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(struct file *file, struct socket *sock, 392 struct poll_table_struct *wait); 393 int tcp_getsockopt(struct sock *sk, int level, int optname, 394 char __user *optval, int __user *optlen); 395 int tcp_setsockopt(struct sock *sk, int level, int optname, 396 char __user *optval, unsigned int optlen); 397 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, int __user *optlen); 399 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 400 char __user *optval, unsigned int optlen); 401 void tcp_set_keepalive(struct sock *sk, int val); 402 void tcp_syn_ack_timeout(const struct request_sock *req); 403 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 404 int flags, int *addr_len); 405 int tcp_set_rcvlowat(struct sock *sk, int val); 406 void tcp_data_ready(struct sock *sk); 407 int tcp_mmap(struct file *file, struct socket *sock, 408 struct vm_area_struct *vma); 409 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 410 struct tcp_options_received *opt_rx, 411 int estab, struct tcp_fastopen_cookie *foc); 412 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 413 414 /* 415 * TCP v4 functions exported for the inet6 API 416 */ 417 418 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 419 void tcp_v4_mtu_reduced(struct sock *sk); 420 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 421 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 422 struct sock *tcp_create_openreq_child(const struct sock *sk, 423 struct request_sock *req, 424 struct sk_buff *skb); 425 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 426 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 427 struct request_sock *req, 428 struct dst_entry *dst, 429 struct request_sock *req_unhash, 430 bool *own_req); 431 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 432 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 433 int tcp_connect(struct sock *sk); 434 enum tcp_synack_type { 435 TCP_SYNACK_NORMAL, 436 TCP_SYNACK_FASTOPEN, 437 TCP_SYNACK_COOKIE, 438 }; 439 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 440 struct request_sock *req, 441 struct tcp_fastopen_cookie *foc, 442 enum tcp_synack_type synack_type); 443 int tcp_disconnect(struct sock *sk, int flags); 444 445 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 446 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 447 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 448 449 /* From syncookies.c */ 450 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 451 struct request_sock *req, 452 struct dst_entry *dst, u32 tsoff); 453 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 454 u32 cookie); 455 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 456 #ifdef CONFIG_SYN_COOKIES 457 458 /* Syncookies use a monotonic timer which increments every 60 seconds. 459 * This counter is used both as a hash input and partially encoded into 460 * the cookie value. A cookie is only validated further if the delta 461 * between the current counter value and the encoded one is less than this, 462 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 463 * the counter advances immediately after a cookie is generated). 464 */ 465 #define MAX_SYNCOOKIE_AGE 2 466 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 467 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 468 469 /* syncookies: remember time of last synqueue overflow 470 * But do not dirty this field too often (once per second is enough) 471 * It is racy as we do not hold a lock, but race is very minor. 472 */ 473 static inline void tcp_synq_overflow(const struct sock *sk) 474 { 475 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 476 unsigned long now = jiffies; 477 478 if (time_after(now, last_overflow + HZ)) 479 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 480 } 481 482 /* syncookies: no recent synqueue overflow on this listening socket? */ 483 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 484 { 485 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 486 487 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 488 } 489 490 static inline u32 tcp_cookie_time(void) 491 { 492 u64 val = get_jiffies_64(); 493 494 do_div(val, TCP_SYNCOOKIE_PERIOD); 495 return val; 496 } 497 498 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 499 u16 *mssp); 500 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 501 u64 cookie_init_timestamp(struct request_sock *req); 502 bool cookie_timestamp_decode(const struct net *net, 503 struct tcp_options_received *opt); 504 bool cookie_ecn_ok(const struct tcp_options_received *opt, 505 const struct net *net, const struct dst_entry *dst); 506 507 /* From net/ipv6/syncookies.c */ 508 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 509 u32 cookie); 510 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 511 512 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 513 const struct tcphdr *th, u16 *mssp); 514 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 515 #endif 516 /* tcp_output.c */ 517 518 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 519 int nonagle); 520 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 521 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 522 void tcp_retransmit_timer(struct sock *sk); 523 void tcp_xmit_retransmit_queue(struct sock *); 524 void tcp_simple_retransmit(struct sock *); 525 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 526 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 527 enum tcp_queue { 528 TCP_FRAG_IN_WRITE_QUEUE, 529 TCP_FRAG_IN_RTX_QUEUE, 530 }; 531 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 532 struct sk_buff *skb, u32 len, 533 unsigned int mss_now, gfp_t gfp); 534 535 void tcp_send_probe0(struct sock *); 536 void tcp_send_partial(struct sock *); 537 int tcp_write_wakeup(struct sock *, int mib); 538 void tcp_send_fin(struct sock *sk); 539 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 540 int tcp_send_synack(struct sock *); 541 void tcp_push_one(struct sock *, unsigned int mss_now); 542 void tcp_send_ack(struct sock *sk); 543 void tcp_send_delayed_ack(struct sock *sk); 544 void tcp_send_loss_probe(struct sock *sk); 545 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 546 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 547 const struct sk_buff *next_skb); 548 549 /* tcp_input.c */ 550 void tcp_rearm_rto(struct sock *sk); 551 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 552 void tcp_reset(struct sock *sk); 553 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 554 void tcp_fin(struct sock *sk); 555 556 /* tcp_timer.c */ 557 void tcp_init_xmit_timers(struct sock *); 558 static inline void tcp_clear_xmit_timers(struct sock *sk) 559 { 560 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 561 __sock_put(sk); 562 563 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 564 __sock_put(sk); 565 566 inet_csk_clear_xmit_timers(sk); 567 } 568 569 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 570 unsigned int tcp_current_mss(struct sock *sk); 571 572 /* Bound MSS / TSO packet size with the half of the window */ 573 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 574 { 575 int cutoff; 576 577 /* When peer uses tiny windows, there is no use in packetizing 578 * to sub-MSS pieces for the sake of SWS or making sure there 579 * are enough packets in the pipe for fast recovery. 