1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP module. 8 * 9 * Version: @(#)tcp.h 1.0.5 05/23/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 */ 14 #ifndef _TCP_H 15 #define _TCP_H 16 17 #define FASTRETRANS_DEBUG 1 18 19 #include <linux/list.h> 20 #include <linux/tcp.h> 21 #include <linux/bug.h> 22 #include <linux/slab.h> 23 #include <linux/cache.h> 24 #include <linux/percpu.h> 25 #include <linux/skbuff.h> 26 #include <linux/cryptohash.h> 27 #include <linux/kref.h> 28 #include <linux/ktime.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/inet_ecn.h> 41 #include <net/dst.h> 42 43 #include <linux/seq_file.h> 44 #include <linux/memcontrol.h> 45 #include <linux/bpf-cgroup.h> 46 47 extern struct inet_hashinfo tcp_hashinfo; 48 49 extern struct percpu_counter tcp_orphan_count; 50 void tcp_time_wait(struct sock *sk, int state, int timeo); 51 52 #define MAX_TCP_HEADER (128 + MAX_HEADER) 53 #define MAX_TCP_OPTION_SPACE 40 54 55 /* 56 * Never offer a window over 32767 without using window scaling. Some 57 * poor stacks do signed 16bit maths! 58 */ 59 #define MAX_TCP_WINDOW 32767U 60 61 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 62 #define TCP_MIN_MSS 88U 63 64 /* The least MTU to use for probing */ 65 #define TCP_BASE_MSS 1024 66 67 /* probing interval, default to 10 minutes as per RFC4821 */ 68 #define TCP_PROBE_INTERVAL 600 69 70 /* Specify interval when tcp mtu probing will stop */ 71 #define TCP_PROBE_THRESHOLD 8 72 73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 74 #define TCP_FASTRETRANS_THRESH 3 75 76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 77 #define TCP_MAX_QUICKACKS 16U 78 79 /* Maximal number of window scale according to RFC1323 */ 80 #define TCP_MAX_WSCALE 14U 81 82 /* urg_data states */ 83 #define TCP_URG_VALID 0x0100 84 #define TCP_URG_NOTYET 0x0200 85 #define TCP_URG_READ 0x0400 86 87 #define TCP_RETR1 3 /* 88 * This is how many retries it does before it 89 * tries to figure out if the gateway is 90 * down. Minimal RFC value is 3; it corresponds 91 * to ~3sec-8min depending on RTO. 92 */ 93 94 #define TCP_RETR2 15 /* 95 * This should take at least 96 * 90 minutes to time out. 97 * RFC1122 says that the limit is 100 sec. 98 * 15 is ~13-30min depending on RTO. 99 */ 100 101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 102 * when active opening a connection. 103 * RFC1122 says the minimum retry MUST 104 * be at least 180secs. Nevertheless 105 * this value is corresponding to 106 * 63secs of retransmission with the 107 * current initial RTO. 108 */ 109 110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 111 * when passive opening a connection. 112 * This is corresponding to 31secs of 113 * retransmission with the current 114 * initial RTO. 115 */ 116 117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 118 * state, about 60 seconds */ 119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 120 /* BSD style FIN_WAIT2 deadlock breaker. 121 * It used to be 3min, new value is 60sec, 122 * to combine FIN-WAIT-2 timeout with 123 * TIME-WAIT timer. 124 */ 125 126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 127 #if HZ >= 100 128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 129 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 130 #else 131 #define TCP_DELACK_MIN 4U 132 #define TCP_ATO_MIN 4U 133 #endif 134 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 135 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 136 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 137 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 138 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 139 * used as a fallback RTO for the 140 * initial data transmission if no 141 * valid RTT sample has been acquired, 142 * most likely due to retrans in 3WHS. 143 */ 144 145 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 146 * for local resources. 147 */ 148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 150 #define TCP_KEEPALIVE_INTVL (75*HZ) 151 152 #define MAX_TCP_KEEPIDLE 32767 153 #define MAX_TCP_KEEPINTVL 32767 154 #define MAX_TCP_KEEPCNT 127 155 #define MAX_TCP_SYNCNT 127 156 157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 158 159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 161 * after this time. It should be equal 162 * (or greater than) TCP_TIMEWAIT_LEN 163 * to provide reliability equal to one 164 * provided by timewait state. 165 */ 166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 167 * timestamps. It must be less than 168 * minimal timewait lifetime. 169 */ 170 /* 171 * TCP option 172 */ 173 174 #define TCPOPT_NOP 1 /* Padding */ 175 #define TCPOPT_EOL 0 /* End of options */ 176 #define TCPOPT_MSS 2 /* Segment size negotiating */ 177 #define TCPOPT_WINDOW 3 /* Window scaling */ 178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 179 #define TCPOPT_SACK 5 /* SACK Block */ 180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 183 #define TCPOPT_EXP 254 /* Experimental */ 184 /* Magic number to be after the option value for sharing TCP 185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 186 */ 187 #define TCPOPT_FASTOPEN_MAGIC 0xF989 188 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 189 190 /* 191 * TCP option lengths 192 */ 193 194 #define TCPOLEN_MSS 4 195 #define TCPOLEN_WINDOW 3 196 #define TCPOLEN_SACK_PERM 2 197 #define TCPOLEN_TIMESTAMP 10 198 #define TCPOLEN_MD5SIG 18 199 #define TCPOLEN_FASTOPEN_BASE 2 200 #define TCPOLEN_EXP_FASTOPEN_BASE 4 201 #define TCPOLEN_EXP_SMC_BASE 6 202 203 /* But this is what stacks really send out. */ 204 #define TCPOLEN_TSTAMP_ALIGNED 12 205 #define TCPOLEN_WSCALE_ALIGNED 4 206 #define TCPOLEN_SACKPERM_ALIGNED 4 207 #define TCPOLEN_SACK_BASE 2 208 #define TCPOLEN_SACK_BASE_ALIGNED 4 209 #define TCPOLEN_SACK_PERBLOCK 8 210 #define TCPOLEN_MD5SIG_ALIGNED 20 211 #define TCPOLEN_MSS_ALIGNED 4 212 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 213 214 /* Flags in tp->nonagle */ 215 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 216 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 217 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 218 219 /* TCP thin-stream limits */ 220 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 221 222 /* TCP initial congestion window as per rfc6928 */ 223 #define TCP_INIT_CWND 10 224 225 /* Bit Flags for sysctl_tcp_fastopen */ 226 #define TFO_CLIENT_ENABLE 1 227 #define TFO_SERVER_ENABLE 2 228 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 229 230 /* Accept SYN data w/o any cookie option */ 231 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 232 233 /* Force enable TFO on all listeners, i.e., not requiring the 234 * TCP_FASTOPEN socket option. 235 */ 236 #define TFO_SERVER_WO_SOCKOPT1 0x400 237 238 239 /* sysctl variables for tcp */ 240 extern int sysctl_tcp_max_orphans; 241 extern long sysctl_tcp_mem[3]; 242 243 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 244 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 245 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 246 247 extern atomic_long_t tcp_memory_allocated; 248 extern struct percpu_counter tcp_sockets_allocated; 249 extern unsigned long tcp_memory_pressure; 250 251 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 252 static inline bool tcp_under_memory_pressure(const struct sock *sk) 253 { 254 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 255 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 256 return true; 257 258 return tcp_memory_pressure; 259 } 260 /* 261 * The next routines deal with comparing 32 bit unsigned ints 262 * and worry about wraparound (automatic with unsigned arithmetic). 