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_v4_err(struct sk_buff *skb, u32); 315 316 extern void tcp_shutdown (struct sock *sk, int how); 317 318 extern int tcp_v4_rcv(struct sk_buff *skb); 319 320 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it); 321 extern void *tcp_v4_tw_get_peer(struct sock *sk); 322 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 323 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 324 size_t size); 325 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 326 size_t size, int flags); 327 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 328 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 329 const struct tcphdr *th, unsigned int len); 330 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 331 const struct tcphdr *th, unsigned int len); 332 extern void tcp_rcv_space_adjust(struct sock *sk); 333 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 334 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 335 extern void tcp_twsk_destructor(struct sock *sk); 336 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 337 struct pipe_inode_info *pipe, size_t len, 338 unsigned int flags); 339 340 static inline void tcp_dec_quickack_mode(struct sock *sk, 341 const unsigned int pkts) 342 { 343 struct inet_connection_sock *icsk = inet_csk(sk); 344 345 if (icsk->icsk_ack.quick) { 346 if (pkts >= icsk->icsk_ack.quick) { 347 icsk->icsk_ack.quick = 0; 348 /* Leaving quickack mode we deflate ATO. */ 349 icsk->icsk_ack.ato = TCP_ATO_MIN; 350 } else 351 icsk->icsk_ack.quick -= pkts; 352 } 353 } 354 355 #define TCP_ECN_OK 1 356 #define TCP_ECN_QUEUE_CWR 2 357 #define TCP_ECN_DEMAND_CWR 4 358 #define TCP_ECN_SEEN 8 359 360 static __inline__ void 361 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 362 { 363 if (sysctl_tcp_ecn && th->ece && th->cwr) 364 inet_rsk(req)->ecn_ok = 1; 365 } 366 367 enum tcp_tw_status { 368 TCP_TW_SUCCESS = 0, 369 TCP_TW_RST = 1, 370 TCP_TW_ACK = 2, 371 TCP_TW_SYN = 3 372 }; 373 374 375 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 376 struct sk_buff *skb, 377 const struct tcphdr *th); 378 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 379 struct request_sock *req, 380 struct request_sock **prev); 381 extern int tcp_child_process(struct sock *parent, struct sock *child, 382 struct sk_buff *skb); 383 extern int tcp_use_frto(struct sock *sk); 384 extern void tcp_enter_frto(struct sock *sk); 385 extern void tcp_enter_loss(struct sock *sk, int how); 386 extern void tcp_clear_retrans(struct tcp_sock *tp); 387 extern void tcp_update_metrics(struct sock *sk); 388 extern void tcp_close(struct sock *sk, long timeout); 389 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 390 struct poll_table_struct *wait); 391 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 392 char __user *optval, int __user *optlen); 393 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 394 char __user *optval, unsigned int optlen); 395 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 396 char __user *optval, int __user *optlen); 397 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, unsigned int optlen); 399 extern void tcp_set_keepalive(struct sock *sk, int val); 400 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 401 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 402 size_t len, int nonblock, int flags, int *addr_len); 403 extern void tcp_parse_options(const struct sk_buff *skb, 404 struct tcp_options_received *opt_rx, const u8 **hvpp, 405 int estab); 406 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 407 408 /* 409 * TCP v4 functions exported for the inet6 API 410 */ 411 412 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 413 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 414 extern struct sock * tcp_create_openreq_child(struct sock *sk, 415 struct request_sock *req, 416 struct sk_buff *skb); 417 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 418 struct request_sock *req, 419 struct dst_entry *dst); 420 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 421 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 422 int addr_len); 423 extern int tcp_connect(struct sock *sk); 424 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 425 struct request_sock *req, 426 struct request_values *rvp); 427 extern int tcp_disconnect(struct sock *sk, int flags); 428 429 430 /* From syncookies.c */ 431 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 432 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 433 struct ip_options *opt); 434 #ifdef CONFIG_SYN_COOKIES 435 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 436 __u16 *mss); 437 #else 438 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 439 struct sk_buff *skb, 440 __u16 *mss) 441 { 442 return 0; 443 } 444 #endif 445 446 extern __u32 cookie_init_timestamp(struct request_sock *req); 447 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 448 449 /* From net/ipv6/syncookies.c */ 450 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 