1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP module. 8 * 9 * Version: @(#)tcp.h 1.0.5 05/23/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 */ 14 #ifndef _TCP_H 15 #define _TCP_H 16 17 #define FASTRETRANS_DEBUG 1 18 19 #include <linux/list.h> 20 #include <linux/tcp.h> 21 #include <linux/bug.h> 22 #include <linux/slab.h> 23 #include <linux/cache.h> 24 #include <linux/percpu.h> 25 #include <linux/skbuff.h> 26 #include <linux/cryptohash.h> 27 #include <linux/kref.h> 28 #include <linux/ktime.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/inet_ecn.h> 41 #include <net/dst.h> 42 43 #include <linux/seq_file.h> 44 #include <linux/memcontrol.h> 45 #include <linux/bpf-cgroup.h> 46 #include <linux/siphash.h> 47 48 extern struct inet_hashinfo tcp_hashinfo; 49 50 extern struct percpu_counter tcp_orphan_count; 51 void tcp_time_wait(struct sock *sk, int state, int timeo); 52 53 #define MAX_TCP_HEADER (128 + MAX_HEADER) 54 #define MAX_TCP_OPTION_SPACE 40 55 #define TCP_MIN_SND_MSS 48 56 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 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 initial 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 READ_ONCE(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 int tcp_v4_err(struct sk_buff *skb, u32); 316 317 void tcp_shutdown(struct sock *sk, int how); 318 319 int tcp_v4_early_demux(struct sk_buff *skb); 320 int tcp_v4_rcv(struct sk_buff *skb); 321 322 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 323 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 324 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 325 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 326 int flags); 327 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 328 size_t size, int flags); 329 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 330 size_t size, int flags); 331 void tcp_release_cb(struct sock *sk); 332 void tcp_wfree(struct sk_buff *skb); 333 void tcp_write_timer_handler(struct sock *sk); 334 void tcp_delack_timer_handler(struct sock *sk); 335 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 336 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 337 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 338 void tcp_rcv_space_adjust(struct sock *sk); 339 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 340 void tcp_twsk_destructor(struct sock *sk); 341 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 342 struct pipe_inode_info *pipe, size_t len, 343 unsigned int flags); 344 345 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 346 static inline void tcp_dec_quickack_mode(struct sock *sk, 347 const unsigned int pkts) 348 { 349 struct inet_connection_sock *icsk = inet_csk(sk); 350 351 if (icsk->icsk_ack.quick) { 352 if (pkts >= icsk->icsk_ack.quick) { 353 icsk->icsk_ack.quick = 0; 354 /* Leaving quickack mode we deflate ATO. */ 355 icsk->icsk_ack.ato = TCP_ATO_MIN; 356 } else 357 icsk->icsk_ack.quick -= pkts; 358 } 359 } 360 361 #define TCP_ECN_OK 1 362 #define TCP_ECN_QUEUE_CWR 2 363 #define TCP_ECN_DEMAND_CWR 4 364 #define TCP_ECN_SEEN 8 365 366 enum tcp_tw_status { 367 TCP_TW_SUCCESS = 0, 368 TCP_TW_RST = 1, 369 TCP_TW_ACK = 2, 370 TCP_TW_SYN = 3 371 }; 372 373 374 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 375 struct sk_buff *skb, 376 const struct tcphdr *th); 377 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 378 struct request_sock *req, bool fastopen, 379 bool *lost_race); 380 int tcp_child_process(struct sock *parent, struct sock *child, 381 struct sk_buff *skb); 382 void tcp_enter_loss(struct sock *sk); 383 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 384 void tcp_clear_retrans(struct tcp_sock *tp); 385 void tcp_update_metrics(struct sock *sk); 386 void tcp_init_metrics(struct sock *sk); 387 void tcp_metrics_init(void); 388 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 389 void tcp_close(struct sock *sk, long timeout); 390 void tcp_init_sock(struct sock *sk); 391 void tcp_init_transfer(struct sock *sk, int bpf_op); 392 __poll_t tcp_poll(struct file *file, struct socket *sock, 393 struct poll_table_struct *wait); 394 int tcp_getsockopt(struct sock *sk, int level, int optname, 395 char __user *optval, int __user *optlen); 396 int tcp_setsockopt(struct sock *sk, int level, int optname, 397 char __user *optval, unsigned int optlen); 398 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 399 char __user *optval, int __user *optlen); 400 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 401 char __user *optval, unsigned int optlen); 402 void tcp_set_keepalive(struct sock *sk, int val); 403 void tcp_syn_ack_timeout(const struct request_sock *req); 404 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 405 int flags, int *addr_len); 406 int tcp_set_rcvlowat(struct sock *sk, int val); 407 void tcp_data_ready(struct sock *sk); 408 #ifdef CONFIG_MMU 409 int tcp_mmap(struct file *file, struct socket *sock, 410 struct vm_area_struct *vma); 411 #endif 412 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 413 struct tcp_options_received *opt_rx, 414 int estab, struct tcp_fastopen_cookie *foc); 415 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 416 417 /* 418 * BPF SKB-less helpers 419 */ 420 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 421 struct tcphdr *th, u32 *cookie); 422 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 423 struct tcphdr *th, u32 *cookie); 424 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 425 const struct tcp_request_sock_ops *af_ops, 426 struct sock *sk, struct tcphdr *th); 427 /* 428 * TCP v4 functions exported for the inet6 API 429 */ 430 431 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 432 void tcp_v4_mtu_reduced(struct sock *sk); 433 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 434 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 435 struct sock *tcp_create_openreq_child(const struct sock *sk, 436 struct request_sock *req, 437 struct sk_buff *skb); 438 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 439 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 440 struct request_sock *req, 441 struct dst_entry *dst, 442 struct request_sock *req_unhash, 443 bool *own_req); 444 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 445 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 446 int tcp_connect(struct sock *sk); 447 enum tcp_synack_type { 448 TCP_SYNACK_NORMAL, 449 TCP_SYNACK_FASTOPEN, 450 TCP_SYNACK_COOKIE, 451 }; 452 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 453 struct request_sock *req, 454 struct tcp_fastopen_cookie *foc, 455 enum tcp_synack_type synack_type); 456 int tcp_disconnect(struct sock *sk, int flags); 457 458 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 459 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 460 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 461 462 /* From syncookies.c */ 463 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 464 struct request_sock *req, 465 struct dst_entry *dst, u32 tsoff); 466 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 467 u32 cookie); 468 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 469 #ifdef CONFIG_SYN_COOKIES 470 471 /* Syncookies use a monotonic timer which increments every 60 seconds. 472 * This counter is used both as a hash input and partially encoded into 473 * the cookie value. A cookie is only validated further if the delta 474 * between the current counter value and the encoded one is less than this, 475 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 476 * the counter advances immediately after a cookie is generated). 477 */ 478 #define MAX_SYNCOOKIE_AGE 2 479 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 480 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 481 482 /* syncookies: remember time of last synqueue overflow 483 * But do not dirty this field too often (once per second is enough) 484 * It is racy as we do not hold a lock, but race is very minor. 