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 const struct tcphdr *th); 339 void tcp_rcv_space_adjust(struct sock *sk); 340 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 341 void tcp_twsk_destructor(struct sock *sk); 342 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 343 struct pipe_inode_info *pipe, size_t len, 344 unsigned int flags); 345 346 static inline void tcp_dec_quickack_mode(struct sock *sk, 347 const unsigned int pkts) 348 { 349 struct inet_connection_sock *icsk = inet_csk(sk); 350 351 if (icsk->icsk_ack.quick) { 352 if (pkts >= icsk->icsk_ack.quick) { 353 icsk->icsk_ack.quick = 0; 354 /* Leaving quickack mode we deflate ATO. */ 355 icsk->icsk_ack.ato = TCP_ATO_MIN; 356 } else 357 icsk->icsk_ack.quick -= pkts; 358 } 359 } 360 361 #define TCP_ECN_OK 1 362 #define TCP_ECN_QUEUE_CWR 2 363 #define TCP_ECN_DEMAND_CWR 4 364 #define TCP_ECN_SEEN 8 365 366 enum tcp_tw_status { 367 TCP_TW_SUCCESS = 0, 368 TCP_TW_RST = 1, 369 TCP_TW_ACK = 2, 370 TCP_TW_SYN = 3 371 }; 372 373 374 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 375 struct sk_buff *skb, 376 const struct tcphdr *th); 377 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 378 struct request_sock *req, bool fastopen, 379 bool *lost_race); 380 int tcp_child_process(struct sock *parent, struct sock *child, 381 struct sk_buff *skb); 382 void tcp_enter_loss(struct sock *sk); 383 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 384 void tcp_clear_retrans(struct tcp_sock *tp); 385 void tcp_update_metrics(struct sock *sk); 386 void tcp_init_metrics(struct sock *sk); 387 void tcp_metrics_init(void); 388 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 389 void tcp_close(struct sock *sk, long timeout); 390 void tcp_init_sock(struct sock *sk); 391 void tcp_init_transfer(struct sock *sk, int bpf_op); 392 __poll_t tcp_poll(struct file *file, struct socket *sock, 393 struct poll_table_struct *wait); 394 int tcp_getsockopt(struct sock *sk, int level, int optname, 395 char __user *optval, int __user *optlen); 396 int tcp_setsockopt(struct sock *sk, int level, int optname, 397 char __user *optval, unsigned int optlen); 398 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 399 char __user *optval, int __user *optlen); 400 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 401 char __user *optval, unsigned int optlen); 402 void tcp_set_keepalive(struct sock *sk, int val); 403 void tcp_syn_ack_timeout(const struct request_sock *req); 404 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 405 int flags, int *addr_len); 406 int tcp_set_rcvlowat(struct sock *sk, int val); 407 void tcp_data_ready(struct sock *sk); 408 int tcp_mmap(struct file *file, struct socket *sock, 409 struct vm_area_struct *vma); 410 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 411 struct tcp_options_received *opt_rx, 412 int estab, struct tcp_fastopen_cookie *foc); 413 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 414 415 /* 416 * TCP v4 functions exported for the inet6 API 417 */ 418 419 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 420 void tcp_v4_mtu_reduced(struct sock *sk); 421 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 422 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 423 struct sock *tcp_create_openreq_child(const struct sock *sk, 424 struct request_sock *req, 425 struct sk_buff *skb); 426 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 427 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 428 struct request_sock *req, 429 struct dst_entry *dst, 430 struct request_sock *req_unhash, 431 bool *own_req); 432 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 433 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 434 int tcp_connect(struct sock *sk); 435 enum tcp_synack_type { 436 TCP_SYNACK_NORMAL, 437 TCP_SYNACK_FASTOPEN, 438 TCP_SYNACK_COOKIE, 439 }; 440 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 441 struct request_sock *req, 442 struct tcp_fastopen_cookie *foc, 443 enum tcp_synack_type synack_type); 444 int tcp_disconnect(struct sock *sk, int flags); 445 446 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 447 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 448 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 449 450 /* From syncookies.c */ 451 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 452 struct request_sock *req, 453 struct dst_entry *dst, u32 tsoff); 454 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 455 u32 cookie); 456 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 457 #ifdef CONFIG_SYN_COOKIES 458 459 /* Syncookies use a monotonic timer which increments every 60 seconds. 460 * This counter is used both as a hash input and partially encoded into 461 * the cookie value. A cookie is only validated further if the delta 462 * between the current counter value and the encoded one is less than this, 463 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 464 * the counter advances immediately after a cookie is generated). 465 */ 466 #define MAX_SYNCOOKIE_AGE 2 467 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 468 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 469 470 /* syncookies: remember time of last synqueue overflow 471 * But do not dirty this field too often (once per second is enough) 472 * It is racy as we do not hold a lock, but race is very minor. 473 */ 474 static inline void tcp_synq_overflow(const struct sock *sk) 475 { 476 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 477 unsigned long now = jiffies; 478 479 if (time_after(now, last_overflow + HZ)) 480 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 481 } 482 483 /* syncookies: no recent synqueue overflow on this listening socket? */ 484 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 485 { 486 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 487 488 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 489 } 490 491 static inline u32 tcp_cookie_time(void) 492 { 493 u64 val = get_jiffies_64(); 494 495 do_div(val, TCP_SYNCOOKIE_PERIOD); 496 return val; 497 } 498 499 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 500 u16 *mssp); 501 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 502 u64 cookie_init_timestamp(struct request_sock *req); 503 bool cookie_timestamp_decode(const struct net *net, 504 struct tcp_options_received *opt); 505 bool cookie_ecn_ok(const struct tcp_options_received *opt, 506 const struct net *net, const struct dst_entry *dst); 507 508 /* From net/ipv6/syncookies.c */ 509 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 510 u32 cookie); 511 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 512 513 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 514 const struct tcphdr *th, u16 *mssp); 515 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 516 #endif 517 /* tcp_output.c */ 518 519 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 520 int nonagle); 521 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 522 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 523 void tcp_retransmit_timer(struct sock *sk); 524 void tcp_xmit_retransmit_queue(struct