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 1024 69 70 /* probing interval, default to 10 minutes as per RFC4821 */ 71 #define TCP_PROBE_INTERVAL 600 72 73 /* Specify interval when tcp mtu probing will stop */ 74 #define TCP_PROBE_THRESHOLD 8 75 76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 77 #define TCP_FASTRETRANS_THRESH 3 78 79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80 #define TCP_MAX_QUICKACKS 16U 81 82 /* urg_data states */ 83 #define TCP_URG_VALID 0x0100 84 #define TCP_URG_NOTYET 0x0200 85 #define TCP_URG_READ 0x0400 86 87 #define TCP_RETR1 3 /* 88 * This is how many retries it does before it 89 * tries to figure out if the gateway is 90 * down. Minimal RFC value is 3; it corresponds 91 * to ~3sec-8min depending on RTO. 92 */ 93 94 #define TCP_RETR2 15 /* 95 * This should take at least 96 * 90 minutes to time out. 97 * RFC1122 says that the limit is 100 sec. 98 * 15 is ~13-30min depending on RTO. 99 */ 100 101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 102 * when active opening a connection. 103 * RFC1122 says the minimum retry MUST 104 * be at least 180secs. Nevertheless 105 * this value is corresponding to 106 * 63secs of retransmission with the 107 * current initial RTO. 108 */ 109 110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 111 * when passive opening a connection. 112 * This is corresponding to 31secs of 113 * retransmission with the current 114 * initial RTO. 115 */ 116 117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 118 * state, about 60 seconds */ 119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 120 /* BSD style FIN_WAIT2 deadlock breaker. 121 * It used to be 3min, new value is 60sec, 122 * to combine FIN-WAIT-2 timeout with 123 * TIME-WAIT timer. 124 */ 125 126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 127 #if HZ >= 100 128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 129 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 130 #else 131 #define TCP_DELACK_MIN 4U 132 #define TCP_ATO_MIN 4U 133 #endif 134 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 135 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 138 * used as a fallback RTO for the 139 * initial data transmission if no 140 * valid RTT sample has been acquired, 141 * most likely due to retrans in 3WHS. 142 */ 143 144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 145 * for local resources. 146 */ 147 148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 150 #define TCP_KEEPALIVE_INTVL (75*HZ) 151 152 #define MAX_TCP_KEEPIDLE 32767 153 #define MAX_TCP_KEEPINTVL 32767 154 #define MAX_TCP_KEEPCNT 127 155 #define MAX_TCP_SYNCNT 127 156 157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 158 159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 161 * after this time. It should be equal 162 * (or greater than) TCP_TIMEWAIT_LEN 163 * to provide reliability equal to one 164 * provided by timewait state. 165 */ 166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 167 * timestamps. It must be less than 168 * minimal timewait lifetime. 169 */ 170 /* 171 * TCP option 172 */ 173 174 #define TCPOPT_NOP 1 /* Padding */ 175 #define TCPOPT_EOL 0 /* End of options */ 176 #define TCPOPT_MSS 2 /* Segment size negotiating */ 177 #define TCPOPT_WINDOW 3 /* Window scaling */ 178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 179 #define TCPOPT_SACK 5 /* SACK Block */ 180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 183 #define TCPOPT_EXP 254 /* Experimental */ 184 /* Magic number to be after the option value for sharing TCP 185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 186 */ 187 #define TCPOPT_FASTOPEN_MAGIC 0xF989 188 189 /* 190 * TCP option lengths 191 */ 192 193 #define TCPOLEN_MSS 4 194 #define TCPOLEN_WINDOW 3 195 #define TCPOLEN_SACK_PERM 2 196 #define TCPOLEN_TIMESTAMP 10 197 #define TCPOLEN_MD5SIG 18 198 #define TCPOLEN_FASTOPEN_BASE 2 199 #define TCPOLEN_EXP_FASTOPEN_BASE 4 200 201 /* But this is what stacks really send out. */ 202 #define TCPOLEN_TSTAMP_ALIGNED 12 203 #define TCPOLEN_WSCALE_ALIGNED 4 204 #define TCPOLEN_SACKPERM_ALIGNED 4 205 #define TCPOLEN_SACK_BASE 2 206 #define TCPOLEN_SACK_BASE_ALIGNED 4 207 #define TCPOLEN_SACK_PERBLOCK 8 208 #define TCPOLEN_MD5SIG_ALIGNED 20 209 #define TCPOLEN_MSS_ALIGNED 4 210 211 /* Flags in tp->nonagle */ 212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 213 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 215 216 /* TCP thin-stream limits */ 217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 218 219 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 220 #define TCP_INIT_CWND 10 221 222 /* Bit Flags for sysctl_tcp_fastopen */ 223 #define TFO_CLIENT_ENABLE 1 224 #define TFO_SERVER_ENABLE 2 225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 226 227 /* Accept SYN data w/o any cookie option */ 228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 229 230 /* Force enable TFO on all listeners, i.e., not requiring the 231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 232 */ 233 #define TFO_SERVER_WO_SOCKOPT1 0x400 234 #define TFO_SERVER_WO_SOCKOPT2 0x800 235 236 extern struct inet_timewait_death_row tcp_death_row; 237 238 /* sysctl variables for tcp */ 239 extern int sysctl_tcp_timestamps; 240 extern int sysctl_tcp_window_scaling; 241 extern int sysctl_tcp_sack; 242 extern int sysctl_tcp_fin_timeout; 243 extern int sysctl_tcp_keepalive_time; 244 extern int sysctl_tcp_keepalive_probes; 245 extern int sysctl_tcp_keepalive_intvl; 246 extern int sysctl_tcp_syn_retries; 247 extern int sysctl_tcp_synack_retries; 248 extern int sysctl_tcp_retries1; 249 extern int sysctl_tcp_retries2; 250 extern int sysctl_tcp_orphan_retries; 251 extern int sysctl_tcp_syncookies; 252 extern int sysctl_tcp_fastopen; 253 extern int sysctl_tcp_retrans_collapse; 254 extern int sysctl_tcp_stdurg; 255 extern int sysctl_tcp_rfc1337; 256 extern int sysctl_tcp_abort_on_overflow; 257 extern int sysctl_tcp_max_orphans; 258 extern int sysctl_tcp_fack; 259 extern int sysctl_tcp_reordering; 260 extern int sysctl_tcp_max_reordering; 261 extern int sysctl_tcp_dsack; 262 extern long sysctl_tcp_mem[3]; 263 extern int sysctl_tcp_wmem[3]; 264 extern int sysctl_tcp_rmem[3]; 265 extern int sysctl_tcp_app_win; 266 extern int sysctl_tcp_adv_win_scale; 267 extern int sysctl_tcp_tw_reuse; 268 extern int sysctl_tcp_frto; 269 extern int sysctl_tcp_low_latency; 270 extern int sysctl_tcp_nometrics_save; 271 extern int sysctl_tcp_moderate_rcvbuf; 272 extern int sysctl_tcp_tso_win_divisor; 273 extern int sysctl_tcp_workaround_signed_windows; 274 extern int sysctl_tcp_slow_start_after_idle; 275 extern int sysctl_tcp_thin_linear_timeouts; 276 extern int sysctl_tcp_thin_dupack; 277 extern int sysctl_tcp_early_retrans; 278 extern int sysctl_tcp_limit_output_bytes; 279 extern int sysctl_tcp_challenge_ack_limit; 280 extern unsigned int sysctl_tcp_notsent_lowat; 281 extern int sysctl_tcp_min_tso_segs; 282 extern int sysctl_tcp_autocorking; 283 extern int sysctl_tcp_invalid_ratelimit; 284 285 extern atomic_long_t tcp_memory_allocated; 286 extern struct percpu_counter tcp_sockets_allocated; 287 extern int tcp_memory_pressure; 288 289 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 290 static inline bool tcp_under_memory_pressure(const struct sock *sk) 291 { 292 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 293 return !!sk->sk_cgrp->memory_pressure; 294 295 return tcp_memory_pressure; 296 } 297 /* 298 * The next routines deal with comparing 32 bit unsigned ints 299 * and worry about wraparound (automatic with unsigned arithmetic). 