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_reuseport.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 #include <linux/bpf-cgroup.h> 50 51 extern struct inet_hashinfo tcp_hashinfo; 52 53 extern struct percpu_counter tcp_orphan_count; 54 void tcp_time_wait(struct sock *sk, int state, int timeo); 55 56 #define MAX_TCP_HEADER (128 + MAX_HEADER) 57 #define MAX_TCP_OPTION_SPACE 40 58 59 /* 60 * Never offer a window over 32767 without using window scaling. Some 61 * poor stacks do signed 16bit maths! 62 */ 63 #define MAX_TCP_WINDOW 32767U 64 65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 66 #define TCP_MIN_MSS 88U 67 68 /* The least MTU to use for probing */ 69 #define TCP_BASE_MSS 1024 70 71 /* probing interval, default to 10 minutes as per RFC4821 */ 72 #define TCP_PROBE_INTERVAL 600 73 74 /* Specify interval when tcp mtu probing will stop */ 75 #define TCP_PROBE_THRESHOLD 8 76 77 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 78 #define TCP_FASTRETRANS_THRESH 3 79 80 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 81 #define TCP_MAX_QUICKACKS 16U 82 83 /* Maximal number of window scale according to RFC1323 */ 84 #define TCP_MAX_WSCALE 14U 85 86 /* urg_data states */ 87 #define TCP_URG_VALID 0x0100 88 #define TCP_URG_NOTYET 0x0200 89 #define TCP_URG_READ 0x0400 90 91 #define TCP_RETR1 3 /* 92 * This is how many retries it does before it 93 * tries to figure out if the gateway is 94 * down. Minimal RFC value is 3; it corresponds 95 * to ~3sec-8min depending on RTO. 96 */ 97 98 #define TCP_RETR2 15 /* 99 * This should take at least 100 * 90 minutes to time out. 101 * RFC1122 says that the limit is 100 sec. 102 * 15 is ~13-30min depending on RTO. 103 */ 104 105 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 106 * when active opening a connection. 107 * RFC1122 says the minimum retry MUST 108 * be at least 180secs. Nevertheless 109 * this value is corresponding to 110 * 63secs of retransmission with the 111 * current initial RTO. 112 */ 113 114 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 115 * when passive opening a connection. 116 * This is corresponding to 31secs of 117 * retransmission with the current 118 * initial RTO. 119 */ 120 121 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 122 * state, about 60 seconds */ 123 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 124 /* BSD style FIN_WAIT2 deadlock breaker. 125 * It used to be 3min, new value is 60sec, 126 * to combine FIN-WAIT-2 timeout with 127 * TIME-WAIT timer. 128 */ 129 130 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 131 #if HZ >= 100 132 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 133 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 134 #else 135 #define TCP_DELACK_MIN 4U 136 #define TCP_ATO_MIN 4U 137 #endif 138 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 139 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 140 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 141 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 142 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 143 * used as a fallback RTO for the 144 * initial data transmission if no 145 * valid RTT sample has been acquired, 146 * most likely due to retrans in 3WHS. 147 */ 148 149 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 150 * for local resources. 151 */ 152 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 153 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 154 #define TCP_KEEPALIVE_INTVL (75*HZ) 155 156 #define MAX_TCP_KEEPIDLE 32767 157 #define MAX_TCP_KEEPINTVL 32767 158 #define MAX_TCP_KEEPCNT 127 159 #define MAX_TCP_SYNCNT 127 160 161 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 162 163 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 164 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 165 * after this time. It should be equal 166 * (or greater than) TCP_TIMEWAIT_LEN 167 * to provide reliability equal to one 168 * provided by timewait state. 169 */ 170 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 171 * timestamps. It must be less than 172 * minimal timewait lifetime. 173 */ 174 /* 175 * TCP option 176 */ 177 178 #define TCPOPT_NOP 1 /* Padding */ 179 #define TCPOPT_EOL 0 /* End of options */ 180 #define TCPOPT_MSS 2 /* Segment size negotiating */ 181 #define TCPOPT_WINDOW 3 /* Window scaling */ 182 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 183 #define TCPOPT_SACK 5 /* SACK Block */ 184 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 185 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 186 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 187 #define TCPOPT_EXP 254 /* Experimental */ 188 /* Magic number to be after the option value for sharing TCP 189 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 190 */ 191 #define TCPOPT_FASTOPEN_MAGIC 0xF989 192 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 193 194 /* 195 * TCP option lengths 196 */ 197 198 #define TCPOLEN_MSS 4 199 #define TCPOLEN_WINDOW 3 200 #define TCPOLEN_SACK_PERM 2 201 #define TCPOLEN_TIMESTAMP 10 202 #define TCPOLEN_MD5SIG 18 203 #define TCPOLEN_FASTOPEN_BASE 2 204 #define TCPOLEN_EXP_FASTOPEN_BASE 4 205 #define TCPOLEN_EXP_SMC_BASE 6 206 207 /* But this is what stacks really send out. */ 208 #define TCPOLEN_TSTAMP_ALIGNED 12 209 #define TCPOLEN_WSCALE_ALIGNED 4 210 #define TCPOLEN_SACKPERM_ALIGNED 4 211 #define TCPOLEN_SACK_BASE 2 212 #define TCPOLEN_SACK_BASE_ALIGNED 4 213 #define TCPOLEN_SACK_PERBLOCK 8 214 #define TCPOLEN_MD5SIG_ALIGNED 20 215 #define TCPOLEN_MSS_ALIGNED 4 216 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 217 218 /* Flags in tp->nonagle */ 219 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 220 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 221 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 222 223 /* TCP thin-stream limits */ 224 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 225 226 /* TCP initial congestion window as per rfc6928 */ 227 #define TCP_INIT_CWND 10 228 229 /* Bit Flags for sysctl_tcp_fastopen */ 230 #define TFO_CLIENT_ENABLE 1 231 #define TFO_SERVER_ENABLE 2 232 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 233 234 /* Accept SYN data w/o any cookie option */ 235 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 236 237 /* Force enable TFO on all listeners, i.e., not requiring the 238 * TCP_FASTOPEN socket option. 239 */ 240 #define TFO_SERVER_WO_SOCKOPT1 0x400 241 242 243 /* sysctl variables for tcp */ 244 extern int sysctl_tcp_max_orphans; 245 extern long sysctl_tcp_mem[3]; 246 247 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 248 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 249 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 250 251 extern atomic_long_t tcp_memory_allocated; 252 extern struct percpu_counter tcp_sockets_allocated; 253 extern unsigned long tcp_memory_pressure; 254 255 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 256 static inline bool tcp_under_memory_pressure(const struct sock *sk) 257 { 258 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 259 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 260 return true; 261 262 return tcp_memory_pressure; 263 } 264 /* 265 * The next routines deal with comparing 32 bit unsigned ints 266 * and worry about wraparound (automatic with unsigned arithmetic). 