451 #ifdef CONFIG_SYN_COOKIES 452 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 453 __u16 *mss); 454 #else 455 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 456 struct sk_buff *skb, 457 __u16 *mss) 458 { 459 return 0; 460 } 461 #endif 462 /* tcp_output.c */ 463 464 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 465 int nonagle); 466 extern int tcp_may_send_now(struct sock *sk); 467 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 468 extern void tcp_retransmit_timer(struct sock *sk); 469 extern void tcp_xmit_retransmit_queue(struct sock *); 470 extern void tcp_simple_retransmit(struct sock *); 471 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 472 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 473 474 extern void tcp_send_probe0(struct sock *); 475 extern void tcp_send_partial(struct sock *); 476 extern int tcp_write_wakeup(struct sock *); 477 extern void tcp_send_fin(struct sock *sk); 478 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 479 extern int tcp_send_synack(struct sock *); 480 extern int tcp_syn_flood_action(struct sock *sk, 481 const struct sk_buff *skb, 482 const char *proto); 483 extern void tcp_push_one(struct sock *, unsigned int mss_now); 484 extern void tcp_send_ack(struct sock *sk); 485 extern void tcp_send_delayed_ack(struct sock *sk); 486 487 /* tcp_input.c */ 488 extern void tcp_cwnd_application_limited(struct sock *sk); 489 490 /* tcp_timer.c */ 491 extern void tcp_init_xmit_timers(struct sock *); 492 static inline void tcp_clear_xmit_timers(struct sock *sk) 493 { 494 inet_csk_clear_xmit_timers(sk); 495 } 496 497 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 498 extern unsigned int tcp_current_mss(struct sock *sk); 499 500 /* Bound MSS / TSO packet size with the half of the window */ 501 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 502 { 503 int cutoff; 504 505 /* When peer uses tiny windows, there is no use in packetizing 506 * to sub-MSS pieces for the sake of SWS or making sure there 507 * are enough packets in the pipe for fast recovery. 508 * 509 * On the other hand, for extremely large MSS devices, handling 510 * smaller than MSS windows in this way does make sense. 511 */ 512 if (tp->max_window >= 512) 513 cutoff = (tp->max_window >> 1); 514 else 515 cutoff = tp->max_window; 516 517 if (cutoff && pktsize > cutoff) 518 return max_t(int, cutoff, 68U - tp->tcp_header_len); 519 else 520 return pktsize; 521 } 522 523 /* tcp.c */ 524 extern void tcp_get_info(const struct sock *, struct tcp_info *); 525 526 /* Read 'sendfile()'-style from a TCP socket */ 527 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 528 unsigned int, size_t); 529 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 530 sk_read_actor_t recv_actor); 531 532 extern void tcp_initialize_rcv_mss(struct sock *sk); 533 534 extern int tcp_mtu_to_mss(const struct sock *sk, int pmtu); 535 extern int tcp_mss_to_mtu(const struct sock *sk, int mss); 536 extern void tcp_mtup_init(struct sock *sk); 537 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt); 538 539 static inline void tcp_bound_rto(const struct sock *sk) 540 { 541 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 542 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 543 } 544 545 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 546 { 547 return (tp->srtt >> 3) + tp->rttvar; 548 } 549 550 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 551 { 552 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 553 ntohl(TCP_FLAG_ACK) | 554 snd_wnd); 555 } 556 557 static inline void tcp_fast_path_on(struct tcp_sock *tp) 558 { 559 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 560 } 561 562 static inline void tcp_fast_path_check(struct sock *sk) 563 { 564 struct tcp_sock *tp = tcp_sk(sk); 565 566 if (skb_queue_empty(&tp->out_of_order_queue) && 567 tp->rcv_wnd && 568 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 569 !tp->urg_data) 570 tcp_fast_path_on(tp); 571 } 572 573 /* Compute the actual rto_min value */ 574 static inline u32 tcp_rto_min(struct sock *sk) 575 { 576 const struct dst_entry *dst = __sk_dst_get(sk); 577 u32 rto_min = TCP_RTO_MIN; 578 579 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 580 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 581 return rto_min; 582 } 583 584 /* Compute the actual receive window we are currently advertising. 585 * Rcv_nxt can be after the window if our peer push more data 586 * than the offered window. 587 */ 588 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 589 { 590 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 591 592 if (win < 0) 593 win = 0; 594 return (u32) win; 595 } 596 597 /* Choose a new window, without checks for shrinking, and without 598 * scaling applied to the result. The caller does these things 599 * if necessary. This is a "raw" window selection. 