300 */ 301 302 static inline bool before(__u32 seq1, __u32 seq2) 303 { 304 return (__s32)(seq1-seq2) < 0; 305 } 306 #define after(seq2, seq1) before(seq1, seq2) 307 308 /* is s2<=s1<=s3 ? */ 309 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 310 { 311 return seq3 - seq2 >= seq1 - seq2; 312 } 313 314 static inline bool tcp_out_of_memory(struct sock *sk) 315 { 316 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 317 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 318 return true; 319 return false; 320 } 321 322 void sk_forced_mem_schedule(struct sock *sk, int size); 323 324 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 325 { 326 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 327 int orphans = percpu_counter_read_positive(ocp); 328 329 if (orphans << shift > sysctl_tcp_max_orphans) { 330 orphans = percpu_counter_sum_positive(ocp); 331 if (orphans << shift > sysctl_tcp_max_orphans) 332 return true; 333 } 334 return false; 335 } 336 337 bool tcp_check_oom(struct sock *sk, int shift); 338 339 340 extern struct proto tcp_prot; 341 342 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 343 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 344 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 345 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 346 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 347 348 void tcp_tasklet_init(void); 349 350 void tcp_v4_err(struct sk_buff *skb, u32); 351 352 void tcp_shutdown(struct sock *sk, int how); 353 354 void tcp_v4_early_demux(struct sk_buff *skb); 355 int tcp_v4_rcv(struct sk_buff *skb); 356 357 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 358 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 359 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 360 int flags); 361 void tcp_release_cb(struct sock *sk); 362 void tcp_wfree(struct sk_buff *skb); 363 void tcp_write_timer_handler(struct sock *sk); 364 void tcp_delack_timer_handler(struct sock *sk); 365 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 366 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 367 const struct tcphdr *th, unsigned int len); 368 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 369 const struct tcphdr *th, unsigned int len); 370 void tcp_rcv_space_adjust(struct sock *sk); 371 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 372 void tcp_twsk_destructor(struct sock *sk); 373 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 374 struct pipe_inode_info *pipe, size_t len, 375 unsigned int flags); 376 377 static inline void tcp_dec_quickack_mode(struct sock *sk, 378 const unsigned int pkts) 379 { 380 struct inet_connection_sock *icsk = inet_csk(sk); 381 382 if (icsk->icsk_ack.quick) { 383 if (pkts >= icsk->icsk_ack.quick) { 384 icsk->icsk_ack.quick = 0; 385 /* Leaving quickack mode we deflate ATO. */ 386 icsk->icsk_ack.ato = TCP_ATO_MIN; 387 } else 388 icsk->icsk_ack.quick -= pkts; 389 } 390 } 391 392 #define TCP_ECN_OK 1 393 #define TCP_ECN_QUEUE_CWR 2 394 #define TCP_ECN_DEMAND_CWR 4 395 #define TCP_ECN_SEEN 8 396 397 enum tcp_tw_status { 398 TCP_TW_SUCCESS = 0, 399 TCP_TW_RST = 1, 400 TCP_TW_ACK = 2, 401 TCP_TW_SYN = 3 402 }; 403 404 405 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 406 struct sk_buff *skb, 407 const struct tcphdr *th); 408 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 409 struct request_sock *req, bool fastopen); 410 int tcp_child_process(struct sock *parent, struct sock *child, 411 struct sk_buff *skb); 412 void tcp_enter_loss(struct sock *sk); 413 void tcp_clear_retrans(struct tcp_sock *tp); 414 void tcp_update_metrics(struct sock *sk); 415 void tcp_init_metrics(struct sock *sk); 416 void tcp_metrics_init(void); 417 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 418 bool paws_check, bool timestamps); 419 bool tcp_remember_stamp(struct sock *sk); 420 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 421 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 422 void tcp_disable_fack(struct tcp_sock *tp); 423 void tcp_close(struct sock *sk, long timeout); 424 void tcp_init_sock(struct sock *sk); 425 unsigned int tcp_poll(struct file *file, struct socket *sock, 426 struct poll_table_struct *wait); 427 int tcp_getsockopt(struct sock *sk, int level, int optname, 428 char __user *optval, int __user *optlen); 429 int tcp_setsockopt(struct sock *sk, int level, int optname, 430 char __user *optval, unsigned int optlen); 431 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 432 char __user *optval, int __user *optlen); 433 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 434 char __user *optval, unsigned int optlen); 435 void tcp_set_keepalive(struct sock *sk, int val); 436 void tcp_syn_ack_timeout(const struct request_sock *req); 437 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 438 int flags, int *addr_len); 439 void tcp_parse_options(const struct sk_buff *skb, 440 struct tcp_options_received *opt_rx, 441 int estab, struct tcp_fastopen_cookie *foc); 442 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 443 444 /* 445 * TCP v4 functions exported for the inet6 API 446 */ 447 448 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 449 void tcp_v4_mtu_reduced(struct sock *sk); 450 void tcp_req_err(struct sock *sk, u32 seq); 451 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 452 struct sock *tcp_create_openreq_child(struct sock *sk, 453 struct request_sock *req, 454 struct sk_buff *skb); 455 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 456 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 457 struct request_sock *req, 458 struct dst_entry *dst); 459 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 460 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 461 int tcp_connect(struct sock *sk); 462 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 463 struct request_sock *req, 464 struct tcp_fastopen_cookie *foc); 465 int tcp_disconnect(struct sock *sk, int flags); 466 467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 470 471 /* From syncookies.c */ 472 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 473 struct request_sock *req, 474 struct dst_entry *dst); 475 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 476 u32 cookie); 477 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 478 #ifdef CONFIG_SYN_COOKIES 479 480 /* Syncookies use a monotonic timer which increments every 60 seconds. 