263 */ 264 265 static inline bool before(__u32 seq1, __u32 seq2) 266 { 267 return (__s32)(seq1-seq2) < 0; 268 } 269 #define after(seq2, seq1) before(seq1, seq2) 270 271 /* is s2<=s1<=s3 ? */ 272 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 273 { 274 return seq3 - seq2 >= seq1 - seq2; 275 } 276 277 static inline bool tcp_out_of_memory(struct sock *sk) 278 { 279 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 280 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 281 return true; 282 return false; 283 } 284 285 void sk_forced_mem_schedule(struct sock *sk, int size); 286 287 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 288 { 289 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 290 int orphans = percpu_counter_read_positive(ocp); 291 292 if (orphans << shift > sysctl_tcp_max_orphans) { 293 orphans = percpu_counter_sum_positive(ocp); 294 if (orphans << shift > sysctl_tcp_max_orphans) 295 return true; 296 } 297 return false; 298 } 299 300 bool tcp_check_oom(struct sock *sk, int shift); 301 302 303 extern struct proto tcp_prot; 304 305 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 306 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 307 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 308 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 309 310 void tcp_tasklet_init(void); 311 312 int tcp_v4_err(struct sk_buff *skb, u32); 313 314 void tcp_shutdown(struct sock *sk, int how); 315 316 int tcp_v4_early_demux(struct sk_buff *skb); 317 int tcp_v4_rcv(struct sk_buff *skb); 318 319 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 320 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 321 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 322 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 323 int flags); 324 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 325 size_t size, int flags); 326 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 327 size_t size, int flags); 328 void tcp_release_cb(struct sock *sk); 329 void tcp_wfree(struct sk_buff *skb); 330 void tcp_write_timer_handler(struct sock *sk); 331 void tcp_delack_timer_handler(struct sock *sk); 332 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 333 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 334 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 335 void tcp_rcv_space_adjust(struct sock *sk); 336 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 337 void tcp_twsk_destructor(struct sock *sk); 338 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 339 struct pipe_inode_info *pipe, size_t len, 340 unsigned int flags); 341 342 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 343 static inline void tcp_dec_quickack_mode(struct sock *sk, 344 const unsigned int pkts) 345 { 346 struct inet_connection_sock *icsk = inet_csk(sk); 347 348 if (icsk->icsk_ack.quick) { 349 if (pkts >= icsk->icsk_ack.quick) { 350 icsk->icsk_ack.quick = 0; 351 /* Leaving quickack mode we deflate ATO. */ 352 icsk->icsk_ack.ato = TCP_ATO_MIN; 353 } else 354 icsk->icsk_ack.quick -= pkts; 355 } 356 } 357 358 #define TCP_ECN_OK 1 359 #define TCP_ECN_QUEUE_CWR 2 360 #define TCP_ECN_DEMAND_CWR 4 361 #define TCP_ECN_SEEN 8 362 363 enum tcp_tw_status { 364 TCP_TW_SUCCESS = 0, 365 TCP_TW_RST = 1, 366 TCP_TW_ACK = 2, 367 TCP_TW_SYN = 3 368 }; 369 370 371 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 372 struct sk_buff *skb, 373 const struct tcphdr *th); 374 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 375 struct request_sock *req, bool fastopen, 376 bool *lost_race); 377 int tcp_child_process(struct sock *parent, struct sock *child, 378 struct sk_buff *skb); 379 void tcp_enter_loss(struct sock *sk); 380 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 381 void tcp_clear_retrans(struct tcp_sock *tp); 382 void tcp_update_metrics(struct sock *sk); 383 void tcp_init_metrics(struct sock *sk); 384 void tcp_metrics_init(void); 385 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 386 void tcp_close(struct sock *sk, long timeout); 387 void tcp_init_sock(struct sock *sk); 388 void tcp_init_transfer(struct sock *sk, int bpf_op); 389 __poll_t tcp_poll(struct file *file, struct socket *sock, 390 struct poll_table_struct *wait); 391 int tcp_getsockopt(struct sock *sk, int level, int optname, 392 char __user *optval, int __user *optlen); 393 int tcp_setsockopt(struct sock *sk, int level, int optname, 394 char __user *optval, unsigned int optlen); 395 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 396 char __user *optval, int __user *optlen); 397 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, unsigned int optlen); 399 void tcp_set_keepalive(struct sock *sk, int val); 400 void tcp_syn_ack_timeout(const struct request_sock *req); 401 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 402 int flags, int *addr_len); 403 int tcp_set_rcvlowat(struct sock *sk, int val); 404 void tcp_data_ready(struct sock *sk); 405 #ifdef CONFIG_MMU 406 int tcp_mmap(struct file *file, struct socket *sock, 407 struct vm_area_struct *vma); 408 #endif 409 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 410 struct tcp_options_received *opt_rx, 411 int estab, struct tcp_fastopen_cookie *foc); 412 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 413 414 /* 415 * TCP v4 functions exported for the inet6 API 416 */ 417 418 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 419 void tcp_v4_mtu_reduced(struct sock *sk); 420 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 421 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 422 struct sock *tcp_create_openreq_child(const struct sock *sk, 423 struct request_sock *req, 424 struct sk_buff *skb); 425 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 426 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 427 struct request_sock *req, 428 struct dst_entry *dst, 429 struct request_sock *req_unhash, 430 bool *own_req); 431 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 432 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 433 int tcp_connect(struct sock *sk); 434 enum tcp_synack_type { 435 TCP_SYNACK_NORMAL, 436 TCP_SYNACK_FASTOPEN, 437 TCP_SYNACK_COOKIE, 438 }; 439 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 440 struct request_sock *req, 441 struct tcp_fastopen_cookie *foc, 442 enum tcp_synack_type synack_type); 443 int tcp_disconnect(struct sock *sk, int flags); 444 445 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 446 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 447 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 448 449 /* From syncookies.c */ 450 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 451 struct request_sock *req, 452 struct dst_entry *dst, u32 tsoff); 453 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 454 u32 cookie); 455 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 456 #ifdef CONFIG_SYN_COOKIES 457 458 /* Syncookies use a monotonic timer which increments every 60 seconds. 459 * This counter is used both as a hash input and partially encoded into 460 * the cookie value. A cookie is only validated further if the delta 461 * between the current counter value and the encoded one is less than this, 462 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 463 * the counter advances immediately after a cookie is generated). 464 */ 465 #define MAX_SYNCOOKIE_AGE 2 466 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 467 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 468 469 /* syncookies: remember time of last synqueue overflow 470 * But do not dirty this field too often (once per second is enough) 471 * It is racy as we do not hold a lock, but race is very minor. 472 */ 473 static inline void tcp_synq_overflow(const struct sock *sk) 474 { 475 unsigned int last_overflow; 476 unsigned int now = jiffies; 477 478 if (sk->sk_reuseport) { 479 struct sock_reuseport *reuse; 480 481 reuse = rcu_dereference(sk->sk_reuseport_cb); 482 if (likely(reuse)) { 483 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 484 if (time_after32(now, last_overflow + HZ)) 485 WRITE_ONCE(reuse->synq_overflow_ts, now); 486 return; 487 } 488 } 489 490 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 491 if (time_after32(now, last_overflow + HZ)) 492 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 493 } 494 495 /* syncookies: no recent synqueue overflow on this listening socket? */ 496 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 497 { 498 unsigned int last_overflow; 499 unsigned int now = jiffies; 500 501 if (sk->sk_reuseport) { 502 struct sock_reuseport *reuse; 503 504 reuse = rcu_dereference(sk->sk_reuseport_cb); 505 if (likely(reuse)) { 506 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 507 return time_after32(now, last_overflow + 508 TCP_SYNCOOKIE_VALID); 509 } 510 } 511 512 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 513 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID); 514 } 515 516 static inline u32 tcp_cookie_time(void) 517 { 518 u64 val = get_jiffies_64(); 519 520 do_div(val, TCP_SYNCOOKIE_PERIOD); 521 return val; 522 } 523 524 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 525 u16 *mssp); 526 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 527 u64 cookie_init_timestamp(struct request_sock *req); 528 bool cookie_timestamp_decode(const struct net *net, 529 struct tcp_options_received *opt); 530 bool cookie_ecn_ok(const struct tcp_options_received *opt, 531 const struct net *net, const struct dst_entry *dst); 532 533 /* From net/ipv6/syncookies.c */ 534 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 535 u32 cookie); 536 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 537 538 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 539 const struct tcphdr *th, u16 *mssp); 540 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 541 #endif 542 /* tcp_output.c */ 543 544 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 545 int nonagle); 546 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 547 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 548 void tcp_retransmit_timer(struct sock *sk); 549 void tcp_xmit_retransmit_queue(struct sock *); 550 void tcp_simple_retransmit(struct sock *); 