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/bug.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/crypto.h> 31 #include <linux/cryptohash.h> 32 #include <linux/kref.h> 33 #include <linux/ktime.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 512 69 70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 71 #define TCP_FASTRETRANS_THRESH 3 72 73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 74 #define TCP_MAX_QUICKACKS 16U 75 76 /* urg_data states */ 77 #define TCP_URG_VALID 0x0100 78 #define TCP_URG_NOTYET 0x0200 79 #define TCP_URG_READ 0x0400 80 81 #define TCP_RETR1 3 /* 82 * This is how many retries it does before it 83 * tries to figure out if the gateway is 84 * down. Minimal RFC value is 3; it corresponds 85 * to ~3sec-8min depending on RTO. 86 */ 87 88 #define TCP_RETR2 15 /* 89 * This should take at least 90 * 90 minutes to time out. 91 * RFC1122 says that the limit is 100 sec. 92 * 15 is ~13-30min depending on RTO. 93 */ 94 95 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 96 * when active opening a connection. 97 * RFC1122 says the minimum retry MUST 98 * be at least 180secs. Nevertheless 99 * this value is corresponding to 100 * 63secs of retransmission with the 101 * current initial RTO. 102 */ 103 104 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 105 * when passive opening a connection. 106 * This is corresponding to 31secs of 107 * retransmission with the current 108 * initial RTO. 109 */ 110 111 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 112 * state, about 60 seconds */ 113 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 114 /* BSD style FIN_WAIT2 deadlock breaker. 115 * It used to be 3min, new value is 60sec, 116 * to combine FIN-WAIT-2 timeout with 117 * TIME-WAIT timer. 118 */ 119 120 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 121 #if HZ >= 100 122 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 123 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 124 #else 125 #define TCP_DELACK_MIN 4U 126 #define TCP_ATO_MIN 4U 127 #endif 128 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 129 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 130 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 131 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 132 * used as a fallback RTO for the 133 * initial data transmission if no 134 * valid RTT sample has been acquired, 135 * most likely due to retrans in 3WHS. 136 */ 137 138 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 139 * for local resources. 140 */ 141 142 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 143 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 144 #define TCP_KEEPALIVE_INTVL (75*HZ) 145 146 #define MAX_TCP_KEEPIDLE 32767 147 #define MAX_TCP_KEEPINTVL 32767 148 #define MAX_TCP_KEEPCNT 127 149 #define MAX_TCP_SYNCNT 127 150 151 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 152 153 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 154 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 155 * after this time. It should be equal 156 * (or greater than) TCP_TIMEWAIT_LEN 157 * to provide reliability equal to one 158 * provided by timewait state. 159 */ 160 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 161 * timestamps. It must be less than 162 * minimal timewait lifetime. 163 */ 164 /* 165 * TCP option 166 */ 167 168 #define TCPOPT_NOP 1 /* Padding */ 169 #define TCPOPT_EOL 0 /* End of options */ 170 #define TCPOPT_MSS 2 /* Segment size negotiating */ 171 #define TCPOPT_WINDOW 3 /* Window scaling */ 172 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 173 #define TCPOPT_SACK 5 /* SACK Block */ 174 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 175 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 176 #define TCPOPT_EXP 254 /* Experimental */ 177 /* Magic number to be after the option value for sharing TCP 178 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 179 */ 180 #define TCPOPT_FASTOPEN_MAGIC 0xF989 181 182 /* 183 * TCP option lengths 184 */ 185 186 #define TCPOLEN_MSS 4 187 #define TCPOLEN_WINDOW 3 188 #define TCPOLEN_SACK_PERM 2 189 #define TCPOLEN_TIMESTAMP 10 190 #define TCPOLEN_MD5SIG 18 191 #define TCPOLEN_EXP_FASTOPEN_BASE 4 192 193 /* But this is what stacks really send out. */ 194 #define TCPOLEN_TSTAMP_ALIGNED 12 195 #define TCPOLEN_WSCALE_ALIGNED 4 196 #define TCPOLEN_SACKPERM_ALIGNED 4 197 #define TCPOLEN_SACK_BASE 2 198 #define TCPOLEN_SACK_BASE_ALIGNED 4 199 #define TCPOLEN_SACK_PERBLOCK 8 200 #define TCPOLEN_MD5SIG_ALIGNED 20 201 #define TCPOLEN_MSS_ALIGNED 4 202 203 /* Flags in tp->nonagle */ 204 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 205 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 206 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 207 208 /* TCP thin-stream limits */ 209 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 210 211 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 212 #define TCP_INIT_CWND 10 213 214 /* Bit Flags for sysctl_tcp_fastopen */ 215 #define TFO_CLIENT_ENABLE 1 216 #define TFO_SERVER_ENABLE 2 217 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 218 219 /* Accept SYN data w/o any cookie option */ 220 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 221 222 /* Force enable TFO on all listeners, i.e., not requiring the 223 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 224 */ 225 #define TFO_SERVER_WO_SOCKOPT1 0x400 226 #define TFO_SERVER_WO_SOCKOPT2 0x800 227 228 extern struct inet_timewait_death_row tcp_death_row; 229 230 /* sysctl variables for tcp */ 231 extern int sysctl_tcp_timestamps; 232 extern int sysctl_tcp_window_scaling; 233 extern int sysctl_tcp_sack; 234 extern int sysctl_tcp_fin_timeout; 235 extern int sysctl_tcp_keepalive_time; 236 extern int sysctl_tcp_keepalive_probes; 237 extern int sysctl_tcp_keepalive_intvl; 238 extern int sysctl_tcp_syn_retries; 239 extern int sysctl_tcp_synack_retries; 240 extern int sysctl_tcp_retries1; 241 extern int sysctl_tcp_retries2; 242 extern int sysctl_tcp_orphan_retries; 243 extern int sysctl_tcp_syncookies; 244 extern int sysctl_tcp_fastopen; 245 extern int sysctl_tcp_retrans_collapse; 246 extern int sysctl_tcp_stdurg; 247 extern int sysctl_tcp_rfc1337; 248 extern int sysctl_tcp_abort_on_overflow; 249 extern int sysctl_tcp_max_orphans; 250 extern int sysctl_tcp_fack; 251 extern int sysctl_tcp_reordering; 252 extern int sysctl_tcp_max_reordering; 253 extern int sysctl_tcp_dsack; 254 extern long sysctl_tcp_mem[3]; 255 extern int sysctl_tcp_wmem[3]; 256 extern int sysctl_tcp_rmem[3]; 257 extern int sysctl_tcp_app_win; 258 extern int sysctl_tcp_adv_win_scale; 259 extern int sysctl_tcp_tw_reuse; 260 extern int sysctl_tcp_frto; 261 extern int sysctl_tcp_low_latency; 262 extern int sysctl_tcp_nometrics_save; 263 extern int sysctl_tcp_moderate_rcvbuf; 264 extern int sysctl_tcp_tso_win_divisor; 265 extern int sysctl_tcp_workaround_signed_windows; 266 extern int sysctl_tcp_slow_start_after_idle; 267 extern int sysctl_tcp_thin_linear_timeouts; 268 extern int sysctl_tcp_thin_dupack; 269 extern int sysctl_tcp_early_retrans; 270 extern int sysctl_tcp_limit_output_bytes; 271 extern int sysctl_tcp_challenge_ack_limit; 272 extern unsigned int sysctl_tcp_notsent_lowat; 273 extern int sysctl_tcp_min_tso_segs; 274 extern int sysctl_tcp_autocorking; 275 extern int sysctl_tcp_invalid_ratelimit; 276 277 extern atomic_long_t tcp_memory_allocated; 278 extern struct percpu_counter tcp_sockets_allocated; 279 extern int tcp_memory_pressure; 280 281 /* 282 * The next routines deal with comparing 32 bit unsigned ints 283 * and worry about wraparound (automatic with unsigned arithmetic). 