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 */ 128 129 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 130 #if HZ >= 100 131 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 132 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 133 #else 134 #define TCP_DELACK_MIN 4U 135 #define TCP_ATO_MIN 4U 136 #endif 137 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 138 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 139 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 140 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 141 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 142 * used as a fallback RTO for the 143 * initial data transmission if no 144 * valid RTT sample has been acquired, 145 * most likely due to retrans in 3WHS. 146 */ 147 148 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 149 * for local resources. 150 */ 151 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 152 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 153 #define TCP_KEEPALIVE_INTVL (75*HZ) 154 155 #define MAX_TCP_KEEPIDLE 32767 156 #define MAX_TCP_KEEPINTVL 32767 157 #define MAX_TCP_KEEPCNT 127 158 #define MAX_TCP_SYNCNT 127 159 160 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 161 162 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 163 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 164 * after this time. It should be equal 165 * (or greater than) TCP_TIMEWAIT_LEN 166 * to provide reliability equal to one 167 * provided by timewait state. 168 */ 169 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 170 * timestamps. It must be less than 171 * minimal timewait lifetime. 172 */ 173 /* 174 * TCP option 175 */ 176 177 #define TCPOPT_NOP 1 /* Padding */ 178 #define TCPOPT_EOL 0 /* End of options */ 179 #define TCPOPT_MSS 2 /* Segment size negotiating */ 180 #define TCPOPT_WINDOW 3 /* Window scaling */ 181 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 182 #define TCPOPT_SACK 5 /* SACK Block */ 183 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 184 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 185 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 186 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 187 #define TCPOPT_EXP 254 /* Experimental */ 188 /* Magic number to be after the option value for sharing TCP 189 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 190 */ 191 #define TCPOPT_FASTOPEN_MAGIC 0xF989 192 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 193 194 /* 195 * TCP option lengths 196 */ 197 198 #define TCPOLEN_MSS 4 199 #define TCPOLEN_WINDOW 3 200 #define TCPOLEN_SACK_PERM 2 201 #define TCPOLEN_TIMESTAMP 10 202 #define TCPOLEN_MD5SIG 18 203 #define TCPOLEN_FASTOPEN_BASE 2 204 #define TCPOLEN_EXP_FASTOPEN_BASE 4 205 #define TCPOLEN_EXP_SMC_BASE 6 206 207 /* But this is what stacks really send out. */ 208 #define TCPOLEN_TSTAMP_ALIGNED 12 209 #define TCPOLEN_WSCALE_ALIGNED 4 210 #define TCPOLEN_SACKPERM_ALIGNED 4 211 #define TCPOLEN_SACK_BASE 2 212 #define TCPOLEN_SACK_BASE_ALIGNED 4 213 #define TCPOLEN_SACK_PERBLOCK 8 214 #define TCPOLEN_MD5SIG_ALIGNED 20 215 #define TCPOLEN_MSS_ALIGNED 4 216 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 217 218 /* Flags in tp->nonagle */ 219 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 220 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 221 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 222 223 /* TCP thin-stream limits */ 224 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 225 226 /* TCP initial congestion window as per rfc6928 */ 227 #define TCP_INIT_CWND 10 228 229 /* Bit Flags for sysctl_tcp_fastopen */ 230 #define TFO_CLIENT_ENABLE 1 231 #define TFO_SERVER_ENABLE 2 232 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 233 234 /* Accept SYN data w/o any cookie option */ 235 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 236 237 /* Force enable TFO on all listeners, i.e., not requiring the 238 * TCP_FASTOPEN socket option. 239 */ 240 #define TFO_SERVER_WO_SOCKOPT1 0x400 241 242 243 /* sysctl variables for tcp */ 244 extern int sysctl_tcp_max_orphans; 245 extern long sysctl_tcp_mem[3]; 246 247 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 248 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 249 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 250 251 extern atomic_long_t tcp_memory_allocated; 252 extern struct percpu_counter tcp_sockets_allocated; 253 extern unsigned long tcp_memory_pressure; 254 255 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 256 static inline bool tcp_under_memory_pressure(const struct sock *sk) 257 { 258 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 259 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 260 return true; 261 262 return READ_ONCE(tcp_memory_pressure); 263 } 264 /* 265 * The next routines deal with comparing 32 bit unsigned ints 266 * and worry about wraparound (automatic with unsigned arithmetic). 267 */ 268 269 static inline bool before(__u32 seq1, __u32 seq2) 270 { 271 return (__s32)(seq1-seq2) < 0; 272 } 273 #define after(seq2, seq1) before(seq1, seq2) 274 275 /* is s2<=s1<=s3 ? */ 276 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 277 { 278 return seq3 - seq2 >= seq1 - seq2; 279 } 280 281 static inline bool tcp_out_of_memory(struct sock *sk) 282 { 283 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 284 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 285 return true; 286 return false; 287 } 288 289 void sk_forced_mem_schedule(struct sock *sk, int size); 290 291 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 292 { 293 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 294 int orphans = percpu_counter_read_positive(ocp); 295 296 if (orphans << shift > sysctl_tcp_max_orphans) { 297 orphans = percpu_counter_sum_positive(ocp); 298 if (orphans << shift > sysctl_tcp_max_orphans) 299 return true; 300 } 301 return false; 302 } 303 304 bool tcp_check_oom(struct sock *sk, int shift); 305 306 307 extern struct proto tcp_prot; 308 309 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 310 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 311 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 312 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 313 314 void tcp_tasklet_init(void); 315 316 int tcp_v4_err(struct sk_buff *skb, u32); 317 318 void tcp_shutdown(struct sock *sk, int how); 319 320 int tcp_v4_early_demux(struct sk_buff *skb); 321 int tcp_v4_rcv(struct sk_buff *skb); 322 323 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 324 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 325 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 326 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 327 int flags); 328 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 329 size_t size, int flags); 330 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 331 size_t size, int flags); 332 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 333 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 334 int size_goal); 335 void tcp_release_cb(struct sock *sk); 336 void tcp_wfree(struct sk_buff *skb); 337 void tcp_write_timer_handler(struct sock *sk); 338 void tcp_delack_timer_handler(struct sock *sk); 339 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 340 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 341 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 342 void tcp_rcv_space_adjust(struct sock *sk); 343 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 344 void tcp_twsk_destructor(struct sock *sk); 345 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 346 struct pipe_inode_info *pipe, size_t len, 347 unsigned int flags); 348 349 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 350 static inline void tcp_dec_quickack_mode(struct sock *sk, 351 const unsigned int pkts) 352 { 353 struct inet_connection_sock *icsk = inet_csk(sk); 354 355 if (icsk->icsk_ack.quick) { 356 if (pkts >= icsk->icsk_ack.quick) { 357 icsk->icsk_ack.quick = 0; 358 /* Leaving quickack mode we deflate ATO. */ 359 icsk->icsk_ack.ato = TCP_ATO_MIN; 360 } else 361 icsk->icsk_ack.quick -= pkts; 362 } 363 } 364 365 #define TCP_ECN_OK 1 366 #define TCP_ECN_QUEUE_CWR 2 367 #define TCP_ECN_DEMAND_CWR 4 368 #define TCP_ECN_SEEN 8 369 370 enum tcp_tw_status { 371 TCP_TW_SUCCESS = 0, 372 TCP_TW_RST = 1, 373 TCP_TW_ACK = 2, 374 TCP_TW_SYN = 3 375 }; 376 377 378 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 379 struct sk_buff *skb, 380 const struct tcphdr *th); 381 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 382 struct request_sock *req, bool fastopen, 383 bool *lost_race); 384 int tcp_child_process(struct sock *parent, struct sock *child, 385 struct sk_buff *skb); 386 void tcp_enter_loss(struct sock *sk); 387 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 388 void tcp_clear_retrans(struct tcp_sock *tp); 389 void tcp_update_metrics(struct sock *sk); 390 void tcp_init_metrics(struct sock *sk); 391 void tcp_metrics_init(void); 392 