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/cryptohash.h> 31 #include <linux/kref.h> 32 #include <linux/ktime.h> 33 34 #include <net/inet_connection_sock.h> 35 #include <net/inet_timewait_sock.h> 36 #include <net/inet_hashtables.h> 37 #include <net/checksum.h> 38 #include <net/request_sock.h> 39 #include <net/sock.h> 40 #include <net/snmp.h> 41 #include <net/ip.h> 42 #include <net/tcp_states.h> 43 #include <net/inet_ecn.h> 44 #include <net/dst.h> 45 46 #include <linux/seq_file.h> 47 #include <linux/memcontrol.h> 48 #include <linux/bpf-cgroup.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 /* Maximal number of window scale according to RFC1323 */ 83 #define TCP_MAX_WSCALE 14U 84 85 /* urg_data states */ 86 #define TCP_URG_VALID 0x0100 87 #define TCP_URG_NOTYET 0x0200 88 #define TCP_URG_READ 0x0400 89 90 #define TCP_RETR1 3 /* 91 * This is how many retries it does before it 92 * tries to figure out if the gateway is 93 * down. Minimal RFC value is 3; it corresponds 94 * to ~3sec-8min depending on RTO. 95 */ 96 97 #define TCP_RETR2 15 /* 98 * This should take at least 99 * 90 minutes to time out. 100 * RFC1122 says that the limit is 100 sec. 101 * 15 is ~13-30min depending on RTO. 102 */ 103 104 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 105 * when active opening a connection. 106 * RFC1122 says the minimum retry MUST 107 * be at least 180secs. Nevertheless 108 * this value is corresponding to 109 * 63secs of retransmission with the 110 * current initial RTO. 111 */ 112 113 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 114 * when passive opening a connection. 115 * This is corresponding to 31secs of 116 * retransmission with the current 117 * initial RTO. 118 */ 119 120 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 121 * state, about 60 seconds */ 122 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 123 /* BSD style FIN_WAIT2 deadlock breaker. 124 * It used to be 3min, new value is 60sec, 125 * to combine FIN-WAIT-2 timeout with 126 * TIME-WAIT timer. 127 */ 128 129 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 130 #if HZ >= 100 131 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 132 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 133 #else 134 #define TCP_DELACK_MIN 4U 135 #define TCP_ATO_MIN 4U 136 #endif 137 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 138 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 139 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 140 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 141 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 142 * used as a fallback RTO for the 143 * initial data transmission if no 144 * valid RTT sample has been acquired, 145 * most likely due to retrans in 3WHS. 146 */ 147 148 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 149 * for local resources. 150 */ 151 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 152 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 153 #define TCP_KEEPALIVE_INTVL (75*HZ) 154 155 #define MAX_TCP_KEEPIDLE 32767 156 #define MAX_TCP_KEEPINTVL 32767 157 #define MAX_TCP_KEEPCNT 127 158 #define MAX_TCP_SYNCNT 127 159 160 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 161 162 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 163 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 164 * after this time. It should be equal 165 * (or greater than) TCP_TIMEWAIT_LEN 166 * to provide reliability equal to one 167 * provided by timewait state. 168 */ 169 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 170 * timestamps. It must be less than 171 * minimal timewait lifetime. 172 */ 173 /* 174 * TCP option 175 */ 176 177 #define TCPOPT_NOP 1 /* Padding */ 178 #define TCPOPT_EOL 0 /* End of options */ 179 #define TCPOPT_MSS 2 /* Segment size negotiating */ 180 #define TCPOPT_WINDOW 3 /* Window scaling */ 181 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 182 #define TCPOPT_SACK 5 /* SACK Block */ 183 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 184 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 185 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 186 #define TCPOPT_EXP 254 /* Experimental */ 187 /* Magic number to be after the option value for sharing TCP 188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 189 */ 190 #define TCPOPT_FASTOPEN_MAGIC 0xF989 191 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 192 193 /* 194 * TCP option lengths 195 */ 196 197 #define TCPOLEN_MSS 4 198 #define TCPOLEN_WINDOW 3 199 #define TCPOLEN_SACK_PERM 2 200 #define TCPOLEN_TIMESTAMP 10 201 #define TCPOLEN_MD5SIG 18 202 #define TCPOLEN_FASTOPEN_BASE 2 203 #define TCPOLEN_EXP_FASTOPEN_BASE 4 204 #define TCPOLEN_EXP_SMC_BASE 6 205 206 /* But this is what stacks really send out. */ 207 #define TCPOLEN_TSTAMP_ALIGNED 12 208 #define TCPOLEN_WSCALE_ALIGNED 4 209 #define TCPOLEN_SACKPERM_ALIGNED 4 210 #define TCPOLEN_SACK_BASE 2 211 #define TCPOLEN_SACK_BASE_ALIGNED 4 212 #define TCPOLEN_SACK_PERBLOCK 8 213 #define TCPOLEN_MD5SIG_ALIGNED 20 214 #define TCPOLEN_MSS_ALIGNED 4 215 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 216 217 /* Flags in tp->nonagle */ 218 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 219 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 220 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 221 222 /* TCP thin-stream limits */ 223 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 224 225 /* TCP initial congestion window as per rfc6928 */ 226 #define TCP_INIT_CWND 10 227 228 /* Bit Flags for sysctl_tcp_fastopen */ 229 #define TFO_CLIENT_ENABLE 1 230 #define TFO_SERVER_ENABLE 2 231 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 232 233 /* Accept SYN data w/o any cookie option */ 234 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 235 236 /* Force enable TFO on all listeners, i.e., not requiring the 237 * TCP_FASTOPEN socket option. 238 */ 239 #define TFO_SERVER_WO_SOCKOPT1 0x400 240 241 242 /* sysctl variables for tcp */ 243 extern int sysctl_tcp_max_orphans; 244 extern long sysctl_tcp_mem[3]; 245 extern int sysctl_tcp_wmem[3]; 246 extern int sysctl_tcp_rmem[3]; 247 248 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 249 250 extern atomic_long_t tcp_memory_allocated; 251 extern struct percpu_counter tcp_sockets_allocated; 252 extern unsigned long tcp_memory_pressure; 253 254 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 255 static inline bool tcp_under_memory_pressure(const struct sock *sk) 256 { 257 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 258 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 259 return true; 260 261 return tcp_memory_pressure; 262 } 263 /* 264 * The next routines deal with comparing 32 bit unsigned ints 265 * and worry about wraparound (automatic with unsigned arithmetic). 266 */ 267 268 static inline bool before(__u32 seq1, __u32 seq2) 269 { 270 return (__s32)(seq1-seq2) < 0; 271 } 272 #define after(seq2, seq1) before(seq1, seq2) 273 274 /* is s2<=s1<=s3 ? */ 275 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 276 { 277 return seq3 - seq2 >= seq1 - seq2; 278 } 279 280 static inline bool tcp_out_of_memory(struct sock *sk) 281 { 282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 283 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 284 return true; 285 return false; 286 } 287 288 void sk_forced_mem_schedule(struct sock *sk, int size); 289 290 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 291 { 292 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 293 int orphans = percpu_counter_read_positive(ocp); 294 295 if (orphans << shift > sysctl_tcp_max_orphans) { 296 orphans = percpu_counter_sum_positive(ocp); 297 if (orphans << shift > sysctl_tcp_max_orphans) 298 return true; 299 } 300 return false; 301 } 302 303 bool tcp_check_oom(struct sock *sk, int shift); 304 305 306 extern struct proto tcp_prot; 307 308 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 309 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 310 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 311 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 312 313 void