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 static inline void tcp_dec_quickack_mode(struct sock *sk, 346 const unsigned int pkts) 347 { 348 struct inet_connection_sock *icsk = inet_csk(sk); 349 350 if (icsk->icsk_ack.quick) { 351 if (pkts >= icsk->icsk_ack.quick) { 352 icsk->icsk_ack.quick = 0; 353 /* Leaving quickack mode we deflate ATO. */ 354 icsk->icsk_ack.ato = TCP_ATO_MIN; 355 } else 356 icsk->icsk_ack.quick -= pkts; 357 } 358 } 359 360 #define TCP_ECN_OK 1 361 #define TCP_ECN_QUEUE_CWR 2 362 #define TCP_ECN_DEMAND_CWR 4 363 #define TCP_ECN_SEEN 8 364 365 enum tcp_tw_status { 366 TCP_TW_SUCCESS = 0, 367 TCP_TW_RST = 1, 368 TCP_TW_ACK = 2, 369 TCP_TW_SYN = 3 370 }; 371 372 373 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 374 struct sk_buff *skb, 375 const struct tcphdr *th); 376 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 377 struct request_sock *req, bool fastopen, 378 bool *lost_race); 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_close(struct sock *sk, long timeout); 389 void tcp_init_sock(struct sock *sk); 390 void tcp_init_transfer(struct sock *sk, int bpf_op); 391 __poll_t tcp_poll_mask(struct socket *sock, __poll_t events); 392 int tcp_getsockopt(struct sock *sk, int level, int optname, 393 char __user *optval, int __user *optlen); 394 int tcp_setsockopt(struct sock *sk, int level, int optname, 395 char __user *optval, unsigned int optlen); 396 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 397 char __user *optval, int __user *optlen); 398 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 399 char __user *optval, unsigned int optlen); 400 void tcp_set_keepalive(struct sock *sk, int val); 401 void tcp_syn_ack_timeout(const struct request_sock *req); 402 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 403 int flags, int *addr_len); 404 int tcp_set_rcvlowat(struct sock *sk, int val); 405 void tcp_data_ready(struct sock *sk); 406 int tcp_mmap(struct file *file, struct socket *sock, 407 struct vm_area_struct *vma); 408 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 409 struct tcp_options_received *opt_rx, 410 int estab, struct tcp_fastopen_cookie *foc); 411 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 412 413 /* 414 * TCP v4 functions exported for the inet6 API 415 */ 416 417 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 418 void tcp_v4_mtu_reduced(struct sock *sk); 419 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 420 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 421 struct sock *tcp_create_openreq_child(const struct sock *sk, 422 struct request_sock *req, 423 struct sk_buff *skb); 424 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 425 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 426 struct request_sock *req, 427 struct dst_entry *dst, 428 struct request_sock *req_unhash, 429 bool *own_req); 430 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 431 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 432 int tcp_connect(struct sock *sk); 433 enum tcp_synack_type { 434 TCP_SYNACK_NORMAL, 435 TCP_SYNACK_FASTOPEN, 436 TCP_SYNACK_COOKIE, 437 }; 438 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 439 struct request_sock *req, 440 struct tcp_fastopen_cookie *foc, 441 enum tcp_synack_type synack_type); 442 int tcp_disconnect(struct sock *sk, int flags); 443 444 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 445 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 446 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 447 448 /* From syncookies.c */ 449 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 450 struct request_sock *req, 451 struct dst_entry *dst, u32 tsoff); 452 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 453 u32 cookie); 454 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 455 #ifdef CONFIG_SYN_COOKIES 456 457 /* Syncookies use a monotonic timer which increments every 60 seconds. 458 * This counter is used both as a hash input and partially encoded into 459 * the cookie value. A cookie is only validated further if the delta 460 * between the current counter value and the encoded one is less than this, 461 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 462 * the counter advances immediately after a cookie is generated). 463 */ 464 #define MAX_SYNCOOKIE_AGE 2 465 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 466 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 467 468 /* syncookies: remember time of last synqueue overflow 469 * But do not dirty this field too often (once per second is enough) 470 * It is racy as we do not hold a lock, but race is very minor. 471 */ 472 static inline void tcp_synq_overflow(const struct sock *sk) 473 { 474 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 475 unsigned long now = jiffies; 476 477 if (time_after(now, last_overflow + HZ)) 478 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 479 } 480 481 /* syncookies: no recent synqueue overflow on this listening socket? */ 482 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 483 { 484 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 485 486 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 487 } 488 489 static inline u32 tcp_cookie_time(void) 490 { 491 u64 val = get_jiffies_64(); 492 493 do_div(val, TCP_SYNCOOKIE_PERIOD); 494 return val; 495 } 496 497 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 498 u16 *mssp); 499 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 500 u64 cookie_init_timestamp(struct request_sock *req); 501 bool cookie_timestamp_decode(const struct net *net, 502 struct tcp_options_received *opt); 503 bool cookie_ecn_ok(const struct tcp_options_received *opt, 504 const struct net *net, const struct dst_entry *dst); 505 506 /* From net/ipv6/syncookies.c */ 507 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 508 u32 cookie); 509 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 510 511 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 512 const struct tcphdr *th, u16 *mssp); 513 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 514 #endif 515 /* tcp_output.c */ 516 517 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, bool advancing_rto); 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 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 560 __sock_put(sk); 561 562 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 563 __sock_put(sk); 564 565 inet_csk_clear_xmit_timers(sk); 566 } 567 568 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 569 unsigned int tcp_current_mss(struct sock *sk); 570 571 /* Bound MSS / TSO packet size with the half of the window */ 572 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 573 { 574 int cutoff; 575 576 /* When peer uses tiny windows, there is no use in packetizing 577 * to sub-MSS pieces for the sake of SWS or making sure there 578 * are enough packets in the pipe for fast recovery. 