600 */ 601 extern u32 __tcp_select_window(struct sock *sk); 602 603 /* TCP timestamps are only 32-bits, this causes a slight 604 * complication on 64-bit systems since we store a snapshot 605 * of jiffies in the buffer control blocks below. We decided 606 * to use only the low 32-bits of jiffies and hide the ugly 607 * casts with the following macro. 608 */ 609 #define tcp_time_stamp ((__u32)(jiffies)) 610 611 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 612 613 #define TCPHDR_FIN 0x01 614 #define TCPHDR_SYN 0x02 615 #define TCPHDR_RST 0x04 616 #define TCPHDR_PSH 0x08 617 #define TCPHDR_ACK 0x10 618 #define TCPHDR_URG 0x20 619 #define TCPHDR_ECE 0x40 620 #define TCPHDR_CWR 0x80 621 622 /* This is what the send packet queuing engine uses to pass 623 * TCP per-packet control information to the transmission code. 624 * We also store the host-order sequence numbers in here too. 625 * This is 44 bytes if IPV6 is enabled. 626 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 627 */ 628 struct tcp_skb_cb { 629 union { 630 struct inet_skb_parm h4; 631 #if IS_ENABLED(CONFIG_IPV6) 632 struct inet6_skb_parm h6; 633 #endif 634 } header; /* For incoming frames */ 635 __u32 seq; /* Starting sequence number */ 636 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 637 __u32 when; /* used to compute rtt's */ 638 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 639 __u8 sacked; /* State flags for SACK/FACK. */ 640 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 641 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 642 #define TCPCB_LOST 0x04 /* SKB is lost */ 643 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 644 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 645 /* 1 byte hole */ 646 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 647 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 648 649 __u32 ack_seq; /* Sequence number ACK'd */ 650 }; 651 652 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 653 654 /* Due to TSO, an SKB can be composed of multiple actual 655 * packets. To keep these tracked properly, we use this. 656 */ 657 static inline int tcp_skb_pcount(const struct sk_buff *skb) 658 { 659 return skb_shinfo(skb)->gso_segs; 660 } 661 662 /* This is valid iff tcp_skb_pcount() > 1. */ 663 static inline int tcp_skb_mss(const struct sk_buff *skb) 664 { 665 return skb_shinfo(skb)->gso_size; 666 } 667 668 /* Events passed to congestion control interface */ 669 enum tcp_ca_event { 670 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 671 CA_EVENT_CWND_RESTART, /* congestion window restart */ 672 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 673 CA_EVENT_FRTO, /* fast recovery timeout */ 674 CA_EVENT_LOSS, /* loss timeout */ 675 CA_EVENT_FAST_ACK, /* in sequence ack */ 676 CA_EVENT_SLOW_ACK, /* other ack */ 677 }; 678 679 /* 680 * Interface for adding new TCP congestion control handlers 681 */ 682 #define TCP_CA_NAME_MAX 16 683 #define TCP_CA_MAX 128 684 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 685 686 #define TCP_CONG_NON_RESTRICTED 0x1 687 #define TCP_CONG_RTT_STAMP 0x2 688 689 struct tcp_congestion_ops { 690 struct list_head list; 691 unsigned long flags; 692 693 /* initialize private data (optional) */ 694 void (*init)(struct sock *sk); 695 /* cleanup private data (optional) */ 696 void (*release)(struct sock *sk); 697 698 /* return slow start threshold (required) */ 699 u32 (*ssthresh)(struct sock *sk); 700 /* lower bound for congestion window (optional) */ 701 u32 (*min_cwnd)(const struct sock *sk); 702 /* do new cwnd calculation (required) */ 703 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 704 /* call before changing ca_state (optional) */ 705 void (*set_state)(struct sock *sk, u8 new_state); 706 /* call when cwnd event occurs (optional) */ 707 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 708 /* new value of cwnd after loss (optional) */ 709 u32 (*undo_cwnd)(struct sock *sk); 710 /* hook for packet ack accounting (optional) */ 711 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 712 /* get info for inet_diag (optional) */ 713 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 714 715 char name[TCP_CA_NAME_MAX]; 716 struct module *owner; 717 }; 718 719 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 720 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 721 722 extern void tcp_init_congestion_control(struct sock *sk); 723 extern void tcp_cleanup_congestion_control(struct sock *sk); 724 extern int tcp_set_default_congestion_control(const char *name); 725 extern void tcp_get_default_congestion_control(char *name); 726 extern void tcp_get_available_congestion_control(char *buf, size_t len); 727 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 728 extern int tcp_set_allowed_congestion_control(char *allowed); 729 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 730 extern void tcp_slow_start(struct tcp_sock *tp); 731 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 732 733 extern struct tcp_congestion_ops tcp_init_congestion_ops; 734 extern u32 tcp_reno_ssthresh(struct sock *sk); 735 