551 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 552 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 553 enum tcp_queue { 554 TCP_FRAG_IN_WRITE_QUEUE, 555 TCP_FRAG_IN_RTX_QUEUE, 556 }; 557 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 558 struct sk_buff *skb, u32 len, 559 unsigned int mss_now, gfp_t gfp); 560 561 void tcp_send_probe0(struct sock *); 562 void tcp_send_partial(struct sock *); 563 int tcp_write_wakeup(struct sock *, int mib); 564 void tcp_send_fin(struct sock *sk); 565 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 566 int tcp_send_synack(struct sock *); 567 void tcp_push_one(struct sock *, unsigned int mss_now); 568 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 569 void tcp_send_ack(struct sock *sk); 570 void tcp_send_delayed_ack(struct sock *sk); 571 void tcp_send_loss_probe(struct sock *sk); 572 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 573 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 574 const struct sk_buff *next_skb); 575 576 /* tcp_input.c */ 577 void tcp_rearm_rto(struct sock *sk); 578 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 579 void tcp_reset(struct sock *sk); 580 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 581 void tcp_fin(struct sock *sk); 582 583 /* tcp_timer.c */ 584 void tcp_init_xmit_timers(struct sock *); 585 static inline void tcp_clear_xmit_timers(struct sock *sk) 586 { 587 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 588 __sock_put(sk); 589 590 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 591 __sock_put(sk); 592 593 inet_csk_clear_xmit_timers(sk); 594 } 595 596 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 597 unsigned int tcp_current_mss(struct sock *sk); 598 599 /* Bound MSS / TSO packet size with the half of the window */ 600 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 601 { 602 int cutoff; 603 604 /* When peer uses tiny windows, there is no use in packetizing 605 * to sub-MSS pieces for the sake of SWS or making sure there 606 * are enough packets in the pipe for fast recovery. 607 * 608 * On the other hand, for extremely large MSS devices, handling 609 * smaller than MSS windows in this way does make sense. 610 */ 611 if (tp->max_window > TCP_MSS_DEFAULT) 612 cutoff = (tp->max_window >> 1); 613 else 614 cutoff = tp->max_window; 615 616 if (cutoff && pktsize > cutoff) 617 return max_t(int, cutoff, 68U - tp->tcp_header_len); 618 else 619 return pktsize; 620 } 621 622 /* tcp.c */ 623 void tcp_get_info(struct sock *, struct tcp_info *); 624 625 /* Read 'sendfile()'-style from a TCP socket */ 626 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 627 sk_read_actor_t recv_actor); 628 629 void tcp_initialize_rcv_mss(struct sock *sk); 630 631 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 632 int tcp_mss_to_mtu(struct sock *sk, int mss); 633 void tcp_mtup_init(struct sock *sk); 634 void tcp_init_buffer_space(struct sock *sk); 635 636 static inline void tcp_bound_rto(const struct sock *sk) 637 { 638 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 639 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 640 } 641 642 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 643 { 644 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 645 } 646 647 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 648 { 649 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 650 ntohl(TCP_FLAG_ACK) | 651 snd_wnd); 652 } 653 654 static inline void tcp_fast_path_on(struct tcp_sock *tp) 655 { 656 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 657 } 658 659 static inline void tcp_fast_path_check(struct sock *sk) 660 { 661 struct tcp_sock *tp = tcp_sk(sk); 662 663 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 664 tp->rcv_wnd && 665 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 666 !tp->urg_data) 667 tcp_fast_path_on(tp); 668 } 669 670 /* Compute the actual rto_min value */ 671 static inline u32 tcp_rto_min(struct sock *sk) 672 { 673 const struct dst_entry *dst = __sk_dst_get(sk); 674 u32 rto_min = TCP_RTO_MIN; 675 676 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 677 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 678 return rto_min; 679 } 680 681 static inline u32 tcp_rto_min_us(struct sock *sk) 682 { 683 return jiffies_to_usecs(tcp_rto_min(sk)); 684 } 685 686 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 687 { 688 return dst_metric_locked(dst, RTAX_CC_ALGO); 689 } 690 691 /* Minimum RTT in usec. ~0 means not available. */ 692 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 693 { 694 return minmax_get(&tp->rtt_min); 695 } 696 697 /* Compute the actual receive window we are currently advertising. 698 * Rcv_nxt can be after the window if our peer push more data 699 * than the offered window. 700 */ 701 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 702 { 703 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 704 705 if (win < 0) 706 win = 0; 707 return (u32) win; 708 } 709 710 /* Choose a new window, without checks for shrinking, and without 711 * scaling applied to the result. The caller does these things 712 * if necessary. This is a "raw" window selection. 713 */ 714 u32 __tcp_select_window(struct sock *sk); 715 716 void tcp_send_window_probe(struct sock *sk); 717 718 /* TCP uses 32bit jiffies to save some space. 719 * Note that this is different from tcp_time_stamp, which 720 * historically has been the same until linux-4.13. 721 */ 722 #define tcp_jiffies32 ((u32)jiffies) 723 724 /* 725 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 726 * It is no longer tied to jiffies, but to 1 ms clock. 727 * Note: double check if you want to use tcp_jiffies32 instead of this. 728 */ 729 #define TCP_TS_HZ 1000 730 731 static inline u64 tcp_clock_ns(void) 732 { 733 return ktime_get_ns(); 734 } 735 736 static inline u64 tcp_clock_us(void) 737 { 738 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 739 } 740 741 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 742 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 743 { 744 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 745 } 746 747 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 748 static inline u32 tcp_time_stamp_raw(void) 749 { 750 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 751 } 752 753 void tcp_mstamp_refresh(struct tcp_sock *tp); 754 755 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 756 { 757 return max_t(s64, t1 - t0, 0); 758 } 759 760 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 761 { 762 return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ); 763 } 764 765 /* provide the departure time in us unit */ 766 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 767 { 768 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 769 } 770 771 772 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 773 774 #define TCPHDR_FIN 0x01 775 #define TCPHDR_SYN 0x02 776 #define TCPHDR_RST 0x04 777 #define TCPHDR_PSH 0x08 778 #define TCPHDR_ACK 0x10 779 #define TCPHDR_URG 0x20 780 #define TCPHDR_ECE 0x40 781 #define TCPHDR_CWR 0x80 782 783 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 784 785 /* This is what the send packet queuing engine uses to pass 786 * TCP per-packet control information to the transmission code. 787 * We also store the host-order sequence numbers in here too. 788 * This is 44 bytes if IPV6 is enabled. 789 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 790 */ 791 struct tcp_skb_cb { 792 __u32 seq; /* Starting sequence number */ 793 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 794 union { 795 /* Note : tcp_tw_isn is used in input path only 796 * (isn chosen by tcp_timewait_state_process()) 797 * 798 * tcp_gso_segs/size are used in write queue only, 799 * cf tcp_skb_pcount()/tcp_skb_mss() 800 */ 801 __u32 tcp_tw_isn; 802 struct { 803 u16 tcp_gso_segs; 804 u16 tcp_gso_size; 805 }; 806 }; 807 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 808 809 __u8 sacked; /* State flags for SACK. */ 810 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 811 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 812 #define TCPCB_LOST 0x04 /* SKB is lost */ 813 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 814 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 815 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 816 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 817 TCPCB_REPAIRED) 818 819 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 820 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 821 eor:1, /* Is skb MSG_EOR marked? */ 822 has_rxtstamp:1, /* SKB has a RX timestamp */ 823 unused:5; 824 __u32 ack_seq; /* Sequence number ACK'd */ 825 union { 826 struct { 827 /* There is space for up to 24 bytes */ 828 __u32 in_flight:30,/* Bytes in flight at transmit */ 829 is_app_limited:1, /* cwnd not fully used? */ 830 unused:1; 831 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 832 __u32 