481 * This counter is used both as a hash input and partially encoded into 482 * the cookie value. A cookie is only validated further if the delta 483 * between the current counter value and the encoded one is less than this, 484 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 485 * the counter advances immediately after a cookie is generated). 486 */ 487 #define MAX_SYNCOOKIE_AGE 2 488 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 489 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 490 491 /* syncookies: remember time of last synqueue overflow 492 * But do not dirty this field too often (once per second is enough) 493 */ 494 static inline void tcp_synq_overflow(struct sock *sk) 495 { 496 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 497 unsigned long now = jiffies; 498 499 if (time_after(now, last_overflow + HZ)) 500 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 501 } 502 503 /* syncookies: no recent synqueue overflow on this listening socket? */ 504 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 505 { 506 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 507 508 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 509 } 510 511 static inline u32 tcp_cookie_time(void) 512 { 513 u64 val = get_jiffies_64(); 514 515 do_div(val, TCP_SYNCOOKIE_PERIOD); 516 return val; 517 } 518 519 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 520 u16 *mssp); 521 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 522 __u16 *mss); 523 __u32 cookie_init_timestamp(struct request_sock *req); 524 bool cookie_timestamp_decode(struct tcp_options_received *opt); 525 bool cookie_ecn_ok(const struct tcp_options_received *opt, 526 const struct net *net, const struct dst_entry *dst); 527 528 /* From net/ipv6/syncookies.c */ 529 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 530 u32 cookie); 531 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 532 533 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 534 const struct tcphdr *th, u16 *mssp); 535 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 536 __u16 *mss); 537 #endif 538 /* tcp_output.c */ 539 540 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 541 int nonagle); 542 bool tcp_may_send_now(struct sock *sk); 543 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 544 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 545 void tcp_retransmit_timer(struct sock *sk); 546 void tcp_xmit_retransmit_queue(struct sock *); 547 void tcp_simple_retransmit(struct sock *); 548 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 549 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 550 551 void tcp_send_probe0(struct sock *); 552 void tcp_send_partial(struct sock *); 553 int tcp_write_wakeup(struct sock *, int mib); 554 void tcp_send_fin(struct sock *sk); 555 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 556 int tcp_send_synack(struct sock *); 557 void tcp_push_one(struct sock *, unsigned int mss_now); 558 void tcp_send_ack(struct sock *sk); 559 void tcp_send_delayed_ack(struct sock *sk); 560 void tcp_send_loss_probe(struct sock *sk); 561 bool tcp_schedule_loss_probe(struct sock *sk); 562 563 /* tcp_input.c */ 564 void tcp_resume_early_retransmit(struct sock *sk); 565 void tcp_rearm_rto(struct sock *sk); 566 void tcp_reset(struct sock *sk); 567 568 /* tcp_timer.c */ 569 void tcp_init_xmit_timers(struct sock *); 570 static inline void tcp_clear_xmit_timers(struct sock *sk) 571 { 572 inet_csk_clear_xmit_timers(sk); 573 } 574 575 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 576 unsigned int tcp_current_mss(struct sock *sk); 577 578 /* Bound MSS / TSO packet size with the half of the window */ 579 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 580 { 581 int cutoff; 582 583 /* When peer uses tiny windows, there is no use in packetizing 584 * to sub-MSS pieces for the sake of SWS or making sure there 585 * are enough packets in the pipe for fast recovery. 586 * 587 * On the other hand, for extremely large MSS devices, handling 588 * smaller than MSS windows in this way does make sense. 589 */ 590 if (tp->max_window >= 512) 591 cutoff = (tp->max_window >> 1); 592 else 593 cutoff = tp->max_window; 594 595 if (cutoff && pktsize > cutoff) 596 return max_t(int, cutoff, 68U - tp->tcp_header_len); 597 else 598 return pktsize; 599 } 600 601 /* tcp.c */ 602 void tcp_get_info(struct sock *, struct tcp_info *); 603 604 /* Read 'sendfile()'-style from a TCP socket */ 605 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 606 unsigned int, size_t); 607 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 608 sk_read_actor_t recv_actor); 609 610 void tcp_initialize_rcv_mss(struct sock *sk); 611 612 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 613 int tcp_mss_to_mtu(struct sock *sk, int mss); 614 void tcp_mtup_init(struct sock *sk); 615 void tcp_init_buffer_space(struct sock *sk); 616 617 static inline void tcp_bound_rto(const struct sock *sk) 618 { 619 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 620 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 621 } 622 623 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 624 { 625 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 626 } 627 628 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 629 { 630 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 631 ntohl(TCP_FLAG_ACK) | 632 snd_wnd); 633 } 634 635 static inline void tcp_fast_path_on(struct tcp_sock *tp) 636 { 637 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 638 } 639 640 static inline void tcp_fast_path_check(struct sock *sk) 641 { 642 struct tcp_sock *tp = tcp_sk(sk); 643 644 if (skb_queue_empty(&tp->out_of_order_queue) && 645 tp->rcv_wnd && 646 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 647 !tp->urg_data) 648 tcp_fast_path_on(tp); 649 } 650 651 /* Compute the actual rto_min value */ 652 static inline u32 tcp_rto_min(struct sock *sk) 653 { 654 const struct dst_entry *dst = __sk_dst_get(sk); 655 u32 rto_min = TCP_RTO_MIN; 656 657 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 658 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 659 return rto_min; 660 } 661 662 static inline u32 tcp_rto_min_us(struct sock *sk) 663 { 664 return jiffies_to_usecs(tcp_rto_min(sk)); 665 } 666 667 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 668 { 669 return dst_metric_locked(dst, RTAX_CC_ALGO); 670 } 671 672 /* Compute the actual receive window we are currently advertising. 