267 */ 268 269 static inline bool before(__u32 seq1, __u32 seq2) 270 { 271 return (__s32)(seq1-seq2) < 0; 272 } 273 #define after(seq2, seq1) before(seq1, seq2) 274 275 /* is s2<=s1<=s3 ? */ 276 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 277 { 278 return seq3 - seq2 >= seq1 - seq2; 279 } 280 281 static inline bool tcp_out_of_memory(struct sock *sk) 282 { 283 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 284 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 285 return true; 286 return false; 287 } 288 289 void sk_forced_mem_schedule(struct sock *sk, int size); 290 291 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 292 { 293 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 294 int orphans = percpu_counter_read_positive(ocp); 295 296 if (orphans << shift > sysctl_tcp_max_orphans) { 297 orphans = percpu_counter_sum_positive(ocp); 298 if (orphans << shift > sysctl_tcp_max_orphans) 299 return true; 300 } 301 return false; 302 } 303 304 bool tcp_check_oom(struct sock *sk, int shift); 305 306 307 extern struct proto tcp_prot; 308 309 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 310 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 311 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 312 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 313 314 void tcp_tasklet_init(void); 315 316 void tcp_v4_err(struct sk_buff *skb, u32); 317 318 void tcp_shutdown(struct sock *sk, int how); 319 320 int tcp_v4_early_demux(struct sk_buff *skb); 321 int tcp_v4_rcv(struct sk_buff *skb); 322 323 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 324 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 325 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 326 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 327 int flags); 328 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 329 size_t size, int flags); 330 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 331 size_t size, int flags); 332 void tcp_release_cb(struct sock *sk); 333 void tcp_wfree(struct sk_buff *skb); 334 void tcp_write_timer_handler(struct sock *sk); 335 void tcp_delack_timer_handler(struct sock *sk); 336 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 337 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 338 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 339 void tcp_rcv_space_adjust(struct sock *sk); 340 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 341 void tcp_twsk_destructor(struct sock *sk); 342 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 343 struct pipe_inode_info *pipe, size_t len, 344 unsigned int flags); 345 346 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 347 static inline void tcp_dec_quickack_mode(struct sock *sk, 348 const unsigned int pkts) 349 { 350 struct inet_connection_sock *icsk = inet_csk(sk); 351 352 if (icsk->icsk_ack.quick) { 353 if (pkts >= icsk->icsk_ack.quick) { 354 icsk->icsk_ack.quick = 0; 355 /* Leaving quickack mode we deflate ATO. */ 356 icsk->icsk_ack.ato = TCP_ATO_MIN; 357 } else 358 icsk->icsk_ack.quick -= pkts; 359 } 360 } 361 362 #define TCP_ECN_OK 1 363 #define TCP_ECN_QUEUE_CWR 2 364 #define TCP_ECN_DEMAND_CWR 4 365 #define TCP_ECN_SEEN 8 366 367 enum tcp_tw_status { 368 TCP_TW_SUCCESS = 0, 369 TCP_TW_RST = 1, 370 TCP_TW_ACK = 2, 371 TCP_TW_SYN = 3 372 }; 373 374 375 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 376 struct sk_buff *skb, 377 const struct tcphdr *th); 378 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 379 struct request_sock *req, bool fastopen, 380 bool *lost_race); 381 int tcp_child_process(struct sock *parent, struct sock *child, 382 struct sk_buff *skb); 383 void tcp_enter_loss(struct sock *sk); 384 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 385 void tcp_clear_retrans(struct tcp_sock *tp); 386 void tcp_update_metrics(struct sock *sk); 387 void tcp_init_metrics(struct sock *sk); 388 void tcp_metrics_init(void); 389 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 390 void tcp_close(struct sock *sk, long timeout); 391 void tcp_init_sock(struct sock *sk); 392 void tcp_init_transfer(struct sock *sk, int bpf_op); 393 __poll_t tcp_poll(struct file *file, struct socket *sock, 394 struct poll_table_struct *wait); 395 int tcp_getsockopt(struct sock *sk, int level, int optname, 396 char __user *optval, int __user *optlen); 397 int tcp_setsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, unsigned int optlen); 399 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 400 char __user *optval, int __user *optlen); 401 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 402 char __user *optval, unsigned int optlen); 403 void tcp_set_keepalive(struct sock *sk, int val); 404 void tcp_syn_ack_timeout(const struct request_sock *req); 405 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 406 int flags, int *addr_len); 407 int tcp_set_rcvlowat(struct sock *sk, int val); 408 void tcp_data_ready(struct sock *sk); 409 int tcp_mmap(struct file *file, struct socket *sock, 410 struct vm_area_struct *vma); 411 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 412 struct tcp_options_received *opt_rx, 413 int estab, struct tcp_fastopen_cookie *foc); 414 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 415 416 /* 417 * TCP v4 functions exported for the inet6 API 418 */ 419 420 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 421 void tcp_v4_mtu_reduced(struct sock *sk); 422 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 423 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 424 struct sock *tcp_create_openreq_child(const struct sock *sk, 425 struct request_sock *req, 426 struct sk_buff *skb); 427 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 428 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 429 struct request_sock *req, 430 struct dst_entry *dst, 431 struct request_sock *req_unhash, 432 bool *own_req); 433 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 434 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 435 int tcp_connect(struct sock *sk); 436 enum tcp_synack_type { 437 TCP_SYNACK_NORMAL, 438 TCP_SYNACK_FASTOPEN, 439 TCP_SYNACK_COOKIE, 440 }; 441 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 442 struct request_sock *req, 443 struct tcp_fastopen_cookie *foc, 444 enum tcp_synack_type synack_type); 445 int tcp_disconnect(struct sock *sk, int flags); 446 447 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 448 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 449 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 450 451 /* From syncookies.c */ 452 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 453 struct request_sock *req, 454 struct dst_entry *dst, u32 tsoff); 455 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 456 u32 cookie); 457 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 458 #ifdef CONFIG_SYN_COOKIES 459 460 /* Syncookies use a monotonic timer which increments every 60 seconds. 461 * This counter is used both as a hash input and partially encoded into 462 * the cookie value. A cookie is only validated further if the delta 463 * between the current counter value and the encoded one is less than this, 464 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 465 * the counter advances immediately after a cookie is generated). 466 */ 467 #define MAX_SYNCOOKIE_AGE 2 468 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 469 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 470 471 /* syncookies: remember time of last synqueue overflow 472 * But do not dirty this field too often (once per second is enough) 473 * It is racy as we do not hold a lock, but race is very minor. 474 */ 475 static inline void tcp_synq_overflow(const struct sock *sk) 476 { 477 unsigned int last_overflow; 478 unsigned int now = jiffies; 479 480 if (sk->sk_reuseport) { 481 struct sock_reuseport *reuse; 482 483 reuse = rcu_dereference(sk->sk_reuseport_cb); 484 if (likely(reuse)) { 485 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 486 if (time_after32(now, last_overflow + HZ)) 487 WRITE_ONCE(reuse->synq_overflow_ts, now); 488 return; 489 } 490 } 491 492 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 493 if (time_after32(now, last_overflow + HZ)) 494 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 495 } 496 497 /* syncookies: no recent synqueue overflow on this listening socket? */ 498 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 499 { 500 unsigned int last_overflow; 501 unsigned int now = jiffies; 502 503 if (sk->sk_reuseport) { 504 struct sock_reuseport *reuse; 505 506 reuse = rcu_dereference(sk->sk_reuseport_cb); 507 if (likely(reuse)) { 508 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 509 return time_after32(now, last_overflow + 510 TCP_SYNCOOKIE_VALID); 511 } 512 } 513 514 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 515 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID); 516 } 517 518 static inline u32 tcp_cookie_time(void) 519 { 520 u64 val = get_jiffies_64(); 521 522 do_div(val, TCP_SYNCOOKIE_PERIOD); 523 return val; 524 } 525 526 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 527 u16 *mssp); 528 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 529 u64 cookie_init_timestamp(struct request_sock *req); 530 bool cookie_timestamp_decode(const struct net *net, 531 struct tcp_options_received *opt); 532 bool cookie_ecn_ok(const struct tcp_options_received *opt, 533 const struct net *net, const struct dst_entry *dst); 534 535 /* From net/ipv6/syncookies.c */ 536 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 537 u32 cookie); 538 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 539 540 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 