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 FASTRETRANS_DEBUG 1 22 23 #include <linux/list.h> 24 #include <linux/tcp.h> 25 #include <linux/slab.h> 26 #include <linux/cache.h> 27 #include <linux/percpu.h> 28 #include <linux/skbuff.h> 29 #include <linux/dmaengine.h> 30 #include <linux/crypto.h> 31 #include <linux/cryptohash.h> 32 #include <linux/kref.h> 33 34 #include <net/inet_connection_sock.h> 35 #include <net/inet_timewait_sock.h> 36 #include <net/inet_hashtables.h> 37 #include <net/checksum.h> 38 #include <net/request_sock.h> 39 #include <net/sock.h> 40 #include <net/snmp.h> 41 #include <net/ip.h> 42 #include <net/tcp_states.h> 43 #include <net/inet_ecn.h> 44 #include <net/dst.h> 45 46 #include <linux/seq_file.h> 47 #include <linux/memcontrol.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)(1*HZ)) /* RFC2988bis initial RTO value */ 126 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 127 * used as a fallback RTO for the 128 * initial data transmission if no 129 * valid RTT sample has been acquired, 130 * most likely due to retrans in 3WHS. 131 */ 132 133 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 134 * for local resources. 135 */ 136 137 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 138 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 139 #define TCP_KEEPALIVE_INTVL (75*HZ) 140 141 #define MAX_TCP_KEEPIDLE 32767 142 #define MAX_TCP_KEEPINTVL 32767 143 #define MAX_TCP_KEEPCNT 127 144 #define MAX_TCP_SYNCNT 127 145 146 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 147 148 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 149 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 150 * after this time. It should be equal 151 * (or greater than) TCP_TIMEWAIT_LEN 152 * to provide reliability equal to one 153 * provided by timewait state. 154 */ 155 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 156 * timestamps. It must be less than 157 * minimal timewait lifetime. 158 */ 159 /* 160 * TCP option 161 */ 162 163 #define TCPOPT_NOP 1 /* Padding */ 164 #define TCPOPT_EOL 0 /* End of options */ 165 #define TCPOPT_MSS 2 /* Segment size negotiating */ 166 #define TCPOPT_WINDOW 3 /* Window scaling */ 167 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 168 #define TCPOPT_SACK 5 /* SACK Block */ 169 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 170 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 171 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 172 173 /* 174 * TCP option lengths 175 */ 176 177 #define TCPOLEN_MSS 4 178 #define TCPOLEN_WINDOW 3 179 #define TCPOLEN_SACK_PERM 2 180 #define TCPOLEN_TIMESTAMP 10 181 #define TCPOLEN_MD5SIG 18 182 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 183 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 184 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 185 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 186 187 /* But this is what stacks really send out. */ 188 #define TCPOLEN_TSTAMP_ALIGNED 12 189 #define TCPOLEN_WSCALE_ALIGNED 4 190 #define TCPOLEN_SACKPERM_ALIGNED 4 191 #define TCPOLEN_SACK_BASE 2 192 #define TCPOLEN_SACK_BASE_ALIGNED 4 193 #define TCPOLEN_SACK_PERBLOCK 8 194 #define TCPOLEN_MD5SIG_ALIGNED 20 195 #define TCPOLEN_MSS_ALIGNED 4 196 197 /* Flags in tp->nonagle */ 198 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 199 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 200 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 201 202 /* TCP thin-stream limits */ 203 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 204 205 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 206 #define TCP_INIT_CWND 10 207 208 extern struct inet_timewait_death_row tcp_death_row; 209 210 /* sysctl variables for tcp */ 211 extern int sysctl_tcp_timestamps; 212 extern int sysctl_tcp_window_scaling; 213 extern int sysctl_tcp_sack; 214 extern int sysctl_tcp_fin_timeout; 215 extern int sysctl_tcp_keepalive_time; 216 extern int sysctl_tcp_keepalive_probes; 217 extern int sysctl_tcp_keepalive_intvl; 218 extern int sysctl_tcp_syn_retries; 219 extern int sysctl_tcp_synack_retries; 220 extern int sysctl_tcp_retries1; 221 extern int sysctl_tcp_retries2; 222 extern int sysctl_tcp_orphan_retries; 223 extern int sysctl_tcp_syncookies; 224 extern int sysctl_tcp_retrans_collapse; 225 extern int sysctl_tcp_stdurg; 226 extern int sysctl_tcp_rfc1337; 227 extern int sysctl_tcp_abort_on_overflow; 228 extern int sysctl_tcp_max_orphans; 229 extern int sysctl_tcp_fack; 230 extern int sysctl_tcp_reordering; 231 extern int sysctl_tcp_ecn; 232 extern int sysctl_tcp_dsack; 233 extern int sysctl_tcp_wmem[3]; 234 extern int sysctl_tcp_rmem[3]; 235 extern int sysctl_tcp_app_win; 236 extern int sysctl_tcp_adv_win_scale; 237 extern int sysctl_tcp_tw_reuse; 238 extern int sysctl_tcp_frto; 239 extern int sysctl_tcp_frto_response; 240 extern int sysctl_tcp_low_latency; 241 extern int sysctl_tcp_dma_copybreak; 242 extern int sysctl_tcp_nometrics_save; 243 extern int sysctl_tcp_moderate_rcvbuf; 244 extern int sysctl_tcp_tso_win_divisor; 245 extern int sysctl_tcp_abc; 246 extern int sysctl_tcp_mtu_probing; 247 extern int sysctl_tcp_base_mss; 248 extern int sysctl_tcp_workaround_signed_windows; 249 extern int sysctl_tcp_slow_start_after_idle; 250 extern int sysctl_tcp_max_ssthresh; 251 extern int sysctl_tcp_cookie_size; 252 extern int sysctl_tcp_thin_linear_timeouts; 253 extern int sysctl_tcp_thin_dupack; 254 255 extern atomic_long_t tcp_memory_allocated; 256 extern struct percpu_counter tcp_sockets_allocated; 257 extern int tcp_memory_pressure; 258 259 /* 260 * The next routines deal with comparing 32 bit unsigned ints 261 * and worry about wraparound (automatic with unsigned arithmetic). 