delivered; 833 /* start of send pipeline phase */ 834 u64 first_tx_mstamp; 835 /* when we reached the "delivered" count */ 836 u64 delivered_mstamp; 837 } tx; /* only used for outgoing skbs */ 838 union { 839 struct inet_skb_parm h4; 840 #if IS_ENABLED(CONFIG_IPV6) 841 struct inet6_skb_parm h6; 842 #endif 843 } header; /* For incoming skbs */ 844 struct { 845 __u32 flags; 846 struct sock *sk_redir; 847 void *data_end; 848 } bpf; 849 }; 850 }; 851 852 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 853 854 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 855 { 856 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 857 } 858 859 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 860 { 861 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 862 } 863 864 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 865 { 866 return TCP_SKB_CB(skb)->bpf.sk_redir; 867 } 868 869 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 870 { 871 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 872 } 873 874 #if IS_ENABLED(CONFIG_IPV6) 875 /* This is the variant of inet6_iif() that must be used by TCP, 876 * as TCP moves IP6CB into a different location in skb->cb[] 877 */ 878 static inline int tcp_v6_iif(const struct sk_buff *skb) 879 { 880 return TCP_SKB_CB(skb)->header.h6.iif; 881 } 882 883 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 884 { 885 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 886 887 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 888 } 889 890 /* TCP_SKB_CB reference means this can not be used from early demux */ 891 static inline int tcp_v6_sdif(const struct sk_buff *skb) 892 { 893 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 894 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 895 return TCP_SKB_CB(skb)->header.h6.iif; 896 #endif 897 return 0; 898 } 899 #endif 900 901 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 902 { 903 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 904 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 905 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 906 return true; 907 #endif 908 return false; 909 } 910 911 /* TCP_SKB_CB reference means this can not be used from early demux */ 912 static inline int tcp_v4_sdif(struct sk_buff *skb) 913 { 914 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 915 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 916 return TCP_SKB_CB(skb)->header.h4.iif; 917 #endif 918 return 0; 919 } 920 921 /* Due to TSO, an SKB can be composed of multiple actual 922 * packets. To keep these tracked properly, we use this. 923 */ 924 static inline int tcp_skb_pcount(const struct sk_buff *skb) 925 { 926 return TCP_SKB_CB(skb)->tcp_gso_segs; 927 } 928 929 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 930 { 931 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 932 } 933 934 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 935 { 936 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 937 } 938 939 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 940 static inline int tcp_skb_mss(const struct sk_buff *skb) 941 { 942 return TCP_SKB_CB(skb)->tcp_gso_size; 943 } 944 945 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 946 { 947 return likely(!TCP_SKB_CB(skb)->eor); 948 } 949 950 /* Events passed to congestion control interface */ 951 enum tcp_ca_event { 952 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 953 CA_EVENT_CWND_RESTART, /* congestion window restart */ 954 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 955 CA_EVENT_LOSS, /* loss timeout */ 956 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 957 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 958 }; 959 960 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 961 enum tcp_ca_ack_event_flags { 962 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 963 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 964 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 965 }; 966 967 /* 968 * Interface for adding new TCP congestion control handlers 969 */ 970 #define TCP_CA_NAME_MAX 16 971 #define TCP_CA_MAX 128 972 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 973 974 #define TCP_CA_UNSPEC 0 975 976 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 977 #define TCP_CONG_NON_RESTRICTED 0x1 978 /* Requires ECN/ECT set on all packets */ 979 #define TCP_CONG_NEEDS_ECN 0x2 980 981 union tcp_cc_info; 982 983 struct ack_sample { 984 u32 pkts_acked; 985 s32 rtt_us; 986 u32 in_flight; 987 }; 988 989 /* A rate sample measures the number of (original/retransmitted) data 990 * packets delivered "delivered" over an interval of time "interval_us". 991 * The tcp_rate.c code fills in the rate sample, and congestion 992 * control modules that define a cong_control function to run at the end 993 * of ACK processing can optionally chose to consult this sample when 994 * setting cwnd and pacing rate. 995 * A sample is invalid if "delivered" or "interval_us" is negative. 996 */ 997 struct rate_sample { 998 u64 prior_mstamp; /* starting timestamp for interval */ 999 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1000 s32 delivered; /* number of packets delivered over interval */ 1001 long interval_us; /* time for tp->delivered to incr "delivered" */ 1002 u32 snd_interval_us; /* snd interval for delivered packets */ 1003 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1004 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1005 int losses; /* number of packets marked lost upon ACK */ 1006 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1007 u32 prior_in_flight; /* in flight before this ACK */ 1008 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1009 bool is_retrans; /* is sample from retransmission? */ 1010 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1011 }; 1012 1013 struct tcp_congestion_ops { 1014 struct list_head list; 1015 u32 key; 1016 u32 flags; 1017 1018 /* initialize private data (optional) */ 1019 void (*init)(struct sock *sk); 1020 /* cleanup private data (optional) */ 1021 void (*release)(struct sock *sk); 1022 1023 /* return slow start threshold (required) */ 1024 u32 (*ssthresh)(struct sock *sk); 1025 /* do new cwnd calculation (required) */ 1026 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1027 /* call before changing ca_state (optional) */ 1028 void (*set_state)(struct sock *sk, u8 new_state); 1029 /* call when cwnd event occurs (optional) */ 1030 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1031 /* call when ack arrives (optional) */ 1032 void (*in_ack_event)(struct sock *sk, u32 flags); 1033 /* new value of cwnd after loss (required) */ 1034 u32 (*undo_cwnd)(struct sock *sk); 1035 /* hook for packet ack accounting (optional) */ 1036 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1037 /* override sysctl_tcp_min_tso_segs */ 1038 u32 (*min_tso_segs)(struct sock *sk); 1039 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1040 u32 (*sndbuf_expand)(struct sock *sk); 1041 /* call when packets are delivered to update cwnd and pacing rate, 1042 * after all the ca_state processing. (optional) 1043 */ 1044 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1045 /* get info for inet_diag (optional) */ 1046 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1047 union tcp_cc_info *info); 1048 1049 char name[TCP_CA_NAME_MAX]; 1050 struct module *owner; 1051 }; 1052 1053 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1054 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1055 1056 void tcp_assign_congestion_control(struct sock *sk); 1057 void tcp_init_congestion_control(struct sock *sk); 1058 void tcp_cleanup_congestion_control(struct sock *sk); 1059 int tcp_set_default_congestion_control(struct net *net, const char *name); 1060 void tcp_get_default_congestion_control(struct net *net, char *name); 1061 void tcp_get_available_congestion_control(char *buf, size_t len); 1062 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1063 int tcp_set_allowed_congestion_control(char *allowed); 1064 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1065 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1066 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1067 1068 u32 tcp_reno_ssthresh(struct sock *sk); 1069 u32 tcp_reno_undo_cwnd(struct sock *sk); 1070 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1071 extern struct tcp_congestion_ops tcp_reno; 1072 1073 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1074 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1075 #ifdef CONFIG_INET 1076 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1077 #else 1078 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1079 { 1080 return NULL; 1081 } 1082 #endif 1083 1084 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1085 { 1086 const struct inet_connection_sock *icsk = inet_csk(sk); 1087 1088 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1089 } 1090 1091 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1092 { 1093 struct inet_connection_sock *icsk = inet_csk(sk); 1094 1095 if (icsk->icsk_ca_ops->set_state) 1096 icsk->icsk_ca_ops->set_state(sk, ca_state); 1097 icsk->icsk_ca_state = ca_state; 1098 } 1099 1100 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1101 { 1102 const struct inet_connection_sock *icsk = inet_csk(sk); 1103 1104 if (icsk->icsk_ca_ops->cwnd_event) 1105 icsk->icsk_ca_ops->cwnd_event(sk, event); 1106 } 1107 1108 /* From tcp_rate.c */ 1109 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1110 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1111 struct rate_sample *rs); 1112 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1113 bool is_sack_reneg, struct rate_sample *rs); 1114 void tcp_rate_check_app_limited(struct sock *sk); 1115 1116 /* These functions determine how the current flow behaves in respect of SACK 1117 * handling. SACK is negotiated with the peer, and therefore it can vary 1118 * between different flows. 