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/kref.h> 27 #include <linux/ktime.h> 28 29 #include <net/inet_connection_sock.h> 30 #include <net/inet_timewait_sock.h> 31 #include <net/inet_hashtables.h> 32 #include <net/checksum.h> 33 #include <net/request_sock.h> 34 #include <net/sock_reuseport.h> 35 #include <net/sock.h> 36 #include <net/snmp.h> 37 #include <net/ip.h> 38 #include <net/tcp_states.h> 39 #include <net/inet_ecn.h> 40 #include <net/dst.h> 41 #include <net/mptcp.h> 42 43 #include <linux/seq_file.h> 44 #include <linux/memcontrol.h> 45 #include <linux/bpf-cgroup.h> 46 #include <linux/siphash.h> 47 48 extern struct inet_hashinfo tcp_hashinfo; 49 50 extern struct percpu_counter tcp_orphan_count; 51 void tcp_time_wait(struct sock *sk, int state, int timeo); 52 53 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 54 #define MAX_TCP_OPTION_SPACE 40 55 #define TCP_MIN_SND_MSS 48 56 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The initial MTU to use for probing */ 68 #define TCP_BASE_MSS 1024 69 70 /* probing interval, default to 10 minutes as per RFC4821 */ 71 #define TCP_PROBE_INTERVAL 600 72 73 /* Specify interval when tcp mtu probing will stop */ 74 #define TCP_PROBE_THRESHOLD 8 75 76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 77 #define TCP_FASTRETRANS_THRESH 3 78 79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80 #define TCP_MAX_QUICKACKS 16U 81 82 /* Maximal number of window scale according to RFC1323 */ 83 #define TCP_MAX_WSCALE 14U 84 85 /* urg_data states */ 86 #define TCP_URG_VALID 0x0100 87 #define TCP_URG_NOTYET 0x0200 88 #define TCP_URG_READ 0x0400 89 90 #define TCP_RETR1 3 /* 91 * This is how many retries it does before it 92 * tries to figure out if the gateway is 93 * down. Minimal RFC value is 3; it corresponds 94 * to ~3sec-8min depending on RTO. 95 */ 96 97 #define TCP_RETR2 15 /* 98 * This should take at least 99 * 90 minutes to time out. 100 * RFC1122 says that the limit is 100 sec. 101 * 15 is ~13-30min depending on RTO. 102 */ 103 104 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 105 * when active opening a connection. 106 * RFC1122 says the minimum retry MUST 107 * be at least 180secs. Nevertheless 108 * this value is corresponding to 109 * 63secs of retransmission with the 110 * current initial RTO. 111 */ 112 113 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 114 * when passive opening a connection. 115 * This is corresponding to 31secs of 116 * retransmission with the current 117 * initial RTO. 118 */ 119 120 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 121 * state, about 60 seconds */ 122 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 123 /* BSD style FIN_WAIT2 deadlock breaker. 124 * It used to be 3min, new value is 60sec, 125 * to combine FIN-WAIT-2 timeout with 126 * TIME-WAIT timer. 127 */ 128 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 129 130 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 131 #if HZ >= 100 132 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 133 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 134 #else 135 #define TCP_DELACK_MIN 4U 136 #define TCP_ATO_MIN 4U 137 #endif 138 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 139 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 140 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 141 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 142 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 143 * used as a fallback RTO for the 144 * initial data transmission if no 145 * valid RTT sample has been acquired, 146 * most likely due to retrans in 3WHS. 147 */ 148 149 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 150 * for local resources. 151 */ 152 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 153 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 154 #define TCP_KEEPALIVE_INTVL (75*HZ) 155 156 #define MAX_TCP_KEEPIDLE 32767 157 #define MAX_TCP_KEEPINTVL 32767 158 #define MAX_TCP_KEEPCNT 127 159 #define MAX_TCP_SYNCNT 127 160 161 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 162 163 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 164 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 165 * after this time. It should be equal 166 * (or greater than) TCP_TIMEWAIT_LEN 167 * to provide reliability equal to one 168 * provided by timewait state. 169 */ 170 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 171 * timestamps. It must be less than 172 * minimal timewait lifetime. 173 */ 174 /* 175 * TCP option 176 */ 177 178 #define TCPOPT_NOP 1 /* Padding */ 179 #define TCPOPT_EOL 0 /* End of options */ 180 #define TCPOPT_MSS 2 /* Segment size negotiating */ 181 #define TCPOPT_WINDOW 3 /* Window scaling */ 182 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 183 #define TCPOPT_SACK 5 /* SACK Block */ 184 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 185 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 186 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 187 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 188 #define TCPOPT_EXP 254 /* Experimental */ 189 /* Magic number to be after the option value for sharing TCP 190 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 191 */ 192 #define TCPOPT_FASTOPEN_MAGIC 0xF989 193 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 194 195 /* 196 * TCP option lengths 197 */ 198 199 #define TCPOLEN_MSS 4 200 #define TCPOLEN_WINDOW 3 201 #define TCPOLEN_SACK_PERM 2 202 #define TCPOLEN_TIMESTAMP 10 203 #define TCPOLEN_MD5SIG 18 204 #define TCPOLEN_FASTOPEN_BASE 2 205 #define TCPOLEN_EXP_FASTOPEN_BASE 4 206 #define TCPOLEN_EXP_SMC_BASE 6 207 208 /* But this is what stacks really send out. */ 209 #define TCPOLEN_TSTAMP_ALIGNED 12 210 #define TCPOLEN_WSCALE_ALIGNED 4 211 #define TCPOLEN_SACKPERM_ALIGNED 4 212 #define TCPOLEN_SACK_BASE 2 213 #define TCPOLEN_SACK_BASE_ALIGNED 4 214 #define TCPOLEN_SACK_PERBLOCK 8 215 #define TCPOLEN_MD5SIG_ALIGNED 20 216 #define TCPOLEN_MSS_ALIGNED 4 217 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 218 219 /* Flags in tp->nonagle */ 220 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 221 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 222 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 223 224 /* TCP thin-stream limits */ 225 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 226 227 /* TCP initial congestion window as per rfc6928 */ 228 #define TCP_INIT_CWND 10 229 230 /* Bit Flags for sysctl_tcp_fastopen */ 231 #define TFO_CLIENT_ENABLE 1 232 #define TFO_SERVER_ENABLE 2 233 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 234 235 /* Accept SYN data w/o any cookie option */ 236 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 237 238 /* Force enable TFO on all listeners, i.e., not requiring the 239 * TCP_FASTOPEN socket option. 240 */ 241 #define TFO_SERVER_WO_SOCKOPT1 0x400 242 243 244 /* sysctl variables for tcp */ 245 extern int sysctl_tcp_max_orphans; 246 extern long sysctl_tcp_mem[3]; 247 248 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 249 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 250 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 251 252 extern atomic_long_t tcp_memory_allocated; 253 extern struct percpu_counter tcp_sockets_allocated; 254 extern unsigned long tcp_memory_pressure; 255 256 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 257 static inline bool tcp_under_memory_pressure(const struct sock *sk) 258 { 259 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 260 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 261 return true; 262 263 return READ_ONCE(tcp_memory_pressure); 264 } 265 /* 266 * The next routines deal with comparing 32 bit unsigned ints 267 * and worry about wraparound (automatic with unsigned arithmetic). 268 */ 269 270 static inline bool before(__u32 seq1, __u32 seq2) 271 { 272 return (__s32)(seq1-seq2) < 0; 273 } 274 #define after(seq2, seq1) before(seq1, seq2) 275 276 /* is s2<=s1<=s3 ? */ 277 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 278 { 279 return seq3 - seq2 >= seq1 - seq2; 280 } 281 282 static inline bool tcp_out_of_memory(struct sock *sk) 283 { 284 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 285 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 286 return true; 287 return false; 288 } 289 290 void sk_forced_mem_schedule(struct sock *sk, int size); 291 292 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 293 { 294 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 295 int orphans = percpu_counter_read_positive(ocp); 296 297 if (orphans << shift > sysctl_tcp_max_orphans) { 298 orphans = percpu_counter_sum_positive(ocp); 299 if (orphans << shift > sysctl_tcp_max_orphans) 300 return true; 301 } 302 return false; 303 } 304 305 bool tcp_check_oom(struct sock *sk, int shift); 306 307 308 extern struct proto tcp_prot; 309 310 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 311 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 312 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 313 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 314 315 void tcp_tasklet_init(void); 316 317 int tcp_v4_err(struct sk_buff *skb, u32); 318 319 void tcp_shutdown(struct sock *sk, int how); 320 321 int tcp_v4_early_demux(struct sk_buff *skb); 322 int tcp_v4_rcv(struct sk_buff *skb); 323 324 