1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define TCP_DEBUG 1 22 #define FASTRETRANS_DEBUG 1 23 24 #include <linux/list.h> 25 #include <linux/tcp.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/dmaengine.h> 31 #include <linux/crypto.h> 32 #include <linux/cryptohash.h> 33 #include <linux/kref.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 extern struct percpu_counter tcp_orphan_count; 52 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54 #define MAX_TCP_HEADER (128 + MAX_HEADER) 55 #define MAX_TCP_OPTION_SPACE 40 56 57 /* 58 * Never offer a window over 32767 without using window scaling. Some 59 * poor stacks do signed 16bit maths! 60 */ 61 #define MAX_TCP_WINDOW 32767U 62 63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 64 #define TCP_MIN_MSS 88U 65 66 /* The least MTU to use for probing */ 67 #define TCP_BASE_MSS 512 68 69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 70 #define TCP_FASTRETRANS_THRESH 3 71 72 /* Maximal reordering. */ 73 #define TCP_MAX_REORDERING 127 74 75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 76 #define TCP_MAX_QUICKACKS 16U 77 78 /* urg_data states */ 79 #define TCP_URG_VALID 0x0100 80 #define TCP_URG_NOTYET 0x0200 81 #define TCP_URG_READ 0x0400 82 83 #define TCP_RETR1 3 /* 84 * This is how many retries it does before it 85 * tries to figure out if the gateway is 86 * down. Minimal RFC value is 3; it corresponds 87 * to ~3sec-8min depending on RTO. 88 */ 89 90 #define TCP_RETR2 15 /* 91 * This should take at least 92 * 90 minutes to time out. 93 * RFC1122 says that the limit is 100 sec. 94 * 15 is ~13-30min depending on RTO. 95 */ 96 97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 98 * connection: ~180sec is RFC minimum */ 99 100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 101 * connection: ~180sec is RFC minimum */ 102 103 104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 105 * socket. 7 is ~50sec-16min. 106 */ 107 108 109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 110 * state, about 60 seconds */ 111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 112 /* BSD style FIN_WAIT2 deadlock breaker. 113 * It used to be 3min, new value is 60sec, 114 * to combine FIN-WAIT-2 timeout with 115 * TIME-WAIT timer. 116 */ 117 118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 119 #if HZ >= 100 120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 121 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 122 #else 123 #define TCP_DELACK_MIN 4U 124 #define TCP_ATO_MIN 4U 125 #endif 126 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 127 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 129 130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 131 * for local resources. 132 */ 133 134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 136 #define TCP_KEEPALIVE_INTVL (75*HZ) 137 138 #define MAX_TCP_KEEPIDLE 32767 139 #define MAX_TCP_KEEPINTVL 32767 140 #define MAX_TCP_KEEPCNT 127 141 #define MAX_TCP_SYNCNT 127 142 143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 144 145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 147 * after this time. It should be equal 148 * (or greater than) TCP_TIMEWAIT_LEN 149 * to provide reliability equal to one 150 * provided by timewait state. 151 */ 152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 153 * timestamps. It must be less than 154 * minimal timewait lifetime. 155 */ 156 /* 157 * TCP option 158 */ 159 160 #define TCPOPT_NOP 1 /* Padding */ 161 #define TCPOPT_EOL 0 /* End of options */ 162 #define TCPOPT_MSS 2 /* Segment size negotiating */ 163 #define TCPOPT_WINDOW 3 /* Window scaling */ 164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 165 #define TCPOPT_SACK 5 /* SACK Block */ 166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 169 170 /* 171 * TCP option lengths 172 */ 173 174 #define TCPOLEN_MSS 4 175 #define TCPOLEN_WINDOW 3 176 #define TCPOLEN_SACK_PERM 2 177 #define TCPOLEN_TIMESTAMP 10 178 #define TCPOLEN_MD5SIG 18 179 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 180 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 181 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 182 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 183 184 /* But this is what stacks really send out. */ 185 #define TCPOLEN_TSTAMP_ALIGNED 12 186 #define TCPOLEN_WSCALE_ALIGNED 4 187 #define TCPOLEN_SACKPERM_ALIGNED 4 188 #define TCPOLEN_SACK_BASE 2 189 #define TCPOLEN_SACK_BASE_ALIGNED 4 190 #define TCPOLEN_SACK_PERBLOCK 8 191 #define TCPOLEN_MD5SIG_ALIGNED 20 192 #define TCPOLEN_MSS_ALIGNED 4 193 194 /* Flags in tp->nonagle */ 195 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 196 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 198 199 /* TCP thin-stream limits */ 200 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 201 202 extern struct inet_timewait_death_row tcp_death_row; 203 204 /* sysctl variables for tcp */ 205 extern int sysctl_tcp_timestamps; 206 extern int sysctl_tcp_window_scaling; 207 extern int sysctl_tcp_sack; 208 extern int sysctl_tcp_fin_timeout; 209 extern int sysctl_tcp_keepalive_time; 210 extern int sysctl_tcp_keepalive_probes; 211 extern int sysctl_tcp_keepalive_intvl; 212 extern int sysctl_tcp_syn_retries; 213 extern int sysctl_tcp_synack_retries; 214 extern int sysctl_tcp_retries1; 215 extern int sysctl_tcp_retries2; 216 extern int sysctl_tcp_orphan_retries; 217 extern int sysctl_tcp_syncookies; 218 extern int sysctl_tcp_retrans_collapse; 219 extern int sysctl_tcp_stdurg; 220 extern int sysctl_tcp_rfc1337; 221 extern int sysctl_tcp_abort_on_overflow; 222 extern int sysctl_tcp_max_orphans; 223 extern int sysctl_tcp_fack; 224 extern int sysctl_tcp_reordering; 225 extern int sysctl_tcp_ecn; 226 extern int sysctl_tcp_dsack; 227 extern int sysctl_tcp_mem[3]; 228 extern int sysctl_tcp_wmem[3]; 229 extern int sysctl_tcp_rmem[3]; 230 extern int sysctl_tcp_app_win; 