673 * Rcv_nxt can be after the window if our peer push more data 674 * than the offered window. 675 */ 676 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 677 { 678 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 679 680 if (win < 0) 681 win = 0; 682 return (u32) win; 683 } 684 685 /* Choose a new window, without checks for shrinking, and without 686 * scaling applied to the result. The caller does these things 687 * if necessary. This is a "raw" window selection. 688 */ 689 u32 __tcp_select_window(struct sock *sk); 690 691 void tcp_send_window_probe(struct sock *sk); 692 693 /* TCP timestamps are only 32-bits, this causes a slight 694 * complication on 64-bit systems since we store a snapshot 695 * of jiffies in the buffer control blocks below. We decided 696 * to use only the low 32-bits of jiffies and hide the ugly 697 * casts with the following macro. 698 */ 699 #define tcp_time_stamp ((__u32)(jiffies)) 700 701 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 702 { 703 return skb->skb_mstamp.stamp_jiffies; 704 } 705 706 707 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 708 709 #define TCPHDR_FIN 0x01 710 #define TCPHDR_SYN 0x02 711 #define TCPHDR_RST 0x04 712 #define TCPHDR_PSH 0x08 713 #define TCPHDR_ACK 0x10 714 #define TCPHDR_URG 0x20 715 #define TCPHDR_ECE 0x40 716 #define TCPHDR_CWR 0x80 717 718 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 719 720 /* This is what the send packet queuing engine uses to pass 721 * TCP per-packet control information to the transmission code. 722 * We also store the host-order sequence numbers in here too. 723 * This is 44 bytes if IPV6 is enabled. 724 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 725 */ 726 struct tcp_skb_cb { 727 __u32 seq; /* Starting sequence number */ 728 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 729 union { 730 /* Note : tcp_tw_isn is used in input path only 731 * (isn chosen by tcp_timewait_state_process()) 732 * 733 * tcp_gso_segs/size are used in write queue only, 734 * cf tcp_skb_pcount()/tcp_skb_mss() 735 */ 736 __u32 tcp_tw_isn; 737 struct { 738 u16 tcp_gso_segs; 739 u16 tcp_gso_size; 740 }; 741 }; 742 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 743 744 __u8 sacked; /* State flags for SACK/FACK. */ 745 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 746 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 747 #define TCPCB_LOST 0x04 /* SKB is lost */ 748 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 749 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 750 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 751 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 752 TCPCB_REPAIRED) 753 754 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 755 /* 1 byte hole */ 756 __u32 ack_seq; /* Sequence number ACK'd */ 757 union { 758 struct inet_skb_parm h4; 759 #if IS_ENABLED(CONFIG_IPV6) 760 struct inet6_skb_parm h6; 761 #endif 762 } header; /* For incoming frames */ 763 }; 764 765 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 766 767 768 #if IS_ENABLED(CONFIG_IPV6) 769 /* This is the variant of inet6_iif() that must be used by TCP, 770 * as TCP moves IP6CB into a different location in skb->cb[] 771 */ 772 static inline int tcp_v6_iif(const struct sk_buff *skb) 773 { 774 return TCP_SKB_CB(skb)->header.h6.iif; 775 } 776 #endif 777 778 /* Due to TSO, an SKB can be composed of multiple actual 779 * packets. To keep these tracked properly, we use this. 780 */ 781 static inline int tcp_skb_pcount(const struct sk_buff *skb) 782 { 783 return TCP_SKB_CB(skb)->tcp_gso_segs; 784 } 785 786 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 787 { 788 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 789 } 790 791 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 792 { 793 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 794 } 795 796 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 797 static inline int tcp_skb_mss(const struct sk_buff *skb) 798 { 799 return TCP_SKB_CB(skb)->tcp_gso_size; 800 } 801 802 /* Events passed to congestion control interface */ 803 enum tcp_ca_event { 804 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 805 CA_EVENT_CWND_RESTART, /* congestion window restart */ 806 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 807 CA_EVENT_LOSS, /* loss timeout */ 808 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 809 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 810 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 811 CA_EVENT_NON_DELAYED_ACK, 812 }; 813 814 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 815 enum tcp_ca_ack_event_flags { 816 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 817 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 818 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 819 }; 820 821 /* 822 * Interface for adding new TCP congestion control handlers 823 */ 824 #define TCP_CA_NAME_MAX 16 825 #define TCP_CA_MAX 128 826 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 827 828 #define TCP_CA_UNSPEC 0 829 830 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 831 #define TCP_CONG_NON_RESTRICTED 0x1 832 /* Requires ECN/ECT set on all packets */ 833 #define TCP_CONG_NEEDS_ECN 0x2 834 835 union tcp_cc_info; 836 837 struct tcp_congestion_ops { 838 struct list_head list; 839 u32 key; 840 u32 flags; 841 842 /* initialize private data (optional) */ 843 void (*init)(struct sock *sk); 844 /* cleanup private data (optional) */ 845 void (*release)(struct sock *sk); 846 847 /* return slow start threshold (required) */ 848 u32 (*ssthresh)(struct sock *sk); 849 /* do new cwnd calculation (required) */ 850 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 851 /* call before changing ca_state (optional) */ 852 void (*set_state)(struct sock *sk, u8 new_state); 853 /* call when cwnd event occurs (optional) */ 854 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 855 /* call when ack arrives (optional) */ 856 void (*in_ack_event)(struct sock *sk, u32 flags); 857 /* new value of cwnd after loss (optional) */ 858 u32 (*undo_cwnd)(struct sock *sk); 859 /* hook for packet ack accounting (optional) */ 860 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 861 /* get info for inet_diag (optional) */ 862 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 863 union tcp_cc_info *info); 864 865 char name[TCP_CA_NAME_MAX]; 866 struct module *owner; 867 }; 868 869 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 870 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 871 872 void tcp_assign_congestion_control(struct sock *sk); 873 void tcp_init_congestion_control(struct sock *sk); 