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 49 #include <linux/bpf.h> 50 #include <linux/filter.h> 51 #include <linux/bpf-cgroup.h> 52 53 extern struct inet_hashinfo tcp_hashinfo; 54 55 extern struct percpu_counter tcp_orphan_count; 56 void tcp_time_wait(struct sock *sk, int state, int timeo); 57 58 #define MAX_TCP_HEADER (128 + MAX_HEADER) 59 #define MAX_TCP_OPTION_SPACE 40 60 61 /* 62 * Never offer a window over 32767 without using window scaling. Some 63 * poor stacks do signed 16bit maths! 64 */ 65 #define MAX_TCP_WINDOW 32767U 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The least MTU to use for probing */ 71 #define TCP_BASE_MSS 1024 72 73 /* probing interval, default to 10 minutes as per RFC4821 */ 74 #define TCP_PROBE_INTERVAL 600 75 76 /* Specify interval when tcp mtu probing will stop */ 77 #define TCP_PROBE_THRESHOLD 8 78 79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 80 #define TCP_FASTRETRANS_THRESH 3 81 82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 83 #define TCP_MAX_QUICKACKS 16U 84 85 /* Maximal number of window scale according to RFC1323 */ 86 #define TCP_MAX_WSCALE 14U 87 88 /* urg_data states */ 89 #define TCP_URG_VALID 0x0100 90 #define TCP_URG_NOTYET 0x0200 91 #define TCP_URG_READ 0x0400 92 93 #define TCP_RETR1 3 /* 94 * This is how many retries it does before it 95 * tries to figure out if the gateway is 96 * down. Minimal RFC value is 3; it corresponds 97 * to ~3sec-8min depending on RTO. 98 */ 99 100 #define TCP_RETR2 15 /* 101 * This should take at least 102 * 90 minutes to time out. 103 * RFC1122 says that the limit is 100 sec. 104 * 15 is ~13-30min depending on RTO. 105 */ 106 107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 108 * when active opening a connection. 109 * RFC1122 says the minimum retry MUST 110 * be at least 180secs. Nevertheless 111 * this value is corresponding to 112 * 63secs of retransmission with the 113 * current initial RTO. 114 */ 115 116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 117 * when passive opening a connection. 118 * This is corresponding to 31secs of 119 * retransmission with the current 120 * initial RTO. 121 */ 122 123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 124 * state, about 60 seconds */ 125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 126 /* BSD style FIN_WAIT2 deadlock breaker. 127 * It used to be 3min, new value is 60sec, 128 * to combine FIN-WAIT-2 timeout with 129 * TIME-WAIT timer. 130 */ 131 132 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 133 #if HZ >= 100 134 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 135 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 136 #else 137 #define TCP_DELACK_MIN 4U 138 #define TCP_ATO_MIN 4U 139 #endif 140 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 141 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 142 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 143 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 144 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 145 * used as a fallback RTO for the 146 * initial data transmission if no 147 * valid RTT sample has been acquired, 148 * most likely due to retrans in 3WHS. 149 */ 150 151 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 152 * for local resources. 153 */ 154 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 155 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 156 #define TCP_KEEPALIVE_INTVL (75*HZ) 157 158 #define MAX_TCP_KEEPIDLE 32767 159 #define MAX_TCP_KEEPINTVL 32767 160 #define MAX_TCP_KEEPCNT 127 161 #define MAX_TCP_SYNCNT 127 162 163 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 164 165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 167 * after this time. It should be equal 168 * (or greater than) TCP_TIMEWAIT_LEN 169 * to provide reliability equal to one 170 * provided by timewait state. 171 */ 172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 173 * timestamps. It must be less than 174 * minimal timewait lifetime. 175 */ 176 /* 177 * TCP option 178 */ 179 180 #define TCPOPT_NOP 1 /* Padding */ 181 #define TCPOPT_EOL 0 /* End of options */ 182 #define TCPOPT_MSS 2 /* Segment size negotiating */ 183 #define TCPOPT_WINDOW 3 /* Window scaling */ 184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 185 #define TCPOPT_SACK 5 /* SACK Block */ 186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 189 #define TCPOPT_EXP 254 /* Experimental */ 190 /* Magic number to be after the option value for sharing TCP 191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 192 */ 193 #define TCPOPT_FASTOPEN_MAGIC 0xF989 194 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 195 196 /* 197 * TCP option lengths 198 */ 199 200 #define TCPOLEN_MSS 4 201 #define TCPOLEN_WINDOW 3 202 #define TCPOLEN_SACK_PERM 2 203 #define TCPOLEN_TIMESTAMP 10 204 #define TCPOLEN_MD5SIG 18 205 #define TCPOLEN_FASTOPEN_BASE 2 206 #define TCPOLEN_EXP_FASTOPEN_BASE 4 207 #define TCPOLEN_EXP_SMC_BASE 6 208 209 /* But this is what stacks really send out. */ 210 #define TCPOLEN_TSTAMP_ALIGNED 12 211 #define TCPOLEN_WSCALE_ALIGNED 4 212 #define TCPOLEN_SACKPERM_ALIGNED 4 213 #define TCPOLEN_SACK_BASE 2 214 #define TCPOLEN_SACK_BASE_ALIGNED 4 215 #define TCPOLEN_SACK_PERBLOCK 8 216 #define TCPOLEN_MD5SIG_ALIGNED 20 217 #define TCPOLEN_MSS_ALIGNED 4 218 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 219 220 /* Flags in tp->nonagle */ 221 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 222 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 223 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 224 225 /* TCP thin-stream limits */ 226 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 227 228 /* TCP initial congestion window as per rfc6928 */ 229 #define TCP_INIT_CWND 10 230 231 /* Bit Flags for sysctl_tcp_fastopen */ 232 #define TFO_CLIENT_ENABLE 1 233 #define TFO_SERVER_ENABLE 2 234 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 235 236 /* Accept SYN data w/o any cookie option */ 237 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 238 239 /* Force enable TFO on all listeners, i.e., not requiring the 240 * TCP_FASTOPEN socket option. 241 */ 242 #define TFO_SERVER_WO_SOCKOPT1 0x400 243 244 245 /* sysctl variables for tcp */ 246 extern int sysctl_tcp_max_orphans; 247 extern long sysctl_tcp_mem[3]; 248 extern int sysctl_tcp_wmem[3]; 249 extern int sysctl_tcp_rmem[3]; 250 251 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 252 253 extern int sysctl_tcp_limit_output_bytes; 254 extern int sysctl_tcp_challenge_ack_limit; 255 extern int sysctl_tcp_min_tso_segs; 256 extern int sysctl_tcp_min_rtt_wlen; 257 extern int sysctl_tcp_autocorking; 258 extern int sysctl_tcp_invalid_ratelimit; 259 extern int sysctl_tcp_pacing_ss_ratio; 260 extern int sysctl_tcp_pacing_ca_ratio; 261 262 extern atomic_long_t tcp_memory_allocated; 263 extern struct percpu_counter tcp_sockets_allocated; 264 extern unsigned long tcp_memory_pressure; 265 266 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 267 static inline bool tcp_under_memory_pressure(const struct sock *sk) 268 { 269 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 270 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 271 return true; 272 273 return tcp_memory_pressure; 274 } 275 /* 276 * The next routines deal with comparing 32 bit unsigned ints 277 * and worry about wraparound (automatic with unsigned arithmetic). 