231 extern int sysctl_tcp_adv_win_scale; 232 extern int sysctl_tcp_tw_reuse; 233 extern int sysctl_tcp_frto; 234 extern int sysctl_tcp_frto_response; 235 extern int sysctl_tcp_low_latency; 236 extern int sysctl_tcp_dma_copybreak; 237 extern int sysctl_tcp_nometrics_save; 238 extern int sysctl_tcp_moderate_rcvbuf; 239 extern int sysctl_tcp_tso_win_divisor; 240 extern int sysctl_tcp_abc; 241 extern int sysctl_tcp_mtu_probing; 242 extern int sysctl_tcp_base_mss; 243 extern int sysctl_tcp_workaround_signed_windows; 244 extern int sysctl_tcp_slow_start_after_idle; 245 extern int sysctl_tcp_max_ssthresh; 246 extern int sysctl_tcp_cookie_size; 247 extern int sysctl_tcp_thin_linear_timeouts; 248 extern int sysctl_tcp_thin_dupack; 249 250 extern atomic_t tcp_memory_allocated; 251 extern struct percpu_counter tcp_sockets_allocated; 252 extern int tcp_memory_pressure; 253 254 /* 255 * The next routines deal with comparing 32 bit unsigned ints 256 * and worry about wraparound (automatic with unsigned arithmetic). 257 */ 258 259 static inline int before(__u32 seq1, __u32 seq2) 260 { 261 return (__s32)(seq1-seq2) < 0; 262 } 263 #define after(seq2, seq1) before(seq1, seq2) 264 265 /* is s2<=s1<=s3 ? */ 266 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 267 { 268 return seq3 - seq2 >= seq1 - seq2; 269 } 270 271 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 272 { 273 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 274 int orphans = percpu_counter_read_positive(ocp); 275 276 if (orphans << shift > sysctl_tcp_max_orphans) { 277 orphans = percpu_counter_sum_positive(ocp); 278 if (orphans << shift > sysctl_tcp_max_orphans) 279 return true; 280 } 281 282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 283 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]) 284 return true; 285 return false; 286 } 287 288 /* syncookies: remember time of last synqueue overflow */ 289 static inline void tcp_synq_overflow(struct sock *sk) 290 { 291 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 292 } 293 294 /* syncookies: no recent synqueue overflow on this listening socket? */ 295 static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 296 { 297 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 298 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT); 299 } 300 301 extern struct proto tcp_prot; 302 303 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 304 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 305 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 306 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 307 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 308 309 extern void tcp_v4_err(struct sk_buff *skb, u32); 310 311 extern void tcp_shutdown (struct sock *sk, int how); 312 313 extern int tcp_v4_rcv(struct sk_buff *skb); 314 315 extern int tcp_v4_remember_stamp(struct sock *sk); 316 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 317 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 318 size_t size); 319 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 320 size_t size, int flags); 321 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 322 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 323 struct tcphdr *th, unsigned len); 324 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 325 struct tcphdr *th, unsigned len); 326 extern void tcp_rcv_space_adjust(struct sock *sk); 327 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 328 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 329 extern void tcp_twsk_destructor(struct sock *sk); 330 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 331 struct pipe_inode_info *pipe, size_t len, 332 unsigned int flags); 333 334 static inline void tcp_dec_quickack_mode(struct sock *sk, 335 const unsigned int pkts) 336 { 337 struct inet_connection_sock *icsk = inet_csk(sk); 338 339 if (icsk->icsk_ack.quick) { 340 if (pkts >= icsk->icsk_ack.quick) { 341 icsk->icsk_ack.quick = 0; 342 /* Leaving quickack mode we deflate ATO. */ 343 icsk->icsk_ack.ato = TCP_ATO_MIN; 344 } else 345 icsk->icsk_ack.quick -= pkts; 346 } 347 } 348 349 extern void tcp_enter_quickack_mode(struct sock *sk); 350 351 #define TCP_ECN_OK 1 352 #define TCP_ECN_QUEUE_CWR 2 353 #define TCP_ECN_DEMAND_CWR 4 354 355 static __inline__ void 356 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 357 { 358 if (sysctl_tcp_ecn && th->ece && th->cwr) 359 inet_rsk(req)->ecn_ok = 1; 360 } 361 362 enum tcp_tw_status { 363 TCP_TW_SUCCESS = 0, 364 TCP_TW_RST = 1, 365 TCP_TW_ACK = 2, 366 TCP_TW_SYN = 3 367 }; 368 369 370 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 371 struct sk_buff *skb, 372 const struct tcphdr *th); 373 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 374 struct request_sock *req, 375 struct request_sock **prev); 376 extern int tcp_child_process(struct sock *parent, struct sock *child, 377 struct sk_buff *skb); 378 extern int tcp_use_frto(struct sock *sk); 379 extern void tcp_enter_frto(struct sock *sk); 380 extern void tcp_enter_loss(struct sock *sk, int how); 381 extern void tcp_clear_retrans(struct tcp_sock *tp); 382 extern void tcp_update_metrics(struct sock *sk); 383 extern void tcp_close(struct sock *sk, long timeout); 384 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 385 struct poll_table_struct *wait); 386 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 387 char __user *optval, int __user *optlen); 388 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 389 char __user *optval, unsigned int optlen); 390 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 391 char __user *optval, int __user *optlen); 392 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 393 char __user *optval, unsigned int optlen); 394 extern void tcp_set_keepalive(struct sock *sk, int val); 395 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 396 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 