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 /* Offer an initial receive window of 10 mss. */ 64 #define TCP_DEFAULT_INIT_RCVWND 10 65 66 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 67 #define TCP_MIN_MSS 88U 68 69 /* The least MTU to use for probing */ 70 #define TCP_BASE_MSS 512 71 72 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 73 #define TCP_FASTRETRANS_THRESH 3 74 75 /* Maximal reordering. */ 76 #define TCP_MAX_REORDERING 127 77 78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 79 #define TCP_MAX_QUICKACKS 16U 80 81 /* urg_data states */ 82 #define TCP_URG_VALID 0x0100 83 #define TCP_URG_NOTYET 0x0200 84 #define TCP_URG_READ 0x0400 85 86 #define TCP_RETR1 3 /* 87 * This is how many retries it does before it 88 * tries to figure out if the gateway is 89 * down. Minimal RFC value is 3; it corresponds 90 * to ~3sec-8min depending on RTO. 91 */ 92 93 #define TCP_RETR2 15 /* 94 * This should take at least 95 * 90 minutes to time out. 96 * RFC1122 says that the limit is 100 sec. 97 * 15 is ~13-30min depending on RTO. 98 */ 99 100 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 101 * connection: ~180sec is RFC minimum */ 102 103 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 104 * connection: ~180sec is RFC minimum */ 105 106 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 107 * state, about 60 seconds */ 108 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 109 /* BSD style FIN_WAIT2 deadlock breaker. 110 * It used to be 3min, new value is 60sec, 111 * to combine FIN-WAIT-2 timeout with 112 * TIME-WAIT timer. 113 */ 114 115 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 116 #if HZ >= 100 117 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 118 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 119 #else 120 #define TCP_DELACK_MIN 4U 121 #define TCP_ATO_MIN 4U 122 #endif 123 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 124 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 125 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 126 127 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 128 * for local resources. 129 */ 130 131 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 132 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 133 #define TCP_KEEPALIVE_INTVL (75*HZ) 134 135 #define MAX_TCP_KEEPIDLE 32767 136 #define MAX_TCP_KEEPINTVL 32767 137 #define MAX_TCP_KEEPCNT 127 138 #define MAX_TCP_SYNCNT 127 139 140 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 141 142 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 143 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 144 * after this time. It should be equal 145 * (or greater than) TCP_TIMEWAIT_LEN 146 * to provide reliability equal to one 147 * provided by timewait state. 148 */ 149 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 150 * timestamps. It must be less than 151 * minimal timewait lifetime. 152 */ 153 /* 154 * TCP option 155 */ 156 157 #define TCPOPT_NOP 1 /* Padding */ 158 #define TCPOPT_EOL 0 /* End of options */ 159 #define TCPOPT_MSS 2 /* Segment size negotiating */ 160 #define TCPOPT_WINDOW 3 /* Window scaling */ 161 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 162 #define TCPOPT_SACK 5 /* SACK Block */ 163 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 164 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 165 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 166 167 /* 168 * TCP option lengths 169 */ 170 171 #define TCPOLEN_MSS 4 172 #define TCPOLEN_WINDOW 3 173 #define TCPOLEN_SACK_PERM 2 174 #define TCPOLEN_TIMESTAMP 10 175 #define TCPOLEN_MD5SIG 18 176 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 177 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 178 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 179 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 180 181 /* But this is what stacks really send out. */ 182 #define TCPOLEN_TSTAMP_ALIGNED 12 183 #define TCPOLEN_WSCALE_ALIGNED 4 184 #define TCPOLEN_SACKPERM_ALIGNED 4 185 #define TCPOLEN_SACK_BASE 2 186 #define TCPOLEN_SACK_BASE_ALIGNED 4 187 #define TCPOLEN_SACK_PERBLOCK 8 188 #define TCPOLEN_MD5SIG_ALIGNED 20 189 #define TCPOLEN_MSS_ALIGNED 4 190 191 /* Flags in tp->nonagle */ 192 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 193 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 194 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 195 196 /* TCP thin-stream limits */ 197 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 198 199 extern struct inet_timewait_death_row tcp_death_row; 200 201 /* sysctl variables for tcp */ 202 extern int sysctl_tcp_timestamps; 203 extern int sysctl_tcp_window_scaling; 204 extern int sysctl_tcp_sack; 205 extern int sysctl_tcp_fin_timeout; 206 extern int sysctl_tcp_keepalive_time; 207 extern int sysctl_tcp_keepalive_probes; 208 extern int sysctl_tcp_keepalive_intvl; 209 extern int sysctl_tcp_syn_retries; 210 extern int sysctl_tcp_synack_retries; 211 extern int sysctl_tcp_retries1; 212 extern int sysctl_tcp_retries2; 213 extern int sysctl_tcp_orphan_retries; 214 extern int sysctl_tcp_syncookies; 215 extern int sysctl_tcp_retrans_collapse; 216 extern int sysctl_tcp_stdurg; 217 extern int sysctl_tcp_rfc1337; 218 extern int sysctl_tcp_abort_on_overflow; 219 extern int sysctl_tcp_max_orphans; 220 extern int sysctl_tcp_fack; 221 extern int sysctl_tcp_reordering; 222 extern int sysctl_tcp_ecn; 223 extern int sysctl_tcp_dsack; 224 extern long sysctl_tcp_mem[3]; 225 extern int sysctl_tcp_wmem[3]; 226 extern int sysctl_tcp_rmem[3]; 227 extern int sysctl_tcp_app_win; 228 extern int sysctl_tcp_adv_win_scale; 229 extern int sysctl_tcp_tw_reuse; 230 extern int sysctl_tcp_frto; 231 extern int sysctl_tcp_frto_response; 232 extern int sysctl_tcp_low_latency; 233 extern int sysctl_tcp_dma_copybreak; 234 extern int sysctl_tcp_nometrics_save; 235 extern int sysctl_tcp_moderate_rcvbuf; 236 extern int sysctl_tcp_tso_win_divisor; 237 extern int sysctl_tcp_abc; 238 extern int sysctl_tcp_mtu_probing; 239 extern int sysctl_tcp_base_mss; 240 extern int sysctl_tcp_workaround_signed_windows; 241 extern int sysctl_tcp_slow_start_after_idle; 242 extern int sysctl_tcp_max_ssthresh; 243 extern int sysctl_tcp_cookie_size; 244 extern int sysctl_tcp_thin_linear_timeouts; 245 extern int sysctl_tcp_thin_dupack; 246 247 extern atomic_long_t tcp_memory_allocated; 248 extern struct percpu_counter tcp_sockets_allocated; 249 extern int tcp_memory_pressure; 250 251 /* 252 * The next routines deal with comparing 32 bit unsigned ints 253 * and worry about wraparound (automatic with unsigned arithmetic). 