sock *); 525 void tcp_simple_retransmit(struct sock *); 526 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 527 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 528 enum tcp_queue { 529 TCP_FRAG_IN_WRITE_QUEUE, 530 TCP_FRAG_IN_RTX_QUEUE, 531 }; 532 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 533 struct sk_buff *skb, u32 len, 534 unsigned int mss_now, gfp_t gfp); 535 536 void tcp_send_probe0(struct sock *); 537 void tcp_send_partial(struct sock *); 538 int tcp_write_wakeup(struct sock *, int mib); 539 void tcp_send_fin(struct sock *sk); 540 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 541 int tcp_send_synack(struct sock *); 542 void tcp_push_one(struct sock *, unsigned int mss_now); 543 void tcp_send_ack(struct sock *sk); 544 void tcp_send_delayed_ack(struct sock *sk); 545 void tcp_send_loss_probe(struct sock *sk); 546 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 547 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 548 const struct sk_buff *next_skb); 549 550 /* tcp_input.c */ 551 void tcp_rearm_rto(struct sock *sk); 552 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 553 void tcp_reset(struct sock *sk); 554 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 555 void tcp_fin(struct sock *sk); 556 557 /* tcp_timer.c */ 558 void tcp_init_xmit_timers(struct sock *); 559 static inline void tcp_clear_xmit_timers(struct sock *sk) 560 { 561 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 562 __sock_put(sk); 563 564 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 565 __sock_put(sk); 566 567 inet_csk_clear_xmit_timers(sk); 568 } 569 570 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 571 unsigned int tcp_current_mss(struct sock *sk); 572 573 /* Bound MSS / TSO packet size with the half of the window */ 574 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 575 { 576 int cutoff; 577 578 /* When peer uses tiny windows, there is no use in packetizing 579 * to sub-MSS pieces for the sake of SWS or making sure there 580 * are enough packets in the pipe for fast recovery. 581 * 582 * On the other hand, for extremely large MSS devices, handling 583 * smaller than MSS windows in this way does make sense. 584 */ 585 if (tp->max_window > TCP_MSS_DEFAULT) 586 cutoff = (tp->max_window >> 1); 587 else 588 cutoff = tp->max_window; 589 590 if (cutoff && pktsize > cutoff) 591 return max_t(int, cutoff, 68U - tp->tcp_header_len); 592 else 593 return pktsize; 594 } 595 596 /* tcp.c */ 597 void tcp_get_info(struct sock *, struct tcp_info *); 598 599 /* Read 'sendfile()'-style from a TCP socket */ 600 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 601 sk_read_actor_t recv_actor); 602 603 void tcp_initialize_rcv_mss(struct sock *sk); 604 605 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 606 int tcp_mss_to_mtu(struct sock *sk, int mss); 607 void tcp_mtup_init(struct sock *sk); 608 void tcp_init_buffer_space(struct sock *sk); 609 610 static inline void tcp_bound_rto(const struct sock *sk) 611 { 612 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 613 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 614 } 615 616 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 617 { 618 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 619 } 620 621 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 622 { 623 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 624 ntohl(TCP_FLAG_ACK) | 625 snd_wnd); 626 } 627 628 static inline void tcp_fast_path_on(struct tcp_sock *tp) 629 { 630 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 631 } 632 633 static inline void tcp_fast_path_check(struct sock *sk) 634 { 635 struct tcp_sock *tp = tcp_sk(sk); 636 637 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 638 tp->rcv_wnd && 639 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 640 !tp->urg_data) 641 tcp_fast_path_on(tp); 642 } 643 644 /* Compute the actual rto_min value */ 645 static inline u32 tcp_rto_min(struct sock *sk) 646 { 647 const struct dst_entry *dst = __sk_dst_get(sk); 648 u32 rto_min = TCP_RTO_MIN; 649 650 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 651 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 652 return rto_min; 653 } 654 655 static inline u32 tcp_rto_min_us(struct sock *sk) 656 { 657 return jiffies_to_usecs(tcp_rto_min(sk)); 658 } 659 660 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 661 { 662 return dst_metric_locked(dst, RTAX_CC_ALGO); 663 } 664 665 /* Minimum RTT in usec. ~0 means not available. */ 666 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 667 { 668 return minmax_get(&tp->rtt_min); 669 } 670 671 /* Compute the actual receive window we are currently advertising. 672 * Rcv_nxt can be after the window if our peer push more data 673 * than the offered window. 674 */ 675 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 676 { 677 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 678 679 if (win < 0) 680 win = 0; 681 return (u32) win; 682 } 683 684 /* Choose a new window, without checks for shrinking, and without 685 * scaling applied to the result. The caller does these things 686 * if necessary. This is a "raw" window selection. 687 */ 688 u32 __tcp_select_window(struct sock *sk); 689 690 void tcp_send_window_probe(struct sock *sk); 691 692 /* TCP uses 32bit jiffies to save some space. 693 * Note that this is different from tcp_time_stamp, which 694 * historically has been the same until linux-4.13. 695 */ 696 #define tcp_jiffies32 ((u32)jiffies) 697 698 /* 699 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 700 * It is no longer tied to jiffies, but to 1 ms clock. 701 * Note: double check if you want to use tcp_jiffies32 instead of this. 702 */ 703 #define TCP_TS_HZ 1000 704 705 static inline u64 tcp_clock_ns(void) 706 { 707 return local_clock(); 708 } 709 710 static inline u64 tcp_clock_us(void) 711 { 712 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 713 } 714 715 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 716 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 717 { 718 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 719 } 720 721 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 722 static inline u32 tcp_time_stamp_raw(void) 723 { 724 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 725 } 726 727 728 /* Refresh 1us clock of a TCP socket, 729 * ensuring monotically increasing values. 730 */ 731 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 732 { 733 u64 val = tcp_clock_us(); 734 735 if (val > tp->tcp_mstamp) 736 tp->tcp_mstamp = val; 737 } 738 739 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 740 { 741 return max_t(s64, t1 - t0, 0); 742 } 743 744 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 745 { 746 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 747 } 748 749 750 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 751 752 #define TCPHDR_FIN 0x01 753 #define TCPHDR_SYN 0x02 754 #define TCPHDR_RST 0x04 755 #define TCPHDR_PSH 0x08 756 #define TCPHDR_ACK 0x10 757 #define TCPHDR_URG 0x20 758 #define TCPHDR_ECE 0x40 759 #define TCPHDR_CWR 0x80 760 761 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 762 763 /* This is what the send packet queuing engine uses to pass 764 * TCP per-packet control information to the transmission code. 765 * We also store the host-order sequence numbers in here too. 766 * This is 44 bytes if IPV6 is enabled. 