397 size_t len, int nonblock, int flags, int *addr_len); 398 extern void tcp_parse_options(struct sk_buff *skb, 399 struct tcp_options_received *opt_rx, u8 **hvpp, 400 int estab); 401 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 402 403 /* 404 * TCP v4 functions exported for the inet6 API 405 */ 406 407 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 408 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 409 extern struct sock * tcp_create_openreq_child(struct sock *sk, 410 struct request_sock *req, 411 struct sk_buff *skb); 412 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 413 struct request_sock *req, 414 struct dst_entry *dst); 415 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 416 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 417 int addr_len); 418 extern int tcp_connect(struct sock *sk); 419 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 420 struct request_sock *req, 421 struct request_values *rvp); 422 extern int tcp_disconnect(struct sock *sk, int flags); 423 424 425 /* From syncookies.c */ 426 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 427 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 428 struct ip_options *opt); 429 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 430 __u16 *mss); 431 432 extern __u32 cookie_init_timestamp(struct request_sock *req); 433 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 434 435 /* From net/ipv6/syncookies.c */ 436 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 437 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 438 __u16 *mss); 439 440 /* tcp_output.c */ 441 442 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 443 int nonagle); 444 extern int tcp_may_send_now(struct sock *sk); 445 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 446 extern void tcp_retransmit_timer(struct sock *sk); 447 extern void tcp_xmit_retransmit_queue(struct sock *); 448 extern void tcp_simple_retransmit(struct sock *); 449 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 450 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 451 452 extern void tcp_send_probe0(struct sock *); 453 extern void tcp_send_partial(struct sock *); 454 extern int tcp_write_wakeup(struct sock *); 455 extern void tcp_send_fin(struct sock *sk); 456 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 457 extern int tcp_send_synack(struct sock *); 458 extern void tcp_push_one(struct sock *, unsigned int mss_now); 459 extern void tcp_send_ack(struct sock *sk); 460 extern void tcp_send_delayed_ack(struct sock *sk); 461 462 /* tcp_input.c */ 463 extern void tcp_cwnd_application_limited(struct sock *sk); 464 465 /* tcp_timer.c */ 466 extern void tcp_init_xmit_timers(struct sock *); 467 static inline void tcp_clear_xmit_timers(struct sock *sk) 468 { 469 inet_csk_clear_xmit_timers(sk); 470 } 471 472 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 473 extern unsigned int tcp_current_mss(struct sock *sk); 474 475 /* Bound MSS / TSO packet size with the half of the window */ 476 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 477 { 478 int cutoff; 479 480 /* When peer uses tiny windows, there is no use in packetizing 481 * to sub-MSS pieces for the sake of SWS or making sure there 482 * are enough packets in the pipe for fast recovery. 483 * 484 * On the other hand, for extremely large MSS devices, handling 485 * smaller than MSS windows in this way does make sense. 486 */ 487 if (tp->max_window >= 512) 488 cutoff = (tp->max_window >> 1); 489 else 490 cutoff = tp->max_window; 491 492 if (cutoff && pktsize > cutoff) 493 return max_t(int, cutoff, 68U - tp->tcp_header_len); 494 else 495 return pktsize; 496 } 497 498 /* tcp.c */ 499 extern void tcp_get_info(struct sock *, struct tcp_info *); 500 501 /* Read 'sendfile()'-style from a TCP socket */ 502 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 503 unsigned int, size_t); 504 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 505 sk_read_actor_t recv_actor); 506 507 extern void tcp_initialize_rcv_mss(struct sock *sk); 508 509 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 510 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 511 extern void tcp_mtup_init(struct sock *sk); 512 513 static inline void tcp_bound_rto(const struct sock *sk) 514 { 515 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 516 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 517 } 518 519 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 520 { 521 return (tp->srtt >> 3) + tp->rttvar; 522 } 523 524 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 525 { 526 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 527 ntohl(TCP_FLAG_ACK) | 528 snd_wnd); 529 } 530 531 static inline void tcp_fast_path_on(struct tcp_sock *tp) 532 { 533 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 534 } 535 536 static inline void tcp_fast_path_check(struct sock *sk) 537 { 538 struct tcp_sock *tp = tcp_sk(sk); 539 540 if (skb_queue_empty(&tp->out_of_order_queue) && 541 tp->rcv_wnd && 542 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 543 !tp->urg_data) 544 tcp_fast_path_on(tp); 545 } 546 547 /* Compute the actual rto_min value */ 548 static inline u32 tcp_rto_min(struct sock *sk) 549 { 550 struct dst_entry *dst = __sk_dst_get(sk); 551 u32 rto_min = TCP_RTO_MIN; 552 553 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 554 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 555 return rto_min; 556 } 557 558 /* Compute the actual receive window we are currently advertising. 559 * Rcv_nxt can be after the window if our peer push more data 560 * than the offered window. 561 */ 562 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 563 { 564 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 565 566 if (win < 0) 567 win = 0; 568 return (u32) win; 569 } 570 571 /* Choose a new window, without checks for shrinking, and without 572 * scaling applied to the result. The caller does these things 573 * if necessary. This is a "raw" window selection. 