278 */ 279 280 static inline bool before(__u32 seq1, __u32 seq2) 281 { 282 return (__s32)(seq1-seq2) < 0; 283 } 284 #define after(seq2, seq1) before(seq1, seq2) 285 286 /* is s2<=s1<=s3 ? */ 287 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 288 { 289 return seq3 - seq2 >= seq1 - seq2; 290 } 291 292 static inline bool tcp_out_of_memory(struct sock *sk) 293 { 294 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 295 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 296 return true; 297 return false; 298 } 299 300 void sk_forced_mem_schedule(struct sock *sk, int size); 301 302 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 303 { 304 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 305 int orphans = percpu_counter_read_positive(ocp); 306 307 if (orphans << shift > sysctl_tcp_max_orphans) { 308 orphans = percpu_counter_sum_positive(ocp); 309 if (orphans << shift > sysctl_tcp_max_orphans) 310 return true; 311 } 312 return false; 313 } 314 315 bool tcp_check_oom(struct sock *sk, int shift); 316 317 318 extern struct proto tcp_prot; 319 320 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 321 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 322 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 323 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 324 325 void tcp_tasklet_init(void); 326 327 void tcp_v4_err(struct sk_buff *skb, u32); 328 329 void tcp_shutdown(struct sock *sk, int how); 330 331 int tcp_v4_early_demux(struct sk_buff *skb); 332 int tcp_v4_rcv(struct sk_buff *skb); 333 334 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 335 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 336 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 337 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 338 int flags); 339 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 340 size_t size, int flags); 341 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 342 size_t size, int flags); 343 void tcp_release_cb(struct sock *sk); 344 void tcp_wfree(struct sk_buff *skb); 345 void tcp_write_timer_handler(struct sock *sk); 346 void tcp_delack_timer_handler(struct sock *sk); 347 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 348 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 349 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 350 const struct tcphdr *th); 351 void tcp_rcv_space_adjust(struct sock *sk); 352 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 353 void tcp_twsk_destructor(struct sock *sk); 354 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 355 struct pipe_inode_info *pipe, size_t len, 356 unsigned int flags); 357 358 static inline void tcp_dec_quickack_mode(struct sock *sk, 359 const unsigned int pkts) 360 { 361 struct inet_connection_sock *icsk = inet_csk(sk); 362 363 if (icsk->icsk_ack.quick) { 364 if (pkts >= icsk->icsk_ack.quick) { 365 icsk->icsk_ack.quick = 0; 366 /* Leaving quickack mode we deflate ATO. */ 367 icsk->icsk_ack.ato = TCP_ATO_MIN; 368 } else 369 icsk->icsk_ack.quick -= pkts; 370 } 371 } 372 373 #define TCP_ECN_OK 1 374 #define TCP_ECN_QUEUE_CWR 2 375 #define TCP_ECN_DEMAND_CWR 4 376 #define TCP_ECN_SEEN 8 377 378 enum tcp_tw_status { 379 TCP_TW_SUCCESS = 0, 380 TCP_TW_RST = 1, 381 TCP_TW_ACK = 2, 382 TCP_TW_SYN = 3 383 }; 384 385 386 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 387 struct sk_buff *skb, 388 const struct tcphdr *th); 389 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 390 struct request_sock *req, bool fastopen); 391 int tcp_child_process(struct sock *parent, struct sock *child, 392 struct sk_buff *skb); 393 void tcp_enter_loss(struct sock *sk); 394 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 395 void tcp_clear_retrans(struct tcp_sock *tp); 396 void tcp_update_metrics(struct sock *sk); 397 void tcp_init_metrics(struct sock *sk); 398 void tcp_metrics_init(void); 399 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 400 void tcp_disable_fack(struct tcp_sock *tp); 401 void tcp_close(struct sock *sk, long timeout); 402 void tcp_init_sock(struct sock *sk); 403 void tcp_init_transfer(struct sock *sk, int bpf_op); 404 unsigned int tcp_poll(struct file *file, struct socket *sock, 405 struct poll_table_struct *wait); 406 int tcp_getsockopt(struct sock *sk, int level, int optname, 407 char __user *optval, int __user *optlen); 408 int tcp_setsockopt(struct sock *sk, int level, int optname, 409 char __user *optval, unsigned int optlen); 410 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 411 char __user *optval, int __user *optlen); 412 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 413 char __user *optval, unsigned int optlen); 414 void tcp_set_keepalive(struct sock *sk, int val); 415 void tcp_syn_ack_timeout(const struct request_sock *req); 416 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 417 int flags, int *addr_len); 418 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 419 struct tcp_options_received *opt_rx, 420 int estab, struct tcp_fastopen_cookie *foc); 421 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 422 423 /* 424 * TCP v4 functions exported for the inet6 API 425 */ 426 427 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 428 void tcp_v4_mtu_reduced(struct sock *sk); 429 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 430 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 431 struct sock *tcp_create_openreq_child(const struct sock *sk, 432 struct request_sock *req, 433 struct sk_buff *skb); 434 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 435 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 436 struct request_sock *req, 437 struct dst_entry *dst, 438 struct request_sock *req_unhash, 439 bool *own_req); 440 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 441 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 442 int tcp_connect(struct sock *sk); 443 enum tcp_synack_type { 444 TCP_SYNACK_NORMAL, 445 TCP_SYNACK_FASTOPEN, 446 TCP_SYNACK_COOKIE, 447 }; 448 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 449 struct request_sock *req, 450 struct tcp_fastopen_cookie *foc, 451 enum tcp_synack_type synack_type); 452 int tcp_disconnect(struct sock *sk, int flags); 453 454 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 455 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 456 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 457 458 /* From syncookies.c */ 459 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 460 struct request_sock *req, 461 struct dst_entry *dst, u32 tsoff); 462 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 463 u32 cookie); 464 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 465 #ifdef CONFIG_SYN_COOKIES 466 467 /* Syncookies use a monotonic timer which increments every 60 seconds. 468 * This counter is used both as a hash input and partially encoded into 469 * the cookie value. A cookie is only validated further if the delta 470 * between the current counter value and the encoded one is less than this, 471 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 472 * the counter advances immediately after a cookie is generated). 473 */ 474 #define MAX_SYNCOOKIE_AGE 2 475 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 476 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 477 478 /* syncookies: remember time of last synqueue overflow 479 * But do not dirty this field too often (once per second is enough) 480 * It is racy as we do not hold a lock, but race is very minor. 481 */ 482 static inline void tcp_synq_overflow(const struct sock *sk) 483 { 484 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 485 unsigned long now = jiffies; 486 487 if (time_after(now, last_overflow + HZ)) 488 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 489 } 490 491 /* syncookies: no recent synqueue overflow on this listening socket? */ 492 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 493 { 494 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 495 496 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 497 } 498 499 static inline u32 tcp_cookie_time(void) 500 { 501 u64 val = get_jiffies_64(); 502 503 do_div(val, TCP_SYNCOOKIE_PERIOD); 504 return val; 505 } 506 507 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 508 u16 *mssp); 509 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 510 u64 cookie_init_timestamp(struct request_sock *req); 511 bool cookie_timestamp_decode(const struct net *net, 512 struct tcp_options_received *opt); 513 bool cookie_ecn_ok(const struct tcp_options_received *opt, 514 const struct net *net, const struct dst_entry *dst); 515 516 /* From net/ipv6/syncookies.c */ 517 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 518 u32 cookie); 519 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 520 521 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 522 const struct tcphdr *th, u16 *mssp); 523 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 524 #endif 525 /* tcp_output.c */ 526 527 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 