767 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 768 */ 769 struct tcp_skb_cb { 770 __u32 seq; /* Starting sequence number */ 771 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 772 union { 773 /* Note : tcp_tw_isn is used in input path only 774 * (isn chosen by tcp_timewait_state_process()) 775 * 776 * tcp_gso_segs/size are used in write queue only, 777 * cf tcp_skb_pcount()/tcp_skb_mss() 778 */ 779 __u32 tcp_tw_isn; 780 struct { 781 u16 tcp_gso_segs; 782 u16 tcp_gso_size; 783 }; 784 }; 785 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 786 787 __u8 sacked; /* State flags for SACK. */ 788 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 789 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 790 #define TCPCB_LOST 0x04 /* SKB is lost */ 791 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 792 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 793 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 794 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 795 TCPCB_REPAIRED) 796 797 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 798 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 799 eor:1, /* Is skb MSG_EOR marked? */ 800 has_rxtstamp:1, /* SKB has a RX timestamp */ 801 unused:5; 802 __u32 ack_seq; /* Sequence number ACK'd */ 803 union { 804 struct { 805 /* There is space for up to 24 bytes */ 806 __u32 in_flight:30,/* Bytes in flight at transmit */ 807 is_app_limited:1, /* cwnd not fully used? */ 808 unused:1; 809 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 810 __u32 delivered; 811 /* start of send pipeline phase */ 812 u64 first_tx_mstamp; 813 /* when we reached the "delivered" count */ 814 u64 delivered_mstamp; 815 } tx; /* only used for outgoing skbs */ 816 union { 817 struct inet_skb_parm h4; 818 #if IS_ENABLED(CONFIG_IPV6) 819 struct inet6_skb_parm h6; 820 #endif 821 } header; /* For incoming skbs */ 822 struct { 823 __u32 flags; 824 struct sock *sk_redir; 825 void *data_end; 826 } bpf; 827 }; 828 }; 829 830 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 831 832 833 #if IS_ENABLED(CONFIG_IPV6) 834 /* This is the variant of inet6_iif() that must be used by TCP, 835 * as TCP moves IP6CB into a different location in skb->cb[] 836 */ 837 static inline int tcp_v6_iif(const struct sk_buff *skb) 838 { 839 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 840 841 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 842 } 843 844 /* TCP_SKB_CB reference means this can not be used from early demux */ 845 static inline int tcp_v6_sdif(const struct sk_buff *skb) 846 { 847 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 848 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 849 return TCP_SKB_CB(skb)->header.h6.iif; 850 #endif 851 return 0; 852 } 853 #endif 854 855 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 856 { 857 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 858 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 859 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 860 return true; 861 #endif 862 return false; 863 } 864 865 /* TCP_SKB_CB reference means this can not be used from early demux */ 866 static inline int tcp_v4_sdif(struct sk_buff *skb) 867 { 868 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 869 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 870 return TCP_SKB_CB(skb)->header.h4.iif; 871 #endif 872 return 0; 873 } 874 875 /* Due to TSO, an SKB can be composed of multiple actual 876 * packets. To keep these tracked properly, we use this. 877 */ 878 static inline int tcp_skb_pcount(const struct sk_buff *skb) 879 { 880 return TCP_SKB_CB(skb)->tcp_gso_segs; 881 } 882 883 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 884 { 885 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 886 } 887 888 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 889 { 890 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 891 } 892 893 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 894 static inline int tcp_skb_mss(const struct sk_buff *skb) 895 { 896 return TCP_SKB_CB(skb)->tcp_gso_size; 897 } 898 899 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 900 { 901 return likely(!TCP_SKB_CB(skb)->eor); 902 } 903 904 /* Events passed to congestion control interface */ 905 enum tcp_ca_event { 906 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 907 CA_EVENT_CWND_RESTART, /* congestion window restart */ 908 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 909 CA_EVENT_LOSS, /* loss timeout */ 910 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 911 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 912 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 913 CA_EVENT_NON_DELAYED_ACK, 914 }; 915 916 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 917 enum tcp_ca_ack_event_flags { 918 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 919 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 920 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 921 }; 922 923 /* 924 * Interface for adding new TCP congestion control handlers 925 */ 926 #define TCP_CA_NAME_MAX 16 927 #define TCP_CA_MAX 128 928 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 929 930 #define TCP_CA_UNSPEC 0 931 932 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 933 #define TCP_CONG_NON_RESTRICTED 0x1 934 /* Requires ECN/ECT set on all packets */ 935 #define TCP_CONG_NEEDS_ECN 0x2 936 937 union tcp_cc_info; 938 939 struct ack_sample { 940 u32 pkts_acked; 941 s32 rtt_us; 942 u32 in_flight; 943 }; 944 945 /* A rate sample measures the number of (original/retransmitted) data 946 * packets delivered "delivered" over an interval of time "interval_us". 947 * The tcp_rate.c code fills in the rate sample, and congestion 948 * control modules that define a cong_control function to run at the end 949 * of ACK processing can optionally chose to consult this sample when 950 * setting cwnd and pacing rate. 951 * A sample is invalid if "delivered" or "interval_us" is negative. 