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]; 245 246 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 247 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 248 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 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 void tcp_rcv_space_adjust(struct sock *sk); 339 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 340 void tcp_twsk_destructor(struct sock *sk); 341 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 342 struct pipe_inode_info *pipe, size_t len, 343 unsigned int flags); 344 345 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 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 bool *lost_race); 380 int tcp_child_process(struct sock *parent, struct sock *child, 381 struct sk_buff *skb); 382 void tcp_enter_loss(struct sock *sk); 383 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 384 void tcp_clear_retrans(struct tcp_sock *tp); 385 void tcp_update_metrics(struct sock *sk); 386 void tcp_init_metrics(struct sock *sk); 387 void tcp_metrics_init(void); 388 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 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 __poll_t 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 int tcp_set_rcvlowat(struct sock *sk, int val); 407 void tcp_data_ready(struct sock *sk); 408 int tcp_mmap(struct file *file, struct socket *sock, 409 struct vm_area_struct *vma); 410 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 411 struct tcp_options_received *opt_rx, 412 int estab, struct tcp_fastopen_cookie *foc); 413 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 414 415 /* 416 * TCP v4 functions exported for the inet6 API 417 */ 418 419 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 420 void tcp_v4_mtu_reduced(struct sock *sk); 421 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 422 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 423 struct sock *tcp_create_openreq_child(const struct sock *sk, 424 struct request_sock *req, 425 struct sk_buff *skb); 426 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 427 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 428 struct request_sock *req, 429 struct dst_entry *dst, 430 struct request_sock *req_unhash, 431 bool *own_req); 432 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 433 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 434 int tcp_connect(struct sock *sk); 435 enum tcp_synack_type { 436 TCP_SYNACK_NORMAL, 437 TCP_SYNACK_FASTOPEN, 438 TCP_SYNACK_COOKIE, 439 }; 440 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 441 struct request_sock *req, 442 struct tcp_fastopen_cookie *foc, 443 enum tcp_synack_type synack_type); 444 int tcp_disconnect(struct sock *sk, int flags); 445 446 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 447 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 448 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 449 450 /* From syncookies.c */ 451 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 452 struct request_sock *req, 453 struct dst_entry *dst, u32 tsoff); 454 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 455 u32 cookie); 456 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 457 #ifdef CONFIG_SYN_COOKIES 458 459 /* Syncookies use a monotonic timer which increments every 60 seconds. 460 * This counter is used both as a hash input and partially encoded into 461 * the cookie value. A cookie is only validated further if the delta 462 * between the current counter value and the encoded one is less than this, 463 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 464 * the counter advances immediately after a cookie is generated). 465 */ 466 #define MAX_SYNCOOKIE_AGE 2 467 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 468 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 469 470 /* syncookies: remember time of last synqueue overflow 471 * But do not dirty this field too often (once per second is enough) 472 * It is racy as we do not hold a lock, but race is very minor. 473 */ 474 static inline void tcp_synq_overflow(const struct sock *sk) 475 { 476 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 477 unsigned long now = jiffies; 478 479 if (time_after(now, last_overflow + HZ)) 480 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 481 } 482 483 /* syncookies: no recent synqueue overflow on this listening socket? */ 484 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 485 { 486 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 487 488 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 489 } 490 491 static inline u32 tcp_cookie_time(void) 492 { 493 u64 val = get_jiffies_64(); 494 495 do_div(val, TCP_SYNCOOKIE_PERIOD); 496 return val; 497 } 498 499 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 500 u16 *mssp); 501 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 502 u64 cookie_init_timestamp(struct request_sock *req); 503 bool cookie_timestamp_decode(const struct net *net, 504 struct tcp_options_received *opt); 505 bool cookie_ecn_ok(const struct tcp_options_received *opt, 506 const struct net *net, const struct dst_entry *dst); 507 508 /* From net/ipv6/syncookies.c */ 509 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 510 u32 cookie); 511 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 512 513 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 514 const struct tcphdr *th, u16 *mssp); 515 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 516 #endif 517 /* tcp_output.c */ 518 519 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 520 int nonagle); 521 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 522 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 523 void tcp_retransmit_timer(struct sock *sk); 524 void tcp_xmit_retransmit_queue(struct sock *); 525 void tcp_simple_retransmit(struct sock *); 526 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 527 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 528 enum tcp_queue { 529 TCP_FRAG_IN_WRITE_QUEUE, 530 TCP_FRAG_IN_RTX_QUEUE, 531 }; 532 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 533 struct sk_buff *skb, u32 len, 534 unsigned int mss_now, gfp_t gfp); 535 536 void tcp_send_probe0(struct sock *); 537 void tcp_send_partial(struct sock *); 538 int tcp_write_wakeup(struct sock *, int mib); 539 void tcp_send_fin(struct sock *sk); 540 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 541 int tcp_send_synack(struct sock *); 542 void tcp_push_one(struct sock *, unsigned int mss_now); 543 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 544 void tcp_send_ack(struct sock *sk); 545 void tcp_send_delayed_ack(struct sock *sk); 546 void tcp_send_loss_probe(struct sock *sk); 547 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 548 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 549 const struct sk_buff *next_skb); 550 551 /* tcp_input.c */ 552 void tcp_rearm_rto(struct sock *sk); 553 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 554 void tcp_reset(struct sock *sk); 555 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 556 void tcp_fin(struct sock *sk); 557 558 /* tcp_timer.c */ 559 void tcp_init_xmit_timers(struct sock *); 560 static inline void tcp_clear_xmit_timers(struct sock *sk) 561 { 562 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 563 __sock_put(sk); 564 565 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 566 __sock_put(sk); 567 568 inet_csk_clear_xmit_timers(sk); 569 } 570 571 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 572 unsigned int tcp_current_mss(struct sock *sk); 573 574 /* Bound MSS / TSO packet size with the half of the window */ 575 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 576 { 577 int cutoff; 578 579 /* When peer uses tiny windows, there is no use in packetizing 580 * to sub-MSS pieces for the sake of SWS or making sure there 581 * are enough packets in the pipe for fast recovery. 582 * 583 * On the other hand, for extremely large MSS devices, handling 584 * smaller than MSS windows in this way does make sense. 585 */ 586 if (tp->max_window > TCP_MSS_DEFAULT) 587 cutoff = (tp->max_window >> 1); 588 else 589 cutoff = tp->max_window; 590 591 if (cutoff && pktsize > cutoff) 592 return max_t(int, cutoff, 68U - tp->tcp_header_len); 593 else 594 return pktsize; 595 } 596 597 /* tcp.c */ 598 void tcp_get_info(struct sock *, struct tcp_info *); 599 600 /* Read 'sendfile()'-style from a TCP socket */ 601 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 602 sk_read_actor_t recv_actor); 603 604 void tcp_initialize_rcv_mss(struct sock *sk); 605 606 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 607 int tcp_mss_to_mtu(struct sock *sk, int mss); 608 void tcp_mtup_init(struct sock *sk); 609 void tcp_init_buffer_space(struct sock *sk); 610 611 static inline void tcp_bound_rto(const struct sock *sk) 612 { 613 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 614 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 615 } 616 617 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 618 { 619 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 620 } 621 622 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 623 { 624 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 625 ntohl(TCP_FLAG_ACK) | 626 snd_wnd); 627 } 628 629 static inline void tcp_fast_path_on(struct tcp_sock *tp) 630 { 631 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 632 } 633 634 static inline void tcp_fast_path_check(struct sock *sk) 635 { 636 struct tcp_sock *tp = tcp_sk(sk); 637 638 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 639 tp->rcv_wnd && 640 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 641 !tp->urg_data) 642 tcp_fast_path_on(tp); 643 } 644 645 /* Compute the actual rto_min value */ 646 static inline u32 tcp_rto_min(struct sock *sk) 647 { 648 const struct dst_entry *dst = __sk_dst_get(sk); 649 u32 rto_min = TCP_RTO_MIN; 650 651 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 652 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 653 return rto_min; 654 } 655 656 static inline u32 tcp_rto_min_us(struct sock *sk) 657 { 658 return jiffies_to_usecs(tcp_rto_min(sk)); 659 } 660 661 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 662 { 663 return dst_metric_locked(dst, RTAX_CC_ALGO); 664 } 665 666 /* Minimum RTT in usec. ~0 means not available. */ 667 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 668 { 669 return minmax_get(&tp->rtt_min); 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 uses 32bit jiffies to save some space. 694 * Note that this is different from tcp_time_stamp, which 695 * historically has been the same until linux-4.13. 696 */ 697 #define tcp_jiffies32 ((u32)jiffies) 698 699 /* 700 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 701 * It is no longer tied to jiffies, but to 1 ms clock. 702 * Note: double check if you want to use tcp_jiffies32 instead of this. 703 */ 704 #define TCP_TS_HZ 1000 705 706 static inline u64 tcp_clock_ns(void) 707 { 708 return local_clock(); 709 } 710 711 static inline u64 tcp_clock_us(void) 712 { 713 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 714 } 715 716 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 717 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 718 { 719 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 720 } 721 722 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 723 static inline u32 tcp_time_stamp_raw(void) 724 { 725 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 726 } 727 728 729 /* Refresh 1us clock of a TCP socket, 730 * ensuring monotically increasing values. 731 */ 732 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 733 { 734 u64 val = tcp_clock_us(); 735 736 if (val > tp->tcp_mstamp) 737 tp->tcp_mstamp = val; 738 } 739 740 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 741 { 742 return max_t(s64, t1 - t0, 0); 743 } 744 745 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 746 { 747 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 748 } 749 750 751 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 752 753 #define TCPHDR_FIN 0x01 754 #define TCPHDR_SYN 0x02 755 #define TCPHDR_RST 0x04 756 #define TCPHDR_PSH 0x08 757 #define TCPHDR_ACK 0x10 758 #define TCPHDR_URG 0x20 759 #define TCPHDR_ECE 0x40 760 #define TCPHDR_CWR 0x80 761 762 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 763 764 /* This is what the send packet queuing engine uses to pass 765 * TCP per-packet control information to the transmission code. 766 * We also store the host-order sequence numbers in here too. 767 * This is 44 bytes if IPV6 is enabled. 768 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 769 */ 770 struct tcp_skb_cb { 771 __u32 seq; /* Starting sequence number */ 772 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 773 union { 774 /* Note : tcp_tw_isn is used in input path only 775 * (isn chosen by tcp_timewait_state_process()) 776 * 777 * tcp_gso_segs/size are used in write queue only, 778 * cf tcp_skb_pcount()/tcp_skb_mss() 779 */ 780 __u32 tcp_tw_isn; 781 struct { 782 u16 tcp_gso_segs; 783 u16 tcp_gso_size; 784 }; 785 }; 786 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 787 788 __u8 sacked; /* State flags for SACK. */ 789 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 790 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 791 #define TCPCB_LOST 0x04 /* SKB is lost */ 792 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 793 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 794 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 795 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 796 TCPCB_REPAIRED) 797 798 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 799 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 