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 393 void tcp_close(struct sock *sk, long timeout); 394 void tcp_init_sock(struct sock *sk); 395 void tcp_init_transfer(struct sock *sk, int bpf_op); 396 __poll_t tcp_poll(struct file *file, struct socket *sock, 397 struct poll_table_struct *wait); 398 int tcp_getsockopt(struct sock *sk, int level, int optname, 399 char __user *optval, int __user *optlen); 400 int tcp_setsockopt(struct sock *sk, int level, int optname, 401 char __user *optval, unsigned int optlen); 402 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 403 char __user *optval, int __user *optlen); 404 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 405 char __user *optval, unsigned int optlen); 406 void tcp_set_keepalive(struct sock *sk, int val); 407 void tcp_syn_ack_timeout(const struct request_sock *req); 408 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 409 int flags, int *addr_len); 410 int tcp_set_rcvlowat(struct sock *sk, int val); 411 void tcp_data_ready(struct sock *sk); 412 #ifdef CONFIG_MMU 413 int tcp_mmap(struct file *file, struct socket *sock, 414 struct vm_area_struct *vma); 415 #endif 416 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 417 struct tcp_options_received *opt_rx, 418 int estab, struct tcp_fastopen_cookie *foc); 419 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 420 421 /* 422 * BPF SKB-less helpers 423 */ 424 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 425 struct tcphdr *th, u32 *cookie); 426 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 427 struct tcphdr *th, u32 *cookie); 428 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 429 const struct tcp_request_sock_ops *af_ops, 430 struct sock *sk, struct tcphdr *th); 431 /* 432 * TCP v4 functions exported for the inet6 API 433 */ 434 435 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 436 void tcp_v4_mtu_reduced(struct sock *sk); 437 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 438 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 439 struct sock *tcp_create_openreq_child(const struct sock *sk, 440 struct request_sock *req, 441 struct sk_buff *skb); 442 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 443 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 444 struct request_sock *req, 445 struct dst_entry *dst, 446 struct request_sock *req_unhash, 447 bool *own_req); 448 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 449 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 450 int tcp_connect(struct sock *sk); 451 enum tcp_synack_type { 452 TCP_SYNACK_NORMAL, 453 TCP_SYNACK_FASTOPEN, 454 TCP_SYNACK_COOKIE, 455 }; 456 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 457 struct request_sock *req, 458 struct tcp_fastopen_cookie *foc, 459 enum tcp_synack_type synack_type); 460 int tcp_disconnect(struct sock *sk, int flags); 461 462 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 463 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 464 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 465 466 /* From syncookies.c */ 467 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 468 struct request_sock *req, 469 struct dst_entry *dst, u32 tsoff); 470 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 471 u32 cookie); 472 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 473 #ifdef CONFIG_SYN_COOKIES 474 475 /* Syncookies use a monotonic timer which increments every 60 seconds. 476 * This counter is used both as a hash input and partially encoded into 477 * the cookie value. A cookie is only validated further if the delta 478 * between the current counter value and the encoded one is less than this, 479 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 480 * the counter advances immediately after a cookie is generated). 481 */ 482 #define MAX_SYNCOOKIE_AGE 2 483 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 484 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 485 486 /* syncookies: remember time of last synqueue overflow 487 * But do not dirty this field too often (once per second is enough) 488 * It is racy as we do not hold a lock, but race is very minor. 489 */ 490 static inline void tcp_synq_overflow(const struct sock *sk) 491 { 492 unsigned int last_overflow; 493 unsigned int now = jiffies; 494 495 if (sk->sk_reuseport) { 496 struct sock_reuseport *reuse; 497 498 reuse = rcu_dereference(sk->sk_reuseport_cb); 499 if (likely(reuse)) { 500 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 501 if (!time_between32(now, last_overflow, 502 last_overflow + HZ)) 503 WRITE_ONCE(reuse->synq_overflow_ts, now); 504 return; 505 } 506 } 507 508 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 509 if (!time_between32(now, last_overflow, last_overflow + HZ)) 510 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 511 } 512 513 /* syncookies: no recent synqueue overflow on this listening socket? */ 514 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 515 { 516 unsigned int last_overflow; 517 unsigned int now = jiffies; 518 519 if (sk->sk_reuseport) { 520 struct sock_reuseport *reuse; 521 522 reuse = rcu_dereference(sk->sk_reuseport_cb); 523 if (likely(reuse)) { 524 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 525 return !time_between32(now, last_overflow - HZ, 526 last_overflow + 527 TCP_SYNCOOKIE_VALID); 528 } 529 } 530 531 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 532 533 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 534 * then we're under synflood. However, we have to use 535 * 'last_overflow - HZ' as lower bound. That's because a concurrent 536 * tcp_synq_overflow() could update .ts_recent_stamp after we read 537 * jiffies but before we store .ts_recent_stamp into last_overflow, 538 * which could lead to rejecting a valid syncookie. 539 */ 540 return !time_between32(now, last_overflow - HZ, 541 last_overflow + TCP_SYNCOOKIE_VALID); 542 } 543 544 static inline u32 tcp_cookie_time(void) 545 { 546 u64 val = get_jiffies_64(); 547 548 do_div(val, TCP_SYNCOOKIE_PERIOD); 549 return val; 550 } 551 552 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 553 u16 *mssp); 554 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 555 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 556 bool cookie_timestamp_decode(const struct net *net, 557 struct tcp_options_received *opt); 558 bool cookie_ecn_ok(const struct tcp_options_received *opt, 559 const struct net *net, const struct dst_entry *dst); 560 561 /* From net/ipv6/syncookies.c */ 562 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 563 u32 cookie); 564 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 565 566 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 567 const struct tcphdr *th, u16 *mssp); 568 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 569 #endif 570 /* tcp_output.c */ 571 572 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 573 int nonagle); 574 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 575 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 576 void tcp_retransmit_timer(struct sock *sk); 577 void tcp_xmit_retransmit_queue(struct sock *); 578 void tcp_simple_retransmit(struct sock *); 579 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 580 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 581 enum tcp_queue { 582 TCP_FRAG_IN_WRITE_QUEUE, 583 TCP_FRAG_IN_RTX_QUEUE, 584 }; 585 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 586 struct sk_buff *skb, u32 len, 587 unsigned int mss_now, gfp_t gfp); 588 589 void tcp_send_probe0(struct sock *); 590 void tcp_send_partial(struct sock *); 591 int tcp_write_wakeup(struct sock *, int mib); 592 void tcp_send_fin(struct sock *sk); 593 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 594 int tcp_send_synack(struct sock *); 595 void tcp_push_one(struct sock *, unsigned int mss_now); 596 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 597 void tcp_send_ack(struct sock *sk); 598 void tcp_send_delayed_ack(struct sock *sk); 599 void tcp_send_loss_probe(struct sock *sk); 600 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 601 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 602 const struct sk_buff *next_skb); 603 604 /* tcp_input.c */ 605 void tcp_rearm_rto(struct sock *sk); 606 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 607 void tcp_reset(struct sock *sk); 608 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 609 void tcp_fin(struct sock *sk); 610 611 /* tcp_timer.c */ 612 void tcp_init_xmit_timers(struct sock *); 613 static inline void tcp_clear_xmit_timers(struct sock *sk) 614 { 615 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 616 __sock_put(sk); 617 618 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 619 __sock_put(sk); 620 621 inet_csk_clear_xmit_timers(sk); 622 } 623 624 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 625 unsigned int tcp_current_mss(struct sock *sk); 626 627 /* Bound MSS / TSO packet size with the half of the window */ 628 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 629 { 630 int cutoff; 631 632 /* When peer uses tiny windows, there is no use in packetizing 633 * to sub-MSS pieces for the sake of SWS or making sure there 634 * are enough packets in the pipe for fast recovery. 