952 */ 953 struct rate_sample { 954 u64 prior_mstamp; /* starting timestamp for interval */ 955 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 956 s32 delivered; /* number of packets delivered over interval */ 957 long interval_us; /* time for tp->delivered to incr "delivered" */ 958 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 959 int losses; /* number of packets marked lost upon ACK */ 960 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 961 u32 prior_in_flight; /* in flight before this ACK */ 962 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 963 bool is_retrans; /* is sample from retransmission? */ 964 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 965 }; 966 967 struct tcp_congestion_ops { 968 struct list_head list; 969 u32 key; 970 u32 flags; 971 972 /* initialize private data (optional) */ 973 void (*init)(struct sock *sk); 974 /* cleanup private data (optional) */ 975 void (*release)(struct sock *sk); 976 977 /* return slow start threshold (required) */ 978 u32 (*ssthresh)(struct sock *sk); 979 /* do new cwnd calculation (required) */ 980 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 981 /* call before changing ca_state (optional) */ 982 void (*set_state)(struct sock *sk, u8 new_state); 983 /* call when cwnd event occurs (optional) */ 984 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 985 /* call when ack arrives (optional) */ 986 void (*in_ack_event)(struct sock *sk, u32 flags); 987 /* new value of cwnd after loss (required) */ 988 u32 (*undo_cwnd)(struct sock *sk); 989 /* hook for packet ack accounting (optional) */ 990 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 991 /* override sysctl_tcp_min_tso_segs */ 992 u32 (*min_tso_segs)(struct sock *sk); 993 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 994 u32 (*sndbuf_expand)(struct sock *sk); 995 /* call when packets are delivered to update cwnd and pacing rate, 996 * after all the ca_state processing. (optional) 997 */ 998 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 999 /* get info for inet_diag (optional) */ 1000 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1001 union tcp_cc_info *info); 1002 1003 char name[TCP_CA_NAME_MAX]; 1004 struct module *owner; 1005 }; 1006 1007 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1008 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1009 1010 void tcp_assign_congestion_control(struct sock *sk); 1011 void tcp_init_congestion_control(struct sock *sk); 1012 void tcp_cleanup_congestion_control(struct sock *sk); 1013 int tcp_set_default_congestion_control(struct net *net, const char *name); 1014 void tcp_get_default_congestion_control(struct net *net, char *name); 1015 void tcp_get_available_congestion_control(char *buf, size_t len); 1016 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1017 int tcp_set_allowed_congestion_control(char *allowed); 1018 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1019 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1020 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1021 1022 u32 tcp_reno_ssthresh(struct sock *sk); 1023 u32 tcp_reno_undo_cwnd(struct sock *sk); 1024 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1025 extern struct tcp_congestion_ops tcp_reno; 1026 1027 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1028 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1029 #ifdef CONFIG_INET 1030 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1031 #else 1032 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1033 { 1034 return NULL; 1035 } 1036 #endif 1037 1038 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1039 { 1040 const struct inet_connection_sock *icsk = inet_csk(sk); 1041 1042 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1043 } 1044 1045 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1046 { 1047 struct inet_connection_sock *icsk = inet_csk(sk); 1048 1049 if (icsk->icsk_ca_ops->set_state) 1050 icsk->icsk_ca_ops->set_state(sk, ca_state); 1051 icsk->icsk_ca_state = ca_state; 1052 } 1053 1054 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1055 { 1056 const struct inet_connection_sock *icsk = inet_csk(sk); 1057 1058 if (icsk->icsk_ca_ops->cwnd_event) 1059 icsk->icsk_ca_ops->cwnd_event(sk, event); 1060 } 1061 1062 /* From tcp_rate.c */ 1063 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1064 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1065 struct rate_sample *rs); 1066 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1067 bool is_sack_reneg, struct rate_sample *rs); 1068 void tcp_rate_check_app_limited(struct sock *sk); 1069 1070 /* These functions determine how the current flow behaves in respect of SACK 1071 * handling. SACK is negotiated with the peer, and therefore it can vary 1072 * between different flows. 1073 * 1074 * tcp_is_sack - SACK enabled 1075 * tcp_is_reno - No SACK 1076 */ 1077 static inline int tcp_is_sack(const struct tcp_sock *tp) 1078 { 1079 return tp->rx_opt.sack_ok; 1080 } 1081 1082 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1083 { 1084 return !tcp_is_sack(tp); 1085 } 1086 1087 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1088 { 1089 return tp->sacked_out + tp->lost_out; 1090 } 1091 1092 /* This determines how many packets are "in the network" to the best 1093 * of our knowledge. In many cases it is conservative, but where 1094 * detailed information is available from the receiver (via SACK 1095 * blocks etc.) we can make more aggressive calculations. 1096 * 1097 * Use this for decisions involving congestion control, use just 1098 * tp->packets_out to determine if the send queue is empty or not. 1099 * 1100 * Read this equation as: 1101 * 1102 * "Packets sent once on transmission queue" MINUS 1103 * "Packets left network, but not honestly ACKed yet" PLUS 1104 * "Packets fast retransmitted" 1105 */ 1106 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1107 { 1108 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1109 } 1110 1111 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1112 1113 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1114 { 1115 return tp->snd_cwnd < tp->snd_ssthresh; 1116 } 1117 1118 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1119 { 1120 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1121 } 1122 1123 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1124 { 1125 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1126 (1 << inet_csk(sk)->icsk_ca_state); 1127 } 1128 1129 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1130 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1131 * ssthresh. 1132 */ 1133 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1134 { 1135 const struct tcp_sock *tp = tcp_sk(sk); 1136 1137 if (tcp_in_cwnd_reduction(sk)) 1138 return tp->snd_ssthresh; 1139 else 1140 return max(tp->snd_ssthresh, 1141 ((tp->snd_cwnd >> 1) + 1142 (tp->snd_cwnd >> 2))); 1143 } 1144 1145 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1146 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1147 1148 void tcp_enter_cwr(struct sock *sk); 1149 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1150 1151 /* The maximum number of MSS of available cwnd for which TSO defers 1152 * sending if not using sysctl_tcp_tso_win_divisor. 1153 */ 1154 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1155 { 1156 return 3; 1157 } 1158 1159 /* Returns end sequence number of the receiver's advertised window */ 1160 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1161 { 1162 return tp->snd_una + tp->snd_wnd; 1163 } 1164 1165 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1166 * flexible approach. The RFC suggests cwnd should not be raised unless 1167 * it was fully used previously. And that's exactly what we do in 1168 * congestion avoidance mode. But in slow start we allow cwnd to grow 1169 * as long as the application has used half the cwnd. 1170 * Example : 1171 * cwnd is 10 (IW10), but application sends 9 frames. 