1119 * 1120 * tcp_is_sack - SACK enabled 1121 * tcp_is_reno - No SACK 1122 */ 1123 static inline int tcp_is_sack(const struct tcp_sock *tp) 1124 { 1125 return likely(tp->rx_opt.sack_ok); 1126 } 1127 1128 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1129 { 1130 return !tcp_is_sack(tp); 1131 } 1132 1133 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1134 { 1135 return tp->sacked_out + tp->lost_out; 1136 } 1137 1138 /* This determines how many packets are "in the network" to the best 1139 * of our knowledge. In many cases it is conservative, but where 1140 * detailed information is available from the receiver (via SACK 1141 * blocks etc.) we can make more aggressive calculations. 1142 * 1143 * Use this for decisions involving congestion control, use just 1144 * tp->packets_out to determine if the send queue is empty or not. 1145 * 1146 * Read this equation as: 1147 * 1148 * "Packets sent once on transmission queue" MINUS 1149 * "Packets left network, but not honestly ACKed yet" PLUS 1150 * "Packets fast retransmitted" 1151 */ 1152 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1153 { 1154 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1155 } 1156 1157 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1158 1159 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1160 { 1161 return tp->snd_cwnd < tp->snd_ssthresh; 1162 } 1163 1164 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1165 { 1166 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1167 } 1168 1169 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1170 { 1171 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1172 (1 << inet_csk(sk)->icsk_ca_state); 1173 } 1174 1175 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1176 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1177 * ssthresh. 1178 */ 1179 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1180 { 1181 const struct tcp_sock *tp = tcp_sk(sk); 1182 1183 if (tcp_in_cwnd_reduction(sk)) 1184 return tp->snd_ssthresh; 1185 else 1186 return max(tp->snd_ssthresh, 1187 ((tp->snd_cwnd >> 1) + 1188 (tp->snd_cwnd >> 2))); 1189 } 1190 1191 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1192 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1193 1194 void tcp_enter_cwr(struct sock *sk); 1195 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1196 1197 /* The maximum number of MSS of available cwnd for which TSO defers 1198 * sending if not using sysctl_tcp_tso_win_divisor. 1199 */ 1200 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1201 { 1202 return 3; 1203 } 1204 1205 /* Returns end sequence number of the receiver's advertised window */ 1206 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1207 { 1208 return tp->snd_una + tp->snd_wnd; 1209 } 1210 1211 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1212 * flexible approach. The RFC suggests cwnd should not be raised unless 1213 * it was fully used previously. And that's exactly what we do in 1214 * congestion avoidance mode. But in slow start we allow cwnd to grow 1215 * as long as the application has used half the cwnd. 1216 * Example : 1217 * cwnd is 10 (IW10), but application sends 9 frames. 1218 * We allow cwnd to reach 18 when all frames are ACKed. 1219 * This check is safe because it's as aggressive as slow start which already 1220 * risks 100% overshoot. The advantage is that we discourage application to 1221 * either send more filler packets or data to artificially blow up the cwnd 1222 * usage, and allow application-limited process to probe bw more aggressively. 1223 */ 1224 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1225 { 1226 const struct tcp_sock *tp = tcp_sk(sk); 1227 1228 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1229 if (tcp_in_slow_start(tp)) 1230 return tp->snd_cwnd < 2 * tp->max_packets_out; 1231 1232 return tp->is_cwnd_limited; 1233 } 1234 1235 /* BBR congestion control needs pacing. 1236 * Same remark for SO_MAX_PACING_RATE. 1237 * sch_fq packet scheduler is efficiently handling pacing, 1238 * but is not always installed/used. 1239 * Return true if TCP stack should pace packets itself. 1240 */ 1241 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1242 { 1243 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1244 } 1245 1246 /* Return in jiffies the delay before one skb is sent. 1247 * If @skb is NULL, we look at EDT for next packet being sent on the socket. 1248 */ 1249 static inline unsigned long tcp_pacing_delay(const struct sock *sk, 1250 const struct sk_buff *skb) 1251 { 1252 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns; 1253 1254 pacing_delay -= tcp_sk(sk)->tcp_clock_cache; 1255 1256 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0; 1257 } 1258 1259 static inline void tcp_reset_xmit_timer(struct sock *sk, 1260 const int what, 1261 unsigned long when, 1262 const unsigned long max_when, 1263 const struct sk_buff *skb) 1264 { 1265 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb), 1266 max_when); 1267 } 1268 1269 /* Something is really bad, we could not queue an additional packet, 1270 * because qdisc is full or receiver sent a 0 window, or we are paced. 1271 * We do not want to add fuel to the fire, or abort too early, 1272 * so make sure the timer we arm now is at least 200ms in the future, 1273 * regardless of current icsk_rto value (as it could be ~2ms) 1274 */ 1275 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1276 { 1277 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1278 } 1279 1280 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1281 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1282 unsigned long max_when) 1283 { 1284 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1285 1286 return (unsigned long)min_t(u64, when, max_when); 1287 } 1288 1289 static inline void tcp_check_probe_timer(struct sock *sk) 1290 { 1291 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1292 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1293 tcp_probe0_base(sk), TCP_RTO_MAX, 1294 NULL); 1295 } 1296 1297 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1298 { 1299 tp->snd_wl1 = seq; 1300 } 1301 1302 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1303 { 1304 tp->snd_wl1 = seq; 1305 } 1306 1307 /* 1308 * Calculate(/check) TCP checksum 1309 */ 1310 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1311 __be32 daddr, __wsum base) 1312 { 1313 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1314 } 1315 1316 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1317 { 1318 return !skb_csum_unnecessary(skb) && 1319 __skb_checksum_complete(skb); 1320 } 1321 1322 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1323 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1324 void tcp_set_state(struct sock *sk, int state); 1325 void tcp_done(struct sock *sk); 1326 int tcp_abort(struct sock *sk, int err); 1327 1328 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1329 { 1330 rx_opt->dsack = 0; 1331 rx_opt->num_sacks = 0; 1332 } 1333 1334 u32 tcp_default_init_rwnd(u32 mss); 1335 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1336 1337 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1338 { 1339 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1340 struct tcp_sock *tp = tcp_sk(sk); 1341 s32 delta; 1342 1343 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1344 ca_ops->cong_control) 1345 return; 1346 delta = tcp_jiffies32 - tp->lsndtime; 1347 if (delta > inet_csk(sk)->icsk_rto) 1348 tcp_cwnd_restart(sk, delta); 1349 } 1350 1351 /* Determine a window scaling and initial window to offer. */ 1352 void tcp_select_initial_window(const struct sock *sk, int __space, 1353 __u32 mss, __u32 *rcv_wnd, 1354 __u32 *window_clamp, int wscale_ok, 1355 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1356 1357 static inline int tcp_win_from_space(const struct sock *sk, int space) 1358 { 1359 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1360 1361 return tcp_adv_win_scale <= 0 ? 