800 eor:1, /* Is skb MSG_EOR marked? */ 801 has_rxtstamp:1, /* SKB has a RX timestamp */ 802 unused:5; 803 __u32 ack_seq; /* Sequence number ACK'd */ 804 union { 805 struct { 806 /* There is space for up to 24 bytes */ 807 __u32 in_flight:30,/* Bytes in flight at transmit */ 808 is_app_limited:1, /* cwnd not fully used? */ 809 unused:1; 810 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 811 __u32 delivered; 812 /* start of send pipeline phase */ 813 u64 first_tx_mstamp; 814 /* when we reached the "delivered" count */ 815 u64 delivered_mstamp; 816 } tx; /* only used for outgoing skbs */ 817 union { 818 struct inet_skb_parm h4; 819 #if IS_ENABLED(CONFIG_IPV6) 820 struct inet6_skb_parm h6; 821 #endif 822 } header; /* For incoming skbs */ 823 struct { 824 __u32 flags; 825 struct sock *sk_redir; 826 void *data_end; 827 } bpf; 828 }; 829 }; 830 831 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 832 833 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 834 { 835 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 836 } 837 838 #if IS_ENABLED(CONFIG_IPV6) 839 /* This is the variant of inet6_iif() that must be used by TCP, 840 * as TCP moves IP6CB into a different location in skb->cb[] 841 */ 842 static inline int tcp_v6_iif(const struct sk_buff *skb) 843 { 844 return TCP_SKB_CB(skb)->header.h6.iif; 845 } 846 847 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 848 { 849 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 850 851 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 852 } 853 854 /* TCP_SKB_CB reference means this can not be used from early demux */ 855 static inline int tcp_v6_sdif(const struct sk_buff *skb) 856 { 857 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 858 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 859 return TCP_SKB_CB(skb)->header.h6.iif; 860 #endif 861 return 0; 862 } 863 #endif 864 865 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 866 { 867 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 868 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 869 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 870 return true; 871 #endif 872 return false; 873 } 874 875 /* TCP_SKB_CB reference means this can not be used from early demux */ 876 static inline int tcp_v4_sdif(struct sk_buff *skb) 877 { 878 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 879 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 880 return TCP_SKB_CB(skb)->header.h4.iif; 881 #endif 882 return 0; 883 } 884 885 /* Due to TSO, an SKB can be composed of multiple actual 886 * packets. To keep these tracked properly, we use this. 887 */ 888 static inline int tcp_skb_pcount(const struct sk_buff *skb) 889 { 890 return TCP_SKB_CB(skb)->tcp_gso_segs; 891 } 892 893 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 894 { 895 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 896 } 897 898 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 899 { 900 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 901 } 902 903 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 904 static inline int tcp_skb_mss(const struct sk_buff *skb) 905 { 906 return TCP_SKB_CB(skb)->tcp_gso_size; 907 } 908 909 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 910 { 911 return likely(!TCP_SKB_CB(skb)->eor); 912 } 913 914 /* Events passed to congestion control interface */ 915 enum tcp_ca_event { 916 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 917 CA_EVENT_CWND_RESTART, /* congestion window restart */ 918 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 919 CA_EVENT_LOSS, /* loss timeout */ 920 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 921 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 922 }; 923 924 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 925 enum tcp_ca_ack_event_flags { 926 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 927 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 928 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 929 }; 930 931 /* 932 * Interface for adding new TCP congestion control handlers 933 */ 934 #define TCP_CA_NAME_MAX 16 935 #define TCP_CA_MAX 128 936 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 937 938 #define TCP_CA_UNSPEC 0 939 940 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 941 #define TCP_CONG_NON_RESTRICTED 0x1 942 /* Requires ECN/ECT set on all packets */ 943 #define TCP_CONG_NEEDS_ECN 0x2 944 945 union tcp_cc_info; 946 947 struct ack_sample { 948 u32 pkts_acked; 949 s32 rtt_us; 950 u32 in_flight; 951 }; 952 953 /* A rate sample measures the number of (original/retransmitted) data 954 * packets delivered "delivered" over an interval of time "interval_us". 955 * The tcp_rate.c code fills in the rate sample, and congestion 956 * control modules that define a cong_control function to run at the end 957 * of ACK processing can optionally chose to consult this sample when 958 * setting cwnd and pacing rate. 959 * A sample is invalid if "delivered" or "interval_us" is negative. 960 */ 961 struct rate_sample { 962 u64 prior_mstamp; /* starting timestamp for interval */ 963 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 964 s32 delivered; /* number of packets delivered over interval */ 965 long interval_us; /* time for tp->delivered to incr "delivered" */ 966 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 967 int losses; /* number of packets marked lost upon ACK */ 968 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 969 u32 prior_in_flight; /* in flight before this ACK */ 970 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 971 bool is_retrans; /* is sample from retransmission? */ 972 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 973 }; 974 975 struct tcp_congestion_ops { 976 struct list_head list; 977 u32 key; 978 u32 flags; 979 980 /* initialize private data (optional) */ 981 void (*init)(struct sock *sk); 982 /* cleanup private data (optional) */ 983 void (*release)(struct sock *sk); 984 985 /* return slow start threshold (required) */ 986 u32 (*ssthresh)(struct sock *sk); 987 /* do new cwnd calculation (required) */ 988 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 989 /* call before changing ca_state (optional) */ 990 void (*set_state)(struct sock *sk, u8 new_state); 991 /* call when cwnd event occurs (optional) */ 992 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 993 /* call when ack arrives (optional) */ 994 void (*in_ack_event)(struct sock *sk, u32 flags); 995 /* new value of cwnd after loss (required) */ 996 u32 (*undo_cwnd)(struct sock *sk); 997 /* hook for packet ack accounting (optional) */ 998 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 999 /* override sysctl_tcp_min_tso_segs */ 1000 u32 (*min_tso_segs)(struct sock *sk); 1001 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1002 u32 (*sndbuf_expand)(struct sock *sk); 1003 /* call when packets are delivered to update cwnd and pacing rate, 1004 * after all the ca_state processing. (optional) 1005 */ 1006 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1007 /* get info for inet_diag (optional) */ 1008 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1009 union tcp_cc_info *info); 1010 1011 char name[TCP_CA_NAME_MAX]; 1012 struct module *owner; 1013 }; 1014 1015 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1016 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1017 1018 void tcp_assign_congestion_control(struct sock *sk); 1019 void tcp_init_congestion_control(struct sock *sk); 1020 void tcp_cleanup_congestion_control(struct sock *sk); 1021 int tcp_set_default_congestion_control(struct net *net, const char *name); 1022 void tcp_get_default_congestion_control(struct net *net, char *name); 1023 void tcp_get_available_congestion_control(char *buf, size_t len); 1024 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1025 int tcp_set_allowed_congestion_control(char *allowed); 1026 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1027 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1028 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1029 1030 u32 tcp_reno_ssthresh(struct sock *sk); 1031 u32 tcp_reno_undo_cwnd(struct sock *sk); 1032 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1033 extern struct tcp_congestion_ops tcp_reno; 1034 1035 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1036 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1037 #ifdef CONFIG_INET 1038 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1039 #else 1040 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1041 { 1042 return NULL; 1043 } 1044 #endif 1045 1046 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1047 { 1048 const struct inet_connection_sock *icsk = inet_csk(sk); 1049 1050 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1051 } 1052 1053 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1054 { 1055 struct inet_connection_sock *icsk = inet_csk(sk); 1056 1057 if (icsk->icsk_ca_ops->set_state) 1058 icsk->icsk_ca_ops->set_state(sk, ca_state); 1059 icsk->icsk_ca_state = ca_state; 1060 } 1061 1062 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1063 { 1064 const struct inet_connection_sock *icsk = inet_csk(sk); 1065 1066 if (icsk->icsk_ca_ops->cwnd_event) 1067 icsk->icsk_ca_ops->cwnd_event(sk, event); 1068 } 1069 1070 /* From tcp_rate.c */ 1071 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1072 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1073 struct rate_sample *rs); 1074 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1075 bool is_sack_reneg, struct rate_sample *rs); 1076 void tcp_rate_check_app_limited(struct sock *sk); 1077 1078 /* These functions determine how the current flow behaves in respect of SACK 1079 * handling. SACK is negotiated with the peer, and therefore it can vary 1080 * between different flows. 1081 * 1082 * tcp_is_sack - SACK enabled 1083 * tcp_is_reno - No SACK 1084 */ 1085 static inline int tcp_is_sack(const struct tcp_sock *tp) 1086 { 1087 return tp->rx_opt.sack_ok; 1088 } 1089 1090 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1091 { 1092 return !tcp_is_sack(tp); 1093 } 1094 1095 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1096 { 1097 return tp->sacked_out + tp->lost_out; 1098 } 1099 1100 /* This determines how many packets are "in the network" to the best 1101 * of our knowledge. In many cases it is conservative, but where 1102 * detailed information is available from the receiver (via SACK 1103 * blocks etc.) we can make more aggressive calculations. 1104 * 1105 * Use this for decisions involving congestion control, use just 1106 * tp->packets_out to determine if the send queue is empty or not. 1107 * 1108 * Read this equation as: 1109 * 1110 * "Packets sent once on transmission queue" MINUS 1111 * "Packets left network, but not honestly ACKed yet" PLUS 1112 * "Packets fast retransmitted" 1113 */ 1114 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1115 { 1116 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1117 } 1118 1119 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1120 1121 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1122 { 1123 return tp->snd_cwnd < tp->snd_ssthresh; 1124 } 1125 1126 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1127 { 1128 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1129 } 1130 1131 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1132 { 1133 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1134 (1 << inet_csk(sk)->icsk_ca_state); 1135 } 1136 1137 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1138 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1139 * ssthresh. 1140 */ 1141 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1142 { 1143 const struct tcp_sock *tp = tcp_sk(sk); 1144 1145 if (tcp_in_cwnd_reduction(sk)) 1146 return tp->snd_ssthresh; 1147 else 1148 return max(tp->snd_ssthresh, 1149 ((tp->snd_cwnd >> 1) + 1150 (tp->snd_cwnd >> 2))); 1151 } 1152 1153 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1154 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1155 1156 void tcp_enter_cwr(struct sock *sk); 1157 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1158 1159 /* The maximum number of MSS of available cwnd for which TSO defers 1160 * sending if not using sysctl_tcp_tso_win_divisor. 1161 */ 1162 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1163 { 1164 return 3; 1165 } 1166 1167 /* Returns end sequence number of the receiver's advertised window */ 1168 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1169 { 1170 return tp->snd_una + tp->snd_wnd; 1171 } 1172 1173 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1174 * flexible approach. The RFC suggests cwnd should not be raised unless 1175 * it was fully used previously. And that's exactly what we do in 1176 * congestion avoidance mode. But in slow start we allow cwnd to grow 1177 * as long as the application has used half the cwnd. 1178 * Example : 1179 * cwnd is 10 (IW10), but application sends 9 frames. 1180 * We allow cwnd to reach 18 when all frames are ACKed. 1181 * This check is safe because it's as aggressive as slow start which already 1182 * risks 100% overshoot. The advantage is that we discourage application to 1183 * either send more filler packets or data to artificially blow up the cwnd 1184 * usage, and allow application-limited process to probe bw more aggressively. 1185 */ 1186 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1187 { 1188 const struct tcp_sock *tp = tcp_sk(sk); 1189 1190 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1191 if (tcp_in_slow_start(tp)) 1192 return tp->snd_cwnd < 2 * tp->max_packets_out; 1193 1194 return tp->is_cwnd_limited; 1195 } 1196 1197 /* Something is really bad, we could not queue an additional packet, 1198 * because qdisc is full or receiver sent a 0 window. 