284 */ 285 286 static inline bool before(__u32 seq1, __u32 seq2) 287 { 288 return (__s32)(seq1-seq2) < 0; 289 } 290 #define after(seq2, seq1) before(seq1, seq2) 291 292 /* is s2<=s1<=s3 ? */ 293 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 294 { 295 return seq3 - seq2 >= seq1 - seq2; 296 } 297 298 static inline bool tcp_out_of_memory(struct sock *sk) 299 { 300 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 301 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 302 return true; 303 return false; 304 } 305 306 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 307 { 308 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 309 int orphans = percpu_counter_read_positive(ocp); 310 311 if (orphans << shift > sysctl_tcp_max_orphans) { 312 orphans = percpu_counter_sum_positive(ocp); 313 if (orphans << shift > sysctl_tcp_max_orphans) 314 return true; 315 } 316 return false; 317 } 318 319 bool tcp_check_oom(struct sock *sk, int shift); 320 321 /* syncookies: remember time of last synqueue overflow */ 322 static inline void tcp_synq_overflow(struct sock *sk) 323 { 324 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 325 } 326 327 /* syncookies: no recent synqueue overflow on this listening socket? */ 328 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 329 { 330 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 331 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 332 } 333 334 extern struct proto tcp_prot; 335 336 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 337 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 338 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 339 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 340 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 341 342 void tcp_tasklet_init(void); 343 344 void tcp_v4_err(struct sk_buff *skb, u32); 345 346 void tcp_shutdown(struct sock *sk, int how); 347 348 void tcp_v4_early_demux(struct sk_buff *skb); 349 int tcp_v4_rcv(struct sk_buff *skb); 350 351 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 352 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 353 size_t size); 354 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 355 int flags); 356 void tcp_release_cb(struct sock *sk); 357 void tcp_wfree(struct sk_buff *skb); 358 void tcp_write_timer_handler(struct sock *sk); 359 void tcp_delack_timer_handler(struct sock *sk); 360 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 361 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 362 const struct tcphdr *th, unsigned int len); 363 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 364 const struct tcphdr *th, unsigned int len); 365 void tcp_rcv_space_adjust(struct sock *sk); 366 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 367 void tcp_twsk_destructor(struct sock *sk); 368 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 369 struct pipe_inode_info *pipe, size_t len, 370 unsigned int flags); 371 372 static inline void tcp_dec_quickack_mode(struct sock *sk, 373 const unsigned int pkts) 374 { 375 struct inet_connection_sock *icsk = inet_csk(sk); 376 377 if (icsk->icsk_ack.quick) { 378 if (pkts >= icsk->icsk_ack.quick) { 379 icsk->icsk_ack.quick = 0; 380 /* Leaving quickack mode we deflate ATO. */ 381 icsk->icsk_ack.ato = TCP_ATO_MIN; 382 } else 383 icsk->icsk_ack.quick -= pkts; 384 } 385 } 386 387 #define TCP_ECN_OK 1 388 #define TCP_ECN_QUEUE_CWR 2 389 #define TCP_ECN_DEMAND_CWR 4 390 #define TCP_ECN_SEEN 8 391 392 enum tcp_tw_status { 393 TCP_TW_SUCCESS = 0, 394 TCP_TW_RST = 1, 395 TCP_TW_ACK = 2, 396 TCP_TW_SYN = 3 397 }; 398 399 400 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 401 struct sk_buff *skb, 402 const struct tcphdr *th); 403 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 404 struct request_sock *req, struct request_sock **prev, 405 bool fastopen); 406 int tcp_child_process(struct sock *parent, struct sock *child, 407 struct sk_buff *skb); 408 void tcp_enter_loss(struct sock *sk); 409 void tcp_clear_retrans(struct tcp_sock *tp); 410 void tcp_update_metrics(struct sock *sk); 411 void tcp_init_metrics(struct sock *sk); 412 void tcp_metrics_init(void); 413 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 414 bool paws_check, bool timestamps); 415 bool tcp_remember_stamp(struct sock *sk); 416 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 417 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 418 void tcp_disable_fack(struct tcp_sock *tp); 419 void tcp_close(struct sock *sk, long timeout); 420 void tcp_init_sock(struct sock *sk); 421 unsigned int tcp_poll(struct file *file, struct socket *sock, 422 struct poll_table_struct *wait); 423 int tcp_getsockopt(struct sock *sk, int level, int optname, 424 char __user *optval, int __user *optlen); 425 int tcp_setsockopt(struct sock *sk, int level, int optname, 426 char __user *optval, unsigned int optlen); 427 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 428 char __user *optval, int __user *optlen); 429 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 430 char __user *optval, unsigned int optlen); 431 void tcp_set_keepalive(struct sock *sk, int val); 432 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 433 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 434 size_t len, int nonblock, int flags, int *addr_len); 435 void tcp_parse_options(const struct sk_buff *skb, 436 struct tcp_options_received *opt_rx, 437 int estab, struct tcp_fastopen_cookie *foc); 438 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 439 440 /* 441 * TCP v4 functions exported for the inet6 API 442 */ 443 444 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 445 void tcp_v4_mtu_reduced(struct sock *sk); 446 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 447 struct sock *tcp_create_openreq_child(struct sock *sk, 448 struct request_sock *req, 449 struct sk_buff *skb); 450 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 451 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 452 struct request_sock *req, 453 struct dst_entry *dst); 454 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 455 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 456 int tcp_connect(struct sock *sk); 457 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 458 struct request_sock *req, 459 struct tcp_fastopen_cookie *foc); 460 int tcp_disconnect(struct sock *sk, int flags); 461 462 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 463 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 464 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 465 466 /* From syncookies.c */ 467 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 468 u32 cookie); 469 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 470 #ifdef CONFIG_SYN_COOKIES 471 472 /* Syncookies use a monotonic timer which increments every 60 seconds. 