1362 (space>>(-tcp_adv_win_scale)) : 1363 space - (space>>tcp_adv_win_scale); 1364 } 1365 1366 /* Note: caller must be prepared to deal with negative returns */ 1367 static inline int tcp_space(const struct sock *sk) 1368 { 1369 return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len - 1370 atomic_read(&sk->sk_rmem_alloc)); 1371 } 1372 1373 static inline int tcp_full_space(const struct sock *sk) 1374 { 1375 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1376 } 1377 1378 extern void tcp_openreq_init_rwin(struct request_sock *req, 1379 const struct sock *sk_listener, 1380 const struct dst_entry *dst); 1381 1382 void tcp_enter_memory_pressure(struct sock *sk); 1383 void tcp_leave_memory_pressure(struct sock *sk); 1384 1385 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1386 { 1387 struct net *net = sock_net((struct sock *)tp); 1388 1389 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1390 } 1391 1392 static inline int keepalive_time_when(const struct tcp_sock *tp) 1393 { 1394 struct net *net = sock_net((struct sock *)tp); 1395 1396 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1397 } 1398 1399 static inline int keepalive_probes(const struct tcp_sock *tp) 1400 { 1401 struct net *net = sock_net((struct sock *)tp); 1402 1403 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1404 } 1405 1406 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1407 { 1408 const struct inet_connection_sock *icsk = &tp->inet_conn; 1409 1410 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1411 tcp_jiffies32 - tp->rcv_tstamp); 1412 } 1413 1414 static inline int tcp_fin_time(const struct sock *sk) 1415 { 1416 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1417 const int rto = inet_csk(sk)->icsk_rto; 1418 1419 if (fin_timeout < (rto << 2) - (rto >> 1)) 1420 fin_timeout = (rto << 2) - (rto >> 1); 1421 1422 return fin_timeout; 1423 } 1424 1425 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1426 int paws_win) 1427 { 1428 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1429 return true; 1430 if (unlikely(!time_before32(ktime_get_seconds(), 1431 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1432 return true; 1433 /* 1434 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1435 * then following tcp messages have valid values. Ignore 0 value, 1436 * or else 'negative' tsval might forbid us to accept their packets. 1437 */ 1438 if (!rx_opt->ts_recent) 1439 return true; 1440 return false; 1441 } 1442 1443 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1444 int rst) 1445 { 1446 if (tcp_paws_check(rx_opt, 0)) 1447 return false; 1448 1449 /* RST segments are not recommended to carry timestamp, 1450 and, if they do, it is recommended to ignore PAWS because 1451 "their cleanup function should take precedence over timestamps." 1452 Certainly, it is mistake. It is necessary to understand the reasons 1453 of this constraint to relax it: if peer reboots, clock may go 1454 out-of-sync and half-open connections will not be reset. 1455 Actually, the problem would be not existing if all 1456 the implementations followed draft about maintaining clock 1457 via reboots. Linux-2.2 DOES NOT! 1458 1459 However, we can relax time bounds for RST segments to MSL. 1460 */ 1461 if (rst && !time_before32(ktime_get_seconds(), 1462 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1463 return false; 1464 return true; 1465 } 1466 1467 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1468 int mib_idx, u32 *last_oow_ack_time); 1469 1470 static inline void tcp_mib_init(struct net *net) 1471 { 1472 /* See RFC 2012 */ 1473 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1474 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1475 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1476 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1477 } 1478 1479 /* from STCP */ 1480 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1481 { 1482 tp->lost_skb_hint = NULL; 1483 } 1484 1485 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1486 { 1487 tcp_clear_retrans_hints_partial(tp); 1488 tp->retransmit_skb_hint = NULL; 1489 } 1490 1491 union tcp_md5_addr { 1492 struct in_addr a4; 1493 #if IS_ENABLED(CONFIG_IPV6) 1494 struct in6_addr a6; 1495 #endif 1496 }; 1497 1498 /* - key database */ 1499 struct tcp_md5sig_key { 1500 struct hlist_node node; 1501 u8 keylen; 1502 u8 family; /* AF_INET or AF_INET6 */ 1503 union tcp_md5_addr addr; 1504 u8 prefixlen; 1505 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1506 struct rcu_head rcu; 1507 }; 1508 1509 /* - sock block */ 1510 struct tcp_md5sig_info { 1511 struct hlist_head head; 1512 struct rcu_head rcu; 1513 }; 1514 1515 /* - pseudo header */ 1516 struct tcp4_pseudohdr { 1517 __be32 saddr; 1518 __be32 daddr; 1519 __u8 pad; 1520 __u8 protocol; 1521 __be16 len; 1522 }; 1523 1524 struct tcp6_pseudohdr { 1525 struct in6_addr saddr; 1526 struct in6_addr daddr; 1527 __be32 len; 1528 __be32 protocol; /* including padding */ 1529 }; 1530 1531 union tcp_md5sum_block { 1532 struct tcp4_pseudohdr ip4; 1533 #if IS_ENABLED(CONFIG_IPV6) 1534 struct tcp6_pseudohdr ip6; 1535 #endif 1536 }; 1537 1538 /* - pool: digest algorithm, hash description and scratch buffer */ 1539 struct tcp_md5sig_pool { 1540 struct ahash_request *md5_req; 1541 void *scratch; 1542 }; 1543 1544 /* - functions */ 1545 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1546 const struct sock *sk, const struct sk_buff *skb); 1547 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1548 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1549 gfp_t gfp); 1550 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1551 int family, u8 prefixlen); 1552 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1553 const struct sock *addr_sk); 1554 1555 #ifdef CONFIG_TCP_MD5SIG 1556 #include <linux/jump_label.h> 1557 extern struct static_key_false tcp_md5_needed; 1558 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, 1559 const union tcp_md5_addr *addr, 1560 int family); 1561 static inline struct tcp_md5sig_key * 1562 tcp_md5_do_lookup(const struct sock *sk, 1563 const union tcp_md5_addr *addr, 1564 int family) 1565 { 1566 if (!static_branch_unlikely(&tcp_md5_needed)) 1567 return NULL; 1568 return __tcp_md5_do_lookup(sk, addr, family); 1569 } 1570 1571 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1572 #else 1573 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1574 const union tcp_md5_addr *addr, 1575 int family) 1576 { 1577 return NULL; 1578 } 1579 #define tcp_twsk_md5_key(twsk) NULL 1580 #endif 1581 1582 bool tcp_alloc_md5sig_pool(void); 1583 1584 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1585 static inline void tcp_put_md5sig_pool(void) 1586 { 1587 local_bh_enable(); 1588 } 1589 1590 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1591 unsigned int header_len); 1592 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1593 const struct tcp_md5sig_key *key); 1594 1595 /* From tcp_fastopen.c */ 1596 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1597 struct tcp_fastopen_cookie *cookie); 1598 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1599 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1600 u16 try_exp); 1601 struct tcp_fastopen_request { 1602 /* Fast Open cookie. Size 0 means a cookie request */ 1603 struct tcp_fastopen_cookie cookie; 1604 struct msghdr *data; /* data in MSG_FASTOPEN */ 1605 size_t size; 1606 int copied; /* queued in tcp_connect() */ 1607 struct ubuf_info *uarg; 1608 }; 1609 void tcp_free_fastopen_req(struct tcp_sock *tp); 1610 void tcp_fastopen_destroy_cipher(struct sock *sk); 1611 void tcp_fastopen_ctx_destroy(struct net *net); 1612 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1613 void *primary_key, void *backup_key, 1614 unsigned int len); 1615 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1616 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1617 struct request_sock *req, 1618 struct tcp_fastopen_cookie *foc, 1619 const struct dst_entry *dst); 1620 void tcp_fastopen_init_key_once(struct net *net); 1621 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1622 struct tcp_fastopen_cookie *cookie); 1623 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1624 #define TCP_FASTOPEN_KEY_LENGTH 16 1625 #define TCP_FASTOPEN_KEY_MAX 2 1626 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1627 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1628 1629 /* Fastopen key context */ 1630 struct tcp_fastopen_context { 1631 struct crypto_cipher *tfm[TCP_FASTOPEN_KEY_MAX]; 1632 __u8 key[TCP_FASTOPEN_KEY_BUF_LENGTH]; 1633 struct rcu_head rcu; 1634 }; 1635 1636 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1637 void tcp_fastopen_active_disable(struct sock *sk); 1638 bool tcp_fastopen_active_should_disable(struct sock *sk); 1639 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1640 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1641 1642 /* Caller needs to wrap with rcu_read_(un)lock() */ 1643 static inline 1644 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1645 { 1646 struct tcp_fastopen_context *ctx; 1647 1648 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1649 if (!ctx) 1650 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1651 return ctx; 1652 } 1653 1654 static inline 1655 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1656 const struct tcp_fastopen_cookie *orig) 1657 { 1658 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1659 orig->len == foc->len && 1660 !memcmp(orig->val, foc->val, foc->len)) 1661 return true; 1662 return false; 1663 } 1664 1665 static inline 1666 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1667 { 1668 if (ctx->tfm[1]) 1669 return 2; 1670 return 1; 1671 } 1672 1673 /* Latencies incurred by various limits for a sender. They are 1674 * chronograph-like stats that are mutually exclusive. 