580 * 581 * On the other hand, for extremely large MSS devices, handling 582 * smaller than MSS windows in this way does make sense. 583 */ 584 if (tp->max_window > TCP_MSS_DEFAULT) 585 cutoff = (tp->max_window >> 1); 586 else 587 cutoff = tp->max_window; 588 589 if (cutoff && pktsize > cutoff) 590 return max_t(int, cutoff, 68U - tp->tcp_header_len); 591 else 592 return pktsize; 593 } 594 595 /* tcp.c */ 596 void tcp_get_info(struct sock *, struct tcp_info *); 597 598 /* Read 'sendfile()'-style from a TCP socket */ 599 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 600 sk_read_actor_t recv_actor); 601 602 void tcp_initialize_rcv_mss(struct sock *sk); 603 604 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 605 int tcp_mss_to_mtu(struct sock *sk, int mss); 606 void tcp_mtup_init(struct sock *sk); 607 void tcp_init_buffer_space(struct sock *sk); 608 609 static inline void tcp_bound_rto(const struct sock *sk) 610 { 611 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 612 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 613 } 614 615 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 616 { 617 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 618 } 619 620 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 621 { 622 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 623 ntohl(TCP_FLAG_ACK) | 624 snd_wnd); 625 } 626 627 static inline void tcp_fast_path_on(struct tcp_sock *tp) 628 { 629 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 630 } 631 632 static inline void tcp_fast_path_check(struct sock *sk) 633 { 634 struct tcp_sock *tp = tcp_sk(sk); 635 636 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 637 tp->rcv_wnd && 638 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 639 !tp->urg_data) 640 tcp_fast_path_on(tp); 641 } 642 643 /* Compute the actual rto_min value */ 644 static inline u32 tcp_rto_min(struct sock *sk) 645 { 646 const struct dst_entry *dst = __sk_dst_get(sk); 647 u32 rto_min = TCP_RTO_MIN; 648 649 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 650 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 651 return rto_min; 652 } 653 654 static inline u32 tcp_rto_min_us(struct sock *sk) 655 { 656 return jiffies_to_usecs(tcp_rto_min(sk)); 657 } 658 659 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 660 { 661 return dst_metric_locked(dst, RTAX_CC_ALGO); 662 } 663 664 /* Minimum RTT in usec. ~0 means not available. */ 665 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 666 { 667 return minmax_get(&tp->rtt_min); 668 } 669 670 /* Compute the actual receive window we are currently advertising. 671 * Rcv_nxt can be after the window if our peer push more data 672 * than the offered window. 673 */ 674 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 675 { 676 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 677 678 if (win < 0) 679 win = 0; 680 return (u32) win; 681 } 682 683 /* Choose a new window, without checks for shrinking, and without 684 * scaling applied to the result. The caller does these things 685 * if necessary. This is a "raw" window selection. 686 */ 687 u32 __tcp_select_window(struct sock *sk); 688 689 void tcp_send_window_probe(struct sock *sk); 690 691 /* TCP uses 32bit jiffies to save some space. 692 * Note that this is different from tcp_time_stamp, which 693 * historically has been the same until linux-4.13. 694 */ 695 #define tcp_jiffies32 ((u32)jiffies) 696 697 /* 698 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 699 * It is no longer tied to jiffies, but to 1 ms clock. 700 * Note: double check if you want to use tcp_jiffies32 instead of this. 701 */ 702 #define TCP_TS_HZ 1000 703 704 static inline u64 tcp_clock_ns(void) 705 { 706 return local_clock(); 707 } 708 709 static inline u64 tcp_clock_us(void) 710 { 711 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 712 } 713 714 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 715 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 716 { 717 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 718 } 719 720 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 721 static inline u32 tcp_time_stamp_raw(void) 722 { 723 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 724 } 725 726 727 /* Refresh 1us clock of a TCP socket, 728 * ensuring monotically increasing values. 729 */ 730 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 731 { 732 u64 val = tcp_clock_us(); 733 734 if (val > tp->tcp_mstamp) 735 tp->tcp_mstamp = val; 736 } 737 738 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 739 { 740 return max_t(s64, t1 - t0, 0); 741 } 742 743 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 744 { 745 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 746 } 747 748 749 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 750 751 #define TCPHDR_FIN 0x01 752 #define TCPHDR_SYN 0x02 753 #define TCPHDR_RST 0x04 754 #define TCPHDR_PSH 0x08 755 #define TCPHDR_ACK 0x10 756 #define TCPHDR_URG 0x20 757 #define TCPHDR_ECE 0x40 758 #define TCPHDR_CWR 0x80 759 760 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 761 762 /* This is what the send packet queuing engine uses to pass 763 * TCP per-packet control information to the transmission code. 764 * We also store the host-order sequence numbers in here too. 765 * This is 44 bytes if IPV6 is enabled. 766 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 767 */ 768 struct tcp_skb_cb { 769 __u32 seq; /* Starting sequence number */ 770 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 771 union { 772 /* Note : tcp_tw_isn is used in input path only 773 * (isn chosen by tcp_timewait_state_process()) 774 * 775 * tcp_gso_segs/size are used in write queue only, 776 * cf tcp_skb_pcount()/tcp_skb_mss() 777 */ 778 __u32 tcp_tw_isn; 779 struct { 780 u16 tcp_gso_segs; 781 u16 tcp_gso_size; 782 }; 783 }; 784 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 785 786 __u8 sacked; /* State flags for SACK. */ 787 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 788 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 789 #define TCPCB_LOST 0x04 /* SKB is lost */ 790 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 791 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 792 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 793 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 794 TCPCB_REPAIRED) 795 796 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 797 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 