254 */ 255 256 static inline int before(__u32 seq1, __u32 seq2) 257 { 258 return (__s32)(seq1-seq2) < 0; 259 } 260 #define after(seq2, seq1) before(seq1, seq2) 261 262 /* is s2<=s1<=s3 ? */ 263 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 264 { 265 return seq3 - seq2 >= seq1 - seq2; 266 } 267 268 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 269 { 270 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 271 int orphans = percpu_counter_read_positive(ocp); 272 273 if (orphans << shift > sysctl_tcp_max_orphans) { 274 orphans = percpu_counter_sum_positive(ocp); 275 if (orphans << shift > sysctl_tcp_max_orphans) 276 return true; 277 } 278 279 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 280 atomic_long_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]) 281 return true; 282 return false; 283 } 284 285 /* syncookies: remember time of last synqueue overflow */ 286 static inline void tcp_synq_overflow(struct sock *sk) 287 { 288 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 289 } 290 291 /* syncookies: no recent synqueue overflow on this listening socket? */ 292 static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 293 { 294 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 295 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT); 296 } 297 298 extern struct proto tcp_prot; 299 300 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 301 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 302 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 303 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 304 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 305 306 extern void tcp_v4_err(struct sk_buff *skb, u32); 307 308 extern void tcp_shutdown (struct sock *sk, int how); 309 310 extern int tcp_v4_rcv(struct sk_buff *skb); 311 312 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it); 313 extern void *tcp_v4_tw_get_peer(struct sock *sk); 314 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 315 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 316 size_t size); 317 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 318 size_t size, int flags); 319 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 320 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 321 struct tcphdr *th, unsigned len); 322 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 323 struct tcphdr *th, unsigned len); 324 extern void tcp_rcv_space_adjust(struct sock *sk); 325 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 326 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 327 extern void tcp_twsk_destructor(struct sock *sk); 328 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 329 struct pipe_inode_info *pipe, size_t len, 330 unsigned int flags); 331 332 static inline void tcp_dec_quickack_mode(struct sock *sk, 333 const unsigned int pkts) 334 { 335 struct inet_connection_sock *icsk = inet_csk(sk); 336 337 if (icsk->icsk_ack.quick) { 338 if (pkts >= icsk->icsk_ack.quick) { 339 icsk->icsk_ack.quick = 0; 340 /* Leaving quickack mode we deflate ATO. */ 341 icsk->icsk_ack.ato = TCP_ATO_MIN; 342 } else 343 icsk->icsk_ack.quick -= pkts; 344 } 345 } 346 347 #define TCP_ECN_OK 1 348 #define TCP_ECN_QUEUE_CWR 2 349 #define TCP_ECN_DEMAND_CWR 4 350 351 static __inline__ void 352 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 353 { 354 if (sysctl_tcp_ecn && th->ece && th->cwr) 355 inet_rsk(req)->ecn_ok = 1; 356 } 357 358 enum tcp_tw_status { 359 TCP_TW_SUCCESS = 0, 360 TCP_TW_RST = 1, 361 TCP_TW_ACK = 2, 362 TCP_TW_SYN = 3 363 }; 364 365 366 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 367 struct sk_buff *skb, 368 const struct tcphdr *th); 369 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 370 struct request_sock *req, 371 struct request_sock **prev); 372 extern int tcp_child_process(struct sock *parent, struct sock *child, 373 struct sk_buff *skb); 374 extern int tcp_use_frto(struct sock *sk); 375 extern void tcp_enter_frto(struct sock *sk); 376 extern void tcp_enter_loss(struct sock *sk, int how); 377 extern void tcp_clear_retrans(struct tcp_sock *tp); 378 extern void tcp_update_metrics(struct sock *sk); 379 extern void tcp_close(struct sock *sk, long timeout); 380 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 381 struct poll_table_struct *wait); 382 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 383 char __user *optval, int __user *optlen); 384 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 385 char __user *optval, unsigned int optlen); 386 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 387 char __user *optval, int __user *optlen); 388 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 389 char __user *optval, unsigned int optlen); 390 extern void tcp_set_keepalive(struct sock *sk, int val); 391 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 392 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 393 size_t len, int nonblock, int flags, int *addr_len); 394 extern void tcp_parse_options(struct sk_buff *skb, 395 struct tcp_options_received *opt_rx, u8 **hvpp, 396 int estab); 397 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 398 399 /* 400 * TCP v4 functions exported for the inet6 API 401 */ 402 403 