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 ktime_get_ns(); 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 void tcp_mstamp_refresh(struct tcp_sock *tp); 756 757 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 758 { 759 return max_t(s64, t1 - t0, 0); 760 } 761 762 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 763 { 764 return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ); 765 } 766 767 /* provide the departure time in us unit */ 768 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 769 { 770 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 771 } 772 773 774 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 775 776 #define TCPHDR_FIN 0x01 777 #define TCPHDR_SYN 0x02 778 #define TCPHDR_RST 0x04 779 #define TCPHDR_PSH 0x08 780 #define TCPHDR_ACK 0x10 781 #define TCPHDR_URG 0x20 782 #define TCPHDR_ECE 0x40 783 #define TCPHDR_CWR 0x80 784 785 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 786 787 /* This is what the send packet queuing engine uses to pass 788 * TCP per-packet control information to the transmission code. 789 * We also store the host-order sequence numbers in here too. 790 * This is 44 bytes if IPV6 is enabled. 791 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 792 */ 793 struct tcp_skb_cb { 794 __u32 seq; /* Starting sequence number */ 795 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 796 union { 797 /* Note : tcp_tw_isn is used in input path only 798 * (isn chosen by tcp_timewait_state_process()) 799 * 800 * tcp_gso_segs/size are used in write queue only, 801 * cf tcp_skb_pcount()/tcp_skb_mss() 802 */ 803 __u32 tcp_tw_isn; 804 struct { 805 u16 tcp_gso_segs; 806 u16 tcp_gso_size; 807 }; 808 }; 809 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 810 811 __u8 sacked; /* State flags for SACK. */ 812 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 813 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 814 #define TCPCB_LOST 0x04 /* SKB is lost */ 815 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 816 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 817 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 818 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 819 TCPCB_REPAIRED) 820 821 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 822 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 823 eor:1, /* Is skb MSG_EOR marked? */ 824 has_rxtstamp:1, /* SKB has a RX timestamp */ 825 unused:5; 826 __u32 ack_seq; /* Sequence number ACK'd */ 827 union { 828 struct { 829 /* There is space for up to 24 bytes */ 830 __u32 in_flight:30,/* Bytes in flight at transmit */ 831 is_app_limited:1, /* cwnd not fully used? */ 832 unused:1; 833 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 834 __u32 delivered; 835 /* start of send pipeline phase */ 836 u64 first_tx_mstamp; 837 /* when we reached the "delivered" count */ 838 u64 delivered_mstamp; 839 } tx; /* only used for outgoing skbs */ 840 union { 841 struct inet_skb_parm h4; 842 #if IS_ENABLED(CONFIG_IPV6) 843 struct inet6_skb_parm h6; 844 #endif 845 } header; /* For incoming skbs */ 846 struct { 847 __u32 flags; 848 struct sock *sk_redir; 849 void *data_end; 850 } bpf; 851 }; 852 }; 853 854 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 855 856 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 857 { 858 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 859 } 860 861 #if IS_ENABLED(CONFIG_IPV6) 862 /* This is the variant of inet6_iif() that must be used by TCP, 863 * as TCP moves IP6CB into a different location in skb->cb[] 864 */ 865 static inline int tcp_v6_iif(const struct sk_buff *skb) 866 { 867 return TCP_SKB_CB(skb)->header.h6.iif; 868 } 869 870 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 871 { 872 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 873 874 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 875 } 876 877 /* TCP_SKB_CB reference means this can not be used from early demux */ 878 static inline int tcp_v6_sdif(const struct sk_buff *skb) 879 { 880 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 881 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 882 return TCP_SKB_CB(skb)->header.h6.iif; 883 #endif 884 return 0; 885 } 886 #endif 887 888 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 889 { 890 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 891 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 892 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 893 return true; 894 #endif 895 return false; 896 } 897 898 /* TCP_SKB_CB reference means this can not be used from early demux */ 899 static inline int tcp_v4_sdif(struct sk_buff *skb) 900 { 901 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 902 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 903 return TCP_SKB_CB(skb)->header.h4.iif; 904 #endif 905 return 0; 906 } 907 908 /* Due to TSO, an SKB can be composed of multiple actual 909 * packets. To keep these tracked properly, we use this. 910 */ 911 static inline int tcp_skb_pcount(const struct sk_buff *skb) 912 { 913 return TCP_SKB_CB(skb)->tcp_gso_segs; 914 } 915 916 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 917 { 918 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 919 } 920 921 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 922 { 923 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 924 } 925 926 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 927 static inline int tcp_skb_mss(const struct sk_buff *skb) 928 { 929 return TCP_SKB_CB(skb)->tcp_gso_size; 930 } 931 