1172 * We allow cwnd to reach 18 when all frames are ACKed. 1173 * This check is safe because it's as aggressive as slow start which already 1174 * risks 100% overshoot. The advantage is that we discourage application to 1175 * either send more filler packets or data to artificially blow up the cwnd 1176 * usage, and allow application-limited process to probe bw more aggressively. 1177 */ 1178 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1179 { 1180 const struct tcp_sock *tp = tcp_sk(sk); 1181 1182 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1183 if (tcp_in_slow_start(tp)) 1184 return tp->snd_cwnd < 2 * tp->max_packets_out; 1185 1186 return tp->is_cwnd_limited; 1187 } 1188 1189 /* Something is really bad, we could not queue an additional packet, 1190 * because qdisc is full or receiver sent a 0 window. 1191 * We do not want to add fuel to the fire, or abort too early, 1192 * so make sure the timer we arm now is at least 200ms in the future, 1193 * regardless of current icsk_rto value (as it could be ~2ms) 1194 */ 1195 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1196 { 1197 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1198 } 1199 1200 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1201 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1202 unsigned long max_when) 1203 { 1204 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1205 1206 return (unsigned long)min_t(u64, when, max_when); 1207 } 1208 1209 static inline void tcp_check_probe_timer(struct sock *sk) 1210 { 1211 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1212 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1213 tcp_probe0_base(sk), TCP_RTO_MAX); 1214 } 1215 1216 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1217 { 1218 tp->snd_wl1 = seq; 1219 } 1220 1221 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1222 { 1223 tp->snd_wl1 = seq; 1224 } 1225 1226 /* 1227 * Calculate(/check) TCP checksum 1228 */ 1229 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1230 __be32 daddr, __wsum base) 1231 { 1232 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1233 } 1234 1235 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1236 { 1237 return __skb_checksum_complete(skb); 1238 } 1239 1240 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1241 { 1242 return !skb_csum_unnecessary(skb) && 1243 __tcp_checksum_complete(skb); 1244 } 1245 1246 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1247 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1248 1249 #undef STATE_TRACE 1250 1251 #ifdef STATE_TRACE 1252 static const char *statename[]={ 1253 "Unused","Established","Syn Sent","Syn Recv", 1254 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1255 "Close Wait","Last ACK","Listen","Closing" 1256 }; 1257 #endif 1258 void tcp_set_state(struct sock *sk, int state); 1259 1260 void tcp_done(struct sock *sk); 1261 1262 int tcp_abort(struct sock *sk, int err); 1263 1264 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1265 { 1266 rx_opt->dsack = 0; 1267 rx_opt->num_sacks = 0; 1268 } 1269 1270 u32 tcp_default_init_rwnd(u32 mss); 1271 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1272 1273 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1274 { 1275 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1276 struct tcp_sock *tp = tcp_sk(sk); 1277 s32 delta; 1278 1279 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1280 ca_ops->cong_control) 1281 return; 1282 delta = tcp_jiffies32 - tp->lsndtime; 1283 if (delta > inet_csk(sk)->icsk_rto) 1284 tcp_cwnd_restart(sk, delta); 1285 } 1286 1287 /* Determine a window scaling and initial window to offer. */ 1288 void tcp_select_initial_window(const struct sock *sk, int __space, 1289 __u32 mss, __u32 *rcv_wnd, 1290 __u32 *window_clamp, int wscale_ok, 1291 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1292 1293 static inline int tcp_win_from_space(const struct sock *sk, int space) 1294 { 1295 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1296 1297 return tcp_adv_win_scale <= 0 ? 1298 (space>>(-tcp_adv_win_scale)) : 1299 space - (space>>tcp_adv_win_scale); 1300 } 1301 1302 /* Note: caller must be prepared to deal with negative returns */ 1303 static inline int tcp_space(const struct sock *sk) 1304 { 1305 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1306 atomic_read(&sk->sk_rmem_alloc)); 1307 } 1308 1309 static inline int tcp_full_space(const struct sock *sk) 1310 { 1311 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1312 } 1313 1314 extern void tcp_openreq_init_rwin(struct request_sock *req, 1315 const struct sock *sk_listener, 1316 const struct dst_entry *dst); 1317 1318 void tcp_enter_memory_pressure(struct sock *sk); 1319 void tcp_leave_memory_pressure(struct sock *sk); 1320 1321 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1322 { 1323 struct net *net = sock_net((struct sock *)tp); 1324 1325 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1326 } 1327 1328 static inline int keepalive_time_when(const struct tcp_sock *tp) 1329 { 1330 struct net *net = sock_net((struct sock *)tp); 1331 1332 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1333 } 1334 1335 static inline int keepalive_probes(const struct tcp_sock *tp) 1336 { 1337 struct net *net = sock_net((struct sock *)tp); 1338 1339 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1340 } 1341 1342 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1343 { 1344 const struct inet_connection_sock *icsk = &tp->inet_conn; 1345 1346 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1347 tcp_jiffies32 - tp->rcv_tstamp); 1348 } 1349 1350 static inline int tcp_fin_time(const struct sock *sk) 1351 { 1352 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1353 const int rto = inet_csk(sk)->icsk_rto; 1354 1355 if (fin_timeout < (rto << 2) - (rto >> 1)) 1356 fin_timeout = (rto << 2) - (rto >> 1); 1357 1358 return fin_timeout; 1359 } 1360 1361 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1362 int paws_win) 1363 { 1364 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1365 return true; 1366 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1367 return true; 1368 /* 1369 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1370 * then following tcp messages have valid values. Ignore 0 value, 1371 * or else 'negative' tsval might forbid us to accept their packets. 1372 */ 1373 if (!rx_opt->ts_recent) 1374 return true; 1375 return false; 1376 } 1377 1378 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1379 int rst) 1380 { 1381 if (tcp_paws_check(rx_opt, 0)) 1382 return false; 1383 1384 /* RST segments are not recommended to carry timestamp, 1385 and, if they do, it is recommended to ignore PAWS because 1386 "their cleanup function should take precedence over timestamps." 1387 Certainly, it is mistake. It is necessary to understand the reasons 1388 of this constraint to relax it: if peer reboots, clock may go 1389 out-of-sync and half-open connections will not be reset. 