635 * 636 * On the other hand, for extremely large MSS devices, handling 637 * smaller than MSS windows in this way does make sense. 638 */ 639 if (tp->max_window > TCP_MSS_DEFAULT) 640 cutoff = (tp->max_window >> 1); 641 else 642 cutoff = tp->max_window; 643 644 if (cutoff && pktsize > cutoff) 645 return max_t(int, cutoff, 68U - tp->tcp_header_len); 646 else 647 return pktsize; 648 } 649 650 /* tcp.c */ 651 void tcp_get_info(struct sock *, struct tcp_info *); 652 653 /* Read 'sendfile()'-style from a TCP socket */ 654 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 655 sk_read_actor_t recv_actor); 656 657 void tcp_initialize_rcv_mss(struct sock *sk); 658 659 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 660 int tcp_mss_to_mtu(struct sock *sk, int mss); 661 void tcp_mtup_init(struct sock *sk); 662 void tcp_init_buffer_space(struct sock *sk); 663 664 static inline void tcp_bound_rto(const struct sock *sk) 665 { 666 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 667 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 668 } 669 670 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 671 { 672 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 673 } 674 675 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 676 { 677 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 678 ntohl(TCP_FLAG_ACK) | 679 snd_wnd); 680 } 681 682 static inline void tcp_fast_path_on(struct tcp_sock *tp) 683 { 684 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 685 } 686 687 static inline void tcp_fast_path_check(struct sock *sk) 688 { 689 struct tcp_sock *tp = tcp_sk(sk); 690 691 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 692 tp->rcv_wnd && 693 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 694 !tp->urg_data) 695 tcp_fast_path_on(tp); 696 } 697 698 /* Compute the actual rto_min value */ 699 static inline u32 tcp_rto_min(struct sock *sk) 700 { 701 const struct dst_entry *dst = __sk_dst_get(sk); 702 u32 rto_min = TCP_RTO_MIN; 703 704 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 705 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 706 return rto_min; 707 } 708 709 static inline u32 tcp_rto_min_us(struct sock *sk) 710 { 711 return jiffies_to_usecs(tcp_rto_min(sk)); 712 } 713 714 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 715 { 716 return dst_metric_locked(dst, RTAX_CC_ALGO); 717 } 718 719 /* Minimum RTT in usec. ~0 means not available. */ 720 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 721 { 722 return minmax_get(&tp->rtt_min); 723 } 724 725 /* Compute the actual receive window we are currently advertising. 726 * Rcv_nxt can be after the window if our peer push more data 727 * than the offered window. 728 */ 729 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 730 { 731 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 732 733 if (win < 0) 734 win = 0; 735 return (u32) win; 736 } 737 738 /* Choose a new window, without checks for shrinking, and without 739 * scaling applied to the result. The caller does these things 740 * if necessary. This is a "raw" window selection. 741 */ 742 u32 __tcp_select_window(struct sock *sk); 743 744 void tcp_send_window_probe(struct sock *sk); 745 746 /* TCP uses 32bit jiffies to save some space. 747 * Note that this is different from tcp_time_stamp, which 748 * historically has been the same until linux-4.13. 749 */ 750 #define tcp_jiffies32 ((u32)jiffies) 751 752 /* 753 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 754 * It is no longer tied to jiffies, but to 1 ms clock. 755 * Note: double check if you want to use tcp_jiffies32 instead of this. 756 */ 757 #define TCP_TS_HZ 1000 758 759 static inline u64 tcp_clock_ns(void) 760 { 761 return ktime_get_ns(); 762 } 763 764 static inline u64 tcp_clock_us(void) 765 { 766 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 767 } 768 769 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 770 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 771 { 772 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 773 } 774 775 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 776 static inline u32 tcp_ns_to_ts(u64 ns) 777 { 778 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 779 } 780 781 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 782 static inline u32 tcp_time_stamp_raw(void) 783 { 784 return tcp_ns_to_ts(tcp_clock_ns()); 785 } 786 787 void tcp_mstamp_refresh(struct tcp_sock *tp); 788 789 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 790 { 791 return max_t(s64, t1 - t0, 0); 792 } 793 794 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 795 { 796 return tcp_ns_to_ts(skb->skb_mstamp_ns); 797 } 798 799 /* provide the departure time in us unit */ 800 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 801 { 802 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 803 } 804 805 806 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 807 808 #define TCPHDR_FIN 0x01 809 #define TCPHDR_SYN 0x02 810 #define TCPHDR_RST 0x04 811 #define TCPHDR_PSH 0x08 812 #define TCPHDR_ACK 0x10 813 #define TCPHDR_URG 0x20 814 #define TCPHDR_ECE 0x40 815 #define TCPHDR_CWR 0x80 816 817 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 818 819 /* This is what the send packet queuing engine uses to pass 820 * TCP per-packet control information to the transmission code. 821 * We also store the host-order sequence numbers in here too. 822 * This is 44 bytes if IPV6 is enabled. 823 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 824 */ 825 struct tcp_skb_cb { 826 __u32 seq; /* Starting sequence number */ 827 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 828 union { 829 /* Note : tcp_tw_isn is used in input path only 830 * (isn chosen by tcp_timewait_state_process()) 831 * 832 * tcp_gso_segs/size are used in write queue only, 833 * cf tcp_skb_pcount()/tcp_skb_mss() 834 */ 835 __u32 tcp_tw_isn; 836 struct { 837 u16 tcp_gso_segs; 838 u16 tcp_gso_size; 839 }; 840 }; 841 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 842 843 __u8 sacked; /* State flags for SACK. */ 844 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 845 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 846 #define TCPCB_LOST 0x04 /* SKB is lost */ 847 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 848 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 849 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 850 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 851 TCPCB_REPAIRED) 852 853 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 854 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 855 eor:1, /* Is skb MSG_EOR marked? */ 856 has_rxtstamp:1, /* SKB has a RX timestamp */ 857 unused:5; 858 __u32 ack_seq; /* Sequence number ACK'd */ 859 union { 860 struct { 861 /* There is space for up to 24 bytes */ 862 __u32 in_flight:30,/* Bytes in flight at transmit */ 863 is_app_limited:1, /* cwnd not fully used? */ 864 unused:1; 865 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 866 __u32 delivered; 867 /* start of send pipeline phase */ 868 u64 first_tx_mstamp; 869 /* when we reached the "delivered" count */ 870 u64 delivered_mstamp; 871 } tx; /* only used for outgoing skbs */ 872 union { 873 struct inet_skb_parm h4; 874 #if IS_ENABLED(CONFIG_IPV6) 875 struct inet6_skb_parm h6; 876 #endif 877 } header; /* For incoming skbs */ 878 struct { 879 __u32 flags; 880 struct sock *sk_redir; 881 void *data_end; 882 } bpf; 883 }; 884 }; 885 886 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 887 888 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 889 { 890 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 891 } 892 893 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 894 { 895 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 896 } 897 898 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 899 { 900 return TCP_SKB_CB(skb)->bpf.sk_redir; 901 } 902 903 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 904 { 905 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 906 } 907 908 #if IS_ENABLED(CONFIG_IPV6) 909 /* This is the variant of inet6_iif() that must be used by TCP, 910 * as TCP moves IP6CB into a different location in skb->cb[] 911 */ 912 static inline int tcp_v6_iif(const struct sk_buff *skb) 913 { 914 return TCP_SKB_CB(skb)->header.h6.iif; 915 } 916 917 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 918 { 919 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 920 921 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 922 } 923 924 /* TCP_SKB_CB reference means this can not be used from early demux */ 925 static inline int tcp_v6_sdif(const struct sk_buff *skb) 926 { 927 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 928 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 929 return TCP_SKB_CB(skb)->header.h6.iif; 930 #endif 931 return 0; 932 } 933 #endif 934 935 