574 */ 575 extern u32 __tcp_select_window(struct sock *sk); 576 577 /* TCP timestamps are only 32-bits, this causes a slight 578 * complication on 64-bit systems since we store a snapshot 579 * of jiffies in the buffer control blocks below. We decided 580 * to use only the low 32-bits of jiffies and hide the ugly 581 * casts with the following macro. 582 */ 583 #define tcp_time_stamp ((__u32)(jiffies)) 584 585 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 586 587 #define TCPHDR_FIN 0x01 588 #define TCPHDR_SYN 0x02 589 #define TCPHDR_RST 0x04 590 #define TCPHDR_PSH 0x08 591 #define TCPHDR_ACK 0x10 592 #define TCPHDR_URG 0x20 593 #define TCPHDR_ECE 0x40 594 #define TCPHDR_CWR 0x80 595 596 /* This is what the send packet queuing engine uses to pass 597 * TCP per-packet control information to the transmission code. 598 * We also store the host-order sequence numbers in here too. 599 * This is 44 bytes if IPV6 is enabled. 600 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 601 */ 602 struct tcp_skb_cb { 603 union { 604 struct inet_skb_parm h4; 605 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 606 struct inet6_skb_parm h6; 607 #endif 608 } header; /* For incoming frames */ 609 __u32 seq; /* Starting sequence number */ 610 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 611 __u32 when; /* used to compute rtt's */ 612 __u8 flags; /* TCP header flags. */ 613 __u8 sacked; /* State flags for SACK/FACK. */ 614 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 615 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 616 #define TCPCB_LOST 0x04 /* SKB is lost */ 617 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 618 619 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 620 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 621 622 __u32 ack_seq; /* Sequence number ACK'd */ 623 }; 624 625 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 626 627 /* Due to TSO, an SKB can be composed of multiple actual 628 * packets. To keep these tracked properly, we use this. 629 */ 630 static inline int tcp_skb_pcount(const struct sk_buff *skb) 631 { 632 return skb_shinfo(skb)->gso_segs; 633 } 634 635 /* This is valid iff tcp_skb_pcount() > 1. */ 636 static inline int tcp_skb_mss(const struct sk_buff *skb) 637 { 638 return skb_shinfo(skb)->gso_size; 639 } 640 641 /* Events passed to congestion control interface */ 642 enum tcp_ca_event { 643 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 644 CA_EVENT_CWND_RESTART, /* congestion window restart */ 645 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 646 CA_EVENT_FRTO, /* fast recovery timeout */ 647 CA_EVENT_LOSS, /* loss timeout */ 648 CA_EVENT_FAST_ACK, /* in sequence ack */ 649 CA_EVENT_SLOW_ACK, /* other ack */ 650 }; 651 652 /* 653 * Interface for adding new TCP congestion control handlers 654 */ 655 #define TCP_CA_NAME_MAX 16 656 #define TCP_CA_MAX 128 657 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 658 659 #define TCP_CONG_NON_RESTRICTED 0x1 660 #define TCP_CONG_RTT_STAMP 0x2 661 662 struct tcp_congestion_ops { 663 struct list_head list; 664 unsigned long flags; 665 666 /* initialize private data (optional) */ 667 void (*init)(struct sock *sk); 668 /* cleanup private data (optional) */ 669 void (*release)(struct sock *sk); 670 671 /* return slow start threshold (required) */ 672 u32 (*ssthresh)(struct sock *sk); 673 /* lower bound for congestion window (optional) */ 674 u32 (*min_cwnd)(const struct sock *sk); 675 /* do new cwnd calculation (required) */ 676 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 677 /* call before changing ca_state (optional) */ 678 void (*set_state)(struct sock *sk, u8 new_state); 679 /* call when cwnd event occurs (optional) */ 680 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 681 /* new value of cwnd after loss (optional) */ 682 u32 (*undo_cwnd)(struct sock *sk); 683 /* hook for packet ack accounting (optional) */ 684 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 685 /* get info for inet_diag (optional) */ 686 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 687 688 char name[TCP_CA_NAME_MAX]; 689 struct module *owner; 690 }; 691 692 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 693 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 694 695 extern void tcp_init_congestion_control(struct sock *sk); 696 extern void tcp_cleanup_congestion_control(struct sock *sk); 697 extern int tcp_set_default_congestion_control(const char *name); 698 extern void tcp_get_default_congestion_control(char *name); 699 extern void tcp_get_available_congestion_control(char *buf, size_t len); 700 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 701 extern int tcp_set_allowed_congestion_control(char *allowed); 702 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 703 extern void tcp_slow_start(struct tcp_sock *tp); 704 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 705 706 extern struct tcp_congestion_ops tcp_init_congestion_ops; 707 extern u32 tcp_reno_ssthresh(struct sock *sk); 708 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 709 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 710 extern struct tcp_congestion_ops tcp_reno; 711 712 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 713 { 714 struct inet_connection_sock *icsk = inet_csk(sk); 715 716 if (icsk->icsk_ca_ops->set_state) 717 icsk->icsk_ca_ops->set_state(sk, ca_state); 718 icsk->icsk_ca_state = ca_state; 719 } 720 721 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 722 { 723 const struct inet_connection_sock *icsk = inet_csk(sk); 724 725 if (icsk->icsk_ca_ops->cwnd_event) 726 icsk->icsk_ca_ops->cwnd_event(sk, event); 727 } 728 729 /* These functions determine how the current flow behaves in respect of SACK 730 * handling. SACK is negotiated with the peer, and therefore it can vary 731 * between different flows. 