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 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 200 #define TCP_INIT_CWND 10 201 202 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 long 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_long_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_long_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 struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it); 316 extern void *tcp_v4_tw_get_peer(struct sock *sk); 317 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 318 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 319 size_t size); 320 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 321 size_t size, int flags); 322 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 323 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 324 struct tcphdr *th, unsigned len); 325 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 326 struct tcphdr *th, unsigned len); 327 extern void tcp_rcv_space_adjust(struct sock *sk); 328 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 329 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 330 extern void tcp_twsk_destructor(struct sock *sk); 331 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 332 struct pipe_inode_info *pipe, size_t len, 333 unsigned int flags); 334 335 static inline void tcp_dec_quickack_mode(struct sock *sk, 336 const unsigned int pkts) 337 { 338 struct inet_connection_sock *icsk = inet_csk(sk); 339 340 if (icsk->icsk_ack.quick) { 341 if (pkts >= icsk->icsk_ack.quick) { 342 icsk->icsk_ack.quick = 0; 343 /* Leaving quickack mode we deflate ATO. */ 344 icsk->icsk_ack.ato = TCP_ATO_MIN; 345 } else 346 icsk->icsk_ack.quick -= pkts; 347 } 348 } 349 350 #define TCP_ECN_OK 1 351 #define TCP_ECN_QUEUE_CWR 2 352 #define TCP_ECN_DEMAND_CWR 4 353 354 static __inline__ void 355 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 356 { 357 if (sysctl_tcp_ecn && th->ece && th->cwr) 358 inet_rsk(req)->ecn_ok = 1; 359 } 360 361 enum tcp_tw_status { 362 TCP_TW_SUCCESS = 0, 363 TCP_TW_RST = 1, 364 TCP_TW_ACK = 2, 365 TCP_TW_SYN = 3 366 }; 367 368 369 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 370 struct sk_buff *skb, 371 const struct tcphdr *th); 372 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 373 struct request_sock *req, 374 struct request_sock **prev); 375 extern int tcp_child_process(struct sock *parent, struct sock *child, 376 struct sk_buff *skb); 377 extern int tcp_use_frto(struct sock *sk); 378 extern void tcp_enter_frto(struct sock *sk); 379 extern void tcp_enter_loss(struct sock *sk, int how); 380 extern void tcp_clear_retrans(struct tcp_sock *tp); 381 extern void tcp_update_metrics(struct sock *sk); 382 extern void tcp_close(struct sock *sk, long timeout); 383 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 384 struct poll_table_struct *wait); 385 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 386 char __user *optval, int __user *optlen); 387 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 388 char __user *optval, unsigned int optlen); 389 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 390 char __user *optval, int __user *optlen); 391 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 392 char __user *optval, unsigned int optlen); 393 extern void tcp_set_keepalive(struct sock *sk, int val); 394 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 395 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 396 size_t len, int nonblock, int flags, int *addr_len); 397 extern void tcp_parse_options(struct sk_buff *skb, 398 struct tcp_options_received *opt_rx, u8 **hvpp, 399 int estab); 400 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 401 402 /* 403 * TCP v4 functions exported for the inet6 API 404 */ 405 406 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 407 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 408 extern struct sock * tcp_create_openreq_child(struct sock *sk, 409 struct request_sock *req, 410 struct sk_buff *skb); 411 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 412 struct request_sock *req, 413 struct dst_entry *dst); 414 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 415 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 416 int addr_len); 417 extern int tcp_connect(struct sock *sk); 418 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 419 struct request_sock *req, 420 struct request_values *rvp); 421 extern int tcp_disconnect(struct sock *sk, int flags); 422 423 424 /* From syncookies.c */ 425 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 426 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 427 struct ip_options *opt); 428 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 429 __u16 *mss); 430 431 extern __u32 cookie_init_timestamp(struct request_sock *req); 432 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 