485 */ 486 static inline void tcp_synq_overflow(const struct sock *sk) 487 { 488 unsigned int last_overflow; 489 unsigned int now = jiffies; 490 491 if (sk->sk_reuseport) { 492 struct sock_reuseport *reuse; 493 494 reuse = rcu_dereference(sk->sk_reuseport_cb); 495 if (likely(reuse)) { 496 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 497 if (time_after32(now, last_overflow + HZ)) 498 WRITE_ONCE(reuse->synq_overflow_ts, now); 499 return; 500 } 501 } 502 503 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 504 if (time_after32(now, last_overflow + HZ)) 505 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 506 } 507 508 /* syncookies: no recent synqueue overflow on this listening socket? */ 509 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 510 { 511 unsigned int last_overflow; 512 unsigned int now = jiffies; 513 514 if (sk->sk_reuseport) { 515 struct sock_reuseport *reuse; 516 517 reuse = rcu_dereference(sk->sk_reuseport_cb); 518 if (likely(reuse)) { 519 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 520 return time_after32(now, last_overflow + 521 TCP_SYNCOOKIE_VALID); 522 } 523 } 524 525 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 526 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID); 527 } 528 529 static inline u32 tcp_cookie_time(void) 530 { 531 u64 val = get_jiffies_64(); 532 533 do_div(val, TCP_SYNCOOKIE_PERIOD); 534 return val; 535 } 536 537 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 538 u16 *mssp); 539 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 540 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 541 bool cookie_timestamp_decode(const struct net *net, 542 struct tcp_options_received *opt); 543 bool cookie_ecn_ok(const struct tcp_options_received *opt, 544 const struct net *net, const struct dst_entry *dst); 545 546 /* From net/ipv6/syncookies.c */ 547 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 548 u32 cookie); 549 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 550 551 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 552 const struct tcphdr *th, u16 *mssp); 553 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 554 #endif 555 /* tcp_output.c */ 556 557 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 558 int nonagle); 559 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 560 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 561 void tcp_retransmit_timer(struct sock *sk); 562 void tcp_xmit_retransmit_queue(struct sock *); 563 void tcp_simple_retransmit(struct sock *); 564 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 565 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 566 enum tcp_queue { 567 TCP_FRAG_IN_WRITE_QUEUE, 568 TCP_FRAG_IN_RTX_QUEUE, 569 }; 570 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 571 struct sk_buff *skb, u32 len, 572 unsigned int mss_now, gfp_t gfp); 573 574 void tcp_send_probe0(struct sock *); 575 void tcp_send_partial(struct sock *); 576 int tcp_write_wakeup(struct sock *, int mib); 577 void tcp_send_fin(struct sock *sk); 578 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 579 int tcp_send_synack(struct sock *); 580 void tcp_push_one(struct sock *, unsigned int mss_now); 581 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 582 void tcp_send_ack(struct sock *sk); 583 void tcp_send_delayed_ack(struct sock *sk); 584 void tcp_send_loss_probe(struct sock *sk); 585 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 586 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 587 const struct sk_buff *next_skb); 588 589 /* tcp_input.c */ 590 void tcp_rearm_rto(struct sock *sk); 591 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 592 void tcp_reset(struct sock *sk); 593 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 594 void tcp_fin(struct sock *sk); 595 596 /* tcp_timer.c */ 597 void tcp_init_xmit_timers(struct sock *); 598 static inline void tcp_clear_xmit_timers(struct sock *sk) 599 { 600 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 601 __sock_put(sk); 602 603 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 604 __sock_put(sk); 605 606 inet_csk_clear_xmit_timers(sk); 607 } 608 609 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 610 unsigned int tcp_current_mss(struct sock *sk); 611 612 /* Bound MSS / TSO packet size with the half of the window */ 613 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 614 { 615 int cutoff; 616 617 /* When peer uses tiny windows, there is no use in packetizing 618 * to sub-MSS pieces for the sake of SWS or making sure there 619 * are enough packets in the pipe for fast recovery. 620 * 621 * On the other hand, for extremely large MSS devices, handling 622 * smaller than MSS windows in this way does make sense. 623 */ 624 if (tp->max_window > TCP_MSS_DEFAULT) 625 cutoff = (tp->max_window >> 1); 626 else 627 cutoff = tp->max_window; 628 629 if (cutoff && pktsize > cutoff) 630 return max_t(int, cutoff, 68U - tp->tcp_header_len); 631 else 632 return pktsize; 633 } 634 635 /* tcp.c */ 636 void tcp_get_info(struct sock *, struct tcp_info *); 637 638 /* Read 'sendfile()'-style from a TCP socket */ 639 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 640 sk_read_actor_t recv_actor); 641 642 void tcp_initialize_rcv_mss(struct sock *sk); 643 644 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 645 int tcp_mss_to_mtu(struct sock *sk, int mss); 646 void tcp_mtup_init(struct sock *sk); 647 void tcp_init_buffer_space(struct sock *sk); 648 649 static inline void tcp_bound_rto(const struct sock *sk) 650 { 651 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 652 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 653 } 654 655 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 656 { 657 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 658 } 659 660 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 661 { 662 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 663 ntohl(TCP_FLAG_ACK) | 664 snd_wnd); 665 } 666 667 static inline void tcp_fast_path_on(struct tcp_sock *tp) 668 { 669 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 670 } 671 672 static inline void tcp_fast_path_check(struct sock *sk) 673 { 674 struct tcp_sock *tp = tcp_sk(sk); 675 676 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 677 tp->rcv_wnd && 678 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 679 !tp->urg_data) 680 tcp_fast_path_on(tp); 681 } 682 683 /* Compute the actual rto_min value */ 684 static inline u32 tcp_rto_min(struct sock *sk) 685 { 686 const struct dst_entry *dst = __sk_dst_get(sk); 687 u32 rto_min = TCP_RTO_MIN; 688 689 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 690 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 691 return rto_min; 692 } 693 694 static inline u32 tcp_rto_min_us(struct sock *sk) 695 { 696 return jiffies_to_usecs(tcp_rto_min(sk)); 697 } 698 699 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 700 { 701 return dst_metric_locked(dst, RTAX_CC_ALGO); 702 } 703 704 /* Minimum RTT in usec. ~0 means not available. */ 705 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 706 { 707 return minmax_get(&tp->rtt_min); 708 } 709 710 /* Compute the actual receive window we are currently advertising. 711 * Rcv_nxt can be after the window if our peer push more data 712 * than the offered window. 713 */ 714 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 715 { 716 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 717 718 if (win < 0) 719 win = 0; 720 return (u32) win; 721 } 722 723 /* Choose a new window, without checks for shrinking, and without 724 * scaling applied to the result. The caller does these things 725 * if necessary. This is a "raw" window selection. 726 */ 727 u32 __tcp_select_window(struct sock *sk); 728 729 void tcp_send_window_probe(struct sock *sk); 730 731 /* TCP uses 32bit jiffies to save some space. 732 * Note that this is different from tcp_time_stamp, which 733 * historically has been the same until linux-4.13. 734 */ 735 #define tcp_jiffies32 ((u32)jiffies) 736 737 /* 738 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 739 * It is no longer tied to jiffies, but to 1 ms clock. 