798 eor:1, /* Is skb MSG_EOR marked? */ 799 has_rxtstamp:1, /* SKB has a RX timestamp */ 800 unused:5; 801 __u32 ack_seq; /* Sequence number ACK'd */ 802 union { 803 struct { 804 /* There is space for up to 24 bytes */ 805 __u32 in_flight:30,/* Bytes in flight at transmit */ 806 is_app_limited:1, /* cwnd not fully used? */ 807 unused:1; 808 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 809 __u32 delivered; 810 /* start of send pipeline phase */ 811 u64 first_tx_mstamp; 812 /* when we reached the "delivered" count */ 813 u64 delivered_mstamp; 814 } tx; /* only used for outgoing skbs */ 815 union { 816 struct inet_skb_parm h4; 817 #if IS_ENABLED(CONFIG_IPV6) 818 struct inet6_skb_parm h6; 819 #endif 820 } header; /* For incoming skbs */ 821 struct { 822 __u32 flags; 823 struct sock *sk_redir; 824 void *data_end; 825 } bpf; 826 }; 827 }; 828 829 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 830 831 832 #if IS_ENABLED(CONFIG_IPV6) 833 /* This is the variant of inet6_iif() that must be used by TCP, 834 * as TCP moves IP6CB into a different location in skb->cb[] 835 */ 836 static inline int tcp_v6_iif(const struct sk_buff *skb) 837 { 838 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 839 840 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 841 } 842 843 /* TCP_SKB_CB reference means this can not be used from early demux */ 844 static inline int tcp_v6_sdif(const struct sk_buff *skb) 845 { 846 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 847 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 848 return TCP_SKB_CB(skb)->header.h6.iif; 849 #endif 850 return 0; 851 } 852 #endif 853 854 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 855 { 856 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 857 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 858 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 859 return true; 860 #endif 861 return false; 862 } 863 864 /* TCP_SKB_CB reference means this can not be used from early demux */ 865 static inline int tcp_v4_sdif(struct sk_buff *skb) 866 { 867 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 868 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 869 return TCP_SKB_CB(skb)->header.h4.iif; 870 #endif 871 return 0; 872 } 873 874 /* Due to TSO, an SKB can be composed of multiple actual 875 * packets. To keep these tracked properly, we use this. 876 */ 877 static inline int tcp_skb_pcount(const struct sk_buff *skb) 878 { 879 return TCP_SKB_CB(skb)->tcp_gso_segs; 880 } 881 882 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 883 { 884 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 885 } 886 887 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 888 { 889 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 890 } 891 892 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 893 static inline int tcp_skb_mss(const struct sk_buff *skb) 894 { 895 return TCP_SKB_CB(skb)->tcp_gso_size; 896 } 897 898 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 899 { 900 return likely(!TCP_SKB_CB(skb)->eor); 901 } 902 903 /* Events passed to congestion control interface */ 904 enum tcp_ca_event { 905 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 906 CA_EVENT_CWND_RESTART, /* congestion window restart */ 907 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 908 CA_EVENT_LOSS, /* loss timeout */ 909 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 910 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 911 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 912 CA_EVENT_NON_DELAYED_ACK, 913 }; 914 915 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 916 enum tcp_ca_ack_event_flags { 917 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 918 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 919 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 920 }; 921 922 /* 923 * Interface for adding new TCP congestion control handlers 924 */ 925 #define TCP_CA_NAME_MAX 16 926 #define TCP_CA_MAX 128 927 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 928 929 #define TCP_CA_UNSPEC 0 930 931 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 932 #define TCP_CONG_NON_RESTRICTED 0x1 933 /* Requires ECN/ECT set on all packets */ 934 #define TCP_CONG_NEEDS_ECN 0x2 935 936 union tcp_cc_info; 937 938 struct ack_sample { 939 u32 pkts_acked; 940 s32 rtt_us; 941 u32 in_flight; 942 }; 943 944 /* A rate sample measures the number of (original/retransmitted) data 945 * packets delivered "delivered" over an interval of time "interval_us". 946 * The tcp_rate.c code fills in the rate sample, and congestion 947 * control modules that define a cong_control function to run at the end 948 * of ACK processing can optionally chose to consult this sample when 949 * setting cwnd and pacing rate. 950 * A sample is invalid if "delivered" or "interval_us" is negative. 951 */ 952 struct rate_sample { 953 u64 prior_mstamp; /* starting timestamp for interval */ 954 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 955 s32 delivered; /* number of packets delivered over interval */ 956 long interval_us; /* time for tp->delivered to incr "delivered" */ 957 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 958 int losses; /* number of packets marked lost upon ACK */ 959 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 960 u32 prior_in_flight; /* in flight before this ACK */ 961 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 962 bool is_retrans; /* is sample from retransmission? */ 963 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 964 }; 965 966 struct tcp_congestion_ops { 967 struct list_head list; 968 u32 key; 969 u32 flags; 970 971 /* initialize private data (optional) */ 972 void (*init)(struct sock *sk); 973 /* cleanup private data (optional) */ 974 void (*release)(struct sock *sk); 975 976 /* return slow start threshold (required) */ 977 u32 (*ssthresh)(struct sock *sk); 978 /* do new cwnd calculation (required) */ 979 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 980 /* call before changing ca_state (optional) */ 981 void (*set_state)(struct sock *sk, u8 new_state); 982 /* call when cwnd event occurs (optional) */ 983 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 984 /* call when ack arrives (optional) */ 985 void (*in_ack_event)(struct sock *sk, u32 flags); 986 /* new value of cwnd after loss (required) */ 987 u32 (*undo_cwnd)(struct sock *sk); 988 /* hook for packet ack accounting (optional) */ 989 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 990 /* override sysctl_tcp_min_tso_segs */ 991 u32 (*min_tso_segs)(struct sock *sk); 992 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 993 u32 (*sndbuf_expand)(struct sock *sk); 994 /* call when packets are delivered to update cwnd and pacing rate, 995 * after all the ca_state processing. (optional) 996 */ 997 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 998 /* get info for inet_diag (optional) */ 999 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1000 union tcp_cc_info *info); 1001 1002 char name[TCP_CA_NAME_MAX]; 1003 struct module *owner; 1004 }; 1005 1006 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1007 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1008 1009 void tcp_assign_congestion_control(struct sock *sk); 1010 void tcp_init_congestion_control(struct sock *sk); 1011 void tcp_cleanup_congestion_control(struct sock *sk); 1012 int tcp_set_default_congestion_control(struct net *net, const char *name); 1013 void tcp_get_default_congestion_control(struct net *net, char *name); 1014 void tcp_get_available_congestion_control(char *buf, size_t len); 1015 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1016 int tcp_set_allowed_congestion_control(char *allowed); 1017 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1018 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1019 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1020 1021 u32 tcp_reno_ssthresh(struct sock *sk); 1022 u32 tcp_reno_undo_cwnd(struct sock *sk); 1023 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1024 extern struct tcp_congestion_ops tcp_reno; 1025 1026 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1027 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1028 #ifdef CONFIG_INET 1029 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1030 #else 1031 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1032 { 1033 return NULL; 1034 } 1035 #endif 1036 1037 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1038 { 1039 const struct inet_connection_sock *icsk = inet_csk(sk); 1040 1041 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1042 } 1043 1044 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1045 { 1046 struct inet_connection_sock *icsk = inet_csk(sk); 1047 1048 if (icsk->icsk_ca_ops->set_state) 1049 icsk->icsk_ca_ops->set_state(sk, ca_state); 1050 icsk->icsk_ca_state = ca_state; 1051 } 1052 1053 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1054 { 1055 const struct inet_connection_sock *icsk = inet_csk(sk); 1056 1057 if (icsk->icsk_ca_ops->cwnd_event) 1058 icsk->icsk_ca_ops->cwnd_event(sk, event); 1059 } 1060 1061 /* From tcp_rate.c */ 1062 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1063 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1064 struct rate_sample *rs); 1065 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1066 bool is_sack_reneg, struct rate_sample *rs); 1067 void tcp_rate_check_app_limited(struct sock *sk); 1068 1069 /* These functions determine how the current flow behaves in respect of SACK 1070 * handling. SACK is negotiated with the peer, and therefore it can vary 1071 * between different flows. 1072 * 1073 * tcp_is_sack - SACK enabled 1074 * tcp_is_reno - No SACK 1075 */ 1076 static inline int tcp_is_sack(const struct tcp_sock *tp) 1077 { 1078 return tp->rx_opt.sack_ok; 1079 } 1080 1081 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1082 { 1083 return !tcp_is_sack(tp); 1084 } 1085 1086 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1087 { 1088 return tp->sacked_out + tp->lost_out; 1089 } 1090 1091 /* This determines how many packets are "in the network" to the best 1092 * of our knowledge. In many cases it is conservative, but where 1093 * detailed information is available from the receiver (via SACK 1094 * blocks etc.) we can make more aggressive calculations. 1095 * 1096 * Use this for decisions involving congestion control, use just 1097 * tp->packets_out to determine if the send queue is empty or not. 1098 * 1099 * Read this equation as: 1100 * 1101 * "Packets sent once on transmission queue" MINUS 1102 * "Packets left network, but not honestly ACKed yet" PLUS 1103 * "Packets fast retransmitted" 1104 */ 1105 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1106 { 1107 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1108 } 1109 1110 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1111 1112 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1113 { 1114 return tp->snd_cwnd < tp->snd_ssthresh; 1115 } 1116 1117 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1118 { 1119 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1120 } 1121 1122 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1123 { 1124 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1125 (1 << inet_csk(sk)->icsk_ca_state); 1126 } 1127 1128 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1129 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1130 * ssthresh. 1131 */ 1132 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1133 { 1134 const struct tcp_sock *tp = tcp_sk(sk); 1135 1136 if (tcp_in_cwnd_reduction(sk)) 1137 return tp->snd_ssthresh; 1138 else 1139 return max(tp->snd_ssthresh, 1140 ((tp->snd_cwnd >> 1) + 1141 (tp->snd_cwnd >> 2))); 1142 } 1143 1144 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1145 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1146 1147 void tcp_enter_cwr(struct sock *sk); 1148 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1149 1150 /* The maximum number of MSS of available cwnd for which TSO defers 1151 * sending if not using sysctl_tcp_tso_win_divisor. 1152 */ 1153 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1154 { 1155 return 3; 1156 } 1157 1158 /* Returns end sequence number of the receiver's advertised window */ 1159 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1160 { 1161 return tp->snd_una + tp->snd_wnd; 1162 } 1163 1164 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1165 * flexible approach. The RFC suggests cwnd should not be raised unless 1166 * it was fully used previously. And that's exactly what we do in 1167 * congestion avoidance mode. But in slow start we allow cwnd to grow 1168 * as long as the application has used half the cwnd. 1169 * Example : 1170 * cwnd is 10 (IW10), but application sends 9 frames. 1171 * We allow cwnd to reach 18 when all frames are ACKed. 1172 * This check is safe because it's as aggressive as slow start which already 1173 * risks 100% overshoot. The advantage is that we discourage application to 1174 * either send more filler packets or data to artificially blow up the cwnd 1175 * usage, and allow application-limited process to probe bw more aggressively. 1176 */ 1177 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1178 { 1179 const struct tcp_sock *tp = tcp_sk(sk); 1180 1181 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1182 if (tcp_in_slow_start(tp)) 1183 return tp->snd_cwnd < 2 * tp->max_packets_out; 1184 1185 return tp->is_cwnd_limited; 1186 } 1187 1188 /* BBR congestion control needs pacing. 1189 * Same remark for SO_MAX_PACING_RATE. 1190 * sch_fq packet scheduler is efficiently handling pacing, 1191 * but is not always installed/used. 1192 * Return true if TCP stack should pace packets itself. 