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 #define TCP_MIN_SND_MSS 48 55 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 56 57 /* 58 * Never offer a window over 32767 without using window scaling. Some 59 * poor stacks do signed 16bit maths! 60 */ 61 #define MAX_TCP_WINDOW 32767U 62 63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 64 #define TCP_MIN_MSS 88U 65 66 /* The least MTU to use for probing */ 67 #define TCP_BASE_MSS 1024 68 69 /* probing interval, default to 10 minutes as per RFC4821 */ 70 #define TCP_PROBE_INTERVAL 600 71 72 /* Specify interval when tcp mtu probing will stop */ 73 #define TCP_PROBE_THRESHOLD 8 74 75 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 76 #define TCP_FASTRETRANS_THRESH 3 77 78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 79 #define TCP_MAX_QUICKACKS 16U 80 81 /* Maximal number of window scale according to RFC1323 */ 82 #define TCP_MAX_WSCALE 14U 83 84 /* urg_data states */ 85 #define TCP_URG_VALID 0x0100 86 #define TCP_URG_NOTYET 0x0200 87 #define TCP_URG_READ 0x0400 88 89 #define TCP_RETR1 3 /* 90 * This is how many retries it does before it 91 * tries to figure out if the gateway is 92 * down. Minimal RFC value is 3; it corresponds 93 * to ~3sec-8min depending on RTO. 94 */ 95 96 #define TCP_RETR2 15 /* 97 * This should take at least 98 * 90 minutes to time out. 99 * RFC1122 says that the limit is 100 sec. 100 * 15 is ~13-30min depending on RTO. 101 */ 102 103 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 104 * when active opening a connection. 105 * RFC1122 says the minimum retry MUST 106 * be at least 180secs. Nevertheless 107 * this value is corresponding to 108 * 63secs of retransmission with the 109 * current initial RTO. 110 */ 111 112 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 113 * when passive opening a connection. 114 * This is corresponding to 31secs of 115 * retransmission with the current 116 * initial RTO. 117 */ 118 119 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 120 * state, about 60 seconds */ 121 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 122 /* BSD style FIN_WAIT2 deadlock breaker. 123 * It used to be 3min, new value is 60sec, 124 * to combine FIN-WAIT-2 timeout with 125 * TIME-WAIT timer. 126 */ 127 128 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 129 #if HZ >= 100 130 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 131 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 132 #else 133 #define TCP_DELACK_MIN 4U 134 #define TCP_ATO_MIN 4U 135 #endif 136 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 137 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 138 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 139 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 140 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 141 * used as a fallback RTO for the 142 * initial data transmission if no 143 * valid RTT sample has been acquired, 144 * most likely due to retrans in 3WHS. 145 */ 146 147 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 148 * for local resources. 149 */ 150 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 151 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 152 #define TCP_KEEPALIVE_INTVL (75*HZ) 153 154 #define MAX_TCP_KEEPIDLE 32767 155 #define MAX_TCP_KEEPINTVL 32767 156 #define MAX_TCP_KEEPCNT 127 157 #define MAX_TCP_SYNCNT 127 158 159 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 160 161 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 162 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 163 * after this time. It should be equal 164 * (or greater than) TCP_TIMEWAIT_LEN 165 * to provide reliability equal to one 166 * provided by timewait state. 167 */ 168 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 169 * timestamps. It must be less than 170 * minimal timewait lifetime. 171 */ 172 /* 173 * TCP option 174 */ 175 176 #define TCPOPT_NOP 1 /* Padding */ 177 #define TCPOPT_EOL 0 /* End of options */ 178 #define TCPOPT_MSS 2 /* Segment size negotiating */ 179 #define TCPOPT_WINDOW 3 /* Window scaling */ 180 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 181 #define TCPOPT_SACK 5 /* SACK Block */ 182 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 183 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 184 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 185 #define TCPOPT_EXP 254 /* Experimental */ 186 /* Magic number to be after the option value for sharing TCP 187 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 188 */ 189 #define TCPOPT_FASTOPEN_MAGIC 0xF989 190 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 191 192 /* 193 * TCP option lengths 194 */ 195 196 #define TCPOLEN_MSS 4 197 #define TCPOLEN_WINDOW 3 198 #define TCPOLEN_SACK_PERM 2 199 #define TCPOLEN_TIMESTAMP 10 200 #define TCPOLEN_MD5SIG 18 201 #define TCPOLEN_FASTOPEN_BASE 2 202 #define TCPOLEN_EXP_FASTOPEN_BASE 4 203 #define TCPOLEN_EXP_SMC_BASE 6 204 205 /* But this is what stacks really send out. */ 206 #define TCPOLEN_TSTAMP_ALIGNED 12 207 #define TCPOLEN_WSCALE_ALIGNED 4 208 #define TCPOLEN_SACKPERM_ALIGNED 4 209 #define TCPOLEN_SACK_BASE 2 210 #define TCPOLEN_SACK_BASE_ALIGNED 4 211 #define TCPOLEN_SACK_PERBLOCK 8 212 #define TCPOLEN_MD5SIG_ALIGNED 20 213 #define TCPOLEN_MSS_ALIGNED 4 214 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 215 216 /* Flags in tp->nonagle */ 217 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 218 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 219 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 220 221 /* TCP thin-stream limits */ 222 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 223 224 /* TCP initial congestion window as per rfc6928 */ 225 #define TCP_INIT_CWND 10 226 227 /* Bit Flags for sysctl_tcp_fastopen */ 228 #define TFO_CLIENT_ENABLE 1 229 #define TFO_SERVER_ENABLE 2 230 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 231 232 /* Accept SYN data w/o any cookie option */ 233 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 234 235 /* Force enable TFO on all listeners, i.e., not requiring the 236 * TCP_FASTOPEN socket option. 237 */ 238 #define TFO_SERVER_WO_SOCKOPT1 0x400 239 240 241 /* sysctl variables for tcp */ 242 extern int sysctl_tcp_max_orphans; 243 extern long sysctl_tcp_mem[3]; 244 245 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 246 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 247 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 248 249 extern atomic_long_t tcp_memory_allocated; 250 extern struct percpu_counter tcp_sockets_allocated; 251 extern unsigned long tcp_memory_pressure; 252 253 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 254 static inline bool tcp_under_memory_pressure(const struct sock *sk) 255 { 256 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 257 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 258 return true; 259 260 return tcp_memory_pressure; 261 } 262 /* 263 * The next routines deal with comparing 32 bit unsigned ints 264 * and worry about wraparound (automatic with unsigned arithmetic). 