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 404 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 405 extern struct sock * tcp_create_openreq_child(struct sock *sk, 406 struct request_sock *req, 407 struct sk_buff *skb); 408 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 409 struct request_sock *req, 410 struct dst_entry *dst); 411 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 412 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 413 int addr_len); 414 extern int tcp_connect(struct sock *sk); 415 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 416 struct request_sock *req, 417 struct request_values *rvp); 418 extern int tcp_disconnect(struct sock *sk, int flags); 419 420 421 /* From syncookies.c */ 422 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 423 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 424 struct ip_options *opt); 425 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 426 __u16 *mss); 427 428 extern __u32 cookie_init_timestamp(struct request_sock *req); 429 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 430 431 /* From net/ipv6/syncookies.c */ 432 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 433 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 434 __u16 *mss); 435 436 /* tcp_output.c */ 437 438 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 439 int nonagle); 440 extern int tcp_may_send_now(struct sock *sk); 441 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 442 extern void tcp_retransmit_timer(struct sock *sk); 443 extern void tcp_xmit_retransmit_queue(struct sock *); 444 extern void tcp_simple_retransmit(struct sock *); 445 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 446 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 447 448 extern void tcp_send_probe0(struct sock *); 449 extern void tcp_send_partial(struct sock *); 450 extern int tcp_write_wakeup(struct sock *); 451 extern void tcp_send_fin(struct sock *sk); 452 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 453 extern int tcp_send_synack(struct sock *); 454 extern void tcp_push_one(struct sock *, unsigned int mss_now); 455 extern void tcp_send_ack(struct sock *sk); 456 extern void tcp_send_delayed_ack(struct sock *sk); 457 458 /* tcp_input.c */ 459 extern void tcp_cwnd_application_limited(struct sock *sk); 460 461 /* tcp_timer.c */ 462 extern void tcp_init_xmit_timers(struct sock *); 463 static inline void tcp_clear_xmit_timers(struct sock *sk) 464 { 465 inet_csk_clear_xmit_timers(sk); 466 } 467 468 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 469 extern unsigned int tcp_current_mss(struct sock *sk); 470 471 /* Bound MSS / TSO packet size with the half of the window */ 472 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 473 { 474 int cutoff; 475 476 /* When peer uses tiny windows, there is no use in packetizing 477 * to sub-MSS pieces for the sake of SWS or making sure there 478 * are enough packets in the pipe for fast recovery. 479 * 480 * On the other hand, for extremely large MSS devices, handling 481 * smaller than MSS windows in this way does make sense. 482 */ 483 if (tp->max_window >= 512) 484 cutoff = (tp->max_window >> 1); 485 else 486 cutoff = tp->max_window; 487 488 if (cutoff && pktsize > cutoff) 489 return max_t(int, cutoff, 68U - tp->tcp_header_len); 490 else 491 return pktsize; 492 } 493 494 /* tcp.c */ 495 extern void tcp_get_info(struct sock *, struct tcp_info *); 496 497 /* Read 'sendfile()'-style from a TCP socket */ 498 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 499 unsigned int, size_t); 500 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 501 sk_read_actor_t recv_actor); 502 503 extern void tcp_initialize_rcv_mss(struct sock *sk); 504 505 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 506 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 507 extern void tcp_mtup_init(struct sock *sk); 508 509 static inline void tcp_bound_rto(const struct sock *sk) 510 { 511 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 512 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 513 } 514 515 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 516 { 517 return (tp->srtt >> 3) + tp->rttvar; 518 } 519 520 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 521 { 522 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 523 ntohl(TCP_FLAG_ACK) | 524 snd_wnd); 525 } 526 527 static inline void tcp_fast_path_on(struct tcp_sock *tp) 528 { 529 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 530 } 531 532 static inline void tcp_fast_path_check(struct sock *sk) 533 { 534 struct tcp_sock *tp = tcp_sk(sk); 535 536 if (skb_queue_empty(&tp->out_of_order_queue) && 537 tp->rcv_wnd && 538 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 539 !tp->urg_data) 540 tcp_fast_path_on(tp); 541 } 542 543 /* Compute the actual rto_min value */ 544 static inline u32 tcp_rto_min(struct sock *sk) 545 { 546 struct dst_entry *dst = __sk_dst_get(sk); 547 u32 rto_min = TCP_RTO_MIN; 548 549 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 550 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 551 return rto_min; 552 } 553 554 /* Compute the actual receive window we are currently advertising. 555 * Rcv_nxt can be after the window if our peer push more data 556 * than the offered window. 557 */ 558 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 559 { 560 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 561 562 if (win < 0) 563 win = 0; 564 return (u32) win; 565 } 566 567 /* Choose a new window, without checks for shrinking, and without 568 * scaling applied to the result. The caller does these things 569 * if necessary. This is a "raw" window selection. 570 */ 571 extern u32 __tcp_select_window(struct sock *sk); 572 573 /* TCP timestamps are only 32-bits, this causes a slight 574 * complication on 64-bit systems since we store a snapshot 575 * of jiffies in the buffer control blocks below. We decided 576 * to use only the low 32-bits of jiffies and hide the ugly 577 * casts with the following macro. 