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 736 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 737 extern struct tcp_congestion_ops tcp_reno; 738 739 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 740 { 741 struct inet_connection_sock *icsk = inet_csk(sk); 742 743 if (icsk->icsk_ca_ops->set_state) 744 icsk->icsk_ca_ops->set_state(sk, ca_state); 745 icsk->icsk_ca_state = ca_state; 746 } 747 748 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 749 { 750 const struct inet_connection_sock *icsk = inet_csk(sk); 751 752 if (icsk->icsk_ca_ops->cwnd_event) 753 icsk->icsk_ca_ops->cwnd_event(sk, event); 754 } 755 756 /* These functions determine how the current flow behaves in respect of SACK 757 * handling. SACK is negotiated with the peer, and therefore it can vary 758 * between different flows. 759 * 760 * tcp_is_sack - SACK enabled 761 * tcp_is_reno - No SACK 762 * tcp_is_fack - FACK enabled, implies SACK enabled 763 */ 764 static inline int tcp_is_sack(const struct tcp_sock *tp) 765 { 766 return tp->rx_opt.sack_ok; 767 } 768 769 static inline int tcp_is_reno(const struct tcp_sock *tp) 770 { 771 return !tcp_is_sack(tp); 772 } 773 774 static inline int tcp_is_fack(const struct tcp_sock *tp) 775 { 776 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 777 } 778 779 static inline void tcp_enable_fack(struct tcp_sock *tp) 780 { 781 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 782 } 783 784 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 785 { 786 return tp->sacked_out + tp->lost_out; 787 } 788 789 /* This determines how many packets are "in the network" to the best 790 * of our knowledge. In many cases it is conservative, but where 791 * detailed information is available from the receiver (via SACK 792 * blocks etc.) we can make more aggressive calculations. 793 * 794 * Use this for decisions involving congestion control, use just 795 * tp->packets_out to determine if the send queue is empty or not. 796 * 797 * Read this equation as: 798 * 799 * "Packets sent once on transmission queue" MINUS 800 * "Packets left network, but not honestly ACKed yet" PLUS 801 * "Packets fast retransmitted" 802 */ 803 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 804 { 805 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 806 } 807 808 #define TCP_INFINITE_SSTHRESH 0x7fffffff 809 810 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 811 { 812 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 813 } 814 815 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 816 * The exception is rate halving phase, when cwnd is decreasing towards 817 * ssthresh. 818 */ 819 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 820 { 821 const struct tcp_sock *tp = tcp_sk(sk); 822 823 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 824 return tp->snd_ssthresh; 825 else 826 return max(tp->snd_ssthresh, 827 ((tp->snd_cwnd >> 1) + 828 (tp->snd_cwnd >> 2))); 829 } 830 831 /* Use define here intentionally to get WARN_ON location shown at the caller */ 832 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 833 834 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 835 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 836 837 /* The maximum number of MSS of available cwnd for which TSO defers 838 * sending if not using sysctl_tcp_tso_win_divisor. 839 */ 840 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 841 { 842 return 3; 843 } 844 845 /* Slow start with delack produces 3 packets of burst, so that 846 * it is safe "de facto". This will be the default - same as 847 * the default reordering threshold - but if reordering increases, 848 * we must be able to allow cwnd to burst at least this much in order 849 * to not pull it back when holes are filled. 850 */ 851 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 852 { 853 return tp->reordering; 854 } 855 856 /* Returns end sequence number of the receiver's advertised window */ 857 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 858 { 859 return tp->snd_una + tp->snd_wnd; 860 } 861 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 862 863 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 864 const struct sk_buff *skb) 865 { 866 if (skb->len < mss) 867 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 868 } 869 870 static inline void tcp_check_probe_timer(struct sock *sk) 871 { 872 const struct tcp_sock *tp = tcp_sk(sk); 873 const struct inet_connection_sock *icsk = inet_csk(sk); 874 875 if (!tp->packets_out && !icsk->icsk_pending) 876 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 877 icsk->icsk_rto, TCP_RTO_MAX); 878 } 879 880 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 881 { 882 tp->snd_wl1 = seq; 883 } 884 885 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 886 { 887 tp->snd_wl1 = seq; 888 } 889 890 /* 891 * Calculate(/check) TCP checksum 892 */ 893 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 894 __be32 daddr, __wsum base) 895 { 896 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 