579 * 580 * On the other hand, for extremely large MSS devices, handling 581 * smaller than MSS windows in this way does make sense. 582 */ 583 if (tp->max_window > TCP_MSS_DEFAULT) 584 cutoff = (tp->max_window >> 1); 585 else 586 cutoff = tp->max_window; 587 588 if (cutoff && pktsize > cutoff) 589 return max_t(int, cutoff, 68U - tp->tcp_header_len); 590 else 591 return pktsize; 592 } 593 594 /* tcp.c */ 595 void tcp_get_info(struct sock *, struct tcp_info *); 596 597 /* Read 'sendfile()'-style from a TCP socket */ 598 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 599 sk_read_actor_t recv_actor); 600 601 void tcp_initialize_rcv_mss(struct sock *sk); 602 603 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 604 int tcp_mss_to_mtu(struct sock *sk, int mss); 605 void tcp_mtup_init(struct sock *sk); 606 void tcp_init_buffer_space(struct sock *sk); 607 608 static inline void tcp_bound_rto(const struct sock *sk) 609 { 610 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 611 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 612 } 613 614 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 615 { 616 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 617 } 618 619 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 620 { 621 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 622 ntohl(TCP_FLAG_ACK) | 623 snd_wnd); 624 } 625 626 static inline void tcp_fast_path_on(struct tcp_sock *tp) 627 { 628 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 629 } 630 631 static inline void tcp_fast_path_check(struct sock *sk) 632 { 633 struct tcp_sock *tp = tcp_sk(sk); 634 635 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 636 tp->rcv_wnd && 637 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 638 !tp->urg_data) 639 tcp_fast_path_on(tp); 640 } 641 642 /* Compute the actual rto_min value */ 643 static inline u32 tcp_rto_min(struct sock *sk) 644 { 645 const struct dst_entry *dst = __sk_dst_get(sk); 646 u32 rto_min = TCP_RTO_MIN; 647 648 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 649 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 650 return rto_min; 651 } 652 653 static inline u32 tcp_rto_min_us(struct sock *sk) 654 { 655 return jiffies_to_usecs(tcp_rto_min(sk)); 656 } 657 658 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 659 { 660 return dst_metric_locked(dst, RTAX_CC_ALGO); 661 } 662 663 /* Minimum RTT in usec. ~0 means not available. */ 664 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 665 { 666 return minmax_get(&tp->rtt_min); 667 } 668 669 /* Compute the actual receive window we are currently advertising. 670 * Rcv_nxt can be after the window if our peer push more data 671 * than the offered window. 672 */ 673 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 674 { 675 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 676 677 if (win < 0) 678 win = 0; 679 return (u32) win; 680 } 681 682 /* Choose a new window, without checks for shrinking, and without 683 * scaling applied to the result. The caller does these things 684 * if necessary. This is a "raw" window selection. 685 */ 686 u32 __tcp_select_window(struct sock *sk); 687 688 void tcp_send_window_probe(struct sock *sk); 689 690 /* TCP uses 32bit jiffies to save some space. 691 * Note that this is different from tcp_time_stamp, which 692 * historically has been the same until linux-4.13. 693 */ 694 #define tcp_jiffies32 ((u32)jiffies) 695 696 /* 697 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 698 * It is no longer tied to jiffies, but to 1 ms clock. 699 * Note: double check if you want to use tcp_jiffies32 instead of this. 700 */ 701 #define TCP_TS_HZ 1000 702 703 static inline u64 tcp_clock_ns(void) 704 { 705 return local_clock(); 706 } 707 708 static inline u64 tcp_clock_us(void) 709 { 710 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 711 } 712 713 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 714 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 715 { 716 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 717 } 718 719 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 720 static inline u32 tcp_time_stamp_raw(void) 721 { 722 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 723 } 724 725 726 /* Refresh 1us clock of a TCP socket, 727 * ensuring monotically increasing values. 728 */ 729 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 730 { 731 u64 val = tcp_clock_us(); 732 733 if (val > tp->tcp_mstamp) 734 tp->tcp_mstamp = val; 735 } 736 737 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 738 { 739 return max_t(s64, t1 - t0, 0); 740 } 741 742 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 743 { 744 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 745 } 746 747 748 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 749 750 #define TCPHDR_FIN 0x01 751 #define TCPHDR_SYN 0x02 752 #define TCPHDR_RST 0x04 753 #define TCPHDR_PSH 0x08 754 #define TCPHDR_ACK 0x10 755 #define TCPHDR_URG 0x20 756 #define TCPHDR_ECE 0x40 757 #define TCPHDR_CWR 0x80 758 759 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 760 761 /* This is what the send packet queuing engine uses to pass 762 * TCP per-packet control information to the transmission code. 763 * We also store the host-order sequence numbers in here too. 764 * This is 44 bytes if IPV6 is enabled. 765 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 766 */ 767 struct tcp_skb_cb { 768 __u32 seq; /* Starting sequence number */ 769 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 770 union { 771 /* Note : tcp_tw_isn is used in input path only 772 * (isn chosen by tcp_timewait_state_process()) 773 * 774 * tcp_gso_segs/size are used in write queue only, 775 * cf tcp_skb_pcount()/tcp_skb_mss() 776 */ 777 __u32 tcp_tw_isn; 778 struct { 779 u16 tcp_gso_segs; 780 u16 tcp_gso_size; 781 }; 782 }; 783 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 784 785 __u8 sacked; /* State flags for SACK. */ 786 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 787 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 788 #define TCPCB_LOST 0x04 /* SKB is lost */ 789 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 790 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 791 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 792 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 793 TCPCB_REPAIRED) 794 795 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 796 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 797 eor:1, /* Is skb MSG_EOR marked? */ 798 has_rxtstamp:1, /* SKB has a RX timestamp */ 799 unused:5; 800 __u32 ack_seq; /* Sequence number ACK'd */ 801 union { 802 struct { 803 /* There is space for up to 24 bytes */ 804 __u32 in_flight:30,/* Bytes in flight at transmit */ 805 is_app_limited:1, /* cwnd not fully used? */ 806 unused:1; 807 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 808 __u32 delivered; 809 /* start of send pipeline phase */ 810 u64 first_tx_mstamp; 811 /* when we reached the "delivered" count */ 812 u64 delivered_mstamp; 813 } tx; /* only used for outgoing skbs */ 814 union { 815 struct inet_skb_parm h4; 816 #if IS_ENABLED(CONFIG_IPV6) 817 struct inet6_skb_parm h6; 818 #endif 819 } header; /* For incoming skbs */ 820 struct { 821 __u32 flags; 822 struct sock *sk_redir; 823 void *data_end; 824 } bpf; 825 }; 826 }; 827 828 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 829 830 831 #if IS_ENABLED(CONFIG_IPV6) 832 /* This is the variant of inet6_iif() that must be used by TCP, 833 * as TCP moves IP6CB into a different location in skb->cb[] 834 */ 835 static inline int tcp_v6_iif(const struct sk_buff *skb) 836 { 837 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 838 839 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 840 } 841 842 /* TCP_SKB_CB reference means this can not be used from early demux */ 843 static inline int tcp_v6_sdif(const struct sk_buff *skb) 844 { 845 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 846 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 847 return TCP_SKB_CB(skb)->header.h6.iif; 848 #endif 849 return 0; 850 } 851 #endif 852 853 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 854 { 855 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 856 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 857 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 858 return true; 859 #endif 860 return false; 861 } 862 863 /* TCP_SKB_CB reference means this can not be used from early demux */ 864 static inline int tcp_v4_sdif(struct sk_buff *skb) 865 { 866 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 867 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 868 return TCP_SKB_CB(skb)->header.h4.iif; 869 #endif 870 return 0; 871 } 872 873 /* Due to TSO, an SKB can be composed of multiple actual 874 * packets. To keep these tracked properly, we use this. 875 */ 876 static inline int tcp_skb_pcount(const struct sk_buff *skb) 877 { 878 return TCP_SKB_CB(skb)->tcp_gso_segs; 879 } 880 881 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 882 { 883 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 884 } 885 886 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 887 { 888 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 889 } 890 891 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 892 static inline int tcp_skb_mss(const struct sk_buff *skb) 893 { 894 return TCP_SKB_CB(skb)->tcp_gso_size; 895 } 896 897 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 898 { 899 return likely(!TCP_SKB_CB(skb)->eor); 900 } 901 902 /* Events passed to congestion control interface */ 903 enum tcp_ca_event { 904 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 905 CA_EVENT_CWND_RESTART, /* congestion window restart */ 906 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 907 CA_EVENT_LOSS, /* loss timeout */ 908 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 909 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 910 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 911 CA_EVENT_NON_DELAYED_ACK, 912 }; 913 914 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 915 enum tcp_ca_ack_event_flags { 916 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 917 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 918 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 919 }; 920 921 /* 922 * Interface for adding new TCP congestion control handlers 923 */ 924 #define TCP_CA_NAME_MAX 16 925 #define TCP_CA_MAX 128 926 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 927 928 #define TCP_CA_UNSPEC 0 929 930 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 931 #define TCP_CONG_NON_RESTRICTED 0x1 932 /* Requires ECN/ECT set on all packets */ 933 #define TCP_CONG_NEEDS_ECN 0x2 934 935 union tcp_cc_info; 936 937 struct ack_sample { 938 u32 pkts_acked; 939 s32 rtt_us; 940 u32 in_flight; 941 }; 942 943 /* A rate sample measures the number of (original/retransmitted) data 944 * packets delivered "delivered" over an interval of time "interval_us". 945 * The tcp_rate.c code fills in the rate sample, and congestion 946 * control modules that define a cong_control function to run at the end 947 * of ACK processing can optionally chose to consult this sample when 948 * setting cwnd and pacing rate. 949 * A sample is invalid if "delivered" or "interval_us" is negative. 950 */ 951 struct rate_sample { 952 u64 prior_mstamp; /* starting timestamp for interval */ 953 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 954 s32 delivered; /* number of packets delivered over interval */ 955 long interval_us; /* time for tp->delivered to incr "delivered" */ 956 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 957 int losses; /* number of packets marked lost upon ACK */ 958 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 959 u32 prior_in_flight; /* in flight before this ACK */ 960 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 961 bool is_retrans; /* is sample from retransmission? */ 962 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 963 }; 964 965 struct tcp_congestion_ops { 966 struct list_head list; 967 u32 key; 968 u32 flags; 969 970 /* initialize private data (optional) */ 971 void (*init)(struct sock *sk); 972 /* cleanup private data (optional) */ 973 void (*release)(struct sock *sk); 974 975 /* return slow start threshold (required) */ 976 u32 (*ssthresh)(struct sock *sk); 977 /* do new cwnd calculation (required) */ 978 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 979 /* call before changing ca_state (optional) */ 980 void (*set_state)(struct sock *sk, u8 new_state); 981 /* call when cwnd event occurs (optional) */ 982 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 983 /* call when ack arrives (optional) */ 984 void (*in_ack_event)(struct sock *sk, u32 flags); 985 /* new value of cwnd after loss (required) */ 986 u32 (*undo_cwnd)(struct sock *sk); 987 /* hook for packet ack accounting (optional) */ 988 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 989 /* override sysctl_tcp_min_tso_segs */ 990 u32 (*min_tso_segs)(struct sock *sk); 991 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 992 u32 (*sndbuf_expand)(struct sock *sk); 993 /* call when packets are delivered to update cwnd and pacing rate, 994 * after all the ca_state processing. (optional) 995 */ 996 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 997 /* get info for inet_diag (optional) */ 998 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 999 union tcp_cc_info *info); 1000 1001 char name[TCP_CA_NAME_MAX]; 1002 struct module *owner; 1003 }; 1004 1005 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1006 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1007 1008 void tcp_assign_congestion_control(struct sock *sk); 1009 void tcp_init_congestion_control(struct sock *sk); 1010 void tcp_cleanup_congestion_control(struct sock *sk); 1011 int tcp_set_default_congestion_control(struct net *net, const char *name); 1012 void tcp_get_default_congestion_control(struct net *net, char *name); 1013 void tcp_get_available_congestion_control(char *buf, size_t len); 1014 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1015 int tcp_set_allowed_congestion_control(char *allowed); 1016 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1017 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1018 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1019 1020 u32 tcp_reno_ssthresh(struct sock *sk); 1021 u32 tcp_reno_undo_cwnd(struct sock *sk); 1022 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1023 extern struct tcp_congestion_ops tcp_reno; 1024 1025 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1026 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1027 #ifdef CONFIG_INET 1028 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1029 #else 1030 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1031 { 1032 return NULL; 1033 } 1034 #endif 1035 1036 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1037 { 1038 const struct inet_connection_sock *icsk = inet_csk(sk); 1039 1040 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1041 } 1042 1043 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1044 { 1045 struct inet_connection_sock *icsk = inet_csk(sk); 1046 1047 if (icsk->icsk_ca_ops->set_state) 1048 icsk->icsk_ca_ops->set_state(sk, ca_state); 1049 icsk->icsk_ca_state = ca_state; 1050 } 1051 1052 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1053 { 1054 const struct inet_connection_sock *icsk = inet_csk(sk); 1055 1056 if (icsk->icsk_ca_ops->cwnd_event) 1057 icsk->icsk_ca_ops->cwnd_event(sk, event); 1058 } 1059 1060 /* From tcp_rate.c */ 1061 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1062 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1063 struct rate_sample *rs); 1064 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1065 bool is_sack_reneg, struct rate_sample *rs); 1066 void tcp_rate_check_app_limited(struct sock *sk); 1067 1068 /* These functions determine how the current flow behaves in respect of SACK 1069 * handling. SACK is negotiated with the peer, and therefore it can vary 1070 * between different flows. 1071 * 1072 * tcp_is_sack - SACK enabled 1073 * tcp_is_reno - No SACK 1074 */ 1075 static inline int tcp_is_sack(const struct tcp_sock *tp) 1076 { 1077 return tp->rx_opt.sack_ok; 1078 } 1079 1080 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1081 { 1082 return !tcp_is_sack(tp); 1083 } 1084 1085 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1086 { 1087 return tp->sacked_out + tp->lost_out; 1088 } 1089 1090 /* This determines how many packets are "in the network" to the best 1091 * of our knowledge. In many cases it is conservative, but where 1092 * detailed information is available from the receiver (via SACK 1093 * blocks etc.) we can make more aggressive calculations. 1094 * 1095 * Use this for decisions involving congestion control, use just 1096 * tp->packets_out to determine if the send queue is empty or not. 1097 * 1098 * Read this equation as: 1099 * 1100 * "Packets sent once on transmission queue" MINUS 1101 * "Packets left network, but not honestly ACKed yet" PLUS 1102 * "Packets fast retransmitted" 1103 */ 1104 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1105 { 1106 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1107 } 1108 1109 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1110 1111 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1112 { 1113 return tp->snd_cwnd < tp->snd_ssthresh; 1114 } 1115 1116 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1117 { 1118 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1119 } 1120 1121 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1122 { 1123 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1124 (1 << inet_csk(sk)->icsk_ca_state); 1125 } 1126 1127 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1128 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1129 * ssthresh. 1130 */ 1131 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1132 { 1133 const struct tcp_sock *tp = tcp_sk(sk); 1134 1135 if (tcp_in_cwnd_reduction(sk)) 1136 return tp->snd_ssthresh; 1137 else 1138 return max(tp->snd_ssthresh, 1139 ((tp->snd_cwnd >> 1) + 1140 (tp->snd_cwnd >> 2))); 1141 } 1142 1143 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1144 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1145 1146 void tcp_enter_cwr(struct sock *sk); 1147 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1148 1149 /* The maximum number of MSS of available cwnd for which TSO defers 1150 * sending if not using sysctl_tcp_tso_win_divisor. 1151 */ 1152 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1153 { 1154 return 3; 1155 } 1156 1157 /* Returns end sequence number of the receiver's advertised window */ 1158 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1159 { 1160 return tp->snd_una + tp->snd_wnd; 1161 } 1162 1163 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1164 * flexible approach. The RFC suggests cwnd should not be raised unless 1165 * it was fully used previously. And that's exactly what we do in 1166 * congestion avoidance mode. But in slow start we allow cwnd to grow 1167 * as long as the application has used half the cwnd. 1168 * Example : 1169 * cwnd is 10 (IW10), but application sends 9 frames. 1170 * We allow cwnd to reach 18 when all frames are ACKed. 1171 * This check is safe because it's as aggressive as slow start which already 1172 * risks 100% overshoot. The advantage is that we discourage application to 1173 * either send more filler packets or data to artificially blow up the cwnd 1174 * usage, and allow application-limited process to probe bw more aggressively. 1175 */ 1176 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1177 { 1178 const struct tcp_sock *tp = tcp_sk(sk); 1179 1180 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1181 if (tcp_in_slow_start(tp)) 1182 return tp->snd_cwnd < 2 * tp->max_packets_out; 1183 1184 return tp->is_cwnd_limited; 1185 } 1186 1187 /* BBR congestion control needs pacing. 1188 * Same remark for SO_MAX_PACING_RATE. 1189 * sch_fq packet scheduler is efficiently handling pacing, 1190 * but is not always installed/used. 1191 * Return true if TCP stack should pace packets itself. 