433 434 /* From net/ipv6/syncookies.c */ 435 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 436 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 437 __u16 *mss); 438 439 /* tcp_output.c */ 440 441 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 442 int nonagle); 443 extern int tcp_may_send_now(struct sock *sk); 444 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 445 extern void tcp_retransmit_timer(struct sock *sk); 446 extern void tcp_xmit_retransmit_queue(struct sock *); 447 extern void tcp_simple_retransmit(struct sock *); 448 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 449 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 450 451 extern void tcp_send_probe0(struct sock *); 452 extern void tcp_send_partial(struct sock *); 453 extern int tcp_write_wakeup(struct sock *); 454 extern void tcp_send_fin(struct sock *sk); 455 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 456 extern int tcp_send_synack(struct sock *); 457 extern void tcp_push_one(struct sock *, unsigned int mss_now); 458 extern void tcp_send_ack(struct sock *sk); 459 extern void tcp_send_delayed_ack(struct sock *sk); 460 461 /* tcp_input.c */ 462 extern void tcp_cwnd_application_limited(struct sock *sk); 463 464 /* tcp_timer.c */ 465 extern void tcp_init_xmit_timers(struct sock *); 466 static inline void tcp_clear_xmit_timers(struct sock *sk) 467 { 468 inet_csk_clear_xmit_timers(sk); 469 } 470 471 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 472 extern unsigned int tcp_current_mss(struct sock *sk); 473 474 /* Bound MSS / TSO packet size with the half of the window */ 475 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 476 { 477 int cutoff; 478 479 /* When peer uses tiny windows, there is no use in packetizing 480 * to sub-MSS pieces for the sake of SWS or making sure there 481 * are enough packets in the pipe for fast recovery. 482 * 483 * On the other hand, for extremely large MSS devices, handling 484 * smaller than MSS windows in this way does make sense. 485 */ 486 if (tp->max_window >= 512) 487 cutoff = (tp->max_window >> 1); 488 else 489 cutoff = tp->max_window; 490 491 if (cutoff && pktsize > cutoff) 492 return max_t(int, cutoff, 68U - tp->tcp_header_len); 493 else 494 return pktsize; 495 } 496 497 /* tcp.c */ 498 extern void tcp_get_info(struct sock *, struct tcp_info *); 499 500 /* Read 'sendfile()'-style from a TCP socket */ 501 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 502 unsigned int, size_t); 503 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 504 sk_read_actor_t recv_actor); 505 506 extern void tcp_initialize_rcv_mss(struct sock *sk); 507 508 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 509 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 510 extern void tcp_mtup_init(struct sock *sk); 511 512 static inline void tcp_bound_rto(const struct sock *sk) 513 { 514 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 515 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 516 } 517 518 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 519 { 520 return (tp->srtt >> 3) + tp->rttvar; 521 } 522 523 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 524 { 525 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 526 ntohl(TCP_FLAG_ACK) | 527 snd_wnd); 528 } 529 530 static inline void tcp_fast_path_on(struct tcp_sock *tp) 531 { 532 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 533 } 534 535 static inline void tcp_fast_path_check(struct sock *sk) 536 { 537 struct tcp_sock *tp = tcp_sk(sk); 538 539 if (skb_queue_empty(&tp->out_of_order_queue) && 540 tp->rcv_wnd && 541 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 542 !tp->urg_data) 543 tcp_fast_path_on(tp); 544 } 545 546 /* Compute the actual rto_min value */ 547 static inline u32 tcp_rto_min(struct sock *sk) 548 { 549 struct dst_entry *dst = __sk_dst_get(sk); 550 u32 rto_min = TCP_RTO_MIN; 551 552 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 553 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 554 return rto_min; 555 } 556 557 /* Compute the actual receive window we are currently advertising. 558 * Rcv_nxt can be after the window if our peer push more data 559 * than the offered window. 560 */ 561 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 562 { 563 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 564 565 if (win < 0) 566 win = 0; 567 return (u32) win; 568 } 569 570 /* Choose a new window, without checks for shrinking, and without 571 * scaling applied to the result. The caller does these things 572 * if necessary. This is a "raw" window selection. 573 */ 574 extern u32 __tcp_select_window(struct sock *sk); 575 576 /* TCP timestamps are only 32-bits, this causes a slight 577 * complication on 64-bit systems since we store a snapshot 578 * of jiffies in the buffer control blocks below. We decided 579 * to use only the low 32-bits of jiffies and hide the ugly 580 * casts with the following macro. 