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 #include <linux/indirect_call_wrapper.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/inet_ecn.h> 41 #include <net/dst.h> 42 #include <net/mptcp.h> 43 44 #include <linux/seq_file.h> 45 #include <linux/memcontrol.h> 46 #include <linux/bpf-cgroup.h> 47 #include <linux/siphash.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 extern struct percpu_counter tcp_orphan_count; 52 void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 55 #define MAX_TCP_OPTION_SPACE 40 56 #define TCP_MIN_SND_MSS 48 57 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 58 59 /* 60 * Never offer a window over 32767 without using window scaling. Some 61 * poor stacks do signed 16bit maths! 62 */ 63 #define MAX_TCP_WINDOW 32767U 64 65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 66 #define TCP_MIN_MSS 88U 67 68 /* The initial MTU to use for probing */ 69 #define TCP_BASE_MSS 1024 70 71 /* probing interval, default to 10 minutes as per RFC4821 */ 72 #define TCP_PROBE_INTERVAL 600 73 74 /* Specify interval when tcp mtu probing will stop */ 75 #define TCP_PROBE_THRESHOLD 8 76 77 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 78 #define TCP_FASTRETRANS_THRESH 3 79 80 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 81 #define TCP_MAX_QUICKACKS 16U 82 83 /* Maximal number of window scale according to RFC1323 */ 84 #define TCP_MAX_WSCALE 14U 85 86 /* urg_data states */ 87 #define TCP_URG_VALID 0x0100 88 #define TCP_URG_NOTYET 0x0200 89 #define TCP_URG_READ 0x0400 90 91 #define TCP_RETR1 3 /* 92 * This is how many retries it does before it 93 * tries to figure out if the gateway is 94 * down. Minimal RFC value is 3; it corresponds 95 * to ~3sec-8min depending on RTO. 96 */ 97 98 #define TCP_RETR2 15 /* 99 * This should take at least 100 * 90 minutes to time out. 101 * RFC1122 says that the limit is 100 sec. 102 * 15 is ~13-30min depending on RTO. 103 */ 104 105 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 106 * when active opening a connection. 107 * RFC1122 says the minimum retry MUST 108 * be at least 180secs. Nevertheless 109 * this value is corresponding to 110 * 63secs of retransmission with the 111 * current initial RTO. 112 */ 113 114 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 115 * when passive opening a connection. 116 * This is corresponding to 31secs of 117 * retransmission with the current 118 * initial RTO. 119 */ 120 121 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 122 * state, about 60 seconds */ 123 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 124 /* BSD style FIN_WAIT2 deadlock breaker. 125 * It used to be 3min, new value is 60sec, 126 * to combine FIN-WAIT-2 timeout with 127 * TIME-WAIT timer. 128 */ 129 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 130 131 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 132 #if HZ >= 100 133 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 134 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 135 #else 136 #define TCP_DELACK_MIN 4U 137 #define TCP_ATO_MIN 4U 138 #endif 139 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 140 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 141 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 142 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 143 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 144 * used as a fallback RTO for the 145 * initial data transmission if no 146 * valid RTT sample has been acquired, 147 * most likely due to retrans in 3WHS. 148 */ 149 150 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 151 * for local resources. 152 */ 153 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 154 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 155 #define TCP_KEEPALIVE_INTVL (75*HZ) 156 157 #define MAX_TCP_KEEPIDLE 32767 158 #define MAX_TCP_KEEPINTVL 32767 159 #define MAX_TCP_KEEPCNT 127 160 #define MAX_TCP_SYNCNT 127 161 162 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 163 164 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 165 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 166 * after this time. It should be equal 167 * (or greater than) TCP_TIMEWAIT_LEN 168 * to provide reliability equal to one 169 * provided by timewait state. 170 */ 171 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 172 * timestamps. It must be less than 173 * minimal timewait lifetime. 