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 936 { 937 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 938 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 939 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 940 return true; 941 #endif 942 return false; 943 } 944 945 /* TCP_SKB_CB reference means this can not be used from early demux */ 946 static inline int tcp_v4_sdif(struct sk_buff *skb) 947 { 948 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 949 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 950 return TCP_SKB_CB(skb)->header.h4.iif; 951 #endif 952 return 0; 953 } 954 955 /* Due to TSO, an SKB can be composed of multiple actual 956 * packets. To keep these tracked properly, we use this. 957 */ 958 static inline int tcp_skb_pcount(const struct sk_buff *skb) 959 { 960 return TCP_SKB_CB(skb)->tcp_gso_segs; 961 } 962 963 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 964 { 965 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 966 } 967 968 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 969 { 970 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 971 } 972 973 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 974 static inline int tcp_skb_mss(const struct sk_buff *skb) 975 { 976 return TCP_SKB_CB(skb)->tcp_gso_size; 977 } 978 979 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 980 { 981 return likely(!TCP_SKB_CB(skb)->eor); 982 } 983 984 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 985 const struct sk_buff *from) 986 { 987 return likely(tcp_skb_can_collapse_to(to) && 988 mptcp_skb_can_collapse(to, from)); 989 } 990 991 /* Events passed to congestion control interface */ 992 enum tcp_ca_event { 993 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 994 CA_EVENT_CWND_RESTART, /* congestion window restart */ 995 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 996 CA_EVENT_LOSS, /* loss timeout */ 997 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 998 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 999 }; 1000 1001 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1002 enum tcp_ca_ack_event_flags { 1003 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1004 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1005 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1006 }; 1007 1008 /* 1009 * Interface for adding new TCP congestion control handlers 1010 */ 1011 #define TCP_CA_NAME_MAX 16 1012 #define TCP_CA_MAX 128 1013 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1014 1015 #define TCP_CA_UNSPEC 0 1016 1017 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1018 #define TCP_CONG_NON_RESTRICTED 0x1 1019 /* Requires ECN/ECT set on all packets */ 1020 #define TCP_CONG_NEEDS_ECN 0x2 1021 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1022 1023 union tcp_cc_info; 1024 1025 struct ack_sample { 1026 u32 pkts_acked; 1027 s32 rtt_us; 1028 u32 in_flight; 1029 }; 1030 1031 /* A rate sample measures the number of (original/retransmitted) data 1032 * packets delivered "delivered" over an interval of time "interval_us". 1033 * The tcp_rate.c code fills in the rate sample, and congestion 1034 * control modules that define a cong_control function to run at the end 1035 * of ACK processing can optionally chose to consult this sample when 1036 * setting cwnd and pacing rate. 1037 * A sample is invalid if "delivered" or "interval_us" is negative. 1038 */ 1039 struct rate_sample { 1040 u64 prior_mstamp; /* starting timestamp for interval */ 1041 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1042 s32 delivered; /* number of packets delivered over interval */ 1043 long interval_us; /* time for tp->delivered to incr "delivered" */ 1044 u32 snd_interval_us; /* snd interval for delivered packets */ 1045 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1046 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1047 int losses; /* number of packets marked lost upon ACK */ 1048 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1049 u32 prior_in_flight; /* in flight before this ACK */ 1050 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1051 bool is_retrans; /* is sample from retransmission? */ 1052 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1053 }; 1054 1055 struct tcp_congestion_ops { 1056 struct list_head list; 1057 u32 key; 1058 u32 flags; 1059 1060 /* initialize private data (optional) */ 1061 void (*init)(struct sock *sk); 1062 /* cleanup private data (optional) */ 1063 void (*release)(struct sock *sk); 1064 1065 /* return slow start threshold (required) */ 1066 u32 (*ssthresh)(struct sock *sk); 1067 /* do new cwnd calculation (required) */ 1068 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1069 /* call before changing ca_state (optional) */ 1070 void (*set_state)(struct sock *sk, u8 new_state); 1071 /* call when cwnd event occurs (optional) */ 1072 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1073 /* call when ack arrives (optional) */ 1074 void (*in_ack_event)(struct sock *sk, u32 flags); 1075 /* new value of cwnd after loss (required) */ 1076 u32 (*undo_cwnd)(struct sock *sk); 1077 /* hook for packet ack accounting (optional) */ 1078 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1079 /* override sysctl_tcp_min_tso_segs */ 1080 u32 (*min_tso_segs)(struct sock *sk); 1081 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1082 u32 (*sndbuf_expand)(struct sock *sk); 1083 /* call when packets are delivered to update cwnd and pacing rate, 1084 * after all the ca_state processing. (optional) 1085 */ 1086 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1087 /* get info for inet_diag (optional) */ 1088 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1089 union tcp_cc_info *info); 1090 1091 char name[TCP_CA_NAME_MAX]; 1092 struct module *owner; 1093 }; 1094 1095 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1096 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1097 1098 void tcp_assign_congestion_control(struct sock *sk); 1099 void tcp_init_congestion_control(struct sock *sk); 1100 void tcp_cleanup_congestion_control(struct sock *sk); 1101 int tcp_set_default_congestion_control(struct net *net, const char *name); 1102 void tcp_get_default_congestion_control(struct net *net, char *name); 1103 void tcp_get_available_congestion_control(char *buf, size_t len); 1104 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1105 int tcp_set_allowed_congestion_control(char *allowed); 1106 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1107 bool reinit, bool cap_net_admin); 1108 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1109 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1110 1111 u32 tcp_reno_ssthresh(struct sock *sk); 1112 u32 tcp_reno_undo_cwnd(struct sock *sk); 1113 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1114 extern struct tcp_congestion_ops tcp_reno; 1115 1116 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1117 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1118 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1119 #ifdef CONFIG_INET 1120 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1121 #else 1122 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1123 { 1124 return NULL; 1125 } 1126 #endif 1127 1128 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1129 { 1130 const struct inet_connection_sock *icsk = inet_csk(sk); 1131 1132 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1133 } 1134 1135 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1136 { 1137 struct inet_connection_sock *icsk = inet_csk(sk); 1138 1139 if (icsk->icsk_ca_ops->set_state) 1140 icsk->icsk_ca_ops->set_state(sk, ca_state); 1141 icsk->icsk_ca_state = ca_state; 1142 } 1143 1144 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1145 { 1146 const struct inet_connection_sock *icsk = inet_csk(sk); 1147 1148 if (icsk->icsk_ca_ops->cwnd_event) 1149 icsk->icsk_ca_ops->cwnd_event(sk, event); 1150 } 1151 1152 /* From tcp_rate.c */ 1153 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1154 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1155 struct rate_sample *rs); 1156 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1157 bool is_sack_reneg, struct rate_sample *rs); 1158 void tcp_rate_check_app_limited(struct sock *sk); 1159 1160 /* These functions determine how the current flow behaves in respect of SACK 1161 * handling. SACK is negotiated with the peer, and therefore it can vary 1162 * between different flows. 1163 * 1164 * tcp_is_sack - SACK enabled 1165 * tcp_is_reno - No SACK 1166 */ 1167 static inline int tcp_is_sack(const struct tcp_sock *tp) 1168 { 1169 return likely(tp->rx_opt.sack_ok); 1170 } 1171 1172 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1173 { 1174 return !tcp_is_sack(tp); 1175 } 1176 1177 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1178 { 1179 return tp->sacked_out + tp->lost_out; 1180 } 1181 1182 /* This determines how many packets are "in the network" to the best 1183 * of our knowledge. In many cases it is conservative, but where 1184 * detailed information is available from the receiver (via SACK 1185 * blocks etc.) we can make more aggressive calculations. 