tcp_tasklet_init(void); 314 315 void tcp_v4_err(struct sk_buff *skb, u32); 316 317 void tcp_shutdown(struct sock *sk, int how); 318 319 int tcp_v4_early_demux(struct sk_buff *skb); 320 int tcp_v4_rcv(struct sk_buff *skb); 321 322 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 323 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 324 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 325 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 326 int flags); 327 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 328 size_t size, int flags); 329 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 330 size_t size, int flags); 331 void tcp_release_cb(struct sock *sk); 332 void tcp_wfree(struct sk_buff *skb); 333 void tcp_write_timer_handler(struct sock *sk); 334 void tcp_delack_timer_handler(struct sock *sk); 335 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 336 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 337 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 338 const struct tcphdr *th); 339 void tcp_rcv_space_adjust(struct sock *sk); 340 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 341 void tcp_twsk_destructor(struct sock *sk); 342 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 343 struct pipe_inode_info *pipe, size_t len, 344 unsigned int flags); 345 346 static inline void tcp_dec_quickack_mode(struct sock *sk, 347 const unsigned int pkts) 348 { 349 struct inet_connection_sock *icsk = inet_csk(sk); 350 351 if (icsk->icsk_ack.quick) { 352 if (pkts >= icsk->icsk_ack.quick) { 353 icsk->icsk_ack.quick = 0; 354 /* Leaving quickack mode we deflate ATO. */ 355 icsk->icsk_ack.ato = TCP_ATO_MIN; 356 } else 357 icsk->icsk_ack.quick -= pkts; 358 } 359 } 360 361 #define TCP_ECN_OK 1 362 #define TCP_ECN_QUEUE_CWR 2 363 #define TCP_ECN_DEMAND_CWR 4 364 #define TCP_ECN_SEEN 8 365 366 enum tcp_tw_status { 367 TCP_TW_SUCCESS = 0, 368 TCP_TW_RST = 1, 369 TCP_TW_ACK = 2, 370 TCP_TW_SYN = 3 371 }; 372 373 374 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 375 struct sk_buff *skb, 376 const struct tcphdr *th); 377 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 378 struct request_sock *req, bool fastopen); 379 int tcp_child_process(struct sock *parent, struct sock *child, 380 struct sk_buff *skb); 381 void tcp_enter_loss(struct sock *sk); 382 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 383 void tcp_clear_retrans(struct tcp_sock *tp); 384 void tcp_update_metrics(struct sock *sk); 385 void tcp_init_metrics(struct sock *sk); 386 void tcp_metrics_init(void); 387 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 388 void tcp_disable_fack(struct tcp_sock *tp); 389 void tcp_close(struct sock *sk, long timeout); 390 void tcp_init_sock(struct sock *sk); 391 void tcp_init_transfer(struct sock *sk, int bpf_op); 392 unsigned int tcp_poll(struct file *file, struct socket *sock, 393 struct poll_table_struct *wait); 394 int tcp_getsockopt(struct sock *sk, int level, int optname, 395 char __user *optval, int __user *optlen); 396 int tcp_setsockopt(struct sock *sk, int level, int optname, 397 char __user *optval, unsigned int optlen); 398 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 399 char __user *optval, int __user *optlen); 400 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 401 char __user *optval, unsigned int optlen); 402 void tcp_set_keepalive(struct sock *sk, int val); 403 void tcp_syn_ack_timeout(const struct request_sock *req); 404 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 405 int flags, int *addr_len); 406 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 407 struct tcp_options_received *opt_rx, 408 int estab, struct tcp_fastopen_cookie *foc); 409 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 410 411 /* 412 * TCP v4 functions exported for the inet6 API 413 */ 414 415 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 416 void tcp_v4_mtu_reduced(struct sock *sk); 417 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 418 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 419 struct sock *tcp_create_openreq_child(const struct sock *sk, 420 struct request_sock *req, 421 struct sk_buff *skb); 422 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 423 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 424 struct request_sock *req, 425 struct dst_entry *dst, 426 struct request_sock *req_unhash, 427 bool *own_req); 428 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 429 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 430 int tcp_connect(struct sock *sk); 431 enum tcp_synack_type { 432 TCP_SYNACK_NORMAL, 433 TCP_SYNACK_FASTOPEN, 434 TCP_SYNACK_COOKIE, 435 }; 436 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 437 struct request_sock *req, 438 struct tcp_fastopen_cookie *foc, 439 enum tcp_synack_type synack_type); 440 int tcp_disconnect(struct sock *sk, int flags); 441 442 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 443 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 444 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 445 446 /* From syncookies.c */ 447 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 448 struct request_sock *req, 449 struct dst_entry *dst, u32 tsoff); 450 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 451 u32 cookie); 452 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 453 #ifdef CONFIG_SYN_COOKIES 454 455 /* Syncookies use a monotonic timer which increments every 60 seconds. 456 * This counter is used both as a hash input and partially encoded into 457 * the cookie value. A cookie is only validated further if the delta 458 * between the current counter value and the encoded one is less than this, 459 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 460 * the counter advances immediately after a cookie is generated). 461 */ 462 #define MAX_SYNCOOKIE_AGE 2 463 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 464 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 465 466 /* syncookies: remember time of last synqueue overflow 467 * But do not dirty this field too often (once per second is enough) 468 * It is racy as we do not hold a lock, but race is very minor. 469 */ 470 static inline void tcp_synq_overflow(const struct sock *sk) 471 { 472 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 473 unsigned long now = jiffies; 474 475 if (time_after(now, last_overflow + HZ)) 476 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 477 } 478 479 /* syncookies: no recent synqueue overflow on this listening socket? */ 480 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 481 { 482 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 483 484 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 485 } 486 487 static inline u32 tcp_cookie_time(void) 488 { 489 u64 val = get_jiffies_64(); 490 491 do_div(val, TCP_SYNCOOKIE_PERIOD); 492 return val; 493 } 494 495 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 496 u16 *mssp); 497 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 498 u64 cookie_init_timestamp(struct request_sock *req); 499 bool cookie_timestamp_decode(const struct net *net, 500 struct tcp_options_received *opt); 501 bool cookie_ecn_ok(const struct tcp_options_received *opt, 502 const struct net *net, const struct dst_entry *dst); 503 504 /* From net/ipv6/syncookies.c */ 505 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 506 u32 cookie); 507 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 508 509 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 510 const struct tcphdr *th, u16 *mssp); 511 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 512 #endif 513 /* tcp_output.c */ 514 515 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 516 int min_tso_segs); 517 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 518 int nonagle); 519 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 520 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 521 void tcp_retransmit_timer(struct sock *sk); 522 void tcp_xmit_retransmit_queue(struct sock *); 523 void tcp_simple_retransmit(struct sock *); 524 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 525 