932 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 933 { 934 return likely(!TCP_SKB_CB(skb)->eor); 935 } 936 937 /* Events passed to congestion control interface */ 938 enum tcp_ca_event { 939 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 940 CA_EVENT_CWND_RESTART, /* congestion window restart */ 941 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 942 CA_EVENT_LOSS, /* loss timeout */ 943 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 944 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 945 }; 946 947 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 948 enum tcp_ca_ack_event_flags { 949 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 950 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 951 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 952 }; 953 954 /* 955 * Interface for adding new TCP congestion control handlers 956 */ 957 #define TCP_CA_NAME_MAX 16 958 #define TCP_CA_MAX 128 959 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 960 961 #define TCP_CA_UNSPEC 0 962 963 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 964 #define TCP_CONG_NON_RESTRICTED 0x1 965 /* Requires ECN/ECT set on all packets */ 966 #define TCP_CONG_NEEDS_ECN 0x2 967 968 union tcp_cc_info; 969 970 struct ack_sample { 971 u32 pkts_acked; 972 s32 rtt_us; 973 u32 in_flight; 974 }; 975 976 /* A rate sample measures the number of (original/retransmitted) data 977 * packets delivered "delivered" over an interval of time "interval_us". 978 * The tcp_rate.c code fills in the rate sample, and congestion 979 * control modules that define a cong_control function to run at the end 980 * of ACK processing can optionally chose to consult this sample when 981 * setting cwnd and pacing rate. 982 * A sample is invalid if "delivered" or "interval_us" is negative. 983 */ 984 struct rate_sample { 985 u64 prior_mstamp; /* starting timestamp for interval */ 986 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 987 s32 delivered; /* number of packets delivered over interval */ 988 long interval_us; /* time for tp->delivered to incr "delivered" */ 989 u32 snd_interval_us; /* snd interval for delivered packets */ 990 u32 rcv_interval_us; /* rcv interval for delivered packets */ 991 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 992 int losses; /* number of packets marked lost upon ACK */ 993 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 994 u32 prior_in_flight; /* in flight before this ACK */ 995 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 996 bool is_retrans; /* is sample from retransmission? */ 997 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 998 }; 999 1000 struct tcp_congestion_ops { 1001 struct list_head list; 1002 u32 key; 1003 u32 flags; 1004 1005 /* initialize private data (optional) */ 1006 void (*init)(struct sock *sk); 1007 /* cleanup private data (optional) */ 1008 void (*release)(struct sock *sk); 1009 1010 /* return slow start threshold (required) */ 1011 u32 (*ssthresh)(struct sock *sk); 1012 /* do new cwnd calculation (required) */ 1013 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1014 /* call before changing ca_state (optional) */ 1015 void (*set_state)(struct sock *sk, u8 new_state); 1016 /* call when cwnd event occurs (optional) */ 1017 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1018 /* call when ack arrives (optional) */ 1019 void (*in_ack_event)(struct sock *sk, u32 flags); 1020 /* new value of cwnd after loss (required) */ 1021 u32 (*undo_cwnd)(struct sock *sk); 1022 /* hook for packet ack accounting (optional) */ 1023 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1024 /* override sysctl_tcp_min_tso_segs */ 1025 u32 (*min_tso_segs)(struct sock *sk); 1026 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1027 u32 (*sndbuf_expand)(struct sock *sk); 1028 /* call when packets are delivered to update cwnd and pacing rate, 1029 * after all the ca_state processing. (optional) 1030 */ 1031 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1032 /* get info for inet_diag (optional) */ 1033 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1034 union tcp_cc_info *info); 1035 1036 char name[TCP_CA_NAME_MAX]; 1037 struct module *owner; 1038 }; 1039 1040 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1041 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1042 1043 void tcp_assign_congestion_control(struct sock *sk); 1044 void tcp_init_congestion_control(struct sock *sk); 1045 void tcp_cleanup_congestion_control(struct sock *sk); 1046 int tcp_set_default_congestion_control(struct net *net, const char *name); 1047 void tcp_get_default_congestion_control(struct net *net, char *name); 1048 void tcp_get_available_congestion_control(char *buf, size_t len); 1049 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1050 int tcp_set_allowed_congestion_control(char *allowed); 1051 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1052 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1053 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1054 1055 u32 tcp_reno_ssthresh(struct sock *sk); 1056 u32 tcp_reno_undo_cwnd(struct sock *sk); 1057 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1058 extern struct tcp_congestion_ops tcp_reno; 1059 1060 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1061 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1062 #ifdef CONFIG_INET 1063 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1064 #else 1065 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1066 { 1067 return NULL; 1068 } 1069 #endif 