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 325 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 326 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 327 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 328 int flags); 329 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 330 size_t size, int flags); 331 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 332 size_t size, int flags); 333 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 334 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 335 int size_goal); 336 void tcp_release_cb(struct sock *sk); 337 void tcp_wfree(struct sk_buff *skb); 338 void tcp_write_timer_handler(struct sock *sk); 339 void tcp_delack_timer_handler(struct sock *sk); 340 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 341 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 342 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_space_adjust(struct sock *sk); 344 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 345 void tcp_twsk_destructor(struct sock *sk); 346 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 347 struct pipe_inode_info *pipe, size_t len, 348 unsigned int flags); 349 350 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 351 static inline void tcp_dec_quickack_mode(struct sock *sk, 352 const unsigned int pkts) 353 { 354 struct inet_connection_sock *icsk = inet_csk(sk); 355 356 if (icsk->icsk_ack.quick) { 357 if (pkts >= icsk->icsk_ack.quick) { 358 icsk->icsk_ack.quick = 0; 359 /* Leaving quickack mode we deflate ATO. */ 360 icsk->icsk_ack.ato = TCP_ATO_MIN; 361 } else 362 icsk->icsk_ack.quick -= pkts; 363 } 364 } 365 366 #define TCP_ECN_OK 1 367 #define TCP_ECN_QUEUE_CWR 2 368 #define TCP_ECN_DEMAND_CWR 4 369 #define TCP_ECN_SEEN 8 370 371 enum tcp_tw_status { 372 TCP_TW_SUCCESS = 0, 373 TCP_TW_RST = 1, 374 TCP_TW_ACK = 2, 375 TCP_TW_SYN = 3 376 }; 377 378 379 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 380 struct sk_buff *skb, 381 const struct tcphdr *th); 382 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 383 struct request_sock *req, bool fastopen, 384 bool *lost_race); 385 int tcp_child_process(struct sock *parent, struct sock *child, 386 struct sk_buff *skb); 387 void tcp_enter_loss(struct sock *sk); 388 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 389 void tcp_clear_retrans(struct tcp_sock *tp); 390 void tcp_update_metrics(struct sock *sk); 391 void tcp_init_metrics(struct sock *sk); 392 void tcp_metrics_init(void); 393 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 394 void tcp_close(struct sock *sk, long timeout); 395 void tcp_init_sock(struct sock *sk); 396 void tcp_init_transfer(struct sock *sk, int bpf_op); 397 __poll_t tcp_poll(struct file *file, struct socket *sock, 398 struct poll_table_struct *wait); 399 int tcp_getsockopt(struct sock *sk, int level, int optname, 400 char __user *optval, int __user *optlen); 401 int tcp_setsockopt(struct sock *sk, int level, int optname, 402 char __user *optval, unsigned int optlen); 403 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 404 char __user *optval, int __user *optlen); 405 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 406 char __user *optval, unsigned int optlen); 407 void tcp_set_keepalive(struct sock *sk, int val); 408 void tcp_syn_ack_timeout(const struct request_sock *req); 409 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 410 int flags, int *addr_len); 411 int tcp_set_rcvlowat(struct sock *sk, int val); 412 void tcp_data_ready(struct sock *sk); 413 #ifdef CONFIG_MMU 414 int tcp_mmap(struct file *file, struct socket *sock, 415 struct vm_area_struct *vma); 416 #endif 417 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 418 struct tcp_options_received *opt_rx, 419 int estab, struct tcp_fastopen_cookie *foc); 420 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 421 422 /* 423 * BPF SKB-less helpers 424 */ 425 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 426 struct tcphdr *th, u32 *cookie); 427 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 428 struct tcphdr *th, u32 *cookie); 429 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 430 const struct tcp_request_sock_ops *af_ops, 431 struct sock *sk, struct tcphdr *th); 432 /* 433 * TCP v4 functions exported for the inet6 API 434 */ 435 436 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 437 void tcp_v4_mtu_reduced(struct sock *sk); 438 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 439 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 440 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 441 struct sock *tcp_create_openreq_child(const struct sock *sk, 442 struct request_sock *req, 443 struct sk_buff *skb); 444 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 445 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 446 struct request_sock *req, 447 struct dst_entry *dst, 448 struct request_sock *req_unhash, 449 bool *own_req); 450 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 451 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 452 int tcp_connect(struct sock *sk); 453 enum tcp_synack_type { 454 TCP_SYNACK_NORMAL, 455 TCP_SYNACK_FASTOPEN, 456 TCP_SYNACK_COOKIE, 457 }; 458 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 459 struct request_sock *req, 460 struct tcp_fastopen_cookie *foc, 461 enum tcp_synack_type synack_type); 462 int tcp_disconnect(struct sock *sk, int flags); 463 464 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 465 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 466 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 467 468 /* From syncookies.c */ 469 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 470 struct request_sock *req, 471 struct dst_entry *dst, u32 tsoff); 472 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 473 u32 cookie); 474 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 475 #ifdef CONFIG_SYN_COOKIES 476 477 /* Syncookies use a monotonic timer which increments every 60 seconds. 478 * This counter is used both as a hash input and partially encoded into 479 * the cookie value. A cookie is only validated further if the delta 480 * between the current counter value and the encoded one is less than this, 481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 482 * the counter advances immediately after a cookie is generated). 483 */ 484 #define MAX_SYNCOOKIE_AGE 2 485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 487 488 /* syncookies: remember time of last synqueue overflow 489 * But do not dirty this field too often (once per second is enough) 490 * It is racy as we do not hold a lock, but race is very minor. 491 */ 492 static inline void tcp_synq_overflow(const struct sock *sk) 493 { 494 unsigned int last_overflow; 495 unsigned int now = jiffies; 496 497 if (sk->sk_reuseport) { 498 struct sock_reuseport *reuse; 499 500 reuse = rcu_dereference(sk->sk_reuseport_cb); 501 if (likely(reuse)) { 502 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 503 if (!time_between32(now, last_overflow, 504 last_overflow + HZ)) 505 WRITE_ONCE(reuse->synq_overflow_ts, now); 506 return; 507 } 508 } 509 510 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 511 if (!time_between32(now, last_overflow, last_overflow + HZ)) 512 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 513 } 514 515 /* syncookies: no recent synqueue overflow on this listening socket? */ 516 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 517 { 518 unsigned int last_overflow; 519 unsigned int now = jiffies; 520 521 if (sk->sk_reuseport) { 522 struct sock_reuseport *reuse; 523 524 reuse = rcu_dereference(sk->sk_reuseport_cb); 525 if (likely(reuse)) { 526 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 527 return !time_between32(now, last_overflow - HZ, 528 last_overflow + 529 TCP_SYNCOOKIE_VALID); 530 } 531 } 532 533 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 534 535 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 536 * then we're under synflood. However, we have to use 537 * 'last_overflow - HZ' as lower bound. That's because a concurrent 538 * tcp_synq_overflow() could update .ts_recent_stamp after we read 539 * jiffies but before we store .ts_recent_stamp into last_overflow, 540 * which could lead to rejecting a valid syncookie. 