740 * Note: double check if you want to use tcp_jiffies32 instead of this. 741 */ 742 #define TCP_TS_HZ 1000 743 744 static inline u64 tcp_clock_ns(void) 745 { 746 return ktime_get_ns(); 747 } 748 749 static inline u64 tcp_clock_us(void) 750 { 751 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 752 } 753 754 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 755 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 756 { 757 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 758 } 759 760 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 761 static inline u32 tcp_ns_to_ts(u64 ns) 762 { 763 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 764 } 765 766 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 767 static inline u32 tcp_time_stamp_raw(void) 768 { 769 return tcp_ns_to_ts(tcp_clock_ns()); 770 } 771 772 void tcp_mstamp_refresh(struct tcp_sock *tp); 773 774 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 775 { 776 return max_t(s64, t1 - t0, 0); 777 } 778 779 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 780 { 781 return tcp_ns_to_ts(skb->skb_mstamp_ns); 782 } 783 784 /* provide the departure time in us unit */ 785 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 786 { 787 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 788 } 789 790 791 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 792 793 #define TCPHDR_FIN 0x01 794 #define TCPHDR_SYN 0x02 795 #define TCPHDR_RST 0x04 796 #define TCPHDR_PSH 0x08 797 #define TCPHDR_ACK 0x10 798 #define TCPHDR_URG 0x20 799 #define TCPHDR_ECE 0x40 800 #define TCPHDR_CWR 0x80 801 802 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 803 804 /* This is what the send packet queuing engine uses to pass 805 * TCP per-packet control information to the transmission code. 806 * We also store the host-order sequence numbers in here too. 807 * This is 44 bytes if IPV6 is enabled. 808 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 809 */ 810 struct tcp_skb_cb { 811 __u32 seq; /* Starting sequence number */ 812 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 813 union { 814 /* Note : tcp_tw_isn is used in input path only 815 * (isn chosen by tcp_timewait_state_process()) 816 * 817 * tcp_gso_segs/size are used in write queue only, 818 * cf tcp_skb_pcount()/tcp_skb_mss() 819 */ 820 __u32 tcp_tw_isn; 821 struct { 822 u16 tcp_gso_segs; 823 u16 tcp_gso_size; 824 }; 825 }; 826 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 827 828 __u8 sacked; /* State flags for SACK. */ 829 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 830 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 831 #define TCPCB_LOST 0x04 /* SKB is lost */ 832 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 833 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 834 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 835 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 836 TCPCB_REPAIRED) 837 838 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 839 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 840 eor:1, /* Is skb MSG_EOR marked? */ 841 has_rxtstamp:1, /* SKB has a RX timestamp */ 842 unused:5; 843 __u32 ack_seq; /* Sequence number ACK'd */ 844 union { 845 struct { 846 /* There is space for up to 24 bytes */ 847 __u32 in_flight:30,/* Bytes in flight at transmit */ 848 is_app_limited:1, /* cwnd not fully used? */ 849 unused:1; 850 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 851 __u32 delivered; 852 /* start of send pipeline phase */ 853 u64 first_tx_mstamp; 854 /* when we reached the "delivered" count */ 855 u64 delivered_mstamp; 856 } tx; /* only used for outgoing skbs */ 857 union { 858 struct inet_skb_parm h4; 859 #if IS_ENABLED(CONFIG_IPV6) 860 struct inet6_skb_parm h6; 861 #endif 862 } header; /* For incoming skbs */ 863 struct { 864 __u32 flags; 865 struct sock *sk_redir; 866 void *data_end; 867 } bpf; 868 }; 869 }; 870 871 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 872 873 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 874 { 875 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 876 } 877 878 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 879 { 880 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 881 } 882 883 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 884 { 885 return TCP_SKB_CB(skb)->bpf.sk_redir; 886 } 887 888 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 889 { 890 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 891 } 892 893 #if IS_ENABLED(CONFIG_IPV6) 894 /* This is the variant of inet6_iif() that must be used by TCP, 895 * as TCP moves IP6CB into a different location in skb->cb[] 896 */ 897 static inline int tcp_v6_iif(const struct sk_buff *skb) 898 { 899 return TCP_SKB_CB(skb)->header.h6.iif; 900 } 901 902 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 903 { 904 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 905 906 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 907 } 908 909 /* TCP_SKB_CB reference means this can not be used from early demux */ 910 static inline int tcp_v6_sdif(const struct sk_buff *skb) 911 { 912 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 913 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 914 return TCP_SKB_CB(skb)->header.h6.iif; 915 #endif 916 return 0; 917 } 918 #endif 919 920 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 921 { 922 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 923 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 924 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 925 return true; 926 #endif 927 return false; 928 } 929 930 /* TCP_SKB_CB reference means this can not be used from early demux */ 931 static inline int tcp_v4_sdif(struct sk_buff *skb) 932 { 933 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 934 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 935 return TCP_SKB_CB(skb)->header.h4.iif; 936 #endif 937 return 0; 938 } 939 940 /* Due to TSO, an SKB can be composed of multiple actual 941 * packets. To keep these tracked properly, we use this. 942 */ 943 static inline int tcp_skb_pcount(const struct sk_buff *skb) 944 { 945 return TCP_SKB_CB(skb)->tcp_gso_segs; 946 } 947 948 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 949 { 950 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 951 } 952 953 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 954 { 955 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 956 } 957 958 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 959 static inline int tcp_skb_mss(const struct sk_buff *skb) 960 { 961 return TCP_SKB_CB(skb)->tcp_gso_size; 962 } 963 964 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 965 { 966 return likely(!TCP_SKB_CB(skb)->eor); 967 } 968 969 /* Events passed to congestion control interface */ 970 enum tcp_ca_event { 971 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 972 CA_EVENT_CWND_RESTART, /* congestion window restart */ 973 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 974 CA_EVENT_LOSS, /* loss timeout */ 975 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 976 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 977 }; 978 979 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 980 enum tcp_ca_ack_event_flags { 981 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 982 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 983 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 984 }; 985 986 /* 987 * Interface for adding new TCP congestion control handlers 988 */ 989 #define TCP_CA_NAME_MAX 16 990 #define TCP_CA_MAX 128 991 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 992 993 #define TCP_CA_UNSPEC 0 994 995 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 996 #define TCP_CONG_NON_RESTRICTED 0x1 997 /* Requires ECN/ECT set on all packets */ 998 #define TCP_CONG_NEEDS_ECN 0x2 999 1000 union tcp_cc_info; 1001 1002 struct ack_sample { 1003 u32 pkts_acked; 1004 s32 rtt_us; 1005 u32 in_flight; 1006 }; 1007 1008 /* A rate sample measures the number of (original/retransmitted) data 1009 * packets delivered "delivered" over an interval of time "interval_us". 