704 705 /* These functions determine how the current flow behaves in respect of SACK 706 * handling. SACK is negotiated with the peer, and therefore it can vary 707 * between different flows. 708 * 709 * tcp_is_sack - SACK enabled 710 * tcp_is_reno - No SACK 711 * tcp_is_fack - FACK enabled, implies SACK enabled 712 */ 713 static inline int tcp_is_sack(const struct tcp_sock *tp) 714 { 715 return tp->rx_opt.sack_ok; 716 } 717 718 static inline int tcp_is_reno(const struct tcp_sock *tp) 719 { 720 return !tcp_is_sack(tp); 721 } 722 723 static inline int tcp_is_fack(const struct tcp_sock *tp) 724 { 725 return tp->rx_opt.sack_ok & 2; 726 } 727 728 static inline void tcp_enable_fack(struct tcp_sock *tp) 729 { 730 tp->rx_opt.sack_ok |= 2; 731 } 732 733 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 734 { 735 return tp->sacked_out + tp->lost_out; 736 } 737 738 /* This determines how many packets are "in the network" to the best 739 * of our knowledge. In many cases it is conservative, but where 740 * detailed information is available from the receiver (via SACK 741 * blocks etc.) we can make more aggressive calculations. 742 * 743 * Use this for decisions involving congestion control, use just 744 * tp->packets_out to determine if the send queue is empty or not. 745 * 746 * Read this equation as: 747 * 748 * "Packets sent once on transmission queue" MINUS 749 * "Packets left network, but not honestly ACKed yet" PLUS 750 * "Packets fast retransmitted" 751 */ 752 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 753 { 754 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 755 } 756 757 #define TCP_INFINITE_SSTHRESH 0x7fffffff 758 759 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 760 { 761 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 762 } 763 764 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 765 * The exception is rate halving phase, when cwnd is decreasing towards 766 * ssthresh. 767 */ 768 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 769 { 770 const struct tcp_sock *tp = tcp_sk(sk); 771 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 772 return tp->snd_ssthresh; 773 else 774 return max(tp->snd_ssthresh, 775 ((tp->snd_cwnd >> 1) + 776 (tp->snd_cwnd >> 2))); 777 } 778 779 /* Use define here intentionally to get WARN_ON location shown at the caller */ 780 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 781 782 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 783 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 784 785 /* Slow start with delack produces 3 packets of burst, so that 786 * it is safe "de facto". This will be the default - same as 787 * the default reordering threshold - but if reordering increases, 788 * we must be able to allow cwnd to burst at least this much in order 789 * to not pull it back when holes are filled. 790 */ 791 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 792 { 793 return tp->reordering; 794 } 795 796 /* Returns end sequence number of the receiver's advertised window */ 797 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 798 { 799 return tp->snd_una + tp->snd_wnd; 800 } 801 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 802 803 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 804 const struct sk_buff *skb) 805 { 806 if (skb->len < mss) 807 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 808 } 809 810 static inline void tcp_check_probe_timer(struct sock *sk) 811 { 812 struct tcp_sock *tp = tcp_sk(sk); 813 const struct inet_connection_sock *icsk = inet_csk(sk); 814 815 if (!tp->packets_out && !icsk->icsk_pending) 816 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 817 icsk->icsk_rto, TCP_RTO_MAX); 818 } 819 820 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 821 { 822 tp->snd_wl1 = seq; 823 } 824 825 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 826 { 827 tp->snd_wl1 = seq; 828 } 829 830 /* 831 * Calculate(/check) TCP checksum 832 */ 833 