581 */ 582 #define tcp_time_stamp ((__u32)(jiffies)) 583 584 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 585 586 #define TCPHDR_FIN 0x01 587 #define TCPHDR_SYN 0x02 588 #define TCPHDR_RST 0x04 589 #define TCPHDR_PSH 0x08 590 #define TCPHDR_ACK 0x10 591 #define TCPHDR_URG 0x20 592 #define TCPHDR_ECE 0x40 593 #define TCPHDR_CWR 0x80 594 595 /* This is what the send packet queuing engine uses to pass 596 * TCP per-packet control information to the transmission code. 597 * We also store the host-order sequence numbers in here too. 598 * This is 44 bytes if IPV6 is enabled. 599 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 600 */ 601 struct tcp_skb_cb { 602 union { 603 struct inet_skb_parm h4; 604 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 605 struct inet6_skb_parm h6; 606 #endif 607 } header; /* For incoming frames */ 608 __u32 seq; /* Starting sequence number */ 609 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 610 __u32 when; /* used to compute rtt's */ 611 __u8 flags; /* TCP header flags. */ 612 __u8 sacked; /* State flags for SACK/FACK. */ 613 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 614 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 615 #define TCPCB_LOST 0x04 /* SKB is lost */ 616 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 617 618 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 619 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 620 621 __u32 ack_seq; /* Sequence number ACK'd */ 622 }; 623 624 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 625 626 /* Due to TSO, an SKB can be composed of multiple actual 627 * packets. To keep these tracked properly, we use this. 628 */ 629 static inline int tcp_skb_pcount(const struct sk_buff *skb) 630 { 631 return skb_shinfo(skb)->gso_segs; 632 } 633 634 /* This is valid iff tcp_skb_pcount() > 1. */ 635 static inline int tcp_skb_mss(const struct sk_buff *skb) 636 { 637 return skb_shinfo(skb)->gso_size; 638 } 639 640 /* Events passed to congestion control interface */ 641 enum tcp_ca_event { 642 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 643 CA_EVENT_CWND_RESTART, /* congestion window restart */ 644 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 645 CA_EVENT_FRTO, /* fast recovery timeout */ 646 CA_EVENT_LOSS, /* loss timeout */ 647 CA_EVENT_FAST_ACK, /* in sequence ack */ 648 CA_EVENT_SLOW_ACK, /* other ack */ 649 }; 650 651 /* 652 * Interface for adding new TCP congestion control handlers 653 */ 654 #define TCP_CA_NAME_MAX 16 655 #define TCP_CA_MAX 128 656 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 657 658 #define TCP_CONG_NON_RESTRICTED 0x1 659 #define TCP_CONG_RTT_STAMP 0x2 660 661 struct tcp_congestion_ops { 662 struct list_head list; 663 unsigned long flags; 664 665 /* initialize private data (optional) */ 666 void (*init)(struct sock *sk); 667 /* cleanup private data (optional) */ 668 void (*release)(struct sock *sk); 669 670 /* return slow start threshold (required) */ 671 u32 (*ssthresh)(struct sock *sk); 672 /* lower bound for congestion window (optional) */ 673 u32 (*min_cwnd)(const struct sock *sk); 674 /* do new cwnd calculation (required) */ 675 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 676 /* call before changing ca_state (optional) */ 677 void (*set_state)(struct sock *sk, u8 new_state); 678 /* call when cwnd event occurs (optional) */ 679 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 680 /* new value of cwnd after loss (optional) */ 681 u32 (*undo_cwnd)(struct sock *sk); 682 /* hook for packet ack accounting (optional) */ 683 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 684 /* get info for inet_diag (optional) */ 685 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 686 687 char name[TCP_CA_NAME_MAX]; 688 struct module *owner; 689 }; 690 691 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 692 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 693 694 extern void tcp_init_congestion_control(struct sock *sk); 695 extern void tcp_cleanup_congestion_control(struct sock *sk); 696 extern int tcp_set_default_congestion_control(const char *name); 697 extern void tcp_get_default_congestion_control(char *name); 698 extern void tcp_get_available_congestion_control(char *buf, size_t len); 699 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 700 extern int tcp_set_allowed_congestion_control(char *allowed); 701 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 702 extern void tcp_slow_start(struct tcp_sock *tp); 703 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 704 705 extern struct tcp_congestion_ops tcp_init_congestion_ops; 706 extern u32 tcp_reno_ssthresh(struct sock *sk); 707 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 708 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 709 extern struct tcp_congestion_ops tcp_reno; 710 711 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 712 { 713 struct inet_connection_sock *icsk = inet_csk(sk); 714 715 if (icsk->icsk_ca_ops->set_state) 716 icsk->icsk_ca_ops->set_state(sk, ca_state); 717 icsk->icsk_ca_state = ca_state; 718 } 719 720 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 721 { 722 const struct inet_connection_sock *icsk = inet_csk(sk); 723 724 if (icsk->icsk_ca_ops->cwnd_event) 725 icsk->icsk_ca_ops->cwnd_event(sk, event); 726 } 727 728 /* These functions determine how the current flow behaves in respect of SACK 729 * handling. SACK is negotiated with the peer, and therefore it can vary 730 * between different flows. 