265 */ 266 267 static inline bool before(__u32 seq1, __u32 seq2) 268 { 269 return (__s32)(seq1-seq2) < 0; 270 } 271 #define after(seq2, seq1) before(seq1, seq2) 272 273 /* is s2<=s1<=s3 ? */ 274 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 275 { 276 return seq3 - seq2 >= seq1 - seq2; 277 } 278 279 static inline bool tcp_out_of_memory(struct sock *sk) 280 { 281 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 282 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 283 return true; 284 return false; 285 } 286 287 void sk_forced_mem_schedule(struct sock *sk, int size); 288 289 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 290 { 291 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 292 int orphans = percpu_counter_read_positive(ocp); 293 294 if (orphans << shift > sysctl_tcp_max_orphans) { 295 orphans = percpu_counter_sum_positive(ocp); 296 if (orphans << shift > sysctl_tcp_max_orphans) 297 return true; 298 } 299 return false; 300 } 301 302 bool tcp_check_oom(struct sock *sk, int shift); 303 304 305 extern struct proto tcp_prot; 306 307 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 308 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 309 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 310 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 311 312 void tcp_tasklet_init(void); 313 314 int tcp_v4_err(struct sk_buff *skb, u32); 315 316 void tcp_shutdown(struct sock *sk, int how); 317 318 int tcp_v4_early_demux(struct sk_buff *skb); 319 int tcp_v4_rcv(struct sk_buff *skb); 320 321 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 322 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 323 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 324 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 325 int flags); 326 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 327 size_t size, int flags); 328 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 329 size_t size, int flags); 330 void tcp_release_cb(struct sock *sk); 331 void tcp_wfree(struct sk_buff *skb); 332 void tcp_write_timer_handler(struct sock *sk); 333 void tcp_delack_timer_handler(struct sock *sk); 334 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 335 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 336 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 337 void tcp_rcv_space_adjust(struct sock *sk); 338 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 339 void tcp_twsk_destructor(struct sock *sk); 340 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 341 struct pipe_inode_info *pipe, size_t len, 342 unsigned int flags); 343 344 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 345 static inline void tcp_dec_quickack_mode(struct sock *sk, 346 const unsigned int pkts) 347 { 348 struct inet_connection_sock *icsk = inet_csk(sk); 349 350 if (icsk->icsk_ack.quick) { 351 if (pkts >= icsk->icsk_ack.quick) { 352 icsk->icsk_ack.quick = 0; 353 /* Leaving quickack mode we deflate ATO. */ 354 icsk->icsk_ack.ato = TCP_ATO_MIN; 355 } else 356 icsk->icsk_ack.quick -= pkts; 357 } 358 } 359 360 #define TCP_ECN_OK 1 361 #define TCP_ECN_QUEUE_CWR 2 362 #define TCP_ECN_DEMAND_CWR 4 363 #define TCP_ECN_SEEN 8 364 365 enum tcp_tw_status { 366 TCP_TW_SUCCESS = 0, 367 TCP_TW_RST = 1, 368 TCP_TW_ACK = 2, 369 TCP_TW_SYN = 3 370 }; 371 372 373 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 374 struct sk_buff *skb, 375 const struct tcphdr *th); 376 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 377 struct request_sock *req, bool fastopen, 378 bool *lost_race); 379 int tcp_child_process(struct sock *parent, struct sock *child, 380 struct sk_buff *skb); 381 void tcp_enter_loss(struct sock *sk); 382 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 383 void tcp_clear_retrans(struct tcp_sock *tp); 384 void tcp_update_metrics(struct sock *sk); 385 void tcp_init_metrics(struct sock *sk); 386 void tcp_metrics_init(void); 387 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 388 void tcp_close(struct sock *sk, long timeout); 389 void tcp_init_sock(struct sock *sk); 390 void tcp_init_transfer(struct sock *sk, int bpf_op); 391 __poll_t tcp_poll(struct file *file, struct socket *sock, 392 struct poll_table_struct *wait); 393 int tcp_getsockopt(struct sock *sk, int level, int optname, 394 char __user *optval, int __user *optlen); 395 int tcp_setsockopt(struct sock *sk, int level, int optname, 396 char __user *optval, unsigned int optlen); 397 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, int __user *optlen); 399 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 400 char __user *optval, unsigned int optlen); 401 void tcp_set_keepalive(struct sock *sk, int val); 402 void tcp_syn_ack_timeout(const struct request_sock *req); 403 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 404 int flags, int *addr_len); 405 int tcp_set_rcvlowat(struct sock *sk, int val); 406 void tcp_data_ready(struct sock *sk); 407 #ifdef CONFIG_MMU 408 int tcp_mmap(struct file *file, struct socket *sock, 409 struct vm_area_struct *vma); 410 #endif 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 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 862 { 863 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 864 } 865 866 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 867 { 868 return TCP_SKB_CB(skb)->bpf.sk_redir; 869 } 870 871 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 872 { 873 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 874 } 875 876 #if IS_ENABLED(CONFIG_IPV6) 877 /* This is the variant of inet6_iif() that must be used by TCP, 878 * as TCP moves IP6CB into a different location in skb->cb[] 879 */ 880 static inline int tcp_v6_iif(const struct sk_buff *skb) 881 { 882 return TCP_SKB_CB(skb)->header.h6.iif; 883 } 884 885 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 886 { 887 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 888 889 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 890 } 891 892 /* TCP_SKB_CB reference means this can not be used from early demux */ 893 static inline int tcp_v6_sdif(const struct sk_buff *skb) 894 { 895 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 896 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 897 return TCP_SKB_CB(skb)->header.h6.iif; 898 #endif 899 return 0; 900 } 901 #endif 902 903 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 904 { 905 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 906 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 907 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 