473 * This counter is used both as a hash input and partially encoded into 474 * the cookie value. A cookie is only validated further if the delta 475 * between the current counter value and the encoded one is less than this, 476 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 477 * the counter advances immediately after a cookie is generated). 478 */ 479 #define MAX_SYNCOOKIE_AGE 2 480 481 static inline u32 tcp_cookie_time(void) 482 { 483 u64 val = get_jiffies_64(); 484 485 do_div(val, 60 * HZ); 486 return val; 487 } 488 489 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 490 u16 *mssp); 491 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 492 __u16 *mss); 493 __u32 cookie_init_timestamp(struct request_sock *req); 494 bool cookie_timestamp_decode(struct tcp_options_received *opt); 495 bool cookie_ecn_ok(const struct tcp_options_received *opt, 496 const struct net *net, const struct dst_entry *dst); 497 498 /* From net/ipv6/syncookies.c */ 499 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 500 u32 cookie); 501 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 502 503 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 504 const struct tcphdr *th, u16 *mssp); 505 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 506 __u16 *mss); 507 #endif 508 /* tcp_output.c */ 509 510 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 511 int nonagle); 512 bool tcp_may_send_now(struct sock *sk); 513 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 514 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 515 void tcp_retransmit_timer(struct sock *sk); 516 void tcp_xmit_retransmit_queue(struct sock *); 517 void tcp_simple_retransmit(struct sock *); 518 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 519 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 520 521 void tcp_send_probe0(struct sock *); 522 void tcp_send_partial(struct sock *); 523 int tcp_write_wakeup(struct sock *); 524 void tcp_send_fin(struct sock *sk); 525 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 526 int tcp_send_synack(struct sock *); 527 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 528 const char *proto); 529 void tcp_push_one(struct sock *, unsigned int mss_now); 530 void tcp_send_ack(struct sock *sk); 531 void tcp_send_delayed_ack(struct sock *sk); 532 void tcp_send_loss_probe(struct sock *sk); 533 bool tcp_schedule_loss_probe(struct sock *sk); 534 535 /* tcp_input.c */ 536 void tcp_resume_early_retransmit(struct sock *sk); 537 void tcp_rearm_rto(struct sock *sk); 538 void tcp_reset(struct sock *sk); 539 540 /* tcp_timer.c */ 541 void tcp_init_xmit_timers(struct sock *); 542 static inline void tcp_clear_xmit_timers(struct sock *sk) 543 { 544 inet_csk_clear_xmit_timers(sk); 545 } 546 547 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 548 unsigned int tcp_current_mss(struct sock *sk); 549 550 /* Bound MSS / TSO packet size with the half of the window */ 551 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 552 { 553 int cutoff; 554 555 /* When peer uses tiny windows, there is no use in packetizing 556 * to sub-MSS pieces for the sake of SWS or making sure there 557 * are enough packets in the pipe for fast recovery. 558 * 559 * On the other hand, for extremely large MSS devices, handling 560 * smaller than MSS windows in this way does make sense. 561 */ 562 if (tp->max_window >= 512) 563 cutoff = (tp->max_window >> 1); 564 else 565 cutoff = tp->max_window; 566 567 if (cutoff && pktsize > cutoff) 568 return max_t(int, cutoff, 68U - tp->tcp_header_len); 569 else 570 return pktsize; 571 } 572 573 /* tcp.c */ 574 void tcp_get_info(const struct sock *, struct tcp_info *); 575 576 /* Read 'sendfile()'-style from a TCP socket */ 577 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 578 unsigned int, size_t); 579 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 580 sk_read_actor_t recv_actor); 581 582 void tcp_initialize_rcv_mss(struct sock *sk); 583 584 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 585 int tcp_mss_to_mtu(struct sock *sk, int mss); 586 void tcp_mtup_init(struct sock *sk); 587 void tcp_init_buffer_space(struct sock *sk); 588 589 static inline void tcp_bound_rto(const struct sock *sk) 590 { 591 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 592 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 593 } 594 595 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 596 { 597 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 598 } 599 600 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 601 { 602 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 603 ntohl(TCP_FLAG_ACK) | 604 snd_wnd); 605 } 606 607 static inline void tcp_fast_path_on(struct tcp_sock *tp) 608 { 609 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 610 } 611 612 static inline void tcp_fast_path_check(struct sock *sk) 613 { 614 struct tcp_sock *tp = tcp_sk(sk); 615 616 if (skb_queue_empty(&tp->out_of_order_queue) && 617 tp->rcv_wnd && 618 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 619 !tp->urg_data) 620 tcp_fast_path_on(tp); 621 } 622 623 /* Compute the actual rto_min value */ 624 static inline u32 tcp_rto_min(struct sock *sk) 625 { 626 const struct dst_entry *dst = __sk_dst_get(sk); 627 u32 rto_min = TCP_RTO_MIN; 628 629 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 630 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 631 return rto_min; 632 } 633 634 static inline u32 tcp_rto_min_us(struct sock *sk) 635 { 636 return jiffies_to_usecs(tcp_rto_min(sk)); 637 } 638 639 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 640 { 641 return dst_metric_locked(dst, RTAX_CC_ALGO); 642 } 643 644 /* Compute the actual receive window we are currently advertising. 645 * Rcv_nxt can be after the window if our peer push more data 646 * than the offered window. 647 */ 648 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 649 { 650 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 651 652 if (win < 0) 653 win = 0; 654 return (u32) win; 655 } 656 657 /* Choose a new window, without checks for shrinking, and without 658 * scaling applied to the result. The caller does these things 659 * if necessary. This is a "raw" window selection. 