1675 */ 1676 enum tcp_chrono { 1677 TCP_CHRONO_UNSPEC, 1678 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1679 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1680 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1681 __TCP_CHRONO_MAX, 1682 }; 1683 1684 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1685 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1686 1687 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1688 * the same memory storage than skb->destructor/_skb_refdst 1689 */ 1690 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1691 { 1692 skb->destructor = NULL; 1693 skb->_skb_refdst = 0UL; 1694 } 1695 1696 #define tcp_skb_tsorted_save(skb) { \ 1697 unsigned long _save = skb->_skb_refdst; \ 1698 skb->_skb_refdst = 0UL; 1699 1700 #define tcp_skb_tsorted_restore(skb) \ 1701 skb->_skb_refdst = _save; \ 1702 } 1703 1704 void tcp_write_queue_purge(struct sock *sk); 1705 1706 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1707 { 1708 return skb_rb_first(&sk->tcp_rtx_queue); 1709 } 1710 1711 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1712 { 1713 return skb_peek(&sk->sk_write_queue); 1714 } 1715 1716 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1717 { 1718 return skb_peek_tail(&sk->sk_write_queue); 1719 } 1720 1721 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1722 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1723 1724 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1725 { 1726 return skb_peek(&sk->sk_write_queue); 1727 } 1728 1729 static inline bool tcp_skb_is_last(const struct sock *sk, 1730 const struct sk_buff *skb) 1731 { 1732 return skb_queue_is_last(&sk->sk_write_queue, skb); 1733 } 1734 1735 static inline bool tcp_write_queue_empty(const struct sock *sk) 1736 { 1737 return skb_queue_empty(&sk->sk_write_queue); 1738 } 1739 1740 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1741 { 1742 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1743 } 1744 1745 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1746 { 1747 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1748 } 1749 1750 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1751 { 1752 __skb_queue_tail(&sk->sk_write_queue, skb); 1753 1754 /* Queue it, remembering where we must start sending. */ 1755 if (sk->sk_write_queue.next == skb) 1756 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1757 } 1758 1759 /* Insert new before skb on the write queue of sk. */ 1760 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1761 struct sk_buff *skb, 1762 struct sock *sk) 1763 { 1764 __skb_queue_before(&sk->sk_write_queue, skb, new); 1765 } 1766 1767 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1768 { 1769 tcp_skb_tsorted_anchor_cleanup(skb); 1770 __skb_unlink(skb, &sk->sk_write_queue); 1771 } 1772 1773 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1774 1775 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1776 { 1777 tcp_skb_tsorted_anchor_cleanup(skb); 1778 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1779 } 1780 1781 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1782 { 1783 list_del(&skb->tcp_tsorted_anchor); 1784 tcp_rtx_queue_unlink(skb, sk); 1785 sk_wmem_free_skb(sk, skb); 1786 } 1787 1788 static inline void tcp_push_pending_frames(struct sock *sk) 1789 { 1790 if (tcp_send_head(sk)) { 1791 struct tcp_sock *tp = tcp_sk(sk); 1792 1793 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1794 } 1795 } 1796 1797 /* Start sequence of the skb just after the highest skb with SACKed 1798 * bit, valid only if sacked_out > 0 or when the caller has ensured 1799 * validity by itself. 1800 */ 1801 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1802 { 1803 if (!tp->sacked_out) 1804 return tp->snd_una; 1805 1806 if (tp->highest_sack == NULL) 1807 return tp->snd_nxt; 1808 1809 return TCP_SKB_CB(tp->highest_sack)->seq; 1810 } 1811 1812 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1813 { 1814 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1815 } 1816 1817 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1818 { 1819 return tcp_sk(sk)->highest_sack; 1820 } 1821 1822 static inline void tcp_highest_sack_reset(struct sock *sk) 1823 { 1824 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1825 } 1826 1827 /* Called when old skb is about to be deleted and replaced by new skb */ 1828 static inline void tcp_highest_sack_replace(struct sock *sk, 1829 struct sk_buff *old, 1830 struct sk_buff *new) 1831 { 1832 if (old == tcp_highest_sack(sk)) 1833 tcp_sk(sk)->highest_sack = new; 1834 } 1835 1836 /* This helper checks if socket has IP_TRANSPARENT set */ 1837 static inline bool inet_sk_transparent(const struct sock *sk) 1838 { 1839 switch (sk->sk_state) { 1840 case TCP_TIME_WAIT: 1841 return inet_twsk(sk)->tw_transparent; 1842 case TCP_NEW_SYN_RECV: 1843 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1844 } 1845 return inet_sk(sk)->transparent; 1846 } 1847 1848 /* Determines whether this is a thin stream (which may suffer from 1849 * increased latency). Used to trigger latency-reducing mechanisms. 1850 */ 1851 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1852 { 1853 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1854 } 1855 1856 /* /proc */ 1857 enum tcp_seq_states { 1858 TCP_SEQ_STATE_LISTENING, 1859 TCP_SEQ_STATE_ESTABLISHED, 1860 }; 1861 1862 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1863 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1864 void tcp_seq_stop(struct seq_file *seq, void *v); 1865 1866 struct tcp_seq_afinfo { 1867 sa_family_t family; 1868 }; 1869 1870 struct tcp_iter_state { 1871 struct seq_net_private p; 1872 enum tcp_seq_states state; 1873 struct sock *syn_wait_sk; 1874 int bucket, offset, sbucket, num; 1875 loff_t last_pos; 1876 }; 1877 1878 extern struct request_sock_ops tcp_request_sock_ops; 1879 extern struct request_sock_ops tcp6_request_sock_ops; 1880 1881 void tcp_v4_destroy_sock(struct sock *sk); 1882 1883 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1884 netdev_features_t features); 1885 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1886 int tcp_gro_complete(struct sk_buff *skb); 1887 1888 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1889 1890 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1891 { 1892 struct net *net = sock_net((struct sock *)tp); 1893 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1894 } 1895 1896 /* @wake is one when sk_stream_write_space() calls us. 1897 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1898 * This mimics the strategy used in sock_def_write_space(). 1899 */ 1900 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1901 { 1902 const struct tcp_sock *tp = tcp_sk(sk); 1903 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1904 1905 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1906 } 1907 1908 #ifdef CONFIG_PROC_FS 1909 int tcp4_proc_init(void); 1910 void tcp4_proc_exit(void); 1911 #endif 1912 1913 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1914 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1915 const struct tcp_request_sock_ops *af_ops, 1916 struct sock *sk, struct sk_buff *skb); 1917 1918 /* TCP af-specific functions */ 1919 struct tcp_sock_af_ops { 1920 #ifdef CONFIG_TCP_MD5SIG 1921 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1922 const struct sock *addr_sk); 1923 int (*calc_md5_hash)(char *location, 1924 const struct tcp_md5sig_key *md5, 1925 const struct sock *sk, 1926 const struct sk_buff *skb); 1927 int (*md5_parse)(struct sock *sk, 1928 int optname, 1929 char __user *optval, 1930 int optlen); 1931 #endif 1932 }; 1933 1934 struct tcp_request_sock_ops { 1935 u16 mss_clamp; 1936 #ifdef CONFIG_TCP_MD5SIG 1937 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1938 const struct sock *addr_sk); 1939 int (*calc_md5_hash) (char *location, 1940 const struct tcp_md5sig_key *md5, 1941 const struct sock *sk, 1942 const struct sk_buff *skb); 1943 #endif 