174 */ 175 /* 176 * TCP option 177 */ 178 179 #define TCPOPT_NOP 1 /* Padding */ 180 #define TCPOPT_EOL 0 /* End of options */ 181 #define TCPOPT_MSS 2 /* Segment size negotiating */ 182 #define TCPOPT_WINDOW 3 /* Window scaling */ 183 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 184 #define TCPOPT_SACK 5 /* SACK Block */ 185 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 186 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 187 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 189 #define TCPOPT_EXP 254 /* Experimental */ 190 /* Magic number to be after the option value for sharing TCP 191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 192 */ 193 #define TCPOPT_FASTOPEN_MAGIC 0xF989 194 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 195 196 /* 197 * TCP option lengths 198 */ 199 200 #define TCPOLEN_MSS 4 201 #define TCPOLEN_WINDOW 3 202 #define TCPOLEN_SACK_PERM 2 203 #define TCPOLEN_TIMESTAMP 10 204 #define TCPOLEN_MD5SIG 18 205 #define TCPOLEN_FASTOPEN_BASE 2 206 #define TCPOLEN_EXP_FASTOPEN_BASE 4 207 #define TCPOLEN_EXP_SMC_BASE 6 208 209 /* But this is what stacks really send out. */ 210 #define TCPOLEN_TSTAMP_ALIGNED 12 211 #define TCPOLEN_WSCALE_ALIGNED 4 212 #define TCPOLEN_SACKPERM_ALIGNED 4 213 #define TCPOLEN_SACK_BASE 2 214 #define TCPOLEN_SACK_BASE_ALIGNED 4 215 #define TCPOLEN_SACK_PERBLOCK 8 216 #define TCPOLEN_MD5SIG_ALIGNED 20 217 #define TCPOLEN_MSS_ALIGNED 4 218 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 219 220 /* Flags in tp->nonagle */ 221 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 222 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 223 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 224 225 /* TCP thin-stream limits */ 226 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 227 228 /* TCP initial congestion window as per rfc6928 */ 229 #define TCP_INIT_CWND 10 230 231 /* Bit Flags for sysctl_tcp_fastopen */ 232 #define TFO_CLIENT_ENABLE 1 233 #define TFO_SERVER_ENABLE 2 234 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 235 236 /* Accept SYN data w/o any cookie option */ 237 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 238 239 /* Force enable TFO on all listeners, i.e., not requiring the 240 * TCP_FASTOPEN socket option. 241 */ 242 #define TFO_SERVER_WO_SOCKOPT1 0x400 243 244 245 /* sysctl variables for tcp */ 246 extern int sysctl_tcp_max_orphans; 247 extern long sysctl_tcp_mem[3]; 248 249 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 250 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 251 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 252 253 extern atomic_long_t tcp_memory_allocated; 254 extern struct percpu_counter tcp_sockets_allocated; 255 extern unsigned long tcp_memory_pressure; 256 257 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 258 static inline bool tcp_under_memory_pressure(const struct sock *sk) 259 { 260 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 261 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 262 return true; 263 264 return READ_ONCE(tcp_memory_pressure); 265 } 266 /* 267 * The next routines deal with comparing 32 bit unsigned ints 268 * and worry about wraparound (automatic with unsigned arithmetic). 269 */ 270 271 static inline bool before(__u32 seq1, __u32 seq2) 272 { 273 return (__s32)(seq1-seq2) < 0; 274 } 275 #define after(seq2, seq1) before(seq1, seq2) 276 277 /* is s2<=s1<=s3 ? */ 278 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 279 { 280 return seq3 - seq2 >= seq1 - seq2; 281 } 282 283 static inline bool tcp_out_of_memory(struct sock *sk) 284 { 285 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 286 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 287 return true; 288 return false; 289 } 290 291 void sk_forced_mem_schedule(struct sock *sk, int size); 292 293 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 294 { 295 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 296 int orphans = percpu_counter_read_positive(ocp); 297 298 if (orphans << shift > sysctl_tcp_max_orphans) { 299 orphans = percpu_counter_sum_positive(ocp); 300 if (orphans << shift > sysctl_tcp_max_orphans) 301 return true; 302 } 303 return false; 304 } 305 306 bool tcp_check_oom(struct sock *sk, int shift); 307 308 309 extern struct proto tcp_prot; 310 311 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 312 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 315 316 void tcp_tasklet_init(void); 317 318 int tcp_v4_err(struct sk_buff *skb, u32); 319 320 void tcp_shutdown(struct sock *sk, int how); 321 322 int tcp_v4_early_demux(struct sk_buff *skb); 323 int tcp_v4_rcv(struct sk_buff *skb); 324 325 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 326 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 