1186 * 1187 * Use this for decisions involving congestion control, use just 1188 * tp->packets_out to determine if the send queue is empty or not. 1189 * 1190 * Read this equation as: 1191 * 1192 * "Packets sent once on transmission queue" MINUS 1193 * "Packets left network, but not honestly ACKed yet" PLUS 1194 * "Packets fast retransmitted" 1195 */ 1196 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1197 { 1198 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1199 } 1200 1201 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1202 1203 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1204 { 1205 return tp->snd_cwnd < tp->snd_ssthresh; 1206 } 1207 1208 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1209 { 1210 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1211 } 1212 1213 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1214 { 1215 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1216 (1 << inet_csk(sk)->icsk_ca_state); 1217 } 1218 1219 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1220 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1221 * ssthresh. 1222 */ 1223 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1224 { 1225 const struct tcp_sock *tp = tcp_sk(sk); 1226 1227 if (tcp_in_cwnd_reduction(sk)) 1228 return tp->snd_ssthresh; 1229 else 1230 return max(tp->snd_ssthresh, 1231 ((tp->snd_cwnd >> 1) + 1232 (tp->snd_cwnd >> 2))); 1233 } 1234 1235 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1236 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1237 1238 void tcp_enter_cwr(struct sock *sk); 1239 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1240 1241 /* The maximum number of MSS of available cwnd for which TSO defers 1242 * sending if not using sysctl_tcp_tso_win_divisor. 1243 */ 1244 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1245 { 1246 return 3; 1247 } 1248 1249 /* Returns end sequence number of the receiver's advertised window */ 1250 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1251 { 1252 return tp->snd_una + tp->snd_wnd; 1253 } 1254 1255 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1256 * flexible approach. The RFC suggests cwnd should not be raised unless 1257 * it was fully used previously. And that's exactly what we do in 1258 * congestion avoidance mode. But in slow start we allow cwnd to grow 1259 * as long as the application has used half the cwnd. 1260 * Example : 1261 * cwnd is 10 (IW10), but application sends 9 frames. 1262 * We allow cwnd to reach 18 when all frames are ACKed. 1263 * This check is safe because it's as aggressive as slow start which already 1264 * risks 100% overshoot. The advantage is that we discourage application to 1265 * either send more filler packets or data to artificially blow up the cwnd 1266 * usage, and allow application-limited process to probe bw more aggressively. 1267 */ 1268 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1269 { 1270 const struct tcp_sock *tp = tcp_sk(sk); 1271 1272 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1273 if (tcp_in_slow_start(tp)) 1274 return tp->snd_cwnd < 2 * tp->max_packets_out; 1275 1276 return tp->is_cwnd_limited; 1277 } 1278 1279 /* BBR congestion control needs pacing. 1280 * Same remark for SO_MAX_PACING_RATE. 1281 * sch_fq packet scheduler is efficiently handling pacing, 1282 * but is not always installed/used. 1283 * Return true if TCP stack should pace packets itself. 1284 */ 1285 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1286 { 1287 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1288 } 1289 1290 /* Return in jiffies the delay before one skb is sent. 1291 * If @skb is NULL, we look at EDT for next packet being sent on the socket. 1292 */ 1293 static inline unsigned long tcp_pacing_delay(const struct sock *sk, 1294 const struct sk_buff *skb) 1295 { 1296 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns; 1297 1298 pacing_delay -= tcp_sk(sk)->tcp_clock_cache; 1299 1300 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0; 1301 } 1302 1303 static inline void tcp_reset_xmit_timer(struct sock *sk, 1304 const int what, 1305 unsigned long when, 1306 const unsigned long max_when, 1307 const struct sk_buff *skb) 1308 { 1309 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb), 1310 max_when); 1311 } 1312 1313 /* Something is really bad, we could not queue an additional packet, 1314 * because qdisc is full or receiver sent a 0 window, or we are paced. 1315 * We do not want to add fuel to the fire, or abort too early, 1316 * so make sure the timer we arm now is at least 200ms in the future, 1317 * regardless of current icsk_rto value (as it could be ~2ms) 1318 */ 1319 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1320 { 1321 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1322 } 1323 1324 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1325 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1326 unsigned long max_when) 1327 { 1328 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1329 1330 return (unsigned long)min_t(u64, when, max_when); 1331 } 1332 1333 static inline void tcp_check_probe_timer(struct sock *sk) 1334 { 1335 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1336 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1337 tcp_probe0_base(sk), TCP_RTO_MAX, 1338 NULL); 1339 } 1340 1341 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1342 { 1343 tp->snd_wl1 = seq; 1344 } 1345 1346 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1347 { 1348 tp->snd_wl1 = seq; 1349 } 1350 1351 /* 1352 * Calculate(/check) TCP checksum 1353 */ 1354 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1355 __be32 daddr, __wsum base) 1356 { 1357 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1358 } 1359 1360 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1361 { 1362 return !skb_csum_unnecessary(skb) && 1363 __skb_checksum_complete(skb); 1364 } 1365 1366 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1367 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1368 void tcp_set_state(struct sock *sk, int state); 1369 void tcp_done(struct sock *sk); 1370 int tcp_abort(struct sock *sk, int err); 1371 1372 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1373 { 1374 rx_opt->dsack = 0; 1375 rx_opt->num_sacks = 0; 1376 } 1377 1378 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1379 1380 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1381 { 1382 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1383 struct tcp_sock *tp = tcp_sk(sk); 1384 s32 delta; 1385 1386 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1387 ca_ops->cong_control) 1388 return; 1389 delta = tcp_jiffies32 - tp->lsndtime; 1390 if (delta > inet_csk(sk)->icsk_rto) 1391 tcp_cwnd_restart(sk, delta); 1392 } 1393 1394 /* Determine a window scaling and initial window to offer. */ 1395 void tcp_select_initial_window(const struct sock *sk, int __space, 1396 __u32 mss, __u32 *rcv_wnd, 1397 __u32 *window_clamp, int wscale_ok, 1398 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1399 1400 static inline int tcp_win_from_space(const struct sock *sk, int space) 1401 { 1402 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1403 1404 return tcp_adv_win_scale <= 0 ? 1405 (space>>(-tcp_adv_win_scale)) : 1406 space - (space>>tcp_adv_win_scale); 1407 } 1408 1409 /* Note: caller must be prepared to deal with negative returns */ 1410 static inline int tcp_space(const struct sock *sk) 1411 { 1412 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1413 READ_ONCE(sk->sk_backlog.len) - 1414 atomic_read(&sk->sk_rmem_alloc)); 1415 } 1416 1417 static inline int tcp_full_space(const struct sock *sk) 1418 { 1419 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1420 } 1421 1422 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1423 * If 87.5 % (7/8) of the space has been consumed, we want to override 1424 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1425 * len/truesize ratio. 