1192 */ 1193 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1194 { 1195 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1196 } 1197 1198 /* Something is really bad, we could not queue an additional packet, 1199 * because qdisc is full or receiver sent a 0 window. 1200 * We do not want to add fuel to the fire, or abort too early, 1201 * so make sure the timer we arm now is at least 200ms in the future, 1202 * regardless of current icsk_rto value (as it could be ~2ms) 1203 */ 1204 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1205 { 1206 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1207 } 1208 1209 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1210 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1211 unsigned long max_when) 1212 { 1213 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1214 1215 return (unsigned long)min_t(u64, when, max_when); 1216 } 1217 1218 static inline void tcp_check_probe_timer(struct sock *sk) 1219 { 1220 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1221 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1222 tcp_probe0_base(sk), TCP_RTO_MAX); 1223 } 1224 1225 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1226 { 1227 tp->snd_wl1 = seq; 1228 } 1229 1230 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1231 { 1232 tp->snd_wl1 = seq; 1233 } 1234 1235 /* 1236 * Calculate(/check) TCP checksum 1237 */ 1238 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1239 __be32 daddr, __wsum base) 1240 { 1241 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1242 } 1243 1244 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1245 { 1246 return __skb_checksum_complete(skb); 1247 } 1248 1249 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1250 { 1251 return !skb_csum_unnecessary(skb) && 1252 __tcp_checksum_complete(skb); 1253 } 1254 1255 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1256 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1257 1258 #undef STATE_TRACE 1259 1260 #ifdef STATE_TRACE 1261 static const char *statename[]={ 1262 "Unused","Established","Syn Sent","Syn Recv", 1263 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1264 "Close Wait","Last ACK","Listen","Closing" 1265 }; 1266 #endif 1267 void tcp_set_state(struct sock *sk, int state); 1268 1269 void tcp_done(struct sock *sk); 1270 1271 int tcp_abort(struct sock *sk, int err); 1272 1273 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1274 { 1275 rx_opt->dsack = 0; 1276 rx_opt->num_sacks = 0; 1277 } 1278 1279 u32 tcp_default_init_rwnd(u32 mss); 1280 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1281 1282 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1283 { 1284 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1285 struct tcp_sock *tp = tcp_sk(sk); 1286 s32 delta; 1287 1288 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1289 ca_ops->cong_control) 1290 return; 1291 delta = tcp_jiffies32 - tp->lsndtime; 1292 if (delta > inet_csk(sk)->icsk_rto) 1293 tcp_cwnd_restart(sk, delta); 1294 } 1295 1296 /* Determine a window scaling and initial window to offer. */ 1297 void tcp_select_initial_window(const struct sock *sk, int __space, 1298 __u32 mss, __u32 *rcv_wnd, 1299 __u32 *window_clamp, int wscale_ok, 1300 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1301 1302 static inline int tcp_win_from_space(const struct sock *sk, int space) 1303 { 1304 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1305 1306 return tcp_adv_win_scale <= 0 ? 1307 (space>>(-tcp_adv_win_scale)) : 1308 space - (space>>tcp_adv_win_scale); 1309 } 1310 1311 /* Note: caller must be prepared to deal with negative returns */ 1312 static inline int tcp_space(const struct sock *sk) 1313 { 1314 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1315 atomic_read(&sk->sk_rmem_alloc)); 1316 } 1317 1318 static inline int tcp_full_space(const struct sock *sk) 1319 { 1320 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1321 } 1322 1323 extern void tcp_openreq_init_rwin(struct request_sock *req, 1324 const struct sock *sk_listener, 1325 const struct dst_entry *dst); 1326 1327 void tcp_enter_memory_pressure(struct sock *sk); 1328 void tcp_leave_memory_pressure(struct sock *sk); 1329 1330 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1331 { 1332 struct net *net = sock_net((struct sock *)tp); 1333 1334 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1335 } 1336 1337 static inline int keepalive_time_when(const struct tcp_sock *tp) 1338 { 1339 struct net *net = sock_net((struct sock *)tp); 1340 1341 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1342 } 1343 1344 static inline int keepalive_probes(const struct tcp_sock *tp) 1345 { 1346 struct net *net = sock_net((struct sock *)tp); 1347 1348 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1349 } 1350 1351 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1352 { 1353 const struct inet_connection_sock *icsk = &tp->inet_conn; 1354 1355 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1356 tcp_jiffies32 - tp->rcv_tstamp); 1357 } 1358 1359 static inline int tcp_fin_time(const struct sock *sk) 1360 { 1361 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1362 const int rto = inet_csk(sk)->icsk_rto; 1363 1364 if (fin_timeout < (rto << 2) - (rto >> 1)) 1365 fin_timeout = (rto << 2) - (rto >> 1); 1366 1367 return fin_timeout; 1368 } 1369 1370 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1371 int paws_win) 1372 { 1373 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1374 return true; 1375 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1376 return true; 1377 /* 1378 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1379 * then following tcp messages have valid values. Ignore 0 value, 1380 * or else 'negative' tsval might forbid us to accept their packets. 1381 */ 1382 if (!rx_opt->ts_recent) 1383 return true; 1384 return false; 1385 } 1386 1387 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1388 int rst) 1389 { 1390 if (tcp_paws_check(rx_opt, 0)) 1391 return false; 1392 1393 /* RST segments are not recommended to carry timestamp, 1394 and, if they do, it is recommended to ignore PAWS because 1395 "their cleanup function should take precedence over timestamps." 1396 Certainly, it is mistake. It is necessary to understand the reasons 1397 of this constraint to relax it: if peer reboots, clock may go 1398 out-of-sync and half-open connections will not be reset. 