262 */ 263 264 static inline int before(__u32 seq1, __u32 seq2) 265 { 266 return (__s32)(seq1-seq2) < 0; 267 } 268 #define after(seq2, seq1) before(seq1, seq2) 269 270 /* is s2<=s1<=s3 ? */ 271 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 272 { 273 return seq3 - seq2 >= seq1 - seq2; 274 } 275 276 static inline bool tcp_out_of_memory(struct sock *sk) 277 { 278 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 279 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 280 return true; 281 return false; 282 } 283 284 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 285 { 286 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 287 int orphans = percpu_counter_read_positive(ocp); 288 289 if (orphans << shift > sysctl_tcp_max_orphans) { 290 orphans = percpu_counter_sum_positive(ocp); 291 if (orphans << shift > sysctl_tcp_max_orphans) 292 return true; 293 } 294 return false; 295 } 296 297 extern bool tcp_check_oom(struct sock *sk, int shift); 298 299 /* syncookies: remember time of last synqueue overflow */ 300 static inline void tcp_synq_overflow(struct sock *sk) 301 { 302 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 303 } 304 305 /* syncookies: no recent synqueue overflow on this listening socket? */ 306 static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 307 { 308 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 309 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 310 } 311 312 extern struct proto tcp_prot; 313 314 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 315 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 316 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 317 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 318 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 319 320 extern void tcp_init_mem(struct net *net); 321 322 extern void tcp_v4_err(struct sk_buff *skb, u32); 323 324 extern void tcp_shutdown (struct sock *sk, int how); 325 326 extern int tcp_v4_rcv(struct sk_buff *skb); 327 328 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it); 329 extern void *tcp_v4_tw_get_peer(struct sock *sk); 330 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 331 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 332 size_t size); 333 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 334 size_t size, int flags); 335 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 336 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 337 const struct tcphdr *th, unsigned int len); 338 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 339 const struct tcphdr *th, unsigned int len); 340 extern void tcp_rcv_space_adjust(struct sock *sk); 341 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 342 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 343 extern void tcp_twsk_destructor(struct sock *sk); 344 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 345 struct pipe_inode_info *pipe, size_t len, 346 unsigned int flags); 347 348 static inline void tcp_dec_quickack_mode(struct sock *sk, 349 const unsigned int pkts) 350 { 351 struct inet_connection_sock *icsk = inet_csk(sk); 352 353 if (icsk->icsk_ack.quick) { 354 if (pkts >= icsk->icsk_ack.quick) { 355 icsk->icsk_ack.quick = 0; 356 /* Leaving quickack mode we deflate ATO. */ 357 icsk->icsk_ack.ato = TCP_ATO_MIN; 358 } else 359 icsk->icsk_ack.quick -= pkts; 360 } 361 } 362 363 #define TCP_ECN_OK 1 364 #define TCP_ECN_QUEUE_CWR 2 365 #define TCP_ECN_DEMAND_CWR 4 366 #define TCP_ECN_SEEN 8 367 368 static __inline__ void 369 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 370 { 371 if (sysctl_tcp_ecn && th->ece && th->cwr) 372 inet_rsk(req)->ecn_ok = 1; 373 } 374 375 enum tcp_tw_status { 376 TCP_TW_SUCCESS = 0, 377 TCP_TW_RST = 1, 378 TCP_TW_ACK = 2, 379 TCP_TW_SYN = 3 380 }; 381 382 383 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 384 struct sk_buff *skb, 385 const struct tcphdr *th); 386 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 387 struct request_sock *req, 388 struct request_sock **prev); 389 extern int tcp_child_process(struct sock *parent, struct sock *child, 390 struct sk_buff *skb); 391 extern int tcp_use_frto(struct sock *sk); 392 extern void tcp_enter_frto(struct sock *sk); 393 extern void tcp_enter_loss(struct sock *sk, int how); 394 extern void tcp_clear_retrans(struct tcp_sock *tp); 395 extern void tcp_update_metrics(struct sock *sk); 396 extern void tcp_close(struct sock *sk, long timeout); 397 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 398 struct poll_table_struct *wait); 399 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 400 char __user *optval, int __user *optlen); 401 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 402 char __user *optval, unsigned int optlen); 403 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 404 char __user *optval, int __user *optlen); 405 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 406 char __user *optval, unsigned int optlen); 407 extern void tcp_set_keepalive(struct sock *sk, int val); 408 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 409 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 410 size_t len, int nonblock, int flags, int *addr_len); 411 extern void tcp_parse_options(const struct sk_buff *skb, 412 struct tcp_options_received *opt_rx, const u8 **hvpp, 413 int estab); 414 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 415 416 /* 417 * TCP v4 functions exported for the inet6 API 418 */ 419 420 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 421 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 422 extern struct sock * tcp_create_openreq_child(struct sock *sk, 423 struct request_sock *req, 424 struct sk_buff *skb); 425 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 426 struct request_sock *req, 427 struct dst_entry *dst); 428 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 429 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 430 int addr_len); 431 extern int tcp_connect(struct sock *sk); 432 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 433 struct request_sock *req, 434 struct request_values *rvp); 435 extern int tcp_disconnect(struct sock *sk, int flags); 