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 526 enum tcp_queue { 527 TCP_FRAG_IN_WRITE_QUEUE, 528 TCP_FRAG_IN_RTX_QUEUE, 529 }; 530 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 531 struct sk_buff *skb, u32 len, 532 unsigned int mss_now, gfp_t gfp); 533 534 void tcp_send_probe0(struct sock *); 535 void tcp_send_partial(struct sock *); 536 int tcp_write_wakeup(struct sock *, int mib); 537 void tcp_send_fin(struct sock *sk); 538 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 539 int tcp_send_synack(struct sock *); 540 void tcp_push_one(struct sock *, unsigned int mss_now); 541 void tcp_send_ack(struct sock *sk); 542 void tcp_send_delayed_ack(struct sock *sk); 543 void tcp_send_loss_probe(struct sock *sk); 544 bool tcp_schedule_loss_probe(struct sock *sk); 545 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 546 const struct sk_buff *next_skb); 547 548 /* tcp_input.c */ 549 void tcp_rearm_rto(struct sock *sk); 550 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 551 void tcp_reset(struct sock *sk); 552 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 553 void tcp_fin(struct sock *sk); 554 555 /* tcp_timer.c */ 556 void tcp_init_xmit_timers(struct sock *); 557 static inline void tcp_clear_xmit_timers(struct sock *sk) 558 { 559 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 560 inet_csk_clear_xmit_timers(sk); 561 } 562 563 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 564 unsigned int tcp_current_mss(struct sock *sk); 565 566 /* Bound MSS / TSO packet size with the half of the window */ 567 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 568 { 569 int cutoff; 570 571 /* When peer uses tiny windows, there is no use in packetizing 572 * to sub-MSS pieces for the sake of SWS or making sure there 573 * are enough packets in the pipe for fast recovery. 574 * 575 * On the other hand, for extremely large MSS devices, handling 576 * smaller than MSS windows in this way does make sense. 577 */ 578 if (tp->max_window > TCP_MSS_DEFAULT) 579 cutoff = (tp->max_window >> 1); 580 else 581 cutoff = tp->max_window; 582 583 if (cutoff && pktsize > cutoff) 584 return max_t(int, cutoff, 68U - tp->tcp_header_len); 585 else 586 return pktsize; 587 } 588 589 /* tcp.c */ 590 void tcp_get_info(struct sock *, struct tcp_info *); 591 592 /* Read 'sendfile()'-style from a TCP socket */ 593 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 594 sk_read_actor_t recv_actor); 595 596 void tcp_initialize_rcv_mss(struct sock *sk); 597 598 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 599 int tcp_mss_to_mtu(struct sock *sk, int mss); 600 void tcp_mtup_init(struct sock *sk); 601 void tcp_init_buffer_space(struct sock *sk); 602 603 static inline void tcp_bound_rto(const struct sock *sk) 604 { 605 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 606 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 607 } 608 609 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 610 { 611 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 612 } 613 614 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 615 { 616 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 617 ntohl(TCP_FLAG_ACK) | 618 snd_wnd); 619 } 620 621 static inline void tcp_fast_path_on(struct tcp_sock *tp) 622 { 623 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 624 } 625 626 static inline void tcp_fast_path_check(struct sock *sk) 627 { 628 struct tcp_sock *tp = tcp_sk(sk); 629 630 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 631 tp->rcv_wnd && 632 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 633 !tp->urg_data) 634 tcp_fast_path_on(tp); 635 } 636 637 /* Compute the actual rto_min value */ 638 static inline u32 tcp_rto_min(struct sock *sk) 639 { 640 const struct dst_entry *dst = __sk_dst_get(sk); 641 u32 rto_min = TCP_RTO_MIN; 642 643 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 644 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 645 return rto_min; 646 } 647 648 static inline u32 tcp_rto_min_us(struct sock *sk) 649 { 650 return jiffies_to_usecs(tcp_rto_min(sk)); 651 } 652 653 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 654 { 655 return dst_metric_locked(dst, RTAX_CC_ALGO); 656 } 657 658 /* Minimum RTT in usec. ~0 means not available. */ 659 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 660 { 661 return minmax_get(&tp->rtt_min); 662 } 663 664 /* Compute the actual receive window we are currently advertising. 665 * Rcv_nxt can be after the window if our peer push more data 666 * than the offered window. 667 */ 668 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 669 { 670 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 671 672 if (win < 0) 673 win = 0; 674 return (u32) win; 675 } 676 677 /* Choose a new window, without checks for shrinking, and without 678 * scaling applied to the result. The caller does these things 679 * if necessary. This is a "raw" window selection. 680 */ 681 u32 __tcp_select_window(struct sock *sk); 682 683 void tcp_send_window_probe(struct sock *sk); 684 685 /* TCP uses 32bit jiffies to save some space. 686 * Note that this is different from tcp_time_stamp, which 687 * historically has been the same until linux-4.13. 688 */ 689 #define tcp_jiffies32 ((u32)jiffies) 690 691 /* 692 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 693 * It is no longer tied to jiffies, but to 1 ms clock. 694 * Note: double check if you want to use tcp_jiffies32 instead of this. 695 */ 696 #define TCP_TS_HZ 1000 697 698 static inline u64 tcp_clock_ns(void) 699 { 700 return local_clock(); 701 } 702 703 static inline u64 tcp_clock_us(void) 704 { 705 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 706 } 707 708 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 709 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 710 { 711 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 712 } 713 714 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 715 static inline u32 tcp_time_stamp_raw(void) 716 { 717 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 718 } 719 720 721 /* Refresh 1us clock of a TCP socket, 722 * ensuring monotically increasing values. 723 */ 724 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 725 { 726 u64 val = tcp_clock_us(); 727 728 if (val > tp->tcp_mstamp) 729 tp->tcp_mstamp = val; 730 } 731 732 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 733 { 734 return max_t(s64, t1 - t0, 0); 735 } 736 737 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 738 { 739 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 740 } 741 742 743 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 744 745 #define TCPHDR_FIN 0x01 746 #define TCPHDR_SYN 0x02 747 #define TCPHDR_RST 0x04 748 #define TCPHDR_PSH 0x08 749 #define TCPHDR_ACK 0x10 750 #define TCPHDR_URG 0x20 751 #define TCPHDR_ECE 0x40 752 #define TCPHDR_CWR 0x80 753 754 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 755 756 /* This is what the send packet queuing engine uses to pass 757 * TCP per-packet control information to the transmission code. 758 * We also store the host-order sequence numbers in here too. 759 * This is 44 bytes if IPV6 is enabled. 