541 const struct tcphdr *th, u16 *mssp); 542 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 543 #endif 544 /* tcp_output.c */ 545 546 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 547 int nonagle); 548 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 549 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 550 void tcp_retransmit_timer(struct sock *sk); 551 void tcp_xmit_retransmit_queue(struct sock *); 552 void tcp_simple_retransmit(struct sock *); 553 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 554 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 555 enum tcp_queue { 556 TCP_FRAG_IN_WRITE_QUEUE, 557 TCP_FRAG_IN_RTX_QUEUE, 558 }; 559 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 560 struct sk_buff *skb, u32 len, 561 unsigned int mss_now, gfp_t gfp); 562 563 void tcp_send_probe0(struct sock *); 564 void tcp_send_partial(struct sock *); 565 int tcp_write_wakeup(struct sock *, int mib); 566 void tcp_send_fin(struct sock *sk); 567 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 568 int tcp_send_synack(struct sock *); 569 void tcp_push_one(struct sock *, unsigned int mss_now); 570 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 571 void tcp_send_ack(struct sock *sk); 572 void tcp_send_delayed_ack(struct sock *sk); 573 void tcp_send_loss_probe(struct sock *sk); 574 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 575 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 576 const struct sk_buff *next_skb); 577 578 /* tcp_input.c */ 579 void tcp_rearm_rto(struct sock *sk); 580 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 581 void tcp_reset(struct sock *sk); 582 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 583 void tcp_fin(struct sock *sk); 584 585 /* tcp_timer.c */ 586 void tcp_init_xmit_timers(struct sock *); 587 static inline void tcp_clear_xmit_timers(struct sock *sk) 588 { 589 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 590 __sock_put(sk); 591 592 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 593 __sock_put(sk); 594 595 inet_csk_clear_xmit_timers(sk); 596 } 597 598 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 599 unsigned int tcp_current_mss(struct sock *sk); 600 601 /* Bound MSS / TSO packet size with the half of the window */ 602 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 603 { 604 int cutoff; 605 606 /* When peer uses tiny windows, there is no use in packetizing 607 * to sub-MSS pieces for the sake of SWS or making sure there 608 * are enough packets in the pipe for fast recovery. 609 * 610 * On the other hand, for extremely large MSS devices, handling 611 * smaller than MSS windows in this way does make sense. 612 */ 613 if (tp->max_window > TCP_MSS_DEFAULT) 614 cutoff = (tp->max_window >> 1); 615 else 616 cutoff = tp->max_window; 617 618 if (cutoff && pktsize > cutoff) 619 return max_t(int, cutoff, 68U - tp->tcp_header_len); 620 else 621 return pktsize; 622 } 623 624 /* tcp.c */ 625 void tcp_get_info(struct sock *, struct tcp_info *); 626 627 /* Read 'sendfile()'-style from a TCP socket */ 628 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 629 sk_read_actor_t recv_actor); 630 631 void tcp_initialize_rcv_mss(struct sock *sk); 632 633 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 634 int tcp_mss_to_mtu(struct sock *sk, int mss); 635 void tcp_mtup_init(struct sock *sk); 636 void tcp_init_buffer_space(struct sock *sk); 637 638 static inline void tcp_bound_rto(const struct sock *sk) 639 { 640 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 641 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 642 } 643 644 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 645 { 646 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 647 } 648 649 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 650 { 651 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 652 ntohl(TCP_FLAG_ACK) | 653 snd_wnd); 654 } 655 656 static inline void tcp_fast_path_on(struct tcp_sock *tp) 657 { 658 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 659 } 660 661 static inline void tcp_fast_path_check(struct sock *sk) 662 { 663 struct tcp_sock *tp = tcp_sk(sk); 664 665 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 666 tp->rcv_wnd && 667 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 668 !tp->urg_data) 669 tcp_fast_path_on(tp); 670 } 671 672 /* Compute the actual rto_min value */ 673 static inline u32 tcp_rto_min(struct sock *sk) 674 { 675 const struct dst_entry *dst = __sk_dst_get(sk); 676 u32 rto_min = TCP_RTO_MIN; 677 678 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 679 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 680 return rto_min; 681 } 682 683 static inline u32 tcp_rto_min_us(struct sock *sk) 684 { 685 return jiffies_to_usecs(tcp_rto_min(sk)); 686 } 687 688 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 689 { 690 return dst_metric_locked(dst, RTAX_CC_ALGO); 691 } 692 693 /* Minimum RTT in usec. ~0 means not available. */ 694 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 695 { 696 return minmax_get(&tp->rtt_min); 697 } 698 699 /* Compute the actual receive window we are currently advertising. 700 * Rcv_nxt can be after the window if our peer push more data 701 * than the offered window. 702 */ 703 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 704 { 705 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 706 707 if (win < 0) 708 win = 0; 709 return (u32) win; 710 } 711 712 /* Choose a new window, without checks for shrinking, and without 713 * scaling applied to the result. The caller does these things 714 * if necessary. This is a "raw" window selection. 715 */ 716 u32 __tcp_select_window(struct sock *sk); 717 718 void tcp_send_window_probe(struct sock *sk); 719 720 /* TCP uses 32bit jiffies to save some space. 721 * Note that this is different from tcp_time_stamp, which 722 * historically has been the same until linux-4.13. 723 */ 724 #define tcp_jiffies32 ((u32)jiffies) 725 726 /* 727 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 728 * It is no longer tied to jiffies, but to 1 ms clock. 729 * Note: double check if you want to use tcp_jiffies32 instead of this. 730 */ 731 #define TCP_TS_HZ 1000 732 733 static inline u64 tcp_clock_ns(void) 734 { 735 return local_clock(); 736 } 737 738 static inline u64 tcp_clock_us(void) 739 { 740 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 741 } 742 743 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 744 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 745 { 746 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 747 } 748 749 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 750 static inline u32 tcp_time_stamp_raw(void) 751 { 752 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 753 } 754 755 756 /* Refresh 1us clock of a TCP socket, 757 * ensuring monotically increasing values. 758 */ 759 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 760 { 761 u64 val = tcp_clock_us(); 762 763 if (val > tp->tcp_mstamp) 764 tp->tcp_mstamp = val; 765 } 766 767 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 768 { 769 return max_t(s64, t1 - t0, 0); 770 } 771 772 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 773 { 774 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 775 } 776 777 778 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 779 780 #define TCPHDR_FIN 0x01 781 #define TCPHDR_SYN 0x02 782 #define TCPHDR_RST 0x04 783 #define TCPHDR_PSH 0x08 784 #define TCPHDR_ACK 0x10 785 #define TCPHDR_URG 0x20 786 #define TCPHDR_ECE 0x40 787 #define TCPHDR_CWR 0x80 788 789 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 790 791 /* This is what the send packet queuing engine uses to pass 792 * TCP per-packet control information to the transmission code. 793 * We also store the host-order sequence numbers in here too. 794 * This is 44 bytes if IPV6 is enabled. 