436 437 438 /* From syncookies.c */ 439 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 440 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 441 struct ip_options *opt); 442 #ifdef CONFIG_SYN_COOKIES 443 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 444 __u16 *mss); 445 #else 446 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 447 struct sk_buff *skb, 448 __u16 *mss) 449 { 450 return 0; 451 } 452 #endif 453 454 extern __u32 cookie_init_timestamp(struct request_sock *req); 455 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 456 457 /* From net/ipv6/syncookies.c */ 458 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 459 #ifdef CONFIG_SYN_COOKIES 460 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 461 __u16 *mss); 462 #else 463 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 464 struct sk_buff *skb, 465 __u16 *mss) 466 { 467 return 0; 468 } 469 #endif 470 /* tcp_output.c */ 471 472 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 473 int nonagle); 474 extern int tcp_may_send_now(struct sock *sk); 475 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 476 extern void tcp_retransmit_timer(struct sock *sk); 477 extern void tcp_xmit_retransmit_queue(struct sock *); 478 extern void tcp_simple_retransmit(struct sock *); 479 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 480 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 481 482 extern void tcp_send_probe0(struct sock *); 483 extern void tcp_send_partial(struct sock *); 484 extern int tcp_write_wakeup(struct sock *); 485 extern void tcp_send_fin(struct sock *sk); 486 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 487 extern int tcp_send_synack(struct sock *); 488 extern int tcp_syn_flood_action(struct sock *sk, 489 const struct sk_buff *skb, 490 const char *proto); 491 extern void tcp_push_one(struct sock *, unsigned int mss_now); 492 extern void tcp_send_ack(struct sock *sk); 493 extern void tcp_send_delayed_ack(struct sock *sk); 494 495 /* tcp_input.c */ 496 extern void tcp_cwnd_application_limited(struct sock *sk); 497 498 /* tcp_timer.c */ 499 extern void tcp_init_xmit_timers(struct sock *); 500 static inline void tcp_clear_xmit_timers(struct sock *sk) 501 { 502 inet_csk_clear_xmit_timers(sk); 503 } 504 505 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 506 extern unsigned int tcp_current_mss(struct sock *sk); 507 508 /* Bound MSS / TSO packet size with the half of the window */ 509 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 510 { 511 int cutoff; 512 513 /* When peer uses tiny windows, there is no use in packetizing 514 * to sub-MSS pieces for the sake of SWS or making sure there 515 * are enough packets in the pipe for fast recovery. 516 * 517 * On the other hand, for extremely large MSS devices, handling 518 * smaller than MSS windows in this way does make sense. 519 */ 520 if (tp->max_window >= 512) 521 cutoff = (tp->max_window >> 1); 522 else 523 cutoff = tp->max_window; 524 525 if (cutoff && pktsize > cutoff) 526 return max_t(int, cutoff, 68U - tp->tcp_header_len); 527 else 528 return pktsize; 529 } 530 531 /* tcp.c */ 532 extern void tcp_get_info(const struct sock *, struct tcp_info *); 533 534 /* Read 'sendfile()'-style from a TCP socket */ 535 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 536 unsigned int, size_t); 537 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 538 sk_read_actor_t recv_actor); 539 540 extern void tcp_initialize_rcv_mss(struct sock *sk); 541 542 extern int tcp_mtu_to_mss(const struct sock *sk, int pmtu); 543 extern int tcp_mss_to_mtu(const struct sock *sk, int mss); 544 extern void tcp_mtup_init(struct sock *sk); 545 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt); 546 547 static inline void tcp_bound_rto(const struct sock *sk) 548 { 549 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 550 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 551 } 552 553 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 554 { 555 return (tp->srtt >> 3) + tp->rttvar; 556 } 557 558 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 559 { 560 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 561 ntohl(TCP_FLAG_ACK) | 562 snd_wnd); 563 } 564 565 static inline void tcp_fast_path_on(struct tcp_sock *tp) 566 { 567 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 568 } 569 570 static inline void tcp_fast_path_check(struct sock *sk) 571 { 572 struct tcp_sock *tp = tcp_sk(sk); 573 574 if (skb_queue_empty(&tp->out_of_order_queue) && 575 tp->rcv_wnd && 576 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 577 !tp->urg_data) 578 tcp_fast_path_on(tp); 579 } 580 581 /* Compute the actual rto_min value */ 582 static inline u32 tcp_rto_min(struct sock *sk) 583 { 584 const struct dst_entry *dst = __sk_dst_get(sk); 585 u32 rto_min = TCP_RTO_MIN; 586 587 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 588 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 589 return rto_min; 590 } 591 592 /* Compute the actual receive window we are currently advertising. 593 * Rcv_nxt can be after the window if our peer push more data 594 * than the offered window. 595 */ 596 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 597 { 598 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 599 600 if (win < 0) 601 win = 0; 602 return (u32) win; 603 } 604 605 /* Choose a new window, without checks for shrinking, and without 606 * scaling applied to the result. The caller does these things 607 * if necessary. This is a "raw" window selection. 608 */ 609 extern u32 __tcp_select_window(struct sock *sk); 610 611 /* TCP timestamps are only 32-bits, this causes a slight 612 * complication on 64-bit systems since we store a snapshot 613 * of jiffies in the buffer control blocks below. We decided 614 * to use only the low 32-bits of jiffies and hide the ugly 615 * casts with the following macro. 