541 */ 542 return !time_between32(now, last_overflow - HZ, 543 last_overflow + TCP_SYNCOOKIE_VALID); 544 } 545 546 static inline u32 tcp_cookie_time(void) 547 { 548 u64 val = get_jiffies_64(); 549 550 do_div(val, TCP_SYNCOOKIE_PERIOD); 551 return val; 552 } 553 554 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 555 u16 *mssp); 556 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 557 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 558 bool cookie_timestamp_decode(const struct net *net, 559 struct tcp_options_received *opt); 560 bool cookie_ecn_ok(const struct tcp_options_received *opt, 561 const struct net *net, const struct dst_entry *dst); 562 563 /* From net/ipv6/syncookies.c */ 564 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 565 u32 cookie); 566 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 567 568 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 569 const struct tcphdr *th, u16 *mssp); 570 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 571 #endif 572 /* tcp_output.c */ 573 574 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 575 int nonagle); 576 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 577 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 578 void tcp_retransmit_timer(struct sock *sk); 579 void tcp_xmit_retransmit_queue(struct sock *); 580 void tcp_simple_retransmit(struct sock *); 581 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 582 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 583 enum tcp_queue { 584 TCP_FRAG_IN_WRITE_QUEUE, 585 TCP_FRAG_IN_RTX_QUEUE, 586 }; 587 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 588 struct sk_buff *skb, u32 len, 589 unsigned int mss_now, gfp_t gfp); 590 591 void tcp_send_probe0(struct sock *); 592 void tcp_send_partial(struct sock *); 593 int tcp_write_wakeup(struct sock *, int mib); 594 void tcp_send_fin(struct sock *sk); 595 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 596 int tcp_send_synack(struct sock *); 597 void tcp_push_one(struct sock *, unsigned int mss_now); 598 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 599 void tcp_send_ack(struct sock *sk); 600 void tcp_send_delayed_ack(struct sock *sk); 601 void tcp_send_loss_probe(struct sock *sk); 602 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 603 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 604 const struct sk_buff *next_skb); 605 606 /* tcp_input.c */ 607 void tcp_rearm_rto(struct sock *sk); 608 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 609 void tcp_reset(struct sock *sk); 610 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 611 void tcp_fin(struct sock *sk); 612 613 /* tcp_timer.c */ 614 void tcp_init_xmit_timers(struct sock *); 615 static inline void tcp_clear_xmit_timers(struct sock *sk) 616 { 617 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 618 __sock_put(sk); 619 620 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 621 __sock_put(sk); 622 623 inet_csk_clear_xmit_timers(sk); 624 } 625 626 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 627 unsigned int tcp_current_mss(struct sock *sk); 628 629 /* Bound MSS / TSO packet size with the half of the window */ 630 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 631 { 632 int cutoff; 633 634 /* When peer uses tiny windows, there is no use in packetizing 635 * to sub-MSS pieces for the sake of SWS or making sure there 636 * are enough packets in the pipe for fast recovery. 637 * 638 * On the other hand, for extremely large MSS devices, handling 639 * smaller than MSS windows in this way does make sense. 640 */ 641 if (tp->max_window > TCP_MSS_DEFAULT) 642 cutoff = (tp->max_window >> 1); 643 else 644 cutoff = tp->max_window; 645 646 if (cutoff && pktsize > cutoff) 647 return max_t(int, cutoff, 68U - tp->tcp_header_len); 648 else 649 return pktsize; 650 } 651 652 /* tcp.c */ 653 void tcp_get_info(struct sock *, struct tcp_info *); 654 655 /* Read 'sendfile()'-style from a TCP socket */ 656 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 657 sk_read_actor_t recv_actor); 658 659 void tcp_initialize_rcv_mss(struct sock *sk); 660 661 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 662 int tcp_mss_to_mtu(struct sock *sk, int mss); 663 void tcp_mtup_init(struct sock *sk); 664 665 static inline void tcp_bound_rto(const struct sock *sk) 666 { 667 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 668 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 669 } 670 671 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 672 { 673 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 674 } 675 676 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 677 { 678 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 679 ntohl(TCP_FLAG_ACK) | 680 snd_wnd); 681 } 682 683 static inline void tcp_fast_path_on(struct tcp_sock *tp) 684 { 685 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 686 } 687 688 static inline void tcp_fast_path_check(struct sock *sk) 689 { 690 struct tcp_sock *tp = tcp_sk(sk); 691 692 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 693 tp->rcv_wnd && 694 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 695 !tp->urg_data) 696 tcp_fast_path_on(tp); 697 } 698 699 /* Compute the actual rto_min value */ 700 static inline u32 tcp_rto_min(struct sock *sk) 701 { 702 const struct dst_entry *dst = __sk_dst_get(sk); 703 u32 rto_min = TCP_RTO_MIN; 704 705 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 706 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 707 return rto_min; 708 } 709 710 static inline u32 tcp_rto_min_us(struct sock *sk) 711 { 712 return jiffies_to_usecs(tcp_rto_min(sk)); 713 } 714 715 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 716 { 717 return dst_metric_locked(dst, RTAX_CC_ALGO); 718 } 719 720 /* Minimum RTT in usec. ~0 means not available. */ 721 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 722 { 723 return minmax_get(&tp->rtt_min); 724 } 725 726 /* Compute the actual receive window we are currently advertising. 727 * Rcv_nxt can be after the window if our peer push more data 728 * than the offered window. 729 */ 730 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 731 { 732 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 733 734 if (win < 0) 735 win = 0; 736 return (u32) win; 737 } 738 739 /* Choose a new window, without checks for shrinking, and without 740 * scaling applied to the result. The caller does these things 741 * if necessary. This is a "raw" window selection. 742 */ 743 u32 __tcp_select_window(struct sock *sk); 744 745 void tcp_send_window_probe(struct sock *sk); 746 747 /* TCP uses 32bit jiffies to save some space. 748 * Note that this is different from tcp_time_stamp, which 749 * historically has been the same until linux-4.13. 750 */ 751 #define tcp_jiffies32 ((u32)jiffies) 752 753 /* 754 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 755 * It is no longer tied to jiffies, but to 1 ms clock. 756 * Note: double check if you want to use tcp_jiffies32 instead of this. 757 */ 758 #define TCP_TS_HZ 1000 759 760 static inline u64 tcp_clock_ns(void) 761 { 762 return ktime_get_ns(); 763 } 764 765 static inline u64 tcp_clock_us(void) 766 { 767 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 768 } 769 770 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 771 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 772 { 773 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 774 } 775 776 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 777 static inline u32 tcp_ns_to_ts(u64 ns) 778 { 779 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 780 } 781 782 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 783 static inline u32 tcp_time_stamp_raw(void) 784 { 785 return tcp_ns_to_ts(tcp_clock_ns()); 786 } 787 788 void tcp_mstamp_refresh(struct tcp_sock *tp); 789 790 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 791 { 792 return max_t(s64, t1 - t0, 0); 793 } 794 795 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 796 { 797 return tcp_ns_to_ts(skb->skb_mstamp_ns); 798 } 799 800 /* provide the departure time in us unit */ 801 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 802 { 803 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 804 } 805 806 807 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 808 809 #define TCPHDR_FIN 0x01 810 #define TCPHDR_SYN 0x02 811 #define TCPHDR_RST 0x04 812 #define TCPHDR_PSH 0x08 813 #define TCPHDR_ACK 0x10 814 #define TCPHDR_URG 0x20 815 #define TCPHDR_ECE 0x40 816 #define TCPHDR_CWR 0x80 817 818 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 819 820 /* This is what the send packet queuing engine uses to pass 821 * TCP per-packet control information to the transmission code. 822 * We also store the host-order sequence numbers in here too. 823 * This is 44 bytes if IPV6 is enabled. 824 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 825 */ 826 struct tcp_skb_cb { 827 __u32 seq; /* Starting sequence number */ 828 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 829 union { 830 /* Note : tcp_tw_isn is used in input path only 831 * (isn chosen by tcp_timewait_state_process()) 832 * 833 * tcp_gso_segs/size are used in write queue only, 834 * cf tcp_skb_pcount()/tcp_skb_mss() 835 */ 836 __u32 tcp_tw_isn; 837 struct { 838 u16 tcp_gso_segs; 839 u16 tcp_gso_size; 840 }; 841 }; 842 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 843 844 __u8 sacked; /* State flags for SACK. */ 845 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 846 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 847 #define TCPCB_LOST 0x04 /* SKB is lost */ 848 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 849 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 850 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 851 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 852 TCPCB_REPAIRED) 853 854 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 855 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 856 eor:1, /* Is skb MSG_EOR marked? */ 857 has_rxtstamp:1, /* SKB has a RX timestamp */ 858 unused:5; 859 __u32 ack_seq; /* Sequence number ACK'd */ 860 union { 861 struct { 862 /* There is space for up to 24 bytes */ 863 __u32 in_flight:30,/* Bytes in flight at transmit */ 864 is_app_limited:1, /* cwnd not fully used? */ 865 unused:1; 866 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 867 __u32 delivered; 868 /* start of send pipeline phase */ 869 u64 first_tx_mstamp; 870 /* when we reached the "delivered" count */ 871 u64 delivered_mstamp; 872 } tx; /* only used for outgoing skbs */ 873 union { 874 struct inet_skb_parm h4; 875 #if IS_ENABLED(CONFIG_IPV6) 876 struct inet6_skb_parm h6; 877 #endif 878 } header; /* For incoming skbs */ 879 struct { 880 __u32 flags; 881 struct sock *sk_redir; 882 void *data_end; 883 } bpf; 884 }; 885 }; 886 887 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 888 889 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 890 { 891 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 892 } 893 894 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 895 { 896 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 897 } 898 899 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 900 { 901 return TCP_SKB_CB(skb)->bpf.sk_redir; 902 } 903 904 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 905 { 906 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 907 } 908 909 #if IS_ENABLED(CONFIG_IPV6) 910 /* This is the variant of inet6_iif() that must be used by TCP, 911 * as TCP moves IP6CB into a different location in skb->cb[] 912 */ 913 static inline int tcp_v6_iif(const struct sk_buff *skb) 914 { 915 return TCP_SKB_CB(skb)->header.h6.iif; 916 } 917 918 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 919 { 920 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 921 922 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 923 } 924 925 /* TCP_SKB_CB reference means this can not be used from early demux */ 926 static inline int tcp_v6_sdif(const struct sk_buff *skb) 927 { 928 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 929 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 930 return TCP_SKB_CB(skb)->header.h6.iif; 931 #endif 932 return 0; 933 } 934 #endif 935 936 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 937 { 938 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 939 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 940 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 941 return true; 942 #endif 943 return false; 944 } 945 946 /* TCP_SKB_CB reference means this can not be used from early demux */ 947 static inline int tcp_v4_sdif(struct sk_buff *skb) 948 { 949 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 950 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 951 return TCP_SKB_CB(skb)->header.h4.iif; 952 #endif 953 return 0; 954 } 955 956 /* Due to TSO, an SKB can be composed of multiple actual 957 * packets. To keep these tracked properly, we use this. 958 */ 959 static inline int tcp_skb_pcount(const struct sk_buff *skb) 960 { 961 return TCP_SKB_CB(skb)->tcp_gso_segs; 962 } 963 964 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 965 { 966 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 967 } 968 969 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 970 { 971 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 972 } 973 974 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 975 static inline int tcp_skb_mss(const struct sk_buff *skb) 976 { 977 return TCP_SKB_CB(skb)->tcp_gso_size; 978 } 979 980 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 981 { 982 return likely(!TCP_SKB_CB(skb)->eor); 983 } 984 985 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 986 const struct sk_buff *from) 987 { 988 return likely(tcp_skb_can_collapse_to(to) && 989 mptcp_skb_can_collapse(to, from)); 990 } 991 992 /* Events passed to congestion control interface */ 993 enum tcp_ca_event { 994 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 995 CA_EVENT_CWND_RESTART, /* congestion window restart */ 996 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 997 CA_EVENT_LOSS, /* loss timeout */ 998 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 999 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1000 }; 1001 1002 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1003 enum tcp_ca_ack_event_flags { 1004 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1005 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1006 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1007 }; 1008 1009 /* 1010 * Interface for adding new TCP congestion control handlers 1011 */ 1012 #define TCP_CA_NAME_MAX 16 1013 #define TCP_CA_MAX 128 1014 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1015 1016 #define TCP_CA_UNSPEC 0 1017 1018 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1019 #define TCP_CONG_NON_RESTRICTED 0x1 1020 /* Requires ECN/ECT set on all packets */ 1021 #define TCP_CONG_NEEDS_ECN 0x2 1022 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1023 1024 union tcp_cc_info; 1025 1026 struct ack_sample { 1027 u32 pkts_acked; 1028 s32 rtt_us; 1029 u32 in_flight; 1030 }; 1031 1032 /* A rate sample measures the number of (original/retransmitted) data 1033 * packets delivered "delivered" over an interval of time "interval_us". 1034 * The tcp_rate.c code fills in the rate sample, and congestion 1035 * control modules that define a cong_control function to run at the end 1036 * of ACK processing can optionally chose to consult this sample when 1037 * setting cwnd and pacing rate. 1038 * A sample is invalid if "delivered" or "interval_us" is negative. 1039 */ 1040 struct rate_sample { 1041 u64 prior_mstamp; /* starting timestamp for interval */ 1042 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1043 s32 delivered; /* number of packets delivered over interval */ 1044 long interval_us; /* time for tp->delivered to incr "delivered" */ 1045 u32 snd_interval_us; /* snd interval for delivered packets */ 1046 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1047 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1048 int losses; /* number of packets marked lost upon ACK */ 1049 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1050 u32 prior_in_flight; /* in flight before this ACK */ 1051 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1052 bool is_retrans; /* is sample from retransmission? */ 1053 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1054 }; 1055 1056 struct tcp_congestion_ops { 1057 struct list_head list; 1058 u32 key; 1059 u32 flags; 1060 1061 /* initialize private data (optional) */ 1062 void (*init)(struct sock *sk); 1063 /* cleanup private data (optional) */ 1064 void (*release)(struct sock *sk); 1065 1066 /* return slow start threshold (required) */ 1067 u32 (*ssthresh)(struct sock *sk); 1068 /* do new cwnd calculation (required) */ 1069 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1070 /* call before changing ca_state (optional) */ 1071 void (*set_state)(struct sock *sk, u8 new_state); 1072 /* call when cwnd event occurs (optional) */ 1073 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1074 /* call when ack arrives (optional) */ 1075 void (*in_ack_event)(struct sock *sk, u32 flags); 1076 /* new value of cwnd after loss (required) */ 1077 u32 (*undo_cwnd)(struct sock *sk); 1078 /* hook for packet ack accounting (optional) */ 1079 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1080 /* override sysctl_tcp_min_tso_segs */ 1081 u32 (*min_tso_segs)(struct sock *sk); 1082 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1083 u32 (*sndbuf_expand)(struct sock *sk); 1084 /* call when packets are delivered to update cwnd and pacing rate, 1085 * after all the ca_state processing. (optional) 1086 */ 1087 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1088 /* get info for inet_diag (optional) */ 1089 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1090 union tcp_cc_info *info); 1091 1092 char name[TCP_CA_NAME_MAX]; 1093 struct module *owner; 1094 }; 1095 1096 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1097 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1098 1099 void tcp_assign_congestion_control(struct sock *sk); 1100 void tcp_init_congestion_control(struct sock *sk); 1101 void tcp_cleanup_congestion_control(struct sock *sk); 1102 int tcp_set_default_congestion_control(struct net *net, const char *name); 1103 void tcp_get_default_congestion_control(struct net *net, char *name); 1104 void tcp_get_available_congestion_control(char *buf, size_t len); 1105 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1106 int tcp_set_allowed_congestion_control(char *allowed); 1107 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1108 bool reinit, bool cap_net_admin); 1109 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1110 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1111 1112 u32 tcp_reno_ssthresh(struct sock *sk); 1113 u32 tcp_reno_undo_cwnd(struct sock *sk); 1114 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1115 extern struct tcp_congestion_ops tcp_reno; 1116 1117 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1118 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1119 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1120 #ifdef CONFIG_INET 1121 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1122 #else 1123 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1124 { 1125 return NULL; 1126 } 1127 #endif 1128 1129 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1130 { 1131 const struct inet_connection_sock *icsk = inet_csk(sk); 1132 1133 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1134 } 1135 1136 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1137 { 1138 struct inet_connection_sock *icsk = inet_csk(sk); 1139 1140 if (icsk->icsk_ca_ops->set_state) 1141 icsk->icsk_ca_ops->set_state(sk, ca_state); 1142 icsk->icsk_ca_state = ca_state; 1143 } 1144 1145 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1146 { 1147 const struct inet_connection_sock *icsk = inet_csk(sk); 1148 1149 if (icsk->icsk_ca_ops->cwnd_event) 1150 icsk->icsk_ca_ops->cwnd_event(sk, event); 1151 } 1152 1153 /* From tcp_rate.c */ 1154 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1155 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1156 struct rate_sample *rs); 1157 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1158 bool is_sack_reneg, struct rate_sample *rs); 1159 void tcp_rate_check_app_limited(struct sock *sk); 1160 1161 /* These functions determine how the current flow behaves in respect of SACK 1162 * handling. SACK is negotiated with the peer, and therefore it can vary 1163 * between different flows. 