1390 Actually, the problem would be not existing if all 1391 the implementations followed draft about maintaining clock 1392 via reboots. Linux-2.2 DOES NOT! 1393 1394 However, we can relax time bounds for RST segments to MSL. 1395 */ 1396 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1397 return false; 1398 return true; 1399 } 1400 1401 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1402 int mib_idx, u32 *last_oow_ack_time); 1403 1404 static inline void tcp_mib_init(struct net *net) 1405 { 1406 /* See RFC 2012 */ 1407 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1408 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1409 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1410 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1411 } 1412 1413 /* from STCP */ 1414 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1415 { 1416 tp->lost_skb_hint = NULL; 1417 } 1418 1419 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1420 { 1421 tcp_clear_retrans_hints_partial(tp); 1422 tp->retransmit_skb_hint = NULL; 1423 } 1424 1425 union tcp_md5_addr { 1426 struct in_addr a4; 1427 #if IS_ENABLED(CONFIG_IPV6) 1428 struct in6_addr a6; 1429 #endif 1430 }; 1431 1432 /* - key database */ 1433 struct tcp_md5sig_key { 1434 struct hlist_node node; 1435 u8 keylen; 1436 u8 family; /* AF_INET or AF_INET6 */ 1437 union tcp_md5_addr addr; 1438 u8 prefixlen; 1439 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1440 struct rcu_head rcu; 1441 }; 1442 1443 /* - sock block */ 1444 struct tcp_md5sig_info { 1445 struct hlist_head head; 1446 struct rcu_head rcu; 1447 }; 1448 1449 /* - pseudo header */ 1450 struct tcp4_pseudohdr { 1451 __be32 saddr; 1452 __be32 daddr; 1453 __u8 pad; 1454 __u8 protocol; 1455 __be16 len; 1456 }; 1457 1458 struct tcp6_pseudohdr { 1459 struct in6_addr saddr; 1460 struct in6_addr daddr; 1461 __be32 len; 1462 __be32 protocol; /* including padding */ 1463 }; 1464 1465 union tcp_md5sum_block { 1466 struct tcp4_pseudohdr ip4; 1467 #if IS_ENABLED(CONFIG_IPV6) 1468 struct tcp6_pseudohdr ip6; 1469 #endif 1470 }; 1471 1472 /* - pool: digest algorithm, hash description and scratch buffer */ 1473 struct tcp_md5sig_pool { 1474 struct ahash_request *md5_req; 1475 void *scratch; 1476 }; 1477 1478 /* - functions */ 1479 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1480 const struct sock *sk, const struct sk_buff *skb); 1481 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1482 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1483 gfp_t gfp); 1484 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1485 int family, u8 prefixlen); 1486 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1487 const struct sock *addr_sk); 1488 1489 #ifdef CONFIG_TCP_MD5SIG 1490 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1491 const union tcp_md5_addr *addr, 1492 int family); 1493 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1494 #else 1495 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1496 const union tcp_md5_addr *addr, 1497 int family) 1498 { 1499 return NULL; 1500 } 1501 #define tcp_twsk_md5_key(twsk) NULL 1502 #endif 1503 1504 bool tcp_alloc_md5sig_pool(void); 1505 1506 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1507 static inline void tcp_put_md5sig_pool(void) 1508 { 1509 local_bh_enable(); 1510 } 1511 1512 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1513 unsigned int header_len); 1514 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1515 const struct tcp_md5sig_key *key); 1516 1517 /* From tcp_fastopen.c */ 1518 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1519 struct tcp_fastopen_cookie *cookie); 1520 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1521 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1522 u16 try_exp); 1523 struct tcp_fastopen_request { 1524 /* Fast Open cookie. Size 0 means a cookie request */ 1525 struct tcp_fastopen_cookie cookie; 1526 struct msghdr *data; /* data in MSG_FASTOPEN */ 1527 size_t size; 1528 int copied; /* queued in tcp_connect() */ 1529 }; 1530 void tcp_free_fastopen_req(struct tcp_sock *tp); 1531 void tcp_fastopen_destroy_cipher(struct sock *sk); 1532 void tcp_fastopen_ctx_destroy(struct net *net); 1533 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1534 void *key, unsigned int len); 1535 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1536 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1537 struct request_sock *req, 1538 struct tcp_fastopen_cookie *foc, 1539 const struct dst_entry *dst); 1540 void tcp_fastopen_init_key_once(struct net *net); 1541 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1542 struct tcp_fastopen_cookie *cookie); 1543 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1544 #define TCP_FASTOPEN_KEY_LENGTH 16 1545 1546 /* Fastopen key context */ 1547 struct tcp_fastopen_context { 1548 struct crypto_cipher *tfm; 1549 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1550 struct rcu_head rcu; 1551 }; 1552 1553 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1554 void tcp_fastopen_active_disable(struct sock *sk); 1555 bool tcp_fastopen_active_should_disable(struct sock *sk); 1556 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1557 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1558 1559 /* Latencies incurred by various limits for a sender. They are 1560 * chronograph-like stats that are mutually exclusive. 1561 */ 1562 enum tcp_chrono { 1563 TCP_CHRONO_UNSPEC, 1564 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1565 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1566 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1567 __TCP_CHRONO_MAX, 1568 }; 1569 1570 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1571 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1572 1573 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1574 * the same memory storage than skb->destructor/_skb_refdst 1575 */ 1576 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1577 { 1578 skb->destructor = NULL; 1579 skb->_skb_refdst = 0UL; 1580 } 1581 1582 #define tcp_skb_tsorted_save(skb) { \ 1583 unsigned long _save = skb->_skb_refdst; \ 1584 skb->_skb_refdst = 0UL; 1585 1586 #define tcp_skb_tsorted_restore(skb) \ 1587 skb->_skb_refdst = _save; \ 1588 } 1589 1590 void tcp_write_queue_purge(struct sock *sk); 1591 1592 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1593 { 1594 return skb_rb_first(&sk->tcp_rtx_queue); 1595 } 1596 1597 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1598 { 1599 return skb_peek(&sk->sk_write_queue); 1600 } 1601 1602 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1603 { 1604 return skb_peek_tail(&sk->sk_write_queue); 1605 } 1606 1607 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1608 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1609 1610 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1611 { 1612 return skb_peek(&sk->sk_write_queue); 1613 } 1614 1615 static inline bool tcp_skb_is_last(const struct sock *sk, 1616 const struct sk_buff *skb) 1617 { 1618 return skb_queue_is_last(&sk->sk_write_queue, skb); 1619 } 1620 1621 static inline bool tcp_write_queue_empty(const struct sock *sk) 1622 { 1623 return