528 int min_tso_segs); 529 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 530 int nonagle); 531 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 532 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 533 void tcp_retransmit_timer(struct sock *sk); 534 void tcp_xmit_retransmit_queue(struct sock *); 535 void tcp_simple_retransmit(struct sock *); 536 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 537 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 538 enum tcp_queue { 539 TCP_FRAG_IN_WRITE_QUEUE, 540 TCP_FRAG_IN_RTX_QUEUE, 541 }; 542 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 543 struct sk_buff *skb, u32 len, 544 unsigned int mss_now, gfp_t gfp); 545 546 void tcp_send_probe0(struct sock *); 547 void tcp_send_partial(struct sock *); 548 int tcp_write_wakeup(struct sock *, int mib); 549 void tcp_send_fin(struct sock *sk); 550 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 551 int tcp_send_synack(struct sock *); 552 void tcp_push_one(struct sock *, unsigned int mss_now); 553 void tcp_send_ack(struct sock *sk); 554 void tcp_send_delayed_ack(struct sock *sk); 555 void tcp_send_loss_probe(struct sock *sk); 556 bool tcp_schedule_loss_probe(struct sock *sk); 557 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 558 const struct sk_buff *next_skb); 559 560 /* tcp_input.c */ 561 void tcp_rearm_rto(struct sock *sk); 562 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 563 void tcp_reset(struct sock *sk); 564 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 565 void tcp_fin(struct sock *sk); 566 567 /* tcp_timer.c */ 568 void tcp_init_xmit_timers(struct sock *); 569 static inline void tcp_clear_xmit_timers(struct sock *sk) 570 { 571 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 572 inet_csk_clear_xmit_timers(sk); 573 } 574 575 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 576 unsigned int tcp_current_mss(struct sock *sk); 577 578 /* Bound MSS / TSO packet size with the half of the window */ 579 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 580 { 581 int cutoff; 582 583 /* When peer uses tiny windows, there is no use in packetizing 584 * to sub-MSS pieces for the sake of SWS or making sure there 585 * are enough packets in the pipe for fast recovery. 586 * 587 * On the other hand, for extremely large MSS devices, handling 588 * smaller than MSS windows in this way does make sense. 589 */ 590 if (tp->max_window > TCP_MSS_DEFAULT) 591 cutoff = (tp->max_window >> 1); 592 else 593 cutoff = tp->max_window; 594 595 if (cutoff && pktsize > cutoff) 596 return max_t(int, cutoff, 68U - tp->tcp_header_len); 597 else 598 return pktsize; 599 } 600 601 /* tcp.c */ 602 void tcp_get_info(struct sock *, struct tcp_info *); 603 604 /* Read 'sendfile()'-style from a TCP socket */ 605 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 606 sk_read_actor_t recv_actor); 607 608 void tcp_initialize_rcv_mss(struct sock *sk); 609 610 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 611 int tcp_mss_to_mtu(struct sock *sk, int mss); 612 void tcp_mtup_init(struct sock *sk); 613 void tcp_init_buffer_space(struct sock *sk); 614 615 static inline void tcp_bound_rto(const struct sock *sk) 616 { 617 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 618 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 619 } 620 621 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 622 { 623 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 624 } 625 626 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 627 { 628 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 629 ntohl(TCP_FLAG_ACK) | 630 snd_wnd); 631 } 632 633 static inline void tcp_fast_path_on(struct tcp_sock *tp) 634 { 635 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 636 } 637 638 static inline void tcp_fast_path_check(struct sock *sk) 639 { 640 struct tcp_sock *tp = tcp_sk(sk); 641 642 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 643 tp->rcv_wnd && 644 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 645 !tp->urg_data) 646 tcp_fast_path_on(tp); 647 } 648 649 /* Compute the actual rto_min value */ 650 static inline u32 tcp_rto_min(struct sock *sk) 651 { 652 const struct dst_entry *dst = __sk_dst_get(sk); 653 u32 rto_min = TCP_RTO_MIN; 654 655 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 656 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 657 return rto_min; 658 } 659 660 static inline u32 tcp_rto_min_us(struct sock *sk) 661 { 662 return jiffies_to_usecs(tcp_rto_min(sk)); 663 } 664 665 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 666 { 667 return dst_metric_locked(dst, RTAX_CC_ALGO); 668 } 669 670 /* Minimum RTT in usec. ~0 means not available. */ 671 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 672 { 673 return minmax_get(&tp->rtt_min); 674 } 675 676 /* Compute the actual receive window we are currently advertising. 677 * Rcv_nxt can be after the window if our peer push more data 678 * than the offered window. 679 */ 680 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 681 { 682 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 683 684 if (win < 0) 685 win = 0; 686 return (u32) win; 687 } 688 689 /* Choose a new window, without checks for shrinking, and without 690 * scaling applied to the result. The caller does these things 691 * if necessary. This is a "raw" window selection. 692 */ 693 u32 __tcp_select_window(struct sock *sk); 694 695 void tcp_send_window_probe(struct sock *sk); 696 697 /* TCP uses 32bit jiffies to save some space. 698 * Note that this is different from tcp_time_stamp, which 699 * historically has been the same until linux-4.13. 700 */ 701 #define tcp_jiffies32 ((u32)jiffies) 702 703 /* 704 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 705 * It is no longer tied to jiffies, but to 1 ms clock. 706 * Note: double check if you want to use tcp_jiffies32 instead of this. 707 */ 708 #define TCP_TS_HZ 1000 709 710 static inline u64 tcp_clock_ns(void) 711 { 712 return local_clock(); 713 } 714 715 static inline u64 tcp_clock_us(void) 716 { 717 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 718 } 719 720 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 721 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 722 { 723 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 724 } 725 726 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 727 static inline u32 tcp_time_stamp_raw(void) 728 { 729 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 730 } 731 732 733 /* Refresh 1us clock of a TCP socket, 734 * ensuring monotically increasing values. 735 */ 736 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 737 { 738 u64 val = tcp_clock_us(); 739 740 if (val > tp->tcp_mstamp) 741 tp->tcp_mstamp = val; 742 } 743 744 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 745 { 746 return max_t(s64, t1 - t0, 0); 747 } 748 749 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 750 { 751 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 752 } 753 754 755 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 756 757 #define TCPHDR_FIN 0x01 758 #define TCPHDR_SYN 0x02 759 #define TCPHDR_RST 0x04 760 #define TCPHDR_PSH 0x08 761 #define TCPHDR_ACK 0x10 762 #define TCPHDR_URG 0x20 763 #define TCPHDR_ECE 0x40 764 #define TCPHDR_CWR 0x80 765 766 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 767 768 /* This is what the send packet queuing engine uses to pass 769 * TCP per-packet control information to the transmission code. 770 * We also store the host-order sequence numbers in here too. 771 * This is 44 bytes if IPV6 is enabled. 