1070 1071 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1072 { 1073 const struct inet_connection_sock *icsk = inet_csk(sk); 1074 1075 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1076 } 1077 1078 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1079 { 1080 struct inet_connection_sock *icsk = inet_csk(sk); 1081 1082 if (icsk->icsk_ca_ops->set_state) 1083 icsk->icsk_ca_ops->set_state(sk, ca_state); 1084 icsk->icsk_ca_state = ca_state; 1085 } 1086 1087 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1088 { 1089 const struct inet_connection_sock *icsk = inet_csk(sk); 1090 1091 if (icsk->icsk_ca_ops->cwnd_event) 1092 icsk->icsk_ca_ops->cwnd_event(sk, event); 1093 } 1094 1095 /* From tcp_rate.c */ 1096 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1097 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1098 struct rate_sample *rs); 1099 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1100 bool is_sack_reneg, struct rate_sample *rs); 1101 void tcp_rate_check_app_limited(struct sock *sk); 1102 1103 /* These functions determine how the current flow behaves in respect of SACK 1104 * handling. SACK is negotiated with the peer, and therefore it can vary 1105 * between different flows. 1106 * 1107 * tcp_is_sack - SACK enabled 1108 * tcp_is_reno - No SACK 1109 */ 1110 static inline int tcp_is_sack(const struct tcp_sock *tp) 1111 { 1112 return tp->rx_opt.sack_ok; 1113 } 1114 1115 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1116 { 1117 return !tcp_is_sack(tp); 1118 } 1119 1120 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1121 { 1122 return tp->sacked_out + tp->lost_out; 1123 } 1124 1125 /* This determines how many packets are "in the network" to the best 1126 * of our knowledge. In many cases it is conservative, but where 1127 * detailed information is available from the receiver (via SACK 1128 * blocks etc.) we can make more aggressive calculations. 1129 * 1130 * Use this for decisions involving congestion control, use just 1131 * tp->packets_out to determine if the send queue is empty or not. 1132 * 1133 * Read this equation as: 1134 * 1135 * "Packets sent once on transmission queue" MINUS 1136 * "Packets left network, but not honestly ACKed yet" PLUS 1137 * "Packets fast retransmitted" 1138 */ 1139 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1140 { 1141 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1142 } 1143 1144 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1145 1146 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1147 { 1148 return tp->snd_cwnd < tp->snd_ssthresh; 1149 } 1150 1151 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1152 { 1153 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1154 } 1155 1156 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1157 { 1158 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1159 (1 << inet_csk(sk)->icsk_ca_state); 1160 } 1161 1162 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1163 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1164 * ssthresh. 1165 */ 1166 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1167 { 1168 const struct tcp_sock *tp = tcp_sk(sk); 1169 1170 if (tcp_in_cwnd_reduction(sk)) 1171 return tp->snd_ssthresh; 1172 else 1173 return max(tp->snd_ssthresh, 1174 ((tp->snd_cwnd >> 1) + 1175 (tp->snd_cwnd >> 2))); 1176 } 1177 1178 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1179 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1180 1181 void tcp_enter_cwr(struct sock *sk); 1182 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1183 1184 /* The maximum number of MSS of available cwnd for which TSO defers 1185 * sending if not using sysctl_tcp_tso_win_divisor. 1186 */ 1187 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1188 { 1189 return 3; 1190 } 1191 1192 /* Returns end sequence number of the receiver's advertised window */ 1193 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1194 { 1195 return tp->snd_una + tp->snd_wnd; 1196 } 1197 1198 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1199 * flexible approach. The RFC suggests cwnd should not be raised unless 1200 * it was fully used previously. And that's exactly what we do in 1201 * congestion avoidance mode. But in slow start we allow cwnd to grow 1202 * as long as the application has used half the cwnd. 1203 * Example : 1204 * cwnd is 10 (IW10), but application sends 9 frames. 1205 * We allow cwnd to reach 18 when all frames are ACKed. 1206 * This check is safe because it's as aggressive as slow start which already 1207 * risks 100% overshoot. The advantage is that we discourage application to 1208 * either send more filler packets or data to artificially blow up the cwnd 1209 * usage, and allow application-limited process to probe bw more aggressively. 1210 */ 1211 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1212 { 1213 const struct tcp_sock *tp = tcp_sk(sk); 1214 1215 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1216 if (tcp_in_slow_start(tp)) 1217 return tp->snd_cwnd < 2 * tp->max_packets_out; 1218 1219 return tp->is_cwnd_limited; 1220 } 1221 1222 /* BBR congestion control needs pacing. 1223 * Same remark for SO_MAX_PACING_RATE. 1224 * sch_fq packet scheduler is efficiently handling pacing, 1225 * but is not always installed/used. 1226 * Return true if TCP stack should pace packets itself. 1227 */ 1228 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1229 { 1230 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1231 } 1232 1233 /* Something is really bad, we could not queue an additional packet, 1234 * because qdisc is full or receiver sent a 0 window. 