795 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 796 */ 797 struct tcp_skb_cb { 798 __u32 seq; /* Starting sequence number */ 799 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 800 union { 801 /* Note : tcp_tw_isn is used in input path only 802 * (isn chosen by tcp_timewait_state_process()) 803 * 804 * tcp_gso_segs/size are used in write queue only, 805 * cf tcp_skb_pcount()/tcp_skb_mss() 806 */ 807 __u32 tcp_tw_isn; 808 struct { 809 u16 tcp_gso_segs; 810 u16 tcp_gso_size; 811 }; 812 }; 813 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 814 815 __u8 sacked; /* State flags for SACK. */ 816 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 817 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 818 #define TCPCB_LOST 0x04 /* SKB is lost */ 819 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 820 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 821 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 822 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 823 TCPCB_REPAIRED) 824 825 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 826 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 827 eor:1, /* Is skb MSG_EOR marked? */ 828 has_rxtstamp:1, /* SKB has a RX timestamp */ 829 unused:5; 830 __u32 ack_seq; /* Sequence number ACK'd */ 831 union { 832 struct { 833 /* There is space for up to 24 bytes */ 834 __u32 in_flight:30,/* Bytes in flight at transmit */ 835 is_app_limited:1, /* cwnd not fully used? */ 836 unused:1; 837 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 838 __u32 delivered; 839 /* start of send pipeline phase */ 840 u64 first_tx_mstamp; 841 /* when we reached the "delivered" count */ 842 u64 delivered_mstamp; 843 } tx; /* only used for outgoing skbs */ 844 union { 845 struct inet_skb_parm h4; 846 #if IS_ENABLED(CONFIG_IPV6) 847 struct inet6_skb_parm h6; 848 #endif 849 } header; /* For incoming skbs */ 850 struct { 851 __u32 flags; 852 struct sock *sk_redir; 853 void *data_end; 854 } bpf; 855 }; 856 }; 857 858 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 859 860 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 861 { 862 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 863 } 864 865 #if IS_ENABLED(CONFIG_IPV6) 866 /* This is the variant of inet6_iif() that must be used by TCP, 867 * as TCP moves IP6CB into a different location in skb->cb[] 868 */ 869 static inline int tcp_v6_iif(const struct sk_buff *skb) 870 { 871 return TCP_SKB_CB(skb)->header.h6.iif; 872 } 873 874 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 875 { 876 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 877 878 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 879 } 880 881 /* TCP_SKB_CB reference means this can not be used from early demux */ 882 static inline int tcp_v6_sdif(const struct sk_buff *skb) 883 { 884 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 885 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 886 return TCP_SKB_CB(skb)->header.h6.iif; 887 #endif 888 return 0; 889 } 890 #endif 891 892 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 893 { 894 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 895 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 896 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 897 return true; 898 #endif 899 return false; 900 } 901 902 /* TCP_SKB_CB reference means this can not be used from early demux */ 903 static inline int tcp_v4_sdif(struct sk_buff *skb) 904 { 905 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 906 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 907 return TCP_SKB_CB(skb)->header.h4.iif; 908 #endif 909 return 0; 910 } 911 912 /* Due to TSO, an SKB can be composed of multiple actual 913 * packets. To keep these tracked properly, we use this. 914 */ 915 static inline int tcp_skb_pcount(const struct sk_buff *skb) 916 { 917 return TCP_SKB_CB(skb)->tcp_gso_segs; 918 } 919 920 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 921 { 922 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 923 } 924 925 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 926 { 927 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 928 } 929 930 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 931 static inline int tcp_skb_mss(const struct sk_buff *skb) 932 { 933 return TCP_SKB_CB(skb)->tcp_gso_size; 934 } 935 936 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 937 { 938 return likely(!TCP_SKB_CB(skb)->eor); 939 } 940 941 /* Events passed to congestion control interface */ 942 enum tcp_ca_event { 943 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 944 CA_EVENT_CWND_RESTART, /* congestion window restart */ 945 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 946 CA_EVENT_LOSS, /* loss timeout */ 947 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 948 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 949 }; 950 951 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 952 enum tcp_ca_ack_event_flags { 953 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 954 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 955 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 956 }; 957 958 /* 959 * Interface for adding new TCP congestion control handlers 960 */ 961 #define TCP_CA_NAME_MAX 16 962 #define TCP_CA_MAX 128 963 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 964 965 #define TCP_CA_UNSPEC 0 966 967 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 968 #define TCP_CONG_NON_RESTRICTED 0x1 969 /* Requires ECN/ECT set on all packets */ 970 #define TCP_CONG_NEEDS_ECN 0x2 971 972 union tcp_cc_info; 973 974 struct ack_sample { 975 u32 pkts_acked; 976 s32 rtt_us; 977 u32 in_flight; 978 }; 979 980 /* A rate sample measures the number of (original/retransmitted) data 981 * packets delivered "delivered" over an interval of time "interval_us". 982 * The tcp_rate.c code fills in the rate sample, and congestion 983 * control modules that define a cong_control function to run at the end 984 * of ACK processing can optionally chose to consult this sample when 985 * setting cwnd and pacing rate. 986 * A sample is invalid if "delivered" or "interval_us" is negative. 987 */ 988 struct rate_sample { 989 u64 prior_mstamp; /* starting timestamp for interval */ 990 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 991 s32 delivered; /* number of packets delivered over interval */ 992 long interval_us; /* time for tp->delivered to incr "delivered" */ 993 u32 snd_interval_us; /* snd interval for delivered packets */ 994 u32 rcv_interval_us; /* rcv interval for delivered packets */ 995 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 996 int losses; /* number of packets marked lost upon ACK */ 997 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 998 u32 prior_in_flight; /* in flight before this ACK */ 999 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1000 bool is_retrans; /* is sample from retransmission? */ 1001 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1002 }; 1003 1004 struct tcp_congestion_ops { 1005 struct list_head list; 1006 u32 key; 1007 u32 flags; 1008 1009 /* initialize private data (optional) */ 1010 void (*init)(struct sock *sk); 1011 /* cleanup private data (optional) */ 1012 void (*release)(struct sock *sk); 1013 1014 /* return slow start threshold (required) */ 1015 u32 (*ssthresh)(struct sock *sk); 1016 /* do new cwnd calculation (required) */ 1017 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1018 /* call before changing ca_state (optional) */ 1019 void (*set_state)(struct sock *sk, u8 new_state); 1020 /* call when cwnd event occurs (optional) */ 1021 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1022 /* call when ack arrives (optional) */ 1023 void (*in_ack_event)(struct sock *sk, u32 flags); 1024 /* new value of cwnd after loss (required) */ 1025 u32 (*undo_cwnd)(struct sock *sk); 1026 /* hook for packet ack accounting (optional) */ 1027 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1028 /* override sysctl_tcp_min_tso_segs */ 1029 u32 (*min_tso_segs)(struct sock *sk); 1030 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1031 u32 (*sndbuf_expand)(struct sock *sk); 1032 /* call when packets are delivered to update cwnd and pacing rate, 1033 * after all the ca_state processing. (optional) 1034 */ 1035 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1036 /* get info for inet_diag (optional) */ 1037 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1038 union tcp_cc_info *info); 1039 1040 char name[TCP_CA_NAME_MAX]; 1041 struct module *owner; 1042 }; 1043 1044 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1045 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1046 1047 void tcp_assign_congestion_control(struct sock *sk); 1048 void tcp_init_congestion_control(struct sock *sk); 1049 void tcp_cleanup_congestion_control(struct sock *sk); 1050 int tcp_set_default_congestion_control(struct net *net, const char *name); 1051 void tcp_get_default_congestion_control(struct net *net, char *name); 1052 void tcp_get_available_congestion_control(char *buf, size_t len); 1053 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1054 