772 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 773 */ 774 struct tcp_skb_cb { 775 __u32 seq; /* Starting sequence number */ 776 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 777 union { 778 /* Note : tcp_tw_isn is used in input path only 779 * (isn chosen by tcp_timewait_state_process()) 780 * 781 * tcp_gso_segs/size are used in write queue only, 782 * cf tcp_skb_pcount()/tcp_skb_mss() 783 */ 784 __u32 tcp_tw_isn; 785 struct { 786 u16 tcp_gso_segs; 787 u16 tcp_gso_size; 788 }; 789 }; 790 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 791 792 __u8 sacked; /* State flags for SACK/FACK. */ 793 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 794 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 795 #define TCPCB_LOST 0x04 /* SKB is lost */ 796 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 797 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 798 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 799 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 800 TCPCB_REPAIRED) 801 802 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 803 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 804 eor:1, /* Is skb MSG_EOR marked? */ 805 has_rxtstamp:1, /* SKB has a RX timestamp */ 806 unused:5; 807 __u32 ack_seq; /* Sequence number ACK'd */ 808 union { 809 struct { 810 /* There is space for up to 24 bytes */ 811 __u32 in_flight:30,/* Bytes in flight at transmit */ 812 is_app_limited:1, /* cwnd not fully used? */ 813 unused:1; 814 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 815 __u32 delivered; 816 /* start of send pipeline phase */ 817 u64 first_tx_mstamp; 818 /* when we reached the "delivered" count */ 819 u64 delivered_mstamp; 820 } tx; /* only used for outgoing skbs */ 821 union { 822 struct inet_skb_parm h4; 823 #if IS_ENABLED(CONFIG_IPV6) 824 struct inet6_skb_parm h6; 825 #endif 826 } header; /* For incoming skbs */ 827 struct { 828 __u32 key; 829 __u32 flags; 830 struct bpf_map *map; 831 } bpf; 832 }; 833 }; 834 835 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 836 837 838 #if IS_ENABLED(CONFIG_IPV6) 839 /* This is the variant of inet6_iif() that must be used by TCP, 840 * as TCP moves IP6CB into a different location in skb->cb[] 841 */ 842 static inline int tcp_v6_iif(const struct sk_buff *skb) 843 { 844 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 845 846 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 847 } 848 849 /* TCP_SKB_CB reference means this can not be used from early demux */ 850 static inline int tcp_v6_sdif(const struct sk_buff *skb) 851 { 852 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 853 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 854 return TCP_SKB_CB(skb)->header.h6.iif; 855 #endif 856 return 0; 857 } 858 #endif 859 860 /* TCP_SKB_CB reference means this can not be used from early demux */ 861 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 862 { 863 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 864 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 865 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 866 return true; 867 #endif 868 return false; 869 } 870 871 /* TCP_SKB_CB reference means this can not be used from early demux */ 872 static inline int tcp_v4_sdif(struct sk_buff *skb) 873 { 874 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 875 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 876 return TCP_SKB_CB(skb)->header.h4.iif; 877 #endif 878 return 0; 879 } 880 881 /* Due to TSO, an SKB can be composed of multiple actual 882 * packets. To keep these tracked properly, we use this. 883 */ 884 static inline int tcp_skb_pcount(const struct sk_buff *skb) 885 { 886 return TCP_SKB_CB(skb)->tcp_gso_segs; 887 } 888 889 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 890 { 891 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 892 } 893 894 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 895 { 896 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 897 } 898 899 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 900 static inline int tcp_skb_mss(const struct sk_buff *skb) 901 { 902 return TCP_SKB_CB(skb)->tcp_gso_size; 903 } 904 905 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 906 { 907 return likely(!TCP_SKB_CB(skb)->eor); 908 } 909 910 /* Events passed to congestion control interface */ 911 enum tcp_ca_event { 912 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 913 CA_EVENT_CWND_RESTART, /* congestion window restart */ 914 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 915 CA_EVENT_LOSS, /* loss timeout */ 916 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 917 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 918 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 919 CA_EVENT_NON_DELAYED_ACK, 920 }; 921 922 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 923 enum tcp_ca_ack_event_flags { 924 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 925 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 926 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 927 }; 928 929 /* 930 * Interface for adding new TCP congestion control handlers 931 */ 932 #define TCP_CA_NAME_MAX 16 933 #define TCP_CA_MAX 128 934 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 935 936 #define TCP_CA_UNSPEC 0 937 938 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 939 #define TCP_CONG_NON_RESTRICTED 0x1 940 /* Requires ECN/ECT set on all packets */ 941 #define TCP_CONG_NEEDS_ECN 0x2 942 943 union tcp_cc_info; 944 945 struct ack_sample { 946 u32 pkts_acked; 947 s32 rtt_us; 948 u32 in_flight; 949 }; 950 951 /* A rate sample measures the number of (original/retransmitted) data 952 * packets delivered "delivered" over an interval of time "interval_us". 953 * The tcp_rate.c code fills in the rate sample, and congestion 954 * control modules that define a cong_control function to run at the end 955 * of ACK processing can optionally chose to consult this sample when 956 * setting cwnd and pacing rate. 957 * A sample is invalid if "delivered" or "interval_us" is negative. 958 */ 959 struct rate_sample { 960 u64 prior_mstamp; /* starting timestamp for interval */ 961 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 962 s32 delivered; /* number of packets delivered over interval */ 963 long interval_us; /* time for tp->delivered to incr "delivered" */ 964 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 965 int losses; /* number of packets marked lost upon ACK */ 966 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 967 u32 prior_in_flight; /* in flight before this ACK */ 968 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 969 bool is_retrans; /* is sample from retransmission? */ 970 }; 971 972 struct tcp_congestion_ops { 973 struct list_head list; 974 u32 key; 975 u32 flags; 976 977 /* initialize private data (optional) */ 978 void (*init)(struct sock *sk); 979 /* cleanup private data (optional) */ 980 void (*release)(struct sock *sk); 981 982 /* return slow start threshold (required) */ 983 u32 (*ssthresh)(struct sock *sk); 984 /* do new cwnd calculation (required) */ 985 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 986 /* call before changing ca_state (optional) */ 987 void (*set_state)(struct sock *sk, u8 new_state); 988 /* call when cwnd event occurs (optional) */ 989 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 990 /* call when ack arrives (optional) */ 991 void (*in_ack_event)(struct sock *sk, u32 flags); 992 /* new value of cwnd after loss (required) */ 993 u32 (*undo_cwnd)(struct sock *sk); 994 /* hook for packet ack accounting (optional) */ 995 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 996 /* suggest number of segments for each skb to transmit (optional) */ 997 u32 (*tso_segs_goal)(struct sock *sk); 998 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 999 u32 (*sndbuf_expand)(struct sock *sk); 1000 /* call when packets are delivered to update cwnd and pacing rate, 1001 * after all the ca_state processing. (optional) 1002 */ 1003 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1004 /* get info for inet_diag (optional) */ 1005 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1006 union tcp_cc_info *info); 1007 1008 char name[TCP_CA_NAME_MAX]; 1009 struct module *owner; 1010 }; 1011 1012 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1013 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1014 1015 void tcp_assign_congestion_control(struct sock *sk); 1016 void tcp_init_congestion_control(struct sock *sk); 1017 void tcp_cleanup_congestion_control(struct sock *sk); 1018 int tcp_set_default_congestion_control(const char *name); 1019 void tcp_get_default_congestion_control(char *name); 1020 void tcp_get_available_congestion_control(char *buf, size_t len); 1021 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1022 