1235 * We do not want to add fuel to the fire, or abort too early, 1236 * so make sure the timer we arm now is at least 200ms in the future, 1237 * regardless of current icsk_rto value (as it could be ~2ms) 1238 */ 1239 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1240 { 1241 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1242 } 1243 1244 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1245 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1246 unsigned long max_when) 1247 { 1248 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1249 1250 return (unsigned long)min_t(u64, when, max_when); 1251 } 1252 1253 static inline void tcp_check_probe_timer(struct sock *sk) 1254 { 1255 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1256 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1257 tcp_probe0_base(sk), TCP_RTO_MAX); 1258 } 1259 1260 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1261 { 1262 tp->snd_wl1 = seq; 1263 } 1264 1265 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1266 { 1267 tp->snd_wl1 = seq; 1268 } 1269 1270 /* 1271 * Calculate(/check) TCP checksum 1272 */ 1273 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1274 __be32 daddr, __wsum base) 1275 { 1276 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1277 } 1278 1279 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1280 { 1281 return __skb_checksum_complete(skb); 1282 } 1283 1284 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1285 { 1286 return !skb_csum_unnecessary(skb) && 1287 __tcp_checksum_complete(skb); 1288 } 1289 1290 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1291 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1292 1293 #undef STATE_TRACE 1294 1295 #ifdef STATE_TRACE 1296 static const char *statename[]={ 1297 "Unused","Established","Syn Sent","Syn Recv", 1298 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1299 "Close Wait","Last ACK","Listen","Closing" 1300 }; 1301 #endif 1302 void tcp_set_state(struct sock *sk, int state); 1303 1304 void tcp_done(struct sock *sk); 1305 1306 int tcp_abort(struct sock *sk, int err); 1307 1308 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1309 { 1310 rx_opt->dsack = 0; 1311 rx_opt->num_sacks = 0; 1312 } 1313 1314 u32 tcp_default_init_rwnd(u32 mss); 1315 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1316 1317 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1318 { 1319 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1320 struct tcp_sock *tp = tcp_sk(sk); 1321 s32 delta; 1322 1323 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1324 ca_ops->cong_control) 1325 return; 1326 delta = tcp_jiffies32 - tp->lsndtime; 1327 if (delta > inet_csk(sk)->icsk_rto) 1328 tcp_cwnd_restart(sk, delta); 1329 } 1330 1331 /* Determine a window scaling and initial window to offer. */ 1332 void tcp_select_initial_window(const struct sock *sk, int __space, 1333 __u32 mss, __u32 *rcv_wnd, 1334 __u32 *window_clamp, int wscale_ok, 1335 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1336 1337 static inline int tcp_win_from_space(const struct sock *sk, int space) 1338 { 1339 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1340 1341 return tcp_adv_win_scale <= 0 ? 1342 (space>>(-tcp_adv_win_scale)) : 1343 space - (space>>tcp_adv_win_scale); 1344 } 1345 1346 /* Note: caller must be prepared to deal with negative returns */ 1347 static inline int tcp_space(const struct sock *sk) 1348 { 1349 return tcp_win_from_space(sk, sk->sk_rcvbuf - 1350 atomic_read(&sk->sk_rmem_alloc)); 1351 } 1352 1353 static inline int tcp_full_space(const struct sock *sk) 1354 { 1355 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1356 } 1357 1358 extern void tcp_openreq_init_rwin(struct request_sock *req, 1359 const struct sock *sk_listener, 1360 const struct dst_entry *dst); 1361 1362 void tcp_enter_memory_pressure(struct sock *sk); 1363 void tcp_leave_memory_pressure(struct sock *sk); 1364 1365 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1366 { 1367 struct net *net = sock_net((struct sock *)tp); 1368 1369 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1370 } 1371 1372 static inline int keepalive_time_when(const struct tcp_sock *tp) 1373 { 1374 struct net *net = sock_net((struct sock *)tp); 1375 1376 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1377 } 1378 1379 static inline int keepalive_probes(const struct tcp_sock *tp) 1380 { 1381 struct net *net = sock_net((struct sock *)tp); 1382 1383 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1384 } 1385 1386 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1387 { 1388 const struct inet_connection_sock *icsk = &tp->inet_conn; 1389 1390 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1391 tcp_jiffies32 - tp->rcv_tstamp); 1392 } 1393 1394 static inline int tcp_fin_time(const struct sock *sk) 1395 { 1396 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1397 const int rto = inet_csk(sk)->icsk_rto; 1398 1399 if (fin_timeout < (rto << 2) - (rto >> 1)) 1400 fin_timeout = (rto << 2) - (rto >> 1); 1401 1402 return fin_timeout; 1403 } 1404 1405 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1406 int paws_win) 1407 { 1408 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1409 return true; 1410 if (unlikely(!time_before32(ktime_get_seconds(), 1411 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1412 return true; 1413 /* 1414 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1415 * then following tcp messages have valid values. Ignore 0 value, 1416 * or else 'negative' tsval might forbid us to accept their packets. 