int tcp_set_allowed_congestion_control(char *allowed); 1055 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1056 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1057 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1058 1059 u32 tcp_reno_ssthresh(struct sock *sk); 1060 u32 tcp_reno_undo_cwnd(struct sock *sk); 1061 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1062 extern struct tcp_congestion_ops tcp_reno; 1063 1064 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1065 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1066 #ifdef CONFIG_INET 1067 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1068 #else 1069 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1070 { 1071 return NULL; 1072 } 1073 #endif 1074 1075 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1076 { 1077 const struct inet_connection_sock *icsk = inet_csk(sk); 1078 1079 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1080 } 1081 1082 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1083 { 1084 struct inet_connection_sock *icsk = inet_csk(sk); 1085 1086 if (icsk->icsk_ca_ops->set_state) 1087 icsk->icsk_ca_ops->set_state(sk, ca_state); 1088 icsk->icsk_ca_state = ca_state; 1089 } 1090 1091 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1092 { 1093 const struct inet_connection_sock *icsk = inet_csk(sk); 1094 1095 if (icsk->icsk_ca_ops->cwnd_event) 1096 icsk->icsk_ca_ops->cwnd_event(sk, event); 1097 } 1098 1099 /* From tcp_rate.c */ 1100 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1101 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1102 struct rate_sample *rs); 1103 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1104 bool is_sack_reneg, struct rate_sample *rs); 1105 void tcp_rate_check_app_limited(struct sock *sk); 1106 1107 /* These functions determine how the current flow behaves in respect of SACK 1108 * handling. SACK is negotiated with the peer, and therefore it can vary 1109 * between different flows. 1110 * 1111 * tcp_is_sack - SACK enabled 1112 * tcp_is_reno - No SACK 1113 */ 1114 static inline int tcp_is_sack(const struct tcp_sock *tp) 1115 { 1116 return tp->rx_opt.sack_ok; 1117 } 1118 1119 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1120 { 1121 return !tcp_is_sack(tp); 1122 } 1123 1124 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1125 { 1126 return tp->sacked_out + tp->lost_out; 1127 } 1128 1129 /* This determines how many packets are "in the network" to the best 1130 * of our knowledge. In many cases it is conservative, but where 1131 * detailed information is available from the receiver (via SACK 1132 * blocks etc.) we can make more aggressive calculations. 1133 * 1134 * Use this for decisions involving congestion control, use just 1135 * tp->packets_out to determine if the send queue is empty or not. 1136 * 1137 * Read this equation as: 1138 * 1139 * "Packets sent once on transmission queue" MINUS 1140 * "Packets left network, but not honestly ACKed yet" PLUS 1141 * "Packets fast retransmitted" 1142 */ 1143 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1144 { 1145 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1146 } 1147 1148 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1149 1150 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1151 { 1152 return tp->snd_cwnd < tp->snd_ssthresh; 1153 } 1154 1155 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1156 { 1157 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1158 } 1159 1160 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1161 { 1162 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1163 (1 << inet_csk(sk)->icsk_ca_state); 1164 } 1165 1166 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1167 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1168 * ssthresh. 1169 */ 1170 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1171 { 1172 const struct tcp_sock *tp = tcp_sk(sk); 1173 1174 if (tcp_in_cwnd_reduction(sk)) 1175 return tp->snd_ssthresh; 1176 else 1177 return max(tp->snd_ssthresh, 1178 ((tp->snd_cwnd >> 1) + 1179 (tp->snd_cwnd >> 2))); 1180 } 1181 1182 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1183 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1184 1185 void tcp_enter_cwr(struct sock *sk); 1186 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1187 1188 /* The maximum number of MSS of available cwnd for which TSO defers 1189 * sending if not using sysctl_tcp_tso_win_divisor. 1190 */ 1191 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1192 { 1193 return 3; 1194 } 1195 1196 /* Returns end sequence number of the receiver's advertised window */ 1197 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1198 { 1199 return tp->snd_una + tp->snd_wnd; 1200 } 1201 1202 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1203 * flexible approach. The RFC suggests cwnd should not be raised unless 1204 * it was fully used previously. And that's exactly what we do in 1205 * congestion avoidance mode. But in slow start we allow cwnd to grow 1206 * as long as the application has used half the cwnd. 1207 * Example : 1208 * cwnd is 10 (IW10), but application sends 9 frames. 1209 * We allow cwnd to reach 18 when all frames are ACKed. 1210 * This check is safe because it's as aggressive as slow start which already 1211 * risks 100% overshoot. The advantage is that we discourage application to 1212 * either send more filler packets or data to artificially blow up the cwnd 1213 * usage, and allow application-limited process to probe bw more aggressively. 1214 */ 1215 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1216 { 1217 const struct tcp_sock *tp = tcp_sk(sk); 1218 1219 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1220 if (tcp_in_slow_start(tp)) 1221 return tp->snd_cwnd < 2 * tp->max_packets_out; 1222 1223 return tp->is_cwnd_limited; 1224 } 1225 1226 /* BBR congestion control needs pacing. 1227 * Same remark for SO_MAX_PACING_RATE. 1228 * sch_fq packet scheduler is efficiently handling pacing, 1229 * but is not always installed/used. 1230 * Return true if TCP stack should pace packets itself. 1231 */ 1232 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1233 { 1234 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1235 } 1236 1237 /* Something is really bad, we could not queue an additional packet, 1238 * because qdisc is full or receiver sent a 0 window. 1239 * We do not want to add fuel to the fire, or abort too early, 1240 * so make sure the timer we arm now is at least 200ms in the future, 1241 * regardless of current icsk_rto value (as it could be ~2ms) 1242 */ 1243 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1244 { 1245 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1246 } 1247 1248 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1249 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1250 unsigned long max_when) 1251 { 1252 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1253 1254 return (unsigned long)min_t(u64, when, max_when); 1255 } 1256 1257 static inline void tcp_check_probe_timer(struct sock *sk) 1258 { 1259 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1260 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1261 tcp_probe0_base(sk), TCP_RTO_MAX); 1262 } 1263 1264 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1265 { 1266 tp->snd_wl1 = seq; 1267 } 1268 1269 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1270 { 1271 tp->snd_wl1 = seq; 1272 } 1273 1274 /* 1275 * Calculate(/check) TCP checksum 1276 */ 1277 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1278 __be32 daddr, __wsum base) 1279 { 1280 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1281 } 1282 1283 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1284 { 1285 return __skb_checksum_complete(skb); 1286 } 1287 1288 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1289 { 1290 return !skb_csum_unnecessary(skb) && 1291 __tcp_checksum_complete(skb); 1292 } 1293 1294 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1295 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1296 1297 #undef STATE_TRACE 1298 1299 #ifdef STATE_TRACE 1300 static const char *statename[]={ 1301 "Unused","Established","Syn Sent","Syn Recv", 1302 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1303 "Close Wait","Last ACK","Listen","Closing" 1304 }; 1305 #endif 1306 void tcp_set_state(struct sock *sk, int state); 1307 1308 void tcp_done(struct sock *sk); 1309 1310 int tcp_abort(struct sock *sk, int err); 1311 1312 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1313 { 1314 rx_opt->dsack = 0; 1315 rx_opt->num_sacks = 0; 1316 } 1317 1318 u32 tcp_default_init_rwnd(u32 mss); 1319 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1320 1321 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1322 { 1323 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1324 struct tcp_sock *tp = tcp_sk(sk); 1325 s32 delta; 1326 1327 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1328 ca_ops->cong_control) 1329 return; 1330 delta = tcp_jiffies32 - tp->lsndtime; 1331 if (delta > inet_csk(sk)->icsk_rto) 1332 tcp_cwnd_restart(sk, delta); 1333 } 1334 1335 /* Determine a window scaling and initial window to offer. */ 1336 void tcp_select_initial_window(const struct sock *sk, int __space, 1337 __u32 mss, __u32 *rcv_wnd, 1338 __u32 *window_clamp, int wscale_ok, 1339 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1340 1341 static inline int tcp_win_from_space(const struct sock *sk, int space) 1342 { 1343 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1344 1345 return tcp_adv_win_scale <= 0 ? 