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