908 return true; 909 #endif 910 return false; 911 } 912 913 /* TCP_SKB_CB reference means this can not be used from early demux */ 914 static inline int tcp_v4_sdif(struct sk_buff *skb) 915 { 916 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 917 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 918 return TCP_SKB_CB(skb)->header.h4.iif; 919 #endif 920 return 0; 921 } 922 923 /* Due to TSO, an SKB can be composed of multiple actual 924 * packets. To keep these tracked properly, we use this. 925 */ 926 static inline int tcp_skb_pcount(const struct sk_buff *skb) 927 { 928 return TCP_SKB_CB(skb)->tcp_gso_segs; 929 } 930 931 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 932 { 933 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 934 } 935 936 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 937 { 938 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 939 } 940 941 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 942 static inline int tcp_skb_mss(const struct sk_buff *skb) 943 { 944 return TCP_SKB_CB(skb)->tcp_gso_size; 945 } 946 947 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 948 { 949 return likely(!TCP_SKB_CB(skb)->eor); 950 } 951 952 /* Events passed to congestion control interface */ 953 enum tcp_ca_event { 954 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 955 CA_EVENT_CWND_RESTART, /* congestion window restart */ 956 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 957 CA_EVENT_LOSS, /* loss timeout */ 958 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 959 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 960 }; 961 962 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 963 enum tcp_ca_ack_event_flags { 964 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 965 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 966 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 967 }; 968 969 /* 970 * Interface for adding new TCP congestion control handlers 971 */ 972 #define TCP_CA_NAME_MAX 16 973 #define TCP_CA_MAX 128 974 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 975 976 #define TCP_CA_UNSPEC 0 977 978 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 979 #define TCP_CONG_NON_RESTRICTED 0x1 980 /* Requires ECN/ECT set on all packets */ 981 #define TCP_CONG_NEEDS_ECN 0x2 982 983 union tcp_cc_info; 984 985 struct ack_sample { 986 u32 pkts_acked; 987 s32 rtt_us; 988 u32 in_flight; 989 }; 990 991 /* A rate sample measures the number of (original/retransmitted) data 992 * packets delivered "delivered" over an interval of time "interval_us". 993 * The tcp_rate.c code fills in the rate sample, and congestion 994 * control modules that define a cong_control function to run at the end 995 * of ACK processing can optionally chose to consult this sample when 996 * setting cwnd and pacing rate. 997 * A sample is invalid if "delivered" or "interval_us" is negative. 998 */ 999 struct rate_sample { 1000 u64 prior_mstamp; /* starting timestamp for interval */ 1001 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1002 s32 delivered; /* number of packets delivered over interval */ 1003 long interval_us; /* time for tp->delivered to incr "delivered" */ 1004 u32 snd_interval_us; /* snd interval for delivered packets */ 1005 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1006 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1007 int losses; /* number of packets marked lost upon ACK */ 1008 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1009 u32 prior_in_flight; /* in flight before this ACK */ 1010 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1011 bool is_retrans; /* is sample from retransmission? */ 1012 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1013 }; 1014 1015 struct tcp_congestion_ops { 1016 struct list_head list; 1017 u32 key; 1018 u32 flags; 1019 1020 /* initialize private data (optional) */ 1021 void (*init)(struct sock *sk); 1022 /* cleanup private data (optional) */ 1023 void (*release)(struct sock *sk); 1024 1025 /* return slow start threshold (required) */ 1026 u32 (*ssthresh)(struct sock *sk); 1027 /* do new cwnd calculation (required) */ 1028 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1029 /* call before changing ca_state (optional) */ 1030 void (*set_state)(struct sock *sk, u8 new_state); 1031 /* call when cwnd event occurs (optional) */ 1032 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1033 /* call when ack arrives (optional) */ 1034 void (*in_ack_event)(struct sock *sk, u32 flags); 1035 /* new value of cwnd after loss (required) */ 1036 u32 (*undo_cwnd)(struct sock *sk); 1037 /* hook for packet ack accounting (optional) */ 1038 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1039 /* override sysctl_tcp_min_tso_segs */ 1040 u32 (*min_tso_segs)(struct sock *sk); 1041 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1042 u32 (*sndbuf_expand)(struct sock *sk); 1043 /* call when packets are delivered to update cwnd and pacing rate, 1044 * after all the ca_state processing. (optional) 1045 */ 1046 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1047 /* get info for inet_diag (optional) */ 1048 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1049 union tcp_cc_info *info); 1050 1051 char name[TCP_CA_NAME_MAX]; 1052 struct module *owner; 1053 }; 1054 1055 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1056 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1057 1058 void tcp_assign_congestion_control(struct sock *sk); 1059 void tcp_init_congestion_control(struct sock *sk); 1060 void tcp_cleanup_congestion_control(struct sock *sk); 1061 int tcp_set_default_congestion_control(struct net *net, const char *name); 1062 void tcp_get_default_congestion_control(struct net *net, char *name); 1063 void tcp_get_available_congestion_control(char *buf, size_t len); 1064 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1065 int tcp_set_allowed_congestion_control(char *allowed); 1066 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 1067 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1068 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1069 1070 u32 tcp_reno_ssthresh(struct sock *sk); 1071 u32 tcp_reno_undo_cwnd(struct sock *sk); 1072 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1073 extern struct tcp_congestion_ops tcp_reno; 1074 1075 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1076 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1077 #ifdef CONFIG_INET 1078 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1079 #else 1080 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1081 { 1082 return NULL; 1083 } 1084 #endif 1085 1086 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1087 { 1088 const struct inet_connection_sock *icsk = inet_csk(sk); 1089 1090 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1091 } 1092 1093 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1094 { 1095 struct inet_connection_sock *icsk = inet_csk(sk); 1096 1097 if (icsk->icsk_ca_ops->set_state) 1098 icsk->icsk_ca_ops->set_state(sk, ca_state); 1099 icsk->icsk_ca_state = ca_state; 1100 } 1101 1102 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1103 { 1104 const struct inet_connection_sock *icsk = inet_csk(sk); 1105 1106 if (icsk->icsk_ca_ops->cwnd_event) 1107 icsk->icsk_ca_ops->cwnd_event(sk, event); 1108 } 1109 1110 /* From tcp_rate.c */ 1111 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1112 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1113 struct rate_sample *rs); 1114 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1115 bool is_sack_reneg, struct rate_sample *rs); 1116 void tcp_rate_check_app_limited(struct sock *sk); 1117 1118 /* These functions determine how the current flow behaves in respect of SACK 1119 * handling. SACK is negotiated with the peer, and therefore it can vary 1120 * between different flows. 1121 * 1122 * tcp_is_sack - SACK enabled 1123 * tcp_is_reno - No SACK 1124 */ 1125 static inline int tcp_is_sack(const struct tcp_sock *tp) 1126 { 1127 return likely(tp->rx_opt.sack_ok); 1128 } 1129 1130 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1131 { 1132 return !tcp_is_sack(tp); 1133 } 1134 1135 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1136 { 1137 return tp->sacked_out + tp->lost_out; 1138 } 1139 1140 /* This determines how many packets are "in the network" to the best 1141 * of our knowledge. In many cases it is conservative, but where 1142 * detailed information is available from the receiver (via SACK 1143 * blocks etc.) we can make more aggressive calculations. 1144 * 1145 * Use this for decisions involving congestion control, use just 1146 * tp->packets_out to determine if the send queue is empty or not. 1147 * 1148 * Read this equation as: 1149 * 1150 * "Packets sent once on transmission queue" MINUS 1151 * "Packets left network, but not honestly ACKed yet" PLUS 1152 * "Packets fast retransmitted" 1153 */ 1154 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1155 { 1156 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1157 } 1158 1159 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1160 1161 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1162 { 1163 return tp->snd_cwnd < tp->snd_ssthresh; 1164 } 1165 1166 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1167 { 1168 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1169 } 1170 1171 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1172 { 1173 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1174 (1 << inet_csk(sk)->icsk_ca_state); 1175 } 1176 1177 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1178 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1179 * ssthresh. 1180 */ 1181 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1182 { 1183 const struct tcp_sock *tp = tcp_sk(sk); 1184 1185 if (tcp_in_cwnd_reduction(sk)) 1186 return tp->snd_ssthresh; 1187 else 1188 return max(tp->snd_ssthresh, 1189 ((tp->snd_cwnd >> 1) + 1190 (tp->snd_cwnd >> 2))); 1191 } 1192 1193 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1194 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1195 1196 void tcp_enter_cwr(struct sock *sk); 1197 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1198 1199 /* The maximum number of MSS of available cwnd for which TSO defers 1200 * sending if not using sysctl_tcp_tso_win_divisor. 1201 */ 1202 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1203 { 1204 return 3; 1205 } 1206 1207 /* Returns end sequence number of the receiver's advertised window */ 1208 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1209 { 1210 return tp->snd_una + tp->snd_wnd; 1211 } 1212 1213 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1214 * flexible approach. The RFC suggests cwnd should not be raised unless 1215 * it was fully used previously. And that's exactly what we do in 1216 * congestion avoidance mode. But in slow start we allow cwnd to grow 1217 * as long as the application has used half the cwnd. 1218 * Example : 1219 * cwnd is 10 (IW10), but application sends 9 frames. 1220 * We allow cwnd to reach 18 when all frames are ACKed. 1221 * This check is safe because it's as aggressive as slow start which already 1222 * risks 100% overshoot. The advantage is that we discourage application to 1223 * either send more filler packets or data to artificially blow up the cwnd 1224 * usage, and allow application-limited process to probe bw more aggressively. 1225 */ 1226 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1227 { 1228 const struct tcp_sock *tp = tcp_sk(sk); 1229 1230 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1231 if (tcp_in_slow_start(tp)) 1232 return tp->snd_cwnd < 2 * tp->max_packets_out; 1233 1234 return tp->is_cwnd_limited; 1235 } 1236 1237 /* BBR congestion control needs pacing. 1238 * Same remark for SO_MAX_PACING_RATE. 1239 * sch_fq packet scheduler is efficiently handling pacing, 1240 * but is not always installed/used. 1241 * Return true if TCP stack should pace packets itself. 1242 */ 1243 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1244 { 1245 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1246 } 1247 1248 /* Return in jiffies the delay before one skb is sent. 1249 * If @skb is NULL, we look at EDT for next packet being sent on the socket. 1250 */ 1251 static inline unsigned long tcp_pacing_delay(const struct sock *sk, 1252 const struct sk_buff *skb) 1253 { 1254 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns; 1255 1256 pacing_delay -= tcp_sk(sk)->tcp_clock_cache; 1257 1258 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0; 1259 } 1260 1261 static inline void tcp_reset_xmit_timer(struct sock *sk, 1262 const int what, 1263 unsigned long when, 1264 const unsigned long max_when, 1265 const struct sk_buff *skb) 1266 { 1267 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb), 1268 max_when); 1269 } 1270 1271 /* Something is really bad, we could not queue an additional packet, 1272 * because qdisc is full or receiver sent a 0 window, or we are paced. 