760 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 761 */ 762 struct tcp_skb_cb { 763 __u32 seq; /* Starting sequence number */ 764 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 765 union { 766 /* Note : tcp_tw_isn is used in input path only 767 * (isn chosen by tcp_timewait_state_process()) 768 * 769 * tcp_gso_segs/size are used in write queue only, 770 * cf tcp_skb_pcount()/tcp_skb_mss() 771 */ 772 __u32 tcp_tw_isn; 773 struct { 774 u16 tcp_gso_segs; 775 u16 tcp_gso_size; 776 }; 777 }; 778 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 779 780 __u8 sacked; /* State flags for SACK/FACK. */ 781 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 782 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 783 #define TCPCB_LOST 0x04 /* SKB is lost */ 784 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 785 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 786 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 787 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 788 TCPCB_REPAIRED) 789 790 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 791 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 792 eor:1, /* Is skb MSG_EOR marked? */ 793 has_rxtstamp:1, /* SKB has a RX timestamp */ 794 unused:5; 795 __u32 ack_seq; /* Sequence number ACK'd */ 796 union { 797 struct { 798 /* There is space for up to 24 bytes */ 799 __u32 in_flight:30,/* Bytes in flight at transmit */ 800 is_app_limited:1, /* cwnd not fully used? */ 801 unused:1; 802 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 803 __u32 delivered; 804 /* start of send pipeline phase */ 805 u64 first_tx_mstamp; 806 /* when we reached the "delivered" count */ 807 u64 delivered_mstamp; 808 } tx; /* only used for outgoing skbs */ 809 union { 810 struct inet_skb_parm h4; 811 #if IS_ENABLED(CONFIG_IPV6) 812 struct inet6_skb_parm h6; 813 #endif 814 } header; /* For incoming skbs */ 815 struct { 816 __u32 key; 817 __u32 flags; 818 struct bpf_map *map; 819 void *data_end; 820 } bpf; 821 }; 822 }; 823 824 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 825 826 827 #if IS_ENABLED(CONFIG_IPV6) 828 /* This is the variant of inet6_iif() that must be used by TCP, 829 * as TCP moves IP6CB into a different location in skb->cb[] 830 */ 831 static inline int tcp_v6_iif(const struct sk_buff *skb) 832 { 833 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 834 835 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 836 } 837 838 /* TCP_SKB_CB reference means this can not be used from early demux */ 839 static inline int tcp_v6_sdif(const struct sk_buff *skb) 840 { 841 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 842 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 843 return TCP_SKB_CB(skb)->header.h6.iif; 844 #endif 845 return 0; 846 } 847 #endif 848 849 /* TCP_SKB_CB reference means this can not be used from early demux */ 850 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 851 { 852 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 853 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 854 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 855 return true; 856 #endif 857 return false; 858 } 859 860 /* TCP_SKB_CB reference means this can not be used from early demux */ 861 static inline int tcp_v4_sdif(struct sk_buff *skb) 862 { 863 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 864 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 865 return TCP_SKB_CB(skb)->header.h4.iif; 866 #endif 867 return 0; 868 } 869 870 /* Due to TSO, an SKB can be composed of multiple actual 871 * packets. To keep these tracked properly, we use this. 872 */ 873 static inline int tcp_skb_pcount(const struct sk_buff *skb) 874 { 875 return TCP_SKB_CB(skb)->tcp_gso_segs; 876 } 877 878 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 879 { 880 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 881 } 882 883 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 884 { 885 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 886 } 887 888 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 889 static inline int tcp_skb_mss(const struct sk_buff *skb) 890 { 891 return TCP_SKB_CB(skb)->tcp_gso_size; 892 } 893 894 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 895 { 896 return likely(!TCP_SKB_CB(skb)->eor); 897 } 898 899 /* Events passed to congestion control interface */ 900 enum tcp_ca_event { 901 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 902 CA_EVENT_CWND_RESTART, /* congestion window restart */ 903 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 904 CA_EVENT_LOSS, /* loss timeout */ 905 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 906 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 907 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 908 CA_EVENT_NON_DELAYED_ACK, 909 }; 910 911 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 912 enum tcp_ca_ack_event_flags { 913 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 914 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 915 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 916 }; 917 918 /* 919 * Interface for adding new TCP congestion control handlers 920 */ 921 #define TCP_CA_NAME_MAX 16 922 #define TCP_CA_MAX 128 923 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 924 925 #define TCP_CA_UNSPEC 0 926 927 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 928 #define TCP_CONG_NON_RESTRICTED 0x1 929 /* Requires ECN/ECT set on all packets */ 930 #define TCP_CONG_NEEDS_ECN 0x2 931 932 union tcp_cc_info; 933 934 struct ack_sample { 935 u32 pkts_acked; 936 s32 rtt_us; 937 u32 in_flight; 938 }; 939 940 /* A rate sample measures the number of (original/retransmitted) data 941 * packets delivered "delivered" over an interval of time "interval_us". 942 * The tcp_rate.c code fills in the rate sample, and congestion 943 * control modules that define a cong_control function to run at the end 944 * of ACK processing can optionally chose to consult this sample when 945 * setting cwnd and pacing rate. 946 * A sample is invalid if "delivered" or "interval_us" is negative. 947 */ 948 struct rate_sample { 949 u64 prior_mstamp; /* starting timestamp for interval */ 950 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 951 s32 delivered; /* number of packets delivered over interval */ 952 long interval_us; /* time for tp->delivered to incr "delivered" */ 953 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 954 int losses; /* number of packets marked lost upon ACK */ 955 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 956 u32 prior_in_flight; /* in flight before this ACK */ 957 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 958 bool is_retrans; /* is sample from retransmission? */ 959 }; 960 961 struct tcp_congestion_ops { 962 struct list_head list; 963 u32 key; 964 u32 flags; 965 966 /* initialize private data (optional) */ 967 void (*init)(struct sock *sk); 968 /* cleanup private data (optional) */ 969 void (*release)(struct sock *sk); 970 971 /* return slow start threshold (required) */ 972 u32 (*ssthresh)(struct sock *sk); 973 /* do new cwnd calculation (required) */ 974 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 975 /* call before changing ca_state (optional) */ 976 void (*set_state)(struct sock *sk, u8 new_state); 977 /* call when cwnd event occurs (optional) */ 978 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 979 /* call when ack arrives (optional) */ 980 void (*in_ack_event)(struct sock *sk, u32 flags); 981 /* new value of cwnd after loss (required) */ 982 u32 (*undo_cwnd)(struct sock *sk); 983 /* hook for packet ack accounting (optional) */ 984 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 985 /* suggest number of segments for each skb to transmit (optional) */ 986 u32 (*tso_segs_goal)(struct sock *sk); 987 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 988 u32 (*sndbuf_expand)(struct sock *sk); 989 /* call when packets are delivered to update cwnd and pacing rate, 990 * after all the ca_state processing. (optional) 991 */ 992 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 993 /* get info for inet_diag (optional) */ 994 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 995 union tcp_cc_info *info); 996 997 char name[TCP_CA_NAME_MAX]; 998 struct module *owner; 999 }; 1000 1001 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1002 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1003 1004 void tcp_assign_congestion_control(struct sock *sk); 1005 void tcp_init_congestion_control(struct sock *sk); 1006 void tcp_cleanup_congestion_control(struct sock *sk); 1007 int tcp_set_default_congestion_control(const char *name); 1008 void tcp_get_default_congestion_control(char *name); 1009 void tcp_get_available_congestion_control(char *buf, size_t len); 1010 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1011 int tcp_set_allowed_congestion_control(char *allowed); 1012 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1013 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1014 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1015 1016 u32 tcp_reno_ssthresh(struct sock *sk); 