616 */ 617 #define tcp_time_stamp ((__u32)(jiffies)) 618 619 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 620 621 #define TCPHDR_FIN 0x01 622 #define TCPHDR_SYN 0x02 623 #define TCPHDR_RST 0x04 624 #define TCPHDR_PSH 0x08 625 #define TCPHDR_ACK 0x10 626 #define TCPHDR_URG 0x20 627 #define TCPHDR_ECE 0x40 628 #define TCPHDR_CWR 0x80 629 630 /* This is what the send packet queuing engine uses to pass 631 * TCP per-packet control information to the transmission code. 632 * We also store the host-order sequence numbers in here too. 633 * This is 44 bytes if IPV6 is enabled. 634 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 635 */ 636 struct tcp_skb_cb { 637 union { 638 struct inet_skb_parm h4; 639 #if IS_ENABLED(CONFIG_IPV6) 640 struct inet6_skb_parm h6; 641 #endif 642 } header; /* For incoming frames */ 643 __u32 seq; /* Starting sequence number */ 644 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 645 __u32 when; /* used to compute rtt's */ 646 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 647 __u8 sacked; /* State flags for SACK/FACK. */ 648 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 649 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 650 #define TCPCB_LOST 0x04 /* SKB is lost */ 651 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 652 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 653 /* 1 byte hole */ 654 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 655 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 656 657 __u32 ack_seq; /* Sequence number ACK'd */ 658 }; 659 660 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 661 662 /* Due to TSO, an SKB can be composed of multiple actual 663 * packets. To keep these tracked properly, we use this. 664 */ 665 static inline int tcp_skb_pcount(const struct sk_buff *skb) 666 { 667 return skb_shinfo(skb)->gso_segs; 668 } 669 670 /* This is valid iff tcp_skb_pcount() > 1. */ 671 static inline int tcp_skb_mss(const struct sk_buff *skb) 672 { 673 return skb_shinfo(skb)->gso_size; 674 } 675 676 /* Events passed to congestion control interface */ 677 enum tcp_ca_event { 678 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 679 CA_EVENT_CWND_RESTART, /* congestion window restart */ 680 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 681 CA_EVENT_FRTO, /* fast recovery timeout */ 682 CA_EVENT_LOSS, /* loss timeout */ 683 CA_EVENT_FAST_ACK, /* in sequence ack */ 684 CA_EVENT_SLOW_ACK, /* other ack */ 685 }; 686 687 /* 688 * Interface for adding new TCP congestion control handlers 689 */ 690 #define TCP_CA_NAME_MAX 16 691 #define TCP_CA_MAX 128 692 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 693 694 #define TCP_CONG_NON_RESTRICTED 0x1 695 #define TCP_CONG_RTT_STAMP 0x2 696 697 struct tcp_congestion_ops { 698 struct list_head list; 699 unsigned long flags; 700 701 /* initialize private data (optional) */ 702 void (*init)(struct sock *sk); 703 /* cleanup private data (optional) */ 704 void (*release)(struct sock *sk); 705 706 /* return slow start threshold (required) */ 707 u32 (*ssthresh)(struct sock *sk); 708 /* lower bound for congestion window (optional) */ 709 u32 (*min_cwnd)(const struct sock *sk); 710 /* do new cwnd calculation (required) */ 711 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 712 /* call before changing ca_state (optional) */ 713 void (*set_state)(struct sock *sk, u8 new_state); 714 /* call when cwnd event occurs (optional) */ 715 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 716 /* new value of cwnd after loss (optional) */ 717 u32 (*undo_cwnd)(struct sock *sk); 718 /* hook for packet ack accounting (optional) */ 719 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 720 /* get info for inet_diag (optional) */ 721 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 722 723 char name[TCP_CA_NAME_MAX]; 724 struct module *owner; 725 }; 726 727 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 728 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 729 730 extern void tcp_init_congestion_control(struct sock *sk); 731 extern void tcp_cleanup_congestion_control(struct sock *sk); 732 extern int tcp_set_default_congestion_control(const char *name); 733 extern void tcp_get_default_congestion_control(char *name); 734 extern void tcp_get_available_congestion_control(char *buf, size_t len); 735 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 736 extern int tcp_set_allowed_congestion_control(char *allowed); 737 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 738 extern void tcp_slow_start(struct tcp_sock *tp); 739 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 740 741 extern struct tcp_congestion_ops tcp_init_congestion_ops; 742 extern u32 tcp_reno_ssthresh(struct sock *sk); 743 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 744 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 745 extern struct tcp_congestion_ops tcp_reno; 746 747 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 748 { 749 struct inet_connection_sock *icsk = inet_csk(sk); 750 751 if (icsk->icsk_ca_ops->set_state) 752 icsk->icsk_ca_ops->set_state(sk, ca_state); 753 icsk->icsk_ca_state = ca_state; 754 } 755 756 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 757 { 758 const struct inet_connection_sock *icsk = inet_csk(sk); 759 760 if (icsk->icsk_ca_ops->cwnd_event) 761 icsk->icsk_ca_ops->cwnd_event(sk, event); 762 } 763 764 /* These functions determine how the current flow behaves in respect of SACK 765 * handling. SACK is negotiated with the peer, and therefore it can vary 766 * between different flows. 