1010 * The tcp_rate.c code fills in the rate sample, and congestion 1011 * control modules that define a cong_control function to run at the end 1012 * of ACK processing can optionally chose to consult this sample when 1013 * setting cwnd and pacing rate. 1014 * A sample is invalid if "delivered" or "interval_us" is negative. 1015 */ 1016 struct rate_sample { 1017 u64 prior_mstamp; /* starting timestamp for interval */ 1018 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1019 s32 delivered; /* number of packets delivered over interval */ 1020 long interval_us; /* time for tp->delivered to incr "delivered" */ 1021 u32 snd_interval_us; /* snd interval for delivered packets */ 1022 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1023 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1024 int losses; /* number of packets marked lost upon ACK */ 1025 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1026 u32 prior_in_flight; /* in flight before this ACK */ 1027 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1028 bool is_retrans; /* is sample from retransmission? */ 1029 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1030 }; 1031 1032 struct tcp_congestion_ops { 1033 struct list_head list; 1034 u32 key; 1035 u32 flags; 1036 1037 /* initialize private data (optional) */ 1038 void (*init)(struct sock *sk); 1039 /* cleanup private data (optional) */ 1040 void (*release)(struct sock *sk); 1041 1042 /* return slow start threshold (required) */ 1043 u32 (*ssthresh)(struct sock *sk); 1044 /* do new cwnd calculation (required) */ 1045 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1046 /* call before changing ca_state (optional) */ 1047 void (*set_state)(struct sock *sk, u8 new_state); 1048 /* call when cwnd event occurs (optional) */ 1049 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1050 /* call when ack arrives (optional) */ 1051 void (*in_ack_event)(struct sock *sk, u32 flags); 1052 /* new value of cwnd after loss (required) */ 1053 u32 (*undo_cwnd)(struct sock *sk); 1054 /* hook for packet ack accounting (optional) */ 1055 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1056 /* override sysctl_tcp_min_tso_segs */ 1057 u32 (*min_tso_segs)(struct sock *sk); 1058 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1059 u32 (*sndbuf_expand)(struct sock *sk); 1060 /* call when packets are delivered to update cwnd and pacing rate, 1061 * after all the ca_state processing. (optional) 1062 */ 1063 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1064 /* get info for inet_diag (optional) */ 1065 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1066 union tcp_cc_info *info); 1067 1068 char name[TCP_CA_NAME_MAX]; 1069 struct module *owner; 1070 }; 1071 1072 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1073 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1074 1075 void tcp_assign_congestion_control(struct sock *sk); 1076 void tcp_init_congestion_control(struct sock *sk); 1077 void tcp_cleanup_congestion_control(struct sock *sk); 1078 int tcp_set_default_congestion_control(struct net *net, const char *name); 1079 void tcp_get_default_congestion_control(struct net *net, char *name); 1080 void tcp_get_available_congestion_control(char *buf, size_t len); 1081 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1082 int tcp_set_allowed_congestion_control(char *allowed); 1083 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1084 bool reinit, bool cap_net_admin); 1085 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1086 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1087 1088 u32 tcp_reno_ssthresh(struct sock *sk); 1089 u32 tcp_reno_undo_cwnd(struct sock *sk); 1090 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1091 extern struct tcp_congestion_ops tcp_reno; 1092 1093 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1094 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1095 #ifdef CONFIG_INET 1096 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1097 #else 1098 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1099 { 1100 return NULL; 1101 } 1102 #endif 1103 1104 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1105 { 1106 const struct inet_connection_sock *icsk = inet_csk(sk); 1107 1108 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1109 } 1110 1111 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1112 { 1113 struct inet_connection_sock *icsk = inet_csk(sk); 1114 1115 if (icsk->icsk_ca_ops->set_state) 1116 icsk->icsk_ca_ops->set_state(sk, ca_state); 1117 icsk->icsk_ca_state = ca_state; 1118 } 1119 1120 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1121 { 1122 const struct inet_connection_sock *icsk = inet_csk(sk); 1123 1124 if (icsk->icsk_ca_ops->cwnd_event) 1125 icsk->icsk_ca_ops->cwnd_event(sk, event); 1126 } 1127 1128 /* From tcp_rate.c */ 1129 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1130 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1131 struct rate_sample *rs); 1132 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1133 bool is_sack_reneg, struct rate_sample *rs); 1134 void tcp_rate_check_app_limited(struct sock *sk); 1135 1136 /* These functions determine how the current flow behaves in respect of SACK 1137 * handling. SACK is negotiated with the peer, and therefore it can vary 1138 * between different flows. 1139 * 1140 * tcp_is_sack - SACK enabled 1141 * tcp_is_reno - No SACK 1142 */ 1143 static inline int tcp_is_sack(const struct tcp_sock *tp) 1144 { 1145 return likely(tp->rx_opt.sack_ok); 1146 } 1147 1148 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1149 { 1150 return !tcp_is_sack(tp); 1151 } 1152 1153 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1154 { 1155 return tp->sacked_out + tp->lost_out; 1156 } 1157 1158 /* This determines how many packets are "in the network" to the best 1159 * of our knowledge. In many cases it is conservative, but where 1160 * detailed information is available from the receiver (via SACK 1161 * blocks etc.) we can make more aggressive calculations. 1162 * 1163 * Use this for decisions involving congestion control, use just 1164 * tp->packets_out to determine if the send queue is empty or not. 1165 * 1166 * Read this equation as: 1167 * 1168 * "Packets sent once on transmission queue" MINUS 1169 * "Packets left network, but not honestly ACKed yet" PLUS 1170 * "Packets fast retransmitted" 1171 */ 1172 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1173 { 1174 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1175 } 1176 1177 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1178 1179 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1180 { 1181 return tp->snd_cwnd < tp->snd_ssthresh; 1182 } 1183 1184 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1185 { 1186 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1187 } 1188 1189 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1190 { 1191 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1192 (1 << inet_csk(sk)->icsk_ca_state); 1193 } 1194 1195 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1196 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1197 * ssthresh. 1198 */ 1199 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1200 { 1201 const struct tcp_sock *tp = tcp_sk(sk); 1202 1203 if (tcp_in_cwnd_reduction(sk)) 1204 return tp->snd_ssthresh; 1205 else 1206 return max(tp->snd_ssthresh, 1207 ((tp->snd_cwnd >> 1) + 1208 (tp->snd_cwnd >> 2))); 1209 } 1210 1211 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1212 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1213 1214 void tcp_enter_cwr(struct sock *sk); 1215 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1216 1217 /* The maximum number of MSS of available cwnd for which TSO defers 1218 * sending if not using sysctl_tcp_tso_win_divisor. 1219 */ 1220 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1221 { 1222 return 3; 1223 } 1224 1225 /* Returns end sequence number of the receiver's advertised window */ 1226 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1227 { 1228 return tp->snd_una + tp->snd_wnd; 1229 } 1230 1231 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1232 * flexible approach. The RFC suggests cwnd should not be raised unless 1233 * it was fully used previously. And that's exactly what we do in 1234 * congestion avoidance mode. But in slow start we allow cwnd to grow 1235 * as long as the application has used half the cwnd. 1236 * Example : 1237 * cwnd is 10 (IW10), but application sends 9 frames. 