int tcp_set_allowed_congestion_control(char *allowed); 1023 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1024 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1025 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1026 1027 u32 tcp_reno_ssthresh(struct sock *sk); 1028 u32 tcp_reno_undo_cwnd(struct sock *sk); 1029 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1030 extern struct tcp_congestion_ops tcp_reno; 1031 1032 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1033 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1034 #ifdef CONFIG_INET 1035 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1036 #else 1037 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1038 { 1039 return NULL; 1040 } 1041 #endif 1042 1043 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1044 { 1045 const struct inet_connection_sock *icsk = inet_csk(sk); 1046 1047 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1048 } 1049 1050 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1051 { 1052 struct inet_connection_sock *icsk = inet_csk(sk); 1053 1054 if (icsk->icsk_ca_ops->set_state) 1055 icsk->icsk_ca_ops->set_state(sk, ca_state); 1056 icsk->icsk_ca_state = ca_state; 1057 } 1058 1059 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1060 { 1061 const struct inet_connection_sock *icsk = inet_csk(sk); 1062 1063 if (icsk->icsk_ca_ops->cwnd_event) 1064 icsk->icsk_ca_ops->cwnd_event(sk, event); 1065 } 1066 1067 /* From tcp_rate.c */ 1068 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1069 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1070 struct rate_sample *rs); 1071 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1072 struct rate_sample *rs); 1073 void tcp_rate_check_app_limited(struct sock *sk); 1074 1075 /* These functions determine how the current flow behaves in respect of SACK 1076 * handling. SACK is negotiated with the peer, and therefore it can vary 1077 * between different flows. 1078 * 1079 * tcp_is_sack - SACK enabled 1080 * tcp_is_reno - No SACK 1081 * tcp_is_fack - FACK enabled, implies SACK enabled 1082 */ 1083 static inline int tcp_is_sack(const struct tcp_sock *tp) 1084 { 1085 return tp->rx_opt.sack_ok; 1086 } 1087 1088 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1089 { 1090 return !tcp_is_sack(tp); 1091 } 1092 1093 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1094 { 1095 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1096 } 1097 1098 static inline void tcp_enable_fack(struct tcp_sock *tp) 1099 { 1100 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1101 } 1102 1103 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1104 { 1105 return tp->sacked_out + tp->lost_out; 1106 } 1107 1108 /* This determines how many packets are "in the network" to the best 1109 * of our knowledge. In many cases it is conservative, but where 1110 * detailed information is available from the receiver (via SACK 1111 * blocks etc.) we can make more aggressive calculations. 1112 * 1113 * Use this for decisions involving congestion control, use just 1114 * tp->packets_out to determine if the send queue is empty or not. 1115 * 1116 * Read this equation as: 1117 * 1118 * "Packets sent once on transmission queue" MINUS 1119 * "Packets left network, but not honestly ACKed yet" PLUS 1120 * "Packets fast retransmitted" 1121 */ 1122 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1123 { 1124 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1125 } 1126 1127 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1128 1129 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1130 { 1131 return tp->snd_cwnd < tp->snd_ssthresh; 1132 } 1133 1134 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1135 { 1136 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1137 } 1138 1139 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1140 { 1141 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1142 (1 << inet_csk(sk)->icsk_ca_state); 1143 } 1144 1145 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1146 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1147 * ssthresh. 1148 */ 1149 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1150 { 1151 const struct tcp_sock *tp = tcp_sk(sk); 1152 1153 if (tcp_in_cwnd_reduction(sk)) 1154 return tp->snd_ssthresh; 1155 else 1156 return max(tp->snd_ssthresh, 1157 ((tp->snd_cwnd >> 1) + 1158 (tp->snd_cwnd >> 2))); 1159 } 1160 1161 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1162 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1163 1164 void tcp_enter_cwr(struct sock *sk); 1165 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1166 1167 /* The maximum number of MSS of available cwnd for which TSO defers 1168 * sending if not using sysctl_tcp_tso_win_divisor. 1169 */ 1170 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1171 { 1172 return 3; 1173 } 1174 1175 /* Returns end sequence number of the receiver's advertised window */ 1176 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1177 { 1178 return tp->snd_una + tp->snd_wnd; 1179 } 1180 1181 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1182 * flexible approach. The RFC suggests cwnd should not be raised unless 1183 * it was fully used previously. And that's exactly what we do in 1184 * congestion avoidance mode. But in slow start we allow cwnd to grow 1185 * as long as the application has used half the cwnd. 1186 * Example : 1187 * cwnd is 10 (IW10), but application sends 9 frames. 1188 * We allow cwnd to reach 18 when all frames are ACKed. 1189 * This check is safe because it's as aggressive as slow start which already 1190 * risks 100% overshoot. The advantage is that we discourage application to 1191 * either send more filler packets or data to artificially blow up the cwnd 1192 * usage, and allow application-limited process to probe bw more aggressively. 1193 */ 1194 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1195 { 1196 const struct tcp_sock *tp = tcp_sk(sk); 1197 1198 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1199 if (tcp_in_slow_start(tp)) 1200 return tp->snd_cwnd < 2 * tp->max_packets_out; 1201 1202 return tp->is_cwnd_limited; 1203 } 1204 1205 /* Something is really bad, we could not queue an additional packet, 1206 * because qdisc is full or receiver sent a 0 window. 1207 * We do not want to add fuel to the fire, or abort too early, 1208 * so make sure the timer we arm now is at least 200ms in the future, 1209 * regardless of current icsk_rto value (as it could be ~2ms) 1210 */ 1211 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1212 { 1213 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1214 } 1215 1216 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1217 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1218 unsigned long max_when) 1219 { 1220 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1221 1222 return (unsigned long)min_t(u64, when, max_when); 1223 } 1224 1225 static inline void tcp_check_probe_timer(struct sock *sk) 1226 { 1227 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1228 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1229 tcp_probe0_base(sk), TCP_RTO_MAX); 1230 } 1231 1232 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1233 { 1234 tp->snd_wl1 = seq; 1235 } 1236 1237 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1238 { 1239 tp->snd_wl1 = seq; 1240 } 1241 1242 /* 1243 * Calculate(/check) TCP checksum 1244 */ 1245 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1246 __be32 daddr, __wsum base) 1247 { 1248 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1249 } 1250 1251 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1252 { 1253 return __skb_checksum_complete(skb); 1254 } 1255 1256 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1257 { 1258 return !skb_csum_unnecessary(skb) && 1259 __tcp_checksum_complete(skb); 1260 } 1261 1262 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1263 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1264 1265 #undef STATE_TRACE 1266 1267 #ifdef STATE_TRACE 1268 static const char *statename[]={ 1269 "Unused","Established","Syn Sent","Syn Recv", 1270 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1271 "Close Wait","Last