1346 (space>>(-tcp_adv_win_scale)) : 1347 space - (space>>tcp_adv_win_scale); 1348 } 1349 1350 /* Note: caller must be prepared to deal with negative returns */ 1351 static inline int tcp_space(const struct sock *sk) 1352 { 1353 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1354 atomic_read(&sk->sk_rmem_alloc)); 1355 } 1356 1357 static inline int tcp_full_space(const struct sock *sk) 1358 { 1359 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1360 } 1361 1362 extern void tcp_openreq_init_rwin(struct request_sock *req, 1363 const struct sock *sk_listener, 1364 const struct dst_entry *dst); 1365 1366 void tcp_enter_memory_pressure(struct sock *sk); 1367 void tcp_leave_memory_pressure(struct sock *sk); 1368 1369 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1370 { 1371 struct net *net = sock_net((struct sock *)tp); 1372 1373 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1374 } 1375 1376 static inline int keepalive_time_when(const struct tcp_sock *tp) 1377 { 1378 struct net *net = sock_net((struct sock *)tp); 1379 1380 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1381 } 1382 1383 static inline int keepalive_probes(const struct tcp_sock *tp) 1384 { 1385 struct net *net = sock_net((struct sock *)tp); 1386 1387 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1388 } 1389 1390 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1391 { 1392 const struct inet_connection_sock *icsk = &tp->inet_conn; 1393 1394 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1395 tcp_jiffies32 - tp->rcv_tstamp); 1396 } 1397 1398 static inline int tcp_fin_time(const struct sock *sk) 1399 { 1400 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1401 const int rto = inet_csk(sk)->icsk_rto; 1402 1403 if (fin_timeout < (rto << 2) - (rto >> 1)) 1404 fin_timeout = (rto << 2) - (rto >> 1); 1405 1406 return fin_timeout; 1407 } 1408 1409 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1410 int paws_win) 1411 { 1412 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1413 return true; 1414 if (unlikely(!time_before32(ktime_get_seconds(), 1415 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1416 return true; 1417 /* 1418 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1419 * then following tcp messages have valid values. Ignore 0 value, 1420 * or else 'negative' tsval might forbid us to accept their packets. 1421 */ 1422 if (!rx_opt->ts_recent) 1423 return true; 1424 return false; 1425 } 1426 1427 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1428 int rst) 1429 { 1430 if (tcp_paws_check(rx_opt, 0)) 1431 return false; 1432 1433 /* RST segments are not recommended to carry timestamp, 1434 and, if they do, it is recommended to ignore PAWS because 1435 "their cleanup function should take precedence over timestamps." 1436 Certainly, it is mistake. It is necessary to understand the reasons 1437 of this constraint to relax it: if peer reboots, clock may go 1438 out-of-sync and half-open connections will not be reset. 1439 Actually, the problem would be not existing if all 1440 the implementations followed draft about maintaining clock 1441 via reboots. Linux-2.2 DOES NOT! 1442 1443 However, we can relax time bounds for RST segments to MSL. 1444 */ 1445 if (rst && !time_before32(ktime_get_seconds(), 1446 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1447 return false; 1448 return true; 1449 } 1450 1451 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1452 int mib_idx, u32 *last_oow_ack_time); 1453 1454 static inline void tcp_mib_init(struct net *net) 1455 { 1456 /* See RFC 2012 */ 1457 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1458 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1459 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1460 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1461 } 1462 1463 /* from STCP */ 1464 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1465 { 1466 tp->lost_skb_hint = NULL; 1467 } 1468 1469 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1470 { 1471 tcp_clear_retrans_hints_partial(tp); 1472 tp->retransmit_skb_hint = NULL; 1473 } 1474 1475 union tcp_md5_addr { 1476 struct in_addr a4; 1477 #if IS_ENABLED(CONFIG_IPV6) 1478 struct in6_addr a6; 1479 #endif 1480 }; 1481 1482 /* - key database */ 1483 struct tcp_md5sig_key { 1484 struct hlist_node node; 1485 u8 keylen; 1486 u8 family; /* AF_INET or AF_INET6 */ 1487 union tcp_md5_addr addr; 1488 u8 prefixlen; 1489 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1490 struct rcu_head rcu; 1491 }; 1492 1493 /* - sock block */ 1494 struct tcp_md5sig_info { 1495 struct hlist_head head; 1496 struct rcu_head rcu; 1497 }; 1498 1499 /* - pseudo header */ 1500 struct tcp4_pseudohdr { 1501 __be32 saddr; 1502 __be32 daddr; 1503 __u8 pad; 1504 __u8 protocol; 1505 __be16 len; 1506 }; 1507 1508 struct tcp6_pseudohdr { 1509 struct in6_addr saddr; 1510 struct in6_addr daddr; 1511 __be32 len; 1512 __be32 protocol; /* including padding */ 1513 }; 1514 1515 union tcp_md5sum_block { 1516 struct tcp4_pseudohdr ip4; 1517 #if IS_ENABLED(CONFIG_IPV6) 1518 struct tcp6_pseudohdr ip6; 1519 #endif 1520 }; 1521 1522 /* - pool: digest algorithm, hash description and scratch buffer */ 1523 struct tcp_md5sig_pool { 1524 struct ahash_request *md5_req; 1525 void *scratch; 1526 }; 1527 1528 /* - functions */ 1529 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1530 const struct sock *sk, const struct sk_buff *skb); 1531 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1532 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1533 gfp_t gfp); 1534 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1535 int family, u8 prefixlen); 1536 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1537 const struct sock *addr_sk); 1538 1539 #ifdef CONFIG_TCP_MD5SIG 1540 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1541 const union tcp_md5_addr *addr, 1542 int family); 1543 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1544 #else 1545 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1546 const union tcp_md5_addr *addr, 1547 int family) 1548 { 1549 return NULL; 1550 } 1551 #define tcp_twsk_md5_key(twsk) NULL 1552 #endif 1553 1554 bool tcp_alloc_md5sig_pool(void); 1555 1556 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1557 static inline void tcp_put_md5sig_pool(void) 1558 { 1559 local_bh_enable(); 1560 } 1561 1562 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1563 unsigned int header_len); 1564 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1565 const struct tcp_md5sig_key *key); 1566 1567 /* From tcp_fastopen.c */ 1568 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1569 struct tcp_fastopen_cookie *cookie); 1570 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1571 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1572 u16 try_exp); 1573 struct tcp_fastopen_request { 1574 /* Fast Open cookie. Size 0 means a cookie request */ 1575 struct tcp_fastopen_cookie cookie; 1576 struct msghdr *data; /* data in MSG_FASTOPEN */ 1577 size_t size; 1578 int copied; /* queued in tcp_connect() */ 1579 }; 1580 void tcp_free_fastopen_req(struct tcp_sock *tp); 1581 void tcp_fastopen_destroy_cipher(struct sock *sk); 1582 void tcp_fastopen_ctx_destroy(struct net *net); 1583 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1584 void *key, unsigned int len); 1585 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1586 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1587 struct request_sock *req, 1588 struct tcp_fastopen_cookie *foc, 1589 const struct dst_entry *dst); 1590 void tcp_fastopen_init_key_once(struct net *net); 1591 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1592 struct tcp_fastopen_cookie *cookie); 1593 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1594 #define TCP_FASTOPEN_KEY_LENGTH 16 1595 1596 /* Fastopen key context */ 1597 struct tcp_fastopen_context { 1598 struct crypto_cipher *tfm; 1599 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1600 struct rcu_head rcu; 1601 }; 1602 1603 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1604 void tcp_fastopen_active_disable(struct sock *sk); 1605 bool tcp_fastopen_active_should_disable(struct sock *sk); 1606 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1607 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1608 1609 /* Latencies incurred by various limits for a sender. They are 1610 * chronograph-like stats that are mutually exclusive. 