1238 * We allow cwnd to reach 18 when all frames are ACKed. 1239 * This check is safe because it's as aggressive as slow start which already 1240 * risks 100% overshoot. The advantage is that we discourage application to 1241 * either send more filler packets or data to artificially blow up the cwnd 1242 * usage, and allow application-limited process to probe bw more aggressively. 1243 */ 1244 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1245 { 1246 const struct tcp_sock *tp = tcp_sk(sk); 1247 1248 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1249 if (tcp_in_slow_start(tp)) 1250 return tp->snd_cwnd < 2 * tp->max_packets_out; 1251 1252 return tp->is_cwnd_limited; 1253 } 1254 1255 /* BBR congestion control needs pacing. 1256 * Same remark for SO_MAX_PACING_RATE. 1257 * sch_fq packet scheduler is efficiently handling pacing, 1258 * but is not always installed/used. 1259 * Return true if TCP stack should pace packets itself. 1260 */ 1261 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1262 { 1263 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1264 } 1265 1266 /* Return in jiffies the delay before one skb is sent. 1267 * If @skb is NULL, we look at EDT for next packet being sent on the socket. 1268 */ 1269 static inline unsigned long tcp_pacing_delay(const struct sock *sk, 1270 const struct sk_buff *skb) 1271 { 1272 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns; 1273 1274 pacing_delay -= tcp_sk(sk)->tcp_clock_cache; 1275 1276 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0; 1277 } 1278 1279 static inline void tcp_reset_xmit_timer(struct sock *sk, 1280 const int what, 1281 unsigned long when, 1282 const unsigned long max_when, 1283 const struct sk_buff *skb) 1284 { 1285 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb), 1286 max_when); 1287 } 1288 1289 /* Something is really bad, we could not queue an additional packet, 1290 * because qdisc is full or receiver sent a 0 window, or we are paced. 1291 * We do not want to add fuel to the fire, or abort too early, 1292 * so make sure the timer we arm now is at least 200ms in the future, 1293 * regardless of current icsk_rto value (as it could be ~2ms) 1294 */ 1295 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1296 { 1297 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1298 } 1299 1300 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1301 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1302 unsigned long max_when) 1303 { 1304 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1305 1306 return (unsigned long)min_t(u64, when, max_when); 1307 } 1308 1309 static inline void tcp_check_probe_timer(struct sock *sk) 1310 { 1311 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1312 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1313 tcp_probe0_base(sk), TCP_RTO_MAX, 1314 NULL); 1315 } 1316 1317 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1318 { 1319 tp->snd_wl1 = seq; 1320 } 1321 1322 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1323 { 1324 tp->snd_wl1 = seq; 1325 } 1326 1327 /* 1328 * Calculate(/check) TCP checksum 1329 */ 1330 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1331 __be32 daddr, __wsum base) 1332 { 1333 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1334 } 1335 1336 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1337 { 1338 return !skb_csum_unnecessary(skb) && 1339 __skb_checksum_complete(skb); 1340 } 1341 1342 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1343 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1344 void tcp_set_state(struct sock *sk, int state); 1345 void tcp_done(struct sock *sk); 1346 int tcp_abort(struct sock *sk, int err); 1347 1348 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1349 { 1350 rx_opt->dsack = 0; 1351 rx_opt->num_sacks = 0; 1352 } 1353 1354 u32 tcp_default_init_rwnd(u32 mss); 1355 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1356 1357 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1358 { 1359 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1360 struct tcp_sock *tp = tcp_sk(sk); 1361 s32 delta; 1362 1363 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1364 ca_ops->cong_control) 1365 return; 1366 delta = tcp_jiffies32 - tp->lsndtime; 1367 if (delta > inet_csk(sk)->icsk_rto) 1368 tcp_cwnd_restart(sk, delta); 1369 } 1370 1371 /* Determine a window scaling and initial window to offer. */ 1372 void tcp_select_initial_window(const struct sock *sk, int __space, 1373 __u32 mss, __u32 *rcv_wnd, 1374 __u32 *window_clamp, int wscale_ok, 1375 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1376 1377 static inline int tcp_win_from_space(const struct sock *sk, int space) 1378 { 1379 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1380 1381 return tcp_adv_win_scale <= 0 ? 1382 (space>>(-tcp_adv_win_scale)) : 1383 space - (space>>tcp_adv_win_scale); 1384 } 1385 1386 /* Note: caller must be prepared to deal with negative returns */ 1387 static inline int tcp_space(const struct sock *sk) 1388 { 1389 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1390 READ_ONCE(sk->sk_backlog.len) - 1391 atomic_read(&sk->sk_rmem_alloc)); 1392 } 1393 1394 static inline int tcp_full_space(const struct sock *sk) 1395 { 1396 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1397 } 1398 1399 extern void tcp_openreq_init_rwin(struct request_sock *req, 1400 const struct sock *sk_listener, 1401 const struct dst_entry *dst); 1402 1403 void tcp_enter_memory_pressure(struct sock *sk); 1404 void tcp_leave_memory_pressure(struct sock *sk); 1405 1406 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1407 { 1408 struct net *net = sock_net((struct sock *)tp); 1409 1410 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1411 } 1412 1413 static inline int keepalive_time_when(const struct tcp_sock *tp) 1414 { 1415 struct net *net = sock_net((struct sock *)tp); 1416 1417 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1418 } 1419 1420 static inline int keepalive_probes(const struct tcp_sock *tp) 1421 { 1422 struct net *net = sock_net((struct sock *)tp); 1423 1424 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1425 } 1426 1427 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1428 { 1429 const struct inet_connection_sock *icsk = &tp->inet_conn; 1430 1431 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1432 tcp_jiffies32 - tp->rcv_tstamp); 1433 } 1434 1435 static inline int tcp_fin_time(const struct sock *sk) 1436 { 1437 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1438 const int rto = inet_csk(sk)->icsk_rto; 1439 1440 if (fin_timeout < (rto << 2) - (rto >> 1)) 1441 fin_timeout = (rto << 2) - (rto >> 1); 1442 1443 return fin_timeout; 1444 } 1445 1446 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1447 int paws_win) 1448 { 1449 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1450 return true; 1451 if (unlikely(!time_before32(ktime_get_seconds(), 1452 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1453 return true; 1454 /* 1455 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1456 * then following tcp messages have valid values. Ignore 0 value, 1457 * or else 'negative' tsval might forbid us to accept their packets. 1458 */ 1459 if (!rx_opt->ts_recent) 1460 return true; 1461 return false; 1462 } 1463 1464 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1465 int rst) 1466 { 1467 if (tcp_paws_check(rx_opt, 0)) 1468 return false; 1469 1470 /* RST segments are not recommended to carry timestamp, 1471 and, if they do, it is recommended to ignore PAWS because 1472 "their cleanup function should take precedence over timestamps." 1473 Certainly, it is mistake. It is necessary to understand the reasons 1474 of this constraint to relax it: if peer reboots, clock may go 1475 out-of-sync and half-open connections will not be reset. 