578 */ 579 #define tcp_time_stamp ((__u32)(jiffies)) 580 581 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 582 583 #define TCPHDR_FIN 0x01 584 #define TCPHDR_SYN 0x02 585 #define TCPHDR_RST 0x04 586 #define TCPHDR_PSH 0x08 587 #define TCPHDR_ACK 0x10 588 #define TCPHDR_URG 0x20 589 #define TCPHDR_ECE 0x40 590 #define TCPHDR_CWR 0x80 591 592 /* This is what the send packet queuing engine uses to pass 593 * TCP per-packet control information to the transmission code. 594 * We also store the host-order sequence numbers in here too. 595 * This is 44 bytes if IPV6 is enabled. 596 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 597 */ 598 struct tcp_skb_cb { 599 union { 600 struct inet_skb_parm h4; 601 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 602 struct inet6_skb_parm h6; 603 #endif 604 } header; /* For incoming frames */ 605 __u32 seq; /* Starting sequence number */ 606 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 607 __u32 when; /* used to compute rtt's */ 608 __u8 flags; /* TCP header flags. */ 609 __u8 sacked; /* State flags for SACK/FACK. */ 610 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 611 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 612 #define TCPCB_LOST 0x04 /* SKB is lost */ 613 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 614 615 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 616 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 617 618 __u32 ack_seq; /* Sequence number ACK'd */ 619 }; 620 621 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 622 623 /* Due to TSO, an SKB can be composed of multiple actual 624 * packets. To keep these tracked properly, we use this. 625 */ 626 static inline int tcp_skb_pcount(const struct sk_buff *skb) 627 { 628 return skb_shinfo(skb)->gso_segs; 629 } 630 631 /* This is valid iff tcp_skb_pcount() > 1. */ 632 static inline int tcp_skb_mss(const struct sk_buff *skb) 633 { 634 return skb_shinfo(skb)->gso_size; 635 } 636 637 /* Events passed to congestion control interface */ 638 enum tcp_ca_event { 639 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 640 CA_EVENT_CWND_RESTART, /* congestion window restart */ 641 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 642 CA_EVENT_FRTO, /* fast recovery timeout */ 643 CA_EVENT_LOSS, /* loss timeout */ 644 CA_EVENT_FAST_ACK, /* in sequence ack */ 645 CA_EVENT_SLOW_ACK, /* other ack */ 646 }; 647 648 /* 649 * Interface for adding new TCP congestion control handlers 650 */ 651 #define TCP_CA_NAME_MAX 16 652 #define TCP_CA_MAX 128 653 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 654 655 #define TCP_CONG_NON_RESTRICTED 0x1 656 #define TCP_CONG_RTT_STAMP 0x2 657 658 struct tcp_congestion_ops { 659 struct list_head list; 660 unsigned long flags; 661 662 /* initialize private data (optional) */ 663 void (*init)(struct sock *sk); 664 /* cleanup private data (optional) */ 665 void (*release)(struct sock *sk); 666 667 /* return slow start threshold (required) */ 668 u32 (*ssthresh)(struct sock *sk); 669 /* lower bound for congestion window (optional) */ 670 u32 (*min_cwnd)(const struct sock *sk); 671 /* do new cwnd calculation (required) */ 672 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 673 /* call before changing ca_state (optional) */ 674 void (*set_state)(struct sock *sk, u8 new_state); 675 /* call when cwnd event occurs (optional) */ 676 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 677 /* new value of cwnd after loss (optional) */ 678 u32 (*undo_cwnd)(struct sock *sk); 679 /* hook for packet ack accounting (optional) */ 680 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 681 /* get info for inet_diag (optional) */ 682 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 683 684 char name[TCP_CA_NAME_MAX]; 685 struct module *owner; 686 }; 687 688 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 689 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 690 691 extern void tcp_init_congestion_control(struct sock *sk); 692 extern void tcp_cleanup_congestion_control(struct sock *sk); 693 extern int tcp_set_default_congestion_control(const char *name); 694 extern void tcp_get_default_congestion_control(char *name); 695 extern void tcp_get_available_congestion_control(char *buf, size_t len); 696 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 697 extern int tcp_set_allowed_congestion_control(char *allowed); 698 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 699 extern void tcp_slow_start(struct tcp_sock *tp); 700 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 701 702 extern struct tcp_congestion_ops tcp_init_congestion_ops; 703 extern u32 tcp_reno_ssthresh(struct sock *sk); 704 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 705 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 706 extern struct tcp_congestion_ops tcp_reno; 707 708 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 709 { 710 struct inet_connection_sock *icsk = inet_csk(sk); 711 712 if (icsk->icsk_ca_ops->set_state) 713 icsk->icsk_ca_ops->set_state(sk, ca_state); 714 icsk->icsk_ca_state = ca_state; 715 } 716 717 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 718 { 719 const struct inet_connection_sock *icsk = inet_csk(sk); 720 721 if (icsk->icsk_ca_ops->cwnd_event) 722 icsk->icsk_ca_ops->cwnd_event(sk, event); 723 } 724 725 /* These functions determine how the current flow behaves in respect of SACK 726 * handling. SACK is negotiated with the peer, and therefore it can vary 727 * between different flows. 