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 834 __be32 daddr, __wsum base) 835 { 836 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 837 } 838 839 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 840 { 841 return __skb_checksum_complete(skb); 842 } 843 844 static inline int tcp_checksum_complete(struct sk_buff *skb) 845 { 846 return !skb_csum_unnecessary(skb) && 847 __tcp_checksum_complete(skb); 848 } 849 850 /* Prequeue for VJ style copy to user, combined with checksumming. */ 851 852 static inline void tcp_prequeue_init(struct tcp_sock *tp) 853 { 854 tp->ucopy.task = NULL; 855 tp->ucopy.len = 0; 856 tp->ucopy.memory = 0; 857 skb_queue_head_init(&tp->ucopy.prequeue); 858 #ifdef CONFIG_NET_DMA 859 tp->ucopy.dma_chan = NULL; 860 tp->ucopy.wakeup = 0; 861 tp->ucopy.pinned_list = NULL; 862 tp->ucopy.dma_cookie = 0; 863 #endif 864 } 865 866 /* Packet is added to VJ-style prequeue for processing in process 867 * context, if a reader task is waiting. Apparently, this exciting 868 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 869 * failed somewhere. Latency? Burstiness? Well, at least now we will 870 * see, why it failed. 8)8) --ANK 871 * 872 * NOTE: is this not too big to inline? 873 */ 874 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 875 { 876 struct tcp_sock *tp = tcp_sk(sk); 877 878 if (sysctl_tcp_low_latency || !tp->ucopy.task) 879 return 0; 880 881 __skb_queue_tail(&tp->ucopy.prequeue, skb); 882 tp->ucopy.memory += skb->truesize; 883 if (tp->ucopy.memory > sk->sk_rcvbuf) { 884 struct sk_buff *skb1; 885 886 BUG_ON(sock_owned_by_user(sk)); 887 888 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 889 sk_backlog_rcv(sk, skb1); 890 NET_INC_STATS_BH(sock_net(sk), 891 LINUX_MIB_TCPPREQUEUEDROPPED); 892 } 893 894 tp->ucopy.memory = 0; 895 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 896 wake_up_interruptible_sync_poll(sk_sleep(sk), 897 POLLIN | POLLRDNORM | POLLRDBAND); 898 if (!inet_csk_ack_scheduled(sk)) 899 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 900 (3 * tcp_rto_min(sk)) / 4, 901 TCP_RTO_MAX); 902 } 903 return 1; 904 } 905 906 907 #undef STATE_TRACE 908 909 #ifdef STATE_TRACE 910 static const char *statename[]={ 911 "Unused","Established","Syn Sent","Syn Recv", 912 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 913 "Close Wait","Last ACK","Listen","Closing" 914 }; 915 #endif 916 extern void tcp_set_state(struct sock *sk, int state); 917 918 extern void tcp_done(struct sock *sk); 919 920 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 921 { 922 rx_opt->dsack = 0; 923 rx_opt->num_sacks = 0; 924 } 925 926 /* Determine a window scaling and initial window to offer. */ 927 extern void tcp_select_initial_window(int __space, __u32 mss, 928 __u32 *rcv_wnd, __u32 *window_clamp, 929 int wscale_ok, __u8 *rcv_wscale, 930 __u32 init_rcv_wnd); 931 932 static inline int tcp_win_from_space(int space) 933 { 934 return sysctl_tcp_adv_win_scale<=0 ? 935 (space>>(-sysctl_tcp_adv_win_scale)) : 936 space - (space>>sysctl_tcp_adv_win_scale); 937 } 938 939 /* Note: caller must be prepared to deal with negative returns */ 940 static inline int tcp_space(const struct sock *sk) 941 { 942 return tcp_win_from_space(sk->sk_rcvbuf - 943 atomic_read(&sk->sk_rmem_alloc)); 944 } 945 946 static inline int tcp_full_space(const struct sock *sk) 947 { 948 return tcp_win_from_space(sk->sk_rcvbuf); 949 } 950 951 static inline void tcp_openreq_init(struct request_sock *req, 952 struct tcp_options_received *rx_opt, 953 struct sk_buff *skb) 954 { 955 struct inet_request_sock *ireq = inet_rsk(req); 956 957 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 958 req->cookie_ts = 0; 959 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 960 req->mss = rx_opt->mss_clamp; 961 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 962 ireq->tstamp_ok = rx_opt->tstamp_ok; 963 ireq->sack_ok = rx_opt->sack_ok; 964 ireq->snd_wscale = rx_opt->snd_wscale; 