731 * 732 * tcp_is_sack - SACK enabled 733 * tcp_is_reno - No SACK 734 * tcp_is_fack - FACK enabled, implies SACK enabled 735 */ 736 static inline int tcp_is_sack(const struct tcp_sock *tp) 737 { 738 return tp->rx_opt.sack_ok; 739 } 740 741 static inline int tcp_is_reno(const struct tcp_sock *tp) 742 { 743 return !tcp_is_sack(tp); 744 } 745 746 static inline int tcp_is_fack(const struct tcp_sock *tp) 747 { 748 return tp->rx_opt.sack_ok & 2; 749 } 750 751 static inline void tcp_enable_fack(struct tcp_sock *tp) 752 { 753 tp->rx_opt.sack_ok |= 2; 754 } 755 756 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 757 { 758 return tp->sacked_out + tp->lost_out; 759 } 760 761 /* This determines how many packets are "in the network" to the best 762 * of our knowledge. In many cases it is conservative, but where 763 * detailed information is available from the receiver (via SACK 764 * blocks etc.) we can make more aggressive calculations. 765 * 766 * Use this for decisions involving congestion control, use just 767 * tp->packets_out to determine if the send queue is empty or not. 768 * 769 * Read this equation as: 770 * 771 * "Packets sent once on transmission queue" MINUS 772 * "Packets left network, but not honestly ACKed yet" PLUS 773 * "Packets fast retransmitted" 774 */ 775 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 776 { 777 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 778 } 779 780 #define TCP_INFINITE_SSTHRESH 0x7fffffff 781 782 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 783 { 784 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 785 } 786 787 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 788 * The exception is rate halving phase, when cwnd is decreasing towards 789 * ssthresh. 790 */ 791 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 792 { 793 const struct tcp_sock *tp = tcp_sk(sk); 794 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 795 return tp->snd_ssthresh; 796 else 797 return max(tp->snd_ssthresh, 798 ((tp->snd_cwnd >> 1) + 799 (tp->snd_cwnd >> 2))); 800 } 801 802 /* Use define here intentionally to get WARN_ON location shown at the caller */ 803 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 804 805 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 806 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 807 808 /* Slow start with delack produces 3 packets of burst, so that 809 * it is safe "de facto". This will be the default - same as 810 * the default reordering threshold - but if reordering increases, 811 * we must be able to allow cwnd to burst at least this much in order 812 * to not pull it back when holes are filled. 813 */ 814 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 815 { 816 return tp->reordering; 817 } 818 819 /* Returns end sequence number of the receiver's advertised window */ 820 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 821 { 822 return tp->snd_una + tp->snd_wnd; 823 } 824 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 825 826 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 827 const struct sk_buff *skb) 828 { 829 if (skb->len < mss) 830 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 831 } 832 833 static inline void tcp_check_probe_timer(struct sock *sk) 834 { 835 struct tcp_sock *tp = tcp_sk(sk); 836 const struct inet_connection_sock *icsk = inet_csk(sk); 837 838 if (!tp->packets_out && !icsk->icsk_pending) 839 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 840 icsk->icsk_rto, TCP_RTO_MAX); 841 } 842 843 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 844 { 845 tp->snd_wl1 = seq; 846 } 847 848 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 849 { 850 tp->snd_wl1 = seq; 851 } 852 853 /* 854 * Calculate(/check) TCP checksum 855 */ 856 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 857 __be32 daddr, __wsum base) 858 { 859 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 860 } 861 862 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 863 { 864 return __skb_checksum_complete(skb); 865 } 866 867 static inline int tcp_checksum_complete(struct sk_buff *skb) 868 { 869 return !skb_csum_unnecessary(skb) && 870 __tcp_checksum_complete(skb); 