1399 Actually, the problem would be not existing if all 1400 the implementations followed draft about maintaining clock 1401 via reboots. Linux-2.2 DOES NOT! 1402 1403 However, we can relax time bounds for RST segments to MSL. 1404 */ 1405 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1406 return false; 1407 return true; 1408 } 1409 1410 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1411 int mib_idx, u32 *last_oow_ack_time); 1412 1413 static inline void tcp_mib_init(struct net *net) 1414 { 1415 /* See RFC 2012 */ 1416 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1417 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1418 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1419 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1420 } 1421 1422 /* from STCP */ 1423 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1424 { 1425 tp->lost_skb_hint = NULL; 1426 } 1427 1428 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1429 { 1430 tcp_clear_retrans_hints_partial(tp); 1431 tp->retransmit_skb_hint = NULL; 1432 } 1433 1434 union tcp_md5_addr { 1435 struct in_addr a4; 1436 #if IS_ENABLED(CONFIG_IPV6) 1437 struct in6_addr a6; 1438 #endif 1439 }; 1440 1441 /* - key database */ 1442 struct tcp_md5sig_key { 1443 struct hlist_node node; 1444 u8 keylen; 1445 u8 family; /* AF_INET or AF_INET6 */ 1446 union tcp_md5_addr addr; 1447 u8 prefixlen; 1448 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1449 struct rcu_head rcu; 1450 }; 1451 1452 /* - sock block */ 1453 struct tcp_md5sig_info { 1454 struct hlist_head head; 1455 struct rcu_head rcu; 1456 }; 1457 1458 /* - pseudo header */ 1459 struct tcp4_pseudohdr { 1460 __be32 saddr; 1461 __be32 daddr; 1462 __u8 pad; 1463 __u8 protocol; 1464 __be16 len; 1465 }; 1466 1467 struct tcp6_pseudohdr { 1468 struct in6_addr saddr; 1469 struct in6_addr daddr; 1470 __be32 len; 1471 __be32 protocol; /* including padding */ 1472 }; 1473 1474 union tcp_md5sum_block { 1475 struct tcp4_pseudohdr ip4; 1476 #if IS_ENABLED(CONFIG_IPV6) 1477 struct tcp6_pseudohdr ip6; 1478 #endif 1479 }; 1480 1481 /* - pool: digest algorithm, hash description and scratch buffer */ 1482 struct tcp_md5sig_pool { 1483 struct ahash_request *md5_req; 1484 void *scratch; 1485 }; 1486 1487 /* - functions */ 1488 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1489 const struct sock *sk, const struct sk_buff *skb); 1490 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1491 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1492 gfp_t gfp); 1493 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1494 int family, u8 prefixlen); 1495 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1496 const struct sock *addr_sk); 1497 1498 #ifdef CONFIG_TCP_MD5SIG 1499 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1500 const union tcp_md5_addr *addr, 1501 int family); 1502 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1503 #else 1504 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1505 const union tcp_md5_addr *addr, 1506 int family) 1507 { 1508 return NULL; 1509 } 1510 #define tcp_twsk_md5_key(twsk) NULL 1511 #endif 1512 1513 bool tcp_alloc_md5sig_pool(void); 1514 1515 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1516 static inline void tcp_put_md5sig_pool(void) 1517 { 1518 local_bh_enable(); 1519 } 1520 1521 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1522 unsigned int header_len); 1523 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1524 const struct tcp_md5sig_key *key); 1525 1526 /* From tcp_fastopen.c */ 1527 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1528 struct tcp_fastopen_cookie *cookie); 1529 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1530 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1531 u16 try_exp); 1532 struct tcp_fastopen_request { 1533 /* Fast Open cookie. Size 0 means a cookie request */ 1534 struct tcp_fastopen_cookie cookie; 1535 struct msghdr *data; /* data in MSG_FASTOPEN */ 1536 size_t size; 1537 int copied; /* queued in tcp_connect() */ 1538 }; 1539 void tcp_free_fastopen_req(struct tcp_sock *tp); 1540 void tcp_fastopen_destroy_cipher(struct sock *sk); 1541 void tcp_fastopen_ctx_destroy(struct net *net); 1542 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1543 void *key, unsigned int len); 1544 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1545 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1546 struct request_sock *req, 1547 struct tcp_fastopen_cookie *foc, 1548 const struct dst_entry *dst); 1549 void tcp_fastopen_init_key_once(struct net *net); 1550 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1551 struct tcp_fastopen_cookie *cookie); 1552 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1553 #define TCP_FASTOPEN_KEY_LENGTH 16 1554 1555 /* Fastopen key context */ 1556 struct tcp_fastopen_context { 1557 struct crypto_cipher *tfm; 1558 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1559 struct rcu_head rcu; 1560 }; 1561 1562 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1563 void tcp_fastopen_active_disable(struct sock *sk); 1564 bool tcp_fastopen_active_should_disable(struct sock *sk); 1565 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1566 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1567 1568 /* Latencies incurred by various limits for a sender. They are 1569 * chronograph-like stats that are mutually exclusive. 1570 */ 1571 enum tcp_chrono { 1572 TCP_CHRONO_UNSPEC, 1573 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1574 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1575 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1576 __TCP_CHRONO_MAX, 1577 }; 1578 1579 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1580 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1581 1582 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1583 * the same memory storage than skb->destructor/_skb_refdst 1584 */ 1585 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1586 { 1587 skb->destructor = NULL; 1588 skb->_skb_refdst = 0UL; 1589 } 1590 1591 #define tcp_skb_tsorted_save(skb) { \ 1592 unsigned long _save = skb->_skb_refdst; \ 1593 skb->_skb_refdst = 0UL; 1594 1595 #define tcp_skb_tsorted_restore(skb) \ 1596 skb->_skb_refdst = _save; \ 1597 } 1598 1599 void tcp_write_queue_purge(struct sock *sk); 1600 1601 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1602 { 1603 return skb_rb_first(&sk->tcp_rtx_queue); 1604 } 1605 1606 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1607 { 1608 return skb_peek(&sk->sk_write_queue); 1609 } 1610 1611 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1612 { 1613 return skb_peek_tail(&sk->sk_write_queue); 1614 } 1615 1616 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1617 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1618 1619 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1620 { 1621 return skb_peek(&sk->sk_write_queue); 1622 } 1623 1624 static inline bool tcp_skb_is_last(const struct sock *sk, 1625 const struct sk_buff *skb) 1626 { 