1476 Actually, the problem would be not existing if all 1477 the implementations followed draft about maintaining clock 1478 via reboots. Linux-2.2 DOES NOT! 1479 1480 However, we can relax time bounds for RST segments to MSL. 1481 */ 1482 if (rst && !time_before32(ktime_get_seconds(), 1483 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1484 return false; 1485 return true; 1486 } 1487 1488 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1489 int mib_idx, u32 *last_oow_ack_time); 1490 1491 static inline void tcp_mib_init(struct net *net) 1492 { 1493 /* See RFC 2012 */ 1494 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1495 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1496 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1497 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1498 } 1499 1500 /* from STCP */ 1501 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1502 { 1503 tp->lost_skb_hint = NULL; 1504 } 1505 1506 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1507 { 1508 tcp_clear_retrans_hints_partial(tp); 1509 tp->retransmit_skb_hint = NULL; 1510 } 1511 1512 union tcp_md5_addr { 1513 struct in_addr a4; 1514 #if IS_ENABLED(CONFIG_IPV6) 1515 struct in6_addr a6; 1516 #endif 1517 }; 1518 1519 /* - key database */ 1520 struct tcp_md5sig_key { 1521 struct hlist_node node; 1522 u8 keylen; 1523 u8 family; /* AF_INET or AF_INET6 */ 1524 union tcp_md5_addr addr; 1525 u8 prefixlen; 1526 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1527 struct rcu_head rcu; 1528 }; 1529 1530 /* - sock block */ 1531 struct tcp_md5sig_info { 1532 struct hlist_head head; 1533 struct rcu_head rcu; 1534 }; 1535 1536 /* - pseudo header */ 1537 struct tcp4_pseudohdr { 1538 __be32 saddr; 1539 __be32 daddr; 1540 __u8 pad; 1541 __u8 protocol; 1542 __be16 len; 1543 }; 1544 1545 struct tcp6_pseudohdr { 1546 struct in6_addr saddr; 1547 struct in6_addr daddr; 1548 __be32 len; 1549 __be32 protocol; /* including padding */ 1550 }; 1551 1552 union tcp_md5sum_block { 1553 struct tcp4_pseudohdr ip4; 1554 #if IS_ENABLED(CONFIG_IPV6) 1555 struct tcp6_pseudohdr ip6; 1556 #endif 1557 }; 1558 1559 /* - pool: digest algorithm, hash description and scratch buffer */ 1560 struct tcp_md5sig_pool { 1561 struct ahash_request *md5_req; 1562 void *scratch; 1563 }; 1564 1565 /* - functions */ 1566 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1567 const struct sock *sk, const struct sk_buff *skb); 1568 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1569 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1570 gfp_t gfp); 1571 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1572 int family, u8 prefixlen); 1573 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1574 const struct sock *addr_sk); 1575 1576 #ifdef CONFIG_TCP_MD5SIG 1577 #include <linux/jump_label.h> 1578 extern struct static_key_false tcp_md5_needed; 1579 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, 1580 const union tcp_md5_addr *addr, 1581 int family); 1582 static inline struct tcp_md5sig_key * 1583 tcp_md5_do_lookup(const struct sock *sk, 1584 const union tcp_md5_addr *addr, 1585 int family) 1586 { 1587 if (!static_branch_unlikely(&tcp_md5_needed)) 1588 return NULL; 1589 return __tcp_md5_do_lookup(sk, addr, family); 1590 } 1591 1592 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1593 #else 1594 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1595 const union tcp_md5_addr *addr, 1596 int family) 1597 { 1598 return NULL; 1599 } 1600 #define tcp_twsk_md5_key(twsk) NULL 1601 #endif 1602 1603 bool tcp_alloc_md5sig_pool(void); 1604 1605 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1606 static inline void tcp_put_md5sig_pool(void) 1607 { 1608 local_bh_enable(); 1609 } 1610 1611 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1612 unsigned int header_len); 1613 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1614 const struct tcp_md5sig_key *key); 1615 1616 /* From tcp_fastopen.c */ 1617 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1618 struct tcp_fastopen_cookie *cookie); 1619 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1620 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1621 u16 try_exp); 1622 struct tcp_fastopen_request { 1623 /* Fast Open cookie. Size 0 means a cookie request */ 1624 struct tcp_fastopen_cookie cookie; 1625 struct msghdr *data; /* data in MSG_FASTOPEN */ 1626 size_t size; 1627 int copied; /* queued in tcp_connect() */ 1628 struct ubuf_info *uarg; 1629 }; 1630 void tcp_free_fastopen_req(struct tcp_sock *tp); 1631 void tcp_fastopen_destroy_cipher(struct sock *sk); 1632 void tcp_fastopen_ctx_destroy(struct net *net); 1633 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1634 void *primary_key, void *backup_key); 1635 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1636 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1637 struct request_sock *req, 1638 struct tcp_fastopen_cookie *foc, 1639 const struct dst_entry *dst); 1640 void tcp_fastopen_init_key_once(struct net *net); 1641 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1642 struct tcp_fastopen_cookie *cookie); 1643 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1644 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1645 #define TCP_FASTOPEN_KEY_MAX 2 1646 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1647 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1648 1649 /* Fastopen key context */ 1650 struct tcp_fastopen_context { 1651 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1652 int num; 1653 struct rcu_head rcu; 1654 }; 1655 1656 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1657 void tcp_fastopen_active_disable(struct sock *sk); 1658 bool tcp_fastopen_active_should_disable(struct sock *sk); 1659 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1660 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1661 1662 /* Caller needs to wrap with rcu_read_(un)lock() */ 1663 static inline 1664 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1665 { 1666 struct tcp_fastopen_context *ctx; 1667 1668 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1669 if (!ctx) 1670 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1671 return ctx; 1672 } 1673 1674 static inline 1675 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1676 const struct tcp_fastopen_cookie *orig) 1677 { 1678 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1679 orig->len == foc->len && 1680 !memcmp(orig->val, foc->val, foc->len)) 1681 return true; 1682 return false; 1683 } 1684 1685 static inline 1686 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1687 { 1688 return ctx->num; 1689 } 1690 1691 /* Latencies incurred by various limits for a sender. They are 1692 * chronograph-like stats that are mutually exclusive. 1693 */ 1694 enum tcp_chrono { 1695 TCP_CHRONO_UNSPEC, 1696 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1697 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1698 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1699 __TCP_CHRONO_MAX, 1700 }; 1701 1702 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1703 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1704 1705 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1706 * the same memory storage than skb->destructor/_skb_refdst 1707 */ 1708 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1709 { 1710 skb->destructor = NULL; 1711 skb->_skb_refdst = 0UL; 1712 } 1713 1714 #define tcp_skb_tsorted_save(skb) { \ 1715 unsigned long _save = skb->_skb_refdst; \ 1716 skb->_skb_refdst = 0UL; 1717 1718 #define tcp_skb_tsorted_restore(skb) \ 1719 skb->_skb_refdst = _save; \ 1720 } 1721 1722 void tcp_write_queue_purge(struct sock *sk); 1723 1724 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1725 { 1726 return skb_rb_first(&sk->tcp_rtx_queue); 1727 } 1728 1729 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1730 { 1731 return skb_rb_last(&sk->tcp_rtx_queue); 1732 } 1733 1734 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1735 { 1736 return skb_peek(&sk->sk_write_queue); 1737 } 1738 1739 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1740 { 