871 } 872 873 /* Prequeue for VJ style copy to user, combined with checksumming. */ 874 875 static inline void tcp_prequeue_init(struct tcp_sock *tp) 876 { 877 tp->ucopy.task = NULL; 878 tp->ucopy.len = 0; 879 tp->ucopy.memory = 0; 880 skb_queue_head_init(&tp->ucopy.prequeue); 881 #ifdef CONFIG_NET_DMA 882 tp->ucopy.dma_chan = NULL; 883 tp->ucopy.wakeup = 0; 884 tp->ucopy.pinned_list = NULL; 885 tp->ucopy.dma_cookie = 0; 886 #endif 887 } 888 889 /* Packet is added to VJ-style prequeue for processing in process 890 * context, if a reader task is waiting. Apparently, this exciting 891 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 892 * failed somewhere. Latency? Burstiness? Well, at least now we will 893 * see, why it failed. 8)8) --ANK 894 * 895 * NOTE: is this not too big to inline? 896 */ 897 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 898 { 899 struct tcp_sock *tp = tcp_sk(sk); 900 901 if (sysctl_tcp_low_latency || !tp->ucopy.task) 902 return 0; 903 904 __skb_queue_tail(&tp->ucopy.prequeue, skb); 905 tp->ucopy.memory += skb->truesize; 906 if (tp->ucopy.memory > sk->sk_rcvbuf) { 907 struct sk_buff *skb1; 908 909 BUG_ON(sock_owned_by_user(sk)); 910 911 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 912 sk_backlog_rcv(sk, skb1); 913 NET_INC_STATS_BH(sock_net(sk), 914 LINUX_MIB_TCPPREQUEUEDROPPED); 915 } 916 917 tp->ucopy.memory = 0; 918 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 919 wake_up_interruptible_sync_poll(sk_sleep(sk), 920 POLLIN | POLLRDNORM | POLLRDBAND); 921 if (!inet_csk_ack_scheduled(sk)) 922 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 923 (3 * tcp_rto_min(sk)) / 4, 924 TCP_RTO_MAX); 925 } 926 return 1; 927 } 928 929 930 #undef STATE_TRACE 931 932 #ifdef STATE_TRACE 933 static const char *statename[]={ 934 "Unused","Established","Syn Sent","Syn Recv", 935 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 936 "Close Wait","Last ACK","Listen","Closing" 937 }; 938 #endif 939 extern void tcp_set_state(struct sock *sk, int state); 940 941 extern void tcp_done(struct sock *sk); 942 943 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 944 { 945 rx_opt->dsack = 0; 946 rx_opt->num_sacks = 0; 947 } 948 949 /* Determine a window scaling and initial window to offer. */ 950 extern void tcp_select_initial_window(int __space, __u32 mss, 951 __u32 *rcv_wnd, __u32 *window_clamp, 952 int wscale_ok, __u8 *rcv_wscale, 953 __u32 init_rcv_wnd); 954 955 static inline int tcp_win_from_space(int space) 956 { 957 return sysctl_tcp_adv_win_scale<=0 ? 958 (space>>(-sysctl_tcp_adv_win_scale)) : 959 space - (space>>sysctl_tcp_adv_win_scale); 960 } 961 962 /* Note: caller must be prepared to deal with negative returns */ 963 static inline int tcp_space(const struct sock *sk) 964 { 965 return tcp_win_from_space(sk->sk_rcvbuf - 966 atomic_read(&sk->sk_rmem_alloc)); 967 } 968 969 static inline int tcp_full_space(const struct sock *sk) 970 { 971 return tcp_win_from_space(sk->sk_rcvbuf); 972 } 973 974 static inline void tcp_openreq_init(struct request_sock *req, 975 struct tcp_options_received *rx_opt, 976 struct sk_buff *skb) 977 { 978 struct inet_request_sock *ireq = inet_rsk(req); 979 980 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 981 req->cookie_ts = 0; 982 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 983 req->mss = rx_opt->mss_clamp; 984 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 985 ireq->tstamp_ok = rx_opt->tstamp_ok; 986 ireq->sack_ok = rx_opt->sack_ok; 987 ireq->snd_wscale = rx_opt->snd_wscale; 988 ireq->wscale_ok = rx_opt->wscale_ok; 989 ireq->acked = 0; 990 ireq->ecn_ok = 0; 991 ireq->rmt_port = tcp_hdr(skb)->source; 992 ireq->loc_port = tcp_hdr(skb)->dest; 993 } 994 995 extern void tcp_enter_memory_pressure(struct sock *sk); 996 997 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 998 { 999 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1000 } 1001 1002 static inline int keepalive_time_when(const struct tcp_sock *tp) 1003 { 1004 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1005 } 1006 1007 static inline int keepalive_probes(const struct tcp_sock *tp) 1008 { 1009 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1010 } 1011 1012 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1013 { 1014 const struct inet_connection_sock *icsk = &tp->inet_conn; 1015 1016 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1017 tcp_time_stamp - tp->rcv_tstamp); 1018 } 1019 1020 static inline int tcp_fin_time(const struct sock *sk) 1021 { 1022 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1023 const int rto = inet_csk(sk)->icsk_rto; 1024 1025 if (fin_timeout < (rto << 2) - (rto >> 1)) 1026 fin_timeout = (rto << 2) - (rto >> 1); 1027 1028 return fin_timeout; 1029 } 1030 1031 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1032 int paws_win) 1033 { 1034 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1035 return 1; 1036 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1037 return 1; 1038 /* 1039 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1040 * then following tcp messages have valid values. Ignore 0 value, 1041 * or else 'negative' tsval might forbid us to accept their packets. 