965 ireq->wscale_ok = rx_opt->wscale_ok; 966 ireq->acked = 0; 967 ireq->ecn_ok = 0; 968 ireq->rmt_port = tcp_hdr(skb)->source; 969 ireq->loc_port = tcp_hdr(skb)->dest; 970 } 971 972 extern void tcp_enter_memory_pressure(struct sock *sk); 973 974 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 975 { 976 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 977 } 978 979 static inline int keepalive_time_when(const struct tcp_sock *tp) 980 { 981 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 982 } 983 984 static inline int keepalive_probes(const struct tcp_sock *tp) 985 { 986 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 987 } 988 989 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 990 { 991 const struct inet_connection_sock *icsk = &tp->inet_conn; 992 993 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 994 tcp_time_stamp - tp->rcv_tstamp); 995 } 996 997 static inline int tcp_fin_time(const struct sock *sk) 998 { 999 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1000 const int rto = inet_csk(sk)->icsk_rto; 1001 1002 if (fin_timeout < (rto << 2) - (rto >> 1)) 1003 fin_timeout = (rto << 2) - (rto >> 1); 1004 1005 return fin_timeout; 1006 } 1007 1008 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1009 int paws_win) 1010 { 1011 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1012 return 1; 1013 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1014 return 1; 1015 1016 return 0; 1017 } 1018 1019 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1020 int rst) 1021 { 1022 if (tcp_paws_check(rx_opt, 0)) 1023 return 0; 1024 1025 /* RST segments are not recommended to carry timestamp, 1026 and, if they do, it is recommended to ignore PAWS because 1027 "their cleanup function should take precedence over timestamps." 1028 Certainly, it is mistake. It is necessary to understand the reasons 1029 of this constraint to relax it: if peer reboots, clock may go 1030 out-of-sync and half-open connections will not be reset. 1031 Actually, the problem would be not existing if all 1032 the implementations followed draft about maintaining clock 1033 via reboots. Linux-2.2 DOES NOT! 1034 1035 However, we can relax time bounds for RST segments to MSL. 1036 */ 1037 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1038 return 0; 1039 return 1; 1040 } 1041 1042 #define TCP_CHECK_TIMER(sk) do { } while (0) 1043 1044 static inline void tcp_mib_init(struct net *net) 1045 { 1046 /* See RFC 2012 */ 1047 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1048 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1049 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1050 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1051 } 1052 1053 /* from STCP */ 1054 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1055 { 1056 tp->lost_skb_hint = NULL; 1057 tp->scoreboard_skb_hint = NULL; 1058 } 1059 1060 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1061 { 1062 tcp_clear_retrans_hints_partial(tp); 1063 tp->retransmit_skb_hint = NULL; 1064 } 1065 1066 /* MD5 Signature */ 1067 struct crypto_hash; 1068 1069 /* - key database */ 1070 struct tcp_md5sig_key { 1071 u8 *key; 1072 u8 keylen; 1073 }; 1074 1075 struct tcp4_md5sig_key { 1076 struct tcp_md5sig_key base; 1077 __be32 addr; 1078 }; 1079 1080 struct tcp6_md5sig_key { 1081 struct tcp_md5sig_key base; 1082 #if 0 1083 u32 scope_id; /* XXX */ 1084 #endif 1085 struct in6_addr addr; 1086 }; 1087 1088 /* - sock block */ 1089 struct tcp_md5sig_info { 1090 struct tcp4_md5sig_key *keys4; 1091 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1092 struct tcp6_md5sig_key *keys6; 1093 u32 entries6; 1094 u32 alloced6; 1095 #endif 1096 u32 entries4; 1097 u32 alloced4; 1098 }; 1099 1100 /* - pseudo header */ 1101 struct tcp4_pseudohdr { 1102 __be32 saddr; 1103 __be32 daddr; 1104 __u8 pad; 1105 __u8 protocol; 1106 __be16 len; 1107 }; 1108 1109 struct tcp6_pseudohdr { 1110 struct in6_addr saddr; 1111 struct