897 } 898 899 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 900 { 901 return __skb_checksum_complete(skb); 902 } 903 904 static inline int tcp_checksum_complete(struct sk_buff *skb) 905 { 906 return !skb_csum_unnecessary(skb) && 907 __tcp_checksum_complete(skb); 908 } 909 910 /* Prequeue for VJ style copy to user, combined with checksumming. */ 911 912 static inline void tcp_prequeue_init(struct tcp_sock *tp) 913 { 914 tp->ucopy.task = NULL; 915 tp->ucopy.len = 0; 916 tp->ucopy.memory = 0; 917 skb_queue_head_init(&tp->ucopy.prequeue); 918 #ifdef CONFIG_NET_DMA 919 tp->ucopy.dma_chan = NULL; 920 tp->ucopy.wakeup = 0; 921 tp->ucopy.pinned_list = NULL; 922 tp->ucopy.dma_cookie = 0; 923 #endif 924 } 925 926 /* Packet is added to VJ-style prequeue for processing in process 927 * context, if a reader task is waiting. Apparently, this exciting 928 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 929 * failed somewhere. Latency? Burstiness? Well, at least now we will 930 * see, why it failed. 8)8) --ANK 931 * 932 * NOTE: is this not too big to inline? 933 */ 934 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 935 { 936 struct tcp_sock *tp = tcp_sk(sk); 937 938 if (sysctl_tcp_low_latency || !tp->ucopy.task) 939 return 0; 940 941 __skb_queue_tail(&tp->ucopy.prequeue, skb); 942 tp->ucopy.memory += skb->truesize; 943 if (tp->ucopy.memory > sk->sk_rcvbuf) { 944 struct sk_buff *skb1; 945 946 BUG_ON(sock_owned_by_user(sk)); 947 948 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 949 sk_backlog_rcv(sk, skb1); 950 NET_INC_STATS_BH(sock_net(sk), 951 LINUX_MIB_TCPPREQUEUEDROPPED); 952 } 953 954 tp->ucopy.memory = 0; 955 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 956 wake_up_interruptible_sync_poll(sk_sleep(sk), 957 POLLIN | POLLRDNORM | POLLRDBAND); 958 if (!inet_csk_ack_scheduled(sk)) 959 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 960 (3 * tcp_rto_min(sk)) / 4, 961 TCP_RTO_MAX); 962 } 963 return 1; 964 } 965 966 967 #undef STATE_TRACE 968 969 #ifdef STATE_TRACE 970 static const char *statename[]={ 971 "Unused","Established","Syn Sent","Syn Recv", 972 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 973 "Close Wait","Last ACK","Listen","Closing" 974 }; 975 #endif 976 extern void tcp_set_state(struct sock *sk, int state); 977 978 extern void tcp_done(struct sock *sk); 979 980 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 981 { 982 rx_opt->dsack = 0; 983 rx_opt->num_sacks = 0; 984 } 985 986 /* Determine a window scaling and initial window to offer. */ 987 extern void tcp_select_initial_window(int __space, __u32 mss, 988 __u32 *rcv_wnd, __u32 *window_clamp, 989 int wscale_ok, __u8 *rcv_wscale, 990 __u32 init_rcv_wnd); 991 992 static inline int tcp_win_from_space(int space) 993 { 994 return sysctl_tcp_adv_win_scale<=0 ? 995 (space>>(-sysctl_tcp_adv_win_scale)) : 996 space - (space>>sysctl_tcp_adv_win_scale); 997 } 998 999 /* Note: caller must be prepared to deal with negative returns */ 1000 static inline int tcp_space(const struct sock *sk) 1001 { 1002 return tcp_win_from_space(sk->sk_rcvbuf - 1003 atomic_read(&sk->sk_rmem_alloc)); 1004 } 1005 1006 static inline int tcp_full_space(const struct sock *sk) 1007 { 1008 return tcp_win_from_space(sk->sk_rcvbuf); 1009 } 1010 1011 static inline void tcp_openreq_init(struct request_sock *req, 1012 struct tcp_options_received *rx_opt, 1013 struct sk_buff *skb) 1014 { 1015 struct inet_request_sock *ireq = inet_rsk(req); 1016 1017 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1018 req->cookie_ts = 0; 1019 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1020 req->mss = rx_opt->mss_clamp; 1021 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1022 ireq->tstamp_ok = rx_opt->tstamp_ok; 1023 ireq->sack_ok = rx_opt->sack_ok; 1024 ireq->snd_wscale = rx_opt->snd_wscale; 1025 ireq->wscale_ok = rx_opt->wscale_ok; 1026 ireq->acked = 0; 1027 ireq->ecn_ok = 0; 1028 ireq->rmt_port = tcp_hdr(skb)->source; 1029 ireq->loc_port = tcp_hdr(skb)->dest; 1030 } 1031 1032 extern void tcp_enter_memory_pressure(struct sock *sk); 1033 1034 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1035 { 1036 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1037 } 1038 1039 static inline int keepalive_time_when(const struct tcp_sock *tp) 1040 { 1041 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1042 } 1043 1044 static inline int keepalive_probes(const struct tcp_sock *tp) 1045 { 1046 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1047 } 1048 1049 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1050 { 1051 const struct inet_connection_sock *icsk = &tp->inet_conn; 1052 1053 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1054 tcp_time_stamp - tp->rcv_tstamp); 1055 } 1056 1057 static inline int tcp_fin_time(const struct sock *sk) 1058 { 1059 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1060 const int rto = inet_csk(sk)->icsk_rto; 1061 1062 if (fin_timeout < (rto << 2) - (rto >> 1)) 1063 fin_timeout = (rto << 2) - (rto >> 1); 1064 1065 return fin_timeout; 1066 } 1067 1068 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1069 int paws_win) 1070 { 1071 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1072 return 1; 1073 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1074 return 1; 1075 /* 1076 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1077 * then following tcp messages have valid values. Ignore 0 value, 1078 * or else 'negative' tsval might forbid us to accept their packets. 