1199 * We do not want to add fuel to the fire, or abort too early, 1200 * so make sure the timer we arm now is at least 200ms in the future, 1201 * regardless of current icsk_rto value (as it could be ~2ms) 1202 */ 1203 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1204 { 1205 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1206 } 1207 1208 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1209 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1210 unsigned long max_when) 1211 { 1212 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1213 1214 return (unsigned long)min_t(u64, when, max_when); 1215 } 1216 1217 static inline void tcp_check_probe_timer(struct sock *sk) 1218 { 1219 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1220 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1221 tcp_probe0_base(sk), TCP_RTO_MAX); 1222 } 1223 1224 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1225 { 1226 tp->snd_wl1 = seq; 1227 } 1228 1229 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1230 { 1231 tp->snd_wl1 = seq; 1232 } 1233 1234 /* 1235 * Calculate(/check) TCP checksum 1236 */ 1237 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1238 __be32 daddr, __wsum base) 1239 { 1240 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1241 } 1242 1243 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1244 { 1245 return __skb_checksum_complete(skb); 1246 } 1247 1248 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1249 { 1250 return !skb_csum_unnecessary(skb) && 1251 __tcp_checksum_complete(skb); 1252 } 1253 1254 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1255 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1256 1257 #undef STATE_TRACE 1258 1259 #ifdef STATE_TRACE 1260 static const char *statename[]={ 1261 "Unused","Established","Syn Sent","Syn Recv", 1262 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1263 "Close Wait","Last ACK","Listen","Closing" 1264 }; 1265 #endif 1266 void tcp_set_state(struct sock *sk, int state); 1267 1268 void tcp_done(struct sock *sk); 1269 1270 int tcp_abort(struct sock *sk, int err); 1271 1272 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1273 { 1274 rx_opt->dsack = 0; 1275 rx_opt->num_sacks = 0; 1276 } 1277 1278 u32 tcp_default_init_rwnd(u32 mss); 1279 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1280 1281 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1282 { 1283 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1284 struct tcp_sock *tp = tcp_sk(sk); 1285 s32 delta; 1286 1287 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1288 ca_ops->cong_control) 1289 return; 1290 delta = tcp_jiffies32 - tp->lsndtime; 1291 if (delta > inet_csk(sk)->icsk_rto) 1292 tcp_cwnd_restart(sk, delta); 1293 } 1294 1295 /* Determine a window scaling and initial window to offer. */ 1296 void tcp_select_initial_window(const struct sock *sk, int __space, 1297 __u32 mss, __u32 *rcv_wnd, 1298 __u32 *window_clamp, int wscale_ok, 1299 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1300 1301 static inline int tcp_win_from_space(const struct sock *sk, int space) 1302 { 1303 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1304 1305 return tcp_adv_win_scale <= 0 ? 1306 (space>>(-tcp_adv_win_scale)) : 1307 space - (space>>tcp_adv_win_scale); 1308 } 1309 1310 /* Note: caller must be prepared to deal with negative returns */ 1311 static inline int tcp_space(const struct sock *sk) 1312 { 1313 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1314 atomic_read(&sk->sk_rmem_alloc)); 1315 } 1316 1317 static inline int tcp_full_space(const struct sock *sk) 1318 { 1319 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1320 } 1321 1322 extern void tcp_openreq_init_rwin(struct request_sock *req, 1323 const struct sock *sk_listener, 1324 const struct dst_entry *dst); 1325 1326 void tcp_enter_memory_pressure(struct sock *sk); 1327 void tcp_leave_memory_pressure(struct sock *sk); 1328 1329 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1330 { 1331 struct net *net = sock_net((struct sock *)tp); 1332 1333 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1334 } 1335 1336 static inline int keepalive_time_when(const struct tcp_sock *tp) 1337 { 1338 struct net *net = sock_net((struct sock *)tp); 1339 1340 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1341 } 1342 1343 static inline int keepalive_probes(const struct tcp_sock *tp) 1344 { 1345 struct net *net = sock_net((struct sock *)tp); 1346 1347 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1348 } 1349 1350 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1351 { 1352 const struct inet_connection_sock *icsk = &tp->inet_conn; 1353 1354 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1355 tcp_jiffies32 - tp->rcv_tstamp); 1356 } 1357 1358 static inline int tcp_fin_time(const struct sock *sk) 1359 { 1360 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1361 const int rto = inet_csk(sk)->icsk_rto; 1362 1363 if (fin_timeout < (rto << 2) - (rto >> 1)) 1364 fin_timeout = (rto << 2) - (rto >> 1); 1365 1366 return fin_timeout; 1367 } 1368 1369 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1370 int paws_win) 1371 { 1372 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1373 return true; 1374 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1375 return true; 1376 /* 1377 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1378 * then following tcp messages have valid values. Ignore 0 value, 1379 * or else 'negative' tsval might forbid us to accept their packets. 1380 */ 1381 if (!rx_opt->ts_recent) 1382 return true; 1383 return false; 1384 } 1385 1386 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1387 int rst) 1388 { 1389 if (tcp_paws_check(rx_opt, 0)) 1390 return false; 1391 1392 /* RST segments are not recommended to carry timestamp, 1393 and, if they do, it is recommended to ignore PAWS because 1394 "their cleanup function should take precedence over timestamps." 1395 Certainly, it is mistake. It is necessary to understand the reasons 1396 of this constraint to relax it: if peer reboots, clock may go 1397 out-of-sync and half-open connections will not be reset. 