728 * 729 * tcp_is_sack - SACK enabled 730 * tcp_is_reno - No SACK 731 * tcp_is_fack - FACK enabled, implies SACK enabled 732 */ 733 static inline int tcp_is_sack(const struct tcp_sock *tp) 734 { 735 return tp->rx_opt.sack_ok; 736 } 737 738 static inline int tcp_is_reno(const struct tcp_sock *tp) 739 { 740 return !tcp_is_sack(tp); 741 } 742 743 static inline int tcp_is_fack(const struct tcp_sock *tp) 744 { 745 return tp->rx_opt.sack_ok & 2; 746 } 747 748 static inline void tcp_enable_fack(struct tcp_sock *tp) 749 { 750 tp->rx_opt.sack_ok |= 2; 751 } 752 753 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 754 { 755 return tp->sacked_out + tp->lost_out; 756 } 757 758 /* This determines how many packets are "in the network" to the best 759 * of our knowledge. In many cases it is conservative, but where 760 * detailed information is available from the receiver (via SACK 761 * blocks etc.) we can make more aggressive calculations. 762 * 763 * Use this for decisions involving congestion control, use just 764 * tp->packets_out to determine if the send queue is empty or not. 765 * 766 * Read this equation as: 767 * 768 * "Packets sent once on transmission queue" MINUS 769 * "Packets left network, but not honestly ACKed yet" PLUS 770 * "Packets fast retransmitted" 771 */ 772 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 773 { 774 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 775 } 776 777 #define TCP_INFINITE_SSTHRESH 0x7fffffff 778 779 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 780 { 781 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 782 } 783 784 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 785 * The exception is rate halving phase, when cwnd is decreasing towards 786 * ssthresh. 787 */ 788 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 789 { 790 const struct tcp_sock *tp = tcp_sk(sk); 791 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 792 return tp->snd_ssthresh; 793 else 794 return max(tp->snd_ssthresh, 795 ((tp->snd_cwnd >> 1) + 796 (tp->snd_cwnd >> 2))); 797 } 798 799 /* Use define here intentionally to get WARN_ON location shown at the caller */ 800 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 801 802 /* 803 * Convert RFC 3390 larger initial window into an equivalent number of packets. 804 * This is based on the numbers specified in RFC 5681, 3.1. 805 */ 806 static inline u32 rfc3390_bytes_to_packets(const u32 smss) 807 { 808 return smss <= 1095 ? 4 : (smss > 2190 ? 2 : 3); 809 } 810 811 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 812 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 813 814 /* Slow start with delack produces 3 packets of burst, so that 815 * it is safe "de facto". This will be the default - same as 816 * the default reordering threshold - but if reordering increases, 817 * we must be able to allow cwnd to burst at least this much in order 818 * to not pull it back when holes are filled. 819 */ 820 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 821 { 822 return tp->reordering; 823 } 824 825 /* Returns end sequence number of the receiver's advertised window */ 826 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 827 { 828 return tp->snd_una + tp->snd_wnd; 829 } 830 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 831 832 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 833 const struct sk_buff *skb) 834 { 835 if (skb->len < mss) 836 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 837 } 838 839 static inline void tcp_check_probe_timer(struct sock *sk) 840 { 841 struct tcp_sock *tp = tcp_sk(sk); 842 const struct inet_connection_sock *icsk = inet_csk(sk); 843 844 if (!tp->packets_out && !icsk->icsk_pending) 845 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 846 icsk->icsk_rto, TCP_RTO_MAX); 847 } 848 849 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 850 { 851 tp->snd_wl1 = seq; 852 } 853 854 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 855 { 856 tp->snd_wl1 = seq; 857 } 858 859 /* 860 * Calculate(/check) TCP checksum 861 */ 862 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 863 __be32 daddr, __wsum base) 864 { 865 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 866 } 867 868 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 869 { 870 return __skb_checksum_complete(skb); 871 } 872 873 static inline int tcp_checksum_complete(struct sk_buff *skb) 874 { 875 return !skb_csum_unnecessary(skb) && 876 __tcp_checksum_complete(skb); 877 } 878 879 /* Prequeue for VJ style copy to user, combined with checksumming. */ 880 881 static inline void tcp_prequeue_init(struct tcp_sock *tp) 882 { 883 tp->ucopy.task = NULL; 884 tp->ucopy.len = 0; 885 tp->ucopy.memory = 0; 886 skb_queue_head_init(&tp->ucopy.prequeue); 887 #ifdef CONFIG_NET_DMA 888 tp->ucopy.dma_chan = NULL; 889 tp->ucopy.wakeup = 0; 890 tp->ucopy.pinned_list = NULL; 891 tp->ucopy.dma_cookie = 0; 892 #endif 893 } 894 895 /* Packet is added to VJ-style prequeue for processing in process 896 * context, if a reader task is waiting. Apparently, this exciting 897 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 898 * failed somewhere. Latency? Burstiness? Well, at least now we will 899 * see, why it failed. 8)8) --ANK 900 * 901 * NOTE: is this not too big to inline? 902 */ 903 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 904 { 905 struct tcp_sock *tp = tcp_sk(sk); 906 907 if (sysctl_tcp_low_latency || !tp->ucopy.task) 908 return 0; 909 910 __skb_queue_tail(&tp->ucopy.prequeue, skb); 911 tp->ucopy.memory += skb->truesize; 912 if (tp->ucopy.memory > sk->sk_rcvbuf) { 913 struct sk_buff *skb1; 914 915 BUG_ON(sock_owned_by_user(sk)); 916 917 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 918 sk_backlog_rcv(sk, skb1); 919 NET_INC_STATS_BH(sock_net(sk), 920 LINUX_MIB_TCPPREQUEUEDROPPED); 921 } 922 923 tp->ucopy.memory = 0; 924 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 925 wake_up_interruptible_sync_poll(sk_sleep(sk), 926 POLLIN | POLLRDNORM | POLLRDBAND); 927 if (!inet_csk_ack_scheduled(sk)) 928 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 929 (3 * tcp_rto_min(sk)) / 4, 930 TCP_RTO_MAX); 931 } 932 return 1; 933 } 934 935 936 #undef STATE_TRACE 937 938 #ifdef STATE_TRACE 939 static const char *statename[]={ 940 "Unused","Established","Syn Sent","Syn Recv", 941 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 942 "Close Wait","Last ACK","Listen","Closing" 943 }; 944 #endif 945 extern void tcp_set_state(struct sock *sk, int state); 946 947 extern void tcp_done(struct sock *sk); 948 949 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 950 { 951 rx_opt->dsack = 0; 952 rx_opt->num_sacks = 0; 953 } 954 955 /* Determine a window scaling and initial window to offer. */ 956 extern void tcp_select_initial_window(int __space, __u32 mss, 957 __u32 *rcv_wnd, __u32 *window_clamp, 958 int wscale_ok, __u8 *rcv_wscale, 959 __u32 init_rcv_wnd); 960 961 static inline int tcp_win_from_space(int space) 962 { 963 return sysctl_tcp_adv_win_scale<=0 ? 