874 void tcp_cleanup_congestion_control(struct sock *sk); 875 int tcp_set_default_congestion_control(const char *name); 876 void tcp_get_default_congestion_control(char *name); 877 void tcp_get_available_congestion_control(char *buf, size_t len); 878 void tcp_get_allowed_congestion_control(char *buf, size_t len); 879 int tcp_set_allowed_congestion_control(char *allowed); 880 int tcp_set_congestion_control(struct sock *sk, const char *name); 881 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 882 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 883 884 u32 tcp_reno_ssthresh(struct sock *sk); 885 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 886 extern struct tcp_congestion_ops tcp_reno; 887 888 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 889 u32 tcp_ca_get_key_by_name(const char *name); 890 #ifdef CONFIG_INET 891 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 892 #else 893 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 894 { 895 return NULL; 896 } 897 #endif 898 899 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 900 { 901 const struct inet_connection_sock *icsk = inet_csk(sk); 902 903 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 904 } 905 906 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 907 { 908 struct inet_connection_sock *icsk = inet_csk(sk); 909 910 if (icsk->icsk_ca_ops->set_state) 911 icsk->icsk_ca_ops->set_state(sk, ca_state); 912 icsk->icsk_ca_state = ca_state; 913 } 914 915 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 916 { 917 const struct inet_connection_sock *icsk = inet_csk(sk); 918 919 if (icsk->icsk_ca_ops->cwnd_event) 920 icsk->icsk_ca_ops->cwnd_event(sk, event); 921 } 922 923 /* These functions determine how the current flow behaves in respect of SACK 924 * handling. SACK is negotiated with the peer, and therefore it can vary 925 * between different flows. 926 * 927 * tcp_is_sack - SACK enabled 928 * tcp_is_reno - No SACK 929 * tcp_is_fack - FACK enabled, implies SACK enabled 930 */ 931 static inline int tcp_is_sack(const struct tcp_sock *tp) 932 { 933 return tp->rx_opt.sack_ok; 934 } 935 936 static inline bool tcp_is_reno(const struct tcp_sock *tp) 937 { 938 return !tcp_is_sack(tp); 939 } 940 941 static inline bool tcp_is_fack(const struct tcp_sock *tp) 942 { 943 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 944 } 945 946 static inline void tcp_enable_fack(struct tcp_sock *tp) 947 { 948 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 949 } 950 951 /* TCP early-retransmit (ER) is similar to but more conservative than 952 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 953 */ 954 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 955 { 956 tp->do_early_retrans = sysctl_tcp_early_retrans && 957 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 958 sysctl_tcp_reordering == 3; 959 } 960 961 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 962 { 963 tp->do_early_retrans = 0; 964 } 965 966 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 967 { 968 return tp->sacked_out + tp->lost_out; 969 } 970 971 /* This determines how many packets are "in the network" to the best 972 * of our knowledge. In many cases it is conservative, but where 973 * detailed information is available from the receiver (via SACK 974 * blocks etc.) we can make more aggressive calculations. 975 * 976 * Use this for decisions involving congestion control, use just 977 * tp->packets_out to determine if the send queue is empty or not. 978 * 979 * Read this equation as: 980 * 981 * "Packets sent once on transmission queue" MINUS 982 * "Packets left network, but not honestly ACKed yet" PLUS 983 * "Packets fast retransmitted" 984 */ 985 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 986 { 987 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 988 } 989 990 #define TCP_INFINITE_SSTHRESH 0x7fffffff 991 992 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 993 { 994 return tp->snd_cwnd < tp->snd_ssthresh; 995 } 996 997 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 998 { 999 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1000 } 1001 1002 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1003 { 1004 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1005 (1 << inet_csk(sk)->icsk_ca_state); 1006 } 1007 1008 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1009 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1010 * ssthresh. 1011 */ 1012 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1013 { 1014 const struct tcp_sock *tp = tcp_sk(sk); 1015 1016 if (tcp_in_cwnd_reduction(sk)) 1017 return tp->snd_ssthresh; 1018 else 1019 return max(tp->snd_ssthresh, 1020 ((tp->snd_cwnd >> 1) + 1021 (tp->snd_cwnd >> 2))); 1022 } 1023 1024 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1025 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1026 1027 void tcp_enter_cwr(struct sock *sk); 1028 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1029 1030 /* The maximum number of MSS of available cwnd for which TSO defers 1031 * sending if not using sysctl_tcp_tso_win_divisor. 1032 */ 1033 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1034 { 1035 return 3; 1036 } 1037 1038 /* Slow start with delack produces 3 packets of burst, so that 1039 * it is safe "de facto". This will be the default - same as 1040 * the default reordering threshold - but if reordering increases, 1041 * we must be able to allow cwnd to burst at least this much in order 1042 * to not pull it back when holes are filled. 1043 */ 1044 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1045 { 1046 return tp->reordering; 1047 } 1048 1049 /* Returns end sequence number of the receiver's advertised window */ 1050 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1051 { 1052 return tp->snd_una + tp->snd_wnd; 1053 } 1054 1055 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1056 * flexible approach. The RFC suggests cwnd should not be raised unless 1057 * it was fully used previously. And that's exactly what we do in 1058 * congestion avoidance mode. But in slow start we allow cwnd to grow 1059 * as long as the application has used half the cwnd. 1060 * Example : 1061 * cwnd is 10 (IW10), but application sends 9 frames. 1062 * We allow cwnd to reach 18 when all frames are ACKed. 1063 * This check is safe because it's as aggressive as slow start which already 1064 * risks 100% overshoot. The advantage is that we discourage application to 1065 * either send more filler packets or data to artificially blow up the cwnd 1066 * usage, and allow application-limited process to probe bw more aggressively. 