1398 Actually, the problem would be not existing if all 1399 the implementations followed draft about maintaining clock 1400 via reboots. Linux-2.2 DOES NOT! 1401 1402 However, we can relax time bounds for RST segments to MSL. 1403 */ 1404 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1405 return false; 1406 return true; 1407 } 1408 1409 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1410 int mib_idx, u32 *last_oow_ack_time); 1411 1412 static inline void tcp_mib_init(struct net *net) 1413 { 1414 /* See RFC 2012 */ 1415 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1416 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1417 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1418 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1419 } 1420 1421 /* from STCP */ 1422 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1423 { 1424 tp->lost_skb_hint = NULL; 1425 } 1426 1427 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1428 { 1429 tcp_clear_retrans_hints_partial(tp); 1430 tp->retransmit_skb_hint = NULL; 1431 } 1432 1433 union tcp_md5_addr { 1434 struct in_addr a4; 1435 #if IS_ENABLED(CONFIG_IPV6) 1436 struct in6_addr a6; 1437 #endif 1438 }; 1439 1440 /* - key database */ 1441 struct tcp_md5sig_key { 1442 struct hlist_node node; 1443 u8 keylen; 1444 u8 family; /* AF_INET or AF_INET6 */ 1445 union tcp_md5_addr addr; 1446 u8 prefixlen; 1447 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1448 struct rcu_head rcu; 1449 }; 1450 1451 /* - sock block */ 1452 struct tcp_md5sig_info { 1453 struct hlist_head head; 1454 struct rcu_head rcu; 1455 }; 1456 1457 /* - pseudo header */ 1458 struct tcp4_pseudohdr { 1459 __be32 saddr; 1460 __be32 daddr; 1461 __u8 pad; 1462 __u8 protocol; 1463 __be16 len; 1464 }; 1465 1466 struct tcp6_pseudohdr { 1467 struct in6_addr saddr; 1468 struct in6_addr daddr; 1469 __be32 len; 1470 __be32 protocol; /* including padding */ 1471 }; 1472 1473 union tcp_md5sum_block { 1474 struct tcp4_pseudohdr ip4; 1475 #if IS_ENABLED(CONFIG_IPV6) 1476 struct tcp6_pseudohdr ip6; 1477 #endif 1478 }; 1479 1480 /* - pool: digest algorithm, hash description and scratch buffer */ 1481 struct tcp_md5sig_pool { 1482 struct ahash_request *md5_req; 1483 void *scratch; 1484 }; 1485 1486 /* - functions */ 1487 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1488 const struct sock *sk, const struct sk_buff *skb); 1489 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1490 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1491 gfp_t gfp); 1492 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1493 int family, u8 prefixlen); 1494 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1495 const struct sock *addr_sk); 1496 1497 #ifdef CONFIG_TCP_MD5SIG 1498 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1499 const union tcp_md5_addr *addr, 1500 int family); 1501 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1502 #else 1503 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1504 const union tcp_md5_addr *addr, 1505 int family) 1506 { 1507 return NULL; 1508 } 1509 #define tcp_twsk_md5_key(twsk) NULL 1510 #endif 1511 1512 bool tcp_alloc_md5sig_pool(void); 1513 1514 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1515 static inline void tcp_put_md5sig_pool(void) 1516 { 1517 local_bh_enable(); 1518 } 1519 1520 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1521 unsigned int header_len); 1522 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1523 const struct tcp_md5sig_key *key); 1524 1525 /* From tcp_fastopen.c */ 1526 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1527 struct tcp_fastopen_cookie *cookie); 1528 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1529 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1530 u16 try_exp); 1531 struct tcp_fastopen_request { 1532 /* Fast Open cookie. Size 0 means a cookie request */ 1533 struct tcp_fastopen_cookie cookie; 1534 struct msghdr *data; /* data in MSG_FASTOPEN */ 1535 size_t size; 1536 int copied; /* queued in tcp_connect() */ 1537 }; 1538 void tcp_free_fastopen_req(struct tcp_sock *tp); 1539 void tcp_fastopen_destroy_cipher(struct sock *sk); 1540 void tcp_fastopen_ctx_destroy(struct net *net); 1541 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1542 void *key, unsigned int len); 1543 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1544 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1545 struct request_sock *req, 1546 struct tcp_fastopen_cookie *foc, 1547 const struct dst_entry *dst); 1548 void tcp_fastopen_init_key_once(struct net *net); 1549 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1550 struct tcp_fastopen_cookie *cookie); 1551 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1552 #define TCP_FASTOPEN_KEY_LENGTH 16 1553 1554 /* Fastopen key context */ 1555 struct tcp_fastopen_context { 1556 struct crypto_cipher *tfm; 1557 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1558 struct rcu_head rcu; 1559 }; 1560 1561 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1562 void tcp_fastopen_active_disable(struct sock *sk); 1563 bool tcp_fastopen_active_should_disable(struct sock *sk); 1564 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1565 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1566 1567 /* Latencies incurred by various limits for a sender. They are 1568 * chronograph-like stats that are mutually exclusive. 1569 */ 1570 enum tcp_chrono { 1571 TCP_CHRONO_UNSPEC, 1572 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1573 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1574 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1575 __TCP_CHRONO_MAX, 1576 }; 1577 1578 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1579 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1580 1581 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1582 * the same memory storage than skb->destructor/_skb_refdst 1583 */ 1584 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1585 { 1586 skb->destructor = NULL; 1587 skb->_skb_refdst = 0UL; 1588 } 1589 1590 #define tcp_skb_tsorted_save(skb) { \ 1591 unsigned long _save = skb->_skb_refdst; \ 1592 skb->_skb_refdst = 0UL; 1593 1594 #define tcp_skb_tsorted_restore(skb) \ 1595 skb->_skb_refdst = _save; \ 1596 } 1597 1598 void tcp_write_queue_purge(struct sock *sk); 1599 1600 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1601 { 1602 return skb_rb_first(&sk->tcp_rtx_queue); 1603 } 1604 1605 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1606 { 1607 return skb_peek(&sk->sk_write_queue); 1608 } 1609 1610 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1611 { 1612 return skb_peek_tail(&sk->sk_write_queue); 1613 } 1614 1615 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1616 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1617 1618 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1619 { 1620 return skb_peek(&sk->sk_write_queue); 1621 } 1622 1623 static inline bool tcp_skb_is_last(const struct sock *sk, 1624 const struct sk_buff *skb) 1625 { 1626 return skb_queue_is_last(&sk->sk_write_queue, skb); 1627 } 1628 1629 static inline bool tcp_write_queue_empty(const struct sock *sk) 1630 { 1631 return skb_queue_empty(&sk->sk_write_queue); 1632 } 1633 1634 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1635 { 1636 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1637 } 1638 1639 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1640 { 1641 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1642 } 1643 1644 