767 * 768 * tcp_is_sack - SACK enabled 769 * tcp_is_reno - No SACK 770 * tcp_is_fack - FACK enabled, implies SACK enabled 771 */ 772 static inline int tcp_is_sack(const struct tcp_sock *tp) 773 { 774 return tp->rx_opt.sack_ok; 775 } 776 777 static inline int tcp_is_reno(const struct tcp_sock *tp) 778 { 779 return !tcp_is_sack(tp); 780 } 781 782 static inline int tcp_is_fack(const struct tcp_sock *tp) 783 { 784 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 785 } 786 787 static inline void tcp_enable_fack(struct tcp_sock *tp) 788 { 789 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 790 } 791 792 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 793 { 794 return tp->sacked_out + tp->lost_out; 795 } 796 797 /* This determines how many packets are "in the network" to the best 798 * of our knowledge. In many cases it is conservative, but where 799 * detailed information is available from the receiver (via SACK 800 * blocks etc.) we can make more aggressive calculations. 801 * 802 * Use this for decisions involving congestion control, use just 803 * tp->packets_out to determine if the send queue is empty or not. 804 * 805 * Read this equation as: 806 * 807 * "Packets sent once on transmission queue" MINUS 808 * "Packets left network, but not honestly ACKed yet" PLUS 809 * "Packets fast retransmitted" 810 */ 811 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 812 { 813 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 814 } 815 816 #define TCP_INFINITE_SSTHRESH 0x7fffffff 817 818 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 819 { 820 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 821 } 822 823 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 824 * The exception is rate halving phase, when cwnd is decreasing towards 825 * ssthresh. 826 */ 827 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 828 { 829 const struct tcp_sock *tp = tcp_sk(sk); 830 831 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 832 return tp->snd_ssthresh; 833 else 834 return max(tp->snd_ssthresh, 835 ((tp->snd_cwnd >> 1) + 836 (tp->snd_cwnd >> 2))); 837 } 838 839 /* Use define here intentionally to get WARN_ON location shown at the caller */ 840 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 841 842 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 843 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 844 845 /* The maximum number of MSS of available cwnd for which TSO defers 846 * sending if not using sysctl_tcp_tso_win_divisor. 847 */ 848 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 849 { 850 return 3; 851 } 852 853 /* Slow start with delack produces 3 packets of burst, so that 854 * it is safe "de facto". This will be the default - same as 855 * the default reordering threshold - but if reordering increases, 856 * we must be able to allow cwnd to burst at least this much in order 857 * to not pull it back when holes are filled. 858 */ 859 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 860 { 861 return tp->reordering; 862 } 863 864 /* Returns end sequence number of the receiver's advertised window */ 865 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 866 { 867 return tp->snd_una + tp->snd_wnd; 868 } 869 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 870 871 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 872 const struct sk_buff *skb) 873 { 874 if (skb->len < mss) 875 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 876 } 877 878 static inline void tcp_check_probe_timer(struct sock *sk) 879 { 880 const struct tcp_sock *tp = tcp_sk(sk); 881 const struct inet_connection_sock *icsk = inet_csk(sk); 882 883 if (!tp->packets_out && !icsk->icsk_pending) 884 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 885 icsk->icsk_rto, TCP_RTO_MAX); 886 } 887 888 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 889 { 890 tp->snd_wl1 = seq; 891 } 892 893 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 894 { 895 tp->snd_wl1 = seq; 896 } 897 898 /* 899 * Calculate(/check) TCP checksum 900 */ 901 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 902 __be32 daddr, __wsum base) 903 { 904 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 905 } 906 907 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 908 { 909 return __skb_checksum_complete(skb); 910 } 911 912 static inline int tcp_checksum_complete(struct sk_buff *skb) 913 { 914 return !skb_csum_unnecessary(skb) && 915 __tcp_checksum_complete(skb); 916 } 917 918 /* Prequeue for VJ style copy to user, combined with checksumming. */ 919 920 static inline void tcp_prequeue_init(struct tcp_sock *tp) 921 { 922 tp->ucopy.task = NULL; 923 tp->ucopy.len = 0; 924 tp->ucopy.memory = 0; 925 skb_queue_head_init(&tp->ucopy.prequeue); 926 #ifdef CONFIG_NET_DMA 927 tp->ucopy.dma_chan = NULL; 928 tp->ucopy.wakeup = 0; 929 tp->ucopy.pinned_list = NULL; 930 tp->ucopy.dma_cookie = 0; 931 #endif 932 } 933 934 /* Packet is added to VJ-style prequeue for processing in process 935 * context, if a reader task is waiting. Apparently, this exciting 936 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 937 * failed somewhere. Latency? Burstiness? Well, at least now we will 938 * see, why it failed. 8)8) --ANK 939 * 940 * NOTE: is this not too big to inline? 941 */ 942 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 943 { 944 struct tcp_sock *tp = tcp_sk(sk); 945 946 if (sysctl_tcp_low_latency || !tp->ucopy.task) 947 return 0; 948 949 __skb_queue_tail(&tp->ucopy.prequeue, skb); 950 tp->ucopy.memory += skb->truesize; 951 if (tp->ucopy.memory > sk->sk_rcvbuf) { 952 struct sk_buff *skb1; 953 954 BUG_ON(sock_owned_by_user(sk)); 955 956 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 957 sk_backlog_rcv(sk, skb1); 958 NET_INC_STATS_BH(sock_net(sk), 959 LINUX_MIB_TCPPREQUEUEDROPPED); 960 } 961 962 tp->ucopy.memory = 0; 963 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 964 wake_up_interruptible_sync_poll(sk_sleep(sk), 965 POLLIN | POLLRDNORM | POLLRDBAND); 966 if (!inet_csk_ack_scheduled(sk)) 967 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 968 (3 * tcp_rto_min(sk)) / 4, 969 TCP_RTO_MAX); 970 } 971 return 1; 972 } 973 974 975 #undef STATE_TRACE 976 977 #ifdef STATE_TRACE 978 static const char *statename[]={ 979 "Unused","Established","Syn Sent","Syn Recv", 980 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 981 "Close Wait","Last ACK","Listen","Closing" 982 }; 983 #endif 984 extern void tcp_set_state(struct sock *sk, int state); 985 986 extern void tcp_done(struct sock *sk); 987 988 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 989 { 990 rx_opt->dsack = 0; 991 rx_opt->num_sacks = 0; 992 } 993 994 /* Determine a window scaling and initial window to offer. */ 995 extern void tcp_select_initial_window(int __space, __u32 mss, 996 __u32 *rcv_wnd, __u32 *window_clamp, 997 int wscale_ok, __u8 *rcv_wscale, 998 __u32 init_rcv_wnd); 999 1000 static inline int tcp_win_from_space(int space) 1001 { 1002 return sysctl_tcp_adv_win_scale<=0 ? 