732 * 733 * tcp_is_sack - SACK enabled 734 * tcp_is_reno - No SACK 735 * tcp_is_fack - FACK enabled, implies SACK enabled 736 */ 737 static inline int tcp_is_sack(const struct tcp_sock *tp) 738 { 739 return tp->rx_opt.sack_ok; 740 } 741 742 static inline int tcp_is_reno(const struct tcp_sock *tp) 743 { 744 return !tcp_is_sack(tp); 745 } 746 747 static inline int tcp_is_fack(const struct tcp_sock *tp) 748 { 749 return tp->rx_opt.sack_ok & 2; 750 } 751 752 static inline void tcp_enable_fack(struct tcp_sock *tp) 753 { 754 tp->rx_opt.sack_ok |= 2; 755 } 756 757 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 758 { 759 return tp->sacked_out + tp->lost_out; 760 } 761 762 /* This determines how many packets are "in the network" to the best 763 * of our knowledge. In many cases it is conservative, but where 764 * detailed information is available from the receiver (via SACK 765 * blocks etc.) we can make more aggressive calculations. 766 * 767 * Use this for decisions involving congestion control, use just 768 * tp->packets_out to determine if the send queue is empty or not. 769 * 770 * Read this equation as: 771 * 772 * "Packets sent once on transmission queue" MINUS 773 * "Packets left network, but not honestly ACKed yet" PLUS 774 * "Packets fast retransmitted" 775 */ 776 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 777 { 778 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 779 } 780 781 #define TCP_INFINITE_SSTHRESH 0x7fffffff 782 783 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 784 { 785 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 786 } 787 788 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 789 * The exception is rate halving phase, when cwnd is decreasing towards 790 * ssthresh. 791 */ 792 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 793 { 794 const struct tcp_sock *tp = tcp_sk(sk); 795 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 796 return tp->snd_ssthresh; 797 else 798 return max(tp->snd_ssthresh, 799 ((tp->snd_cwnd >> 1) + 800 (tp->snd_cwnd >> 2))); 801 } 802 803 /* Use define here intentionally to get WARN_ON location shown at the caller */ 804 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 805 806 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 807 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 808 809 /* Slow start with delack produces 3 packets of burst, so that 810 * it is safe "de facto". This will be the default - same as 811 * the default reordering threshold - but if reordering increases, 812 * we must be able to allow cwnd to burst at least this much in order 813 * to not pull it back when holes are filled. 814 */ 815 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 816 { 817 return tp->reordering; 818 } 819 820 /* Returns end sequence number of the receiver's advertised window */ 821 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 822 { 823 return tp->snd_una + tp->snd_wnd; 824 } 825 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 826 827 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 828 const struct sk_buff *skb) 829 { 830 if (skb->len < mss) 831 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 832 } 833 834 static inline void tcp_check_probe_timer(struct sock *sk) 835 { 836 struct tcp_sock *tp = tcp_sk(sk); 837 const struct inet_connection_sock *icsk = inet_csk(sk); 838 839 if (!tp->packets_out && !icsk->icsk_pending) 840 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 841 icsk->icsk_rto, TCP_RTO_MAX); 842 } 843 844 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 845 { 846 tp->snd_wl1 = seq; 847 } 848 849 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 850 { 851 tp->snd_wl1 = seq; 852 } 853 854 /* 855 * Calculate(/check) TCP checksum 856 */ 857 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 858 __be32 daddr, __wsum base) 859 { 860 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 861 } 862 863 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 864 { 865 return __skb_checksum_complete(skb); 866 } 867 868 static inline int tcp_checksum_complete(struct sk_buff *skb) 869 { 870 return !skb_csum_unnecessary(skb) && 871 __tcp_checksum_complete(skb); 872 } 873 874 /* Prequeue for VJ style copy to user, combined with checksumming. */ 875 876 static inline void tcp_prequeue_init(struct tcp_sock *tp) 877 { 878 tp->ucopy.task = NULL; 879 tp->ucopy.len = 0; 880 tp->ucopy.memory = 0; 881 skb_queue_head_init(&tp->ucopy.prequeue); 882 #ifdef CONFIG_NET_DMA 883 tp->ucopy.dma_chan = NULL; 884 tp->ucopy.wakeup = 0; 885 tp->ucopy.pinned_list = NULL; 886 tp->ucopy.dma_cookie = 0; 887 #endif 888 } 889 890 /* Packet is added to VJ-style prequeue for processing in process 891 * context, if a reader task is waiting. Apparently, this exciting 892 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 893 * failed somewhere. Latency? Burstiness? Well, at least now we will 894 * see, why it failed. 8)8) --ANK 895 * 896 * NOTE: is this not too big to inline? 