1079 */ 1080 if (!rx_opt->ts_recent) 1081 return 1; 1082 return 0; 1083 } 1084 1085 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1086 int rst) 1087 { 1088 if (tcp_paws_check(rx_opt, 0)) 1089 return 0; 1090 1091 /* RST segments are not recommended to carry timestamp, 1092 and, if they do, it is recommended to ignore PAWS because 1093 "their cleanup function should take precedence over timestamps." 1094 Certainly, it is mistake. It is necessary to understand the reasons 1095 of this constraint to relax it: if peer reboots, clock may go 1096 out-of-sync and half-open connections will not be reset. 1097 Actually, the problem would be not existing if all 1098 the implementations followed draft about maintaining clock 1099 via reboots. Linux-2.2 DOES NOT! 1100 1101 However, we can relax time bounds for RST segments to MSL. 1102 */ 1103 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1104 return 0; 1105 return 1; 1106 } 1107 1108 static inline void tcp_mib_init(struct net *net) 1109 { 1110 /* See RFC 2012 */ 1111 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1112 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1113 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1114 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1115 } 1116 1117 /* from STCP */ 1118 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1119 { 1120 tp->lost_skb_hint = NULL; 1121 tp->scoreboard_skb_hint = NULL; 1122 } 1123 1124 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1125 { 1126 tcp_clear_retrans_hints_partial(tp); 1127 tp->retransmit_skb_hint = NULL; 1128 } 1129 1130 /* MD5 Signature */ 1131 struct crypto_hash; 1132 1133 /* - key database */ 1134 struct tcp_md5sig_key { 1135 u8 *key; 1136 u8 keylen; 1137 }; 1138 1139 struct tcp4_md5sig_key { 1140 struct tcp_md5sig_key base; 1141 __be32 addr; 1142 }; 1143 1144 struct tcp6_md5sig_key { 1145 struct tcp_md5sig_key base; 1146 #if 0 1147 u32 scope_id; /* XXX */ 1148 #endif 1149 struct in6_addr addr; 1150 }; 1151 1152 /* - sock block */ 1153 struct tcp_md5sig_info { 1154 struct tcp4_md5sig_key *keys4; 1155 #if IS_ENABLED(CONFIG_IPV6) 1156 struct tcp6_md5sig_key *keys6; 1157 u32 entries6; 1158 u32 alloced6; 1159 #endif 1160 u32 entries4; 1161 u32 alloced4; 1162 }; 1163 1164 /* - pseudo header */ 1165 struct tcp4_pseudohdr { 1166 __be32 saddr; 1167 __be32 daddr; 1168 __u8 pad; 1169 __u8 protocol; 1170 __be16 len; 1171 }; 1172 1173 struct tcp6_pseudohdr { 1174 struct in6_addr saddr; 1175 struct in6_addr daddr; 1176 __be32 len; 1177 __be32 protocol; /* including padding */ 1178 }; 1179 1180 union tcp_md5sum_block { 1181 struct tcp4_pseudohdr ip4; 1182 #if IS_ENABLED(CONFIG_IPV6) 1183 struct tcp6_pseudohdr ip6; 1184 #endif 1185 }; 1186 1187 /* - pool: digest algorithm, hash description and scratch buffer */ 1188 struct tcp_md5sig_pool { 1189 struct hash_desc md5_desc; 1190 union tcp_md5sum_block md5_blk; 1191 }; 1192 1193 /* - functions */ 1194 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1195 const struct sock *sk, 1196 const struct request_sock *req, 1197 const struct sk_buff *skb); 1198 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1199 struct sock *addr_sk); 1200 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1201 u8 newkeylen); 1202 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1203 1204 #ifdef CONFIG_TCP_MD5SIG 1205 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1206 &(struct tcp_md5sig_key) { \ 1207 .key = (twsk)->tw_md5_key, \ 1208 .keylen = (twsk)->tw_md5_keylen, \ 1209 } : NULL) 1210 #else 1211 #define tcp_twsk_md5_key(twsk) NULL 1212 #endif 1213 1214 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *); 1215 extern void tcp_free_md5sig_pool(void); 1216 1217 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1218 extern void tcp_put_md5sig_pool(void); 1219 1220 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1221 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1222 unsigned header_len); 1223 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1224 const struct tcp_md5sig_key *key); 1225 1226 /* write queue abstraction */ 1227 static inline void tcp_write_queue_purge(struct sock *sk) 1228 { 1229 struct sk_buff *skb; 1230 1231 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1232 sk_wmem_free_skb(sk, skb); 1233 sk_mem_reclaim(sk); 1234 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1235 } 1236 1237 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1238 { 1239 return skb_peek(&sk->sk_write_queue); 