skb_queue_empty(&sk->sk_write_queue); 1624 } 1625 1626 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1627 { 1628 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1629 } 1630 1631 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1632 { 1633 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1634 } 1635 1636 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1637 { 1638 if (tcp_write_queue_empty(sk)) 1639 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1640 } 1641 1642 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1643 { 1644 __skb_queue_tail(&sk->sk_write_queue, skb); 1645 } 1646 1647 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1648 { 1649 __tcp_add_write_queue_tail(sk, skb); 1650 1651 /* Queue it, remembering where we must start sending. */ 1652 if (sk->sk_write_queue.next == skb) 1653 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1654 } 1655 1656 /* Insert new before skb on the write queue of sk. */ 1657 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1658 struct sk_buff *skb, 1659 struct sock *sk) 1660 { 1661 __skb_queue_before(&sk->sk_write_queue, skb, new); 1662 } 1663 1664 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1665 { 1666 tcp_skb_tsorted_anchor_cleanup(skb); 1667 __skb_unlink(skb, &sk->sk_write_queue); 1668 } 1669 1670 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1671 1672 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1673 { 1674 tcp_skb_tsorted_anchor_cleanup(skb); 1675 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1676 } 1677 1678 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1679 { 1680 list_del(&skb->tcp_tsorted_anchor); 1681 tcp_rtx_queue_unlink(skb, sk); 1682 sk_wmem_free_skb(sk, skb); 1683 } 1684 1685 static inline void tcp_push_pending_frames(struct sock *sk) 1686 { 1687 if (tcp_send_head(sk)) { 1688 struct tcp_sock *tp = tcp_sk(sk); 1689 1690 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1691 } 1692 } 1693 1694 /* Start sequence of the skb just after the highest skb with SACKed 1695 * bit, valid only if sacked_out > 0 or when the caller has ensured 1696 * validity by itself. 1697 */ 1698 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1699 { 1700 if (!tp->sacked_out) 1701 return tp->snd_una; 1702 1703 if (tp->highest_sack == NULL) 1704 return tp->snd_nxt; 1705 1706 return TCP_SKB_CB(tp->highest_sack)->seq; 1707 } 1708 1709 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1710 { 1711 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1712 } 1713 1714 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1715 { 1716 return tcp_sk(sk)->highest_sack; 1717 } 1718 1719 static inline void tcp_highest_sack_reset(struct sock *sk) 1720 { 1721 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1722 } 1723 1724 /* Called when old skb is about to be deleted and replaced by new skb */ 1725 static inline void tcp_highest_sack_replace(struct sock *sk, 1726 struct sk_buff *old, 1727 struct sk_buff *new) 1728 { 1729 if (old == tcp_highest_sack(sk)) 1730 tcp_sk(sk)->highest_sack = new; 1731 } 1732 1733 /* This helper checks if socket has IP_TRANSPARENT set */ 1734 static inline bool inet_sk_transparent(const struct sock *sk) 1735 { 1736 switch (sk->sk_state) { 1737 case TCP_TIME_WAIT: 1738 return inet_twsk(sk)->tw_transparent; 1739 case TCP_NEW_SYN_RECV: 1740 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1741 } 1742 return inet_sk(sk)->transparent; 1743 } 1744 1745 /* Determines whether this is a thin stream (which may suffer from 1746 * increased latency). Used to trigger latency-reducing mechanisms. 1747 */ 1748 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1749 { 1750 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1751 } 1752 1753 /* /proc */ 1754 enum tcp_seq_states { 1755 TCP_SEQ_STATE_LISTENING, 1756 TCP_SEQ_STATE_ESTABLISHED, 1757 }; 1758 1759 int tcp_seq_open(struct inode *inode, struct file *file); 1760 1761 struct tcp_seq_afinfo { 1762 char *name; 1763 sa_family_t family; 1764 const struct file_operations *seq_fops; 1765 struct seq_operations seq_ops; 1766 }; 1767 1768 struct tcp_iter_state { 1769 struct seq_net_private p; 1770 sa_family_t family; 1771 enum tcp_seq_states state; 1772 struct sock *syn_wait_sk; 1773 int bucket, offset, sbucket, num; 1774 loff_t last_pos; 1775 }; 1776 1777 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1778 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1779 1780 extern struct request_sock_ops tcp_request_sock_ops; 1781 extern struct request_sock_ops tcp6_request_sock_ops; 1782 1783 void tcp_v4_destroy_sock(struct sock *sk); 1784 1785 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1786 netdev_features_t features); 1787 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1788 int tcp_gro_complete(struct sk_buff *skb); 1789 1790 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1791 1792 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1793 { 1794 struct net *net = sock_net((struct sock *)tp); 1795 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1796 } 1797 1798 static inline bool tcp_stream_memory_free(const struct sock *sk) 1799 { 1800 const struct tcp_sock *tp = tcp_sk(sk); 1801 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1802 1803 return notsent_bytes < tcp_notsent_lowat(tp); 1804 } 1805 1806 #ifdef CONFIG_PROC_FS 1807 int tcp4_proc_init(void); 1808 void tcp4_proc_exit(void); 1809 #endif 1810 1811 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1812 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1813 const struct tcp_request_sock_ops *af_ops, 1814 struct sock *sk, struct sk_buff *skb); 1815 1816 /* TCP af-specific functions */ 1817 struct tcp_sock_af_ops { 1818 #ifdef CONFIG_TCP_MD5SIG 1819 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1820 const struct sock *addr_sk); 1821 int (*calc_md5_hash)(char *location, 1822 const struct tcp_md5sig_key *md5, 1823 const struct sock *sk, 1824 const struct sk_buff *skb); 1825 int (*md5_parse)(struct sock *sk, 1826 int optname, 1827 char __user *optval, 1828 int optlen); 1829 #endif 1830 }; 1831 1832 struct tcp_request_sock_ops { 1833 u16 mss_clamp; 1834 #ifdef CONFIG_TCP_MD5SIG 1835 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1836 const struct sock *addr_sk); 1837 int (*calc_md5_hash) (char *location, 1838 const struct tcp_md5sig_key *md5, 1839 const struct sock *sk, 1840 const struct sk_buff *skb); 1841 #endif 1842 void (*init_req)(struct request_sock *req, 1843 const struct sock *sk_listener, 1844 struct sk_buff *skb); 1845 #ifdef CONFIG_SYN_COOKIES 1846 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1847 __u16 *mss); 1848 #endif 1849 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1850 const struct request_sock *req); 1851 u32 (*init_seq)(const struct sk_buff *skb); 1852 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1853 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1854 struct flowi *fl, struct request_sock *req, 1855 struct tcp_fastopen_cookie *foc, 1856 enum tcp_synack_type synack_type); 1857 }; 1858 1859 #ifdef CONFIG_SYN_COOKIES 1860 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1861 