1193 */ 1194 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1195 { 1196 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1197 } 1198 1199 /* Something is really bad, we could not queue an additional packet, 1200 * because qdisc is full or receiver sent a 0 window. 1201 * We do not want to add fuel to the fire, or abort too early, 1202 * so make sure the timer we arm now is at least 200ms in the future, 1203 * regardless of current icsk_rto value (as it could be ~2ms) 1204 */ 1205 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1206 { 1207 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1208 } 1209 1210 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1211 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1212 unsigned long max_when) 1213 { 1214 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1215 1216 return (unsigned long)min_t(u64, when, max_when); 1217 } 1218 1219 static inline void tcp_check_probe_timer(struct sock *sk) 1220 { 1221 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1222 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1223 tcp_probe0_base(sk), TCP_RTO_MAX); 1224 } 1225 1226 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1227 { 1228 tp->snd_wl1 = seq; 1229 } 1230 1231 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1232 { 1233 tp->snd_wl1 = seq; 1234 } 1235 1236 /* 1237 * Calculate(/check) TCP checksum 1238 */ 1239 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1240 __be32 daddr, __wsum base) 1241 { 1242 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1243 } 1244 1245 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1246 { 1247 return __skb_checksum_complete(skb); 1248 } 1249 1250 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1251 { 1252 return !skb_csum_unnecessary(skb) && 1253 __tcp_checksum_complete(skb); 1254 } 1255 1256 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1257 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1258 1259 #undef STATE_TRACE 1260 1261 #ifdef STATE_TRACE 1262 static const char *statename[]={ 1263 "Unused","Established","Syn Sent","Syn Recv", 1264 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1265 "Close Wait","Last ACK","Listen","Closing" 1266 }; 1267 #endif 1268 void tcp_set_state(struct sock *sk, int state); 1269 1270 void tcp_done(struct sock *sk); 1271 1272 int tcp_abort(struct sock *sk, int err); 1273 1274 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1275 { 1276 rx_opt->dsack = 0; 1277 rx_opt->num_sacks = 0; 1278 } 1279 1280 u32 tcp_default_init_rwnd(u32 mss); 1281 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1282 1283 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1284 { 1285 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1286 struct tcp_sock *tp = tcp_sk(sk); 1287 s32 delta; 1288 1289 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1290 ca_ops->cong_control) 1291 return; 1292 delta = tcp_jiffies32 - tp->lsndtime; 1293 if (delta > inet_csk(sk)->icsk_rto) 1294 tcp_cwnd_restart(sk, delta); 1295 } 1296 1297 /* Determine a window scaling and initial window to offer. */ 1298 void tcp_select_initial_window(const struct sock *sk, int __space, 1299 __u32 mss, __u32 *rcv_wnd, 1300 __u32 *window_clamp, int wscale_ok, 1301 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1302 1303 static inline int tcp_win_from_space(const struct sock *sk, int space) 1304 { 1305 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1306 1307 return tcp_adv_win_scale <= 0 ? 1308 (space>>(-tcp_adv_win_scale)) : 1309 space - (space>>tcp_adv_win_scale); 1310 } 1311 1312 /* Note: caller must be prepared to deal with negative returns */ 1313 static inline int tcp_space(const struct sock *sk) 1314 { 1315 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1316 atomic_read(&sk->sk_rmem_alloc)); 1317 } 1318 1319 static inline int tcp_full_space(const struct sock *sk) 1320 { 1321 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1322 } 1323 1324 extern void tcp_openreq_init_rwin(struct request_sock *req, 1325 const struct sock *sk_listener, 1326 const struct dst_entry *dst); 1327 1328 void tcp_enter_memory_pressure(struct sock *sk); 1329 void tcp_leave_memory_pressure(struct sock *sk); 1330 1331 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1332 { 1333 struct net *net = sock_net((struct sock *)tp); 1334 1335 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1336 } 1337 1338 static inline int keepalive_time_when(const struct tcp_sock *tp) 1339 { 1340 struct net *net = sock_net((struct sock *)tp); 1341 1342 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1343 } 1344 1345 static inline int keepalive_probes(const struct tcp_sock *tp) 1346 { 1347 struct net *net = sock_net((struct sock *)tp); 1348 1349 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1350 } 1351 1352 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1353 { 1354 const struct inet_connection_sock *icsk = &tp->inet_conn; 1355 1356 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1357 tcp_jiffies32 - tp->rcv_tstamp); 1358 } 1359 1360 static inline int tcp_fin_time(const struct sock *sk) 1361 { 1362 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1363 const int rto = inet_csk(sk)->icsk_rto; 1364 1365 if (fin_timeout < (rto << 2) - (rto >> 1)) 1366 fin_timeout = (rto << 2) - (rto >> 1); 1367 1368 return fin_timeout; 1369 } 1370 1371 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1372 int paws_win) 1373 { 1374 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1375 return true; 1376 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1377 return true; 1378 /* 1379 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1380 * then following tcp messages have valid values. Ignore 0 value, 1381 * or else 'negative' tsval might forbid us to accept their packets. 1382 */ 1383 if (!rx_opt->ts_recent) 1384 return true; 1385 return false; 1386 } 1387 1388 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1389 int rst) 1390 { 1391 if (tcp_paws_check(rx_opt, 0)) 1392 return false; 1393 1394 /* RST segments are not recommended to carry timestamp, 1395 and, if they do, it is recommended to ignore PAWS because 1396 "their cleanup function should take precedence over timestamps." 1397 Certainly, it is mistake. It is necessary to understand the reasons 1398 of this constraint to relax it: if peer reboots, clock may go 1399 out-of-sync and half-open connections will not be reset. 