1067 */ 1068 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1069 { 1070 const struct tcp_sock *tp = tcp_sk(sk); 1071 1072 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1073 if (tcp_in_slow_start(tp)) 1074 return tp->snd_cwnd < 2 * tp->max_packets_out; 1075 1076 return tp->is_cwnd_limited; 1077 } 1078 1079 /* Something is really bad, we could not queue an additional packet, 1080 * because qdisc is full or receiver sent a 0 window. 1081 * We do not want to add fuel to the fire, or abort too early, 1082 * so make sure the timer we arm now is at least 200ms in the future, 1083 * regardless of current icsk_rto value (as it could be ~2ms) 1084 */ 1085 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1086 { 1087 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1088 } 1089 1090 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1091 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1092 unsigned long max_when) 1093 { 1094 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1095 1096 return (unsigned long)min_t(u64, when, max_when); 1097 } 1098 1099 static inline void tcp_check_probe_timer(struct sock *sk) 1100 { 1101 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1102 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1103 tcp_probe0_base(sk), TCP_RTO_MAX); 1104 } 1105 1106 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1107 { 1108 tp->snd_wl1 = seq; 1109 } 1110 1111 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1112 { 1113 tp->snd_wl1 = seq; 1114 } 1115 1116 /* 1117 * Calculate(/check) TCP checksum 1118 */ 1119 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1120 __be32 daddr, __wsum base) 1121 { 1122 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1123 } 1124 1125 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1126 { 1127 return __skb_checksum_complete(skb); 1128 } 1129 1130 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1131 { 1132 return !skb_csum_unnecessary(skb) && 1133 __tcp_checksum_complete(skb); 1134 } 1135 1136 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1137 1138 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1139 { 1140 tp->ucopy.task = NULL; 1141 tp->ucopy.len = 0; 1142 tp->ucopy.memory = 0; 1143 skb_queue_head_init(&tp->ucopy.prequeue); 1144 } 1145 1146 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1147 1148 #undef STATE_TRACE 1149 1150 #ifdef STATE_TRACE 1151 static const char *statename[]={ 1152 "Unused","Established","Syn Sent","Syn Recv", 1153 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1154 "Close Wait","Last ACK","Listen","Closing" 1155 }; 1156 #endif 1157 void tcp_set_state(struct sock *sk, int state); 1158 1159 void tcp_done(struct sock *sk); 1160 1161 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1162 { 1163 rx_opt->dsack = 0; 1164 rx_opt->num_sacks = 0; 1165 } 1166 1167 u32 tcp_default_init_rwnd(u32 mss); 1168 1169 /* Determine a window scaling and initial window to offer. */ 1170 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1171 __u32 *window_clamp, int wscale_ok, 1172 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1173 1174 static inline int tcp_win_from_space(int space) 1175 { 1176 return sysctl_tcp_adv_win_scale<=0 ? 1177 (space>>(-sysctl_tcp_adv_win_scale)) : 1178 space - (space>>sysctl_tcp_adv_win_scale); 1179 } 1180 1181 /* Note: caller must be prepared to deal with negative returns */ 1182 static inline int tcp_space(const struct sock *sk) 1183 { 1184 return tcp_win_from_space(sk->sk_rcvbuf - 1185 atomic_read(&sk->sk_rmem_alloc)); 1186 } 1187 1188 static inline int tcp_full_space(const struct sock *sk) 1189 { 1190 return tcp_win_from_space(sk->sk_rcvbuf); 1191 } 1192 1193 extern void tcp_openreq_init_rwin(struct request_sock *req, 1194 struct sock *sk, struct dst_entry *dst); 1195 1196 void tcp_enter_memory_pressure(struct sock *sk); 1197 1198 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1199 { 1200 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1201 } 1202 1203 static inline int keepalive_time_when(const struct tcp_sock *tp) 1204 { 1205 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1206 } 1207 1208 static inline int keepalive_probes(const struct tcp_sock *tp) 1209 { 1210 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1211 } 1212 1213 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1214 { 1215 const struct inet_connection_sock *icsk = &tp->inet_conn; 1216 1217 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1218 tcp_time_stamp - tp->rcv_tstamp); 1219 } 1220 1221 static inline int tcp_fin_time(const struct sock *sk) 1222 { 1223 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1224 const int rto = inet_csk(sk)->icsk_rto; 1225 1226 if (fin_timeout < (rto << 2) - (rto >> 1)) 1227 fin_timeout = (rto << 2) - (rto >> 1); 1228 1229 return fin_timeout; 1230 } 1231 1232 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1233 int paws_win) 1234 { 1235 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1236 return true; 1237 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1238 return true; 1239 /* 1240 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1241 * then following tcp messages have valid values. Ignore 0 value, 1242 * or else 'negative' tsval might forbid us to accept their packets. 1243 */ 1244 if (!rx_opt->ts_recent) 1245 return true; 1246 return false; 1247 } 1248 1249 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1250 int rst) 1251 { 1252 if (tcp_paws_check(rx_opt, 0)) 1253 return false; 1254 1255 /* RST segments are not recommended to carry timestamp, 1256 and, if they do, it is recommended to ignore PAWS because 1257 "their cleanup function should take precedence over timestamps." 1258 Certainly, it is mistake. It is necessary to understand the reasons 1259 of this constraint to relax it: if peer reboots, clock may go 1260 out-of-sync and half-open connections will not be reset. 1261 Actually, the problem would be not existing if all 1262 the implementations followed draft about maintaining clock 1263 via reboots. Linux-2.2 DOES NOT! 1264 1265 However, we can relax time bounds for RST segments to MSL. 1266 */ 1267 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1268 return false; 1269 return true; 1270 } 1271 1272 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1273 int mib_idx, u32 *last_oow_ack_time); 1274 1275 static inline void tcp_mib_init(struct net *net) 1276 { 1277 /* See RFC 2012 */ 1278 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1279 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1280 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1281 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1282 } 1283 1284 /* from STCP */ 1285 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1286 { 1287 tp->lost_skb_hint = NULL; 1288 } 1289 1290 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1291 { 1292 tcp_clear_retrans_hints_partial(tp); 1293 tp->retransmit_skb_hint = NULL; 