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1645 { 1646 if (tcp_write_queue_empty(sk)) 1647 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1648 } 1649 1650 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1651 { 1652 __skb_queue_tail(&sk->sk_write_queue, skb); 1653 } 1654 1655 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1656 { 1657 __tcp_add_write_queue_tail(sk, skb); 1658 1659 /* Queue it, remembering where we must start sending. */ 1660 if (sk->sk_write_queue.next == skb) 1661 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1662 } 1663 1664 /* Insert new before skb on the write queue of sk. */ 1665 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1666 struct sk_buff *skb, 1667 struct sock *sk) 1668 { 1669 __skb_queue_before(&sk->sk_write_queue, skb, new); 1670 } 1671 1672 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1673 { 1674 tcp_skb_tsorted_anchor_cleanup(skb); 1675 __skb_unlink(skb, &sk->sk_write_queue); 1676 } 1677 1678 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1679 1680 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1681 { 1682 tcp_skb_tsorted_anchor_cleanup(skb); 1683 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1684 } 1685 1686 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1687 { 1688 list_del(&skb->tcp_tsorted_anchor); 1689 tcp_rtx_queue_unlink(skb, sk); 1690 sk_wmem_free_skb(sk, skb); 1691 } 1692 1693 static inline void tcp_push_pending_frames(struct sock *sk) 1694 { 1695 if (tcp_send_head(sk)) { 1696 struct tcp_sock *tp = tcp_sk(sk); 1697 1698 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1699 } 1700 } 1701 1702 /* Start sequence of the skb just after the highest skb with SACKed 1703 * bit, valid only if sacked_out > 0 or when the caller has ensured 1704 * validity by itself. 1705 */ 1706 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1707 { 1708 if (!tp->sacked_out) 1709 return tp->snd_una; 1710 1711 if (tp->highest_sack == NULL) 1712 return tp->snd_nxt; 1713 1714 return TCP_SKB_CB(tp->highest_sack)->seq; 1715 } 1716 1717 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1718 { 1719 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1720 } 1721 1722 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1723 { 1724 return tcp_sk(sk)->highest_sack; 1725 } 1726 1727 static inline void tcp_highest_sack_reset(struct sock *sk) 1728 { 1729 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1730 } 1731 1732 /* Called when old skb is about to be deleted and replaced by new skb */ 1733 static inline void tcp_highest_sack_replace(struct sock *sk, 1734 struct sk_buff *old, 1735 struct sk_buff *new) 1736 { 1737 if (old == tcp_highest_sack(sk)) 1738 tcp_sk(sk)->highest_sack = new; 1739 } 1740 1741 /* This helper checks if socket has IP_TRANSPARENT set */ 1742 static inline bool inet_sk_transparent(const struct sock *sk) 1743 { 1744 switch (sk->sk_state) { 1745 case TCP_TIME_WAIT: 1746 return inet_twsk(sk)->tw_transparent; 1747 case TCP_NEW_SYN_RECV: 1748 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1749 } 1750 return inet_sk(sk)->transparent; 1751 } 1752 1753 /* Determines whether this is a thin stream (which may suffer from 1754 * increased latency). Used to trigger latency-reducing mechanisms. 1755 */ 1756 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1757 { 1758 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1759 } 1760 1761 /* /proc */ 1762 enum tcp_seq_states { 1763 TCP_SEQ_STATE_LISTENING, 1764 TCP_SEQ_STATE_ESTABLISHED, 1765 }; 1766 1767 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1768 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1769 void tcp_seq_stop(struct seq_file *seq, void *v); 1770 1771 struct tcp_seq_afinfo { 1772 sa_family_t family; 1773 }; 1774 1775 struct tcp_iter_state { 1776 struct seq_net_private p; 1777 enum tcp_seq_states state; 1778 struct sock *syn_wait_sk; 1779 int bucket, offset, sbucket, num; 1780 loff_t last_pos; 1781 }; 1782 1783 extern struct request_sock_ops tcp_request_sock_ops; 1784 extern struct request_sock_ops tcp6_request_sock_ops; 1785 1786 void tcp_v4_destroy_sock(struct sock *sk); 1787 1788 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1789 netdev_features_t features); 1790 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1791 int tcp_gro_complete(struct sk_buff *skb); 1792 1793 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1794 1795 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1796 { 1797 struct net *net = sock_net((struct sock *)tp); 1798 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1799 } 1800 1801 static inline bool tcp_stream_memory_free(const struct sock *sk) 1802 { 1803 const struct tcp_sock *tp = tcp_sk(sk); 1804 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1805 1806 return notsent_bytes < tcp_notsent_lowat(tp); 1807 } 1808 1809 #ifdef CONFIG_PROC_FS 1810 int tcp4_proc_init(void); 1811 void tcp4_proc_exit(void); 1812 #endif 1813 1814 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1815 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1816 const struct tcp_request_sock_ops *af_ops, 1817 struct sock *sk, struct sk_buff *skb); 1818 1819 /* TCP af-specific functions */ 1820 struct tcp_sock_af_ops { 1821 #ifdef CONFIG_TCP_MD5SIG 1822 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1823 const struct sock *addr_sk); 1824 int (*calc_md5_hash)(char *location, 1825 const struct tcp_md5sig_key *md5, 1826 const struct sock *sk, 1827 const struct sk_buff *skb); 1828 int (*md5_parse)(struct sock *sk, 1829 int optname, 1830 char __user *optval, 1831 int optlen); 1832 #endif 1833 }; 1834 1835 struct tcp_request_sock_ops { 1836 u16 mss_clamp; 1837 #ifdef CONFIG_TCP_MD5SIG 1838 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1839 const struct sock *addr_sk); 1840 int (*calc_md5_hash) (char *location, 1841 const struct tcp_md5sig_key *md5, 1842 const struct sock *sk, 1843 const struct sk_buff *skb); 1844 #endif 1845 void (*init_req)(struct request_sock *req, 1846 const struct sock *sk_listener, 1847 struct sk_buff *skb); 1848 #ifdef CONFIG_SYN_COOKIES 1849 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1850 __u16 *mss); 1851 #endif 1852 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1853 const struct request_sock *req); 1854 u32 (*init_seq)(const struct sk_buff *skb); 1855 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1856 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1857 struct flowi *fl, struct request_sock *req, 1858 struct tcp_fastopen_cookie *foc, 1859 enum tcp_synack_type synack_type); 1860 }; 1861 1862 #ifdef CONFIG_SYN_COOKIES 1863 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1864 const struct sock *sk, struct sk_buff *skb, 1865 __u16 *mss) 1866 { 1867 tcp_synq_overflow(sk); 1868 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1869 return ops->cookie_init_seq(skb, mss); 1870 } 1871 #else 1872 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1873 const struct sock *sk, struct sk_buff *skb, 1874 __u16 *mss) 1875 { 1876 return 0; 1877 } 1878 #endif 1879 1880 int tcpv4_offload_init(void); 1881 1882 void tcp_v4_init(void); 1883 void tcp_init(void); 1884 1885 /* tcp_recovery.c */ 1886 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1887 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1888 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1889 u32 reo_wnd); 1890 extern void tcp_rack_mark_lost(struct sock *sk); 1891 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1892 u64 xmit_time); 1893 extern void tcp_rack_reo_timeout(struct sock *sk); 1894 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1895 1896 /* At how many usecs into the future should the RTO fire? */ 1897 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1898 { 1899 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1900 u32 rto = inet_csk(sk)->icsk_rto; 1901 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1902 1903 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1904 } 1905 1906 /* 1907 * Save and compile IPv4 options, return a pointer to it 1908 */ 1909 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1910 struct sk_buff *skb) 1911 { 1912 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1913 struct ip_options_rcu *dopt = NULL; 1914 1915 if (opt->optlen) { 1916 int opt_size = sizeof(*dopt) + opt->optlen; 1917 1918 dopt = kmalloc(opt_size, GFP_ATOMIC); 1919 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1920 kfree(dopt); 1921 dopt = NULL; 1922 } 1923 } 1924 return dopt; 1925 } 1926 1927 /* locally generated TCP pure ACKs have skb->truesize == 2 1928 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1929 * This is much faster than dissecting the packet to find out. 1930 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1931 */ 1932 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1933 { 1934 return skb->truesize == 2; 1935 } 1936 1937 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1938 { 1939 skb->truesize = 2; 1940 } 1941 1942 static inline int tcp_inq(struct sock *sk) 1943 { 1944 struct tcp_sock *tp = tcp_sk(sk); 1945 int answ; 1946 1947 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1948 answ = 0; 1949 } else if (sock_flag(sk, SOCK_URGINLINE) || 1950 !tp->urg_data || 1951 before(tp->urg_seq, tp->copied_seq) || 1952 !before(tp->urg_seq, tp->rcv_nxt)) { 1953 1954 answ = tp->rcv_nxt - tp->copied_seq; 1955 1956 /* Subtract 1, if FIN was received */ 1957 if (answ && sock_flag(sk, SOCK_DONE)) 1958 answ--; 1959 } else { 1960 answ = tp->urg_seq - tp->copied_seq; 1961 } 1962 1963 return answ; 1964 } 1965 1966 int tcp_peek_len(struct socket *sock); 1967 1968 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1969 { 1970 u16 segs_in; 1971 1972 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1973 tp->segs_in += segs_in; 1974 if (skb->len > tcp_hdrlen(skb)) 1975 tp->data_segs_in += segs_in; 1976 } 1977 1978 /* 1979 * TCP listen path runs lockless. 1980 * We forced "struct sock" to be const qualified to make sure 1981 * we don't modify one of its field by mistake. 1982 * Here, we increment sk_drops which is an atomic_t, so we can safely 1983 * make sock writable again. 1984 */ 1985 static inline void tcp_listendrop(const struct sock *sk) 1986 { 1987 atomic_inc(&((struct sock *)sk)->sk_drops); 1988 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1989 } 1990 1991 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 1992 1993 /* 1994 * Interface for adding Upper Level Protocols over TCP 1995 */ 1996 1997 #define TCP_ULP_NAME_MAX 16 1998 #define TCP_ULP_MAX 128 1999 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2000 2001 enum { 2002 TCP_ULP_TLS, 2003 TCP_ULP_BPF, 2004 }; 2005 2006 struct tcp_ulp_ops { 2007 struct list_head list; 2008 2009 /* initialize ulp */ 2010 int (*init)(struct sock *sk); 2011 /* cleanup ulp */ 2012 void (*release)(struct sock *sk); 2013 2014 int uid; 2015 char name[TCP_ULP_NAME_MAX]; 2016 bool user_visible; 2017 struct module *owner; 2018 }; 2019 int tcp_register_ulp(struct tcp_ulp_ops *type); 2020 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2021 int tcp_set_ulp(struct sock *sk, const char *name); 2022 int tcp_set_ulp_id(struct sock *sk, const int ulp); 2023 void tcp_get_available_ulp(char *buf, size_t len); 2024 void tcp_cleanup_ulp(struct sock *sk); 2025 2026 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2027 * is < 0, then the BPF op failed (for example if the loaded BPF 2028 * program does not support the chosen operation or there is no BPF 2029 * program loaded). 2030 */ 2031 #ifdef CONFIG_BPF 2032 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2033 { 2034 struct bpf_sock_ops_kern sock_ops; 2035 int ret; 2036 2037 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2038 if (sk_fullsock(sk)) { 2039 sock_ops.is_fullsock = 1; 2040 sock_owned_by_me(sk); 2041 } 2042 2043 sock_ops.sk = sk; 2044 sock_ops.op = op; 2045 if (nargs > 0) 2046 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2047 2048 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2049 if (ret == 0) 2050 ret = sock_ops.reply; 2051 else 2052 ret = -1; 2053 return ret; 2054 } 2055 2056 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2057 { 2058 u32 args[2] = {arg1, arg2}; 2059 2060 return tcp_call_bpf(sk, op, 2, args); 2061 } 2062 2063 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2064 u32 arg3) 2065 { 2066 u32 args[3] = {arg1, arg2, arg3}; 2067 2068 return tcp_call_bpf(sk, op, 3, args); 2069 } 2070 2071 #else 2072 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2073 { 2074 return -EPERM; 2075 } 2076 2077 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2078 { 2079 return -EPERM; 2080 } 2081 2082 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2083 u32 arg3) 2084 { 2085 return -EPERM; 2086 } 2087 2088 #endif 2089 2090 static inline u32 tcp_timeout_init(struct sock *sk) 2091 { 2092 int timeout; 2093 2094 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2095 2096 if (timeout <= 0) 2097 timeout = TCP_TIMEOUT_INIT; 2098 return timeout; 2099 } 2100 2101 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2102 { 2103 int rwnd; 2104 2105 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2106 2107 if (rwnd < 0) 2108 rwnd = 0; 2109 return rwnd; 2110 } 2111 2112 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2113 { 2114 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2115 } 2116 2117 #if IS_ENABLED(CONFIG_SMC) 2118 extern struct static_key_false tcp_have_smc; 2119 #endif 2120 2121 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2122 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2123 void (*cad)(struct sock *sk, u32 ack_seq)); 2124 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2125 2126 #endif 2127 2128 #endif /* _TCP_H */ 2129