1400 Actually, the problem would be not existing if all 1401 the implementations followed draft about maintaining clock 1402 via reboots. Linux-2.2 DOES NOT! 1403 1404 However, we can relax time bounds for RST segments to MSL. 1405 */ 1406 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1407 return false; 1408 return true; 1409 } 1410 1411 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1412 int mib_idx, u32 *last_oow_ack_time); 1413 1414 static inline void tcp_mib_init(struct net *net) 1415 { 1416 /* See RFC 2012 */ 1417 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1418 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1419 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1420 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1421 } 1422 1423 /* from STCP */ 1424 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1425 { 1426 tp->lost_skb_hint = NULL; 1427 } 1428 1429 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1430 { 1431 tcp_clear_retrans_hints_partial(tp); 1432 tp->retransmit_skb_hint = NULL; 1433 } 1434 1435 union tcp_md5_addr { 1436 struct in_addr a4; 1437 #if IS_ENABLED(CONFIG_IPV6) 1438 struct in6_addr a6; 1439 #endif 1440 }; 1441 1442 /* - key database */ 1443 struct tcp_md5sig_key { 1444 struct hlist_node node; 1445 u8 keylen; 1446 u8 family; /* AF_INET or AF_INET6 */ 1447 union tcp_md5_addr addr; 1448 u8 prefixlen; 1449 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1450 struct rcu_head rcu; 1451 }; 1452 1453 /* - sock block */ 1454 struct tcp_md5sig_info { 1455 struct hlist_head head; 1456 struct rcu_head rcu; 1457 }; 1458 1459 /* - pseudo header */ 1460 struct tcp4_pseudohdr { 1461 __be32 saddr; 1462 __be32 daddr; 1463 __u8 pad; 1464 __u8 protocol; 1465 __be16 len; 1466 }; 1467 1468 struct tcp6_pseudohdr { 1469 struct in6_addr saddr; 1470 struct in6_addr daddr; 1471 __be32 len; 1472 __be32 protocol; /* including padding */ 1473 }; 1474 1475 union tcp_md5sum_block { 1476 struct tcp4_pseudohdr ip4; 1477 #if IS_ENABLED(CONFIG_IPV6) 1478 struct tcp6_pseudohdr ip6; 1479 #endif 1480 }; 1481 1482 /* - pool: digest algorithm, hash description and scratch buffer */ 1483 struct tcp_md5sig_pool { 1484 struct ahash_request *md5_req; 1485 void *scratch; 1486 }; 1487 1488 /* - functions */ 1489 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1490 const struct sock *sk, const struct sk_buff *skb); 1491 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1492 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1493 gfp_t gfp); 1494 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1495 int family, u8 prefixlen); 1496 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1497 const struct sock *addr_sk); 1498 1499 #ifdef CONFIG_TCP_MD5SIG 1500 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1501 const union tcp_md5_addr *addr, 1502 int family); 1503 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1504 #else 1505 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1506 const union tcp_md5_addr *addr, 1507 int family) 1508 { 1509 return NULL; 1510 } 1511 #define tcp_twsk_md5_key(twsk) NULL 1512 #endif 1513 1514 bool tcp_alloc_md5sig_pool(void); 1515 1516 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1517 static inline void tcp_put_md5sig_pool(void) 1518 { 1519 local_bh_enable(); 1520 } 1521 1522 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1523 unsigned int header_len); 1524 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1525 const struct tcp_md5sig_key *key); 1526 1527 /* From tcp_fastopen.c */ 1528 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1529 struct tcp_fastopen_cookie *cookie); 1530 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1531 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1532 u16 try_exp); 1533 struct tcp_fastopen_request { 1534 /* Fast Open cookie. Size 0 means a cookie request */ 1535 struct tcp_fastopen_cookie cookie; 1536 struct msghdr *data; /* data in MSG_FASTOPEN */ 1537 size_t size; 1538 int copied; /* queued in tcp_connect() */ 1539 }; 1540 void tcp_free_fastopen_req(struct tcp_sock *tp); 1541 void tcp_fastopen_destroy_cipher(struct sock *sk); 1542 void tcp_fastopen_ctx_destroy(struct net *net); 1543 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1544 void *key, unsigned int len); 1545 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1546 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1547 struct request_sock *req, 1548 struct tcp_fastopen_cookie *foc, 1549 const struct dst_entry *dst); 1550 void tcp_fastopen_init_key_once(struct net *net); 1551 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1552 struct tcp_fastopen_cookie *cookie); 1553 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1554 #define TCP_FASTOPEN_KEY_LENGTH 16 1555 1556 /* Fastopen key context */ 1557 struct tcp_fastopen_context { 1558 struct crypto_cipher *tfm; 1559 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1560 struct rcu_head rcu; 1561 }; 1562 1563 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1564 void tcp_fastopen_active_disable(struct sock *sk); 1565 bool tcp_fastopen_active_should_disable(struct sock *sk); 1566 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1567 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1568 1569 /* Latencies incurred by various limits for a sender. They are 1570 * chronograph-like stats that are mutually exclusive. 1571 */ 1572 enum tcp_chrono { 1573 TCP_CHRONO_UNSPEC, 1574 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1575 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1576 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1577 __TCP_CHRONO_MAX, 1578 }; 1579 1580 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1581 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1582 1583 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1584 * the same memory storage than skb->destructor/_skb_refdst 1585 */ 1586 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1587 { 1588 skb->destructor = NULL; 1589 skb->_skb_refdst = 0UL; 1590 } 1591 1592 #define tcp_skb_tsorted_save(skb) { \ 1593 unsigned long _save = skb->_skb_refdst; \ 1594 skb->_skb_refdst = 0UL; 1595 1596 #define tcp_skb_tsorted_restore(skb) \ 1597 skb->_skb_refdst = _save; \ 1598 } 1599 1600 void tcp_write_queue_purge(struct sock *sk); 1601 1602 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1603 { 1604 return skb_rb_first(&sk->tcp_rtx_queue); 1605 } 1606 1607 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1608 { 1609 return skb_peek(&sk->sk_write_queue); 1610 } 1611 1612 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1613 { 1614 return skb_peek_tail(&sk->sk_write_queue); 1615 } 1616 1617 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1618 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1619 1620 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1621 { 1622 return skb_peek(&sk->sk_write_queue); 1623 } 1624 1625 static inline bool tcp_skb_is_last(const struct sock *sk, 1626 const struct sk_buff *skb) 1627 { 1628 return skb_queue_is_last(&sk->sk_write_queue, skb); 1629 } 1630 1631 static inline bool tcp_write_queue_empty(const struct sock *sk) 1632 { 1633 return skb_queue_empty(&sk->sk_write_queue); 1634 } 1635 1636 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1637 { 1638 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1639 } 1640 1641 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1642 { 1643 