660 */ 661 u32 __tcp_select_window(struct sock *sk); 662 663 void tcp_send_window_probe(struct sock *sk); 664 665 /* TCP timestamps are only 32-bits, this causes a slight 666 * complication on 64-bit systems since we store a snapshot 667 * of jiffies in the buffer control blocks below. We decided 668 * to use only the low 32-bits of jiffies and hide the ugly 669 * casts with the following macro. 670 */ 671 #define tcp_time_stamp ((__u32)(jiffies)) 672 673 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 674 { 675 return skb->skb_mstamp.stamp_jiffies; 676 } 677 678 679 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 680 681 #define TCPHDR_FIN 0x01 682 #define TCPHDR_SYN 0x02 683 #define TCPHDR_RST 0x04 684 #define TCPHDR_PSH 0x08 685 #define TCPHDR_ACK 0x10 686 #define TCPHDR_URG 0x20 687 #define TCPHDR_ECE 0x40 688 #define TCPHDR_CWR 0x80 689 690 /* This is what the send packet queuing engine uses to pass 691 * TCP per-packet control information to the transmission code. 692 * We also store the host-order sequence numbers in here too. 693 * This is 44 bytes if IPV6 is enabled. 694 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 695 */ 696 struct tcp_skb_cb { 697 __u32 seq; /* Starting sequence number */ 698 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 699 union { 700 /* Note : tcp_tw_isn is used in input path only 701 * (isn chosen by tcp_timewait_state_process()) 702 * 703 * tcp_gso_segs is used in write queue only, 704 * cf tcp_skb_pcount() 705 */ 706 __u32 tcp_tw_isn; 707 __u32 tcp_gso_segs; 708 }; 709 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 710 711 __u8 sacked; /* State flags for SACK/FACK. */ 712 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 713 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 714 #define TCPCB_LOST 0x04 /* SKB is lost */ 715 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 716 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 717 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 718 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 719 TCPCB_REPAIRED) 720 721 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 722 /* 1 byte hole */ 723 __u32 ack_seq; /* Sequence number ACK'd */ 724 union { 725 struct inet_skb_parm h4; 726 #if IS_ENABLED(CONFIG_IPV6) 727 struct inet6_skb_parm h6; 728 #endif 729 } header; /* For incoming frames */ 730 }; 731 732 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 733 734 735 #if IS_ENABLED(CONFIG_IPV6) 736 /* This is the variant of inet6_iif() that must be used by TCP, 737 * as TCP moves IP6CB into a different location in skb->cb[] 738 */ 739 static inline int tcp_v6_iif(const struct sk_buff *skb) 740 { 741 return TCP_SKB_CB(skb)->header.h6.iif; 742 } 743 #endif 744 745 /* Due to TSO, an SKB can be composed of multiple actual 746 * packets. To keep these tracked properly, we use this. 747 */ 748 static inline int tcp_skb_pcount(const struct sk_buff *skb) 749 { 750 return TCP_SKB_CB(skb)->tcp_gso_segs; 751 } 752 753 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 754 { 755 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 756 } 757 758 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 759 { 760 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 761 } 762 763 /* This is valid iff tcp_skb_pcount() > 1. */ 764 static inline int tcp_skb_mss(const struct sk_buff *skb) 765 { 766 return skb_shinfo(skb)->gso_size; 767 } 768 769 /* Events passed to congestion control interface */ 770 enum tcp_ca_event { 771 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 772 CA_EVENT_CWND_RESTART, /* congestion window restart */ 773 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 774 CA_EVENT_LOSS, /* loss timeout */ 775 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 776 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 777 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 778 CA_EVENT_NON_DELAYED_ACK, 779 }; 780 781 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 782 enum tcp_ca_ack_event_flags { 783 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 784 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 785 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 786 }; 787 788 /* 789 * Interface for adding new TCP congestion control handlers 790 */ 791 #define TCP_CA_NAME_MAX 16 792 #define TCP_CA_MAX 128 793 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 794 795 #define TCP_CA_UNSPEC 0 796 797 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 798 #define TCP_CONG_NON_RESTRICTED 0x1 799 /* Requires ECN/ECT set on all packets */ 800 #define TCP_CONG_NEEDS_ECN 0x2 801 802 struct tcp_congestion_ops { 803 struct list_head list; 804 u32 key; 805 u32 flags; 806 807 /* initialize private data (optional) */ 808 void (*init)(struct sock *sk); 809 /* cleanup private data (optional) */ 810 void (*release)(struct sock *sk); 811 812 /* return slow start threshold (required) */ 813 u32 (*ssthresh)(struct sock *sk); 814 /* do new cwnd calculation (required) */ 815 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 816 /* call before changing ca_state (optional) */ 817 void (*set_state)(struct sock *sk, u8 new_state); 818 /* call when cwnd event occurs (optional) */ 819 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 820 /* call when ack arrives (optional) */ 821 void (*in_ack_event)(struct sock *sk, u32 flags); 822 /* new value of cwnd after loss (optional) */ 823 u32 (*undo_cwnd)(struct sock *sk); 824 /* hook for packet ack accounting (optional) */ 825 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 826 /* get info for inet_diag (optional) */ 827 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 828 829 char name[TCP_CA_NAME_MAX]; 830 struct module *owner; 831 }; 832 833 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 834 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 835 836 void tcp_assign_congestion_control(struct sock *sk); 837 void tcp_init_congestion_control(struct sock *sk); 838 void tcp_cleanup_congestion_control(struct sock *sk); 839 int tcp_set_default_congestion_control(const char *name); 840 void tcp_get_default_congestion_control(char *name); 841 void tcp_get_available_congestion_control(char *buf, size_t len); 842 void tcp_get_allowed_congestion_control(char *buf, size_t len); 843 int tcp_set_allowed_congestion_control(char *allowed); 844 int tcp_set_congestion_control(struct sock *sk, const char *name); 845 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 846 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 847 848 u32 tcp_reno_ssthresh(struct sock *sk); 849 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 850 extern struct tcp_congestion_ops tcp_reno; 851 852 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 853 u32 tcp_ca_get_key_by_name(const char *name); 854 #ifdef CONFIG_INET 855 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 856 #else 857 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 858 { 859 return NULL; 860 } 861 #endif 862 863 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 864 { 865 const struct inet_connection_sock *icsk = inet_csk(sk); 866 867 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 868 } 869 870 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 871 { 872 struct inet_connection_sock *icsk = inet_csk(sk); 873 874 if (icsk->icsk_ca_ops->set_state) 875 icsk->icsk_ca_ops->set_state(sk, ca_state); 876 icsk->icsk_ca_state = ca_state; 877 } 878 879 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 880 { 881 const struct inet_connection_sock *icsk = inet_csk(sk); 882 883 if (icsk->icsk_ca_ops->cwnd_event) 884 icsk->icsk_ca_ops->cwnd_event(sk, event); 885 } 886 887 /* These functions determine how the current flow behaves in respect of SACK 888 * handling. SACK is negotiated with the peer, and therefore it can vary 889 * between different flows. 