327 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 328 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 329 int flags); 330 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 331 size_t size, int flags); 332 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 333 size_t size, int flags); 334 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 335 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 336 int size_goal); 337 void tcp_release_cb(struct sock *sk); 338 void tcp_wfree(struct sk_buff *skb); 339 void tcp_write_timer_handler(struct sock *sk); 340 void tcp_delack_timer_handler(struct sock *sk); 341 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 342 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 344 void tcp_rcv_space_adjust(struct sock *sk); 345 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 346 void tcp_twsk_destructor(struct sock *sk); 347 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 348 struct pipe_inode_info *pipe, size_t len, 349 unsigned int flags); 350 351 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 352 static inline void tcp_dec_quickack_mode(struct sock *sk, 353 const unsigned int pkts) 354 { 355 struct inet_connection_sock *icsk = inet_csk(sk); 356 357 if (icsk->icsk_ack.quick) { 358 if (pkts >= icsk->icsk_ack.quick) { 359 icsk->icsk_ack.quick = 0; 360 /* Leaving quickack mode we deflate ATO. */ 361 icsk->icsk_ack.ato = TCP_ATO_MIN; 362 } else 363 icsk->icsk_ack.quick -= pkts; 364 } 365 } 366 367 #define TCP_ECN_OK 1 368 #define TCP_ECN_QUEUE_CWR 2 369 #define TCP_ECN_DEMAND_CWR 4 370 #define TCP_ECN_SEEN 8 371 372 enum tcp_tw_status { 373 TCP_TW_SUCCESS = 0, 374 TCP_TW_RST = 1, 375 TCP_TW_ACK = 2, 376 TCP_TW_SYN = 3 377 }; 378 379 380 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 381 struct sk_buff *skb, 382 const struct tcphdr *th); 383 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 384 struct request_sock *req, bool fastopen, 385 bool *lost_race); 386 int tcp_child_process(struct sock *parent, struct sock *child, 387 struct sk_buff *skb); 388 void tcp_enter_loss(struct sock *sk); 389 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 390 void tcp_clear_retrans(struct tcp_sock *tp); 391 void tcp_update_metrics(struct sock *sk); 392 void tcp_init_metrics(struct sock *sk); 393 void tcp_metrics_init(void); 394 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 395 void tcp_close(struct sock *sk, long timeout); 396 void tcp_init_sock(struct sock *sk); 397 void tcp_init_transfer(struct sock *sk, int bpf_op); 398 __poll_t tcp_poll(struct file *file, struct socket *sock, 399 struct poll_table_struct *wait); 400 int tcp_getsockopt(struct sock *sk, int level, int optname, 401 char __user *optval, int __user *optlen); 402 int tcp_setsockopt(struct sock *sk, int level, int optname, 403 char __user *optval, unsigned int optlen); 404 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 405 char __user *optval, int __user *optlen); 406 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 407 char __user *optval, unsigned int optlen); 408 void tcp_set_keepalive(struct sock *sk, int val); 409 void tcp_syn_ack_timeout(const struct request_sock *req); 410 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 411 int flags, int *addr_len); 412 int tcp_set_rcvlowat(struct sock *sk, int val); 413 void tcp_data_ready(struct sock *sk); 414 #ifdef CONFIG_MMU 415 int tcp_mmap(struct file *file, struct socket *sock, 416 struct vm_area_struct *vma); 417 #endif 418 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 419 struct tcp_options_received *opt_rx, 420 int estab, struct tcp_fastopen_cookie *foc); 421 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 422 423 /* 424 * BPF SKB-less helpers 425 */ 426 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 427 struct tcphdr *th, u32 *cookie); 428 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 429 struct tcphdr *th, u32 *cookie); 430 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 431 const struct tcp_request_sock_ops *af_ops, 432 struct sock *sk, struct tcphdr *th); 433 /* 434 * TCP v4 functions exported for the inet6 API 435 */ 436 437 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 438 void tcp_v4_mtu_reduced(struct sock *sk); 439 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 440 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 441 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 442 struct sock *tcp_create_openreq_child(const struct sock *sk, 443 struct request_sock *req, 444 struct sk_buff *skb); 445 