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1644 } 1645 1646 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1647 { 1648 if (tcp_write_queue_empty(sk)) 1649 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1650 } 1651 1652 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1653 { 1654 __skb_queue_tail(&sk->sk_write_queue, skb); 1655 } 1656 1657 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1658 { 1659 __tcp_add_write_queue_tail(sk, skb); 1660 1661 /* Queue it, remembering where we must start sending. */ 1662 if (sk->sk_write_queue.next == skb) 1663 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1664 } 1665 1666 /* Insert new before skb on the write queue of sk. */ 1667 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1668 struct sk_buff *skb, 1669 struct sock *sk) 1670 { 1671 __skb_queue_before(&sk->sk_write_queue, skb, new); 1672 } 1673 1674 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1675 { 1676 tcp_skb_tsorted_anchor_cleanup(skb); 1677 __skb_unlink(skb, &sk->sk_write_queue); 1678 } 1679 1680 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1681 1682 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1683 { 1684 tcp_skb_tsorted_anchor_cleanup(skb); 1685 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1686 } 1687 1688 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1689 { 1690 list_del(&skb->tcp_tsorted_anchor); 1691 tcp_rtx_queue_unlink(skb, sk); 1692 sk_wmem_free_skb(sk, skb); 1693 } 1694 1695 static inline void tcp_push_pending_frames(struct sock *sk) 1696 { 1697 if (tcp_send_head(sk)) { 1698 struct tcp_sock *tp = tcp_sk(sk); 1699 1700 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1701 } 1702 } 1703 1704 /* Start sequence of the skb just after the highest skb with SACKed 1705 * bit, valid only if sacked_out > 0 or when the caller has ensured 1706 * validity by itself. 1707 */ 1708 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1709 { 1710 if (!tp->sacked_out) 1711 return tp->snd_una; 1712 1713 if (tp->highest_sack == NULL) 1714 return tp->snd_nxt; 1715 1716 return TCP_SKB_CB(tp->highest_sack)->seq; 1717 } 1718 1719 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1720 { 1721 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1722 } 1723 1724 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1725 { 1726 return tcp_sk(sk)->highest_sack; 1727 } 1728 1729 static inline void tcp_highest_sack_reset(struct sock *sk) 1730 { 1731 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1732 } 1733 1734 /* Called when old skb is about to be deleted and replaced by new skb */ 1735 static inline void tcp_highest_sack_replace(struct sock *sk, 1736 struct sk_buff *old, 1737 struct sk_buff *new) 1738 { 1739 if (old == tcp_highest_sack(sk)) 1740 tcp_sk(sk)->highest_sack = new; 1741 } 1742 1743 /* This helper checks if socket has IP_TRANSPARENT set */ 1744 static inline bool inet_sk_transparent(const struct sock *sk) 1745 { 1746 switch (sk->sk_state) { 1747 case TCP_TIME_WAIT: 1748 return inet_twsk(sk)->tw_transparent; 1749 case TCP_NEW_SYN_RECV: 1750 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1751 } 1752 return inet_sk(sk)->transparent; 1753 } 1754 1755 /* Determines whether this is a thin stream (which may suffer from 1756 * increased latency). Used to trigger latency-reducing mechanisms. 1757 */ 1758 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1759 { 1760 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1761 } 1762 1763 /* /proc */ 1764 enum tcp_seq_states { 1765 TCP_SEQ_STATE_LISTENING, 1766 TCP_SEQ_STATE_ESTABLISHED, 1767 }; 1768 1769 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1770 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1771 void tcp_seq_stop(struct seq_file *seq, void *v); 1772 1773 struct tcp_seq_afinfo { 1774 sa_family_t family; 1775 }; 1776 1777 struct tcp_iter_state { 1778 struct seq_net_private p; 1779 enum tcp_seq_states state; 1780 struct sock *syn_wait_sk; 1781 int bucket, offset, sbucket, num; 1782 loff_t last_pos; 1783 }; 1784 1785 extern struct request_sock_ops tcp_request_sock_ops; 1786 extern struct request_sock_ops tcp6_request_sock_ops; 1787 1788 void tcp_v4_destroy_sock(struct sock *sk); 1789 1790 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1791 netdev_features_t features); 1792 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1793 int tcp_gro_complete(struct sk_buff *skb); 1794 1795 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1796 1797 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1798 { 1799 struct net *net = sock_net((struct sock *)tp); 1800 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1801 } 1802 1803 static inline bool tcp_stream_memory_free(const struct sock *sk) 1804 { 1805 const struct tcp_sock *tp = tcp_sk(sk); 1806 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1807 1808 return notsent_bytes < tcp_notsent_lowat(tp); 1809 } 1810 1811 #ifdef CONFIG_PROC_FS 1812 int tcp4_proc_init(void); 1813 void tcp4_proc_exit(void); 1814 #endif 1815 1816 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1817 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1818 const struct tcp_request_sock_ops *af_ops, 1819 struct sock *sk, struct sk_buff *skb); 1820 1821 /* TCP af-specific functions */ 1822 struct tcp_sock_af_ops { 1823 #ifdef CONFIG_TCP_MD5SIG 1824 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1825 const struct sock *addr_sk); 1826 int (*calc_md5_hash)(char *location, 1827 const struct tcp_md5sig_key *md5, 1828 const struct sock *sk, 1829 const struct sk_buff *skb); 1830 int (*md5_parse)(struct sock *sk, 1831 int optname, 1832 char __user *optval, 1833 int optlen); 1834 #endif 1835 }; 1836 1837 struct tcp_request_sock_ops { 1838 u16 mss_clamp; 1839 #ifdef CONFIG_TCP_MD5SIG 1840 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1841 const struct sock *addr_sk); 1842 int (*calc_md5_hash) (char *location, 1843 const struct tcp_md5sig_key *md5, 1844 const struct sock *sk, 1845 const struct sk_buff *skb); 1846 #endif 1847 void (*init_req)(struct request_sock *req, 1848 const struct sock *sk_listener, 1849 struct sk_buff *skb); 1850 #ifdef CONFIG_SYN_COOKIES 1851 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1852 __u16 *mss); 1853 #endif 1854 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1855 const struct request_sock *req); 1856 u32 (*init_seq)(const struct sk_buff *skb); 1857 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1858 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1859 struct flowi *fl, struct request_sock *req, 1860 struct tcp_fastopen_cookie *foc, 1861 enum tcp_synack_type synack_type); 1862 }; 1863 1864 #ifdef CONFIG_SYN_COOKIES 1865 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1866 const struct sock *sk, struct sk_buff *skb, 1867 __u16 *mss) 1868 { 1869 tcp_synq_overflow(sk); 1870 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1871 return