1944 void (*init_req)(struct request_sock *req, 1945 const struct sock *sk_listener, 1946 struct sk_buff *skb); 1947 #ifdef CONFIG_SYN_COOKIES 1948 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1949 __u16 *mss); 1950 #endif 1951 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1952 const struct request_sock *req); 1953 u32 (*init_seq)(const struct sk_buff *skb); 1954 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1955 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1956 struct flowi *fl, struct request_sock *req, 1957 struct tcp_fastopen_cookie *foc, 1958 enum tcp_synack_type synack_type); 1959 }; 1960 1961 #ifdef CONFIG_SYN_COOKIES 1962 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1963 const struct sock *sk, struct sk_buff *skb, 1964 __u16 *mss) 1965 { 1966 tcp_synq_overflow(sk); 1967 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1968 return ops->cookie_init_seq(skb, mss); 1969 } 1970 #else 1971 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1972 const struct sock *sk, struct sk_buff *skb, 1973 __u16 *mss) 1974 { 1975 return 0; 1976 } 1977 #endif 1978 1979 int tcpv4_offload_init(void); 1980 1981 void tcp_v4_init(void); 1982 void tcp_init(void); 1983 1984 /* tcp_recovery.c */ 1985 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1986 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1987 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1988 u32 reo_wnd); 1989 extern void tcp_rack_mark_lost(struct sock *sk); 1990 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1991 u64 xmit_time); 1992 extern void tcp_rack_reo_timeout(struct sock *sk); 1993 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1994 1995 /* At how many usecs into the future should the RTO fire? */ 1996 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1997 { 1998 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1999 u32 rto = inet_csk(sk)->icsk_rto; 2000 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2001 2002 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2003 } 2004 2005 /* 2006 * Save and compile IPv4 options, return a pointer to it 2007 */ 2008 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2009 struct sk_buff *skb) 2010 { 2011 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2012 struct ip_options_rcu *dopt = NULL; 2013 2014 if (opt->optlen) { 2015 int opt_size = sizeof(*dopt) + opt->optlen; 2016 2017 dopt = kmalloc(opt_size, GFP_ATOMIC); 2018 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2019 kfree(dopt); 2020 dopt = NULL; 2021 } 2022 } 2023 return dopt; 2024 } 2025 2026 /* locally generated TCP pure ACKs have skb->truesize == 2 2027 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2028 * This is much faster than dissecting the packet to find out. 2029 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2030 */ 2031 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2032 { 2033 return skb->truesize == 2; 2034 } 2035 2036 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2037 { 2038 skb->truesize = 2; 2039 } 2040 2041 static inline int tcp_inq(struct sock *sk) 2042 { 2043 struct tcp_sock *tp = tcp_sk(sk); 2044 int answ; 2045 2046 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2047 answ = 0; 2048 } else if (sock_flag(sk, SOCK_URGINLINE) || 2049 !tp->urg_data || 2050 before(tp->urg_seq, tp->copied_seq) || 2051 !before(tp->urg_seq, tp->rcv_nxt)) { 2052 2053 answ = tp->rcv_nxt - tp->copied_seq; 2054 2055 /* Subtract 1, if FIN was received */ 2056 if (answ && sock_flag(sk, SOCK_DONE)) 2057 answ--; 2058 } else { 2059 answ = tp->urg_seq - tp->copied_seq; 2060 } 2061 2062 return answ; 2063 } 2064 2065 int tcp_peek_len(struct socket *sock); 2066 2067 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2068 { 2069 u16 segs_in; 2070 2071 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2072 tp->segs_in += segs_in; 2073 if (skb->len > tcp_hdrlen(skb)) 2074 tp->data_segs_in += segs_in; 2075 } 2076 2077 /* 2078 * TCP listen path runs lockless. 2079 * We forced "struct sock" to be const qualified to make sure 2080 * we don't modify one of its field by mistake. 2081 * Here, we increment sk_drops which is an atomic_t, so we can safely 2082 * make sock writable again. 2083 */ 2084 static inline void tcp_listendrop(const struct sock *sk) 2085 { 2086 atomic_inc(&((struct sock *)sk)->sk_drops); 2087 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2088 } 2089 2090 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2091 2092 /* 2093 * Interface for adding Upper Level Protocols over TCP 2094 */ 2095 2096 #define TCP_ULP_NAME_MAX 16 2097 #define TCP_ULP_MAX 128 2098 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2099 2100 struct tcp_ulp_ops { 2101 struct list_head list; 2102 2103 /* initialize ulp */ 2104 int (*init)(struct sock *sk); 2105 /* cleanup ulp */ 2106 void (*release)(struct sock *sk); 2107 2108 char name[TCP_ULP_NAME_MAX]; 2109 struct module *owner; 2110 }; 2111 int tcp_register_ulp(struct tcp_ulp_ops *type); 2112 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2113 int tcp_set_ulp(struct sock *sk, const char *name); 2114 void tcp_get_available_ulp(char *buf, size_t len); 2115 void tcp_cleanup_ulp(struct sock *sk); 2116 2117 #define MODULE_ALIAS_TCP_ULP(name) \ 2118 __MODULE_INFO(alias, alias_userspace, name); \ 2119 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2120 2121 struct sk_msg; 2122 struct sk_psock; 2123 2124 int tcp_bpf_init(struct sock *sk); 2125 void tcp_bpf_reinit(struct sock *sk); 2126 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2127 int flags); 2128 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 2129 int nonblock, int flags, int *addr_len); 2130 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2131 struct msghdr *msg, int len, int flags); 2132 2133 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2134 * is < 0, then the BPF op failed (for example if the loaded BPF 2135 * program does not support the chosen operation or there is no BPF 2136 * program loaded). 2137 */ 2138 #ifdef CONFIG_BPF 2139 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2140 { 2141 struct bpf_sock_ops_kern sock_ops; 2142 int ret; 2143 2144 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2145 if (sk_fullsock(sk)) { 2146 sock_ops.is_fullsock = 1; 2147 sock_owned_by_me(sk); 2148 } 2149 2150 sock_ops.sk = sk; 2151 sock_ops.op = op; 2152 if (nargs > 0) 2153 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2154 2155 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2156 if (ret == 0) 2157 ret = sock_ops.reply; 2158 else 2159 ret = -1; 2160 return ret; 2161 } 2162 2163 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2164 { 2165 u32 args[2] = {arg1, arg2}; 2166 2167 return tcp_call_bpf(sk, op, 2, args); 2168 } 2169 2170 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2171 u32 arg3) 2172 { 2173 u32 args[3] = {arg1, arg2, arg3}; 2174 2175 return tcp_call_bpf(sk, op, 3, args); 2176 } 2177 2178 #else 2179 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2180 { 2181 return -EPERM; 2182 } 2183 2184 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2185 { 2186 return -EPERM; 2187 } 2188 2189 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2190 u32 arg3) 2191 { 2192 return -EPERM; 2193 } 2194 2195 #endif 2196 2197 static inline u32 tcp_timeout_init(struct sock *sk) 2198 { 2199 int timeout; 2200 2201 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2202 2203 if (timeout <= 0) 2204 timeout = TCP_TIMEOUT_INIT; 2205 return timeout; 2206 } 2207 2208 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2209 { 2210 int rwnd; 2211 2212 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2213 2214 if (rwnd < 0) 2215 rwnd = 0; 2216 return rwnd; 2217 } 2218 2219 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2220 { 2221 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2222 } 2223 2224 #if IS_ENABLED(CONFIG_SMC) 2225 extern struct static_key_false tcp_have_smc; 2226 #endif 2227 2228 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2229 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2230 void (*cad)(struct sock *sk, u32 ack_seq)); 2231 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2232 void clean_acked_data_flush(void); 2233 #endif 2234 2235 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2236 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2237 const struct tcp_sock *tp) 2238 { 2239 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2240 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2241 } 2242 2243 static inline void tcp_set_tx_time(struct sk_buff *skb, 2244 const struct sock *sk) 2245 { 2246 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2247 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2248 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2249 2250 skb->skb_mstamp_ns = tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2251 } 2252 } 2253 2254 #endif /* _TCP_H */ 2255