897 */ 898 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 899 { 900 struct tcp_sock *tp = tcp_sk(sk); 901 902 if (sysctl_tcp_low_latency || !tp->ucopy.task) 903 return 0; 904 905 __skb_queue_tail(&tp->ucopy.prequeue, skb); 906 tp->ucopy.memory += skb->truesize; 907 if (tp->ucopy.memory > sk->sk_rcvbuf) { 908 struct sk_buff *skb1; 909 910 BUG_ON(sock_owned_by_user(sk)); 911 912 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 913 sk_backlog_rcv(sk, skb1); 914 NET_INC_STATS_BH(sock_net(sk), 915 LINUX_MIB_TCPPREQUEUEDROPPED); 916 } 917 918 tp->ucopy.memory = 0; 919 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 920 wake_up_interruptible_sync_poll(sk_sleep(sk), 921 POLLIN | POLLRDNORM | POLLRDBAND); 922 if (!inet_csk_ack_scheduled(sk)) 923 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 924 (3 * tcp_rto_min(sk)) / 4, 925 TCP_RTO_MAX); 926 } 927 return 1; 928 } 929 930 931 #undef STATE_TRACE 932 933 #ifdef STATE_TRACE 934 static const char *statename[]={ 935 "Unused","Established","Syn Sent","Syn Recv", 936 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 937 "Close Wait","Last ACK","Listen","Closing" 938 }; 939 #endif 940 extern void tcp_set_state(struct sock *sk, int state); 941 942 extern void tcp_done(struct sock *sk); 943 944 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 945 { 946 rx_opt->dsack = 0; 947 rx_opt->num_sacks = 0; 948 } 949 950 /* Determine a window scaling and initial window to offer. */ 951 extern void tcp_select_initial_window(int __space, __u32 mss, 952 __u32 *rcv_wnd, __u32 *window_clamp, 953 int wscale_ok, __u8 *rcv_wscale, 954 __u32 init_rcv_wnd); 955 956 static inline int tcp_win_from_space(int space) 957 { 958 return sysctl_tcp_adv_win_scale<=0 ? 959 (space>>(-sysctl_tcp_adv_win_scale)) : 960 space - (space>>sysctl_tcp_adv_win_scale); 961 } 962 963 /* Note: caller must be prepared to deal with negative returns */ 964 static inline int tcp_space(const struct sock *sk) 965 { 966 return tcp_win_from_space(sk->sk_rcvbuf - 967 atomic_read(&sk->sk_rmem_alloc)); 968 } 969 970 static inline int tcp_full_space(const struct sock *sk) 971 { 972 return tcp_win_from_space(sk->sk_rcvbuf); 973 } 974 975 static inline void tcp_openreq_init(struct request_sock *req, 976 struct tcp_options_received *rx_opt, 977 struct sk_buff *skb) 978 { 979 struct inet_request_sock *ireq = inet_rsk(req); 980 981 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 982 req->cookie_ts = 0; 983 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 984 req->mss = rx_opt->mss_clamp; 985 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 986 ireq->tstamp_ok = rx_opt->tstamp_ok; 987 ireq->sack_ok = rx_opt->sack_ok; 988 ireq->snd_wscale = rx_opt->snd_wscale; 989 ireq->wscale_ok = rx_opt->wscale_ok; 990 ireq->acked = 0; 991 ireq->ecn_ok = 0; 992 ireq->rmt_port = tcp_hdr(skb)->source; 993 ireq->loc_port = tcp_hdr(skb)->dest; 994 } 995 996 extern void tcp_enter_memory_pressure(struct sock *sk); 997 998 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 999 { 1000 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1001 } 1002 1003 static inline int keepalive_time_when(const struct tcp_sock *tp) 1004 { 1005 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1006 } 1007 1008 static inline int keepalive_probes(const struct tcp_sock *tp) 1009 { 1010 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1011 } 1012 1013 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1014 { 1015 const struct inet_connection_sock *icsk = &tp->inet_conn; 1016 1017 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1018 tcp_time_stamp - tp->rcv_tstamp); 1019 } 1020 1021 static inline int tcp_fin_time(const struct sock *sk) 1022 { 1023 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1024 const int rto = inet_csk(sk)->icsk_rto; 1025 1026 if (fin_timeout < (rto << 2) - (rto >> 1)) 1027 fin_timeout = (rto << 2) - (rto >> 1); 1028 1029 return fin_timeout; 1030 } 1031 1032 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1033 int paws_win) 1034 { 1035 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1036 return 1; 1037 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1038 return 1; 1039 1040 return 0; 1041 } 1042 1043 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1044 int rst) 1045 { 1046 if (tcp_paws_check(rx_opt, 0)) 1047 return 0; 1048 1049 /* RST segments are not recommended to carry timestamp, 1050 and, if they do, it is recommended to ignore PAWS because 1051 "their cleanup function should take precedence over timestamps." 1052 Certainly, it is mistake. It is necessary to understand the reasons 1053 of this constraint to relax it: if peer reboots, clock may go 1054 out-of-sync and half-open connections will not be reset. 1055 Actually, the problem would be not existing if all 1056 the implementations followed draft about maintaining clock 1057 via reboots. Linux-2.2 DOES NOT! 1058 1059 However, we can relax time bounds for RST segments to MSL. 1060 */ 1061 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1062 return 0; 1063 return 1; 1064 } 1065 1066 #define TCP_CHECK_TIMER(sk) do { } while (0) 1067 1068 static inline void tcp_mib_init(struct net *net) 1069 { 1070 /* See RFC 2012 */ 1071 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1072 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1073 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1074 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1075 } 1076 1077 /* from STCP */ 1078 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1079 { 1080 tp->lost_skb_hint = NULL; 1081 tp->scoreboard_skb_hint = NULL; 1082 } 1083 1084 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1085 { 1086 tcp_clear_retrans_hints_partial(tp); 1087 tp->retransmit_skb_hint = NULL; 1088 } 1089 1090 /* MD5 Signature */ 1091 struct crypto_hash; 1092 1093 /* - key database */ 1094 struct tcp_md5sig_key { 1095 u8 *key; 1096 u8 keylen; 1097 }; 1098 1099 struct tcp4_md5sig_key { 1100 struct tcp_md5sig_key base; 1101 __be32 addr; 1102 }; 1103 1104 struct tcp6_md5sig_key { 1105 struct tcp_md5sig_key base; 1106 #if 0 1107 u32 scope_id; /* XXX */ 1108 #endif 1109 struct in6_addr addr; 1110 }; 1111 1112 /* - sock block */ 1113 struct tcp_md5sig_info { 1114 struct tcp4_md5sig_key *keys4; 1115 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1116 struct tcp6_md5sig_key *keys6; 1117 u32 entries6; 1118 u32 alloced6; 1119 #endif 1120 u32 entries4; 1121 u32 alloced4; 1122 }; 1123 1124 /* - pseudo header */ 1125 struct tcp4_pseudohdr { 1126 __be32 saddr; 1127 __be32 daddr; 1128 __u8 pad; 1129 __u8 protocol; 1130 __be16 len; 1131 }; 1132 1133 struct tcp6_pseudohdr { 1134 struct in6_addr saddr; 1135 struct in6_addr daddr; 1136 __be32 len; 1137 __be32 