1426 */ 1427 static inline bool tcp_rmem_pressure(const struct sock *sk) 1428 { 1429 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1430 int threshold = rcvbuf - (rcvbuf >> 3); 1431 1432 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1433 } 1434 1435 extern void tcp_openreq_init_rwin(struct request_sock *req, 1436 const struct sock *sk_listener, 1437 const struct dst_entry *dst); 1438 1439 void tcp_enter_memory_pressure(struct sock *sk); 1440 void tcp_leave_memory_pressure(struct sock *sk); 1441 1442 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1443 { 1444 struct net *net = sock_net((struct sock *)tp); 1445 1446 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1447 } 1448 1449 static inline int keepalive_time_when(const struct tcp_sock *tp) 1450 { 1451 struct net *net = sock_net((struct sock *)tp); 1452 1453 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1454 } 1455 1456 static inline int keepalive_probes(const struct tcp_sock *tp) 1457 { 1458 struct net *net = sock_net((struct sock *)tp); 1459 1460 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1461 } 1462 1463 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1464 { 1465 const struct inet_connection_sock *icsk = &tp->inet_conn; 1466 1467 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1468 tcp_jiffies32 - tp->rcv_tstamp); 1469 } 1470 1471 static inline int tcp_fin_time(const struct sock *sk) 1472 { 1473 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1474 const int rto = inet_csk(sk)->icsk_rto; 1475 1476 if (fin_timeout < (rto << 2) - (rto >> 1)) 1477 fin_timeout = (rto << 2) - (rto >> 1); 1478 1479 return fin_timeout; 1480 } 1481 1482 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1483 int paws_win) 1484 { 1485 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1486 return true; 1487 if (unlikely(!time_before32(ktime_get_seconds(), 1488 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1489 return true; 1490 /* 1491 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1492 * then following tcp messages have valid values. Ignore 0 value, 1493 * or else 'negative' tsval might forbid us to accept their packets. 1494 */ 1495 if (!rx_opt->ts_recent) 1496 return true; 1497 return false; 1498 } 1499 1500 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1501 int rst) 1502 { 1503 if (tcp_paws_check(rx_opt, 0)) 1504 return false; 1505 1506 /* RST segments are not recommended to carry timestamp, 1507 and, if they do, it is recommended to ignore PAWS because 1508 "their cleanup function should take precedence over timestamps." 1509 Certainly, it is mistake. It is necessary to understand the reasons 1510 of this constraint to relax it: if peer reboots, clock may go 1511 out-of-sync and half-open connections will not be reset. 1512 Actually, the problem would be not existing if all 1513 the implementations followed draft about maintaining clock 1514 via reboots. Linux-2.2 DOES NOT! 1515 1516 However, we can relax time bounds for RST segments to MSL. 1517 */ 1518 if (rst && !time_before32(ktime_get_seconds(), 1519 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1520 return false; 1521 return true; 1522 } 1523 1524 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1525 int mib_idx, u32 *last_oow_ack_time); 1526 1527 static inline void tcp_mib_init(struct net *net) 1528 { 1529 /* See RFC 2012 */ 1530 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1531 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1532 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1533 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1534 } 1535 1536 /* from STCP */ 1537 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1538 { 1539 tp->lost_skb_hint = NULL; 1540 } 1541 1542 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1543 { 1544 tcp_clear_retrans_hints_partial(tp); 1545 tp->retransmit_skb_hint = NULL; 1546 } 1547 1548 union tcp_md5_addr { 1549 struct in_addr a4; 1550 #if IS_ENABLED(CONFIG_IPV6) 1551 struct in6_addr a6; 1552 #endif 1553 }; 1554 1555 /* - key database */ 1556 struct tcp_md5sig_key { 1557 struct hlist_node node; 1558 u8 keylen; 1559 u8 family; /* AF_INET or AF_INET6 */ 1560 u8 prefixlen; 1561 union tcp_md5_addr addr; 1562 int l3index; /* set if key added with L3 scope */ 1563 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1564 struct rcu_head rcu; 1565 }; 1566 1567 /* - sock block */ 1568 struct tcp_md5sig_info { 1569 struct hlist_head head; 1570 struct rcu_head rcu; 1571 }; 1572 1573 /* - pseudo header */ 1574 struct tcp4_pseudohdr { 1575 __be32 saddr; 1576 __be32 daddr; 1577 __u8 pad; 1578 __u8 protocol; 1579 __be16 len; 1580 }; 1581 1582 struct tcp6_pseudohdr { 1583 struct in6_addr saddr; 1584 struct in6_addr daddr; 1585 __be32 len; 1586 __be32 protocol; /* including padding */ 1587 }; 1588 1589 union tcp_md5sum_block { 1590 struct tcp4_pseudohdr ip4; 1591 #if IS_ENABLED(CONFIG_IPV6) 1592 struct tcp6_pseudohdr ip6; 1593 #endif 1594 }; 1595 1596 /* - pool: digest algorithm, hash description and scratch buffer */ 1597 struct tcp_md5sig_pool { 1598 struct ahash_request *md5_req; 1599 void *scratch; 1600 }; 1601 1602 /* - functions */ 1603 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1604 const struct sock *sk, const struct sk_buff *skb); 1605 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1606 int family, u8 prefixlen, int l3index, 1607 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1608 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1609 int family, u8 prefixlen, int l3index); 1610 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1611 const struct sock *addr_sk); 1612 1613 #ifdef CONFIG_TCP_MD5SIG 1614 #include <linux/jump_label.h> 1615 extern struct static_key_false tcp_md5_needed; 1616 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1617 const union tcp_md5_addr *addr, 1618 int family); 1619 static inline struct tcp_md5sig_key * 1620 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1621 const union tcp_md5_addr *addr, int family) 1622 { 1623 if (!static_branch_unlikely(&tcp_md5_needed)) 1624 return NULL; 1625 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1626 } 1627 1628 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1629 #else 1630 static inline struct tcp_md5sig_key * 1631 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1632 const union tcp_md5_addr *addr, int family) 1633 { 1634 return NULL; 1635 } 1636 #define tcp_twsk_md5_key(twsk) NULL 1637 #endif 1638 1639 bool tcp_alloc_md5sig_pool(void); 1640 1641 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1642 static inline void tcp_put_md5sig_pool(void) 1643 { 1644 local_bh_enable(); 1645 } 1646 1647 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1648 unsigned int header_len); 1649 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1650 const struct tcp_md5sig_key *key); 1651 1652 /* From tcp_fastopen.c */ 1653 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1654 struct tcp_fastopen_cookie *cookie); 1655 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1656 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1657 u16 try_exp); 1658 struct tcp_fastopen_request { 1659 /* Fast Open cookie. Size 0 means a cookie request */ 1660 struct tcp_fastopen_cookie cookie; 1661 struct msghdr *data; /* data in MSG_FASTOPEN */ 1662 size_t size; 1663 int copied; /* queued in tcp_connect() */ 1664 struct ubuf_info *uarg; 1665 }; 1666 void tcp_free_fastopen_req(struct tcp_sock *tp); 1667 void tcp_fastopen_destroy_cipher(struct sock *sk); 1668 void tcp_fastopen_ctx_destroy(struct net *net); 1669 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1670 void *primary_key, void *backup_key); 1671 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1672 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1673 struct request_sock *req, 1674 struct tcp_fastopen_cookie *foc, 1675 const struct dst_entry *dst); 1676 void tcp_fastopen_init_key_once(struct net *net); 1677 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1678 struct tcp_fastopen_cookie *cookie); 1679 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1680 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1681 #define TCP_FASTOPEN_KEY_MAX 2 1682 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1683 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1684 1685 /* Fastopen key context */ 1686 struct tcp_fastopen_context { 1687 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1688 int num; 1689 struct rcu_head rcu; 1690 }; 1691 1692 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1693 void tcp_fastopen_active_disable(struct sock *sk); 1694 bool tcp_fastopen_active_should_disable(struct sock *sk); 1695 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1696 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1697 1698 /* Caller needs to wrap with rcu_read_(un)lock() */ 1699 static inline 1700 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1701 { 1702 struct tcp_fastopen_context *ctx; 1703 1704 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1705 if (!ctx) 1706 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1707 return ctx; 1708 } 1709 1710 static inline 1711 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1712 const struct tcp_fastopen_cookie *orig) 1713 { 1714 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1715 orig->len == foc->len && 1716 !memcmp(orig->val, foc->val, foc->len)) 1717 return true; 1718 return false; 1719 } 1720 1721 static inline 1722 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1723 { 1724 return ctx->num; 1725 } 1726 1727 /* Latencies incurred by various limits for a sender. They are 1728 * chronograph-like stats that are mutually exclusive. 