1003 (space>>(-sysctl_tcp_adv_win_scale)) : 1004 space - (space>>sysctl_tcp_adv_win_scale); 1005 } 1006 1007 /* Note: caller must be prepared to deal with negative returns */ 1008 static inline int tcp_space(const struct sock *sk) 1009 { 1010 return tcp_win_from_space(sk->sk_rcvbuf - 1011 atomic_read(&sk->sk_rmem_alloc)); 1012 } 1013 1014 static inline int tcp_full_space(const struct sock *sk) 1015 { 1016 return tcp_win_from_space(sk->sk_rcvbuf); 1017 } 1018 1019 static inline void tcp_openreq_init(struct request_sock *req, 1020 struct tcp_options_received *rx_opt, 1021 struct sk_buff *skb) 1022 { 1023 struct inet_request_sock *ireq = inet_rsk(req); 1024 1025 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1026 req->cookie_ts = 0; 1027 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1028 req->mss = rx_opt->mss_clamp; 1029 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1030 ireq->tstamp_ok = rx_opt->tstamp_ok; 1031 ireq->sack_ok = rx_opt->sack_ok; 1032 ireq->snd_wscale = rx_opt->snd_wscale; 1033 ireq->wscale_ok = rx_opt->wscale_ok; 1034 ireq->acked = 0; 1035 ireq->ecn_ok = 0; 1036 ireq->rmt_port = tcp_hdr(skb)->source; 1037 ireq->loc_port = tcp_hdr(skb)->dest; 1038 } 1039 1040 extern void tcp_enter_memory_pressure(struct sock *sk); 1041 1042 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1043 { 1044 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1045 } 1046 1047 static inline int keepalive_time_when(const struct tcp_sock *tp) 1048 { 1049 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1050 } 1051 1052 static inline int keepalive_probes(const struct tcp_sock *tp) 1053 { 1054 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1055 } 1056 1057 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1058 { 1059 const struct inet_connection_sock *icsk = &tp->inet_conn; 1060 1061 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1062 tcp_time_stamp - tp->rcv_tstamp); 1063 } 1064 1065 static inline int tcp_fin_time(const struct sock *sk) 1066 { 1067 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1068 const int rto = inet_csk(sk)->icsk_rto; 1069 1070 if (fin_timeout < (rto << 2) - (rto >> 1)) 1071 fin_timeout = (rto << 2) - (rto >> 1); 1072 1073 return fin_timeout; 1074 } 1075 1076 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1077 int paws_win) 1078 { 1079 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1080 return 1; 1081 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1082 return 1; 1083 /* 1084 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1085 * then following tcp messages have valid values. Ignore 0 value, 1086 * or else 'negative' tsval might forbid us to accept their packets. 1087 */ 1088 if (!rx_opt->ts_recent) 1089 return 1; 1090 return 0; 1091 } 1092 1093 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1094 int rst) 1095 { 1096 if (tcp_paws_check(rx_opt, 0)) 1097 return 0; 1098 1099 /* RST segments are not recommended to carry timestamp, 1100 and, if they do, it is recommended to ignore PAWS because 1101 "their cleanup function should take precedence over timestamps." 1102 Certainly, it is mistake. It is necessary to understand the reasons 1103 of this constraint to relax it: if peer reboots, clock may go 1104 out-of-sync and half-open connections will not be reset. 1105 Actually, the problem would be not existing if all 1106 the implementations followed draft about maintaining clock 1107 via reboots. Linux-2.2 DOES NOT! 1108 1109 However, we can relax time bounds for RST segments to MSL. 1110 */ 1111 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1112 return 0; 1113 return 1; 1114 } 1115 1116 static inline void tcp_mib_init(struct net *net) 1117 { 1118 /* See RFC 2012 */ 1119 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1120 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1121 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1122 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1123 } 1124 1125 /* from STCP */ 1126 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1127 { 1128 tp->lost_skb_hint = NULL; 1129 tp->scoreboard_skb_hint = NULL; 1130 } 1131 1132 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1133 { 1134 tcp_clear_retrans_hints_partial(tp); 1135 tp->retransmit_skb_hint = NULL; 1136 } 1137 1138 /* MD5 Signature */ 1139 struct crypto_hash; 1140 1141 /* - key database */ 1142 struct tcp_md5sig_key { 1143 u8 *key; 1144 u8 keylen; 1145 }; 1146 1147 struct tcp4_md5sig_key { 1148 struct tcp_md5sig_key base; 1149 __be32 addr; 1150 }; 1151 1152 struct tcp6_md5sig_key { 1153 struct tcp_md5sig_key base; 1154 #if 0 1155 u32 scope_id; /* XXX */ 1156 #endif 1157 struct in6_addr addr; 1158 }; 1159 1160 /* - sock block */ 1161 struct tcp_md5sig_info { 1162 struct tcp4_md5sig_key *keys4; 1163 #if IS_ENABLED(CONFIG_IPV6) 1164 struct tcp6_md5sig_key *keys6; 1165 u32 entries6; 1166 u32 alloced6; 1167 #endif 1168 u32 entries4; 1169 u32 alloced4; 1170 }; 1171 1172 /* - pseudo header */ 1173 struct tcp4_pseudohdr { 1174 __be32 saddr; 1175 __be32 daddr; 1176 __u8 pad; 1177 __u8 protocol; 1178 __be16 len; 1179 }; 1180 1181 struct tcp6_pseudohdr { 1182 struct in6_addr saddr; 1183 struct in6_addr daddr; 1184 __be32 len; 1185 __be32 protocol; /* including padding */ 1186 }; 1187 1188 union tcp_md5sum_block { 1189 