1164 * 1165 * tcp_is_sack - SACK enabled 1166 * tcp_is_reno - No SACK 1167 */ 1168 static inline int tcp_is_sack(const struct tcp_sock *tp) 1169 { 1170 return likely(tp->rx_opt.sack_ok); 1171 } 1172 1173 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1174 { 1175 return !tcp_is_sack(tp); 1176 } 1177 1178 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1179 { 1180 return tp->sacked_out + tp->lost_out; 1181 } 1182 1183 /* This determines how many packets are "in the network" to the best 1184 * of our knowledge. In many cases it is conservative, but where 1185 * detailed information is available from the receiver (via SACK 1186 * blocks etc.) we can make more aggressive calculations. 1187 * 1188 * Use this for decisions involving congestion control, use just 1189 * tp->packets_out to determine if the send queue is empty or not. 1190 * 1191 * Read this equation as: 1192 * 1193 * "Packets sent once on transmission queue" MINUS 1194 * "Packets left network, but not honestly ACKed yet" PLUS 1195 * "Packets fast retransmitted" 1196 */ 1197 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1198 { 1199 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1200 } 1201 1202 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1203 1204 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1205 { 1206 return tp->snd_cwnd < tp->snd_ssthresh; 1207 } 1208 1209 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1210 { 1211 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1212 } 1213 1214 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1215 { 1216 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1217 (1 << inet_csk(sk)->icsk_ca_state); 1218 } 1219 1220 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1221 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1222 * ssthresh. 1223 */ 1224 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1225 { 1226 const struct tcp_sock *tp = tcp_sk(sk); 1227 1228 if (tcp_in_cwnd_reduction(sk)) 1229 return tp->snd_ssthresh; 1230 else 1231 return max(tp->snd_ssthresh, 1232 ((tp->snd_cwnd >> 1) + 1233 (tp->snd_cwnd >> 2))); 1234 } 1235 1236 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1237 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1238 1239 void tcp_enter_cwr(struct sock *sk); 1240 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1241 1242 /* The maximum number of MSS of available cwnd for which TSO defers 1243 * sending if not using sysctl_tcp_tso_win_divisor. 1244 */ 1245 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1246 { 1247 return 3; 1248 } 1249 1250 /* Returns end sequence number of the receiver's advertised window */ 1251 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1252 { 1253 return tp->snd_una + tp->snd_wnd; 1254 } 1255 1256 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1257 * flexible approach. The RFC suggests cwnd should not be raised unless 1258 * it was fully used previously. And that's exactly what we do in 1259 * congestion avoidance mode. But in slow start we allow cwnd to grow 1260 * as long as the application has used half the cwnd. 1261 * Example : 1262 * cwnd is 10 (IW10), but application sends 9 frames. 1263 * We allow cwnd to reach 18 when all frames are ACKed. 1264 * This check is safe because it's as aggressive as slow start which already 1265 * risks 100% overshoot. The advantage is that we discourage application to 1266 * either send more filler packets or data to artificially blow up the cwnd 1267 * usage, and allow application-limited process to probe bw more aggressively. 1268 */ 1269 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1270 { 1271 const struct tcp_sock *tp = tcp_sk(sk); 1272 1273 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1274 if (tcp_in_slow_start(tp)) 1275 return tp->snd_cwnd < 2 * tp->max_packets_out; 1276 1277 return tp->is_cwnd_limited; 1278 } 1279 1280 /* BBR congestion control needs pacing. 1281 * Same remark for SO_MAX_PACING_RATE. 1282 * sch_fq packet scheduler is efficiently handling pacing, 1283 * but is not always installed/used. 1284 * Return true if TCP stack should pace packets itself. 1285 */ 1286 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1287 { 1288 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1289 } 1290 1291 /* Estimates in how many jiffies next packet for this flow can be sent. 1292 * Scheduling a retransmit timer too early would be silly. 1293 */ 1294 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1295 { 1296 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1297 1298 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1299 } 1300 1301 static inline void tcp_reset_xmit_timer(struct sock *sk, 1302 const int what, 1303 unsigned long when, 1304 const unsigned long max_when) 1305 { 1306 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1307 max_when); 1308 } 1309 1310 /* Something is really bad, we could not queue an additional packet, 1311 * because qdisc is full or receiver sent a 0 window, or we are paced. 1312 * We do not want to add fuel to the fire, or abort too early, 1313 * so make sure the timer we arm now is at least 200ms in the future, 1314 * regardless of current icsk_rto value (as it could be ~2ms) 1315 */ 1316 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1317 { 1318 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1319 } 1320 1321 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1322 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1323 unsigned long max_when) 1324 { 1325 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1326 1327 return (unsigned long)min_t(u64, when, max_when); 1328 } 1329 1330 static inline void tcp_check_probe_timer(struct sock *sk) 1331 { 1332 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1333 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1334 tcp_probe0_base(sk), TCP_RTO_MAX); 1335 } 1336 1337 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1338 { 1339 tp->snd_wl1 = seq; 1340 } 1341 1342 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1343 { 1344 tp->snd_wl1 = seq; 1345 } 1346 1347 /* 1348 * Calculate(/check) TCP checksum 1349 */ 1350 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1351 __be32 daddr, __wsum base) 1352 { 1353 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1354 } 1355 1356 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1357 { 1358 return !skb_csum_unnecessary(skb) && 1359 __skb_checksum_complete(skb); 1360 } 1361 1362 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1363 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1364 void tcp_set_state(struct sock *sk, int state); 1365 void tcp_done(struct sock *sk); 1366 int tcp_abort(struct sock *sk, int err); 1367 1368 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1369 { 1370 rx_opt->dsack = 0; 1371 rx_opt->num_sacks = 0; 1372 } 1373 1374 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1375 1376 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1377 { 1378 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1379 struct tcp_sock *tp = tcp_sk(sk); 1380 s32 delta; 1381 1382 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1383 ca_ops->cong_control) 1384 return; 1385 delta = tcp_jiffies32 - tp->lsndtime; 1386 if (delta > inet_csk(sk)->icsk_rto) 1387 tcp_cwnd_restart(sk, delta); 1388 } 1389 1390 /* Determine a window scaling and initial window to offer. */ 1391 void tcp_select_initial_window(const struct sock *sk, int __space, 1392 __u32 mss, __u32 *rcv_wnd, 1393 __u32 *window_clamp, int wscale_ok, 1394 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1395 1396 static inline int tcp_win_from_space(const struct sock *sk, int space) 1397 { 1398 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1399 1400 return tcp_adv_win_scale <= 0 ? 1401 (space>>(-tcp_adv_win_scale)) : 1402 space - (space>>tcp_adv_win_scale); 1403 } 1404 1405 /* Note: caller must be prepared to deal with negative returns */ 1406 static inline int tcp_space(const struct sock *sk) 1407 { 1408 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1409 READ_ONCE(sk->sk_backlog.len) - 1410 atomic_read(&sk->sk_rmem_alloc)); 1411 } 1412 1413 static inline int tcp_full_space(const struct sock *sk) 1414 { 1415 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1416 } 1417 1418 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1419 * If 87.5 % (7/8) of the space has been consumed, we want to override 1420 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1421 * len/truesize ratio. 