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 446 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 447 struct request_sock *req, 448 struct dst_entry *dst, 449 struct request_sock *req_unhash, 450 bool *own_req); 451 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 452 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 453 int tcp_connect(struct sock *sk); 454 enum tcp_synack_type { 455 TCP_SYNACK_NORMAL, 456 TCP_SYNACK_FASTOPEN, 457 TCP_SYNACK_COOKIE, 458 }; 459 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 460 struct request_sock *req, 461 struct tcp_fastopen_cookie *foc, 462 enum tcp_synack_type synack_type); 463 int tcp_disconnect(struct sock *sk, int flags); 464 465 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 466 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 467 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 468 469 /* From syncookies.c */ 470 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 471 struct request_sock *req, 472 struct dst_entry *dst, u32 tsoff); 473 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 474 u32 cookie); 475 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 476 #ifdef CONFIG_SYN_COOKIES 477 478 /* Syncookies use a monotonic timer which increments every 60 seconds. 479 * This counter is used both as a hash input and partially encoded into 480 * the cookie value. A cookie is only validated further if the delta 481 * between the current counter value and the encoded one is less than this, 482 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 483 * the counter advances immediately after a cookie is generated). 484 */ 485 #define MAX_SYNCOOKIE_AGE 2 486 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 487 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 488 489 /* syncookies: remember time of last synqueue overflow 490 * But do not dirty this field too often (once per second is enough) 491 * It is racy as we do not hold a lock, but race is very minor. 492 */ 493 static inline void tcp_synq_overflow(const struct sock *sk) 494 { 495 unsigned int last_overflow; 496 unsigned int now = jiffies; 497 498 if (sk->sk_reuseport) { 499 struct sock_reuseport *reuse; 500 501 reuse = rcu_dereference(sk->sk_reuseport_cb); 502 if (likely(reuse)) { 503 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 504 if (!time_between32(now, last_overflow, 505 last_overflow + HZ)) 506 WRITE_ONCE(reuse->synq_overflow_ts, now); 507 return; 508 } 509 } 510 511 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 512 if (!time_between32(now, last_overflow, last_overflow + HZ)) 513 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 514 } 515 516 /* syncookies: no recent synqueue overflow on this listening socket? */ 517 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 518 { 519 unsigned int last_overflow; 520 unsigned int now = jiffies; 521 522 if (sk->sk_reuseport) { 523 struct sock_reuseport *reuse; 524 525 reuse = rcu_dereference(sk->sk_reuseport_cb); 526 if (likely(reuse)) { 527 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 528 return !time_between32(now, last_overflow - HZ, 529 last_overflow + 530 TCP_SYNCOOKIE_VALID); 531 } 532 } 533 534 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 535 536 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 537 * then we're under synflood. However, we have to use 538 * 'last_overflow - HZ' as lower bound. That's because a concurrent 539 * tcp_synq_overflow() could update .ts_recent_stamp after we read 540 * jiffies but before we store .ts_recent_stamp into last_overflow, 541 * which could lead to rejecting a valid syncookie. 542 */ 543 return !time_between32(now, last_overflow - HZ, 544 last_overflow + TCP_SYNCOOKIE_VALID); 545 } 546 547 static inline u32 tcp_cookie_time(void) 548 { 549 u64 val = get_jiffies_64(); 550 551 do_div(val, TCP_SYNCOOKIE_PERIOD); 552 return val; 553 } 554 555 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 556 u16 *mssp); 557 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 558 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 559 bool cookie_timestamp_decode(const struct net *net, 560 struct tcp_options_received *opt); 561 bool cookie_ecn_ok(const struct tcp_options_received *opt, 562 const struct net *net, const struct dst_entry *dst); 563 564 /* From net/ipv6/syncookies.c */ 565 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 566 u32 cookie); 567 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 568 569 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 570 const struct tcphdr *th, u16 *mssp); 571 