struct tcp4_pseudohdr ip4; 1190 #if IS_ENABLED(CONFIG_IPV6) 1191 struct tcp6_pseudohdr ip6; 1192 #endif 1193 }; 1194 1195 /* - pool: digest algorithm, hash description and scratch buffer */ 1196 struct tcp_md5sig_pool { 1197 struct hash_desc md5_desc; 1198 union tcp_md5sum_block md5_blk; 1199 }; 1200 1201 /* - functions */ 1202 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1203 const struct sock *sk, 1204 const struct request_sock *req, 1205 const struct sk_buff *skb); 1206 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1207 struct sock *addr_sk); 1208 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1209 u8 newkeylen); 1210 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1211 1212 #ifdef CONFIG_TCP_MD5SIG 1213 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1214 &(struct tcp_md5sig_key) { \ 1215 .key = (twsk)->tw_md5_key, \ 1216 .keylen = (twsk)->tw_md5_keylen, \ 1217 } : NULL) 1218 #else 1219 #define tcp_twsk_md5_key(twsk) NULL 1220 #endif 1221 1222 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *); 1223 extern void tcp_free_md5sig_pool(void); 1224 1225 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1226 extern void tcp_put_md5sig_pool(void); 1227 1228 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1229 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1230 unsigned header_len); 1231 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1232 const struct tcp_md5sig_key *key); 1233 1234 /* write queue abstraction */ 1235 static inline void tcp_write_queue_purge(struct sock *sk) 1236 { 1237 struct sk_buff *skb; 1238 1239 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1240 sk_wmem_free_skb(sk, skb); 1241 sk_mem_reclaim(sk); 1242 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1243 } 1244 1245 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1246 { 1247 return skb_peek(&sk->sk_write_queue); 1248 } 1249 1250 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1251 { 1252 return skb_peek_tail(&sk->sk_write_queue); 1253 } 1254 1255 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1256 const struct sk_buff *skb) 1257 { 1258 return skb_queue_next(&sk->sk_write_queue, skb); 1259 } 1260 1261 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1262 const struct sk_buff *skb) 1263 { 1264 return skb_queue_prev(&sk->sk_write_queue, skb); 1265 } 1266 1267 #define tcp_for_write_queue(skb, sk) \ 1268 skb_queue_walk(&(sk)->sk_write_queue, skb) 1269 1270 #define tcp_for_write_queue_from(skb, sk) \ 1271 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1272 1273 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1274 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1275 1276 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1277 { 1278 return sk->sk_send_head; 1279 } 1280 1281 static inline bool tcp_skb_is_last(const struct sock *sk, 1282 const struct sk_buff *skb) 1283 { 1284 return skb_queue_is_last(&sk->sk_write_queue, skb); 1285 } 1286 1287 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1288 { 1289 if (tcp_skb_is_last(sk, skb)) 1290 sk->sk_send_head = NULL; 1291 else 1292 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1293 } 1294 1295 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1296 { 1297 if (sk->sk_send_head == skb_unlinked) 1298 sk->sk_send_head = NULL; 1299 } 1300 1301 static inline void tcp_init_send_head(struct sock *sk) 1302 { 1303 sk->sk_send_head = NULL; 1304 } 1305 1306 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1307 { 1308 __skb_queue_tail(&sk->sk_write_queue, skb); 1309 } 1310 1311 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1312 { 1313 __tcp_add_write_queue_tail(sk, skb); 1314 1315 /* Queue it, remembering where we must start sending. */ 1316 if (sk->sk_send_head == NULL) { 1317 sk->sk_send_head = skb; 1318 1319 if (tcp_sk(sk)->highest_sack == NULL) 1320 tcp_sk(sk)->highest_sack = skb; 1321 } 1322 } 1323 1324 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1325 { 1326 __skb_queue_head(&sk->sk_write_queue, skb); 1327 } 1328 1329 /* Insert buff after skb on the write queue of sk. */ 1330 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1331 struct sk_buff *buff, 1332 struct sock *sk) 1333 { 1334 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1335 } 1336 1337 /* Insert new before skb on the write queue of sk. */ 1338 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1339 struct sk_buff *skb, 1340 struct sock *sk) 1341 { 1342 __skb_queue_before(&sk->sk_write_queue, skb, new); 1343 1344 if (sk->sk_send_head == skb) 1345 sk->sk_send_head = new; 1346 } 1347 1348 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1349 { 1350 __skb_unlink(skb, &sk->sk_write_queue); 1351 } 1352 1353 static inline int tcp_write_queue_empty(struct sock *sk) 1354 { 1355 return skb_queue_empty(&sk->sk_write_queue); 1356 } 1357 1358 static inline void tcp_push_pending_frames(struct sock *sk) 1359 { 1360 if (tcp_send_head(sk)) { 1361 struct tcp_sock *tp = tcp_sk(sk); 1362 1363 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1364 } 1365 } 1366 1367 /* Start sequence of the highest skb with SACKed bit, valid only if 1368 * sacked > 0 or when the caller has ensured validity by itself. 1369 */ 1370 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1371 { 1372 if (!tp->sacked_out) 1373 return tp->snd_una; 1374 1375 if (tp->highest_sack == NULL) 1376 return tp->snd_nxt; 1377 1378 return TCP_SKB_CB(tp->highest_sack)->seq; 1379 } 1380 1381 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1382 { 1383 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1384 tcp_write_queue_next(sk, skb); 1385 } 1386 1387 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1388 { 1389 return tcp_sk(sk)->highest_sack; 1390 } 1391 1392 static inline void tcp_highest_sack_reset(struct sock *sk) 1393 { 1394 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1395 } 1396 1397 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1398 static inline void tcp_highest_sack_combine(struct sock *sk, 1399 struct sk_buff *old, 1400 struct sk_buff *new) 1401 { 1402 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1403 tcp_sk(sk)->highest_sack = new; 1404 } 1405 1406 /* Determines whether this is a thin stream (which may suffer from 1407 * increased latency). Used to trigger latency-reducing mechanisms. 