ACK","Listen","Closing" 1272 }; 1273 #endif 1274 void tcp_set_state(struct sock *sk, int state); 1275 1276 void tcp_done(struct sock *sk); 1277 1278 int tcp_abort(struct sock *sk, int err); 1279 1280 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1281 { 1282 rx_opt->dsack = 0; 1283 rx_opt->num_sacks = 0; 1284 } 1285 1286 u32 tcp_default_init_rwnd(u32 mss); 1287 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1288 1289 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1290 { 1291 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1292 struct tcp_sock *tp = tcp_sk(sk); 1293 s32 delta; 1294 1295 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1296 ca_ops->cong_control) 1297 return; 1298 delta = tcp_jiffies32 - tp->lsndtime; 1299 if (delta > inet_csk(sk)->icsk_rto) 1300 tcp_cwnd_restart(sk, delta); 1301 } 1302 1303 /* Determine a window scaling and initial window to offer. */ 1304 void tcp_select_initial_window(const struct sock *sk, int __space, 1305 __u32 mss, __u32 *rcv_wnd, 1306 __u32 *window_clamp, int wscale_ok, 1307 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1308 1309 static inline int tcp_win_from_space(const struct sock *sk, int space) 1310 { 1311 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1312 1313 return tcp_adv_win_scale <= 0 ? 1314 (space>>(-tcp_adv_win_scale)) : 1315 space - (space>>tcp_adv_win_scale); 1316 } 1317 1318 /* Note: caller must be prepared to deal with negative returns */ 1319 static inline int tcp_space(const struct sock *sk) 1320 { 1321 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1322 atomic_read(&sk->sk_rmem_alloc)); 1323 } 1324 1325 static inline int tcp_full_space(const struct sock *sk) 1326 { 1327 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1328 } 1329 1330 extern void tcp_openreq_init_rwin(struct request_sock *req, 1331 const struct sock *sk_listener, 1332 const struct dst_entry *dst); 1333 1334 void tcp_enter_memory_pressure(struct sock *sk); 1335 void tcp_leave_memory_pressure(struct sock *sk); 1336 1337 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1338 { 1339 struct net *net = sock_net((struct sock *)tp); 1340 1341 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1342 } 1343 1344 static inline int keepalive_time_when(const struct tcp_sock *tp) 1345 { 1346 struct net *net = sock_net((struct sock *)tp); 1347 1348 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1349 } 1350 1351 static inline int keepalive_probes(const struct tcp_sock *tp) 1352 { 1353 struct net *net = sock_net((struct sock *)tp); 1354 1355 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1356 } 1357 1358 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1359 { 1360 const struct inet_connection_sock *icsk = &tp->inet_conn; 1361 1362 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1363 tcp_jiffies32 - tp->rcv_tstamp); 1364 } 1365 1366 static inline int tcp_fin_time(const struct sock *sk) 1367 { 1368 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1369 const int rto = inet_csk(sk)->icsk_rto; 1370 1371 if (fin_timeout < (rto << 2) - (rto >> 1)) 1372 fin_timeout = (rto << 2) - (rto >> 1); 1373 1374 return fin_timeout; 1375 } 1376 1377 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1378 int paws_win) 1379 { 1380 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1381 return true; 1382 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1383 return true; 1384 /* 1385 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1386 * then following tcp messages have valid values. Ignore 0 value, 1387 * or else 'negative' tsval might forbid us to accept their packets. 1388 */ 1389 if (!rx_opt->ts_recent) 1390 return true; 1391 return false; 1392 } 1393 1394 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1395 int rst) 1396 { 1397 if (tcp_paws_check(rx_opt, 0)) 1398 return false; 1399 1400 /* RST segments are not recommended to carry timestamp, 1401 and, if they do, it is recommended to ignore PAWS because 1402 "their cleanup function should take precedence over timestamps." 1403 Certainly, it is mistake. It is necessary to understand the reasons 1404 of this constraint to relax it: if peer reboots, clock may go 1405 out-of-sync and half-open connections will not be reset. 1406 Actually, the problem would be not existing if all 1407 the implementations followed draft about maintaining clock 1408 via reboots. Linux-2.2 DOES NOT! 1409 1410 However, we can relax time bounds for RST segments to MSL. 1411 */ 1412 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1413 return false; 1414 return true; 1415 } 1416 1417 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1418 int mib_idx, u32 *last_oow_ack_time); 1419 1420 static inline void tcp_mib_init(struct net *net) 1421 { 1422 /* See RFC 2012 */ 1423 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1424 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1425 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1426 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1427 } 1428 1429 /* from STCP */ 1430 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1431 { 1432 tp->lost_skb_hint = NULL; 1433 } 1434 1435 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1436 { 1437 tcp_clear_retrans_hints_partial(tp); 1438 tp->retransmit_skb_hint = NULL; 1439 } 1440 1441 union tcp_md5_addr { 1442 struct in_addr a4; 1443 #if IS_ENABLED(CONFIG_IPV6) 1444 struct in6_addr a6; 1445 #endif 1446 }; 1447 1448 /* - key database */ 1449 struct tcp_md5sig_key { 1450 struct hlist_node node; 1451 u8 keylen; 1452 u8 family; /* AF_INET or AF_INET6 */ 1453 union tcp_md5_addr addr; 1454 u8 prefixlen; 1455 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1456 struct rcu_head rcu; 1457 }; 1458 1459 /* - sock block */ 1460 struct tcp_md5sig_info { 1461 struct hlist_head head; 1462 struct rcu_head rcu; 1463 }; 1464 1465 /* - pseudo header */ 1466 struct tcp4_pseudohdr { 1467 __be32 saddr; 1468 __be32 daddr; 1469 __u8 pad; 1470 __u8 protocol; 1471 __be16 len; 1472 }; 1473 1474 struct tcp6_pseudohdr { 1475 struct in6_addr saddr; 1476 struct in6_addr daddr; 1477 __be32 len; 1478 __be32 protocol; /* including padding */ 1479 }; 1480 1481 union tcp_md5sum_block { 1482 struct tcp4_pseudohdr ip4; 1483 #if IS_ENABLED(CONFIG_IPV6) 1484 struct tcp6_pseudohdr ip6; 1485 #endif 1486 }; 1487 1488 /* - pool: digest algorithm, hash description and scratch buffer */ 1489 struct tcp_md5sig_pool { 1490 struct ahash_request *md5_req; 1491 void *scratch; 1492 }; 1493 1494 /* - functions */ 1495 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1496 const struct sock *sk, const struct sk_buff *skb); 1497 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1498 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1499 gfp_t gfp); 1500 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1501 int family, u8 prefixlen); 1502 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1503 const struct sock *addr_sk); 1504 1505 #ifdef CONFIG_TCP_MD5SIG 1506 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1507 const union tcp_md5_addr *addr, 1508 int family); 1509 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1510 #else 1511 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1512 const union tcp_md5_addr *addr, 1513 int family) 1514 { 1515 return NULL; 1516 } 1517 #define tcp_twsk_md5_key(twsk) NULL 1518 #endif 1519 1520 bool tcp_alloc_md5sig_pool(void); 1521 1522 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1523 static inline void tcp_put_md5sig_pool(void) 1524 { 1525 local_bh_enable(); 1526 } 1527 1528 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1529 unsigned int header_len); 1530 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1531 const struct tcp_md5sig_key *key); 1532 1533 /* From tcp_fastopen.c */ 1534 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1535 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1536 unsigned long *last_syn_loss); 1537 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1538 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1539 u16 try_exp); 1540 struct tcp_fastopen_request { 1541 /* Fast Open cookie. Size 0 means a cookie request */ 1542 struct tcp_fastopen_cookie cookie; 1543 struct msghdr *data; /* data in MSG_FASTOPEN */ 1544 size_t size; 1545 int copied; /* queued in tcp_connect() */ 1546 }; 1547 void tcp_free_fastopen_req(struct tcp_sock *tp); 1548 void tcp_fastopen_destroy_cipher(struct sock *sk); 1549 void tcp_fastopen_ctx_destroy(struct net *net); 1550 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1551 void *key, unsigned int len); 1552 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1553 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1554 struct request_sock *req, 1555 struct tcp_fastopen_cookie *foc, 1556 const struct dst_entry *dst); 1557 void tcp_fastopen_init_key_once(struct net *net); 1558 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1559 struct tcp_fastopen_cookie *cookie); 1560 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1561 #define TCP_FASTOPEN_KEY_LENGTH 16 1562 1563 /* Fastopen key context */ 1564 struct tcp_fastopen_context { 1565 struct crypto_cipher *tfm; 1566 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1567 struct rcu_head rcu; 1568 }; 1569 1570 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1571 void tcp_fastopen_active_disable(struct sock *sk); 1572 bool tcp_fastopen_active_should_disable(struct sock *sk); 1573 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1574 void tcp_fastopen_active_timeout_reset(void); 1575 1576 /* Latencies incurred by various limits for a sender. They are 1577 * chronograph-like stats that are mutually exclusive. 1578 */ 1579 enum tcp_chrono { 1580 TCP_CHRONO_UNSPEC, 1581 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1582 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1583 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1584 __TCP_CHRONO_MAX, 1585 }; 1586 1587 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1588 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1589 1590 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1591 * the same memory storage than skb->destructor/_skb_refdst 1592 */ 1593 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1594 { 1595 skb->destructor = NULL; 1596 skb->_skb_refdst = 0UL; 1597 } 1598 1599 #define tcp_skb_tsorted_save(skb) { \ 1600 unsigned long _save = skb->_skb_refdst; \ 1601 skb->_skb_refdst = 0UL; 1602 1603 #define tcp_skb_tsorted_restore(skb) \ 1604 skb->_skb_refdst = _save; \ 1605 } 1606 1607 void tcp_write_queue_purge(struct sock *sk); 1608 1609 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1610 { 1611 return skb_rb_first(&sk->tcp_rtx_queue); 1612 } 1613 1614 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1615 { 1616 return skb_peek(&sk->sk_write_queue); 1617 } 1618 1619 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1620 { 1621 return skb_peek_tail(&sk->sk_write_queue); 1622 } 1623 1624 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1625 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1626 1627 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1628 { 1629 return skb_peek(&sk->sk_write_queue); 1630 } 1631 1632 static inline bool tcp_skb_is_last(const struct sock *sk, 1633 const struct sk_buff *skb) 1634 { 1635 return skb_queue_is_last(&sk->sk_write_queue, skb); 1636 } 1637 1638 static inline bool tcp_write_queue_empty(const struct sock *sk) 1639 { 1640 return skb_queue_empty(&sk->sk_write_queue); 1641 } 1642 1643 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1644 { 1645 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1646 } 1647 1648 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1649 { 1650 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1651 } 1652 1653 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1654 { 1655 if (tcp_write_queue_empty(sk)) 1656 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1657 1658 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1659 tcp_sk(sk)->highest_sack = NULL; 1660 } 1661 1662 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1663 { 1664 __skb_queue_tail(&sk->sk_write_queue, skb); 1665 } 1666 1667 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1668 { 1669 __tcp_add_write_queue_tail(sk, skb); 1670 1671 /* Queue it, remembering where we must start sending. */ 1672 if (sk->sk_write_queue.next == skb) { 1673 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1674 1675 if (tcp_sk(sk)->highest_sack == NULL) 1676 tcp_sk(sk)->highest_sack = skb; 1677 } 1678 } 1679 1680 /* Insert new before skb on the write queue of sk. */ 1681 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1682 struct sk_buff *skb, 1683 struct sock *sk) 1684 { 1685 __skb_queue_before(&sk->sk_write_queue, skb, new); 1686 } 1687 1688 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1689 { 1690 tcp_skb_tsorted_anchor_cleanup(skb); 1691 __skb_unlink(skb, &sk->sk_write_queue); 1692 } 1693 1694 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1695 1696 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1697 { 1698 tcp_skb_tsorted_anchor_cleanup(skb); 1699 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1700 } 1701 1702 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1703 { 1704 list_del(&skb->tcp_tsorted_anchor); 1705 tcp_rtx_queue_unlink(skb, sk); 1706 sk_wmem_free_skb(sk, skb); 1707 } 1708 1709 static inline void tcp_push_pending_frames(struct sock *sk) 1710 { 1711 if (tcp_send_head(sk)) { 1712 struct tcp_sock *tp = tcp_sk(sk); 1713 1714 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1715 } 1716 } 1717 1718 /* Start sequence of the skb just after the highest skb with SACKed 1719 * bit, valid only if sacked_out > 0 or when the caller has ensured 1720 * validity by itself. 1721 */ 1722 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1723 { 1724 if (!tp->sacked_out) 1725 return tp->snd_una; 1726 1727 if (tp->highest_sack == NULL) 1728 return tp->snd_nxt; 1729 1730 return TCP_SKB_CB(tp->highest_sack)->seq; 1731 } 1732 1733 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1734 { 1735 struct sk_buff *next = skb_rb_next(skb); 1736 1737 tcp_sk(sk)->highest_sack = next ?: tcp_send_head(sk); 1738 } 1739 1740 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1741 { 1742 return tcp_sk(sk)->highest_sack; 1743 } 1744 1745 static inline void tcp_highest_sack_reset(struct sock *sk) 1746 { 1747 struct sk_buff *skb = tcp_rtx_queue_head(sk); 1748 1749 tcp_sk(sk)->highest_sack = skb ?: tcp_send_head(sk); 1750 } 1751 1752 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1753 static inline void tcp_highest_sack_combine(struct sock *sk, 1754 struct sk_buff *old, 1755 struct sk_buff *new) 1756 { 1757 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1758 tcp_sk(sk)->highest_sack = new; 1759 } 1760 1761 /* This helper checks if socket has IP_TRANSPARENT set */ 1762 static inline bool inet_sk_transparent(const struct sock *sk) 1763 { 1764 switch (sk->sk_state) { 1765 case TCP_TIME_WAIT: 1766 return inet_twsk(sk)->tw_transparent; 1767 case TCP_NEW_SYN_RECV: 1768 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1769 } 1770 return inet_sk(sk)->transparent; 1771 } 1772 1773 /* Determines whether this is a thin stream (which may suffer from 1774 * increased latency). Used to trigger latency-reducing mechanisms. 