1017 u32 tcp_reno_undo_cwnd(struct sock *sk); 1018 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1019 extern struct tcp_congestion_ops tcp_reno; 1020 1021 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1022 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1023 #ifdef CONFIG_INET 1024 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1025 #else 1026 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1027 { 1028 return NULL; 1029 } 1030 #endif 1031 1032 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1033 { 1034 const struct inet_connection_sock *icsk = inet_csk(sk); 1035 1036 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1037 } 1038 1039 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1040 { 1041 struct inet_connection_sock *icsk = inet_csk(sk); 1042 1043 if (icsk->icsk_ca_ops->set_state) 1044 icsk->icsk_ca_ops->set_state(sk, ca_state); 1045 icsk->icsk_ca_state = ca_state; 1046 } 1047 1048 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1049 { 1050 const struct inet_connection_sock *icsk = inet_csk(sk); 1051 1052 if (icsk->icsk_ca_ops->cwnd_event) 1053 icsk->icsk_ca_ops->cwnd_event(sk, event); 1054 } 1055 1056 /* From tcp_rate.c */ 1057 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1058 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1059 struct rate_sample *rs); 1060 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1061 struct rate_sample *rs); 1062 void tcp_rate_check_app_limited(struct sock *sk); 1063 1064 /* These functions determine how the current flow behaves in respect of SACK 1065 * handling. SACK is negotiated with the peer, and therefore it can vary 1066 * between different flows. 1067 * 1068 * tcp_is_sack - SACK enabled 1069 * tcp_is_reno - No SACK 1070 * tcp_is_fack - FACK enabled, implies SACK enabled 1071 */ 1072 static inline int tcp_is_sack(const struct tcp_sock *tp) 1073 { 1074 return tp->rx_opt.sack_ok; 1075 } 1076 1077 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1078 { 1079 return !tcp_is_sack(tp); 1080 } 1081 1082 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1083 { 1084 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1085 } 1086 1087 static inline void tcp_enable_fack(struct tcp_sock *tp) 1088 { 1089 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1090 } 1091 1092 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1093 { 1094 return tp->sacked_out + tp->lost_out; 1095 } 1096 1097 /* This determines how many packets are "in the network" to the best 1098 * of our knowledge. In many cases it is conservative, but where 1099 * detailed information is available from the receiver (via SACK 1100 * blocks etc.) we can make more aggressive calculations. 1101 * 1102 * Use this for decisions involving congestion control, use just 1103 * tp->packets_out to determine if the send queue is empty or not. 1104 * 1105 * Read this equation as: 1106 * 1107 * "Packets sent once on transmission queue" MINUS 1108 * "Packets left network, but not honestly ACKed yet" PLUS 1109 * "Packets fast retransmitted" 1110 */ 1111 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1112 { 1113 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1114 } 1115 1116 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1117 1118 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1119 { 1120 return tp->snd_cwnd < tp->snd_ssthresh; 1121 } 1122 1123 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1124 { 1125 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1126 } 1127 1128 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1129 { 1130 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1131 (1 << inet_csk(sk)->icsk_ca_state); 1132 } 1133 1134 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1135 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1136 * ssthresh. 1137 */ 1138 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1139 { 1140 const struct tcp_sock *tp = tcp_sk(sk); 1141 1142 if (tcp_in_cwnd_reduction(sk)) 1143 return tp->snd_ssthresh; 1144 else 1145 return max(tp->snd_ssthresh, 1146 ((tp->snd_cwnd >> 1) + 1147 (tp->snd_cwnd >> 2))); 1148 } 1149 1150 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1151 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1152 1153 void tcp_enter_cwr(struct sock *sk); 1154 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1155 1156 /* The maximum number of MSS of available cwnd for which TSO defers 1157 * sending if not using sysctl_tcp_tso_win_divisor. 1158 */ 1159 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1160 { 1161 return 3; 1162 } 1163 1164 /* Returns end sequence number of the receiver's advertised window */ 1165 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1166 { 1167 return tp->snd_una + tp->snd_wnd; 1168 } 1169 1170 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1171 * flexible approach. The RFC suggests cwnd should not be raised unless 1172 * it was fully used previously. And that's exactly what we do in 1173 * congestion avoidance mode. But in slow start we allow cwnd to grow 1174 * as long as the application has used half the cwnd. 1175 * Example : 1176 * cwnd is 10 (IW10), but application sends 9 frames. 1177 * We allow cwnd to reach 18 when all frames are ACKed. 1178 * This check is safe because it's as aggressive as slow start which already 1179 * risks 100% overshoot. The advantage is that we discourage application to 1180 * either send more filler packets or data to artificially blow up the cwnd 1181 * usage, and allow application-limited process to probe bw more aggressively. 1182 */ 1183 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1184 { 1185 const struct tcp_sock *tp = tcp_sk(sk); 1186 1187 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1188 if (tcp_in_slow_start(tp)) 1189 return tp->snd_cwnd < 2 * tp->max_packets_out; 1190 1191 return tp->is_cwnd_limited; 1192 } 1193 1194 /* Something is really bad, we could not queue an additional packet, 1195 * because qdisc is full or receiver sent a 0 window. 1196 * We do not want to add fuel to the fire, or abort too early, 1197 * so make sure the timer we arm now is at least 200ms in the future, 1198 * regardless of current icsk_rto value (as it could be ~2ms) 1199 */ 1200 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1201 { 1202 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1203 } 1204 1205 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1206 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1207 unsigned long max_when) 1208 { 1209 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1210 1211 return (unsigned long)min_t(u64, when, max_when); 1212 } 1213 1214 static inline void tcp_check_probe_timer(struct sock *sk) 1215 { 1216 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1217 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1218 tcp_probe0_base(sk), TCP_RTO_MAX); 1219 } 1220 1221 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1222 { 1223 tp->snd_wl1 = seq; 1224 } 1225 1226 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1227 { 1228 tp->snd_wl1 = seq; 1229 } 1230 1231 /* 1232 * Calculate(/check) TCP checksum 1233 */ 1234 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1235 __be32 daddr, __wsum base) 1236 { 1237 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1238 } 1239 1240 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1241 { 1242 return __skb_checksum_complete(skb); 1243 } 1244 1245 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1246 { 1247 return !skb_csum_unnecessary(skb) && 1248 __tcp_checksum_complete(skb); 1249 } 1250 1251 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1252 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1253 1254 #undef STATE_TRACE 1255 1256 #ifdef STATE_TRACE 1257 static const char *statename[]={ 1258 "Unused","Established","Syn Sent","Syn Recv", 1259 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1260 "Close Wait","Last ACK","Listen","Closing" 1261 }; 1262 #endif 1263 void tcp_set_state(struct sock *sk, int state); 1264 1265 void tcp_done(struct sock *sk); 1266 1267 int tcp_abort(struct sock *sk, int err); 1268 1269 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1270 { 1271 rx_opt->dsack = 0; 1272 rx_opt->num_sacks = 0; 1273 } 1274 1275 u32 tcp_default_init_rwnd(u32 mss); 1276 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1277 1278 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1279 { 1280 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1281 struct tcp_sock *tp = tcp_sk(sk); 1282 s32 delta; 1283 1284 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1285 ca_ops->cong_control) 1286 return; 1287 delta = tcp_jiffies32 - tp->lsndtime; 1288 if (delta > inet_csk(sk)->icsk_rto) 1289 tcp_cwnd_restart(sk, delta); 1290 } 1291 1292 /* Determine a window scaling and initial window to offer. */ 1293 void tcp_select_initial_window(const struct sock *sk, int __space, 1294 __u32 mss, __u32 *rcv_wnd, 1295 __u32 *window_clamp, int wscale_ok, 1296 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1297 1298 static inline int tcp_win_from_space(const struct sock *sk, int space) 1299 { 1300 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1301 1302 return tcp_adv_win_scale <= 0 ? 