ops->cookie_init_seq(skb, mss); 1872 } 1873 #else 1874 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1875 const struct sock *sk, struct sk_buff *skb, 1876 __u16 *mss) 1877 { 1878 return 0; 1879 } 1880 #endif 1881 1882 int tcpv4_offload_init(void); 1883 1884 void tcp_v4_init(void); 1885 void tcp_init(void); 1886 1887 /* tcp_recovery.c */ 1888 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1889 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1890 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1891 u32 reo_wnd); 1892 extern void tcp_rack_mark_lost(struct sock *sk); 1893 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1894 u64 xmit_time); 1895 extern void tcp_rack_reo_timeout(struct sock *sk); 1896 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1897 1898 /* At how many usecs into the future should the RTO fire? */ 1899 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1900 { 1901 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1902 u32 rto = inet_csk(sk)->icsk_rto; 1903 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1904 1905 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1906 } 1907 1908 /* 1909 * Save and compile IPv4 options, return a pointer to it 1910 */ 1911 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1912 struct sk_buff *skb) 1913 { 1914 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1915 struct ip_options_rcu *dopt = NULL; 1916 1917 if (opt->optlen) { 1918 int opt_size = sizeof(*dopt) + opt->optlen; 1919 1920 dopt = kmalloc(opt_size, GFP_ATOMIC); 1921 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1922 kfree(dopt); 1923 dopt = NULL; 1924 } 1925 } 1926 return dopt; 1927 } 1928 1929 /* locally generated TCP pure ACKs have skb->truesize == 2 1930 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1931 * This is much faster than dissecting the packet to find out. 1932 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1933 */ 1934 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1935 { 1936 return skb->truesize == 2; 1937 } 1938 1939 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1940 { 1941 skb->truesize = 2; 1942 } 1943 1944 static inline int tcp_inq(struct sock *sk) 1945 { 1946 struct tcp_sock *tp = tcp_sk(sk); 1947 int answ; 1948 1949 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1950 answ = 0; 1951 } else if (sock_flag(sk, SOCK_URGINLINE) || 1952 !tp->urg_data || 1953 before(tp->urg_seq, tp->copied_seq) || 1954 !before(tp->urg_seq, tp->rcv_nxt)) { 1955 1956 answ = tp->rcv_nxt - tp->copied_seq; 1957 1958 /* Subtract 1, if FIN was received */ 1959 if (answ && sock_flag(sk, SOCK_DONE)) 1960 answ--; 1961 } else { 1962 answ = tp->urg_seq - tp->copied_seq; 1963 } 1964 1965 return answ; 1966 } 1967 1968 int tcp_peek_len(struct socket *sock); 1969 1970 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1971 { 1972 u16 segs_in; 1973 1974 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1975 tp->segs_in += segs_in; 1976 if (skb->len > tcp_hdrlen(skb)) 1977 tp->data_segs_in += segs_in; 1978 } 1979 1980 /* 1981 * TCP listen path runs lockless. 1982 * We forced "struct sock" to be const qualified to make sure 1983 * we don't modify one of its field by mistake. 1984 * Here, we increment sk_drops which is an atomic_t, so we can safely 1985 * make sock writable again. 1986 */ 1987 static inline void tcp_listendrop(const struct sock *sk) 1988 { 1989 atomic_inc(&((struct sock *)sk)->sk_drops); 1990 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1991 } 1992 1993 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 1994 1995 /* 1996 * Interface for adding Upper Level Protocols over TCP 1997 */ 1998 1999 #define TCP_ULP_NAME_MAX 16 2000 #define TCP_ULP_MAX 128 2001 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2002 2003 enum { 2004 TCP_ULP_TLS, 2005 TCP_ULP_BPF, 2006 }; 2007 2008 struct tcp_ulp_ops { 2009 struct list_head list; 2010 2011 /* initialize ulp */ 2012 int (*init)(struct sock *sk); 2013 /* cleanup ulp */ 2014 void (*release)(struct sock *sk); 2015 2016 int uid; 2017 char name[TCP_ULP_NAME_MAX]; 2018 bool user_visible; 2019 struct module *owner; 2020 }; 2021 int tcp_register_ulp(struct tcp_ulp_ops *type); 2022 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2023 int tcp_set_ulp(struct sock *sk, const char *name); 2024 int tcp_set_ulp_id(struct sock *sk, const int ulp); 2025 void tcp_get_available_ulp(char *buf, size_t len); 2026 void tcp_cleanup_ulp(struct sock *sk); 2027 2028 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2029 * is < 0, then the BPF op failed (for example if the loaded BPF 2030 * program does not support the chosen operation or there is no BPF 2031 * program loaded). 2032 */ 2033 #ifdef CONFIG_BPF 2034 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2035 { 2036 struct bpf_sock_ops_kern sock_ops; 2037 int ret; 2038 2039 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2040 if (sk_fullsock(sk)) { 2041 sock_ops.is_fullsock = 1; 2042 sock_owned_by_me(sk); 2043 } 2044 2045 sock_ops.sk = sk; 2046 sock_ops.op = op; 2047 if (nargs > 0) 2048 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2049 2050 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2051 if (ret == 0) 2052 ret = sock_ops.reply; 2053 else 2054 ret = -1; 2055 return ret; 2056 } 2057 2058 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2059 { 2060 u32 args[2] = {arg1, arg2}; 2061 2062 return tcp_call_bpf(sk, op, 2, args); 2063 } 2064 2065 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2066 u32 arg3) 2067 { 2068 u32 args[3] = {arg1, arg2, arg3}; 2069 2070 return tcp_call_bpf(sk, op, 3, args); 2071 } 2072 2073 #else 2074 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2075 { 2076 return -EPERM; 2077 } 2078 2079 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2080 { 2081 return -EPERM; 2082 } 2083 2084 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2085 u32 arg3) 2086 { 2087 return -EPERM; 2088 } 2089 2090 #endif 2091 2092 static inline u32 tcp_timeout_init(struct sock *sk) 2093 { 2094 int timeout; 2095 2096 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2097 2098 if (timeout <= 0) 2099 timeout = TCP_TIMEOUT_INIT; 2100 return timeout; 2101 } 2102 2103 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2104 { 2105 int rwnd; 2106 2107 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2108 2109 if (rwnd < 0) 2110 rwnd = 0; 2111 return rwnd; 2112 } 2113 2114 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2115 { 2116 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2117 } 2118 2119 #if IS_ENABLED(CONFIG_SMC) 2120 extern struct static_key_false tcp_have_smc; 2121 #endif 2122 2123 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2124 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2125 void (*cad)(struct sock *sk, u32 ack_seq)); 2126 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2127 2128 #endif 2129 2130 #endif /* _TCP_H */ 2131