const struct sock *sk, struct sk_buff *skb, 1862 __u16 *mss) 1863 { 1864 tcp_synq_overflow(sk); 1865 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1866 return ops->cookie_init_seq(skb, mss); 1867 } 1868 #else 1869 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1870 const struct sock *sk, struct sk_buff *skb, 1871 __u16 *mss) 1872 { 1873 return 0; 1874 } 1875 #endif 1876 1877 int tcpv4_offload_init(void); 1878 1879 void tcp_v4_init(void); 1880 void tcp_init(void); 1881 1882 /* tcp_recovery.c */ 1883 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1884 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1885 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1886 u32 reo_wnd); 1887 extern void tcp_rack_mark_lost(struct sock *sk); 1888 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1889 u64 xmit_time); 1890 extern void tcp_rack_reo_timeout(struct sock *sk); 1891 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1892 1893 /* At how many usecs into the future should the RTO fire? */ 1894 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1895 { 1896 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1897 u32 rto = inet_csk(sk)->icsk_rto; 1898 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1899 1900 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1901 } 1902 1903 /* 1904 * Save and compile IPv4 options, return a pointer to it 1905 */ 1906 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1907 struct sk_buff *skb) 1908 { 1909 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1910 struct ip_options_rcu *dopt = NULL; 1911 1912 if (opt->optlen) { 1913 int opt_size = sizeof(*dopt) + opt->optlen; 1914 1915 dopt = kmalloc(opt_size, GFP_ATOMIC); 1916 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1917 kfree(dopt); 1918 dopt = NULL; 1919 } 1920 } 1921 return dopt; 1922 } 1923 1924 /* locally generated TCP pure ACKs have skb->truesize == 2 1925 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1926 * This is much faster than dissecting the packet to find out. 1927 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1928 */ 1929 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1930 { 1931 return skb->truesize == 2; 1932 } 1933 1934 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1935 { 1936 skb->truesize = 2; 1937 } 1938 1939 static inline int tcp_inq(struct sock *sk) 1940 { 1941 struct tcp_sock *tp = tcp_sk(sk); 1942 int answ; 1943 1944 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1945 answ = 0; 1946 } else if (sock_flag(sk, SOCK_URGINLINE) || 1947 !tp->urg_data || 1948 before(tp->urg_seq, tp->copied_seq) || 1949 !before(tp->urg_seq, tp->rcv_nxt)) { 1950 1951 answ = tp->rcv_nxt - tp->copied_seq; 1952 1953 /* Subtract 1, if FIN was received */ 1954 if (answ && sock_flag(sk, SOCK_DONE)) 1955 answ--; 1956 } else { 1957 answ = tp->urg_seq - tp->copied_seq; 1958 } 1959 1960 return answ; 1961 } 1962 1963 int tcp_peek_len(struct socket *sock); 1964 1965 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1966 { 1967 u16 segs_in; 1968 1969 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1970 tp->segs_in += segs_in; 1971 if (skb->len > tcp_hdrlen(skb)) 1972 tp->data_segs_in += segs_in; 1973 } 1974 1975 /* 1976 * TCP listen path runs lockless. 1977 * We forced "struct sock" to be const qualified to make sure 1978 * we don't modify one of its field by mistake. 1979 * Here, we increment sk_drops which is an atomic_t, so we can safely 1980 * make sock writable again. 1981 */ 1982 static inline void tcp_listendrop(const struct sock *sk) 1983 { 1984 atomic_inc(&((struct sock *)sk)->sk_drops); 1985 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1986 } 1987 1988 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 1989 1990 /* 1991 * Interface for adding Upper Level Protocols over TCP 1992 */ 1993 1994 #define TCP_ULP_NAME_MAX 16 1995 #define TCP_ULP_MAX 128 1996 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 1997 1998 enum { 1999 TCP_ULP_TLS, 2000 TCP_ULP_BPF, 2001 }; 2002 2003 struct tcp_ulp_ops { 2004 struct list_head list; 2005 2006 /* initialize ulp */ 2007 int (*init)(struct sock *sk); 2008 /* cleanup ulp */ 2009 void (*release)(struct sock *sk); 2010 2011 int uid; 2012 char name[TCP_ULP_NAME_MAX]; 2013 bool user_visible; 2014 struct module *owner; 2015 }; 2016 int tcp_register_ulp(struct tcp_ulp_ops *type); 2017 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2018 int tcp_set_ulp(struct sock *sk, const char *name); 2019 int tcp_set_ulp_id(struct sock *sk, const int ulp); 2020 void tcp_get_available_ulp(char *buf, size_t len); 2021 void tcp_cleanup_ulp(struct sock *sk); 2022 2023 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2024 * is < 0, then the BPF op failed (for example if the loaded BPF 2025 * program does not support the chosen operation or there is no BPF 2026 * program loaded). 2027 */ 2028 #ifdef CONFIG_BPF 2029 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2030 { 2031 struct bpf_sock_ops_kern sock_ops; 2032 int ret; 2033 2034 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2035 if (sk_fullsock(sk)) { 2036 sock_ops.is_fullsock = 1; 2037 sock_owned_by_me(sk); 2038 } 2039 2040 sock_ops.sk = sk; 2041 sock_ops.op = op; 2042 if (nargs > 0) 2043 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2044 2045 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2046 if (ret == 0) 2047 ret = sock_ops.reply; 2048 else 2049 ret = -1; 2050 return ret; 2051 } 2052 2053 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2054 { 2055 u32 args[2] = {arg1, arg2}; 2056 2057 return tcp_call_bpf(sk, op, 2, args); 2058 } 2059 2060 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2061 u32 arg3) 2062 { 2063 u32 args[3] = {arg1, arg2, arg3}; 2064 2065 return tcp_call_bpf(sk, op, 3, args); 2066 } 2067 2068 #else 2069 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2070 { 2071 return -EPERM; 2072 } 2073 2074 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2075 { 2076 return -EPERM; 2077 } 2078 2079 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2080 u32 arg3) 2081 { 2082 return -EPERM; 2083 } 2084 2085 #endif 2086 2087 static inline u32 tcp_timeout_init(struct sock *sk) 2088 { 2089 int timeout; 2090 2091 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2092 2093 if (timeout <= 0) 2094 timeout = TCP_TIMEOUT_INIT; 2095 return timeout; 2096 } 2097 2098 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2099 { 2100 int rwnd; 2101 2102 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2103 2104 if (rwnd < 0) 2105 rwnd = 0; 2106 return rwnd; 2107 } 2108 2109 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2110 { 2111 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2112 } 2113 2114 #if IS_ENABLED(CONFIG_SMC) 2115 extern struct static_key_false tcp_have_smc; 2116 #endif 2117 2118 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2119 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2120 void (*cad)(struct sock *sk, u32 ack_seq)); 2121 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2122 2123 #endif 2124 2125 #endif /* _TCP_H */ 2126