1273 * We do not want to add fuel to the fire, or abort too early, 1274 * so make sure the timer we arm now is at least 200ms in the future, 1275 * regardless of current icsk_rto value (as it could be ~2ms) 1276 */ 1277 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1278 { 1279 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1280 } 1281 1282 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1283 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1284 unsigned long max_when) 1285 { 1286 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1287 1288 return (unsigned long)min_t(u64, when, max_when); 1289 } 1290 1291 static inline void tcp_check_probe_timer(struct sock *sk) 1292 { 1293 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1294 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1295 tcp_probe0_base(sk), TCP_RTO_MAX, 1296 NULL); 1297 } 1298 1299 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1300 { 1301 tp->snd_wl1 = seq; 1302 } 1303 1304 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1305 { 1306 tp->snd_wl1 = seq; 1307 } 1308 1309 /* 1310 * Calculate(/check) TCP checksum 1311 */ 1312 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1313 __be32 daddr, __wsum base) 1314 { 1315 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1316 } 1317 1318 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1319 { 1320 return !skb_csum_unnecessary(skb) && 1321 __skb_checksum_complete(skb); 1322 } 1323 1324 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1325 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1326 void tcp_set_state(struct sock *sk, int state); 1327 void tcp_done(struct sock *sk); 1328 int tcp_abort(struct sock *sk, int err); 1329 1330 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1331 { 1332 rx_opt->dsack = 0; 1333 rx_opt->num_sacks = 0; 1334 } 1335 1336 u32 tcp_default_init_rwnd(u32 mss); 1337 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1338 1339 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1340 { 1341 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1342 struct tcp_sock *tp = tcp_sk(sk); 1343 s32 delta; 1344 1345 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1346 ca_ops->cong_control) 1347 return; 1348 delta = tcp_jiffies32 - tp->lsndtime; 1349 if (delta > inet_csk(sk)->icsk_rto) 1350 tcp_cwnd_restart(sk, delta); 1351 } 1352 1353 /* Determine a window scaling and initial window to offer. */ 1354 void tcp_select_initial_window(const struct sock *sk, int __space, 1355 __u32 mss, __u32 *rcv_wnd, 1356 __u32 *window_clamp, int wscale_ok, 1357 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1358 1359 static inline int tcp_win_from_space(const struct sock *sk, int space) 1360 { 1361 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1362 1363 return tcp_adv_win_scale <= 0 ? 1364 (space>>(-tcp_adv_win_scale)) : 1365 space - (space>>tcp_adv_win_scale); 1366 } 1367 1368 /* Note: caller must be prepared to deal with negative returns */ 1369 static inline int tcp_space(const struct sock *sk) 1370 { 1371 return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len - 1372 atomic_read(&sk->sk_rmem_alloc)); 1373 } 1374 1375 static inline int tcp_full_space(const struct sock *sk) 1376 { 1377 return tcp_win_from_space(sk, sk->sk_rcvbuf); 1378 } 1379 1380 extern void tcp_openreq_init_rwin(struct request_sock *req, 1381 const struct sock *sk_listener, 1382 const struct dst_entry *dst); 1383 1384 void tcp_enter_memory_pressure(struct sock *sk); 1385 void tcp_leave_memory_pressure(struct sock *sk); 1386 1387 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1388 { 1389 struct net *net = sock_net((struct sock *)tp); 1390 1391 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1392 } 1393 1394 static inline int keepalive_time_when(const struct tcp_sock *tp) 1395 { 1396 struct net *net = sock_net((struct sock *)tp); 1397 1398 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1399 } 1400 1401 static inline int keepalive_probes(const struct tcp_sock *tp) 1402 { 1403 struct net *net = sock_net((struct sock *)tp); 1404 1405 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1406 } 1407 1408 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1409 { 1410 const struct inet_connection_sock *icsk = &tp->inet_conn; 1411 1412 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1413 tcp_jiffies32 - tp->rcv_tstamp); 1414 } 1415 1416 static inline int tcp_fin_time(const struct sock *sk) 1417 { 1418 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1419 const int rto = inet_csk(sk)->icsk_rto; 1420 1421 if (fin_timeout < (rto << 2) - (rto >> 1)) 1422 fin_timeout = (rto << 2) - (rto >> 1); 1423 1424 return fin_timeout; 1425 } 1426 1427 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1428 int paws_win) 1429 { 1430 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1431 return true; 1432 if (unlikely(!time_before32(ktime_get_seconds(), 1433 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1434 return true; 1435 /* 1436 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1437 * then following tcp messages have valid values. Ignore 0 value, 1438 * or else 'negative' tsval might forbid us to accept their packets. 1439 */ 1440 if (!rx_opt->ts_recent) 1441 return true; 1442 return false; 1443 } 1444 1445 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1446 int rst) 1447 { 1448 if (tcp_paws_check(rx_opt, 0)) 1449 return false; 1450 1451 /* RST segments are not recommended to carry timestamp, 1452 and, if they do, it is recommended to ignore PAWS because 1453 "their cleanup function should take precedence over timestamps." 1454 Certainly, it is mistake. It is necessary to understand the reasons 1455 of this constraint to relax it: if peer reboots, clock may go 1456 out-of-sync and half-open connections will not be reset. 1457 Actually, the problem would be not existing if all 1458 the implementations followed draft about maintaining clock 1459 via reboots. Linux-2.2 DOES NOT! 1460 1461 However, we can relax time bounds for RST segments to MSL. 1462 */ 1463 if (rst && !time_before32(ktime_get_seconds(), 1464 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1465 return false; 1466 return true; 1467 } 1468 1469 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1470 int mib_idx, u32 *last_oow_ack_time); 1471 1472 static inline void tcp_mib_init(struct net *net) 1473 { 1474 /* See RFC 2012 */ 1475 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1476 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1477 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1478 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1479 } 1480 1481 /* from STCP */ 1482 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1483 { 1484 tp->lost_skb_hint = NULL; 1485 } 1486 1487 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1488 { 1489 tcp_clear_retrans_hints_partial(tp); 1490 tp->retransmit_skb_hint = NULL; 1491 } 1492 1493 union tcp_md5_addr { 1494 struct in_addr a4; 1495 #if IS_ENABLED(CONFIG_IPV6) 1496 struct in6_addr a6; 1497 #endif 1498 }; 1499 1500 /* - key database */ 1501 struct tcp_md5sig_key { 1502 struct hlist_node node; 1503 u8 keylen; 1504 u8 family; /* AF_INET or AF_INET6 */ 1505 union tcp_md5_addr addr; 1506 u8 prefixlen; 1507 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1508 struct rcu_head rcu; 1509 }; 1510 1511 /* - sock block */ 1512 struct tcp_md5sig_info { 1513 struct hlist_head