964 (space>>(-sysctl_tcp_adv_win_scale)) : 965 space - (space>>sysctl_tcp_adv_win_scale); 966 } 967 968 /* Note: caller must be prepared to deal with negative returns */ 969 static inline int tcp_space(const struct sock *sk) 970 { 971 return tcp_win_from_space(sk->sk_rcvbuf - 972 atomic_read(&sk->sk_rmem_alloc)); 973 } 974 975 static inline int tcp_full_space(const struct sock *sk) 976 { 977 return tcp_win_from_space(sk->sk_rcvbuf); 978 } 979 980 static inline void tcp_openreq_init(struct request_sock *req, 981 struct tcp_options_received *rx_opt, 982 struct sk_buff *skb) 983 { 984 struct inet_request_sock *ireq = inet_rsk(req); 985 986 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 987 req->cookie_ts = 0; 988 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 989 req->mss = rx_opt->mss_clamp; 990 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 991 ireq->tstamp_ok = rx_opt->tstamp_ok; 992 ireq->sack_ok = rx_opt->sack_ok; 993 ireq->snd_wscale = rx_opt->snd_wscale; 994 ireq->wscale_ok = rx_opt->wscale_ok; 995 ireq->acked = 0; 996 ireq->ecn_ok = 0; 997 ireq->rmt_port = tcp_hdr(skb)->source; 998 ireq->loc_port = tcp_hdr(skb)->dest; 999 } 1000 1001 extern void tcp_enter_memory_pressure(struct sock *sk); 1002 1003 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1004 { 1005 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1006 } 1007 1008 static inline int keepalive_time_when(const struct tcp_sock *tp) 1009 { 1010 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1011 } 1012 1013 static inline int keepalive_probes(const struct tcp_sock *tp) 1014 { 1015 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1016 } 1017 1018 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1019 { 1020 const struct inet_connection_sock *icsk = &tp->inet_conn; 1021 1022 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1023 tcp_time_stamp - tp->rcv_tstamp); 1024 } 1025 1026 static inline int tcp_fin_time(const struct sock *sk) 1027 { 1028 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1029 const int rto = inet_csk(sk)->icsk_rto; 1030 1031 if (fin_timeout < (rto << 2) - (rto >> 1)) 1032 fin_timeout = (rto << 2) - (rto >> 1); 1033 1034 return fin_timeout; 1035 } 1036 1037 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1038 int paws_win) 1039 { 1040 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1041 return 1; 1042 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1043 return 1; 1044 /* 1045 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1046 * then following tcp messages have valid values. Ignore 0 value, 1047 * or else 'negative' tsval might forbid us to accept their packets. 1048 */ 1049 if (!rx_opt->ts_recent) 1050 return 1; 1051 return 0; 1052 } 1053 1054 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1055 int rst) 1056 { 1057 if (tcp_paws_check(rx_opt, 0)) 1058 return 0; 1059 1060 /* RST segments are not recommended to carry timestamp, 1061 and, if they do, it is recommended to ignore PAWS because 1062 "their cleanup function should take precedence over timestamps." 1063 Certainly, it is mistake. It is necessary to understand the reasons 1064 of this constraint to relax it: if peer reboots, clock may go 1065 out-of-sync and half-open connections will not be reset. 1066 Actually, the problem would be not existing if all 1067 the implementations followed draft about maintaining clock 1068 via reboots. Linux-2.2 DOES NOT! 1069 1070 However, we can relax time bounds for RST segments to MSL. 1071 */ 1072 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1073 return 0; 1074 return 1; 1075 } 1076 1077 #define TCP_CHECK_TIMER(sk) do { } while (0) 1078 1079 static inline void tcp_mib_init(struct net *net) 1080 { 1081 /* See RFC 2012 */ 1082 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1083 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1084 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1085 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1086 } 1087 1088 /* from STCP */ 1089 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1090 { 1091 tp->lost_skb_hint = NULL; 1092 tp->scoreboard_skb_hint = NULL; 1093 } 1094 1095 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1096 { 1097 tcp_clear_retrans_hints_partial(tp); 1098 tp->retransmit_skb_hint = NULL; 1099 } 1100 1101 /* MD5 Signature */ 1102 struct crypto_hash; 1103 1104 /* - key database */ 1105 struct tcp_md5sig_key { 1106 u8 *key; 1107 u8 keylen; 1108 }; 1109 1110 struct tcp4_md5sig_key { 1111 struct tcp_md5sig_key base; 1112 __be32 addr; 1113 }; 1114 1115 struct tcp6_md5sig_key { 1116 struct tcp_md5sig_key base; 1117 #if 0 1118 u32 scope_id; /* XXX */ 1119 #endif 1120 struct in6_addr addr; 1121 }; 1122 1123 /* - sock block */ 1124 struct tcp_md5sig_info { 1125 struct tcp4_md5sig_key *keys4; 1126 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1127 struct tcp6_md5sig_key *keys6; 1128 u32 entries6; 1129 u32 alloced6; 1130 #endif 1131 u32 entries4; 1132 u32 alloced4; 1133 }; 1134 1135 /* - pseudo header */ 1136 struct tcp4_pseudohdr { 1137 __be32 saddr; 1138 __be32 daddr; 1139 __u8 pad; 1140 __u8 protocol; 1141 __be16 len; 1142 }; 1143 1144 