1303 (space>>(-tcp_adv_win_scale)) : 1304 space - (space>>tcp_adv_win_scale); 1305 } 1306 1307 /* Note: caller must be prepared to deal with negative returns */ 1308 static inline int tcp_space(const struct sock *sk) 1309 { 1310 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1311 atomic_read(&sk->sk_rmem_alloc)); 1312 } 1313 1314 static inline int tcp_full_space(const struct sock *sk) 1315 { 1316 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1317 } 1318 1319 extern void tcp_openreq_init_rwin(struct request_sock *req, 1320 const struct sock *sk_listener, 1321 const struct dst_entry *dst); 1322 1323 void tcp_enter_memory_pressure(struct sock *sk); 1324 void tcp_leave_memory_pressure(struct sock *sk); 1325 1326 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1327 { 1328 struct net *net = sock_net((struct sock *)tp); 1329 1330 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1331 } 1332 1333 static inline int keepalive_time_when(const struct tcp_sock *tp) 1334 { 1335 struct net *net = sock_net((struct sock *)tp); 1336 1337 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1338 } 1339 1340 static inline int keepalive_probes(const struct tcp_sock *tp) 1341 { 1342 struct net *net = sock_net((struct sock *)tp); 1343 1344 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1345 } 1346 1347 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1348 { 1349 const struct inet_connection_sock *icsk = &tp->inet_conn; 1350 1351 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1352 tcp_jiffies32 - tp->rcv_tstamp); 1353 } 1354 1355 static inline int tcp_fin_time(const struct sock *sk) 1356 { 1357 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1358 const int rto = inet_csk(sk)->icsk_rto; 1359 1360 if (fin_timeout < (rto << 2) - (rto >> 1)) 1361 fin_timeout = (rto << 2) - (rto >> 1); 1362 1363 return fin_timeout; 1364 } 1365 1366 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1367 int paws_win) 1368 { 1369 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1370 return true; 1371 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1372 return true; 1373 /* 1374 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1375 * then following tcp messages have valid values. Ignore 0 value, 1376 * or else 'negative' tsval might forbid us to accept their packets. 1377 */ 1378 if (!rx_opt->ts_recent) 1379 return true; 1380 return false; 1381 } 1382 1383 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1384 int rst) 1385 { 1386 if (tcp_paws_check(rx_opt, 0)) 1387 return false; 1388 1389 /* RST segments are not recommended to carry timestamp, 1390 and, if they do, it is recommended to ignore PAWS because 1391 "their cleanup function should take precedence over timestamps." 1392 Certainly, it is mistake. It is necessary to understand the reasons 1393 of this constraint to relax it: if peer reboots, clock may go 1394 out-of-sync and half-open connections will not be reset. 1395 Actually, the problem would be not existing if all 1396 the implementations followed draft about maintaining clock 1397 via reboots. Linux-2.2 DOES NOT! 1398 1399 However, we can relax time bounds for RST segments to MSL. 1400 */ 1401 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1402 return false; 1403 return true; 1404 } 1405 1406 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1407 int mib_idx, u32 *last_oow_ack_time); 1408 1409 static inline void tcp_mib_init(struct net *net) 1410 { 1411 /* See RFC 2012 */ 1412 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1413 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1414 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1415 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1416 } 1417 1418 /* from STCP */ 1419 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1420 { 1421 tp->lost_skb_hint = NULL; 1422 } 1423 1424 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1425 { 1426 tcp_clear_retrans_hints_partial(tp); 1427 tp->retransmit_skb_hint = NULL; 1428 } 1429 1430 union tcp_md5_addr { 1431 struct in_addr a4; 1432 #if IS_ENABLED(CONFIG_IPV6) 1433 struct in6_addr a6; 1434 #endif 1435 }; 1436 1437 /* - key database */ 1438 struct tcp_md5sig_key { 1439 struct hlist_node node; 1440 u8 keylen; 1441 u8 family; /* AF_INET or AF_INET6 */ 1442 union tcp_md5_addr addr; 1443 u8 prefixlen; 1444 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1445 struct rcu_head rcu; 1446 }; 1447 1448 /* - sock block */ 1449 struct tcp_md5sig_info { 1450 struct hlist_head head; 1451 struct rcu_head rcu; 1452 }; 1453 1454 /* - pseudo header */ 1455 struct tcp4_pseudohdr { 1456 __be32 saddr; 1457 __be32 daddr; 1458 __u8 pad; 1459 __u8 protocol; 1460 __be16 len; 1461 }; 1462 1463 struct tcp6_pseudohdr { 1464 struct in6_addr saddr; 1465 struct in6_addr daddr; 1466 __be32 len; 1467 __be32 protocol; /* including padding */ 1468 }; 1469 1470 union tcp_md5sum_block { 1471 struct tcp4_pseudohdr ip4; 1472 #if IS_ENABLED(CONFIG_IPV6) 1473 struct tcp6_pseudohdr ip6; 1474 #endif 1475 }; 1476 1477 /* - pool: digest algorithm, hash description and scratch buffer */ 1478 struct tcp_md5sig_pool { 1479 struct ahash_request *md5_req; 1480 void *scratch; 1481 }; 1482 1483 /* - functions */ 1484 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1485 const struct sock *sk, const struct sk_buff *skb); 1486 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1487 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1488 gfp_t gfp); 1489 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1490 int family, u8 prefixlen); 1491 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1492 const struct sock *addr_sk); 1493 1494 #ifdef CONFIG_TCP_MD5SIG 1495 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1496 const union tcp_md5_addr *addr, 1497 int family); 1498 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1499 #else 1500 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1501 const union tcp_md5_addr *addr, 1502 int family) 1503 { 1504 return NULL; 1505 } 1506 #define tcp_twsk_md5_key(twsk) NULL 1507 #endif 1508 1509 bool tcp_alloc_md5sig_pool(void); 1510 1511 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1512 static inline void tcp_put_md5sig_pool(void) 1513 { 1514 local_bh_enable(); 1515 } 1516 1517 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1518 unsigned int header_len); 1519 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1520 const struct tcp_md5sig_key *key); 1521 1522 /* From tcp_fastopen.c */ 1523 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1524 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1525 unsigned long *last_syn_loss); 1526 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1527 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1528 u16 try_exp); 1529 struct tcp_fastopen_request { 1530 /* Fast Open cookie. Size 0 means a cookie request */ 1531 struct tcp_fastopen_cookie cookie; 1532 struct msghdr *data; /* data in MSG_FASTOPEN */ 1533 size_t size; 1534 int copied; /* queued in tcp_connect() */ 1535 }; 1536 void tcp_free_fastopen_req(struct