1611 */ 1612 enum tcp_chrono { 1613 TCP_CHRONO_UNSPEC, 1614 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1615 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1616 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1617 __TCP_CHRONO_MAX, 1618 }; 1619 1620 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1621 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1622 1623 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1624 * the same memory storage than skb->destructor/_skb_refdst 1625 */ 1626 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1627 { 1628 skb->destructor = NULL; 1629 skb->_skb_refdst = 0UL; 1630 } 1631 1632 #define tcp_skb_tsorted_save(skb) { \ 1633 unsigned long _save = skb->_skb_refdst; \ 1634 skb->_skb_refdst = 0UL; 1635 1636 #define tcp_skb_tsorted_restore(skb) \ 1637 skb->_skb_refdst = _save; \ 1638 } 1639 1640 void tcp_write_queue_purge(struct sock *sk); 1641 1642 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1643 { 1644 return skb_rb_first(&sk->tcp_rtx_queue); 1645 } 1646 1647 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1648 { 1649 return skb_peek(&sk->sk_write_queue); 1650 } 1651 1652 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1653 { 1654 return skb_peek_tail(&sk->sk_write_queue); 1655 } 1656 1657 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1658 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1659 1660 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1661 { 1662 return skb_peek(&sk->sk_write_queue); 1663 } 1664 1665 static inline bool tcp_skb_is_last(const struct sock *sk, 1666 const struct sk_buff *skb) 1667 { 1668 return skb_queue_is_last(&sk->sk_write_queue, skb); 1669 } 1670 1671 static inline bool tcp_write_queue_empty(const struct sock *sk) 1672 { 1673 return skb_queue_empty(&sk->sk_write_queue); 1674 } 1675 1676 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1677 { 1678 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1679 } 1680 1681 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1682 { 1683 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1684 } 1685 1686 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1687 { 1688 if (tcp_write_queue_empty(sk)) 1689 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1690 } 1691 1692 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1693 { 1694 __skb_queue_tail(&sk->sk_write_queue, skb); 1695 } 1696 1697 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1698 { 1699 __tcp_add_write_queue_tail(sk, skb); 1700 1701 /* Queue it, remembering where we must start sending. */ 1702 if (sk->sk_write_queue.next == skb) 1703 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1704 } 1705 1706 /* Insert new before skb on the write queue of sk. */ 1707 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1708 struct sk_buff *skb, 1709 struct sock *sk) 1710 { 1711 __skb_queue_before(&sk->sk_write_queue, skb, new); 1712 } 1713 1714 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1715 { 1716 tcp_skb_tsorted_anchor_cleanup(skb); 1717 __skb_unlink(skb, &sk->sk_write_queue); 1718 } 1719 1720 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1721 1722 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1723 { 1724 tcp_skb_tsorted_anchor_cleanup(skb); 1725 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1726 } 1727 1728 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1729 { 1730 list_del(&skb->tcp_tsorted_anchor); 1731 tcp_rtx_queue_unlink(skb, sk); 1732 sk_wmem_free_skb(sk, skb); 1733 } 1734 1735 static inline void tcp_push_pending_frames(struct sock *sk) 1736 { 1737 if (tcp_send_head(sk)) { 1738 struct tcp_sock *tp = tcp_sk(sk); 1739 1740 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1741 } 1742 } 1743 1744 /* Start sequence of the skb just after the highest skb with SACKed 1745 * bit, valid only if sacked_out > 0 or when the caller has ensured 1746 * validity by itself. 1747 */ 1748 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1749 { 1750 if (!tp->sacked_out) 1751 return tp->snd_una; 1752 1753 if (tp->highest_sack == NULL) 1754 return tp->snd_nxt; 1755 1756 return TCP_SKB_CB(tp->highest_sack)->seq; 1757 } 1758 1759 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1760 { 1761 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1762 } 1763 1764 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1765 { 1766 return tcp_sk(sk)->highest_sack; 1767 } 1768 1769 static inline void tcp_highest_sack_reset(struct sock *sk) 1770 { 1771 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1772 } 1773 1774 /* Called when old skb is about to be deleted and replaced by new skb */ 1775 static inline void tcp_highest_sack_replace(struct sock *sk, 1776 struct sk_buff *old, 1777 struct sk_buff *new) 1778 { 1779 if (old == tcp_highest_sack(sk)) 1780 tcp_sk(sk)->highest_sack = new; 1781 } 1782 1783 /* This helper checks if socket has IP_TRANSPARENT set */ 1784 static inline bool inet_sk_transparent(const struct sock *sk) 1785 { 1786 switch (sk->sk_state) { 1787 case TCP_TIME_WAIT: 1788 return inet_twsk(sk)->tw_transparent; 1789 case TCP_NEW_SYN_RECV: 1790 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1791 } 1792 return inet_sk(sk)->transparent; 1793 } 1794 1795 /* Determines whether this is a thin stream (which may suffer from 1796 * increased latency). Used to trigger latency-reducing mechanisms. 1797 */ 1798 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1799 { 1800 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1801 } 1802 1803 /* /proc */ 1804 enum tcp_seq_states { 1805 TCP_SEQ_STATE_LISTENING, 1806 TCP_SEQ_STATE_ESTABLISHED, 1807 }; 1808 1809 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1810 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1811 void tcp_seq_stop(struct seq_file *seq, void *v); 1812 1813 struct tcp_seq_afinfo { 1814 sa_family_t family; 1815 }; 1816 1817 struct tcp_iter_state { 1818 struct seq_net_private p; 1819 enum tcp_seq_states state; 1820 struct sock *syn_wait_sk; 1821 int bucket, offset, sbucket, num; 1822 loff_t last_pos; 1823 }; 1824 1825 extern struct request_sock_ops tcp_request_sock_ops; 1826 extern struct request_sock_ops tcp6_request_sock_ops; 1827 1828 void tcp_v4_destroy_sock(struct sock *sk); 1829 1830 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1831 netdev_features_t features); 1832 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1833 int tcp_gro_complete(struct sk_buff *skb); 1834 1835 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1836 1837 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1838 { 1839 struct net *net = sock_net((struct sock *)tp); 1840 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1841 } 1842 1843 static inline bool tcp_stream_memory_free(const struct sock *sk) 1844 { 1845 const struct tcp_sock *tp = tcp_sk(sk); 1846 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1847 1848 return notsent_bytes < tcp_notsent_lowat(tp); 1849 } 1850 1851 #ifdef CONFIG_PROC_FS 1852 int tcp4_proc_init(void); 1853 void tcp4_proc_exit(void); 1854 #endif 1855 1856 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1857 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1858 const struct tcp_request_sock_ops *af_ops, 1859 struct sock *sk, struct sk_buff *skb); 1860 1861 /* TCP af-specific functions */ 1862 struct tcp_sock_af_ops { 1863 #ifdef CONFIG_TCP_MD5SIG 1864 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1865 const struct sock *addr_sk); 1866 int (*calc_md5_hash)(char *location, 1867 const