1417 */ 1418 if (!rx_opt->ts_recent) 1419 return true; 1420 return false; 1421 } 1422 1423 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1424 int rst) 1425 { 1426 if (tcp_paws_check(rx_opt, 0)) 1427 return false; 1428 1429 /* RST segments are not recommended to carry timestamp, 1430 and, if they do, it is recommended to ignore PAWS because 1431 "their cleanup function should take precedence over timestamps." 1432 Certainly, it is mistake. It is necessary to understand the reasons 1433 of this constraint to relax it: if peer reboots, clock may go 1434 out-of-sync and half-open connections will not be reset. 1435 Actually, the problem would be not existing if all 1436 the implementations followed draft about maintaining clock 1437 via reboots. Linux-2.2 DOES NOT! 1438 1439 However, we can relax time bounds for RST segments to MSL. 1440 */ 1441 if (rst && !time_before32(ktime_get_seconds(), 1442 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1443 return false; 1444 return true; 1445 } 1446 1447 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1448 int mib_idx, u32 *last_oow_ack_time); 1449 1450 static inline void tcp_mib_init(struct net *net) 1451 { 1452 /* See RFC 2012 */ 1453 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1454 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1455 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1456 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1457 } 1458 1459 /* from STCP */ 1460 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1461 { 1462 tp->lost_skb_hint = NULL; 1463 } 1464 1465 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1466 { 1467 tcp_clear_retrans_hints_partial(tp); 1468 tp->retransmit_skb_hint = NULL; 1469 } 1470 1471 union tcp_md5_addr { 1472 struct in_addr a4; 1473 #if IS_ENABLED(CONFIG_IPV6) 1474 struct in6_addr a6; 1475 #endif 1476 }; 1477 1478 /* - key database */ 1479 struct tcp_md5sig_key { 1480 struct hlist_node node; 1481 u8 keylen; 1482 u8 family; /* AF_INET or AF_INET6 */ 1483 union tcp_md5_addr addr; 1484 u8 prefixlen; 1485 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1486 struct rcu_head rcu; 1487 }; 1488 1489 /* - sock block */ 1490 struct tcp_md5sig_info { 1491 struct hlist_head head; 1492 struct rcu_head rcu; 1493 }; 1494 1495 /* - pseudo header */ 1496 struct tcp4_pseudohdr { 1497 __be32 saddr; 1498 __be32 daddr; 1499 __u8 pad; 1500 __u8 protocol; 1501 __be16 len; 1502 }; 1503 1504 struct tcp6_pseudohdr { 1505 struct in6_addr saddr; 1506 struct in6_addr daddr; 1507 __be32 len; 1508 __be32 protocol; /* including padding */ 1509 }; 1510 1511 union tcp_md5sum_block { 1512 struct tcp4_pseudohdr ip4; 1513 #if IS_ENABLED(CONFIG_IPV6) 1514 struct tcp6_pseudohdr ip6; 1515 #endif 1516 }; 1517 1518 /* - pool: digest algorithm, hash description and scratch buffer */ 1519 struct tcp_md5sig_pool { 1520 struct ahash_request *md5_req; 1521 void *scratch; 1522 }; 1523 1524 /* - functions */ 1525 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1526 const struct sock *sk, const struct sk_buff *skb); 1527 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1528 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1529 gfp_t gfp); 1530 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1531 int family, u8 prefixlen); 1532 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1533 const struct sock *addr_sk); 1534 1535 #ifdef CONFIG_TCP_MD5SIG 1536 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1537 const union tcp_md5_addr *addr, 1538 int family); 1539 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1540 #else 1541 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1542 const union tcp_md5_addr *addr, 1543 int family) 1544 { 1545 return NULL; 1546 } 1547 #define tcp_twsk_md5_key(twsk) NULL 1548 #endif 1549 1550 bool tcp_alloc_md5sig_pool(void); 1551 1552 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1553 static inline void tcp_put_md5sig_pool(void) 1554 { 1555 local_bh_enable(); 1556 } 1557 1558 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1559 unsigned int header_len); 1560 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1561 const struct tcp_md5sig_key *key); 1562 1563 /* From tcp_fastopen.c */ 1564 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1565 struct tcp_fastopen_cookie *cookie); 1566 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1567 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1568 u16 try_exp); 1569 struct tcp_fastopen_request { 1570 /* Fast Open cookie. Size 0 means a cookie request */ 1571 struct tcp_fastopen_cookie cookie; 1572 struct msghdr *data; /* data in MSG_FASTOPEN */ 1573 size_t size; 1574 int copied; /* queued in tcp_connect() */ 1575 }; 1576 void tcp_free_fastopen_req(struct tcp_sock *tp); 1577 void tcp_fastopen_destroy_cipher(struct sock *sk); 1578 void tcp_fastopen_ctx_destroy(struct net *net); 1579 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1580 void *key, unsigned int len); 1581 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1582 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1583 struct request_sock *req, 1584 struct tcp_fastopen_cookie *foc, 1585 const struct dst_entry *dst); 1586 void tcp_fastopen_init_key_once(struct net *net); 1587 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1588 struct tcp_fastopen_cookie *cookie); 1589 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1590 #define TCP_FASTOPEN_KEY_LENGTH 16 1591 1592 /* Fastopen key context */ 1593 struct tcp_fastopen_context { 1594 struct crypto_cipher *tfm; 1595 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1596 struct rcu_head rcu; 1597 }; 1598 1599 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1600 void tcp_fastopen_active_disable(struct sock *sk); 1601 bool tcp_fastopen_active_should_disable(struct sock *sk); 1602 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1603 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1604 1605 /* Latencies incurred by various limits for a sender. They are 1606 * chronograph-like stats that are mutually exclusive. 