tcp_sock *tp); 1537 void tcp_fastopen_destroy_cipher(struct sock *sk); 1538 void tcp_fastopen_ctx_destroy(struct net *net); 1539 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1540 void *key, unsigned int len); 1541 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1542 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1543 struct request_sock *req, 1544 struct tcp_fastopen_cookie *foc, 1545 const struct dst_entry *dst); 1546 void tcp_fastopen_init_key_once(struct net *net); 1547 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1548 struct tcp_fastopen_cookie *cookie); 1549 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1550 #define TCP_FASTOPEN_KEY_LENGTH 16 1551 1552 /* Fastopen key context */ 1553 struct tcp_fastopen_context { 1554 struct crypto_cipher *tfm; 1555 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1556 struct rcu_head rcu; 1557 }; 1558 1559 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1560 void tcp_fastopen_active_disable(struct sock *sk); 1561 bool tcp_fastopen_active_should_disable(struct sock *sk); 1562 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1563 void tcp_fastopen_active_timeout_reset(void); 1564 1565 /* Latencies incurred by various limits for a sender. They are 1566 * chronograph-like stats that are mutually exclusive. 1567 */ 1568 enum tcp_chrono { 1569 TCP_CHRONO_UNSPEC, 1570 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1571 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1572 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1573 __TCP_CHRONO_MAX, 1574 }; 1575 1576 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1577 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1578 1579 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1580 * the same memory storage than skb->destructor/_skb_refdst 1581 */ 1582 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1583 { 1584 skb->destructor = NULL; 1585 skb->_skb_refdst = 0UL; 1586 } 1587 1588 #define tcp_skb_tsorted_save(skb) { \ 1589 unsigned long _save = skb->_skb_refdst; \ 1590 skb->_skb_refdst = 0UL; 1591 1592 #define tcp_skb_tsorted_restore(skb) \ 1593 skb->_skb_refdst = _save; \ 1594 } 1595 1596 void tcp_write_queue_purge(struct sock *sk); 1597 1598 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1599 { 1600 return skb_rb_first(&sk->tcp_rtx_queue); 1601 } 1602 1603 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1604 { 1605 return skb_peek(&sk->sk_write_queue); 1606 } 1607 1608 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1609 { 1610 return skb_peek_tail(&sk->sk_write_queue); 1611 } 1612 1613 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1614 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1615 1616 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1617 { 1618 return skb_peek(&sk->sk_write_queue); 1619 } 1620 1621 static inline bool tcp_skb_is_last(const struct sock *sk, 1622 const struct sk_buff *skb) 1623 { 1624 return skb_queue_is_last(&sk->sk_write_queue, skb); 1625 } 1626 1627 static inline bool tcp_write_queue_empty(const struct sock *sk) 1628 { 1629 return skb_queue_empty(&sk->sk_write_queue); 1630 } 1631 1632 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1633 { 1634 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1635 } 1636 1637 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1638 { 1639 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1640 } 1641 1642 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1643 { 1644 if (tcp_write_queue_empty(sk)) 1645 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1646 1647 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1648 tcp_sk(sk)->highest_sack = NULL; 1649 } 1650 1651 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1652 { 1653 __skb_queue_tail(&sk->sk_write_queue, skb); 1654 } 1655 1656 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1657 { 1658 __tcp_add_write_queue_tail(sk, skb); 1659 1660 /* Queue it, remembering where we must start sending. */ 1661 if (sk->sk_write_queue.next == skb) { 1662 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1663 1664 if (tcp_sk(sk)->highest_sack == NULL) 1665 tcp_sk(sk)->highest_sack = skb; 1666 } 1667 } 1668 1669 /* Insert new before skb on the write queue of sk. */ 1670 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1671 struct sk_buff *skb, 1672 struct sock *sk) 1673 { 1674 __skb_queue_before(&sk->sk_write_queue, skb, new); 1675 } 1676 1677 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1678 { 1679 tcp_skb_tsorted_anchor_cleanup(skb); 1680 __skb_unlink(skb, &sk->sk_write_queue); 1681 } 1682 1683 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1684 1685 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1686 { 1687 tcp_skb_tsorted_anchor_cleanup(skb); 1688 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1689 } 1690 1691 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1692 { 1693 list_del(&skb->tcp_tsorted_anchor); 1694 tcp_rtx_queue_unlink(skb, sk); 1695 sk_wmem_free_skb(sk, skb); 1696 } 1697 1698 static inline void tcp_push_pending_frames(struct sock *sk) 1699 { 1700 if (tcp_send_head(sk)) { 1701 struct tcp_sock *tp = tcp_sk(sk); 1702 1703 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1704 } 1705 } 1706 1707 /* Start sequence of the skb just after the highest skb with SACKed 1708 * bit, valid only if sacked_out > 0 or when the caller has ensured 1709 * validity by itself. 1710 */ 1711 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1712 { 1713 if (!tp->sacked_out) 1714 return tp->snd_una; 1715 1716 if (tp->highest_sack == NULL) 1717 return tp->snd_nxt; 1718 1719 return TCP_SKB_CB(tp->highest_sack)->seq; 1720 } 1721 1722 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1723 { 1724 struct sk_buff *next = skb_rb_next(skb); 1725 1726 tcp_sk(sk)->highest_sack = next ?: tcp_send_head(sk); 1727 } 1728 1729 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1730 { 1731 return tcp_sk(sk)->highest_sack; 1732 } 1733 1734 static inline void tcp_highest_sack_reset(struct sock *sk) 1735 { 1736 struct sk_buff *skb = tcp_rtx_queue_head(sk); 1737 1738 tcp_sk(sk)->highest_sack = skb ?: tcp_send_head(sk); 1739 } 1740 1741 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1742 static inline void tcp_highest_sack_combine(struct sock *sk, 1743 struct sk_buff *old, 1744 struct sk_buff *new) 1745 { 1746 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1747 tcp_sk(sk)->highest_sack = new; 1748 } 1749 1750 /* This helper checks if socket has IP_TRANSPARENT set */ 1751 static inline bool inet_sk_transparent(const struct sock *sk) 1752 { 1753 switch (sk->sk_state) { 1754 case TCP_TIME_WAIT: 1755 return inet_twsk(sk)->tw_transparent; 1756 case TCP_NEW_SYN_RECV: 1757 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1758 } 1759 return inet_sk(sk)->transparent; 1760 } 1761 1762 /* Determines whether this is a thin stream (which may suffer from 1763 * increased latency). Used to trigger latency-reducing mechanisms. 