struct tcp6_pseudohdr { 1145 struct in6_addr saddr; 1146 struct in6_addr daddr; 1147 __be32 len; 1148 __be32 protocol; /* including padding */ 1149 }; 1150 1151 union tcp_md5sum_block { 1152 struct tcp4_pseudohdr ip4; 1153 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1154 struct tcp6_pseudohdr ip6; 1155 #endif 1156 }; 1157 1158 /* - pool: digest algorithm, hash description and scratch buffer */ 1159 struct tcp_md5sig_pool { 1160 struct hash_desc md5_desc; 1161 union tcp_md5sum_block md5_blk; 1162 }; 1163 1164 /* - functions */ 1165 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1166 struct sock *sk, struct request_sock *req, 1167 struct sk_buff *skb); 1168 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1169 struct sock *addr_sk); 1170 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1171 u8 newkeylen); 1172 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1173 1174 #ifdef CONFIG_TCP_MD5SIG 1175 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1176 &(struct tcp_md5sig_key) { \ 1177 .key = (twsk)->tw_md5_key, \ 1178 .keylen = (twsk)->tw_md5_keylen, \ 1179 } : NULL) 1180 #else 1181 #define tcp_twsk_md5_key(twsk) NULL 1182 #endif 1183 1184 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1185 extern void tcp_free_md5sig_pool(void); 1186 1187 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1188 extern void tcp_put_md5sig_pool(void); 1189 1190 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1191 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1192 unsigned header_len); 1193 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1194 struct tcp_md5sig_key *key); 1195 1196 /* write queue abstraction */ 1197 static inline void tcp_write_queue_purge(struct sock *sk) 1198 { 1199 struct sk_buff *skb; 1200 1201 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1202 sk_wmem_free_skb(sk, skb); 1203 sk_mem_reclaim(sk); 1204 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1205 } 1206 1207 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1208 { 1209 return skb_peek(&sk->sk_write_queue); 1210 } 1211 1212 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1213 { 1214 return skb_peek_tail(&sk->sk_write_queue); 1215 } 1216 1217 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1218 { 1219 return skb_queue_next(&sk->sk_write_queue, skb); 1220 } 1221 1222 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1223 { 1224 return skb_queue_prev(&sk->sk_write_queue, skb); 1225 } 1226 1227 #define tcp_for_write_queue(skb, sk) \ 1228 skb_queue_walk(&(sk)->sk_write_queue, skb) 1229 1230 #define tcp_for_write_queue_from(skb, sk) \ 1231 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1232 1233 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1234 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1235 1236 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1237 { 1238 return sk->sk_send_head; 1239 } 1240 1241 static inline bool tcp_skb_is_last(const struct sock *sk, 1242 const struct sk_buff *skb) 1243 { 1244 return skb_queue_is_last(&sk->sk_write_queue, skb); 1245 } 1246 1247 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1248 { 1249 if (tcp_skb_is_last(sk, skb)) 1250 sk->sk_send_head = NULL; 1251 else 1252 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1253 } 1254 1255 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1256 { 1257 if (sk->sk_send_head == skb_unlinked) 1258 sk->sk_send_head = NULL; 1259 } 1260 1261 static inline void tcp_init_send_head(struct sock *sk) 1262 { 1263 sk->sk_send_head = NULL; 1264 } 1265 1266 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1267 { 1268 __skb_queue_tail(&sk->sk_write_queue, skb); 1269 } 1270 1271 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1272 { 1273 __tcp_add_write_queue_tail(sk, skb); 1274 1275 /* Queue it, remembering where we must start sending. */ 1276 if (sk->sk_send_head == NULL) { 1277 sk->sk_send_head = skb; 1278 1279 if (tcp_sk(sk)->highest_sack == NULL) 1280 tcp_sk(sk)->highest_sack = skb; 1281 } 1282 } 1283 1284 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1285 { 1286 __skb_queue_head(&sk->sk_write_queue, skb); 1287 } 1288 1289 /* Insert buff after skb on the write queue of sk. */ 1290 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1291 struct sk_buff *buff, 1292 struct sock *sk) 1293 { 1294 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1295 } 1296 1297 /* Insert new before skb on the write queue of sk. */ 1298 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1299 struct sk_buff *skb, 1300 struct sock *sk) 1301 { 1302 __skb_queue_before(&sk->sk_write_queue, skb, new); 1303 1304 if (sk->sk_send_head == skb) 1305 sk->sk_send_head = new; 1306 } 1307 1308 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1309 { 1310 __skb_unlink(skb, &sk->sk_write_queue); 1311 } 1312 1313 static inline int tcp_write_queue_empty(struct sock *sk) 1314 { 1315 return skb_queue_empty(&sk->sk_write_queue); 1316 } 1317 1318 static inline void tcp_push_pending_frames(struct sock *sk) 1319 { 1320 if (tcp_send_head(sk)) { 1321 struct tcp_sock *tp = tcp_sk(sk); 1322 1323 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1324 } 1325 } 1326 1327 /* Start sequence of the highest skb with SACKed bit, valid only if 1328 * sacked > 0 or when the caller has ensured validity by itself. 1329 */ 1330 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1331 { 1332 if (!tp->sacked_out) 1333 return tp->snd_una; 1334 1335 if (tp->highest_sack == NULL) 1336 return tp->snd_nxt; 1337 1338 return TCP_SKB_CB(tp->highest_sack)->seq; 1339 } 1340 1341 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1342 { 1343 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1344 tcp_write_queue_next(sk, skb); 1345 } 1346 1347 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1348 { 1349 return tcp_sk(sk)->highest_sack; 1350 } 1351 1352 static inline void tcp_highest_sack_reset(struct sock *sk) 1353 { 1354 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1355 } 1356 1357 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1358 static inline void tcp_highest_sack_combine(struct sock *sk, 1359 struct sk_buff *old, 1360 struct sk_buff *new) 1361 { 1362 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1363 tcp_sk(sk)->highest_sack = new; 1364 } 1365 1366 /* Determines whether this is a thin stream (which may suffer from 1367 * increased latency). Used to trigger latency-reducing mechanisms. 