1042 */ 1043 if (!rx_opt->ts_recent) 1044 return 1; 1045 return 0; 1046 } 1047 1048 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1049 int rst) 1050 { 1051 if (tcp_paws_check(rx_opt, 0)) 1052 return 0; 1053 1054 /* RST segments are not recommended to carry timestamp, 1055 and, if they do, it is recommended to ignore PAWS because 1056 "their cleanup function should take precedence over timestamps." 1057 Certainly, it is mistake. It is necessary to understand the reasons 1058 of this constraint to relax it: if peer reboots, clock may go 1059 out-of-sync and half-open connections will not be reset. 1060 Actually, the problem would be not existing if all 1061 the implementations followed draft about maintaining clock 1062 via reboots. Linux-2.2 DOES NOT! 1063 1064 However, we can relax time bounds for RST segments to MSL. 1065 */ 1066 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1067 return 0; 1068 return 1; 1069 } 1070 1071 static inline void tcp_mib_init(struct net *net) 1072 { 1073 /* See RFC 2012 */ 1074 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1075 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1076 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1077 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1078 } 1079 1080 /* from STCP */ 1081 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1082 { 1083 tp->lost_skb_hint = NULL; 1084 tp->scoreboard_skb_hint = NULL; 1085 } 1086 1087 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1088 { 1089 tcp_clear_retrans_hints_partial(tp); 1090 tp->retransmit_skb_hint = NULL; 1091 } 1092 1093 /* MD5 Signature */ 1094 struct crypto_hash; 1095 1096 /* - key database */ 1097 struct tcp_md5sig_key { 1098 u8 *key; 1099 u8 keylen; 1100 }; 1101 1102 struct tcp4_md5sig_key { 1103 struct tcp_md5sig_key base; 1104 __be32 addr; 1105 }; 1106 1107 struct tcp6_md5sig_key { 1108 struct tcp_md5sig_key base; 1109 #if 0 1110 u32 scope_id; /* XXX */ 1111 #endif 1112 struct in6_addr addr; 1113 }; 1114 1115 /* - sock block */ 1116 struct tcp_md5sig_info { 1117 struct tcp4_md5sig_key *keys4; 1118 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1119 struct tcp6_md5sig_key *keys6; 1120 u32 entries6; 1121 u32 alloced6; 1122 #endif 1123 u32 entries4; 1124 u32 alloced4; 1125 }; 1126 1127 /* - pseudo header */ 1128 struct tcp4_pseudohdr { 1129 __be32 saddr; 1130 __be32 daddr; 1131 __u8 pad; 1132 __u8 protocol; 1133 __be16 len; 1134 }; 1135 1136 struct tcp6_pseudohdr { 1137 struct in6_addr saddr; 1138 struct in6_addr daddr; 1139 __be32 len; 1140 __be32 protocol; /* including padding */ 1141 }; 1142 1143 union tcp_md5sum_block { 1144 struct tcp4_pseudohdr ip4; 1145 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1146 struct tcp6_pseudohdr ip6; 1147 #endif 1148 }; 1149 1150 /* - pool: digest algorithm, hash description and scratch buffer */ 1151 struct tcp_md5sig_pool { 1152 struct hash_desc md5_desc; 1153 union tcp_md5sum_block md5_blk; 1154 }; 1155 1156 /* - functions */ 1157 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1158 struct sock *sk, struct request_sock *req, 1159 struct sk_buff *skb); 1160 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1161 struct sock *addr_sk); 1162 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1163 u8 newkeylen); 1164 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1165 1166 #ifdef CONFIG_TCP_MD5SIG 1167 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1168 &(struct tcp_md5sig_key) { \ 1169 .key = (twsk)->tw_md5_key, \ 1170 .keylen = (twsk)->tw_md5_keylen, \ 1171 } : NULL) 1172 #else 1173 #define tcp_twsk_md5_key(twsk) NULL 1174 #endif 1175 1176 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1177 extern void tcp_free_md5sig_pool(void); 1178 1179 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1180 extern void tcp_put_md5sig_pool(void); 1181 1182 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1183 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1184 unsigned header_len); 1185 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1186 struct tcp_md5sig_key *key); 1187 1188 /* write queue abstraction */ 1189 static inline void tcp_write_queue_purge(struct sock *sk) 1190 { 1191 struct sk_buff *skb; 1192 1193 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1194 sk_wmem_free_skb(sk, skb); 1195 sk_mem_reclaim(sk); 1196 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1197 } 1198 1199 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1200 { 1201 return skb_peek(&sk->sk_write_queue); 1202 } 1203 1204 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1205 { 1206 return skb_peek_tail(&sk->sk_write_queue); 1207 } 1208 1209 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1210 { 1211 return skb_queue_next(&sk->sk_write_queue, skb); 1212 } 1213 1214 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1215 { 1216 return skb_queue_prev(&sk->sk_write_queue, skb); 1217 } 1218 1219 #define tcp_for_write_queue(skb, sk) \ 1220 skb_queue_walk(&(sk)->sk_write_queue, skb) 1221 1222 #define tcp_for_write_queue_from(skb, sk) \ 1223 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1224 1225 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1226 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1227 1228 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1229 { 1230 return sk->sk_send_head; 1231 } 1232 1233 static inline bool tcp_skb_is_last(const struct sock *sk, 1234 const struct sk_buff *skb) 1235 { 1236 return skb_queue_is_last(&sk->sk_write_queue, skb); 1237 } 1238 1239 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1240 { 1241 if (tcp_skb_is_last(sk, skb)) 1242 sk->sk_send_head = NULL; 1243 else 1244 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1245 } 1246 1247 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1248 { 1249 if (sk->sk_send_head == skb_unlinked) 1250 sk->sk_send_head = NULL; 1251 } 1252 1253 static inline void tcp_init_send_head(struct sock *sk) 1254 { 1255 sk->sk_send_head = NULL; 1256 } 1257 1258 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1259 { 1260 __skb_queue_tail(&sk->sk_write_queue, skb); 1261 } 1262 1263 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1264 { 1265 __tcp_add_write_queue_tail(sk, skb); 1266 1267 /* Queue it, remembering where we must start sending. */ 1268 if (sk->sk_send_head == NULL) { 1269 sk->sk_send_head = skb; 1270 1271 if (tcp_sk(sk)->highest_sack == NULL) 1272 tcp_sk(sk)->highest_sack = skb; 1273 } 1274 } 1275 1276 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1277 { 1278 __skb_queue_head(&sk->sk_write_queue, skb); 1279 } 1280 1281 /* Insert buff after skb on the write queue of sk. */ 1282 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1283 struct sk_buff *buff, 1284 struct sock *sk) 1285 { 1286 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1287 } 1288 1289 /* Insert new before skb on the write queue of sk. */ 1290 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1291 struct sk_buff *skb, 1292 struct sock *sk) 1293 { 1294 __skb_queue_before(&sk->sk_write_queue, skb, new); 1295 1296 if (sk->sk_send_head == skb) 1297 sk->sk_send_head = new; 1298 } 1299 1300 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1301 { 1302 __skb_unlink(skb, &sk->sk_write_queue); 1303 } 1304 1305 static inline int tcp_write_queue_empty(struct sock *sk) 1306 { 1307 return skb_queue_empty(&sk->sk_write_queue); 1308 } 1309 1310 static inline void tcp_push_pending_frames(struct sock *sk) 1311 { 1312 if (tcp_send_head(sk)) { 1313 struct tcp_sock *tp = tcp_sk(sk); 1314 1315 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1316 } 1317 } 1318 1319 /* Start sequence of the highest skb with SACKed bit, valid only if 1320 * sacked > 0 or when the caller has ensured validity by itself. 1321 */ 1322 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1323 { 1324 if (!tp->sacked_out) 1325 return tp->snd_una; 1326 1327 if (tp->highest_sack == NULL) 1328 return tp->snd_nxt; 1329 1330 return TCP_SKB_CB(tp->highest_sack)->seq; 1331 } 1332 1333 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1334 { 1335 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1336 tcp_write_queue_next(sk, skb); 1337 } 1338 1339 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1340 { 1341 return tcp_sk(sk)->highest_sack; 1342 } 1343 1344 static inline void tcp_highest_sack_reset(struct sock *sk) 1345 { 1346 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1347 } 1348 1349 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1350 static inline void tcp_highest_sack_combine(struct sock *sk, 1351 struct sk_buff *old, 1352 struct sk_buff *new) 1353 { 1354 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1355 tcp_sk(sk)->highest_sack = new; 1356 } 1357 1358 /* Determines whether this is a thin stream (which may suffer from 1359 * increased latency). Used to trigger latency-reducing mechanisms. 