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 572 #endif 573 /* tcp_output.c */ 574 575 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 576 int nonagle); 577 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 578 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 579 void tcp_retransmit_timer(struct sock *sk); 580 void tcp_xmit_retransmit_queue(struct sock *); 581 void tcp_simple_retransmit(struct sock *); 582 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 583 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 584 enum tcp_queue { 585 TCP_FRAG_IN_WRITE_QUEUE, 586 TCP_FRAG_IN_RTX_QUEUE, 587 }; 588 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 589 struct sk_buff *skb, u32 len, 590 unsigned int mss_now, gfp_t gfp); 591 592 void tcp_send_probe0(struct sock *); 593 void tcp_send_partial(struct sock *); 594 int tcp_write_wakeup(struct sock *, int mib); 595 void tcp_send_fin(struct sock *sk); 596 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 597 int tcp_send_synack(struct sock *); 598 void tcp_push_one(struct sock *, unsigned int mss_now); 599 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 600 void tcp_send_ack(struct sock *sk); 601 void tcp_send_delayed_ack(struct sock *sk); 602 void tcp_send_loss_probe(struct sock *sk); 603 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 604 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 605 const struct sk_buff *next_skb); 606 607 /* tcp_input.c */ 608 void tcp_rearm_rto(struct sock *sk); 609 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 610 void tcp_reset(struct sock *sk); 611 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 612 void tcp_fin(struct sock *sk); 613 614 /* tcp_timer.c */ 615 void tcp_init_xmit_timers(struct sock *); 616 static inline void tcp_clear_xmit_timers(struct sock *sk) 617 { 618 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 619 __sock_put(sk); 620 621 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 622 __sock_put(sk); 623 624 inet_csk_clear_xmit_timers(sk); 625 } 626 627 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 628 unsigned int tcp_current_mss(struct sock *sk); 629 630 /* Bound MSS / TSO packet size with the half of the window */ 631 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 632 { 633 int cutoff; 634 635 /* When peer uses tiny windows, there is no use in packetizing 636 * to sub-MSS pieces for the sake of SWS or making sure there 637 * are enough packets in the pipe for fast recovery. 638 * 639 * On the other hand, for extremely large MSS devices, handling 640 * smaller than MSS windows in this way does make sense. 641 */ 642 if (tp->max_window > TCP_MSS_DEFAULT) 643 cutoff = (tp->max_window >> 1); 644 else 645 cutoff = tp->max_window; 646 647 if (cutoff && pktsize > cutoff) 648 return max_t(int, cutoff, 68U - tp->tcp_header_len); 649 else 650 return pktsize; 651 } 652 653 /* tcp.c */ 654 void tcp_get_info(struct sock *, struct tcp_info *); 655 656 /* Read 'sendfile()'-style from a TCP socket */ 657 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 658 sk_read_actor_t recv_actor); 659 660 void tcp_initialize_rcv_mss(struct sock *sk); 661 662 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 663 int tcp_mss_to_mtu(struct sock *sk, int mss); 664 void tcp_mtup_init(struct sock *sk); 665 666 static inline void tcp_bound_rto(const struct sock *sk) 667 { 668 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 669 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 670 } 671 672 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 673 { 674 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 675 } 676 677 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 678 { 679 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 680 ntohl(TCP_FLAG_ACK) | 681 snd_wnd); 682 } 683 684 static inline void tcp_fast_path_on(struct tcp_sock *tp) 685 { 686 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 687 } 688 689 static inline void tcp_fast_path_check(struct sock *sk) 690 { 691 struct tcp_sock *tp = tcp_sk(sk); 692 693 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 694 tp->rcv_wnd && 695 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 696 !tp->urg_data) 697 tcp_fast_path_on(tp); 698 } 699 700 /* Compute the actual rto_min value */ 701 static inline u32 tcp_rto_min(struct sock *sk) 702 { 703 const struct dst_entry *dst = __sk_dst_get(sk); 704 u32 rto_min = TCP_RTO_MIN; 705 706 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 707 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 708 return rto_min; 709 } 710 711 static inline u32 tcp_rto_min_us(struct sock *sk) 712 { 713 return jiffies_to_usecs(tcp_rto_min(sk)); 714 } 715 716 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 717 { 718 return dst_metric_locked(dst, RTAX_CC_ALGO); 719 } 720 721 /* Minimum RTT in usec. ~0 means not available. */ 722 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 723 { 724 return minmax_get(&tp->rtt_min); 725 } 726 727 /* Compute the actual receive window we are currently advertising. 