in6_addr daddr; 1112 __be32 len; 1113 __be32 protocol; /* including padding */ 1114 }; 1115 1116 union tcp_md5sum_block { 1117 struct tcp4_pseudohdr ip4; 1118 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1119 struct tcp6_pseudohdr ip6; 1120 #endif 1121 }; 1122 1123 /* - pool: digest algorithm, hash description and scratch buffer */ 1124 struct tcp_md5sig_pool { 1125 struct hash_desc md5_desc; 1126 union tcp_md5sum_block md5_blk; 1127 }; 1128 1129 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1130 1131 /* - functions */ 1132 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1133 struct sock *sk, struct request_sock *req, 1134 struct sk_buff *skb); 1135 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1136 struct sock *addr_sk); 1137 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1138 u8 newkeylen); 1139 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1140 1141 #ifdef CONFIG_TCP_MD5SIG 1142 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1143 &(struct tcp_md5sig_key) { \ 1144 .key = (twsk)->tw_md5_key, \ 1145 .keylen = (twsk)->tw_md5_keylen, \ 1146 } : NULL) 1147 #else 1148 #define tcp_twsk_md5_key(twsk) NULL 1149 #endif 1150 1151 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1152 extern void tcp_free_md5sig_pool(void); 1153 1154 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1155 extern void tcp_put_md5sig_pool(void); 1156 1157 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1158 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1159 unsigned header_len); 1160 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1161 struct tcp_md5sig_key *key); 1162 1163 /* write queue abstraction */ 1164 static inline void tcp_write_queue_purge(struct sock *sk) 1165 { 1166 struct sk_buff *skb; 1167 1168 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1169 sk_wmem_free_skb(sk, skb); 1170 sk_mem_reclaim(sk); 1171 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1172 } 1173 1174 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1175 { 1176 return skb_peek(&sk->sk_write_queue); 1177 } 1178 1179 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1180 { 1181 return skb_peek_tail(&sk->sk_write_queue); 1182 } 1183 1184 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1185 { 1186 return skb_queue_next(&sk->sk_write_queue, skb); 1187 } 1188 1189 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1190 { 1191 return skb_queue_prev(&sk->sk_write_queue, skb); 1192 } 1193 1194 #define tcp_for_write_queue(skb, sk) \ 1195 skb_queue_walk(&(sk)->sk_write_queue, skb) 1196 1197 #define tcp_for_write_queue_from(skb, sk) \ 1198 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1199 1200 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1201 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1202 1203 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1204 { 1205 return sk->sk_send_head; 1206 } 1207 1208 static inline bool tcp_skb_is_last(const struct sock *sk, 1209 const struct sk_buff *skb) 1210 { 1211 return skb_queue_is_last(&sk->sk_write_queue, skb); 1212 } 1213 1214 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1215 { 1216 if (tcp_skb_is_last(sk, skb)) 1217 sk->sk_send_head = NULL; 1218 else 1219 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1220 } 1221 1222 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1223 { 1224 if (sk->sk_send_head == skb_unlinked) 1225 sk->sk_send_head = NULL; 1226 } 1227 1228 static inline void tcp_init_send_head(struct sock *sk) 1229 { 1230 sk->sk_send_head = NULL; 1231 } 1232 1233 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1234 { 1235 __skb_queue_tail(&sk->sk_write_queue, skb); 1236 } 1237 1238 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1239 { 1240 __tcp_add_write_queue_tail(sk, skb); 1241 1242 /* Queue it, remembering where we must start sending. */ 1243 if (sk->sk_send_head == NULL) { 1244 sk->sk_send_head = skb; 1245 1246 if (tcp_sk(sk)->highest_sack == NULL) 1247 tcp_sk(sk)->highest_sack = skb; 1248 } 1249 } 1250 1251 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1252 { 1253 __skb_queue_head(&sk->sk_write_queue, skb); 1254 } 1255 1256 /* Insert buff after skb on the write queue of sk. */ 1257 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1258 struct sk_buff *buff, 1259 struct sock *sk) 1260 { 1261 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1262 } 1263 1264 /* Insert new before skb on the write queue of sk. */ 1265 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1266 struct sk_buff *skb, 1267 struct sock *sk) 1268 { 1269 __skb_queue_before(&sk->sk_write_queue, skb, new); 1270 1271 if (sk->sk_send_head == skb) 1272 sk->sk_send_head = new; 1273 } 1274 1275 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1276 { 1277 __skb_unlink(skb, &sk->sk_write_queue); 1278 } 1279 1280 static inline int tcp_write_queue_empty(struct sock *sk) 1281 { 1282 return skb_queue_empty(&sk->sk_write_queue); 1283 } 1284 1285 static inline void tcp_push_pending_frames(struct sock *sk) 1286 { 1287 if (tcp_send_head(sk)) { 1288 struct tcp_sock *tp = tcp_sk(sk); 1289 1290 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1291 } 1292 } 1293 1294 /* Start sequence of the highest skb with SACKed bit, valid only if 1295 * sacked > 0 or when the caller has ensured validity by itself. 1296 */ 1297 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1298 { 1299 if (!tp->sacked_out) 1300 return tp->snd_una; 1301 1302 if (tp->highest_sack == NULL) 1303 return tp->snd_nxt; 1304 1305 return TCP_SKB_CB(tp->highest_sack)->seq; 1306 } 1307 1308 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1309 { 1310 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1311 tcp_write_queue_next(sk, skb); 1312 } 1313 1314 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1315 { 1316 return tcp_sk(sk)->highest_sack; 1317 } 1318 1319 static inline void tcp_highest_sack_reset(struct sock *sk) 1320 { 1321 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1322 } 1323 1324 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1325 static inline void tcp_highest_sack_combine(struct sock *sk, 1326 struct sk_buff *old, 1327 struct sk_buff *new) 1328 { 1329 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1330 tcp_sk(sk)->highest_sack = new; 1331 } 1332 1333 /* Determines whether this is a thin stream (which may suffer from 1334 * increased latency). Used to trigger latency-reducing mechanisms. 1335 */ 1336 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1337 { 1338 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1339 } 1340 1341 /* /proc */ 1342 enum tcp_seq_states { 1343 TCP_SEQ_STATE_LISTENING, 1344 TCP_SEQ_STATE_OPENREQ, 1345 TCP_SEQ_STATE_ESTABLISHED, 1346 TCP_SEQ_STATE_TIME_WAIT, 1347 }; 1348 1349 struct tcp_seq_afinfo { 1350 char *name; 1351 sa_family_t family; 1352 struct file_operations seq_fops; 1353 struct seq_operations seq_ops; 1354 }; 1355 1356 struct tcp_iter_state { 1357 struct seq_net_private p; 1358 sa_family_t family; 1359 enum tcp_seq_states state; 1360 struct sock *syn_wait_sk; 1361 int bucket, offset, sbucket, num, uid; 1362 loff_t last_pos; 1363 }; 1364 1365 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1366 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1367 1368 extern struct request_sock_ops tcp_request_sock_ops; 1369 extern struct request_sock_ops tcp6_request_sock_ops; 1370 1371 extern void tcp_v4_destroy_sock(struct sock *sk); 1372 1373 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1374 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1375 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1376 struct sk_buff *skb); 1377 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1378 struct sk_buff *skb); 1379 extern int tcp_gro_complete(struct sk_buff *skb); 1380 extern int tcp4_gro_complete(struct sk_buff *skb); 1381 1382 #ifdef CONFIG_PROC_FS 1383 extern int tcp4_proc_init(void); 