1775 */ 1776 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1777 { 1778 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1779 } 1780 1781 /* /proc */ 1782 enum tcp_seq_states { 1783 TCP_SEQ_STATE_LISTENING, 1784 TCP_SEQ_STATE_ESTABLISHED, 1785 }; 1786 1787 int tcp_seq_open(struct inode *inode, struct file *file); 1788 1789 struct tcp_seq_afinfo { 1790 char *name; 1791 sa_family_t family; 1792 const struct file_operations *seq_fops; 1793 struct seq_operations seq_ops; 1794 }; 1795 1796 struct tcp_iter_state { 1797 struct seq_net_private p; 1798 sa_family_t family; 1799 enum tcp_seq_states state; 1800 struct sock *syn_wait_sk; 1801 int bucket, offset, sbucket, num; 1802 loff_t last_pos; 1803 }; 1804 1805 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1806 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1807 1808 extern struct request_sock_ops tcp_request_sock_ops; 1809 extern struct request_sock_ops tcp6_request_sock_ops; 1810 1811 void tcp_v4_destroy_sock(struct sock *sk); 1812 1813 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1814 netdev_features_t features); 1815 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1816 int tcp_gro_complete(struct sk_buff *skb); 1817 1818 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1819 1820 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1821 { 1822 struct net *net = sock_net((struct sock *)tp); 1823 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1824 } 1825 1826 static inline bool tcp_stream_memory_free(const struct sock *sk) 1827 { 1828 const struct tcp_sock *tp = tcp_sk(sk); 1829 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1830 1831 return notsent_bytes < tcp_notsent_lowat(tp); 1832 } 1833 1834 #ifdef CONFIG_PROC_FS 1835 int tcp4_proc_init(void); 1836 void tcp4_proc_exit(void); 1837 #endif 1838 1839 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1840 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1841 const struct tcp_request_sock_ops *af_ops, 1842 struct sock *sk, struct sk_buff *skb); 1843 1844 /* TCP af-specific functions */ 1845 struct tcp_sock_af_ops { 1846 #ifdef CONFIG_TCP_MD5SIG 1847 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1848 const struct sock *addr_sk); 1849 int (*calc_md5_hash)(char *location, 1850 const struct tcp_md5sig_key *md5, 1851 const struct sock *sk, 1852 const struct sk_buff *skb); 1853 int (*md5_parse)(struct sock *sk, 1854 int optname, 1855 char __user *optval, 1856 int optlen); 1857 #endif 1858 }; 1859 1860 struct tcp_request_sock_ops { 1861 u16 mss_clamp; 1862 #ifdef CONFIG_TCP_MD5SIG 1863 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1864 const struct sock *addr_sk); 1865 int (*calc_md5_hash) (char *location, 1866 const struct tcp_md5sig_key *md5, 1867 const struct sock *sk, 1868 const struct sk_buff *skb); 1869 #endif 1870 void (*init_req)(struct request_sock *req, 1871 const struct sock *sk_listener, 1872 struct sk_buff *skb); 1873 #ifdef CONFIG_SYN_COOKIES 1874 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1875 __u16 *mss); 1876 #endif 1877 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1878 const struct request_sock *req); 1879 u32 (*init_seq)(const struct sk_buff *skb); 1880 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1881 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1882 struct flowi *fl, struct request_sock *req, 1883 struct tcp_fastopen_cookie *foc, 1884 enum tcp_synack_type synack_type); 1885 }; 1886 1887 #ifdef CONFIG_SYN_COOKIES 1888 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1889 const struct sock *sk, struct sk_buff *skb, 1890 __u16 *mss) 1891 { 1892 tcp_synq_overflow(sk); 1893 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1894 return ops->cookie_init_seq(skb, mss); 1895 } 1896 #else 1897 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1898 const struct sock *sk, struct sk_buff *skb, 1899 __u16 *mss) 1900 { 1901 return 0; 1902 } 1903 #endif 1904 1905 int tcpv4_offload_init(void); 1906 1907 void tcp_v4_init(void); 1908 void tcp_init(void); 1909 1910 /* tcp_recovery.c */ 1911 extern void tcp_rack_mark_lost(struct sock *sk); 1912 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1913 u64 xmit_time); 1914 extern void tcp_rack_reo_timeout(struct sock *sk); 1915 1916 /* At how many usecs into the future should the RTO fire? */ 1917 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1918 { 1919 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1920 u32 rto = inet_csk(sk)->icsk_rto; 1921 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1922 1923 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1924 } 1925 1926 /* 1927 * Save and compile IPv4 options, return a pointer to it 1928 */ 1929 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1930 struct sk_buff *skb) 1931 { 1932 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1933 struct ip_options_rcu *dopt = NULL; 1934 1935 if (opt->optlen) { 1936 int opt_size = sizeof(*dopt) + opt->optlen; 1937 1938 dopt = kmalloc(opt_size, GFP_ATOMIC); 1939 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1940 kfree(dopt); 1941 dopt = NULL; 1942 } 1943 } 1944 return dopt; 1945 } 1946 1947 /* locally generated TCP pure ACKs have skb->truesize == 2 1948 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1949 * This is much faster than dissecting the packet to find out. 1950 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1951 */ 1952 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1953 { 1954 return skb->truesize == 2; 1955 } 1956 1957 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1958 { 1959 skb->truesize = 2; 1960 } 1961 1962 static inline int tcp_inq(struct sock *sk) 1963 { 1964 struct tcp_sock *tp = tcp_sk(sk); 1965 int answ; 1966 1967 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1968 answ = 0; 1969 } else if (sock_flag(sk, SOCK_URGINLINE) || 1970 !tp->urg_data || 1971 before(tp->urg_seq, tp->copied_seq) || 1972 !before(tp->urg_seq, tp->rcv_nxt)) { 1973 1974 answ = tp->rcv_nxt - tp->copied_seq; 1975 1976 /* Subtract 1, if FIN was received */ 1977 if (answ && sock_flag(sk, SOCK_DONE)) 1978 answ--; 1979 } else { 1980 answ = tp->urg_seq - tp->copied_seq; 1981 } 1982 1983 return answ; 1984 } 1985 1986 int tcp_peek_len(struct socket *sock); 1987 1988 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1989 { 1990 u16 segs_in; 1991 1992 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1993 tp->segs_in += segs_in; 1994 if (skb->len > tcp_hdrlen(skb)) 1995 tp->data_segs_in += segs_in; 1996 } 1997 1998 /* 1999 * TCP listen path runs lockless. 2000 * We forced "struct sock" to be const qualified to make sure 2001 * we don't modify one of its field by mistake. 2002 * Here, we increment sk_drops which is an atomic_t, so we can safely 2003 * make sock writable again. 2004 */ 2005 static inline void tcp_listendrop(const struct sock *sk) 2006 { 2007 atomic_inc(&((struct sock *)sk)->sk_drops); 2008 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2009 } 2010 2011 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2012 2013 /* 2014 * Interface for adding Upper Level Protocols over TCP 2015 */ 2016 2017 #define TCP_ULP_NAME_MAX 16 2018 #define TCP_ULP_MAX 128 2019 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2020 2021 struct tcp_ulp_ops { 2022 struct list_head list; 2023 2024 /* initialize ulp */ 2025 int (*init)(struct sock *sk); 2026 /* cleanup ulp */ 2027 void (*release)(struct sock *sk); 2028 2029 char name[TCP_ULP_NAME_MAX]; 2030 struct module *owner; 2031 }; 2032 int tcp_register_ulp(struct tcp_ulp_ops *type); 2033 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2034 int tcp_set_ulp(struct sock *sk, const char *name); 2035 void tcp_get_available_ulp(char *buf, size_t len); 2036 void tcp_cleanup_ulp(struct sock *sk); 2037 2038 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2039 * is < 0, then the BPF op failed (for example if the loaded BPF 2040 * program does not support the chosen operation or there is no BPF 2041 * program loaded). 2042 */ 2043 #ifdef CONFIG_BPF 2044 static inline int tcp_call_bpf(struct sock *sk, int op) 2045 { 2046 struct bpf_sock_ops_kern sock_ops; 2047 int ret; 2048 2049 if (sk_fullsock(sk)) 2050 sock_owned_by_me(sk); 2051 2052 memset(&sock_ops, 0, sizeof(sock_ops)); 2053 sock_ops.sk = sk; 2054 sock_ops.op = op; 2055 2056 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2057 if (ret == 0) 2058 ret = sock_ops.reply; 2059 else 2060 ret = -1; 2061 return ret; 2062 } 2063 #else 2064 static inline int tcp_call_bpf(struct sock *sk, int op) 2065 { 2066 return -EPERM; 2067 } 2068 #endif 2069 2070 static inline u32 tcp_timeout_init(struct sock *sk) 2071 { 2072 int timeout; 2073 2074 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2075 2076 if (timeout <= 0) 2077 timeout = TCP_TIMEOUT_INIT; 2078 return timeout; 2079 } 2080 2081 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2082 { 2083 int rwnd; 2084 2085 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2086 2087 if (rwnd < 0) 2088 rwnd = 0; 2089 return rwnd; 2090 } 2091 2092 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2093 { 2094 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2095 } 2096 2097 #if IS_ENABLED(CONFIG_SMC) 2098 extern struct static_key_false tcp_have_smc; 2099 #endif 2100 #endif /* _TCP_H */ 2101