890 * 891 * tcp_is_sack - SACK enabled 892 * tcp_is_reno - No SACK 893 * tcp_is_fack - FACK enabled, implies SACK enabled 894 */ 895 static inline int tcp_is_sack(const struct tcp_sock *tp) 896 { 897 return tp->rx_opt.sack_ok; 898 } 899 900 static inline bool tcp_is_reno(const struct tcp_sock *tp) 901 { 902 return !tcp_is_sack(tp); 903 } 904 905 static inline bool tcp_is_fack(const struct tcp_sock *tp) 906 { 907 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 908 } 909 910 static inline void tcp_enable_fack(struct tcp_sock *tp) 911 { 912 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 913 } 914 915 /* TCP early-retransmit (ER) is similar to but more conservative than 916 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 917 */ 918 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 919 { 920 tp->do_early_retrans = sysctl_tcp_early_retrans && 921 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 922 sysctl_tcp_reordering == 3; 923 } 924 925 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 926 { 927 tp->do_early_retrans = 0; 928 } 929 930 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 931 { 932 return tp->sacked_out + tp->lost_out; 933 } 934 935 /* This determines how many packets are "in the network" to the best 936 * of our knowledge. In many cases it is conservative, but where 937 * detailed information is available from the receiver (via SACK 938 * blocks etc.) we can make more aggressive calculations. 939 * 940 * Use this for decisions involving congestion control, use just 941 * tp->packets_out to determine if the send queue is empty or not. 942 * 943 * Read this equation as: 944 * 945 * "Packets sent once on transmission queue" MINUS 946 * "Packets left network, but not honestly ACKed yet" PLUS 947 * "Packets fast retransmitted" 948 */ 949 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 950 { 951 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 952 } 953 954 #define TCP_INFINITE_SSTHRESH 0x7fffffff 955 956 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 957 { 958 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 959 } 960 961 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 962 { 963 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 964 (1 << inet_csk(sk)->icsk_ca_state); 965 } 966 967 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 968 * The exception is cwnd reduction phase, when cwnd is decreasing towards 969 * ssthresh. 970 */ 971 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 972 { 973 const struct tcp_sock *tp = tcp_sk(sk); 974 975 if (tcp_in_cwnd_reduction(sk)) 976 return tp->snd_ssthresh; 977 else 978 return max(tp->snd_ssthresh, 979 ((tp->snd_cwnd >> 1) + 980 (tp->snd_cwnd >> 2))); 981 } 982 983 /* Use define here intentionally to get WARN_ON location shown at the caller */ 984 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 985 986 void tcp_enter_cwr(struct sock *sk); 987 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 988 989 /* The maximum number of MSS of available cwnd for which TSO defers 990 * sending if not using sysctl_tcp_tso_win_divisor. 991 */ 992 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 993 { 994 return 3; 995 } 996 997 /* Slow start with delack produces 3 packets of burst, so that 998 * it is safe "de facto". This will be the default - same as 999 * the default reordering threshold - but if reordering increases, 1000 * we must be able to allow cwnd to burst at least this much in order 1001 * to not pull it back when holes are filled. 1002 */ 1003 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1004 { 1005 return tp->reordering; 1006 } 1007 1008 /* Returns end sequence number of the receiver's advertised window */ 1009 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1010 { 1011 return tp->snd_una + tp->snd_wnd; 1012 } 1013 1014 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1015 * flexible approach. The RFC suggests cwnd should not be raised unless 1016 * it was fully used previously. And that's exactly what we do in 1017 * congestion avoidance mode. But in slow start we allow cwnd to grow 1018 * as long as the application has used half the cwnd. 1019 * Example : 1020 * cwnd is 10 (IW10), but application sends 9 frames. 1021 * We allow cwnd to reach 18 when all frames are ACKed. 1022 * This check is safe because it's as aggressive as slow start which already 1023 * risks 100% overshoot. The advantage is that we discourage application to 1024 * either send more filler packets or data to artificially blow up the cwnd 1025 * usage, and allow application-limited process to probe bw more aggressively. 1026 */ 1027 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1028 { 1029 const struct tcp_sock *tp = tcp_sk(sk); 1030 1031 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1032 if (tp->snd_cwnd <= tp->snd_ssthresh) 1033 return tp->snd_cwnd < 2 * tp->max_packets_out; 1034 1035 return tp->is_cwnd_limited; 1036 } 1037 1038 static inline void tcp_check_probe_timer(struct sock *sk) 1039 { 1040 const struct tcp_sock *tp = tcp_sk(sk); 1041 const struct inet_connection_sock *icsk = inet_csk(sk); 1042 1043 if (!tp->packets_out && !icsk->icsk_pending) 1044 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1045 icsk->icsk_rto, TCP_RTO_MAX); 1046 } 1047 1048 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1049 { 1050 tp->snd_wl1 = seq; 1051 } 1052 1053 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1054 { 1055 tp->snd_wl1 = seq; 1056 } 1057 1058 /* 1059 * Calculate(/check) TCP checksum 1060 */ 1061 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1062 __be32 daddr, __wsum base) 1063 { 1064 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1065 } 1066 1067 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1068 { 1069 return __skb_checksum_complete(skb); 1070 } 1071 1072 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1073 { 1074 return !skb_csum_unnecessary(skb) && 