728 * Rcv_nxt can be after the window if our peer push more data 729 * than the offered window. 730 */ 731 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 732 { 733 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 734 735 if (win < 0) 736 win = 0; 737 return (u32) win; 738 } 739 740 /* Choose a new window, without checks for shrinking, and without 741 * scaling applied to the result. The caller does these things 742 * if necessary. This is a "raw" window selection. 743 */ 744 u32 __tcp_select_window(struct sock *sk); 745 746 void tcp_send_window_probe(struct sock *sk); 747 748 /* TCP uses 32bit jiffies to save some space. 749 * Note that this is different from tcp_time_stamp, which 750 * historically has been the same until linux-4.13. 751 */ 752 #define tcp_jiffies32 ((u32)jiffies) 753 754 /* 755 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 756 * It is no longer tied to jiffies, but to 1 ms clock. 757 * Note: double check if you want to use tcp_jiffies32 instead of this. 758 */ 759 #define TCP_TS_HZ 1000 760 761 static inline u64 tcp_clock_ns(void) 762 { 763 return ktime_get_ns(); 764 } 765 766 static inline u64 tcp_clock_us(void) 767 { 768 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 769 } 770 771 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 772 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 773 { 774 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 775 } 776 777 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 778 static inline u32 tcp_ns_to_ts(u64 ns) 779 { 780 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 781 } 782 783 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 784 static inline u32 tcp_time_stamp_raw(void) 785 { 786 return tcp_ns_to_ts(tcp_clock_ns()); 787 } 788 789 void tcp_mstamp_refresh(struct tcp_sock *tp); 790 791 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 792 { 793 return max_t(s64, t1 - t0, 0); 794 } 795 796 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 797 { 798 return tcp_ns_to_ts(skb->skb_mstamp_ns); 799 } 800 801 /* provide the departure time in us unit */ 802 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 803 { 804 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 805 } 806 807 808 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 809 810 #define TCPHDR_FIN 0x01 811 #define TCPHDR_SYN 0x02 812 #define TCPHDR_RST 0x04 813 #define TCPHDR_PSH 0x08 814 #define TCPHDR_ACK 0x10 815 #define TCPHDR_URG 0x20 816 #define TCPHDR_ECE 0x40 817 #define TCPHDR_CWR 0x80 818 819 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 820 821 /* This is what the send packet queuing engine uses to pass 822 * TCP per-packet control information to the transmission code. 823 * We also store the host-order sequence numbers in here too. 824 * This is 44 bytes if IPV6 is enabled. 825 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 826 */ 827 struct tcp_skb_cb { 828 __u32 seq; /* Starting sequence number */ 829 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 830 union { 831 /* Note : tcp_tw_isn is used in input path only 832 * (isn chosen by tcp_timewait_state_process()) 833 * 834 * tcp_gso_segs/size are used in write queue only, 835 * cf tcp_skb_pcount()/tcp_skb_mss() 836 */ 837 __u32 tcp_tw_isn; 838 struct { 839 u16 tcp_gso_segs; 840 u16 tcp_gso_size; 841 }; 842 }; 843 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 844 845 __u8 sacked; /* State flags for SACK. */ 846 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 847 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 848 #define TCPCB_LOST 0x04 /* SKB is lost */ 849 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 850 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 851 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 852 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 853 TCPCB_REPAIRED) 854 855 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 856 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 857 eor:1, /* Is skb MSG_EOR marked? */ 858 has_rxtstamp:1, /* SKB has a RX timestamp */ 859 unused:5; 860 __u32 ack_seq; /* Sequence number ACK'd */ 