1607 */ 1608 enum tcp_chrono { 1609 TCP_CHRONO_UNSPEC, 1610 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1611 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1612 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1613 __TCP_CHRONO_MAX, 1614 }; 1615 1616 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1617 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1618 1619 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1620 * the same memory storage than skb->destructor/_skb_refdst 1621 */ 1622 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1623 { 1624 skb->destructor = NULL; 1625 skb->_skb_refdst = 0UL; 1626 } 1627 1628 #define tcp_skb_tsorted_save(skb) { \ 1629 unsigned long _save = skb->_skb_refdst; \ 1630 skb->_skb_refdst = 0UL; 1631 1632 #define tcp_skb_tsorted_restore(skb) \ 1633 skb->_skb_refdst = _save; \ 1634 } 1635 1636 void tcp_write_queue_purge(struct sock *sk); 1637 1638 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1639 { 1640 return skb_rb_first(&sk->tcp_rtx_queue); 1641 } 1642 1643 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1644 { 1645 return skb_peek(&sk->sk_write_queue); 1646 } 1647 1648 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1649 { 1650 return skb_peek_tail(&sk->sk_write_queue); 1651 } 1652 1653 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1654 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1655 1656 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1657 { 1658 return skb_peek(&sk->sk_write_queue); 1659 } 1660 1661 static inline bool tcp_skb_is_last(const struct sock *sk, 1662 const struct sk_buff *skb) 1663 { 1664 return skb_queue_is_last(&sk->sk_write_queue, skb); 1665 } 1666 1667 static inline bool tcp_write_queue_empty(const struct sock *sk) 1668 { 1669 return skb_queue_empty(&sk->sk_write_queue); 1670 } 1671 1672 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1673 { 1674 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1675 } 1676 1677 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1678 { 1679 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1680 } 1681 1682 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1683 { 1684 if (tcp_write_queue_empty(sk)) 1685 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1686 } 1687 1688 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1689 { 1690 __skb_queue_tail(&sk->sk_write_queue, skb); 1691 } 1692 1693 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1694 { 1695 __tcp_add_write_queue_tail(sk, skb); 1696 1697 /* Queue it, remembering where we must start sending. */ 1698 if (sk->sk_write_queue.next == skb) 1699 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1700 } 1701 1702 /* Insert new before skb on the write queue of sk. */ 1703 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1704 struct sk_buff *skb, 1705 struct sock *sk) 1706 { 1707 __skb_queue_before(&sk->sk_write_queue, skb, new); 1708 } 1709 1710 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1711 { 1712 tcp_skb_tsorted_anchor_cleanup(skb); 1713 __skb_unlink(skb, &sk->sk_write_queue); 1714 } 1715 1716 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1717 1718 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1719 { 1720 tcp_skb_tsorted_anchor_cleanup(skb); 1721 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1722 } 1723 1724 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1725 { 1726 list_del(&skb->tcp_tsorted_anchor); 1727 tcp_rtx_queue_unlink(skb, sk); 1728 sk_wmem_free_skb(sk, skb); 1729 } 1730 1731 static inline void tcp_push_pending_frames(struct sock *sk) 1732 { 1733 if (tcp_send_head(sk)) { 1734 struct tcp_sock *tp = tcp_sk(sk); 1735 1736 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1737 } 1738 } 1739 1740 /* Start sequence of the skb just after the highest skb with SACKed 1741 * bit, valid only if sacked_out > 0 or when the caller has ensured 1742 * validity by itself. 1743 */ 1744 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1745 { 1746 if (!tp->sacked_out) 1747 return tp->snd_una; 1748 1749 if (tp->highest_sack == NULL) 1750 return tp->snd_nxt; 1751 1752 return TCP_SKB_CB(tp->highest_sack)->seq; 1753 } 1754 1755 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1756 { 1757 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1758 } 1759 1760 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1761 { 1762 return tcp_sk(sk)->highest_sack; 1763 } 1764 1765 static inline void tcp_highest_sack_reset(struct sock *sk) 1766 { 1767 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1768 } 1769 1770 /* Called when old skb is about to be deleted and replaced by new skb */ 1771 static inline void tcp_highest_sack_replace(struct sock *sk, 1772 struct sk_buff *old, 1773 struct sk_buff *new) 1774 { 1775 if (old == tcp_highest_sack(sk)) 1776 tcp_sk(sk)->highest_sack = new; 1777 } 1778 1779 /* This helper checks if socket has IP_TRANSPARENT set */ 1780 static inline bool inet_sk_transparent(const struct sock *sk) 1781 { 1782 switch (sk->sk_state) { 1783 case TCP_TIME_WAIT: 1784 return inet_twsk(sk)->tw_transparent; 1785 case TCP_NEW_SYN_RECV: 1786 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1787 } 1788 return inet_sk(sk)->transparent; 1789 } 1790 1791 /* Determines whether this is a thin stream (which may suffer from 1792 * increased latency). Used to trigger latency-reducing mechanisms. 