1075 __tcp_checksum_complete(skb); 1076 } 1077 1078 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1079 1080 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1081 { 1082 tp->ucopy.task = NULL; 1083 tp->ucopy.len = 0; 1084 tp->ucopy.memory = 0; 1085 skb_queue_head_init(&tp->ucopy.prequeue); 1086 } 1087 1088 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1089 1090 #undef STATE_TRACE 1091 1092 #ifdef STATE_TRACE 1093 static const char *statename[]={ 1094 "Unused","Established","Syn Sent","Syn Recv", 1095 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1096 "Close Wait","Last ACK","Listen","Closing" 1097 }; 1098 #endif 1099 void tcp_set_state(struct sock *sk, int state); 1100 1101 void tcp_done(struct sock *sk); 1102 1103 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1104 { 1105 rx_opt->dsack = 0; 1106 rx_opt->num_sacks = 0; 1107 } 1108 1109 u32 tcp_default_init_rwnd(u32 mss); 1110 1111 /* Determine a window scaling and initial window to offer. */ 1112 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1113 __u32 *window_clamp, int wscale_ok, 1114 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1115 1116 static inline int tcp_win_from_space(int space) 1117 { 1118 return sysctl_tcp_adv_win_scale<=0 ? 1119 (space>>(-sysctl_tcp_adv_win_scale)) : 1120 space - (space>>sysctl_tcp_adv_win_scale); 1121 } 1122 1123 /* Note: caller must be prepared to deal with negative returns */ 1124 static inline int tcp_space(const struct sock *sk) 1125 { 1126 return tcp_win_from_space(sk->sk_rcvbuf - 1127 atomic_read(&sk->sk_rmem_alloc)); 1128 } 1129 1130 static inline int tcp_full_space(const struct sock *sk) 1131 { 1132 return tcp_win_from_space(sk->sk_rcvbuf); 1133 } 1134 1135 static inline void tcp_openreq_init(struct request_sock *req, 1136 struct tcp_options_received *rx_opt, 1137 struct sk_buff *skb, struct sock *sk) 1138 { 1139 struct inet_request_sock *ireq = inet_rsk(req); 1140 1141 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1142 req->cookie_ts = 0; 1143 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1144 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1145 tcp_rsk(req)->snt_synack = tcp_time_stamp; 1146 tcp_rsk(req)->last_oow_ack_time = 0; 1147 req->mss = rx_opt->mss_clamp; 1148 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1149 ireq->tstamp_ok = rx_opt->tstamp_ok; 1150 ireq->sack_ok = rx_opt->sack_ok; 1151 ireq->snd_wscale = rx_opt->snd_wscale; 1152 ireq->wscale_ok = rx_opt->wscale_ok; 1153 ireq->acked = 0; 1154 ireq->ecn_ok = 0; 1155 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1156 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1157 ireq->ir_mark = inet_request_mark(sk, skb); 1158 } 1159 1160 extern void tcp_openreq_init_rwin(struct request_sock *req, 1161 struct sock *sk, struct dst_entry *dst); 1162 1163 void tcp_enter_memory_pressure(struct sock *sk); 1164 1165 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1166 { 1167 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1168 } 1169 1170 static inline int keepalive_time_when(const struct tcp_sock *tp) 1171 { 1172 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1173 } 1174 1175 static inline int keepalive_probes(const struct tcp_sock *tp) 1176 { 1177 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1178 } 1179 1180 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1181 { 1182 const struct inet_connection_sock *icsk = &tp->inet_conn; 1183 1184 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1185 tcp_time_stamp - tp->rcv_tstamp); 1186 } 1187 1188 static inline int tcp_fin_time(const struct sock *sk) 1189 { 1190 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1191 const int rto = inet_csk(sk)->icsk_rto; 1192 1193 if (fin_timeout < (rto << 2) - (rto >> 1)) 1194 fin_timeout = (rto << 2) - (rto >> 1); 1195 1196 return fin_timeout; 1197 } 1198 1199 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1200 int paws_win) 1201 { 1202 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1203 return true; 1204 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1205 return true; 1206 /* 1207 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1208 * then following tcp messages have valid values. Ignore 0 value, 1209 * or else 'negative' tsval might forbid us to accept their packets. 1210 */ 1211 if (!rx_opt->ts_recent) 1212 return true; 1213 return false; 1214 } 1215 1216 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1217 int rst) 1218 { 1219 if (tcp_paws_check(rx_opt, 0)) 1220 return false; 1221 1222 /* RST segments are not recommended to carry timestamp, 1223 and, if they do, it is recommended to ignore PAWS because 1224 "their cleanup function should take precedence over timestamps." 1225 Certainly, it is mistake. It is necessary to understand the reasons 1226 of this constraint to relax it: if peer reboots, clock may go 1227 out-of-sync and half-open connections will not be reset. 1228 Actually, the problem would be not existing if all 1229 the implementations followed draft about maintaining clock 1230 via reboots. Linux-2.2 DOES NOT! 1231 1232 However, we can relax time bounds for RST segments to MSL. 1233 */ 1234 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1235 return false; 1236 return true; 1237 } 1238 1239 /* Return true if we're currently rate-limiting out-of-window ACKs and 1240 * thus shouldn't send a dupack right now. We rate-limit dupacks in 1241 * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS 1242 * attacks that send repeated SYNs or ACKs for the same connection. To 1243 * do this, we do not send a duplicate SYNACK or ACK if the remote 1244 * endpoint is sending out-of-window SYNs or pure ACKs at a high rate. 1245 */ 1246 static inline bool tcp_oow_rate_limited(struct net *net, 1247 const struct sk_buff *skb, 1248 int mib_idx, u32 *last_oow_ack_time) 1249 { 1250 /* Data packets without SYNs are not likely part of an ACK loop. */ 1251 if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) && 1252 !tcp_hdr(skb)->syn) 1253 goto not_rate_limited; 1254 1255 if (*last_oow_ack_time) { 1256 s32 elapsed = (s32)(tcp_time_stamp - *last_oow_ack_time); 1257 1258 if (0 <= elapsed && elapsed < sysctl_tcp_invalid_ratelimit) { 1259 NET_INC_STATS_BH(net, mib_idx); 1260 return true; /* rate-limited: don't send yet! */ 1261 } 1262 } 1263 1264 *last_oow_ack_time = tcp_time_stamp; 1265 1266 not_rate_limited: 1267 return false; /* not rate-limited: go ahead, send dupack now! */ 1268 } 1269 1270 static inline void tcp_mib_init(struct net *net) 1271 { 1272 /* See RFC 2012 */ 1273 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1274 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1275 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1276 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1277 } 1278 1279 /* from STCP */ 1280 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1281 { 1282 