861 union { 862 struct { 863 /* There is space for up to 24 bytes */ 864 __u32 in_flight:30,/* Bytes in flight at transmit */ 865 is_app_limited:1, /* cwnd not fully used? */ 866 unused:1; 867 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 868 __u32 delivered; 869 /* start of send pipeline phase */ 870 u64 first_tx_mstamp; 871 /* when we reached the "delivered" count */ 872 u64 delivered_mstamp; 873 } tx; /* only used for outgoing skbs */ 874 union { 875 struct inet_skb_parm h4; 876 #if IS_ENABLED(CONFIG_IPV6) 877 struct inet6_skb_parm h6; 878 #endif 879 } header; /* For incoming skbs */ 880 struct { 881 __u32 flags; 882 struct sock *sk_redir; 883 void *data_end; 884 } bpf; 885 }; 886 }; 887 888 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 889 890 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 891 { 892 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 893 } 894 895 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 896 { 897 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 898 } 899 900 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 901 { 902 return TCP_SKB_CB(skb)->bpf.sk_redir; 903 } 904 905 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 906 { 907 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 908 } 909 910 extern const struct inet_connection_sock_af_ops ipv4_specific; 911 912 #if IS_ENABLED(CONFIG_IPV6) 913 /* This is the variant of inet6_iif() that must be used by TCP, 914 * as TCP moves IP6CB into a different location in skb->cb[] 915 */ 916 static inline int tcp_v6_iif(const struct sk_buff *skb) 917 { 918 return TCP_SKB_CB(skb)->header.h6.iif; 919 } 920 921 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 922 { 923 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 924 925 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 926 } 927 928 /* TCP_SKB_CB reference means this can not be used from early demux */ 929 static inline int tcp_v6_sdif(const struct sk_buff *skb) 930 { 931 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 932 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 933 return TCP_SKB_CB(skb)->header.h6.iif; 934 #endif 935 return 0; 936 } 937 938 extern const struct inet_connection_sock_af_ops ipv6_specific; 939 940 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 941 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 942 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb)); 943 944 #endif 945 946 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 947 { 948 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 949 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 950 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 951 return true; 952 #endif 953 return false; 954 } 955 956 /* TCP_SKB_CB reference means this can not be used from early demux */ 957 static inline int tcp_v4_sdif(struct sk_buff *skb) 958 { 959 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 960 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 961 return TCP_SKB_CB(skb)->header.h4.iif; 962 #endif 963 return 0; 964 } 965 966 /* Due to TSO, an SKB can be composed of multiple actual 967 * packets. To keep these tracked properly, we use this. 968 */ 969 static inline int tcp_skb_pcount(const struct sk_buff *skb) 970 { 971 return TCP_SKB_CB(skb)->tcp_gso_segs; 972 } 973 974 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 975 { 976 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 977 } 978 979 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 980 { 981 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 982 } 983 984 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 985 static inline int tcp_skb_mss(const struct sk_buff *skb) 986 { 987 return TCP_SKB_CB(skb)->tcp_gso_size; 988 } 989 990 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 991 { 992 return likely(!TCP_SKB_CB(skb)->eor); 993 } 994 995 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 996 const struct sk_buff *from) 997 { 998 return likely(tcp_skb_can_collapse_to(to) && 999 mptcp_skb_can_collapse(to, from)); 1000 } 1001 1002 /* Events passed to congestion control interface */ 1003 enum tcp_ca_event { 1004 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1005 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1006 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1007 CA_EVENT_LOSS, /* loss timeout */ 1008 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1009 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1010 }; 1011 1012 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1013 enum tcp_ca_ack_event_flags { 1014 