1294 } 1295 1296 /* MD5 Signature */ 1297 struct crypto_hash; 1298 1299 union tcp_md5_addr { 1300 struct in_addr a4; 1301 #if IS_ENABLED(CONFIG_IPV6) 1302 struct in6_addr a6; 1303 #endif 1304 }; 1305 1306 /* - key database */ 1307 struct tcp_md5sig_key { 1308 struct hlist_node node; 1309 u8 keylen; 1310 u8 family; /* AF_INET or AF_INET6 */ 1311 union tcp_md5_addr addr; 1312 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1313 struct rcu_head rcu; 1314 }; 1315 1316 /* - sock block */ 1317 struct tcp_md5sig_info { 1318 struct hlist_head head; 1319 struct rcu_head rcu; 1320 }; 1321 1322 /* - pseudo header */ 1323 struct tcp4_pseudohdr { 1324 __be32 saddr; 1325 __be32 daddr; 1326 __u8 pad; 1327 __u8 protocol; 1328 __be16 len; 1329 }; 1330 1331 struct tcp6_pseudohdr { 1332 struct in6_addr saddr; 1333 struct in6_addr daddr; 1334 __be32 len; 1335 __be32 protocol; /* including padding */ 1336 }; 1337 1338 union tcp_md5sum_block { 1339 struct tcp4_pseudohdr ip4; 1340 #if IS_ENABLED(CONFIG_IPV6) 1341 struct tcp6_pseudohdr ip6; 1342 #endif 1343 }; 1344 1345 /* - pool: digest algorithm, hash description and scratch buffer */ 1346 struct tcp_md5sig_pool { 1347 struct hash_desc md5_desc; 1348 union tcp_md5sum_block md5_blk; 1349 }; 1350 1351 /* - functions */ 1352 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1353 const struct sock *sk, const struct sk_buff *skb); 1354 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1355 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1356 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1357 int family); 1358 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1359 const struct sock *addr_sk); 1360 1361 #ifdef CONFIG_TCP_MD5SIG 1362 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1363 const union tcp_md5_addr *addr, 1364 int family); 1365 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1366 #else 1367 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1368 const union tcp_md5_addr *addr, 1369 int family) 1370 { 1371 return NULL; 1372 } 1373 #define tcp_twsk_md5_key(twsk) NULL 1374 #endif 1375 1376 bool tcp_alloc_md5sig_pool(void); 1377 1378 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1379 static inline void tcp_put_md5sig_pool(void) 1380 { 1381 local_bh_enable(); 1382 } 1383 1384 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1385 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1386 unsigned int header_len); 1387 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1388 const struct tcp_md5sig_key *key); 1389 1390 /* From tcp_fastopen.c */ 1391 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1392 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1393 unsigned long *last_syn_loss); 1394 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1395 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1396 u16 try_exp); 1397 struct tcp_fastopen_request { 1398 /* Fast Open cookie. Size 0 means a cookie request */ 1399 struct tcp_fastopen_cookie cookie; 1400 struct msghdr *data; /* data in MSG_FASTOPEN */ 1401 size_t size; 1402 int copied; /* queued in tcp_connect() */ 1403 }; 1404 void tcp_free_fastopen_req(struct tcp_sock *tp); 1405 1406 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1407 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1408 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1409 struct request_sock *req, 1410 struct tcp_fastopen_cookie *foc, 1411 struct dst_entry *dst); 1412 void tcp_fastopen_init_key_once(bool publish); 1413 #define TCP_FASTOPEN_KEY_LENGTH 16 1414 1415 /* Fastopen key context */ 1416 struct tcp_fastopen_context { 1417 struct crypto_cipher *tfm; 1418 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1419 struct rcu_head rcu; 1420 }; 1421 1422 /* write queue abstraction */ 1423 static inline void tcp_write_queue_purge(struct sock *sk) 1424 { 1425 struct sk_buff *skb; 1426 1427 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1428 sk_wmem_free_skb(sk, skb); 1429 sk_mem_reclaim(sk); 1430 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1431 } 1432 1433 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1434 { 1435 return skb_peek(&sk->sk_write_queue); 1436 } 1437 1438 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1439 { 1440 return skb_peek_tail(&sk->sk_write_queue); 1441 } 1442 1443 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1444 const struct sk_buff *skb) 1445 { 1446 return skb_queue_next(&sk->sk_write_queue, skb); 1447 } 1448 1449 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1450 const struct sk_buff *skb) 1451 { 1452 return skb_queue_prev(&sk->sk_write_queue, skb); 1453 } 1454 1455 #define tcp_for_write_queue(skb, sk) \ 1456 skb_queue_walk(&(sk)->sk_write_queue, skb) 1457 1458 #define tcp_for_write_queue_from(skb, sk) \ 1459 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1460 1461 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1462 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1463 1464 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1465 { 1466 return sk->sk_send_head; 1467 } 1468 1469 static inline bool tcp_skb_is_last(const struct sock *sk, 1470 const struct sk_buff *skb) 1471 { 1472 return skb_queue_is_last(&sk->sk_write_queue, skb); 1473 } 1474 1475 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1476 { 1477 if (tcp_skb_is_last(sk, skb)) 1478 sk->sk_send_head = NULL; 1479 else 1480 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1481 } 1482 1483 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1484 { 1485 if (sk->sk_send_head == skb_unlinked) 1486 sk->sk_send_head = NULL; 1487 } 1488 1489 static inline void tcp_init_send_head(struct sock *sk) 1490 { 1491 sk->sk_send_head = NULL; 1492 } 1493 1494 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1495 { 1496 __skb_queue_tail(&sk->sk_write_queue, skb); 1497 } 1498 1499 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1500 { 1501 __tcp_add_write_queue_tail(sk, skb); 1502 1503 /* Queue it, remembering where we must start sending. */ 1504 if (sk->sk_send_head == NULL) { 1505 sk->sk_send_head = skb; 1506 1507 if (tcp_sk(sk)->highest_sack == NULL) 1508 tcp_sk(sk)->highest_sack = skb; 1509 } 1510 } 1511 1512 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1513 { 1514 __skb_queue_head(&sk->sk_write_queue, skb); 1515 } 1516 1517 /* Insert buff after skb on the write queue of sk. */ 1518 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1519 struct sk_buff *buff, 1520 struct sock *sk) 1521 { 1522 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1523 } 1524 1525 /* Insert new before skb on the write queue of sk. */ 1526 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1527 struct sk_buff *skb, 1528 struct sock *sk) 1529 { 1530 __skb_queue_before(&sk->sk_write_queue, skb, new); 1531 1532 if (sk->sk_send_head == skb) 1533 sk->sk_send_head = new; 1534 } 1535 1536 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1537 { 1538 __skb_unlink(skb, &sk->sk_write_queue); 1539 } 1540 1541 static inline bool tcp_write_queue_empty(struct sock *sk) 1542 { 1543 return skb_queue_empty(&sk->sk_write_queue); 1544 } 1545 1546 static inline void tcp_push_pending_frames(struct sock *sk) 1547 { 1548 if (tcp_send_head(sk)) { 1549 struct tcp_sock *tp = tcp_sk(sk); 1550 1551 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1552 } 1553 } 1554 1555 /* Start sequence of the skb just after the highest skb with SACKed 1556 * bit, valid only if sacked_out > 0 or when the caller has ensured 1557 * validity by itself. 