1764 */ 1765 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1766 { 1767 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1768 } 1769 1770 /* /proc */ 1771 enum tcp_seq_states { 1772 TCP_SEQ_STATE_LISTENING, 1773 TCP_SEQ_STATE_ESTABLISHED, 1774 }; 1775 1776 int tcp_seq_open(struct inode *inode, struct file *file); 1777 1778 struct tcp_seq_afinfo { 1779 char *name; 1780 sa_family_t family; 1781 const struct file_operations *seq_fops; 1782 struct seq_operations seq_ops; 1783 }; 1784 1785 struct tcp_iter_state { 1786 struct seq_net_private p; 1787 sa_family_t family; 1788 enum tcp_seq_states state; 1789 struct sock *syn_wait_sk; 1790 int bucket, offset, sbucket, num; 1791 loff_t last_pos; 1792 }; 1793 1794 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1795 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1796 1797 extern struct request_sock_ops tcp_request_sock_ops; 1798 extern struct request_sock_ops tcp6_request_sock_ops; 1799 1800 void tcp_v4_destroy_sock(struct sock *sk); 1801 1802 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1803 netdev_features_t features); 1804 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1805 int tcp_gro_complete(struct sk_buff *skb); 1806 1807 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1808 1809 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1810 { 1811 struct net *net = sock_net((struct sock *)tp); 1812 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1813 } 1814 1815 static inline bool tcp_stream_memory_free(const struct sock *sk) 1816 { 1817 const struct tcp_sock *tp = tcp_sk(sk); 1818 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1819 1820 return notsent_bytes < tcp_notsent_lowat(tp); 1821 } 1822 1823 #ifdef CONFIG_PROC_FS 1824 int tcp4_proc_init(void); 1825 void tcp4_proc_exit(void); 1826 #endif 1827 1828 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1829 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1830 const struct tcp_request_sock_ops *af_ops, 1831 struct sock *sk, struct sk_buff *skb); 1832 1833 /* TCP af-specific functions */ 1834 struct tcp_sock_af_ops { 1835 #ifdef CONFIG_TCP_MD5SIG 1836 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1837 const struct sock *addr_sk); 1838 int (*calc_md5_hash)(char *location, 1839 const struct tcp_md5sig_key *md5, 1840 const struct sock *sk, 1841 const struct sk_buff *skb); 1842 int (*md5_parse)(struct sock *sk, 1843 int optname, 1844 char __user *optval, 1845 int optlen); 1846 #endif 1847 }; 1848 1849 struct tcp_request_sock_ops { 1850 u16 mss_clamp; 1851 #ifdef CONFIG_TCP_MD5SIG 1852 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1853 const struct sock *addr_sk); 1854 int (*calc_md5_hash) (char *location, 1855 const struct tcp_md5sig_key *md5, 1856 const struct sock *sk, 1857 const struct sk_buff *skb); 1858 #endif 1859 void (*init_req)(struct request_sock *req, 1860 const struct sock *sk_listener, 1861 struct sk_buff *skb); 1862 #ifdef CONFIG_SYN_COOKIES 1863 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1864 __u16 *mss); 1865 #endif 1866 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1867 const struct request_sock *req); 1868 u32 (*init_seq)(const struct sk_buff *skb); 1869 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1870 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1871 struct flowi *fl, struct request_sock *req, 1872 struct tcp_fastopen_cookie *foc, 1873 enum tcp_synack_type synack_type); 1874 }; 1875 1876 #ifdef CONFIG_SYN_COOKIES 1877 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1878 const struct sock *sk, struct sk_buff *skb, 1879 __u16 *mss) 1880 { 1881 tcp_synq_overflow(sk); 1882 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1883 return ops->cookie_init_seq(skb, mss); 1884 } 1885 #else 1886 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1887 const struct sock *sk, struct sk_buff *skb, 1888 __u16 *mss) 1889 { 1890 return 0; 1891 } 1892 #endif 1893 1894 int tcpv4_offload_init(void); 1895 1896 void tcp_v4_init(void); 1897 void tcp_init(void); 1898 1899 /* tcp_recovery.c */ 1900 extern void tcp_rack_mark_lost(struct sock *sk); 1901 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1902 u64 xmit_time); 1903 extern void tcp_rack_reo_timeout(struct sock *sk); 1904 1905 /* At how many usecs into the future should the RTO fire? */ 1906 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1907 { 1908 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1909 u32 rto = inet_csk(sk)->icsk_rto; 1910 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1911 1912 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1913 } 1914 1915 /* 1916 * Save and compile IPv4 options, return a pointer to it 1917 */ 1918 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1919 struct sk_buff *skb) 1920 { 1921 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1922 struct ip_options_rcu *dopt = NULL; 1923 1924 if (opt->optlen) { 1925 int opt_size = sizeof(*dopt) + opt->optlen; 1926 1927 dopt = kmalloc(opt_size, GFP_ATOMIC); 1928 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1929 kfree(dopt); 1930 dopt = NULL; 1931 } 1932 } 1933 return dopt; 1934 } 1935 1936 /* locally generated TCP pure ACKs have skb->truesize == 2 1937 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1938 * This is much faster than dissecting the packet to find out. 1939 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1940 */ 1941 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1942 { 1943 return skb->truesize == 2; 1944 } 1945 1946 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1947 { 1948 skb->truesize = 2; 1949 } 1950 1951 static inline int tcp_inq(struct sock *sk) 1952 { 1953 struct tcp_sock *tp = tcp_sk(sk); 1954 int answ; 1955 1956 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1957 answ = 0; 1958 } else if (sock_flag(sk, SOCK_URGINLINE) || 1959 !tp->urg_data || 1960 before(tp->urg_seq, tp->copied_seq) || 1961 !before(tp->urg_seq, tp->rcv_nxt)) { 1962 1963 answ = tp->rcv_nxt - tp->copied_seq; 1964 1965 /* Subtract 1, if FIN was received */ 1966 if (answ && sock_flag(sk, SOCK_DONE)) 1967 answ--; 1968 } else { 1969 answ = tp->urg_seq - tp->copied_seq; 1970 } 1971 1972 return answ; 1973 } 1974 1975 int tcp_peek_len(struct socket *sock); 1976 1977 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1978 { 1979 u16 segs_in; 1980 1981 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1982 tp->segs_in += segs_in; 1983 if (skb->len > tcp_hdrlen(skb)) 1984 tp->data_segs_in += segs_in; 1985 } 1986 1987 /* 1988 * TCP listen path runs lockless. 1989 * We forced "struct sock" to be const qualified to make sure 1990 * we don't modify one of its field by mistake. 1991 * Here, we increment sk_drops which is an atomic_t, so we can safely 1992 * make sock writable again. 1993 */ 1994 static inline void tcp_listendrop(const struct sock *sk) 1995 { 1996 atomic_inc(&((struct sock *)sk)->sk_drops); 1997 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1998 } 1999 2000 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2001 2002 /* 2003 * Interface for adding Upper Level Protocols over TCP 2004 */ 2005 2006 #define TCP_ULP_NAME_MAX 16 2007 #define TCP_ULP_MAX 128 2008 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2009 2010 struct tcp_ulp_ops { 2011 struct list_head list; 2012 2013 /* initialize ulp */ 2014 int (*init)(struct sock *sk); 2015 /* cleanup ulp */ 2016 void (*release)(struct sock *sk); 2017 2018 char name[TCP_ULP_NAME_MAX]; 2019 struct module *owner; 2020 }; 2021 int tcp_register_ulp(struct tcp_ulp_ops *type); 2022 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2023 int tcp_set_ulp(struct sock *sk, const char *name); 2024 void tcp_get_available_ulp(char *buf, size_t len); 2025 void tcp_cleanup_ulp(struct sock *sk); 2026 2027 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2028 * is < 0, then the BPF op failed (for example if the loaded BPF 2029 * program does not support the chosen operation or there is no BPF 2030 * program loaded). 2031 */ 2032 #ifdef CONFIG_BPF 2033 static inline int tcp_call_bpf(struct sock *sk, int op) 2034 { 2035 struct bpf_sock_ops_kern sock_ops; 2036 int ret; 2037 2038 if (sk_fullsock(sk)) 2039 sock_owned_by_me(sk); 2040 2041 memset(&sock_ops, 0, sizeof(sock_ops)); 2042 sock_ops.sk = sk; 2043 sock_ops.op = op; 2044 2045 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2046 if (ret == 0) 2047 ret = sock_ops.reply; 2048 else 2049 ret = -1; 2050 return ret; 2051 } 2052 #else 2053 static inline int tcp_call_bpf(struct sock *sk, int op) 2054 { 2055 return -EPERM; 2056 } 2057 #endif 2058 2059 static inline u32 tcp_timeout_init(struct sock *sk) 2060 { 2061 int timeout; 2062 2063 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2064 2065 if (timeout <= 0) 2066 timeout = TCP_TIMEOUT_INIT; 2067 return timeout; 2068 } 2069 2070 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2071 { 2072 int rwnd; 2073 2074 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2075 2076 if (rwnd < 0) 2077 rwnd = 0; 2078 return rwnd; 2079 } 2080 2081 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2082 { 2083 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2084 } 2085 2086 #if IS_ENABLED(CONFIG_SMC) 2087 extern struct static_key_false tcp_have_smc; 2088 #endif 2089 #endif /* _TCP_H */ 2090