1422 */ 1423 static inline bool tcp_rmem_pressure(const struct sock *sk) 1424 { 1425 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1426 int threshold = rcvbuf - (rcvbuf >> 3); 1427 1428 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1429 } 1430 1431 extern void tcp_openreq_init_rwin(struct request_sock *req, 1432 const struct sock *sk_listener, 1433 const struct dst_entry *dst); 1434 1435 void tcp_enter_memory_pressure(struct sock *sk); 1436 void tcp_leave_memory_pressure(struct sock *sk); 1437 1438 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1439 { 1440 struct net *net = sock_net((struct sock *)tp); 1441 1442 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1443 } 1444 1445 static inline int keepalive_time_when(const struct tcp_sock *tp) 1446 { 1447 struct net *net = sock_net((struct sock *)tp); 1448 1449 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1450 } 1451 1452 static inline int keepalive_probes(const struct tcp_sock *tp) 1453 { 1454 struct net *net = sock_net((struct sock *)tp); 1455 1456 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1457 } 1458 1459 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1460 { 1461 const struct inet_connection_sock *icsk = &tp->inet_conn; 1462 1463 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1464 tcp_jiffies32 - tp->rcv_tstamp); 1465 } 1466 1467 static inline int tcp_fin_time(const struct sock *sk) 1468 { 1469 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1470 const int rto = inet_csk(sk)->icsk_rto; 1471 1472 if (fin_timeout < (rto << 2) - (rto >> 1)) 1473 fin_timeout = (rto << 2) - (rto >> 1); 1474 1475 return fin_timeout; 1476 } 1477 1478 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1479 int paws_win) 1480 { 1481 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1482 return true; 1483 if (unlikely(!time_before32(ktime_get_seconds(), 1484 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1485 return true; 1486 /* 1487 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1488 * then following tcp messages have valid values. Ignore 0 value, 1489 * or else 'negative' tsval might forbid us to accept their packets. 1490 */ 1491 if (!rx_opt->ts_recent) 1492 return true; 1493 return false; 1494 } 1495 1496 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1497 int rst) 1498 { 1499 if (tcp_paws_check(rx_opt, 0)) 1500 return false; 1501 1502 /* RST segments are not recommended to carry timestamp, 1503 and, if they do, it is recommended to ignore PAWS because 1504 "their cleanup function should take precedence over timestamps." 1505 Certainly, it is mistake. It is necessary to understand the reasons 1506 of this constraint to relax it: if peer reboots, clock may go 1507 out-of-sync and half-open connections will not be reset. 1508 Actually, the problem would be not existing if all 1509 the implementations followed draft about maintaining clock 1510 via reboots. Linux-2.2 DOES NOT! 1511 1512 However, we can relax time bounds for RST segments to MSL. 1513 */ 1514 if (rst && !time_before32(ktime_get_seconds(), 1515 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1516 return false; 1517 return true; 1518 } 1519 1520 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1521 int mib_idx, u32 *last_oow_ack_time); 1522 1523 static inline void tcp_mib_init(struct net *net) 1524 { 1525 /* See RFC 2012 */ 1526 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1527 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1528 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1529 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1530 } 1531 1532 /* from STCP */ 1533 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1534 { 1535 tp->lost_skb_hint = NULL; 1536 } 1537 1538 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1539 { 1540 tcp_clear_retrans_hints_partial(tp); 1541 tp->retransmit_skb_hint = NULL; 1542 } 1543 1544 union tcp_md5_addr { 1545 struct in_addr a4; 1546 #if IS_ENABLED(CONFIG_IPV6) 1547 struct in6_addr a6; 1548 #endif 1549 }; 1550 1551 /* - key database */ 1552 struct tcp_md5sig_key { 1553 struct hlist_node node; 1554 u8 keylen; 1555 u8 family; /* AF_INET or AF_INET6 */ 1556 u8 prefixlen; 1557 union tcp_md5_addr addr; 1558 int l3index; /* set if key added with L3 scope */ 1559 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1560 struct rcu_head rcu; 1561 }; 1562 1563 /* - sock block */ 1564 struct tcp_md5sig_info { 1565 struct hlist_head head; 1566 struct rcu_head rcu; 1567 }; 1568 1569 /* - pseudo header */ 1570 struct tcp4_pseudohdr { 1571 __be32 saddr; 1572 __be32 daddr; 1573 __u8 pad; 1574 __u8 protocol; 1575 __be16 len; 1576 }; 1577 1578 struct tcp6_pseudohdr { 1579 struct in6_addr saddr; 1580 struct in6_addr daddr; 1581 __be32 len; 1582 __be32 protocol; /* including padding */ 1583 }; 1584 1585 union tcp_md5sum_block { 1586 struct tcp4_pseudohdr ip4; 1587 #if IS_ENABLED(CONFIG_IPV6) 1588 struct tcp6_pseudohdr ip6; 1589 #endif 1590 }; 1591 1592 /* - pool: digest algorithm, hash description and scratch buffer */ 1593 struct tcp_md5sig_pool { 1594 struct ahash_request *md5_req; 1595 void *scratch; 1596 }; 1597 1598 /* - functions */ 1599 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1600 const struct sock *sk, const struct sk_buff *skb); 1601 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1602 int family, u8 prefixlen, int l3index, 1603 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1604 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1605 int family, u8 prefixlen, int l3index); 1606 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1607 const struct sock *addr_sk); 1608 1609 #ifdef CONFIG_TCP_MD5SIG 1610 #include <linux/jump_label.h> 1611 extern struct static_key_false tcp_md5_needed; 1612 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1613 const union tcp_md5_addr *addr, 1614 int family); 1615 static inline struct tcp_md5sig_key * 1616 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1617 const union tcp_md5_addr *addr, int family) 1618 { 1619 if (!static_branch_unlikely(&tcp_md5_needed)) 1620 return NULL; 1621 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1622 } 1623 1624 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1625 #else 1626 static inline struct tcp_md5sig_key * 1627 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1628 const union tcp_md5_addr *addr, int family) 1629 { 1630 return NULL; 1631 } 1632 #define tcp_twsk_md5_key(twsk) NULL 1633 #endif 1634 1635 bool tcp_alloc_md5sig_pool(void); 1636 1637 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1638 static inline void tcp_put_md5sig_pool(void) 1639 { 1640 local_bh_enable(); 1641 } 1642 1643 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1644 unsigned int header_len); 1645 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1646 const struct tcp_md5sig_key *key); 1647 1648 /* From tcp_fastopen.c */ 1649 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1650 struct tcp_fastopen_cookie *cookie); 1651 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1652 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1653 u16 try_exp); 1654 struct tcp_fastopen_request { 1655 /* Fast Open cookie. Size 0 means a cookie request */ 1656 struct tcp_fastopen_cookie cookie; 1657 struct msghdr *data; /* data in MSG_FASTOPEN */ 1658 size_t size; 1659 int copied; /* queued in tcp_connect() */ 1660 struct ubuf_info *uarg; 1661 }; 1662 void tcp_free_fastopen_req(struct tcp_sock *tp); 1663 void tcp_fastopen_destroy_cipher(struct sock *sk); 1664 void tcp_fastopen_ctx_destroy(struct net *net); 1665 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1666 void *primary_key, void *backup_key); 1667 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1668 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1669 struct request_sock *req, 1670 struct tcp_fastopen_cookie *foc, 1671 const struct dst_entry *dst); 1672 void tcp_fastopen_init_key_once(struct net *net); 1673 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1674 struct tcp_fastopen_cookie *cookie); 1675 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1676 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1677 #define TCP_FASTOPEN_KEY_MAX 2 1678 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1679 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1680 1681 /* Fastopen key context */ 1682 struct tcp_fastopen_context { 1683 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1684 int num; 1685 struct rcu_head rcu; 1686 }; 1687 1688 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1689 void tcp_fastopen_active_disable(struct sock *sk); 1690 bool tcp_fastopen_active_should_disable(struct sock *sk); 1691 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1692 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1693 1694 /* Caller needs to wrap with rcu_read_(un)lock() */ 1695 static inline 1696 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1697 { 1698 struct tcp_fastopen_context *ctx; 1699 1700 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1701 if (!ctx) 1702 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1703 return ctx; 1704 } 1705 1706 static inline 1707 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1708 const struct tcp_fastopen_cookie *orig) 1709 { 1710 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1711 orig->len == foc->len && 1712 !memcmp(orig->val, foc->val, foc->len)) 1713 return true; 1714 return false; 1715 } 1716 1717 static inline 1718 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1719 { 1720 return ctx->num; 1721 } 1722 1723 /* Latencies incurred by various limits for a sender. They are 1724 * chronograph-like stats that are mutually exclusive. 