1240 } 1241 1242 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1243 { 1244 return skb_peek_tail(&sk->sk_write_queue); 1245 } 1246 1247 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1248 const struct sk_buff *skb) 1249 { 1250 return skb_queue_next(&sk->sk_write_queue, skb); 1251 } 1252 1253 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1254 const struct sk_buff *skb) 1255 { 1256 return skb_queue_prev(&sk->sk_write_queue, skb); 1257 } 1258 1259 #define tcp_for_write_queue(skb, sk) \ 1260 skb_queue_walk(&(sk)->sk_write_queue, skb) 1261 1262 #define tcp_for_write_queue_from(skb, sk) \ 1263 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1264 1265 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1266 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1267 1268 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1269 { 1270 return sk->sk_send_head; 1271 } 1272 1273 static inline bool tcp_skb_is_last(const struct sock *sk, 1274 const struct sk_buff *skb) 1275 { 1276 return skb_queue_is_last(&sk->sk_write_queue, skb); 1277 } 1278 1279 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1280 { 1281 if (tcp_skb_is_last(sk, skb)) 1282 sk->sk_send_head = NULL; 1283 else 1284 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1285 } 1286 1287 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1288 { 1289 if (sk->sk_send_head == skb_unlinked) 1290 sk->sk_send_head = NULL; 1291 } 1292 1293 static inline void tcp_init_send_head(struct sock *sk) 1294 { 1295 sk->sk_send_head = NULL; 1296 } 1297 1298 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1299 { 1300 __skb_queue_tail(&sk->sk_write_queue, skb); 1301 } 1302 1303 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1304 { 1305 __tcp_add_write_queue_tail(sk, skb); 1306 1307 /* Queue it, remembering where we must start sending. */ 1308 if (sk->sk_send_head == NULL) { 1309 sk->sk_send_head = skb; 1310 1311 if (tcp_sk(sk)->highest_sack == NULL) 1312 tcp_sk(sk)->highest_sack = skb; 1313 } 1314 } 1315 1316 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1317 { 1318 __skb_queue_head(&sk->sk_write_queue, skb); 1319 } 1320 1321 /* Insert buff after skb on the write queue of sk. */ 1322 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1323 struct sk_buff *buff, 1324 struct sock *sk) 1325 { 1326 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1327 } 1328 1329 /* Insert new before skb on the write queue of sk. */ 1330 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1331 struct sk_buff *skb, 1332 struct sock *sk) 1333 { 1334 __skb_queue_before(&sk->sk_write_queue, skb, new); 1335 1336 if (sk->sk_send_head == skb) 1337 sk->sk_send_head = new; 1338 } 1339 1340 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1341 { 1342 __skb_unlink(skb, &sk->sk_write_queue); 1343 } 1344 1345 static inline int tcp_write_queue_empty(struct sock *sk) 1346 { 1347 return skb_queue_empty(&sk->sk_write_queue); 1348 } 1349 1350 static inline void tcp_push_pending_frames(struct sock *sk) 1351 { 1352 if (tcp_send_head(sk)) { 1353 struct tcp_sock *tp = tcp_sk(sk); 1354 1355 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1356 } 1357 } 1358 1359 /* Start sequence of the highest skb with SACKed bit, valid only if 1360 * sacked > 0 or when the caller has ensured validity by itself. 1361 */ 1362 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1363 { 1364 if (!tp->sacked_out) 1365 return tp->snd_una; 1366 1367 if (tp->highest_sack == NULL) 1368 return tp->snd_nxt; 1369 1370 return TCP_SKB_CB(tp->highest_sack)->seq; 1371 } 1372 1373 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1374 { 1375 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1376 tcp_write_queue_next(sk, skb); 1377 } 1378 1379 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1380 { 1381 return tcp_sk(sk)->highest_sack; 1382 } 1383 1384 static inline void tcp_highest_sack_reset(struct sock *sk) 1385 { 1386 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1387 } 1388 1389 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1390 static inline void tcp_highest_sack_combine(struct sock *sk, 1391 struct sk_buff *old, 1392 struct sk_buff *new) 1393 { 1394 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1395 tcp_sk(sk)->highest_sack = new; 1396 } 1397 1398 /* Determines whether this is a thin stream (which may suffer from 1399 * increased latency). Used to trigger latency-reducing mechanisms. 