protocol; /* including padding */ 1138 }; 1139 1140 union tcp_md5sum_block { 1141 struct tcp4_pseudohdr ip4; 1142 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1143 struct tcp6_pseudohdr ip6; 1144 #endif 1145 }; 1146 1147 /* - pool: digest algorithm, hash description and scratch buffer */ 1148 struct tcp_md5sig_pool { 1149 struct hash_desc md5_desc; 1150 union tcp_md5sum_block md5_blk; 1151 }; 1152 1153 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1154 1155 /* - functions */ 1156 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1157 struct sock *sk, struct request_sock *req, 1158 struct sk_buff *skb); 1159 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1160 struct sock *addr_sk); 1161 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1162 u8 newkeylen); 1163 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1164 1165 #ifdef CONFIG_TCP_MD5SIG 1166 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1167 &(struct tcp_md5sig_key) { \ 1168 .key = (twsk)->tw_md5_key, \ 1169 .keylen = (twsk)->tw_md5_keylen, \ 1170 } : NULL) 1171 #else 1172 #define tcp_twsk_md5_key(twsk) NULL 1173 #endif 1174 1175 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1176 extern void tcp_free_md5sig_pool(void); 1177 1178 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1179 extern void tcp_put_md5sig_pool(void); 1180 1181 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1182 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1183 unsigned header_len); 1184 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1185 struct tcp_md5sig_key *key); 1186 1187 /* write queue abstraction */ 1188 static inline void tcp_write_queue_purge(struct sock *sk) 1189 { 1190 struct sk_buff *skb; 1191 1192 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1193 sk_wmem_free_skb(sk, skb); 1194 sk_mem_reclaim(sk); 1195 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1196 } 1197 1198 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1199 { 1200 return skb_peek(&sk->sk_write_queue); 1201 } 1202 1203 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1204 { 1205 return skb_peek_tail(&sk->sk_write_queue); 1206 } 1207 1208 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1209 { 1210 return skb_queue_next(&sk->sk_write_queue, skb); 1211 } 1212 1213 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1214 { 1215 return skb_queue_prev(&sk->sk_write_queue, skb); 1216 } 1217 1218 #define tcp_for_write_queue(skb, sk) \ 1219 skb_queue_walk(&(sk)->sk_write_queue, skb) 1220 1221 #define tcp_for_write_queue_from(skb, sk) \ 1222 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1223 1224 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1225 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1226 1227 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1228 { 1229 return sk->sk_send_head; 1230 } 1231 1232 static inline bool tcp_skb_is_last(const struct sock *sk, 1233 const struct sk_buff *skb) 1234 { 1235 return skb_queue_is_last(&sk->sk_write_queue, skb); 1236 } 1237 1238 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1239 { 1240 if (tcp_skb_is_last(sk, skb)) 1241 sk->sk_send_head = NULL; 1242 else 1243 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1244 } 1245 1246 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1247 { 1248 if (sk->sk_send_head == skb_unlinked) 1249 sk->sk_send_head = NULL; 1250 } 1251 1252 static inline void tcp_init_send_head(struct sock *sk) 1253 { 1254 sk->sk_send_head = NULL; 1255 } 1256 1257 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1258 { 1259 __skb_queue_tail(&sk->sk_write_queue, skb); 1260 } 1261 1262 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1263 { 1264 __tcp_add_write_queue_tail(sk, skb); 1265 1266 /* Queue it, remembering where we must start sending. */ 1267 if (sk->sk_send_head == NULL) { 1268 sk->sk_send_head = skb; 1269 1270 if (tcp_sk(sk)->highest_sack == NULL) 1271 tcp_sk(sk)->highest_sack = skb; 1272 } 1273 } 1274 1275 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1276 { 1277 __skb_queue_head(&sk->sk_write_queue, skb); 1278 } 1279 1280 /* Insert buff after skb on the write queue of sk. */ 1281 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1282 struct sk_buff *buff, 1283 struct sock *sk) 1284 { 1285 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1286 } 1287 1288 /* Insert new before skb on the write queue of sk. */ 1289 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1290 struct sk_buff *skb, 1291 struct sock *sk) 1292 { 1293 __skb_queue_before(&sk->sk_write_queue, skb, new); 1294 1295 if (sk->sk_send_head == skb) 1296 sk->sk_send_head = new; 1297 } 1298 1299 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1300 { 1301 __skb_unlink(skb, &sk->sk_write_queue); 1302 } 1303 1304 static inline int tcp_write_queue_empty(struct sock *sk) 1305 { 1306 return skb_queue_empty(&sk->sk_write_queue); 1307 } 1308 1309 static inline void tcp_push_pending_frames(struct sock *sk) 1310 { 1311 if (tcp_send_head(sk)) { 1312 struct tcp_sock *tp = tcp_sk(sk); 1313 1314 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1315 } 1316 } 1317 1318 /* Start sequence of the highest skb with SACKed bit, valid only if 1319 * sacked > 0 or when the caller has ensured validity by itself. 1320 */ 1321 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1322 { 1323 if (!tp->sacked_out) 1324 return tp->snd_una; 1325 1326 if (tp->highest_sack == NULL) 1327 return tp->snd_nxt; 1328 1329 return TCP_SKB_CB(tp->highest_sack)->seq; 1330 } 1331 1332 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1333 { 1334 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1335 tcp_write_queue_next(sk, skb); 1336 } 1337 1338 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1339 { 1340 return tcp_sk(sk)->highest_sack; 1341 } 1342 1343 static inline void tcp_highest_sack_reset(struct sock *sk) 1344 { 1345 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1346 } 1347 1348 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1349 static inline void tcp_highest_sack_combine(struct sock *sk, 1350 struct sk_buff *old, 1351 struct sk_buff *new) 1352 { 1353 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1354 tcp_sk(sk)->highest_sack = new; 1355 } 1356 1357 /* Determines whether this is a thin stream (which may suffer from 1358 * increased latency). Used to trigger latency-reducing mechanisms. 