1408 */ 1409 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1410 { 1411 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1412 } 1413 1414 /* /proc */ 1415 enum tcp_seq_states { 1416 TCP_SEQ_STATE_LISTENING, 1417 TCP_SEQ_STATE_OPENREQ, 1418 TCP_SEQ_STATE_ESTABLISHED, 1419 TCP_SEQ_STATE_TIME_WAIT, 1420 }; 1421 1422 int tcp_seq_open(struct inode *inode, struct file *file); 1423 1424 struct tcp_seq_afinfo { 1425 char *name; 1426 sa_family_t family; 1427 const struct file_operations *seq_fops; 1428 struct seq_operations seq_ops; 1429 }; 1430 1431 struct tcp_iter_state { 1432 struct seq_net_private p; 1433 sa_family_t family; 1434 enum tcp_seq_states state; 1435 struct sock *syn_wait_sk; 1436 int bucket, offset, sbucket, num, uid; 1437 loff_t last_pos; 1438 }; 1439 1440 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1441 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1442 1443 extern struct request_sock_ops tcp_request_sock_ops; 1444 extern struct request_sock_ops tcp6_request_sock_ops; 1445 1446 extern void tcp_v4_destroy_sock(struct sock *sk); 1447 1448 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1449 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, 1450 netdev_features_t features); 1451 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1452 struct sk_buff *skb); 1453 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1454 struct sk_buff *skb); 1455 extern int tcp_gro_complete(struct sk_buff *skb); 1456 extern int tcp4_gro_complete(struct sk_buff *skb); 1457 1458 #ifdef CONFIG_PROC_FS 1459 extern int tcp4_proc_init(void); 1460 extern void tcp4_proc_exit(void); 1461 #endif 1462 1463 /* TCP af-specific functions */ 1464 struct tcp_sock_af_ops { 1465 #ifdef CONFIG_TCP_MD5SIG 1466 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1467 struct sock *addr_sk); 1468 int (*calc_md5_hash) (char *location, 1469 struct tcp_md5sig_key *md5, 1470 const struct sock *sk, 1471 const struct request_sock *req, 1472 const struct sk_buff *skb); 1473 int (*md5_add) (struct sock *sk, 1474 struct sock *addr_sk, 1475 u8 *newkey, 1476 u8 len); 1477 int (*md5_parse) (struct sock *sk, 1478 char __user *optval, 1479 int optlen); 1480 #endif 1481 }; 1482 1483 struct tcp_request_sock_ops { 1484 #ifdef CONFIG_TCP_MD5SIG 1485 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1486 struct request_sock *req); 1487 int (*calc_md5_hash) (char *location, 1488 struct tcp_md5sig_key *md5, 1489 const struct sock *sk, 1490 const struct request_sock *req, 1491 const struct sk_buff *skb); 1492 #endif 1493 }; 1494 1495 /* Using SHA1 for now, define some constants. 1496 */ 1497 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1498 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1499 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1500 1501 extern int tcp_cookie_generator(u32 *bakery); 1502 1503 /** 1504 * struct tcp_cookie_values - each socket needs extra space for the 1505 * cookies, together with (optional) space for any SYN data. 1506 * 1507 * A tcp_sock contains a pointer to the current value, and this is 1508 * cloned to the tcp_timewait_sock. 1509 * 1510 * @cookie_pair: variable data from the option exchange. 1511 * 1512 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1513 * indicates default (sysctl_tcp_cookie_size). 1514 * After cookie sent, remembers size of cookie. 1515 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1516 * 1517 * @s_data_desired: user specified tcpct_s_data_desired. When the 1518 * constant payload is specified (@s_data_constant), 1519 * holds its length instead. 1520 * Range 0 to TCP_MSS_DESIRED. 1521 * 1522 * @s_data_payload: constant data that is to be included in the 1523 * payload of SYN or SYNACK segments when the 1524 * cookie option is present. 1525 */ 1526 struct tcp_cookie_values { 1527 struct kref kref; 1528 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1529 u8 cookie_pair_size; 1530 u8 cookie_desired; 1531 u16 s_data_desired:11, 1532 s_data_constant:1, 1533 s_data_in:1, 1534 s_data_out:1, 1535 s_data_unused:2; 1536 u8 s_data_payload[0]; 1537 }; 1538 1539 static inline void tcp_cookie_values_release(struct kref *kref) 1540 { 1541 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1542 } 1543 1544 /* The length of constant payload data. Note that s_data_desired is 1545 * overloaded, depending on s_data_constant: either the length of constant 1546 * data (returned here) or the limit on variable data. 1547 */ 1548 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1549 { 1550 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1551 ? tp->cookie_values->s_data_desired 1552 : 0; 1553 } 1554 1555 /** 1556 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1557 * 1558 * As tcp_request_sock has already been extended in other places, the 1559 * only remaining method is to pass stack values along as function 1560 * parameters. These parameters are not needed after sending SYNACK. 1561 * 1562 * @cookie_bakery: cryptographic secret and message workspace. 1563 * 1564 * @cookie_plus: bytes in authenticator/cookie option, copied from 1565 * struct tcp_options_received (above). 1566 */ 1567 struct tcp_extend_values { 1568 struct request_values rv; 1569 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1570 u8 cookie_plus:6, 1571 cookie_out_never:1, 1572 cookie_in_always:1; 1573 }; 1574 1575 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1576 { 1577 return (struct tcp_extend_values *)rvp; 1578 } 1579 1580 extern void tcp_v4_init(void); 1581 extern void tcp_init(void); 1582 1583 #endif /* _TCP_H */ 1584