head; 1514 struct rcu_head rcu; 1515 }; 1516 1517 /* - pseudo header */ 1518 struct tcp4_pseudohdr { 1519 __be32 saddr; 1520 __be32 daddr; 1521 __u8 pad; 1522 __u8 protocol; 1523 __be16 len; 1524 }; 1525 1526 struct tcp6_pseudohdr { 1527 struct in6_addr saddr; 1528 struct in6_addr daddr; 1529 __be32 len; 1530 __be32 protocol; /* including padding */ 1531 }; 1532 1533 union tcp_md5sum_block { 1534 struct tcp4_pseudohdr ip4; 1535 #if IS_ENABLED(CONFIG_IPV6) 1536 struct tcp6_pseudohdr ip6; 1537 #endif 1538 }; 1539 1540 /* - pool: digest algorithm, hash description and scratch buffer */ 1541 struct tcp_md5sig_pool { 1542 struct ahash_request *md5_req; 1543 void *scratch; 1544 }; 1545 1546 /* - functions */ 1547 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1548 const struct sock *sk, const struct sk_buff *skb); 1549 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1550 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1551 gfp_t gfp); 1552 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1553 int family, u8 prefixlen); 1554 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1555 const struct sock *addr_sk); 1556 1557 #ifdef CONFIG_TCP_MD5SIG 1558 #include <linux/jump_label.h> 1559 extern struct static_key_false tcp_md5_needed; 1560 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, 1561 const union tcp_md5_addr *addr, 1562 int family); 1563 static inline struct tcp_md5sig_key * 1564 tcp_md5_do_lookup(const struct sock *sk, 1565 const union tcp_md5_addr *addr, 1566 int family) 1567 { 1568 if (!static_branch_unlikely(&tcp_md5_needed)) 1569 return NULL; 1570 return __tcp_md5_do_lookup(sk, addr, family); 1571 } 1572 1573 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1574 #else 1575 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1576 const union tcp_md5_addr *addr, 1577 int family) 1578 { 1579 return NULL; 1580 } 1581 #define tcp_twsk_md5_key(twsk) NULL 1582 #endif 1583 1584 bool tcp_alloc_md5sig_pool(void); 1585 1586 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1587 static inline void tcp_put_md5sig_pool(void) 1588 { 1589 local_bh_enable(); 1590 } 1591 1592 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1593 unsigned int header_len); 1594 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1595 const struct tcp_md5sig_key *key); 1596 1597 /* From tcp_fastopen.c */ 1598 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1599 struct tcp_fastopen_cookie *cookie); 1600 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1601 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1602 u16 try_exp); 1603 struct tcp_fastopen_request { 1604 /* Fast Open cookie. Size 0 means a cookie request */ 1605 struct tcp_fastopen_cookie cookie; 1606 struct msghdr *data; /* data in MSG_FASTOPEN */ 1607 size_t size; 1608 int copied; /* queued in tcp_connect() */ 1609 struct ubuf_info *uarg; 1610 }; 1611 void tcp_free_fastopen_req(struct tcp_sock *tp); 1612 void tcp_fastopen_destroy_cipher(struct sock *sk); 1613 void tcp_fastopen_ctx_destroy(struct net *net); 1614 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1615 void *key, unsigned int len); 1616 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1617 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1618 struct request_sock *req, 1619 struct tcp_fastopen_cookie *foc, 1620 const struct dst_entry *dst); 1621 void tcp_fastopen_init_key_once(struct net *net); 1622 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1623 struct tcp_fastopen_cookie *cookie); 1624 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1625 #define TCP_FASTOPEN_KEY_LENGTH 16 1626 1627 /* Fastopen key context */ 1628 struct tcp_fastopen_context { 1629 struct crypto_cipher *tfm; 1630 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1631 struct rcu_head rcu; 1632 }; 1633 1634 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1635 void tcp_fastopen_active_disable(struct sock *sk); 1636 bool tcp_fastopen_active_should_disable(struct sock *sk); 1637 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1638 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1639 1640 /* Latencies incurred by various limits for a sender. They are 1641 * chronograph-like stats that are mutually exclusive. 1642 */ 1643 enum tcp_chrono { 1644 TCP_CHRONO_UNSPEC, 1645 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1646 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1647 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1648 __TCP_CHRONO_MAX, 1649 }; 1650 1651 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1652 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1653 1654 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1655 * the same memory storage than skb->destructor/_skb_refdst 1656 */ 1657 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1658 { 1659 skb->destructor = NULL; 1660 skb->_skb_refdst = 0UL; 1661 } 1662 1663 #define tcp_skb_tsorted_save(skb) { \ 1664 unsigned long _save = skb->_skb_refdst; \ 1665 skb->_skb_refdst = 0UL; 1666 1667 #define tcp_skb_tsorted_restore(skb) \ 1668 skb->_skb_refdst = _save; \ 1669 } 1670 1671 void tcp_write_queue_purge(struct sock *sk); 1672 1673 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1674 { 1675 return skb_rb_first(&sk->tcp_rtx_queue); 1676 } 1677 1678 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1679 { 1680 return skb_peek(&sk->sk_write_queue); 1681 } 1682 1683 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1684 { 1685 return skb_peek_tail(&sk->sk_write_queue); 1686 } 1687 1688 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1689 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1690 1691 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1692 { 1693 return skb_peek(&sk->sk_write_queue); 1694 } 1695 1696 static inline bool tcp_skb_is_last(const struct sock *sk, 1697 const struct sk_buff *skb) 1698 { 1699 return skb_queue_is_last(&sk->sk_write_queue, skb); 1700 } 1701 1702 static inline bool tcp_write_queue_empty(const struct sock *sk) 1703 { 1704 return skb_queue_empty(&sk->sk_write_queue); 1705 } 1706 1707 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1708 { 1709 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1710 } 1711 1712 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1713 { 1714 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1715 } 1716 1717 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1718 { 1719 __skb_queue_tail(&sk->sk_write_queue, skb); 1720 1721 /* Queue it, remembering where we must start sending. */ 1722 if (sk->sk_write_queue.next == skb) 1723 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1724 } 1725 1726 /* Insert new before skb on the write queue of sk. */ 1727 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1728 struct sk_buff *skb, 1729 struct sock *sk) 1730 { 1731 __skb_queue_before(&sk->sk_write_queue, skb, new); 1732 } 1733 1734 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1735 { 1736 tcp_skb_tsorted_anchor_cleanup(skb); 1737 __skb_unlink(skb, &sk->sk_write_queue); 1738 } 1739 1740 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1741 1742 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1743 { 1744 tcp_skb_tsorted_anchor_cleanup(skb); 1745 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1746 } 1747 1748 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1749 { 1750 list_del(&skb->tcp_tsorted_anchor); 1751 tcp_rtx_queue_unlink(skb, sk); 1752 sk_wmem_free_skb(sk, skb); 1753 } 1754 1755 static inline void tcp_push_pending_frames(struct sock *sk) 1756 { 1757 if (tcp_send_head(sk)) { 1758 struct tcp_sock *tp = tcp_sk(sk); 1759 1760 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1761 } 1762 } 1763 1764 /* Start sequence of the skb just after the highest skb with SACKed 1765 * bit, valid only if sacked_out > 0 or when the caller has ensured 1766 * validity by itself. 1767 */ 1768 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1769 { 1770 if (!tp->sacked_out) 1771 return tp->snd_una; 1772 1773 if (tp->highest_sack == NULL) 1774 return tp->snd_nxt; 1775 1776 return TCP_SKB_CB(tp->highest_sack)->seq; 1777 } 1778 1779 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1780 { 1781 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1782 } 1783 1784 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1785 { 1786 return tcp_sk(sk)->highest_sack; 1787 } 1788 1789 static inline void tcp_highest_sack_reset(struct sock *sk) 1790 { 1791 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1792 } 1793 1794 /* Called when old skb is about to be deleted and replaced by new skb */ 1795 static inline void tcp_highest_sack_replace(struct sock *sk, 1796 struct sk_buff *old, 1797 struct sk_buff *new) 1798 { 1799 if (old == tcp_highest_sack(sk)) 1800 tcp_sk(sk)->highest_sack = new; 1801 } 1802 1803 /* This helper checks if socket has IP_TRANSPARENT set */ 1804 static inline bool inet_sk_transparent(const struct sock *sk) 1805 { 1806 switch (sk->sk_state) { 1807 case TCP_TIME_WAIT: 1808 return inet_twsk(sk)->tw_transparent; 1809 case TCP_NEW_SYN_RECV: 1810 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1811 } 1812 return inet_sk(sk)->transparent; 1813 } 1814 1815 /* Determines whether this is a thin stream (which may suffer from 1816 * increased latency). Used to trigger latency-reducing mechanisms. 1817 */ 1818 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1819 { 1820 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1821 } 1822 1823 /* /proc */ 1824 enum tcp_seq_states { 1825 TCP_SEQ_STATE_LISTENING, 1826 TCP_SEQ_STATE_ESTABLISHED, 1827 }; 1828 1829 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1830 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1831 void tcp_seq_stop(struct seq_file *seq, void *v); 1832 1833 struct tcp_seq_afinfo { 1834 sa_family_t family; 1835 }; 1836 1837 struct tcp_iter_state { 1838 struct seq_net_private p; 1839 enum tcp_seq_states state; 1840 struct sock *syn_wait_sk; 1841 int bucket, offset, sbucket, num; 1842 loff_t last_pos; 1843 }; 1844 1845 extern struct request_sock_ops tcp_request_sock_ops; 1846 extern struct request_sock_ops tcp6_request_sock_ops; 1847 1848 void tcp_v4_destroy_sock(struct sock *sk); 1849 1850 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1851 netdev_features_t features); 1852 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1853 int tcp_gro_complete(struct sk_buff *skb); 1854 1855 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1856 1857 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1858 { 1859 struct net *net = sock_net((struct sock *)tp); 1860 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1861 } 1862 1863 /* @wake is one when sk_stream_write_space() calls us. 1864 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1865 * This mimics the strategy used in sock_def_write_space(). 1866 */ 1867 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1868 { 1869 const struct tcp_sock *tp = tcp_sk(sk); 1870 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1871 1872 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1873 } 1874 1875 #ifdef CONFIG_PROC_FS 1876 int tcp4_proc_init(void); 1877 void tcp4_proc_exit(void); 1878 #endif 1879 1880 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1881 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1882 const struct tcp_request_sock_ops *af_ops, 1883 struct sock *sk, struct sk_buff *skb); 1884 1885 /* TCP af-specific functions */ 1886 struct tcp_sock_af_ops { 1887 #ifdef CONFIG_TCP_MD5SIG 1888 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1889 const struct sock *addr_sk); 1890 int (*calc_md5_hash)(char *location, 1891 const struct tcp_md5sig_key *md5, 1892 const struct sock *sk, 1893 const struct sk_buff *skb); 1894 int (*md5_parse)(struct sock *sk, 1895 int optname, 1896 char __user *optval, 1897 int optlen); 1898 #endif 1899 }; 1900 1901 struct tcp_request_sock_ops { 1902 u16 mss_clamp; 1903 #ifdef CONFIG_TCP_MD5SIG 1904 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1905 const struct sock *addr_sk); 1906 int (*calc_md5_hash) (char *location, 1907 const struct tcp_md5sig_key *md5, 1908 const struct sock *sk, 1909 const struct sk_buff *skb); 1910 #endif 1911 void (*init_req)(struct request_sock *req, 1912 const struct sock *sk_listener, 1913 struct sk_buff *skb); 1914 #ifdef CONFIG_SYN_COOKIES 1915 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1916 __u16 *mss); 1917 #endif 1918 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1919 const struct request_sock *req); 1920 u32 (*init_seq)(const struct sk_buff *skb); 1921 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1922 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1923 struct flowi *fl, struct request_sock *req, 1924 struct tcp_fastopen_cookie *foc, 1925 enum tcp_synack_type synack_type); 1926 }; 1927 1928 #ifdef CONFIG_SYN_COOKIES 1929 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1930 const struct sock *sk, struct sk_buff *skb, 1931 __u16 *mss) 1932 { 1933 tcp_synq_overflow(sk); 1934 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1935 return ops->cookie_init_seq(skb, mss); 1936 } 1937 #else 1938 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1939 const struct sock *sk, struct sk_buff *skb, 1940 __u16 *mss) 1941 { 1942 return 0; 1943 } 1944 #endif 1945 1946 int tcpv4_offload_init(void); 1947 1948 void tcp_v4_init(void); 1949 void tcp_init(void); 1950 1951 /* tcp_recovery.c */ 1952 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 1953 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 1954 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 1955 u32 reo_wnd); 1956 extern void