1400 */ 1401 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1402 { 1403 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1404 } 1405 1406 /* /proc */ 1407 enum tcp_seq_states { 1408 TCP_SEQ_STATE_LISTENING, 1409 TCP_SEQ_STATE_OPENREQ, 1410 TCP_SEQ_STATE_ESTABLISHED, 1411 TCP_SEQ_STATE_TIME_WAIT, 1412 }; 1413 1414 int tcp_seq_open(struct inode *inode, struct file *file); 1415 1416 struct tcp_seq_afinfo { 1417 char *name; 1418 sa_family_t family; 1419 const struct file_operations *seq_fops; 1420 struct seq_operations seq_ops; 1421 }; 1422 1423 struct tcp_iter_state { 1424 struct seq_net_private p; 1425 sa_family_t family; 1426 enum tcp_seq_states state; 1427 struct sock *syn_wait_sk; 1428 int bucket, offset, sbucket, num, uid; 1429 loff_t last_pos; 1430 }; 1431 1432 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1433 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1434 1435 extern struct request_sock_ops tcp_request_sock_ops; 1436 extern struct request_sock_ops tcp6_request_sock_ops; 1437 1438 extern void tcp_v4_destroy_sock(struct sock *sk); 1439 1440 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1441 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, 1442 netdev_features_t features); 1443 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1444 struct sk_buff *skb); 1445 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1446 struct sk_buff *skb); 1447 extern int tcp_gro_complete(struct sk_buff *skb); 1448 extern int tcp4_gro_complete(struct sk_buff *skb); 1449 1450 #ifdef CONFIG_PROC_FS 1451 extern int tcp4_proc_init(void); 1452 extern void tcp4_proc_exit(void); 1453 #endif 1454 1455 /* TCP af-specific functions */ 1456 struct tcp_sock_af_ops { 1457 #ifdef CONFIG_TCP_MD5SIG 1458 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1459 struct sock *addr_sk); 1460 int (*calc_md5_hash) (char *location, 1461 struct tcp_md5sig_key *md5, 1462 const struct sock *sk, 1463 const struct request_sock *req, 1464 const struct sk_buff *skb); 1465 int (*md5_add) (struct sock *sk, 1466 struct sock *addr_sk, 1467 u8 *newkey, 1468 u8 len); 1469 int (*md5_parse) (struct sock *sk, 1470 char __user *optval, 1471 int optlen); 1472 #endif 1473 }; 1474 1475 struct tcp_request_sock_ops { 1476 #ifdef CONFIG_TCP_MD5SIG 1477 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1478 struct request_sock *req); 1479 int (*calc_md5_hash) (char *location, 1480 struct tcp_md5sig_key *md5, 1481 const struct sock *sk, 1482 const struct request_sock *req, 1483 const struct sk_buff *skb); 1484 #endif 1485 }; 1486 1487 /* Using SHA1 for now, define some constants. 1488 */ 1489 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1490 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1491 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1492 1493 extern int tcp_cookie_generator(u32 *bakery); 1494 1495 /** 1496 * struct tcp_cookie_values - each socket needs extra space for the 1497 * cookies, together with (optional) space for any SYN data. 1498 * 1499 * A tcp_sock contains a pointer to the current value, and this is 1500 * cloned to the tcp_timewait_sock. 1501 * 1502 * @cookie_pair: variable data from the option exchange. 1503 * 1504 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1505 * indicates default (sysctl_tcp_cookie_size). 1506 * After cookie sent, remembers size of cookie. 1507 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1508 * 1509 * @s_data_desired: user specified tcpct_s_data_desired. When the 1510 * constant payload is specified (@s_data_constant), 1511 * holds its length instead. 1512 * Range 0 to TCP_MSS_DESIRED. 1513 * 1514 * @s_data_payload: constant data that is to be included in the 1515 * payload of SYN or SYNACK segments when the 1516 * cookie option is present. 1517 */ 1518 struct tcp_cookie_values { 1519 struct kref kref; 1520 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1521 u8 cookie_pair_size; 1522 u8 cookie_desired; 1523 u16 s_data_desired:11, 1524 s_data_constant:1, 1525 s_data_in:1, 1526 s_data_out:1, 1527 s_data_unused:2; 1528 u8 s_data_payload[0]; 1529 }; 1530 1531 static inline void tcp_cookie_values_release(struct kref *kref) 1532 { 1533 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1534 } 1535 1536 /* The length of constant payload data. Note that s_data_desired is 1537 * overloaded, depending on s_data_constant: either the length of constant 1538 * data (returned here) or the limit on variable data. 1539 */ 1540 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1541 { 1542 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1543 ? tp->cookie_values->s_data_desired 1544 : 0; 1545 } 1546 1547 /** 1548 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1549 * 1550 * As tcp_request_sock has already been extended in other places, the 1551 * only remaining method is to pass stack values along as function 1552 * parameters. These parameters are not needed after sending SYNACK. 1553 * 1554 * @cookie_bakery: cryptographic secret and message workspace. 1555 * 1556 * @cookie_plus: bytes in authenticator/cookie option, copied from 1557 * struct tcp_options_received (above). 1558 */ 1559 struct tcp_extend_values { 1560 struct request_values rv; 1561 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1562 u8 cookie_plus:6, 1563 cookie_out_never:1, 1564 cookie_in_always:1; 1565 }; 1566 1567 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1568 { 1569 return (struct tcp_extend_values *)rvp; 1570 } 1571 1572 extern void tcp_v4_init(void); 1573 extern void tcp_init(void); 1574 1575 #endif /* _TCP_H */ 1576