tp->lost_skb_hint = NULL; 1283 } 1284 1285 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1286 { 1287 tcp_clear_retrans_hints_partial(tp); 1288 tp->retransmit_skb_hint = NULL; 1289 } 1290 1291 /* MD5 Signature */ 1292 struct crypto_hash; 1293 1294 union tcp_md5_addr { 1295 struct in_addr a4; 1296 #if IS_ENABLED(CONFIG_IPV6) 1297 struct in6_addr a6; 1298 #endif 1299 }; 1300 1301 /* - key database */ 1302 struct tcp_md5sig_key { 1303 struct hlist_node node; 1304 u8 keylen; 1305 u8 family; /* AF_INET or AF_INET6 */ 1306 union tcp_md5_addr addr; 1307 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1308 struct rcu_head rcu; 1309 }; 1310 1311 /* - sock block */ 1312 struct tcp_md5sig_info { 1313 struct hlist_head head; 1314 struct rcu_head rcu; 1315 }; 1316 1317 /* - pseudo header */ 1318 struct tcp4_pseudohdr { 1319 __be32 saddr; 1320 __be32 daddr; 1321 __u8 pad; 1322 __u8 protocol; 1323 __be16 len; 1324 }; 1325 1326 struct tcp6_pseudohdr { 1327 struct in6_addr saddr; 1328 struct in6_addr daddr; 1329 __be32 len; 1330 __be32 protocol; /* including padding */ 1331 }; 1332 1333 union tcp_md5sum_block { 1334 struct tcp4_pseudohdr ip4; 1335 #if IS_ENABLED(CONFIG_IPV6) 1336 struct tcp6_pseudohdr ip6; 1337 #endif 1338 }; 1339 1340 /* - pool: digest algorithm, hash description and scratch buffer */ 1341 struct tcp_md5sig_pool { 1342 struct hash_desc md5_desc; 1343 union tcp_md5sum_block md5_blk; 1344 }; 1345 1346 /* - functions */ 1347 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1348 const struct sock *sk, const struct request_sock *req, 1349 const struct sk_buff *skb); 1350 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1351 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1352 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1353 int family); 1354 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1355 struct sock *addr_sk); 1356 1357 #ifdef CONFIG_TCP_MD5SIG 1358 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1359 const union tcp_md5_addr *addr, 1360 int family); 1361 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1362 #else 1363 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1364 const union tcp_md5_addr *addr, 1365 int family) 1366 { 1367 return NULL; 1368 } 1369 #define tcp_twsk_md5_key(twsk) NULL 1370 #endif 1371 1372 bool tcp_alloc_md5sig_pool(void); 1373 1374 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1375 static inline void tcp_put_md5sig_pool(void) 1376 { 1377 local_bh_enable(); 1378 } 1379 1380 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1381 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1382 unsigned int header_len); 1383 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1384 const struct tcp_md5sig_key *key); 1385 1386 /* From tcp_fastopen.c */ 1387 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1388 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1389 unsigned long *last_syn_loss); 1390 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1391 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1392 struct tcp_fastopen_request { 1393 /* Fast Open cookie. Size 0 means a cookie request */ 1394 struct tcp_fastopen_cookie cookie; 1395 struct msghdr *data; /* data in MSG_FASTOPEN */ 1396 size_t size; 1397 int copied; /* queued in tcp_connect() */ 1398 }; 1399 void tcp_free_fastopen_req(struct tcp_sock *tp); 1400 1401 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1402 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1403 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1404 struct request_sock *req, 1405 struct tcp_fastopen_cookie *foc, 1406 struct dst_entry *dst); 1407 void tcp_fastopen_init_key_once(bool publish); 1408 #define TCP_FASTOPEN_KEY_LENGTH 16 1409 1410 /* Fastopen key context */ 1411 struct tcp_fastopen_context { 1412 struct crypto_cipher *tfm; 1413 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1414 struct rcu_head rcu; 1415 }; 1416 1417 /* write queue abstraction */ 1418 static inline void tcp_write_queue_purge(struct sock *sk) 1419 { 1420 struct sk_buff *skb; 1421 1422 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1423 sk_wmem_free_skb(sk, skb); 1424 sk_mem_reclaim(sk); 1425 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1426 } 1427 1428 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1429 { 1430 return skb_peek(&sk->sk_write_queue); 1431 } 1432 1433 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1434 { 1435 return skb_peek_tail(&sk->sk_write_queue); 1436 } 1437 1438 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1439 const struct sk_buff *skb) 1440 { 1441 return skb_queue_next(&sk->sk_write_queue, skb); 1442 } 1443 1444 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1445 const struct sk_buff *skb) 1446 { 1447 return skb_queue_prev(&sk->sk_write_queue, skb); 1448 } 1449 1450 #define tcp_for_write_queue(skb, sk) \ 1451 skb_queue_walk(&(sk)->sk_write_queue, skb) 1452 1453 #define tcp_for_write_queue_from(skb, sk) \ 1454 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1455 1456 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1457 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1458 1459 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1460 { 1461 return sk->sk_send_head; 1462 } 1463 1464 static inline bool tcp_skb_is_last(const struct sock *sk, 1465 const struct sk_buff *skb) 1466 { 1467 return skb_queue_is_last(&sk->sk_write_queue, skb); 1468 } 1469 1470 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1471 { 1472 if (tcp_skb_is_last(sk, skb)) 1473 sk->sk_send_head = NULL; 1474 else 1475 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1476 } 1477 1478 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1479 { 1480 if (sk->sk_send_head == skb_unlinked) 1481 sk->sk_send_head = NULL; 1482 } 1483 1484 static inline void tcp_init_send_head(struct sock *sk) 1485 { 1486 sk->sk_send_head = NULL; 1487 } 1488 1489 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1490 { 1491 __skb_queue_tail(&sk->sk_write_queue, skb); 1492 } 1493 1494 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1495 { 1496 __tcp_add_write_queue_tail(sk, skb); 1497 1498 /* Queue it, remembering where we must start sending. */ 1499 if (sk->sk_send_head == NULL) { 1500 sk->sk_send_head = skb; 1501 1502 if (tcp_sk(sk)->highest_sack == NULL) 1503 tcp_sk(sk)->highest_sack = skb; 1504 } 1505 } 1506 1507 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1508 { 1509 __skb_queue_head(&sk->sk_write_queue, skb); 1510 } 1511 1512 /* Insert buff after skb on the write queue of sk. */ 1513 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1514 struct sk_buff *buff, 1515 struct sock *sk) 1516 { 1517 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1518 } 1519 1520 /* Insert new before skb on the write queue of sk. */ 1521 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1522 struct sk_buff *skb, 1523 struct sock *sk) 1524 { 1525 __skb_queue_before(&sk->sk_write_queue, skb, new); 1526 1527 if (sk->sk_send_head == skb) 1528 sk->sk_send_head = new; 1529 } 1530 1531 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1532 { 1533 __skb_unlink(skb, &sk->sk_write_queue); 1534 } 1535 1536 static inline bool tcp_write_queue_empty(struct sock *sk) 1537 { 1538 return skb_queue_empty(&sk->sk_write_queue); 1539 } 1540 1541 static inline void tcp_push_pending_frames(struct sock *sk) 1542 { 1543 if (tcp_send_head(sk)) { 1544 struct tcp_sock *tp = tcp_sk(sk); 1545 1546 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1547 } 1548 } 1549 1550 /* Start sequence of the skb just after the highest skb with SACKed 1551 * bit, valid only if sacked_out > 0 or when the caller has ensured 1552 * validity by itself. 