1627 return skb_queue_is_last(&sk->sk_write_queue, skb); 1628 } 1629 1630 static inline bool tcp_write_queue_empty(const struct sock *sk) 1631 { 1632 return skb_queue_empty(&sk->sk_write_queue); 1633 } 1634 1635 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1636 { 1637 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1638 } 1639 1640 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1641 { 1642 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1643 } 1644 1645 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1646 { 1647 if (tcp_write_queue_empty(sk)) 1648 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 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 1665 /* Insert new before skb on the write queue of sk. */ 1666 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1667 struct sk_buff *skb, 1668 struct sock *sk) 1669 { 1670 __skb_queue_before(&sk->sk_write_queue, skb, new); 1671 } 1672 1673 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1674 { 1675 tcp_skb_tsorted_anchor_cleanup(skb); 1676 __skb_unlink(skb, &sk->sk_write_queue); 1677 } 1678 1679 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1680 1681 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1682 { 1683 tcp_skb_tsorted_anchor_cleanup(skb); 1684 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1685 } 1686 1687 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1688 { 1689 list_del(&skb->tcp_tsorted_anchor); 1690 tcp_rtx_queue_unlink(skb, sk); 1691 sk_wmem_free_skb(sk, skb); 1692 } 1693 1694 static inline void tcp_push_pending_frames(struct sock *sk) 1695 { 1696 if (tcp_send_head(sk)) { 1697 struct tcp_sock *tp = tcp_sk(sk); 1698 1699 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1700 } 1701 } 1702 1703 /* Start sequence of the skb just after the highest skb with SACKed 1704 * bit, valid only if sacked_out > 0 or when the caller has ensured 1705 * validity by itself. 1706 */ 1707 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1708 { 1709 if (!tp->sacked_out) 1710 return tp->snd_una; 1711 1712 if (tp->highest_sack == NULL) 1713 return tp->snd_nxt; 1714 1715 return TCP_SKB_CB(tp->highest_sack)->seq; 1716 } 1717 1718 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1719 { 1720 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1721 } 1722 1723 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1724 { 1725 return tcp_sk(sk)->highest_sack; 1726 } 1727 1728 static inline void tcp_highest_sack_reset(struct sock *sk) 1729 { 1730 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1731 } 1732 1733 /* Called when old skb is about to be deleted and replaced by new skb */ 1734 static inline void tcp_highest_sack_replace(struct sock *sk, 1735 struct sk_buff *old, 1736 struct sk_buff *new) 1737 { 1738 if (old == tcp_highest_sack(sk)) 1739 tcp_sk(sk)->highest_sack = new; 1740 } 1741 1742 /* This helper checks if socket has IP_TRANSPARENT set */ 1743 static inline bool inet_sk_transparent(const struct sock *sk) 1744 { 1745 switch (sk->sk_state) { 1746 case TCP_TIME_WAIT: 1747 return inet_twsk(sk)->tw_transparent; 1748 case TCP_NEW_SYN_RECV: 1749 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1750 } 1751 return inet_sk(sk)->transparent; 1752 } 1753 1754 /* Determines whether this is a thin stream (which may suffer from 1755 * increased latency). Used to trigger latency-reducing mechanisms. 1756 */ 1757 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1758 { 1759 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1760 } 1761 1762 /* /proc */ 1763 enum tcp_seq_states { 1764 TCP_SEQ_STATE_LISTENING, 1765 TCP_SEQ_STATE_ESTABLISHED, 1766 }; 1767 1768 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1769 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1770 void tcp_seq_stop(struct seq_file *seq, void *v); 1771 1772 struct tcp_seq_afinfo { 1773 sa_family_t family; 1774 }; 1775 1776 struct tcp_iter_state { 1777 struct seq_net_private p; 1778 enum tcp_seq_states state; 1779 struct sock *syn_wait_sk; 1780 int bucket, offset, sbucket, num; 1781 loff_t last_pos; 1782 }; 1783 1784 extern struct request_sock_ops tcp_request_sock_ops; 1785 extern struct request_sock_ops tcp6_request_sock_ops; 1786 1787 void tcp_v4_destroy_sock(struct sock *sk); 1788 1789 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1790 netdev_features_t features); 1791 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1792 int tcp_gro_complete(struct sk_buff *skb); 1793 1794 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1795 1796 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1797 { 1798 struct net *net = sock_net((struct sock *)tp); 1799 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1800 } 1801 1802 static inline bool tcp_stream_memory_free(const struct sock *sk) 1803 { 1804 const struct tcp_sock *tp = tcp_sk(sk); 1805 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1806 1807 return notsent_bytes < tcp_notsent_lowat(tp); 1808 } 1809 1810 #ifdef CONFIG_PROC_FS 1811 int tcp4_proc_init(void); 1812 void tcp4_proc_exit(void); 1813 #endif 1814 1815 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1816 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1817 const struct tcp_request_sock_ops *af_ops, 1818 struct sock *sk, struct sk_buff *skb); 1819 1820 /* TCP af-specific functions */ 1821 struct tcp_sock_af_ops { 1822 #ifdef CONFIG_TCP_MD5SIG 1823 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1824 const struct sock *addr_sk); 1825 int (*calc_md5_hash)(char *location, 1826 const struct tcp_md5sig_key *md5, 1827 const struct sock *sk, 1828 const struct sk_buff *skb); 1829 int (*md5_parse)(struct sock *sk, 1830 int optname, 1831 char __user *optval, 1832 int optlen); 1833 #endif 1834 }; 1835 1836 struct tcp_request_sock_ops { 1837 u16 mss_clamp; 1838 #ifdef CONFIG_TCP_MD5SIG 1839 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1840 const struct sock *addr_sk); 1841 int (*calc_md5_hash) (char *location, 1842 const struct tcp_md5sig_key *md5, 1843 const struct sock *sk, 1844 const struct sk_buff *skb); 1845 #endif 1846 void (*init_req)(struct request_sock *req, 1847 const struct sock *sk_listener, 1848 struct sk_buff *skb); 1849 #ifdef CONFIG_SYN_COOKIES 1850 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1851 __u16 *mss); 1852 #endif 1853 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1854 const struct request_sock *req); 1855 u32 (*init_seq)(const struct sk_buff *skb); 1856 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1857 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1858 struct flowi *fl, struct request_sock *req, 1859 struct tcp_fastopen_cookie *foc, 1860 enum tcp_synack_type synack_type); 1861 }; 1862 1863 #ifdef CONFIG_SYN_COOKIES 1864 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1865 