1558 */ 1559 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1560 { 1561 if (!tp->sacked_out) 1562 return tp->snd_una; 1563 1564 if (tp->highest_sack == NULL) 1565 return tp->snd_nxt; 1566 1567 return TCP_SKB_CB(tp->highest_sack)->seq; 1568 } 1569 1570 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1571 { 1572 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1573 tcp_write_queue_next(sk, skb); 1574 } 1575 1576 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1577 { 1578 return tcp_sk(sk)->highest_sack; 1579 } 1580 1581 static inline void tcp_highest_sack_reset(struct sock *sk) 1582 { 1583 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1584 } 1585 1586 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1587 static inline void tcp_highest_sack_combine(struct sock *sk, 1588 struct sk_buff *old, 1589 struct sk_buff *new) 1590 { 1591 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1592 tcp_sk(sk)->highest_sack = new; 1593 } 1594 1595 /* Determines whether this is a thin stream (which may suffer from 1596 * increased latency). Used to trigger latency-reducing mechanisms. 1597 */ 1598 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1599 { 1600 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1601 } 1602 1603 /* /proc */ 1604 enum tcp_seq_states { 1605 TCP_SEQ_STATE_LISTENING, 1606 TCP_SEQ_STATE_OPENREQ, 1607 TCP_SEQ_STATE_ESTABLISHED, 1608 }; 1609 1610 int tcp_seq_open(struct inode *inode, struct file *file); 1611 1612 struct tcp_seq_afinfo { 1613 char *name; 1614 sa_family_t family; 1615 const struct file_operations *seq_fops; 1616 struct seq_operations seq_ops; 1617 }; 1618 1619 struct tcp_iter_state { 1620 struct seq_net_private p; 1621 sa_family_t family; 1622 enum tcp_seq_states state; 1623 struct sock *syn_wait_sk; 1624 int bucket, offset, sbucket, num; 1625 kuid_t uid; 1626 loff_t last_pos; 1627 }; 1628 1629 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1630 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1631 1632 extern struct request_sock_ops tcp_request_sock_ops; 1633 extern struct request_sock_ops tcp6_request_sock_ops; 1634 1635 void tcp_v4_destroy_sock(struct sock *sk); 1636 1637 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1638 netdev_features_t features); 1639 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1640 int tcp_gro_complete(struct sk_buff *skb); 1641 1642 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1643 1644 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1645 { 1646 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1647 } 1648 1649 static inline bool tcp_stream_memory_free(const struct sock *sk) 1650 { 1651 const struct tcp_sock *tp = tcp_sk(sk); 1652 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1653 1654 return notsent_bytes < tcp_notsent_lowat(tp); 1655 } 1656 1657 #ifdef CONFIG_PROC_FS 1658 int tcp4_proc_init(void); 1659 void tcp4_proc_exit(void); 1660 #endif 1661 1662 int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1663 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1664 const struct tcp_request_sock_ops *af_ops, 1665 struct sock *sk, struct sk_buff *skb); 1666 1667 /* TCP af-specific functions */ 1668 struct tcp_sock_af_ops { 1669 #ifdef CONFIG_TCP_MD5SIG 1670 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1671 const struct sock *addr_sk); 1672 int (*calc_md5_hash)(char *location, 1673 const struct tcp_md5sig_key *md5, 1674 const struct sock *sk, 1675 const struct sk_buff *skb); 1676 int (*md5_parse)(struct sock *sk, 1677 char __user *optval, 1678 int optlen); 1679 #endif 1680 }; 1681 1682 struct tcp_request_sock_ops { 1683 u16 mss_clamp; 1684 #ifdef CONFIG_TCP_MD5SIG 1685 struct tcp_md5sig_key *(*req_md5_lookup)(struct sock *sk, 1686 const struct sock *addr_sk); 1687 int (*calc_md5_hash) (char *location, 1688 const struct tcp_md5sig_key *md5, 1689 const struct sock *sk, 1690 const struct sk_buff *skb); 1691 #endif 1692 void (*init_req)(struct request_sock *req, struct sock *sk, 1693 struct sk_buff *skb); 1694 #ifdef CONFIG_SYN_COOKIES 1695 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1696 __u16 *mss); 1697 #endif 1698 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1699 const struct request_sock *req, 1700 bool *strict); 1701 __u32 (*init_seq)(const struct sk_buff *skb); 1702 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1703 struct flowi *fl, struct request_sock *req, 1704 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1705 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1706 const unsigned long timeout); 1707 }; 1708 1709 #ifdef CONFIG_SYN_COOKIES 1710 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1711 struct sock *sk, struct sk_buff *skb, 1712 __u16 *mss) 1713 { 1714 return ops->cookie_init_seq(sk, skb, mss); 1715 } 1716 #else 1717 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1718 struct sock *sk, struct sk_buff *skb, 1719 __u16 *mss) 1720 { 1721 return 0; 1722 } 1723 #endif 1724 1725 int tcpv4_offload_init(void); 1726 1727 void tcp_v4_init(void); 1728 void tcp_init(void); 1729 1730 /* 1731 * Save and compile IPv4 options, return a pointer to it 1732 */ 1733 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1734 { 1735 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1736 struct ip_options_rcu *dopt = NULL; 1737 1738 if (opt->optlen) { 1739 int opt_size = sizeof(*dopt) + opt->optlen; 1740 1741 dopt = kmalloc(opt_size, GFP_ATOMIC); 1742 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1743 kfree(dopt); 1744 dopt = NULL; 1745 } 1746 } 1747 return dopt; 1748 } 1749 1750 /* locally generated TCP pure ACKs have skb->truesize == 2 1751 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1752 * This is much faster than dissecting the packet to find out. 1753 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1754 */ 1755 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1756 { 1757 return skb->truesize == 2; 1758 } 1759 1760 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1761 { 1762 skb->truesize = 2; 1763 } 1764 1765 #endif /* _TCP_H */ 1766