tcp_rack_mark_lost(struct sock *sk); 1957 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1958 u64 xmit_time); 1959 extern void tcp_rack_reo_timeout(struct sock *sk); 1960 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1961 1962 /* At how many usecs into the future should the RTO fire? */ 1963 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1964 { 1965 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1966 u32 rto = inet_csk(sk)->icsk_rto; 1967 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 1968 1969 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1970 } 1971 1972 /* 1973 * Save and compile IPv4 options, return a pointer to it 1974 */ 1975 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1976 struct sk_buff *skb) 1977 { 1978 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1979 struct ip_options_rcu *dopt = NULL; 1980 1981 if (opt->optlen) { 1982 int opt_size = sizeof(*dopt) + opt->optlen; 1983 1984 dopt = kmalloc(opt_size, GFP_ATOMIC); 1985 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1986 kfree(dopt); 1987 dopt = NULL; 1988 } 1989 } 1990 return dopt; 1991 } 1992 1993 /* locally generated TCP pure ACKs have skb->truesize == 2 1994 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1995 * This is much faster than dissecting the packet to find out. 1996 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1997 */ 1998 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1999 { 2000 return skb->truesize == 2; 2001 } 2002 2003 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2004 { 2005 skb->truesize = 2; 2006 } 2007 2008 static inline int tcp_inq(struct sock *sk) 2009 { 2010 struct tcp_sock *tp = tcp_sk(sk); 2011 int answ; 2012 2013 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2014 answ = 0; 2015 } else if (sock_flag(sk, SOCK_URGINLINE) || 2016 !tp->urg_data || 2017 before(tp->urg_seq, tp->copied_seq) || 2018 !before(tp->urg_seq, tp->rcv_nxt)) { 2019 2020 answ = tp->rcv_nxt - tp->copied_seq; 2021 2022 /* Subtract 1, if FIN was received */ 2023 if (answ && sock_flag(sk, SOCK_DONE)) 2024 answ--; 2025 } else { 2026 answ = tp->urg_seq - tp->copied_seq; 2027 } 2028 2029 return answ; 2030 } 2031 2032 int tcp_peek_len(struct socket *sock); 2033 2034 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2035 { 2036 u16 segs_in; 2037 2038 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2039 tp->segs_in += segs_in; 2040 if (skb->len > tcp_hdrlen(skb)) 2041 tp->data_segs_in += segs_in; 2042 } 2043 2044 /* 2045 * TCP listen path runs lockless. 2046 * We forced "struct sock" to be const qualified to make sure 2047 * we don't modify one of its field by mistake. 2048 * Here, we increment sk_drops which is an atomic_t, so we can safely 2049 * make sock writable again. 2050 */ 2051 static inline void tcp_listendrop(const struct sock *sk) 2052 { 2053 atomic_inc(&((struct sock *)sk)->sk_drops); 2054 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2055 } 2056 2057 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2058 2059 /* 2060 * Interface for adding Upper Level Protocols over TCP 2061 */ 2062 2063 #define TCP_ULP_NAME_MAX 16 2064 #define TCP_ULP_MAX 128 2065 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2066 2067 struct tcp_ulp_ops { 2068 struct list_head list; 2069 2070 /* initialize ulp */ 2071 int (*init)(struct sock *sk); 2072 /* cleanup ulp */ 2073 void (*release)(struct sock *sk); 2074 2075 char name[TCP_ULP_NAME_MAX]; 2076 struct module *owner; 2077 }; 2078 int tcp_register_ulp(struct tcp_ulp_ops *type); 2079 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2080 int tcp_set_ulp(struct sock *sk, const char *name); 2081 void tcp_get_available_ulp(char *buf, size_t len); 2082 void tcp_cleanup_ulp(struct sock *sk); 2083 2084 #define MODULE_ALIAS_TCP_ULP(name) \ 2085 __MODULE_INFO(alias, alias_userspace, name); \ 2086 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2087 2088 struct sk_msg; 2089 struct sk_psock; 2090 2091 int tcp_bpf_init(struct sock *sk); 2092 void tcp_bpf_reinit(struct sock *sk); 2093 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2094 int flags); 2095 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 2096 int nonblock, int flags, int *addr_len); 2097 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2098 struct msghdr *msg, int len, int flags); 2099 2100 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2101 * is < 0, then the BPF op failed (for example if the loaded BPF 2102 * program does not support the chosen operation or there is no BPF 2103 * program loaded). 2104 */ 2105 #ifdef CONFIG_BPF 2106 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2107 { 2108 struct bpf_sock_ops_kern sock_ops; 2109 int ret; 2110 2111 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2112 if (sk_fullsock(sk)) { 2113 sock_ops.is_fullsock = 1; 2114 sock_owned_by_me(sk); 2115 } 2116 2117 sock_ops.sk = sk; 2118 sock_ops.op = op; 2119 if (nargs > 0) 2120 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2121 2122 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2123 if (ret == 0) 2124 ret = sock_ops.reply; 2125 else 2126 ret = -1; 2127 return ret; 2128 } 2129 2130 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2131 { 2132 u32 args[2] = {arg1, arg2}; 2133 2134 return tcp_call_bpf(sk, op, 2, args); 2135 } 2136 2137 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2138 u32 arg3) 2139 { 2140 u32 args[3] = {arg1, arg2, arg3}; 2141 2142 return tcp_call_bpf(sk, op, 3, args); 2143 } 2144 2145 #else 2146 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2147 { 2148 return -EPERM; 2149 } 2150 2151 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2152 { 2153 return -EPERM; 2154 } 2155 2156 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2157 u32 arg3) 2158 { 2159 return -EPERM; 2160 } 2161 2162 #endif 2163 2164 static inline u32 tcp_timeout_init(struct sock *sk) 2165 { 2166 int timeout; 2167 2168 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2169 2170 if (timeout <= 0) 2171 timeout = TCP_TIMEOUT_INIT; 2172 return timeout; 2173 } 2174 2175 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2176 { 2177 int rwnd; 2178 2179 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2180 2181 if (rwnd < 0) 2182 rwnd = 0; 2183 return rwnd; 2184 } 2185 2186 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2187 { 2188 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2189 } 2190 2191 #if IS_ENABLED(CONFIG_SMC) 2192 extern struct static_key_false tcp_have_smc; 2193 #endif 2194 2195 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2196 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2197 void (*cad)(struct sock *sk, u32 ack_seq)); 2198 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2199 void clean_acked_data_flush(void); 2200 #endif 2201 2202 #endif /* _TCP_H */ 2203