1729 */ 1730 enum tcp_chrono { 1731 TCP_CHRONO_UNSPEC, 1732 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1733 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1734 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1735 __TCP_CHRONO_MAX, 1736 }; 1737 1738 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1739 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1740 1741 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1742 * the same memory storage than skb->destructor/_skb_refdst 1743 */ 1744 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1745 { 1746 skb->destructor = NULL; 1747 skb->_skb_refdst = 0UL; 1748 } 1749 1750 #define tcp_skb_tsorted_save(skb) { \ 1751 unsigned long _save = skb->_skb_refdst; \ 1752 skb->_skb_refdst = 0UL; 1753 1754 #define tcp_skb_tsorted_restore(skb) \ 1755 skb->_skb_refdst = _save; \ 1756 } 1757 1758 void tcp_write_queue_purge(struct sock *sk); 1759 1760 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1761 { 1762 return skb_rb_first(&sk->tcp_rtx_queue); 1763 } 1764 1765 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1766 { 1767 return skb_rb_last(&sk->tcp_rtx_queue); 1768 } 1769 1770 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1771 { 1772 return skb_peek(&sk->sk_write_queue); 1773 } 1774 1775 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1776 { 1777 return skb_peek_tail(&sk->sk_write_queue); 1778 } 1779 1780 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1781 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1782 1783 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1784 { 1785 return skb_peek(&sk->sk_write_queue); 1786 } 1787 1788 static inline bool tcp_skb_is_last(const struct sock *sk, 1789 const struct sk_buff *skb) 1790 { 1791 return skb_queue_is_last(&sk->sk_write_queue, skb); 1792 } 1793 1794 /** 1795 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1796 * @sk: socket 1797 * 1798 * Since the write queue can have a temporary empty skb in it, 1799 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1800 */ 1801 static inline bool tcp_write_queue_empty(const struct sock *sk) 1802 { 1803 const struct tcp_sock *tp = tcp_sk(sk); 1804 1805 return tp->write_seq == tp->snd_nxt; 1806 } 1807 1808 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1809 { 1810 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1811 } 1812 1813 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1814 { 1815 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1816 } 1817 1818 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1819 { 1820 __skb_queue_tail(&sk->sk_write_queue, skb); 1821 1822 /* Queue it, remembering where we must start sending. */ 1823 if (sk->sk_write_queue.next == skb) 1824 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1825 } 1826 1827 /* Insert new before skb on the write queue of sk. */ 1828 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1829 struct sk_buff *skb, 1830 struct sock *sk) 1831 { 1832 __skb_queue_before(&sk->sk_write_queue, skb, new); 1833 } 1834 1835 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1836 { 1837 tcp_skb_tsorted_anchor_cleanup(skb); 1838 __skb_unlink(skb, &sk->sk_write_queue); 1839 } 1840 1841 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1842 1843 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1844 { 1845 tcp_skb_tsorted_anchor_cleanup(skb); 1846 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1847 } 1848 1849 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1850 { 1851 list_del(&skb->tcp_tsorted_anchor); 1852 tcp_rtx_queue_unlink(skb, sk); 1853 sk_wmem_free_skb(sk, skb); 1854 } 1855 1856 static inline void tcp_push_pending_frames(struct sock *sk) 1857 { 1858 if (tcp_send_head(sk)) { 1859 struct tcp_sock *tp = tcp_sk(sk); 1860 1861 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1862 } 1863 } 1864 1865 /* Start sequence of the skb just after the highest skb with SACKed 1866 * bit, valid only if sacked_out > 0 or when the caller has ensured 1867 * validity by itself. 1868 */ 1869 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1870 { 1871 if (!tp->sacked_out) 1872 return tp->snd_una; 1873 1874 if (tp->highest_sack == NULL) 1875 return tp->snd_nxt; 1876 1877 return TCP_SKB_CB(tp->highest_sack)->seq; 1878 } 1879 1880 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1881 { 1882 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1883 } 1884 1885 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1886 { 1887 return tcp_sk(sk)->highest_sack; 1888 } 1889 1890 static inline void tcp_highest_sack_reset(struct sock *sk) 1891 { 1892 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1893 } 1894 1895 /* Called when old skb is about to be deleted and replaced by new skb */ 1896 static inline void tcp_highest_sack_replace(struct sock *sk, 1897 struct sk_buff *old, 1898 struct sk_buff *new) 1899 { 1900 if (old == tcp_highest_sack(sk)) 1901 tcp_sk(sk)->highest_sack = new; 1902 } 1903 1904 /* This helper checks if socket has IP_TRANSPARENT set */ 1905 static inline bool inet_sk_transparent(const struct sock *sk) 1906 { 1907 switch (sk->sk_state) { 1908 case TCP_TIME_WAIT: 1909 return inet_twsk(sk)->tw_transparent; 1910 case TCP_NEW_SYN_RECV: 1911 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1912 } 1913 return inet_sk(sk)->transparent; 1914 } 1915 1916 /* Determines whether this is a thin stream (which may suffer from 1917 * increased latency). Used to trigger latency-reducing mechanisms. 1918 */ 1919 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1920 { 1921 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1922 } 1923 1924 /* /proc */ 1925 enum tcp_seq_states { 1926 TCP_SEQ_STATE_LISTENING, 1927 TCP_SEQ_STATE_ESTABLISHED, 1928 }; 1929 1930 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1931 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1932 void tcp_seq_stop(struct seq_file *seq, void *v); 1933 1934 struct tcp_seq_afinfo { 1935 sa_family_t family; 1936 }; 1937 1938 struct tcp_iter_state { 1939 struct seq_net_private p; 1940 enum tcp_seq_states state; 1941 struct sock *syn_wait_sk; 1942 int bucket, offset, sbucket, num; 1943 loff_t last_pos; 1944 }; 1945 1946 extern struct request_sock_ops tcp_request_sock_ops; 1947 extern struct request_sock_ops tcp6_request_sock_ops; 1948 1949 void tcp_v4_destroy_sock(struct sock *sk); 1950 1951 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1952 netdev_features_t features); 1953 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1954 int tcp_gro_complete(struct sk_buff *skb); 1955 1956 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1957 1958 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1959 { 1960 struct net *net = sock_net((struct sock *)tp); 1961 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1962 } 1963 1964 /* @wake is one when sk_stream_write_space() calls us. 1965 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1966 * This mimics the strategy used in sock_def_write_space(). 1967 */ 1968 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1969 { 1970 const struct tcp_sock *tp = tcp_sk(sk); 1971 u32 notsent_bytes = READ_ONCE(tp->write_seq) - 1972 READ_ONCE(tp->snd_nxt); 1973 1974 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1975 } 1976 1977 #ifdef CONFIG_PROC_FS 1978 int tcp4_proc_init(void); 1979 void tcp4_proc_exit(void); 1980 #endif 1981 1982 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1983 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1984 const struct tcp_request_sock_ops *af_ops, 1985 struct sock *sk, struct sk_buff *skb); 1986 1987 /* TCP af-specific functions */ 1988 struct tcp_sock_af_ops { 1989 #ifdef CONFIG_TCP_MD5SIG 1990 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1991 const struct sock *addr_sk); 1992 int (*calc_md5_hash)(char *location, 1993 const struct tcp_md5sig_key *md5, 1994 const struct sock *sk, 1995 const struct sk_buff *skb); 1996 int (*md5_parse)(struct sock *sk, 1997 int optname, 1998 char __user *optval, 1999 int optlen); 2000 #endif 2001 }; 2002 2003 struct tcp_request_sock_ops { 2004 u16 mss_clamp; 2005 #ifdef CONFIG_TCP_MD5SIG 2006 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2007 const struct sock *addr_sk); 2008 int (*calc_md5_hash) (char *location, 2009 const struct tcp_md5sig_key *md5, 2010 const struct sock *sk, 2011 const struct sk_buff *skb); 2012 #endif 2013 void (*init_req)(struct request_sock *req, 2014 const struct sock *sk_listener, 2015 struct sk_buff *skb); 2016 #ifdef CONFIG_SYN_COOKIES 2017 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2018 __u16 *mss); 2019 #endif 2020 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 2021 const struct request_sock *req); 2022 u32 (*init_seq)(const struct sk_buff *skb); 2023 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2024 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2025 struct flowi *fl, struct request_sock *req, 2026 struct tcp_fastopen_cookie *foc, 2027 enum tcp_synack_type synack_type); 2028 }; 2029 2030 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2031 #if IS_ENABLED(CONFIG_IPV6) 2032 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2033 #endif 2034 2035 #ifdef CONFIG_SYN_COOKIES 2036 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2037 const struct sock *sk, struct sk_buff *skb, 2038 __u16 *mss) 2039 { 2040 tcp_synq_overflow(sk); 