1553 */ 1554 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1555 { 1556 if (!tp->sacked_out) 1557 return tp->snd_una; 1558 1559 if (tp->highest_sack == NULL) 1560 return tp->snd_nxt; 1561 1562 return TCP_SKB_CB(tp->highest_sack)->seq; 1563 } 1564 1565 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1566 { 1567 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1568 tcp_write_queue_next(sk, skb); 1569 } 1570 1571 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1572 { 1573 return tcp_sk(sk)->highest_sack; 1574 } 1575 1576 static inline void tcp_highest_sack_reset(struct sock *sk) 1577 { 1578 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1579 } 1580 1581 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1582 static inline void tcp_highest_sack_combine(struct sock *sk, 1583 struct sk_buff *old, 1584 struct sk_buff *new) 1585 { 1586 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1587 tcp_sk(sk)->highest_sack = new; 1588 } 1589 1590 /* Determines whether this is a thin stream (which may suffer from 1591 * increased latency). Used to trigger latency-reducing mechanisms. 1592 */ 1593 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1594 { 1595 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1596 } 1597 1598 /* /proc */ 1599 enum tcp_seq_states { 1600 TCP_SEQ_STATE_LISTENING, 1601 TCP_SEQ_STATE_OPENREQ, 1602 TCP_SEQ_STATE_ESTABLISHED, 1603 }; 1604 1605 int tcp_seq_open(struct inode *inode, struct file *file); 1606 1607 struct tcp_seq_afinfo { 1608 char *name; 1609 sa_family_t family; 1610 const struct file_operations *seq_fops; 1611 struct seq_operations seq_ops; 1612 }; 1613 1614 struct tcp_iter_state { 1615 struct seq_net_private p; 1616 sa_family_t family; 1617 enum tcp_seq_states state; 1618 struct sock *syn_wait_sk; 1619 int bucket, offset, sbucket, num; 1620 kuid_t uid; 1621 loff_t last_pos; 1622 }; 1623 1624 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1625 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1626 1627 extern struct request_sock_ops tcp_request_sock_ops; 1628 extern struct request_sock_ops tcp6_request_sock_ops; 1629 1630 void tcp_v4_destroy_sock(struct sock *sk); 1631 1632 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1633 netdev_features_t features); 1634 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1635 int tcp_gro_complete(struct sk_buff *skb); 1636 1637 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1638 1639 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1640 { 1641 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1642 } 1643 1644 static inline bool tcp_stream_memory_free(const struct sock *sk) 1645 { 1646 const struct tcp_sock *tp = tcp_sk(sk); 1647 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1648 1649 return notsent_bytes < tcp_notsent_lowat(tp); 1650 } 1651 1652 #ifdef CONFIG_PROC_FS 1653 int tcp4_proc_init(void); 1654 void tcp4_proc_exit(void); 1655 #endif 1656 1657 int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1658 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1659 const struct tcp_request_sock_ops *af_ops, 1660 struct sock *sk, struct sk_buff *skb); 1661 1662 /* TCP af-specific functions */ 1663 struct tcp_sock_af_ops { 1664 #ifdef CONFIG_TCP_MD5SIG 1665 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1666 struct sock *addr_sk); 1667 int (*calc_md5_hash) (char *location, 1668 struct tcp_md5sig_key *md5, 1669 const struct sock *sk, 1670 const struct request_sock *req, 1671 const struct sk_buff *skb); 1672 int (*md5_parse) (struct sock *sk, 1673 char __user *optval, 1674 int optlen); 1675 #endif 1676 }; 1677 1678 struct tcp_request_sock_ops { 1679 u16 mss_clamp; 1680 #ifdef CONFIG_TCP_MD5SIG 1681 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1682 struct request_sock *req); 1683 int (*calc_md5_hash) (char *location, 1684 struct tcp_md5sig_key *md5, 1685 const struct sock *sk, 1686 const struct request_sock *req, 1687 const struct sk_buff *skb); 1688 #endif 1689 void (*init_req)(struct request_sock *req, struct sock *sk, 1690 struct sk_buff *skb); 1691 #ifdef CONFIG_SYN_COOKIES 1692 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1693 __u16 *mss); 1694 #endif 1695 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1696 const struct request_sock *req, 1697 bool *strict); 1698 __u32 (*init_seq)(const struct sk_buff *skb); 1699 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1700 struct flowi *fl, struct request_sock *req, 1701 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1702 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1703 const unsigned long timeout); 1704 }; 1705 1706 #ifdef CONFIG_SYN_COOKIES 1707 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1708 struct sock *sk, struct sk_buff *skb, 1709 __u16 *mss) 1710 { 1711 return ops->cookie_init_seq(sk, skb, mss); 1712 } 1713 #else 1714 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1715 struct sock *sk, struct sk_buff *skb, 1716 __u16 *mss) 1717 { 1718 return 0; 1719 } 1720 #endif 1721 1722 int tcpv4_offload_init(void); 1723 1724 void tcp_v4_init(void); 1725 void tcp_init(void); 1726 1727 /* 1728 * Save and compile IPv4 options, return a pointer to it 1729 */ 1730 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1731 { 1732 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1733 struct ip_options_rcu *dopt = NULL; 1734 1735 if (opt->optlen) { 1736 int opt_size = sizeof(*dopt) + opt->optlen; 1737 1738 dopt = kmalloc(opt_size, GFP_ATOMIC); 1739 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1740 kfree(dopt); 1741 dopt = NULL; 1742 } 1743 } 1744 return dopt; 1745 } 1746 1747 /* locally generated TCP pure ACKs have skb->truesize == 2 1748 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1749 * This is much faster than dissecting the packet to find out. 1750 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1751 */ 1752 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1753 { 1754 return skb->truesize == 2; 1755 } 1756 1757 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1758 { 1759 skb->truesize = 2; 1760 } 1761 1762 #endif /* _TCP_H */ 1763