1360 */ 1361 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1362 { 1363 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1364 } 1365 1366 /* /proc */ 1367 enum tcp_seq_states { 1368 TCP_SEQ_STATE_LISTENING, 1369 TCP_SEQ_STATE_OPENREQ, 1370 TCP_SEQ_STATE_ESTABLISHED, 1371 TCP_SEQ_STATE_TIME_WAIT, 1372 }; 1373 1374 struct tcp_seq_afinfo { 1375 char *name; 1376 sa_family_t family; 1377 struct file_operations seq_fops; 1378 struct seq_operations seq_ops; 1379 }; 1380 1381 struct tcp_iter_state { 1382 struct seq_net_private p; 1383 sa_family_t family; 1384 enum tcp_seq_states state; 1385 struct sock *syn_wait_sk; 1386 int bucket, offset, sbucket, num, uid; 1387 loff_t last_pos; 1388 }; 1389 1390 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1391 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1392 1393 extern struct request_sock_ops tcp_request_sock_ops; 1394 extern struct request_sock_ops tcp6_request_sock_ops; 1395 1396 extern void tcp_v4_destroy_sock(struct sock *sk); 1397 1398 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1399 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, u32 features); 1400 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1401 struct sk_buff *skb); 1402 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1403 struct sk_buff *skb); 1404 extern int tcp_gro_complete(struct sk_buff *skb); 1405 extern int tcp4_gro_complete(struct sk_buff *skb); 1406 1407 #ifdef CONFIG_PROC_FS 1408 extern int tcp4_proc_init(void); 1409 extern void tcp4_proc_exit(void); 1410 #endif 1411 1412 /* TCP af-specific functions */ 1413 struct tcp_sock_af_ops { 1414 #ifdef CONFIG_TCP_MD5SIG 1415 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1416 struct sock *addr_sk); 1417 int (*calc_md5_hash) (char *location, 1418 struct tcp_md5sig_key *md5, 1419 struct sock *sk, 1420 struct request_sock *req, 1421 struct sk_buff *skb); 1422 int (*md5_add) (struct sock *sk, 1423 struct sock *addr_sk, 1424 u8 *newkey, 1425 u8 len); 1426 int (*md5_parse) (struct sock *sk, 1427 char __user *optval, 1428 int optlen); 1429 #endif 1430 }; 1431 1432 struct tcp_request_sock_ops { 1433 #ifdef CONFIG_TCP_MD5SIG 1434 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1435 struct request_sock *req); 1436 int (*calc_md5_hash) (char *location, 1437 struct tcp_md5sig_key *md5, 1438 struct sock *sk, 1439 struct request_sock *req, 1440 struct sk_buff *skb); 1441 #endif 1442 }; 1443 1444 /* Using SHA1 for now, define some constants. 1445 */ 1446 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1447 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1448 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1449 1450 extern int tcp_cookie_generator(u32 *bakery); 1451 1452 /** 1453 * struct tcp_cookie_values - each socket needs extra space for the 1454 * cookies, together with (optional) space for any SYN data. 1455 * 1456 * A tcp_sock contains a pointer to the current value, and this is 1457 * cloned to the tcp_timewait_sock. 1458 * 1459 * @cookie_pair: variable data from the option exchange. 1460 * 1461 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1462 * indicates default (sysctl_tcp_cookie_size). 1463 * After cookie sent, remembers size of cookie. 1464 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1465 * 1466 * @s_data_desired: user specified tcpct_s_data_desired. When the 1467 * constant payload is specified (@s_data_constant), 1468 * holds its length instead. 1469 * Range 0 to TCP_MSS_DESIRED. 1470 * 1471 * @s_data_payload: constant data that is to be included in the 1472 * payload of SYN or SYNACK segments when the 1473 * cookie option is present. 1474 */ 1475 struct tcp_cookie_values { 1476 struct kref kref; 1477 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1478 u8 cookie_pair_size; 1479 u8 cookie_desired; 1480 u16 s_data_desired:11, 1481 s_data_constant:1, 1482 s_data_in:1, 1483 s_data_out:1, 1484 s_data_unused:2; 1485 u8 s_data_payload[0]; 1486 }; 1487 1488 static inline void tcp_cookie_values_release(struct kref *kref) 1489 { 1490 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1491 } 1492 1493 /* The length of constant payload data. Note that s_data_desired is 1494 * overloaded, depending on s_data_constant: either the length of constant 1495 * data (returned here) or the limit on variable data. 1496 */ 1497 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1498 { 1499 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1500 ? tp->cookie_values->s_data_desired 1501 : 0; 1502 } 1503 1504 /** 1505 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1506 * 1507 * As tcp_request_sock has already been extended in other places, the 1508 * only remaining method is to pass stack values along as function 1509 * parameters. These parameters are not needed after sending SYNACK. 1510 * 1511 * @cookie_bakery: cryptographic secret and message workspace. 1512 * 1513 * @cookie_plus: bytes in authenticator/cookie option, copied from 1514 * struct tcp_options_received (above). 1515 */ 1516 struct tcp_extend_values { 1517 struct request_values rv; 1518 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1519 u8 cookie_plus:6, 1520 cookie_out_never:1, 1521 cookie_in_always:1; 1522 }; 1523 1524 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1525 { 1526 return (struct tcp_extend_values *)rvp; 1527 } 1528 1529 extern void tcp_v4_init(void); 1530 extern void tcp_init(void); 1531 1532 #endif /* _TCP_H */ 1533