struct tcp_md5sig_key *md5, 1868 const struct sock *sk, 1869 const struct sk_buff *skb); 1870 int (*md5_parse)(struct sock *sk, 1871 int optname, 1872 char __user *optval, 1873 int optlen); 1874 #endif 1875 }; 1876 1877 struct tcp_request_sock_ops { 1878 u16 mss_clamp; 1879 #ifdef CONFIG_TCP_MD5SIG 1880 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1881 const struct sock *addr_sk); 1882 int (*calc_md5_hash) (char *location, 1883 const struct tcp_md5sig_key *md5, 1884 const struct sock *sk, 1885 const struct sk_buff *skb); 1886 #endif 1887 void (*init_req)(struct request_sock *req, 1888 const struct sock *sk_listener, 1889 struct sk_buff *skb); 1890 #ifdef CONFIG_SYN_COOKIES 1891 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1892 __u16 *mss); 1893 #endif 1894 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1895 const struct request_sock *req); 1896 u32 (*init_seq)(const struct sk_buff *skb); 1897 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1898 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1899 struct flowi *fl, struct request_sock *req, 1900 struct tcp_fastopen_cookie *foc, 1901 enum tcp_synack_type synack_type); 1902 }; 1903 1904 #ifdef CONFIG_SYN_COOKIES 1905 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1906 const struct sock *sk, struct sk_buff *skb, 1907 __u16 *mss) 1908 { 1909 tcp_synq_overflow(sk); 1910 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1911 return ops->cookie_init_seq(skb, mss); 1912 } 1913 #else 1914 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1915 const struct sock *sk, struct sk_buff *skb, 1916 __u16 *mss) 1917 { 1918 return 0; 1919 } 1920 #endif 1921 1922 int tcpv4_offload_init(void); 1923 1924 void tcp_v4_init(void); 1925 void tcp_init(void); 1926 1927 /* tcp_recovery.c */ 1928 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1929 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1930 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1931 u32 reo_wnd); 1932 extern void tcp_rack_mark_lost(struct sock *sk); 1933 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1934 u64 xmit_time); 1935 extern void tcp_rack_reo_timeout(struct sock *sk); 1936 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1937 1938 /* At how many usecs into the future should the RTO fire? */ 1939 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1940 { 1941 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1942 u32 rto = inet_csk(sk)->icsk_rto; 1943 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1944 1945 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1946 } 1947 1948 /* 1949 * Save and compile IPv4 options, return a pointer to it 1950 */ 1951 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1952 struct sk_buff *skb) 1953 { 1954 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1955 struct ip_options_rcu *dopt = NULL; 1956 1957 if (opt->optlen) { 1958 int opt_size = sizeof(*dopt) + opt->optlen; 1959 1960 dopt = kmalloc(opt_size, GFP_ATOMIC); 1961 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1962 kfree(dopt); 1963 dopt = NULL; 1964 } 1965 } 1966 return dopt; 1967 } 1968 1969 /* locally generated TCP pure ACKs have skb->truesize == 2 1970 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1971 * This is much faster than dissecting the packet to find out. 1972 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1973 */ 1974 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1975 { 1976 return skb->truesize == 2; 1977 } 1978 1979 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1980 { 1981 skb->truesize = 2; 1982 } 1983 1984 static inline int tcp_inq(struct sock *sk) 1985 { 1986 struct tcp_sock *tp = tcp_sk(sk); 1987 int answ; 1988 1989 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1990 answ = 0; 1991 } else if (sock_flag(sk, SOCK_URGINLINE) || 1992 !tp->urg_data || 1993 before(tp->urg_seq, tp->copied_seq) || 1994 !before(tp->urg_seq, tp->rcv_nxt)) { 1995 1996 answ = tp->rcv_nxt - tp->copied_seq; 1997 1998 /* Subtract 1, if FIN was received */ 1999 if (answ && sock_flag(sk, SOCK_DONE)) 2000 answ--; 2001 } else { 2002 answ = tp->urg_seq - tp->copied_seq; 2003 } 2004 2005 return answ; 2006 } 2007 2008 int tcp_peek_len(struct socket *sock); 2009 2010 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2011 { 2012 u16 segs_in; 2013 2014 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2015 tp->segs_in += segs_in; 2016 if (skb->len > tcp_hdrlen(skb)) 2017 tp->data_segs_in += segs_in; 2018 } 2019 2020 /* 2021 * TCP listen path runs lockless. 2022 * We forced "struct sock" to be const qualified to make sure 2023 * we don't modify one of its field by mistake. 2024 * Here, we increment sk_drops which is an atomic_t, so we can safely 2025 * make sock writable again. 2026 */ 2027 static inline void tcp_listendrop(const struct sock *sk) 2028 { 2029 atomic_inc(&((struct sock *)sk)->sk_drops); 2030 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2031 } 2032 2033 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2034 2035 /* 2036 * Interface for adding Upper Level Protocols over TCP 2037 */ 2038 2039 #define TCP_ULP_NAME_MAX 16 2040 #define TCP_ULP_MAX 128 2041 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2042 2043 enum { 2044 TCP_ULP_TLS, 2045 TCP_ULP_BPF, 2046 }; 2047 2048 struct tcp_ulp_ops { 2049 struct list_head list; 2050 2051 /* initialize ulp */ 2052 int (*init)(struct sock *sk); 2053 /* cleanup ulp */ 2054 void (*release)(struct sock *sk); 2055 2056 int uid; 2057 char name[TCP_ULP_NAME_MAX]; 2058 bool user_visible; 2059 struct module *owner; 2060 }; 2061 int tcp_register_ulp(struct tcp_ulp_ops *type); 2062 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2063 int tcp_set_ulp(struct sock *sk, const char *name); 2064 int tcp_set_ulp_id(struct sock *sk, const int ulp); 2065 void tcp_get_available_ulp(char *buf, size_t len); 2066 void tcp_cleanup_ulp(struct sock *sk); 2067 2068 #define MODULE_ALIAS_TCP_ULP(name) \ 2069 __MODULE_INFO(alias, alias_userspace, name); \ 2070 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2071 2072 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2073 * is < 0, then the BPF op failed (for example if the loaded BPF 2074 * program does not support the chosen operation or there is no BPF 2075 * program loaded). 2076 */ 2077 #ifdef CONFIG_BPF 2078 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2079 { 2080 struct bpf_sock_ops_kern sock_ops; 2081 int ret; 2082 2083 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2084 if (sk_fullsock(sk)) { 2085 sock_ops.is_fullsock = 1; 2086 sock_owned_by_me(sk); 2087 } 2088 2089 sock_ops.sk = sk; 2090 sock_ops.op = op; 2091 if (nargs > 0) 2092 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2093 2094 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2095 if (ret == 0) 2096 ret = sock_ops.reply; 2097 else 2098 ret = -1; 2099 return ret; 2100 } 2101 2102 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2103 { 2104 u32 args[2] = {arg1, arg2}; 2105 2106 return tcp_call_bpf(sk, op, 2, args); 2107 } 2108 2109 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2110 u32 arg3) 2111 { 2112 u32 args[3] = {arg1, arg2, arg3}; 2113 2114 return tcp_call_bpf(sk, op, 3, args); 2115 } 2116 2117 #else 2118 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2119 { 2120 return -EPERM; 2121 } 2122 2123 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2124 { 2125 return -EPERM; 2126 } 2127 2128 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2129 u32 arg3) 2130 { 2131 return -EPERM; 2132 } 2133 2134 #endif 2135 2136 static inline u32 tcp_timeout_init(struct sock *sk) 2137 { 2138 int timeout; 2139 2140 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2141 2142 if (timeout <= 0) 2143 timeout = TCP_TIMEOUT_INIT; 2144 return timeout; 2145 } 2146 2147 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2148 { 2149 int rwnd; 2150 2151 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2152 2153 if (rwnd < 0) 2154 rwnd = 0; 2155 return rwnd; 2156 } 2157 2158 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2159 { 2160 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2161 } 2162 2163 #if IS_ENABLED(CONFIG_SMC) 2164 extern struct static_key_false tcp_have_smc; 2165 #endif 2166 2167 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2168 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2169 void (*cad)(struct sock *sk, u32 ack_seq)); 2170 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2171 2172 #endif 2173 2174 #endif /* _TCP_H */ 2175