2041 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2042 return ops->cookie_init_seq(skb, mss); 2043 } 2044 #else 2045 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2046 const struct sock *sk, struct sk_buff *skb, 2047 __u16 *mss) 2048 { 2049 return 0; 2050 } 2051 #endif 2052 2053 int tcpv4_offload_init(void); 2054 2055 void tcp_v4_init(void); 2056 void tcp_init(void); 2057 2058 /* tcp_recovery.c */ 2059 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2060 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2061 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2062 u32 reo_wnd); 2063 extern void tcp_rack_mark_lost(struct sock *sk); 2064 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2065 u64 xmit_time); 2066 extern void tcp_rack_reo_timeout(struct sock *sk); 2067 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2068 2069 /* At how many usecs into the future should the RTO fire? */ 2070 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2071 { 2072 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2073 u32 rto = inet_csk(sk)->icsk_rto; 2074 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2075 2076 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2077 } 2078 2079 /* 2080 * Save and compile IPv4 options, return a pointer to it 2081 */ 2082 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2083 struct sk_buff *skb) 2084 { 2085 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2086 struct ip_options_rcu *dopt = NULL; 2087 2088 if (opt->optlen) { 2089 int opt_size = sizeof(*dopt) + opt->optlen; 2090 2091 dopt = kmalloc(opt_size, GFP_ATOMIC); 2092 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2093 kfree(dopt); 2094 dopt = NULL; 2095 } 2096 } 2097 return dopt; 2098 } 2099 2100 /* locally generated TCP pure ACKs have skb->truesize == 2 2101 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2102 * This is much faster than dissecting the packet to find out. 2103 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2104 */ 2105 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2106 { 2107 return skb->truesize == 2; 2108 } 2109 2110 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2111 { 2112 skb->truesize = 2; 2113 } 2114 2115 static inline int tcp_inq(struct sock *sk) 2116 { 2117 struct tcp_sock *tp = tcp_sk(sk); 2118 int answ; 2119 2120 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2121 answ = 0; 2122 } else if (sock_flag(sk, SOCK_URGINLINE) || 2123 !tp->urg_data || 2124 before(tp->urg_seq, tp->copied_seq) || 2125 !before(tp->urg_seq, tp->rcv_nxt)) { 2126 2127 answ = tp->rcv_nxt - tp->copied_seq; 2128 2129 /* Subtract 1, if FIN was received */ 2130 if (answ && sock_flag(sk, SOCK_DONE)) 2131 answ--; 2132 } else { 2133 answ = tp->urg_seq - tp->copied_seq; 2134 } 2135 2136 return answ; 2137 } 2138 2139 int tcp_peek_len(struct socket *sock); 2140 2141 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2142 { 2143 u16 segs_in; 2144 2145 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2146 tp->segs_in += segs_in; 2147 if (skb->len > tcp_hdrlen(skb)) 2148 tp->data_segs_in += segs_in; 2149 } 2150 2151 /* 2152 * TCP listen path runs lockless. 2153 * We forced "struct sock" to be const qualified to make sure 2154 * we don't modify one of its field by mistake. 2155 * Here, we increment sk_drops which is an atomic_t, so we can safely 2156 * make sock writable again. 2157 */ 2158 static inline void tcp_listendrop(const struct sock *sk) 2159 { 2160 atomic_inc(&((struct sock *)sk)->sk_drops); 2161 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2162 } 2163 2164 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2165 2166 /* 2167 * Interface for adding Upper Level Protocols over TCP 2168 */ 2169 2170 #define TCP_ULP_NAME_MAX 16 2171 #define TCP_ULP_MAX 128 2172 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2173 2174 struct tcp_ulp_ops { 2175 struct list_head list; 2176 2177 /* initialize ulp */ 2178 int (*init)(struct sock *sk); 2179 /* update ulp */ 2180 void (*update)(struct sock *sk, struct proto *p, 2181 void (*write_space)(struct sock *sk)); 2182 /* cleanup ulp */ 2183 void (*release)(struct sock *sk); 2184 /* diagnostic */ 2185 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2186 size_t (*get_info_size)(const struct sock *sk); 2187 /* clone ulp */ 2188 void (*clone)(const struct request_sock *req, struct sock *newsk, 2189 const gfp_t priority); 2190 2191 char name[TCP_ULP_NAME_MAX]; 2192 struct module *owner; 2193 }; 2194 int tcp_register_ulp(struct tcp_ulp_ops *type); 2195 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2196 int tcp_set_ulp(struct sock *sk, const char *name); 2197 void tcp_get_available_ulp(char *buf, size_t len); 2198 void tcp_cleanup_ulp(struct sock *sk); 2199 void tcp_update_ulp(struct sock *sk, struct proto *p, 2200 void (*write_space)(struct sock *sk)); 2201 2202 #define MODULE_ALIAS_TCP_ULP(name) \ 2203 __MODULE_INFO(alias, alias_userspace, name); \ 2204 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2205 2206 struct sk_msg; 2207 struct sk_psock; 2208 2209 #ifdef CONFIG_BPF_STREAM_PARSER 2210 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2211 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2212 #else 2213 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2214 { 2215 } 2216 #endif /* CONFIG_BPF_STREAM_PARSER */ 2217 2218 #ifdef CONFIG_NET_SOCK_MSG 2219 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2220 int flags); 2221 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2222 struct msghdr *msg, int len, int flags); 2223 #endif /* CONFIG_NET_SOCK_MSG */ 2224 2225 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2226 * is < 0, then the BPF op failed (for example if the loaded BPF 2227 * program does not support the chosen operation or there is no BPF 2228 * program loaded). 2229 */ 2230 #ifdef CONFIG_BPF 2231 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2232 { 2233 struct bpf_sock_ops_kern sock_ops; 2234 int ret; 2235 2236 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2237 if (sk_fullsock(sk)) { 2238 sock_ops.is_fullsock = 1; 2239 sock_owned_by_me(sk); 2240 } 2241 2242 sock_ops.sk = sk; 2243 sock_ops.op = op; 2244 if (nargs > 0) 2245 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2246 2247 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2248 if (ret == 0) 2249 ret = sock_ops.reply; 2250 else 2251 ret = -1; 2252 return ret; 2253 } 2254 2255 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2256 { 2257 u32 args[2] = {arg1, arg2}; 2258 2259 return tcp_call_bpf(sk, op, 2, args); 2260 } 2261 2262 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2263 u32 arg3) 2264 { 2265 u32 args[3] = {arg1, arg2, arg3}; 2266 2267 return tcp_call_bpf(sk, op, 3, args); 2268 } 2269 2270 #else 2271 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2272 { 2273 return -EPERM; 2274 } 2275 2276 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2277 { 2278 return -EPERM; 2279 } 2280 2281 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2282 u32 arg3) 2283 { 2284 return -EPERM; 2285 } 2286 2287 #endif 2288 2289 static inline u32 tcp_timeout_init(struct sock *sk) 2290 { 2291 int timeout; 2292 2293 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2294 2295 if (timeout <= 0) 2296 timeout = TCP_TIMEOUT_INIT; 2297 return timeout; 2298 } 2299 2300 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2301 { 2302 int rwnd; 2303 2304 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2305 2306 if (rwnd < 0) 2307 rwnd = 0; 2308 return rwnd; 2309 } 2310 2311 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2312 { 2313 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2314 } 2315 2316 static inline void tcp_bpf_rtt(struct sock *sk) 2317 { 2318 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2319 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2320 } 2321 2322 #if IS_ENABLED(CONFIG_SMC) 2323 extern struct static_key_false tcp_have_smc; 2324 #endif 2325 2326 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2327 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2328 void (*cad)(struct sock *sk, u32 ack_seq)); 2329 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2330 void clean_acked_data_flush(void); 2331 #endif 2332 2333 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2334 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2335 const struct tcp_sock *tp) 2336 { 2337 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2338 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2339 } 2340 2341 /* Compute Earliest Departure Time for some control packets 2342 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2343 */ 2344 static inline u64 tcp_transmit_time(const struct sock *sk) 2345 { 2346 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2347 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2348 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2349 2350 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2351 } 2352 return 0; 2353 } 2354 2355 #endif /* _TCP_H */ 2356