1384 extern void tcp4_proc_exit(void); 1385 #endif 1386 1387 /* TCP af-specific functions */ 1388 struct tcp_sock_af_ops { 1389 #ifdef CONFIG_TCP_MD5SIG 1390 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1391 struct sock *addr_sk); 1392 int (*calc_md5_hash) (char *location, 1393 struct tcp_md5sig_key *md5, 1394 struct sock *sk, 1395 struct request_sock *req, 1396 struct sk_buff *skb); 1397 int (*md5_add) (struct sock *sk, 1398 struct sock *addr_sk, 1399 u8 *newkey, 1400 u8 len); 1401 int (*md5_parse) (struct sock *sk, 1402 char __user *optval, 1403 int optlen); 1404 #endif 1405 }; 1406 1407 struct tcp_request_sock_ops { 1408 #ifdef CONFIG_TCP_MD5SIG 1409 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1410 struct request_sock *req); 1411 int (*calc_md5_hash) (char *location, 1412 struct tcp_md5sig_key *md5, 1413 struct sock *sk, 1414 struct request_sock *req, 1415 struct sk_buff *skb); 1416 #endif 1417 }; 1418 1419 /* Using SHA1 for now, define some constants. 1420 */ 1421 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1422 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1423 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1424 1425 extern int tcp_cookie_generator(u32 *bakery); 1426 1427 /** 1428 * struct tcp_cookie_values - each socket needs extra space for the 1429 * cookies, together with (optional) space for any SYN data. 1430 * 1431 * A tcp_sock contains a pointer to the current value, and this is 1432 * cloned to the tcp_timewait_sock. 1433 * 1434 * @cookie_pair: variable data from the option exchange. 1435 * 1436 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1437 * indicates default (sysctl_tcp_cookie_size). 1438 * After cookie sent, remembers size of cookie. 1439 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1440 * 1441 * @s_data_desired: user specified tcpct_s_data_desired. When the 1442 * constant payload is specified (@s_data_constant), 1443 * holds its length instead. 1444 * Range 0 to TCP_MSS_DESIRED. 1445 * 1446 * @s_data_payload: constant data that is to be included in the 1447 * payload of SYN or SYNACK segments when the 1448 * cookie option is present. 1449 */ 1450 struct tcp_cookie_values { 1451 struct kref kref; 1452 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1453 u8 cookie_pair_size; 1454 u8 cookie_desired; 1455 u16 s_data_desired:11, 1456 s_data_constant:1, 1457 s_data_in:1, 1458 s_data_out:1, 1459 s_data_unused:2; 1460 u8 s_data_payload[0]; 1461 }; 1462 1463 static inline void tcp_cookie_values_release(struct kref *kref) 1464 { 1465 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1466 } 1467 1468 /* The length of constant payload data. Note that s_data_desired is 1469 * overloaded, depending on s_data_constant: either the length of constant 1470 * data (returned here) or the limit on variable data. 1471 */ 1472 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1473 { 1474 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1475 ? tp->cookie_values->s_data_desired 1476 : 0; 1477 } 1478 1479 /** 1480 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1481 * 1482 * As tcp_request_sock has already been extended in other places, the 1483 * only remaining method is to pass stack values along as function 1484 * parameters. These parameters are not needed after sending SYNACK. 1485 * 1486 * @cookie_bakery: cryptographic secret and message workspace. 1487 * 1488 * @cookie_plus: bytes in authenticator/cookie option, copied from 1489 * struct tcp_options_received (above). 1490 */ 1491 struct tcp_extend_values { 1492 struct request_values rv; 1493 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1494 u8 cookie_plus:6, 1495 cookie_out_never:1, 1496 cookie_in_always:1; 1497 }; 1498 1499 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1500 { 1501 return (struct tcp_extend_values *)rvp; 1502 } 1503 1504 extern void tcp_v4_init(void); 1505 extern void tcp_init(void); 1506 1507 #endif /* _TCP_H */ 1508