1741 return skb_peek_tail(&sk->sk_write_queue); 1742 } 1743 1744 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1745 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1746 1747 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1748 { 1749 return skb_peek(&sk->sk_write_queue); 1750 } 1751 1752 static inline bool tcp_skb_is_last(const struct sock *sk, 1753 const struct sk_buff *skb) 1754 { 1755 return skb_queue_is_last(&sk->sk_write_queue, skb); 1756 } 1757 1758 static inline bool tcp_write_queue_empty(const struct sock *sk) 1759 { 1760 return skb_queue_empty(&sk->sk_write_queue); 1761 } 1762 1763 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1764 { 1765 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1766 } 1767 1768 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1769 { 1770 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1771 } 1772 1773 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1774 { 1775 __skb_queue_tail(&sk->sk_write_queue, skb); 1776 1777 /* Queue it, remembering where we must start sending. */ 1778 if (sk->sk_write_queue.next == skb) 1779 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1780 } 1781 1782 /* Insert new before skb on the write queue of sk. */ 1783 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1784 struct sk_buff *skb, 1785 struct sock *sk) 1786 { 1787 __skb_queue_before(&sk->sk_write_queue, skb, new); 1788 } 1789 1790 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1791 { 1792 tcp_skb_tsorted_anchor_cleanup(skb); 1793 __skb_unlink(skb, &sk->sk_write_queue); 1794 } 1795 1796 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1797 1798 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1799 { 1800 tcp_skb_tsorted_anchor_cleanup(skb); 1801 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1802 } 1803 1804 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1805 { 1806 list_del(&skb->tcp_tsorted_anchor); 1807 tcp_rtx_queue_unlink(skb, sk); 1808 sk_wmem_free_skb(sk, skb); 1809 } 1810 1811 static inline void tcp_push_pending_frames(struct sock *sk) 1812 { 1813 if (tcp_send_head(sk)) { 1814 struct tcp_sock *tp = tcp_sk(sk); 1815 1816 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1817 } 1818 } 1819 1820 /* Start sequence of the skb just after the highest skb with SACKed 1821 * bit, valid only if sacked_out > 0 or when the caller has ensured 1822 * validity by itself. 1823 */ 1824 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1825 { 1826 if (!tp->sacked_out) 1827 return tp->snd_una; 1828 1829 if (tp->highest_sack == NULL) 1830 return tp->snd_nxt; 1831 1832 return TCP_SKB_CB(tp->highest_sack)->seq; 1833 } 1834 1835 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1836 { 1837 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1838 } 1839 1840 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1841 { 1842 return tcp_sk(sk)->highest_sack; 1843 } 1844 1845 static inline void tcp_highest_sack_reset(struct sock *sk) 1846 { 1847 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1848 } 1849 1850 /* Called when old skb is about to be deleted and replaced by new skb */ 1851 static inline void tcp_highest_sack_replace(struct sock *sk, 1852 struct sk_buff *old, 1853 struct sk_buff *new) 1854 { 1855 if (old == tcp_highest_sack(sk)) 1856 tcp_sk(sk)->highest_sack = new; 1857 } 1858 1859 /* This helper checks if socket has IP_TRANSPARENT set */ 1860 static inline bool inet_sk_transparent(const struct sock *sk) 1861 { 1862 switch (sk->sk_state) { 1863 case TCP_TIME_WAIT: 1864 return inet_twsk(sk)->tw_transparent; 1865 case TCP_NEW_SYN_RECV: 1866 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1867 } 1868 return inet_sk(sk)->transparent; 1869 } 1870 1871 /* Determines whether this is a thin stream (which may suffer from 1872 * increased latency). Used to trigger latency-reducing mechanisms. 1873 */ 1874 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1875 { 1876 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1877 } 1878 1879 /* /proc */ 1880 enum tcp_seq_states { 1881 TCP_SEQ_STATE_LISTENING, 1882 TCP_SEQ_STATE_ESTABLISHED, 1883 }; 1884 1885 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1886 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1887 void tcp_seq_stop(struct seq_file *seq, void *v); 1888 1889 struct tcp_seq_afinfo { 1890 sa_family_t family; 1891 }; 1892 1893 struct tcp_iter_state { 1894 struct seq_net_private p; 1895 enum tcp_seq_states state; 1896 struct sock *syn_wait_sk; 1897 int bucket, offset, sbucket, num; 1898 loff_t last_pos; 1899 }; 1900 1901 extern struct request_sock_ops tcp_request_sock_ops; 1902 extern struct request_sock_ops tcp6_request_sock_ops; 1903 1904 void tcp_v4_destroy_sock(struct sock *sk); 1905 1906 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1907 netdev_features_t features); 1908 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1909 int tcp_gro_complete(struct sk_buff *skb); 1910 1911 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1912 1913 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1914 { 1915 struct net *net = sock_net((struct sock *)tp); 1916 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1917 } 1918 1919 /* @wake is one when sk_stream_write_space() calls us. 1920 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1921 * This mimics the strategy used in sock_def_write_space(). 1922 */ 1923 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1924 { 1925 const struct tcp_sock *tp = tcp_sk(sk); 1926 u32 notsent_bytes = READ_ONCE(tp->write_seq) - 1927 READ_ONCE(tp->snd_nxt); 1928 1929 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1930 } 1931 1932 #ifdef CONFIG_PROC_FS 1933 int tcp4_proc_init(void); 1934 void tcp4_proc_exit(void); 1935 #endif 1936 1937 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1938 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1939 const struct tcp_request_sock_ops *af_ops, 1940 struct sock *sk, struct sk_buff *skb); 1941 1942 /* TCP af-specific functions */ 1943 struct tcp_sock_af_ops { 1944 #ifdef CONFIG_TCP_MD5SIG 1945 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1946 const struct sock *addr_sk); 1947 int (*calc_md5_hash)(char *location, 1948 const struct tcp_md5sig_key *md5, 1949 const struct sock *sk, 1950 const struct sk_buff *skb); 1951 int (*md5_parse)(struct sock *sk, 1952 int optname, 1953 char __user *optval, 1954 int optlen); 1955 #endif 1956 }; 1957 1958 struct tcp_request_sock_ops { 1959 u16 mss_clamp; 1960 #ifdef CONFIG_TCP_MD5SIG 1961 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1962 const struct sock *addr_sk); 1963 int (*calc_md5_hash) (char *location, 1964 const struct tcp_md5sig_key *md5, 1965 const struct sock *sk, 1966 const struct sk_buff *skb); 1967 #endif 1968 void (*init_req)(struct request_sock *req, 1969 const struct sock *sk_listener, 1970 struct sk_buff *skb); 1971 #ifdef CONFIG_SYN_COOKIES 1972 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1973 __u16 *mss); 1974 #endif 1975 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1976 const struct request_sock *req); 1977 u32 (*init_seq)(const struct sk_buff *skb); 1978 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1979 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1980 struct flowi *fl, struct request_sock *req, 1981 struct tcp_fastopen_cookie *foc, 1982 enum tcp_synack_type synack_type); 1983 }; 1984 1985 #ifdef CONFIG_SYN_COOKIES 1986 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1987 const struct sock *sk, struct sk_buff *skb, 1988 __u16 *mss) 1989 { 1990 tcp_synq_overflow(sk); 1991 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1992 return ops->cookie_init_seq(skb, mss); 1993 } 1994 #else 1995 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1996 const struct sock *sk, struct sk_buff *skb, 1997 __u16 *mss) 1998 { 1999 return 0; 2000 } 2001 #endif 2002 2003 int tcpv4_offload_init(void); 2004 2005 void tcp_v4_init(void); 2006 void tcp_init(void); 2007 2008 /* tcp_recovery.c */ 2009 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2010 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2011 