1793 */ 1794 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1795 { 1796 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1797 } 1798 1799 /* /proc */ 1800 enum tcp_seq_states { 1801 TCP_SEQ_STATE_LISTENING, 1802 TCP_SEQ_STATE_ESTABLISHED, 1803 }; 1804 1805 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1806 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1807 void tcp_seq_stop(struct seq_file *seq, void *v); 1808 1809 struct tcp_seq_afinfo { 1810 sa_family_t family; 1811 }; 1812 1813 struct tcp_iter_state { 1814 struct seq_net_private p; 1815 enum tcp_seq_states state; 1816 struct sock *syn_wait_sk; 1817 int bucket, offset, sbucket, num; 1818 loff_t last_pos; 1819 }; 1820 1821 extern struct request_sock_ops tcp_request_sock_ops; 1822 extern struct request_sock_ops tcp6_request_sock_ops; 1823 1824 void tcp_v4_destroy_sock(struct sock *sk); 1825 1826 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1827 netdev_features_t features); 1828 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1829 int tcp_gro_complete(struct sk_buff *skb); 1830 1831 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1832 1833 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1834 { 1835 struct net *net = sock_net((struct sock *)tp); 1836 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1837 } 1838 1839 static inline bool tcp_stream_memory_free(const struct sock *sk) 1840 { 1841 const struct tcp_sock *tp = tcp_sk(sk); 1842 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1843 1844 return notsent_bytes < tcp_notsent_lowat(tp); 1845 } 1846 1847 #ifdef CONFIG_PROC_FS 1848 int tcp4_proc_init(void); 1849 void tcp4_proc_exit(void); 1850 #endif 1851 1852 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1853 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1854 const struct tcp_request_sock_ops *af_ops, 1855 struct sock *sk, struct sk_buff *skb); 1856 1857 /* TCP af-specific functions */ 1858 struct tcp_sock_af_ops { 1859 #ifdef CONFIG_TCP_MD5SIG 1860 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1861 const struct sock *addr_sk); 1862 int (*calc_md5_hash)(char *location, 1863 const struct tcp_md5sig_key *md5, 1864 const struct sock *sk, 1865 const struct sk_buff *skb); 1866 int (*md5_parse)(struct sock *sk, 1867 int optname, 1868 char __user *optval, 1869 int optlen); 1870 #endif 1871 }; 1872 1873 struct tcp_request_sock_ops { 1874 u16 mss_clamp; 1875 #ifdef CONFIG_TCP_MD5SIG 1876 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1877 const struct sock *addr_sk); 1878 int (*calc_md5_hash) (char *location, 1879 const struct tcp_md5sig_key *md5, 1880 const struct sock *sk, 1881 const struct sk_buff *skb); 1882 #endif 1883 void (*init_req)(struct request_sock *req, 1884 const struct sock *sk_listener, 1885 struct sk_buff *skb); 1886 #ifdef CONFIG_SYN_COOKIES 1887 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1888 __u16 *mss); 1889 #endif 1890 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1891 const struct request_sock *req); 1892 u32 (*init_seq)(const struct sk_buff *skb); 1893 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1894 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1895 struct flowi *fl, struct request_sock *req, 1896 struct tcp_fastopen_cookie *foc, 1897 enum tcp_synack_type synack_type); 1898 }; 1899 1900 #ifdef CONFIG_SYN_COOKIES 1901 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1902 const struct sock *sk, struct sk_buff *skb, 1903 __u16 *mss) 1904 { 1905 tcp_synq_overflow(sk); 1906 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1907 return ops->cookie_init_seq(skb, mss); 1908 } 1909 #else 1910 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1911 const struct sock *sk, struct sk_buff *skb, 1912 __u16 *mss) 1913 { 1914 return 0; 1915 } 1916 #endif 1917 1918 int tcpv4_offload_init(void); 1919 1920 void tcp_v4_init(void); 1921 void tcp_init(void); 1922 1923 /* tcp_recovery.c */ 1924 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1925 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1926 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1927 u32 reo_wnd); 1928 extern void tcp_rack_mark_lost(struct sock *sk); 1929 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1930 u64 xmit_time); 1931 extern void tcp_rack_reo_timeout(struct sock *sk); 1932 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1933 1934 /* At how many usecs into the future should the RTO fire? */ 1935 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1936 { 1937 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1938 u32 rto = inet_csk(sk)->icsk_rto; 1939 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 1940 1941 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1942 } 1943 1944 /* 1945 * Save and compile IPv4 options, return a pointer to it 1946 */ 1947 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1948 struct sk_buff *skb) 1949 { 1950 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1951 struct ip_options_rcu *dopt = NULL; 1952 1953 if (opt->optlen) { 1954 int opt_size = sizeof(*dopt) + opt->optlen; 1955 1956 dopt = kmalloc(opt_size, GFP_ATOMIC); 