1725 */ 1726 enum tcp_chrono { 1727 TCP_CHRONO_UNSPEC, 1728 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1729 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1730 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1731 __TCP_CHRONO_MAX, 1732 }; 1733 1734 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1735 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1736 1737 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1738 * the same memory storage than skb->destructor/_skb_refdst 1739 */ 1740 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1741 { 1742 skb->destructor = NULL; 1743 skb->_skb_refdst = 0UL; 1744 } 1745 1746 #define tcp_skb_tsorted_save(skb) { \ 1747 unsigned long _save = skb->_skb_refdst; \ 1748 skb->_skb_refdst = 0UL; 1749 1750 #define tcp_skb_tsorted_restore(skb) \ 1751 skb->_skb_refdst = _save; \ 1752 } 1753 1754 void tcp_write_queue_purge(struct sock *sk); 1755 1756 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1757 { 1758 return skb_rb_first(&sk->tcp_rtx_queue); 1759 } 1760 1761 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1762 { 1763 return skb_rb_last(&sk->tcp_rtx_queue); 1764 } 1765 1766 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1767 { 1768 return skb_peek(&sk->sk_write_queue); 1769 } 1770 1771 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1772 { 1773 return skb_peek_tail(&sk->sk_write_queue); 1774 } 1775 1776 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1777 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1778 1779 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1780 { 1781 return skb_peek(&sk->sk_write_queue); 1782 } 1783 1784 static inline bool tcp_skb_is_last(const struct sock *sk, 1785 const struct sk_buff *skb) 1786 { 1787 return skb_queue_is_last(&sk->sk_write_queue, skb); 1788 } 1789 1790 /** 1791 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1792 * @sk: socket 1793 * 1794 * Since the write queue can have a temporary empty skb in it, 1795 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1796 */ 1797 static inline bool tcp_write_queue_empty(const struct sock *sk) 1798 { 1799 const struct tcp_sock *tp = tcp_sk(sk); 1800 1801 return tp->write_seq == tp->snd_nxt; 1802 } 1803 1804 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1805 { 1806 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1807 } 1808 1809 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1810 { 1811 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1812 } 1813 1814 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1815 { 1816 __skb_queue_tail(&sk->sk_write_queue, skb); 1817 1818 /* Queue it, remembering where we must start sending. */ 1819 if (sk->sk_write_queue.next == skb) 1820 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1821 } 1822 1823 /* Insert new before skb on the write queue of sk. */ 1824 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1825 struct sk_buff *skb, 1826 struct sock *sk) 1827 { 1828 __skb_queue_before(&sk->sk_write_queue, skb, new); 1829 } 1830 1831 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1832 { 1833 tcp_skb_tsorted_anchor_cleanup(skb); 1834 __skb_unlink(skb, &sk->sk_write_queue); 1835 } 1836 1837 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1838 1839 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1840 { 1841 tcp_skb_tsorted_anchor_cleanup(skb); 1842 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1843 } 1844 1845 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1846 { 1847 list_del(&skb->tcp_tsorted_anchor); 1848 tcp_rtx_queue_unlink(skb, sk); 1849 sk_wmem_free_skb(sk, skb); 1850 } 1851 1852 static inline void tcp_push_pending_frames(struct sock *sk) 1853 { 1854 if (tcp_send_head(sk)) { 1855 struct tcp_sock *tp = tcp_sk(sk); 1856 1857 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1858 } 1859 } 1860 1861 /* Start sequence of the skb just after the highest skb with SACKed 1862 * bit, valid only if sacked_out > 0 or when the caller has ensured 1863 * validity by itself. 1864 */ 1865 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1866 { 1867 if (!tp->sacked_out) 1868 return tp->snd_una; 1869 1870 if (tp->highest_sack == NULL) 1871 return tp->snd_nxt; 1872 1873 return TCP_SKB_CB(tp->highest_sack)->seq; 1874 } 1875 1876 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1877 { 1878 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1879 } 1880 1881 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1882 { 1883 return tcp_sk(sk)->highest_sack; 1884 } 1885 1886 static inline void tcp_highest_sack_reset(struct sock *sk) 1887 { 1888 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1889 } 1890 1891 /* Called when old skb is about to be deleted and replaced by new skb */ 1892 static inline void tcp_highest_sack_replace(struct sock *sk, 1893 struct sk_buff *old, 1894 struct sk_buff *new) 1895 { 1896 if (old == tcp_highest_sack(sk)) 1897 tcp_sk(sk)->highest_sack = new; 1898 } 1899 1900 /* This helper checks if socket has IP_TRANSPARENT set */ 1901 static inline bool inet_sk_transparent(const struct sock *sk) 1902 { 1903 switch (sk->sk_state) { 1904 case TCP_TIME_WAIT: 1905 return inet_twsk(sk)->tw_transparent; 1906 case TCP_NEW_SYN_RECV: 1907 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1908 } 1909 return inet_sk(sk)->transparent; 1910 } 1911 1912 /* Determines whether this is a thin stream (which may suffer from 1913 * increased latency). Used to trigger latency-reducing mechanisms. 1914 */ 1915 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1916 { 1917 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1918 } 1919 1920 /* /proc */ 1921 enum tcp_seq_states { 1922 TCP_SEQ_STATE_LISTENING, 1923 TCP_SEQ_STATE_ESTABLISHED, 1924 }; 1925 1926 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1927 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1928 void tcp_seq_stop(struct seq_file *seq, void *v); 1929 1930 struct tcp_seq_afinfo { 1931 sa_family_t family; 1932 }; 1933 1934 struct tcp_iter_state { 1935 struct seq_net_private p; 1936 enum tcp_seq_states state; 1937 struct sock *syn_wait_sk; 1938 int bucket, offset, sbucket, num; 1939 loff_t last_pos; 1940 }; 1941 1942 extern struct request_sock_ops tcp_request_sock_ops; 1943 extern struct request_sock_ops tcp6_request_sock_ops; 1944 1945 void tcp_v4_destroy_sock(struct sock *sk); 1946 1947 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1948 netdev_features_t features); 1949 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1950 int tcp_gro_complete(struct sk_buff *skb); 1951 1952 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1953 1954 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1955 { 1956 struct net *net = sock_net((struct sock *)tp); 1957 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1958 } 1959 1960 /* @wake is one when sk_stream_write_space() calls us. 1961 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1962 * This mimics the strategy used in sock_def_write_space(). 1963 */ 1964 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1965 { 1966 const struct tcp_sock *tp = tcp_sk(sk); 1967 u32 notsent_bytes = READ_ONCE(tp->write_seq) - 1968 READ_ONCE(tp->snd_nxt); 1969 1970 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1971 } 1972 1973 #ifdef CONFIG_PROC_FS 1974 int tcp4_proc_init(void); 1975 void tcp4_proc_exit(void); 1976 #endif 1977 1978 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1979 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1980 const struct tcp_request_sock_ops *af_ops, 1981 struct sock *sk, struct sk_buff *skb); 1982 1983 /* TCP af-specific functions */ 1984 struct tcp_sock_af_ops { 1985 #ifdef CONFIG_TCP_MD5SIG 1986 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1987 const struct sock *addr_sk); 1988 int (*calc_md5_hash)(char *location, 1989 const struct tcp_md5sig_key *md5, 1990 const struct sock *sk, 1991 const struct sk_buff *skb); 1992 int (*md5_parse)(struct sock *sk, 1993 int optname, 1994 char __user *optval, 1995 int optlen); 1996 #endif 1997 }; 1998 1999 struct tcp_request_sock_ops { 2000 u16 mss_clamp; 2001 #ifdef CONFIG_TCP_MD5SIG 2002 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2003 const struct sock *addr_sk); 2004 int (*calc_md5_hash) (char *location, 2005 const struct tcp_md5sig_key *md5, 2006 const struct sock *sk, 2007 const struct sk_buff *skb); 2008 #endif 2009 void (*init_req)(struct request_sock *req, 2010 const struct sock *sk_listener, 2011 struct sk_buff *skb); 2012 #ifdef CONFIG_SYN_COOKIES 2013 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2014 __u16 *mss); 2015 #endif 2016 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 2017 const struct request_sock *req); 2018 u32 (*init_seq)(const struct sk_buff *skb); 2019 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2020 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2021 struct flowi *fl, struct request_sock *req, 2022 struct tcp_fastopen_cookie *foc, 2023 enum tcp_synack_type synack_type); 2024 }; 2025 2026 