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2012 u32 reo_wnd); 2013 extern void tcp_rack_mark_lost(struct sock *sk); 2014 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2015 u64 xmit_time); 2016 extern void tcp_rack_reo_timeout(struct sock *sk); 2017 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2018 2019 /* At how many usecs into the future should the RTO fire? */ 2020 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2021 { 2022 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2023 u32 rto = inet_csk(sk)->icsk_rto; 2024 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2025 2026 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2027 } 2028 2029 /* 2030 * Save and compile IPv4 options, return a pointer to it 2031 */ 2032 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2033 struct sk_buff *skb) 2034 { 2035 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2036 struct ip_options_rcu *dopt = NULL; 2037 2038 if (opt->optlen) { 2039 int opt_size = sizeof(*dopt) + opt->optlen; 2040 2041 dopt = kmalloc(opt_size, GFP_ATOMIC); 2042 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2043 kfree(dopt); 2044 dopt = NULL; 2045 } 2046 } 2047 return dopt; 2048 } 2049 2050 /* locally generated TCP pure ACKs have skb->truesize == 2 2051 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2052 * This is much faster than dissecting the packet to find out. 2053 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2054 */ 2055 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2056 { 2057 return skb->truesize == 2; 2058 } 2059 2060 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2061 { 2062 skb->truesize = 2; 2063 } 2064 2065 static inline int tcp_inq(struct sock *sk) 2066 { 2067 struct tcp_sock *tp = tcp_sk(sk); 2068 int answ; 2069 2070 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2071 answ = 0; 2072 } else if (sock_flag(sk, SOCK_URGINLINE) || 2073 !tp->urg_data || 2074 before(tp->urg_seq, tp->copied_seq) || 2075 !before(tp->urg_seq, tp->rcv_nxt)) { 2076 2077 answ = tp->rcv_nxt - tp->copied_seq; 2078 2079 /* Subtract 1, if FIN was received */ 2080 if (answ && sock_flag(sk, SOCK_DONE)) 2081 answ--; 2082 } else { 2083 answ = tp->urg_seq - tp->copied_seq; 2084 } 2085 2086 return answ; 2087 } 2088 2089 int tcp_peek_len(struct socket *sock); 2090 2091 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2092 { 2093 u16 segs_in; 2094 2095 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2096 tp->segs_in += segs_in; 2097 if (skb->len > tcp_hdrlen(skb)) 2098 tp->data_segs_in += segs_in; 2099 } 2100 2101 /* 2102 * TCP listen path runs lockless. 2103 * We forced "struct sock" to be const qualified to make sure 2104 * we don't modify one of its field by mistake. 2105 * Here, we increment sk_drops which is an atomic_t, so we can safely 2106 * make sock writable again. 2107 */ 2108 static inline void tcp_listendrop(const struct sock *sk) 2109 { 2110 atomic_inc(&((struct sock *)sk)->sk_drops); 2111 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2112 } 2113 2114 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2115 2116 /* 2117 * Interface for adding Upper Level Protocols over TCP 2118 */ 2119 2120 #define TCP_ULP_NAME_MAX 16 2121 #define TCP_ULP_MAX 128 2122 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2123 2124 struct tcp_ulp_ops { 2125 struct list_head list; 2126 2127 /* initialize ulp */ 2128 int (*init)(struct sock *sk); 2129 /* update ulp */ 2130 void (*update)(struct sock *sk, struct proto *p); 2131 /* cleanup ulp */ 2132 void (*release)(struct sock *sk); 2133 /* diagnostic */ 2134 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2135 size_t (*get_info_size)(const struct sock *sk); 2136 2137 char name[TCP_ULP_NAME_MAX]; 2138 struct module *owner; 2139 }; 2140 int tcp_register_ulp(struct tcp_ulp_ops *type); 2141 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2142 int tcp_set_ulp(struct sock *sk, const char *name); 2143 void tcp_get_available_ulp(char *buf, size_t len); 2144 void tcp_cleanup_ulp(struct sock *sk); 2145 void tcp_update_ulp(struct sock *sk, struct proto *p); 2146 2147 #define MODULE_ALIAS_TCP_ULP(name) \ 2148 __MODULE_INFO(alias, alias_userspace, name); \ 2149 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2150 2151 struct sk_msg; 2152 struct sk_psock; 2153 2154 int tcp_bpf_init(struct sock *sk); 2155 void tcp_bpf_reinit(struct sock *sk); 2156 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2157 int flags); 2158 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 2159 int nonblock, int flags, int *addr_len); 2160 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2161 struct msghdr *msg, int len, int flags); 2162 2163 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2164 * is < 0, then the BPF op failed (for example if the loaded BPF 2165 * program does not support the chosen operation or there is no BPF 2166 * program loaded). 2167 */ 2168 #ifdef CONFIG_BPF 2169 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2170 { 2171 struct bpf_sock_ops_kern sock_ops; 2172 int ret; 2173 2174 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2175 if (sk_fullsock(sk)) { 2176 sock_ops.is_fullsock = 1; 2177 sock_owned_by_me(sk); 2178 } 2179 2180 sock_ops.sk = sk; 2181 sock_ops.op = op; 2182 if (nargs > 0) 2183 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2184 2185 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2186 if (ret == 0) 2187 ret = sock_ops.reply; 2188 else 2189 ret = -1; 2190 return ret; 2191 } 2192 2193 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2194 { 2195 u32 args[2] = {arg1, arg2}; 2196 2197 return tcp_call_bpf(sk, op, 2, args); 2198 } 2199 2200 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2201 u32 arg3) 2202 { 2203 u32 args[3] = {arg1, arg2, arg3}; 2204 2205 return tcp_call_bpf(sk, op, 3, args); 2206 } 2207 2208 #else 2209 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2210 { 2211 return -EPERM; 2212 } 2213 2214 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2215 { 2216 return -EPERM; 2217 } 2218 2219 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2220 u32 arg3) 2221 { 2222 return -EPERM; 2223 } 2224 2225 #endif 2226 2227 static inline u32 tcp_timeout_init(struct sock *sk) 2228 { 2229 int timeout; 2230 2231 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2232 2233 if (timeout <= 0) 2234 timeout = TCP_TIMEOUT_INIT; 2235 return timeout; 2236 } 2237 2238 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2239 { 2240 int rwnd; 2241 2242 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2243 2244 if (rwnd < 0) 2245 rwnd = 0; 2246 return rwnd; 2247 } 2248 2249 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2250 { 2251 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2252 } 2253 2254 static inline void tcp_bpf_rtt(struct sock *sk) 2255 { 2256 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2257 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2258 } 2259 2260 #if IS_ENABLED(CONFIG_SMC) 2261 extern struct static_key_false tcp_have_smc; 2262 #endif 2263 2264 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2265 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2266 void (*cad)(struct sock *sk, u32 ack_seq)); 2267 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2268 void clean_acked_data_flush(void); 2269 #endif 2270 2271 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2272 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2273 const struct tcp_sock *tp) 2274 { 2275 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2276 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2277 } 2278 2279 /* Compute Earliest Departure Time for some control packets 2280 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2281 */ 2282 static inline u64 tcp_transmit_time(const struct sock *sk) 2283 { 2284 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2285 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2286 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2287 2288 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2289 } 2290 return 0; 2291 } 2292 2293 #endif /* _TCP_H */ 2294