1957 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1958 kfree(dopt); 1959 dopt = NULL; 1960 } 1961 } 1962 return dopt; 1963 } 1964 1965 /* locally generated TCP pure ACKs have skb->truesize == 2 1966 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1967 * This is much faster than dissecting the packet to find out. 1968 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1969 */ 1970 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1971 { 1972 return skb->truesize == 2; 1973 } 1974 1975 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1976 { 1977 skb->truesize = 2; 1978 } 1979 1980 static inline int tcp_inq(struct sock *sk) 1981 { 1982 struct tcp_sock *tp = tcp_sk(sk); 1983 int answ; 1984 1985 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1986 answ = 0; 1987 } else if (sock_flag(sk, SOCK_URGINLINE) || 1988 !tp->urg_data || 1989 before(tp->urg_seq, tp->copied_seq) || 1990 !before(tp->urg_seq, tp->rcv_nxt)) { 1991 1992 answ = tp->rcv_nxt - tp->copied_seq; 1993 1994 /* Subtract 1, if FIN was received */ 1995 if (answ && sock_flag(sk, SOCK_DONE)) 1996 answ--; 1997 } else { 1998 answ = tp->urg_seq - tp->copied_seq; 1999 } 2000 2001 return answ; 2002 } 2003 2004 int tcp_peek_len(struct socket *sock); 2005 2006 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2007 { 2008 u16 segs_in; 2009 2010 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2011 tp->segs_in += segs_in; 2012 if (skb->len > tcp_hdrlen(skb)) 2013 tp->data_segs_in += segs_in; 2014 } 2015 2016 /* 2017 * TCP listen path runs lockless. 2018 * We forced "struct sock" to be const qualified to make sure 2019 * we don't modify one of its field by mistake. 2020 * Here, we increment sk_drops which is an atomic_t, so we can safely 2021 * make sock writable again. 2022 */ 2023 static inline void tcp_listendrop(const struct sock *sk) 2024 { 2025 atomic_inc(&((struct sock *)sk)->sk_drops); 2026 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2027 } 2028 2029 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2030 2031 /* 2032 * Interface for adding Upper Level Protocols over TCP 2033 */ 2034 2035 #define TCP_ULP_NAME_MAX 16 2036 #define TCP_ULP_MAX 128 2037 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2038 2039 enum { 2040 TCP_ULP_TLS, 2041 TCP_ULP_BPF, 2042 }; 2043 2044 struct tcp_ulp_ops { 2045 struct list_head list; 2046 2047 /* initialize ulp */ 2048 int (*init)(struct sock *sk); 2049 /* cleanup ulp */ 2050 void (*release)(struct sock *sk); 2051 2052 int uid; 2053 char name[TCP_ULP_NAME_MAX]; 2054 bool user_visible; 2055 struct module *owner; 2056 }; 2057 int tcp_register_ulp(struct tcp_ulp_ops *type); 2058 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2059 int tcp_set_ulp(struct sock *sk, const char *name); 2060 int tcp_set_ulp_id(struct sock *sk, const int ulp); 2061 void tcp_get_available_ulp(char *buf, size_t len); 2062 void tcp_cleanup_ulp(struct sock *sk); 2063 2064 #define MODULE_ALIAS_TCP_ULP(name) \ 2065 __MODULE_INFO(alias, alias_userspace, name); \ 2066 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2067 2068 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2069 * is < 0, then the BPF op failed (for example if the loaded BPF 2070 * program does not support the chosen operation or there is no BPF 2071 * program loaded). 2072 */ 2073 #ifdef CONFIG_BPF 2074 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2075 { 2076 struct bpf_sock_ops_kern sock_ops; 2077 int ret; 2078 2079 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2080 if (sk_fullsock(sk)) { 2081 sock_ops.is_fullsock = 1; 2082 sock_owned_by_me(sk); 2083 } 2084 2085 sock_ops.sk = sk; 2086 sock_ops.op = op; 2087 if (nargs > 0) 2088 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2089 2090 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2091 if (ret == 0) 2092 ret = sock_ops.reply; 2093 else 2094 ret = -1; 2095 return ret; 2096 } 2097 2098 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2099 { 2100 u32 args[2] = {arg1, arg2}; 2101 2102 return tcp_call_bpf(sk, op, 2, args); 2103 } 2104 2105 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2106 u32 arg3) 2107 { 2108 u32 args[3] = {arg1, arg2, arg3}; 2109 2110 return tcp_call_bpf(sk, op, 3, args); 2111 } 2112 2113 #else 2114 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2115 { 2116 return -EPERM; 2117 } 2118 2119 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2120 { 2121 return -EPERM; 2122 } 2123 2124 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2125 u32 arg3) 2126 { 2127 return -EPERM; 2128 } 2129 2130 #endif 2131 2132 static inline u32 tcp_timeout_init(struct sock *sk) 2133 { 2134 int timeout; 2135 2136 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2137 2138 if (timeout <= 0) 2139 timeout = TCP_TIMEOUT_INIT; 2140 return timeout; 2141 } 2142 2143 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2144 { 2145 int rwnd; 2146 2147 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2148 2149 if (rwnd < 0) 2150 rwnd = 0; 2151 return rwnd; 2152 } 2153 2154 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2155 { 2156 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2157 } 2158 2159 #if IS_ENABLED(CONFIG_SMC) 2160 extern struct static_key_false tcp_have_smc; 2161 #endif 2162 2163 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2164 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2165 void (*cad)(struct sock *sk, u32 ack_seq)); 2166 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2167 2168 #endif 2169 2170 #endif /* _TCP_H */ 2171