1359 */ 1360 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1361 { 1362 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1363 } 1364 1365 /* /proc */ 1366 enum tcp_seq_states { 1367 TCP_SEQ_STATE_LISTENING, 1368 TCP_SEQ_STATE_OPENREQ, 1369 TCP_SEQ_STATE_ESTABLISHED, 1370 TCP_SEQ_STATE_TIME_WAIT, 1371 }; 1372 1373 struct tcp_seq_afinfo { 1374 char *name; 1375 sa_family_t family; 1376 struct file_operations seq_fops; 1377 struct seq_operations seq_ops; 1378 }; 1379 1380 struct tcp_iter_state { 1381 struct seq_net_private p; 1382 sa_family_t family; 1383 enum tcp_seq_states state; 1384 struct sock *syn_wait_sk; 1385 int bucket, offset, sbucket, num, uid; 1386 loff_t last_pos; 1387 }; 1388 1389 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1390 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1391 1392 extern struct request_sock_ops tcp_request_sock_ops; 1393 extern struct request_sock_ops tcp6_request_sock_ops; 1394 1395 extern void tcp_v4_destroy_sock(struct sock *sk); 1396 1397 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1398 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1399 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1400 struct sk_buff *skb); 1401 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1402 struct sk_buff *skb); 1403 extern int tcp_gro_complete(struct sk_buff *skb); 1404 extern int tcp4_gro_complete(struct sk_buff *skb); 1405 1406 #ifdef CONFIG_PROC_FS 1407 extern int tcp4_proc_init(void); 1408 extern void tcp4_proc_exit(void); 1409 #endif 1410 1411 /* TCP af-specific functions */ 1412 struct tcp_sock_af_ops { 1413 #ifdef CONFIG_TCP_MD5SIG 1414 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1415 struct sock *addr_sk); 1416 int (*calc_md5_hash) (char *location, 1417 struct tcp_md5sig_key *md5, 1418 struct sock *sk, 1419 struct request_sock *req, 1420 struct sk_buff *skb); 1421 int (*md5_add) (struct sock *sk, 1422 struct sock *addr_sk, 1423 u8 *newkey, 1424 u8 len); 1425 int (*md5_parse) (struct sock *sk, 1426 char __user *optval, 1427 int optlen); 1428 #endif 1429 }; 1430 1431 struct tcp_request_sock_ops { 1432 #ifdef CONFIG_TCP_MD5SIG 1433 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1434 struct request_sock *req); 1435 int (*calc_md5_hash) (char *location, 1436 struct tcp_md5sig_key *md5, 1437 struct sock *sk, 1438 struct request_sock *req, 1439 struct sk_buff *skb); 1440 #endif 1441 }; 1442 1443 /* Using SHA1 for now, define some constants. 1444 */ 1445 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1446 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1447 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1448 1449 extern int tcp_cookie_generator(u32 *bakery); 1450 1451 /** 1452 * struct tcp_cookie_values - each socket needs extra space for the 1453 * cookies, together with (optional) space for any SYN data. 1454 * 1455 * A tcp_sock contains a pointer to the current value, and this is 1456 * cloned to the tcp_timewait_sock. 1457 * 1458 * @cookie_pair: variable data from the option exchange. 1459 * 1460 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1461 * indicates default (sysctl_tcp_cookie_size). 1462 * After cookie sent, remembers size of cookie. 1463 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1464 * 1465 * @s_data_desired: user specified tcpct_s_data_desired. When the 1466 * constant payload is specified (@s_data_constant), 1467 * holds its length instead. 1468 * Range 0 to TCP_MSS_DESIRED. 1469 * 1470 * @s_data_payload: constant data that is to be included in the 1471 * payload of SYN or SYNACK segments when the 1472 * cookie option is present. 1473 */ 1474 struct tcp_cookie_values { 1475 struct kref kref; 1476 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1477 u8 cookie_pair_size; 1478 u8 cookie_desired; 1479 u16 s_data_desired:11, 1480 s_data_constant:1, 1481 s_data_in:1, 1482 s_data_out:1, 1483 s_data_unused:2; 1484 u8 s_data_payload[0]; 1485 }; 1486 1487 static inline void tcp_cookie_values_release(struct kref *kref) 1488 { 1489 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1490 } 1491 1492 /* The length of constant payload data. Note that s_data_desired is 1493 * overloaded, depending on s_data_constant: either the length of constant 1494 * data (returned here) or the limit on variable data. 1495 */ 1496 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1497 { 1498 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1499 ? tp->cookie_values->s_data_desired 1500 : 0; 1501 } 1502 1503 /** 1504 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1505 * 1506 * As tcp_request_sock has already been extended in other places, the 1507 * only remaining method is to pass stack values along as function 1508 * parameters. These parameters are not needed after sending SYNACK. 1509 * 1510 * @cookie_bakery: cryptographic secret and message workspace. 1511 * 1512 * @cookie_plus: bytes in authenticator/cookie option, copied from 1513 * struct tcp_options_received (above). 1514 */ 1515 struct tcp_extend_values { 1516 struct request_values rv; 1517 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1518 u8 cookie_plus:6, 1519 cookie_out_never:1, 1520 cookie_in_always:1; 1521 }; 1522 1523 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1524 { 1525 return (struct tcp_extend_values *)rvp; 1526 } 1527 1528 extern void tcp_v4_init(void); 1529 extern void tcp_init(void); 1530 1531 #endif /* _TCP_H */ 1532