1368 */ 1369 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1370 { 1371 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1372 } 1373 1374 /* /proc */ 1375 enum tcp_seq_states { 1376 TCP_SEQ_STATE_LISTENING, 1377 TCP_SEQ_STATE_OPENREQ, 1378 TCP_SEQ_STATE_ESTABLISHED, 1379 TCP_SEQ_STATE_TIME_WAIT, 1380 }; 1381 1382 struct tcp_seq_afinfo { 1383 char *name; 1384 sa_family_t family; 1385 struct file_operations seq_fops; 1386 struct seq_operations seq_ops; 1387 }; 1388 1389 struct tcp_iter_state { 1390 struct seq_net_private p; 1391 sa_family_t family; 1392 enum tcp_seq_states state; 1393 struct sock *syn_wait_sk; 1394 int bucket, offset, sbucket, num, uid; 1395 loff_t last_pos; 1396 }; 1397 1398 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1399 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1400 1401 extern struct request_sock_ops tcp_request_sock_ops; 1402 extern struct request_sock_ops tcp6_request_sock_ops; 1403 1404 extern void tcp_v4_destroy_sock(struct sock *sk); 1405 1406 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1407 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1408 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1409 struct sk_buff *skb); 1410 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1411 struct sk_buff *skb); 1412 extern int tcp_gro_complete(struct sk_buff *skb); 1413 extern int tcp4_gro_complete(struct sk_buff *skb); 1414 1415 #ifdef CONFIG_PROC_FS 1416 extern int tcp4_proc_init(void); 1417 extern void tcp4_proc_exit(void); 1418 #endif 1419 1420 /* TCP af-specific functions */ 1421 struct tcp_sock_af_ops { 1422 #ifdef CONFIG_TCP_MD5SIG 1423 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1424 struct sock *addr_sk); 1425 int (*calc_md5_hash) (char *location, 1426 struct tcp_md5sig_key *md5, 1427 struct sock *sk, 1428 struct request_sock *req, 1429 struct sk_buff *skb); 1430 int (*md5_add) (struct sock *sk, 1431 struct sock *addr_sk, 1432 u8 *newkey, 1433 u8 len); 1434 int (*md5_parse) (struct sock *sk, 1435 char __user *optval, 1436 int optlen); 1437 #endif 1438 }; 1439 1440 struct tcp_request_sock_ops { 1441 #ifdef CONFIG_TCP_MD5SIG 1442 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1443 struct request_sock *req); 1444 int (*calc_md5_hash) (char *location, 1445 struct tcp_md5sig_key *md5, 1446 struct sock *sk, 1447 struct request_sock *req, 1448 struct sk_buff *skb); 1449 #endif 1450 }; 1451 1452 /* Using SHA1 for now, define some constants. 1453 */ 1454 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1455 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1456 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1457 1458 extern int tcp_cookie_generator(u32 *bakery); 1459 1460 /** 1461 * struct tcp_cookie_values - each socket needs extra space for the 1462 * cookies, together with (optional) space for any SYN data. 1463 * 1464 * A tcp_sock contains a pointer to the current value, and this is 1465 * cloned to the tcp_timewait_sock. 1466 * 1467 * @cookie_pair: variable data from the option exchange. 1468 * 1469 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1470 * indicates default (sysctl_tcp_cookie_size). 1471 * After cookie sent, remembers size of cookie. 1472 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1473 * 1474 * @s_data_desired: user specified tcpct_s_data_desired. When the 1475 * constant payload is specified (@s_data_constant), 1476 * holds its length instead. 1477 * Range 0 to TCP_MSS_DESIRED. 1478 * 1479 * @s_data_payload: constant data that is to be included in the 1480 * payload of SYN or SYNACK segments when the 1481 * cookie option is present. 1482 */ 1483 struct tcp_cookie_values { 1484 struct kref kref; 1485 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1486 u8 cookie_pair_size; 1487 u8 cookie_desired; 1488 u16 s_data_desired:11, 1489 s_data_constant:1, 1490 s_data_in:1, 1491 s_data_out:1, 1492 s_data_unused:2; 1493 u8 s_data_payload[0]; 1494 }; 1495 1496 static inline void tcp_cookie_values_release(struct kref *kref) 1497 { 1498 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1499 } 1500 1501 /* The length of constant payload data. Note that s_data_desired is 1502 * overloaded, depending on s_data_constant: either the length of constant 1503 * data (returned here) or the limit on variable data. 1504 */ 1505 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1506 { 1507 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1508 ? tp->cookie_values->s_data_desired 1509 : 0; 1510 } 1511 1512 /** 1513 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1514 * 1515 * As tcp_request_sock has already been extended in other places, the 1516 * only remaining method is to pass stack values along as function 1517 * parameters. These parameters are not needed after sending SYNACK. 1518 * 1519 * @cookie_bakery: cryptographic secret and message workspace. 1520 * 1521 * @cookie_plus: bytes in authenticator/cookie option, copied from 1522 * struct tcp_options_received (above). 1523 */ 1524 struct tcp_extend_values { 1525 struct request_values rv; 1526 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1527 u8 cookie_plus:6, 1528 cookie_out_never:1, 1529 cookie_in_always:1; 1530 }; 1531 1532 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1533 { 1534 return (struct tcp_extend_values *)rvp; 1535 } 1536 1537 extern void tcp_v4_init(void); 1538 extern void tcp_init(void); 1539 1540 #endif /* _TCP_H */ 1541