const struct sock *sk, struct sk_buff *skb, 1866 __u16 *mss) 1867 { 1868 tcp_synq_overflow(sk); 1869 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1870 return ops->cookie_init_seq(skb, mss); 1871 } 1872 #else 1873 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1874 const struct sock *sk, struct sk_buff *skb, 1875 __u16 *mss) 1876 { 1877 return 0; 1878 } 1879 #endif 1880 1881 int tcpv4_offload_init(void); 1882 1883 void tcp_v4_init(void); 1884 void tcp_init(void); 1885 1886 /* tcp_recovery.c */ 1887 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1888 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1889 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1890 u32 reo_wnd); 1891 extern void tcp_rack_mark_lost(struct sock *sk); 1892 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1893 u64 xmit_time); 1894 extern void tcp_rack_reo_timeout(struct sock *sk); 1895 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1896 1897 /* At how many usecs into the future should the RTO fire? */ 1898 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1899 { 1900 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1901 u32 rto = inet_csk(sk)->icsk_rto; 1902 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1903 1904 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1905 } 1906 1907 /* 1908 * Save and compile IPv4 options, return a pointer to it 1909 */ 1910 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1911 struct sk_buff *skb) 1912 { 1913 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1914 struct ip_options_rcu *dopt = NULL; 1915 1916 if (opt->optlen) { 1917 int opt_size = sizeof(*dopt) + opt->optlen; 1918 1919 dopt = kmalloc(opt_size, GFP_ATOMIC); 1920 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1921 kfree(dopt); 1922 dopt = NULL; 1923 } 1924 } 1925 return dopt; 1926 } 1927 1928 /* locally generated TCP pure ACKs have skb->truesize == 2 1929 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1930 * This is much faster than dissecting the packet to find out. 1931 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1932 */ 1933 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1934 { 1935 return skb->truesize == 2; 1936 } 1937 1938 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1939 { 1940 skb->truesize = 2; 1941 } 1942 1943 static inline int tcp_inq(struct sock *sk) 1944 { 1945 struct tcp_sock *tp = tcp_sk(sk); 1946 int answ; 1947 1948 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1949 answ = 0; 1950 } else if (sock_flag(sk, SOCK_URGINLINE) || 1951 !tp->urg_data || 1952 before(tp->urg_seq, tp->copied_seq) || 1953 !before(tp->urg_seq, tp->rcv_nxt)) { 1954 1955 answ = tp->rcv_nxt - tp->copied_seq; 1956 1957 /* Subtract 1, if FIN was received */ 1958 if (answ && sock_flag(sk, SOCK_DONE)) 1959 answ--; 1960 } else { 1961 answ = tp->urg_seq - tp->copied_seq; 1962 } 1963 1964 return answ; 1965 } 1966 1967 int tcp_peek_len(struct socket *sock); 1968 1969 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1970 { 1971 u16 segs_in; 1972 1973 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1974 tp->segs_in += segs_in; 1975 if (skb->len > tcp_hdrlen(skb)) 1976 tp->data_segs_in += segs_in; 1977 } 1978 1979 /* 1980 * TCP listen path runs lockless. 1981 * We forced "struct sock" to be const qualified to make sure 1982 * we don't modify one of its field by mistake. 1983 * Here, we increment sk_drops which is an atomic_t, so we can safely 1984 * make sock writable again. 1985 */ 1986 static inline void tcp_listendrop(const struct sock *sk) 1987 { 1988 atomic_inc(&((struct sock *)sk)->sk_drops); 1989 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1990 } 1991 1992 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 1993 1994 /* 1995 * Interface for adding Upper Level Protocols over TCP 1996 */ 1997 1998 #define TCP_ULP_NAME_MAX 16 1999 #define TCP_ULP_MAX 128 2000 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2001 2002 enum { 2003 TCP_ULP_TLS, 2004 TCP_ULP_BPF, 2005 }; 2006 2007 struct tcp_ulp_ops { 2008 struct list_head list; 2009 2010 /* initialize ulp */ 2011 int (*init)(struct sock *sk); 2012 /* cleanup ulp */ 2013 void (*release)(struct sock *sk); 2014 2015 int uid; 2016 char name[TCP_ULP_NAME_MAX]; 2017 bool user_visible; 2018 struct module *owner; 2019 }; 2020 int tcp_register_ulp(struct tcp_ulp_ops *type); 2021 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2022 int tcp_set_ulp(struct sock *sk, const char *name); 2023 int tcp_set_ulp_id(struct sock *sk, const int ulp); 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, u32 nargs, u32 *args) 2034 { 2035 struct bpf_sock_ops_kern sock_ops; 2036 int ret; 2037 2038 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2039 if (sk_fullsock(sk)) { 2040 sock_ops.is_fullsock = 1; 2041 sock_owned_by_me(sk); 2042 } 2043 2044 sock_ops.sk = sk; 2045 sock_ops.op = op; 2046 if (nargs > 0) 2047 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2048 2049 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2050 if (ret == 0) 2051 ret = sock_ops.reply; 2052 else 2053 ret = -1; 2054 return ret; 2055 } 2056 2057 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2058 { 2059 u32 args[2] = {arg1, arg2}; 2060 2061 return tcp_call_bpf(sk, op, 2, args); 2062 } 2063 2064 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2065 u32 arg3) 2066 { 2067 u32 args[3] = {arg1, arg2, arg3}; 2068 2069 return tcp_call_bpf(sk, op, 3, args); 2070 } 2071 2072 #else 2073 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2074 { 2075 return -EPERM; 2076 } 2077 2078 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2079 { 2080 return -EPERM; 2081 } 2082 2083 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2084 u32 arg3) 2085 { 2086 return -EPERM; 2087 } 2088 2089 #endif 2090 2091 static inline u32 tcp_timeout_init(struct sock *sk) 2092 { 2093 int timeout; 2094 2095 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2096 2097 if (timeout <= 0) 2098 timeout = TCP_TIMEOUT_INIT; 2099 return timeout; 2100 } 2101 2102 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2103 { 2104 int rwnd; 2105 2106 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2107 2108 if (rwnd < 0) 2109 rwnd = 0; 2110 return rwnd; 2111 } 2112 2113 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2114 { 2115 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2116 } 2117 2118 #if IS_ENABLED(CONFIG_SMC) 2119 extern struct static_key_false tcp_have_smc; 2120 #endif 2121 2122 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2123 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2124 void (*cad)(struct sock *sk, u32 ack_seq)); 2125 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2126 2127 #endif 2128 2129 #endif /* _TCP_H */ 2130