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2027 #if IS_ENABLED(CONFIG_IPV6) 2028 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2029 #endif 2030 2031 #ifdef CONFIG_SYN_COOKIES 2032 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2033 const struct sock *sk, struct sk_buff *skb, 2034 __u16 *mss) 2035 { 2036 tcp_synq_overflow(sk); 2037 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2038 return ops->cookie_init_seq(skb, mss); 2039 } 2040 #else 2041 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2042 const struct sock *sk, struct sk_buff *skb, 2043 __u16 *mss) 2044 { 2045 return 0; 2046 } 2047 #endif 2048 2049 int tcpv4_offload_init(void); 2050 2051 void tcp_v4_init(void); 2052 void tcp_init(void); 2053 2054 /* tcp_recovery.c */ 2055 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2056 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2057 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2058 u32 reo_wnd); 2059 extern void tcp_rack_mark_lost(struct sock *sk); 2060 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2061 u64 xmit_time); 2062 extern void tcp_rack_reo_timeout(struct sock *sk); 2063 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2064 2065 /* At how many usecs into the future should the RTO fire? */ 2066 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2067 { 2068 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2069 u32 rto = inet_csk(sk)->icsk_rto; 2070 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2071 2072 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2073 } 2074 2075 /* 2076 * Save and compile IPv4 options, return a pointer to it 2077 */ 2078 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2079 struct sk_buff *skb) 2080 { 2081 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2082 struct ip_options_rcu *dopt = NULL; 2083 2084 if (opt->optlen) { 2085 int opt_size = sizeof(*dopt) + opt->optlen; 2086 2087 dopt = kmalloc(opt_size, GFP_ATOMIC); 2088 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2089 kfree(dopt); 2090 dopt = NULL; 2091 } 2092 } 2093 return dopt; 2094 } 2095 2096 /* locally generated TCP pure ACKs have skb->truesize == 2 2097 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2098 * This is much faster than dissecting the packet to find out. 2099 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2100 */ 2101 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2102 { 2103 return skb->truesize == 2; 2104 } 2105 2106 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2107 { 2108 skb->truesize = 2; 2109 } 2110 2111 static inline int tcp_inq(struct sock *sk) 2112 { 2113 struct tcp_sock *tp = tcp_sk(sk); 2114 int answ; 2115 2116 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2117 answ = 0; 2118 } else if (sock_flag(sk, SOCK_URGINLINE) || 2119 !tp->urg_data || 2120 before(tp->urg_seq, tp->copied_seq) || 2121 !before(tp->urg_seq, tp->rcv_nxt)) { 2122 2123 answ = tp->rcv_nxt - tp->copied_seq; 2124 2125 /* Subtract 1, if FIN was received */ 2126 if (answ && sock_flag(sk, SOCK_DONE)) 2127 answ--; 2128 } else { 2129 answ = tp->urg_seq - tp->copied_seq; 2130 } 2131 2132 return answ; 2133 } 2134 2135 int tcp_peek_len(struct socket *sock); 2136 2137 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2138 { 2139 u16 segs_in; 2140 2141 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2142 tp->segs_in += segs_in; 2143 if (skb->len > tcp_hdrlen(skb)) 2144 tp->data_segs_in += segs_in; 2145 } 2146 2147 /* 2148 * TCP listen path runs lockless. 2149 * We forced "struct sock" to be const qualified to make sure 2150 * we don't modify one of its field by mistake. 2151 * Here, we increment sk_drops which is an atomic_t, so we can safely 2152 * make sock writable again. 2153 */ 2154 static inline void tcp_listendrop(const struct sock *sk) 2155 { 2156 atomic_inc(&((struct sock *)sk)->sk_drops); 2157 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2158 } 2159 2160 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2161 2162 /* 2163 * Interface for adding Upper Level Protocols over TCP 2164 */ 2165 2166 #define TCP_ULP_NAME_MAX 16 2167 #define TCP_ULP_MAX 128 2168 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2169 2170 struct tcp_ulp_ops { 2171 struct list_head list; 2172 2173 /* initialize ulp */ 2174 int (*init)(struct sock *sk); 2175 /* update ulp */ 2176 void (*update)(struct sock *sk, struct proto *p, 2177 void (*write_space)(struct sock *sk)); 2178 /* cleanup ulp */ 2179 void (*release)(struct sock *sk); 2180 /* diagnostic */ 2181 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2182 size_t (*get_info_size)(const struct sock *sk); 2183 /* clone ulp */ 2184 void (*clone)(const struct request_sock *req, struct sock *newsk, 2185 const gfp_t priority); 2186 2187 char name[TCP_ULP_NAME_MAX]; 2188 struct module *owner; 2189 }; 2190 int tcp_register_ulp(struct tcp_ulp_ops *type); 2191 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2192 int tcp_set_ulp(struct sock *sk, const char *name); 2193 void tcp_get_available_ulp(char *buf, size_t len); 2194 void tcp_cleanup_ulp(struct sock *sk); 2195 void tcp_update_ulp(struct sock *sk, struct proto *p, 2196 void (*write_space)(struct sock *sk)); 2197 2198 #define MODULE_ALIAS_TCP_ULP(name) \ 2199 __MODULE_INFO(alias, alias_userspace, name); \ 2200 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2201 2202 struct sk_msg; 2203 struct sk_psock; 2204 2205 #ifdef CONFIG_BPF_STREAM_PARSER 2206 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2207 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2208 #else 2209 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2210 { 2211 } 2212 #endif /* CONFIG_BPF_STREAM_PARSER */ 2213 2214 #ifdef CONFIG_NET_SOCK_MSG 2215 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2216 int flags); 2217 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2218 struct msghdr *msg, int len, int flags); 2219 #endif /* CONFIG_NET_SOCK_MSG */ 2220 2221 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2222 * is < 0, then the BPF op failed (for example if the loaded BPF 2223 * program does not support the chosen operation or there is no BPF 2224 * program loaded). 2225 */ 2226 #ifdef CONFIG_BPF 2227 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2228 { 2229 struct bpf_sock_ops_kern sock_ops; 2230 int ret; 2231 2232 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2233 if (sk_fullsock(sk)) { 2234 sock_ops.is_fullsock = 1; 2235 sock_owned_by_me(sk); 2236 } 2237 2238 sock_ops.sk = sk; 2239 sock_ops.op = op; 2240 if (nargs > 0) 2241 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2242 2243 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2244 if (ret == 0) 2245 ret = sock_ops.reply; 2246 else 2247 ret = -1; 2248 return ret; 2249 } 2250 2251 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2252 { 2253 u32 args[2] = {arg1, arg2}; 2254 2255 return tcp_call_bpf(sk, op, 2, args); 2256 } 2257 2258 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2259 u32 arg3) 2260 { 2261 u32 args[3] = {arg1, arg2, arg3}; 2262 2263 return tcp_call_bpf(sk, op, 3, args); 2264 } 2265 2266 #else 2267 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2268 { 2269 return -EPERM; 2270 } 2271 2272 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2273 { 2274 return -EPERM; 2275 } 2276 2277 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2278 u32 arg3) 2279 { 2280 return -EPERM; 2281 } 2282 2283 #endif 2284 2285 static inline u32 tcp_timeout_init(struct sock *sk) 2286 { 2287 int timeout; 2288 2289 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2290 2291 if (timeout <= 0) 2292 timeout = TCP_TIMEOUT_INIT; 2293 return timeout; 2294 } 2295 2296 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2297 { 2298 int rwnd; 2299 2300 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2301 2302 if (rwnd < 0) 2303 rwnd = 0; 2304 return rwnd; 2305 } 2306 2307 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2308 { 2309 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2310 } 2311 2312 static inline void tcp_bpf_rtt(struct sock *sk) 2313 { 2314 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2315 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2316 } 2317 2318 #if IS_ENABLED(CONFIG_SMC) 2319 extern struct static_key_false tcp_have_smc; 2320 #endif 2321 2322 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2323 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2324 void (*cad)(struct sock *sk, u32 ack_seq)); 2325 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2326 void clean_acked_data_flush(void); 2327 #endif 2328 2329 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2330 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2331 const struct tcp_sock *tp) 2332 { 2333 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2334 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2335 } 2336 2337 /* Compute Earliest Departure Time for some control packets 2338 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2339 */ 2340 static inline u64 tcp_transmit_time(const struct sock *sk) 2341 { 2342 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2343 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2344 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2345 2346 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2347 } 2348 return 0; 2349 } 2350 2351 #endif /* _TCP_H */ 2352