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1015 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1016 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1017 }; 1018 1019 /* 1020 * Interface for adding new TCP congestion control handlers 1021 */ 1022 #define TCP_CA_NAME_MAX 16 1023 #define TCP_CA_MAX 128 1024 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1025 1026 #define TCP_CA_UNSPEC 0 1027 1028 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1029 #define TCP_CONG_NON_RESTRICTED 0x1 1030 /* Requires ECN/ECT set on all packets */ 1031 #define TCP_CONG_NEEDS_ECN 0x2 1032 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1033 1034 union tcp_cc_info; 1035 1036 struct ack_sample { 1037 u32 pkts_acked; 1038 s32 rtt_us; 1039 u32 in_flight; 1040 }; 1041 1042 /* A rate sample measures the number of (original/retransmitted) data 1043 * packets delivered "delivered" over an interval of time "interval_us". 1044 * The tcp_rate.c code fills in the rate sample, and congestion 1045 * control modules that define a cong_control function to run at the end 1046 * of ACK processing can optionally chose to consult this sample when 1047 * setting cwnd and pacing rate. 1048 * A sample is invalid if "delivered" or "interval_us" is negative. 1049 */ 1050 struct rate_sample { 1051 u64 prior_mstamp; /* starting timestamp for interval */ 1052 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1053 s32 delivered; /* number of packets delivered over interval */ 1054 long interval_us; /* time for tp->delivered to incr "delivered" */ 1055 u32 snd_interval_us; /* snd interval for delivered packets */ 1056 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1057 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1058 int losses; /* number of packets marked lost upon ACK */ 1059 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1060 u32 prior_in_flight; /* in flight before this ACK */ 1061 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1062 bool is_retrans; /* is sample from retransmission? */ 1063 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1064 }; 1065 1066 struct tcp_congestion_ops { 1067 struct list_head list; 1068 u32 key; 1069 u32 flags; 1070 1071 /* initialize private data (optional) */ 1072 void (*init)(struct sock *sk); 1073 /* cleanup private data (optional) */ 1074 void (*release)(struct sock *sk); 1075 1076 /* return slow start threshold (required) */ 1077 u32 (*ssthresh)(struct sock *sk); 1078 /* do new cwnd calculation (required) */ 1079 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1080 /* call before changing ca_state (optional) */ 1081 void (*set_state)(struct sock *sk, u8 new_state); 1082 /* call when cwnd event occurs (optional) */ 1083 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1084 /* call when ack arrives (optional) */ 1085 void (*in_ack_event)(struct sock *sk, u32 flags); 1086 /* new value of cwnd after loss (required) */ 1087 u32 (*undo_cwnd)(struct sock *sk); 1088 /* hook for packet ack accounting (optional) */ 1089 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1090 /* override sysctl_tcp_min_tso_segs */ 1091 u32 (*min_tso_segs)(struct sock *sk); 1092 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1093 u32 (*sndbuf_expand)(struct sock *sk); 1094 /* call when packets are delivered to update cwnd and pacing rate, 1095 * after all the ca_state processing. (optional) 1096 */ 1097 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1098 /* get info for inet_diag (optional) */ 1099 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1100 union tcp_cc_info *info); 1101 1102 char name[TCP_CA_NAME_MAX]; 1103 struct module *owner; 1104 }; 1105 1106 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1107 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1108 1109 void tcp_assign_congestion_control(struct sock *sk); 1110 void tcp_init_congestion_control(struct sock *sk); 1111 void tcp_cleanup_congestion_control(struct sock *sk); 1112 int tcp_set_default_congestion_control(struct net *net, const char *name); 1113 void tcp_get_default_congestion_control(struct net *net, char *name); 1114 void tcp_get_available_congestion_control(char *buf, size_t len); 1115 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1116 int tcp_set_allowed_congestion_control(char *allowed); 1117 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1118 bool reinit, bool cap_net_admin); 1119 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1120 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1121 1122 u32 tcp_reno_ssthresh(struct sock *sk); 1123 u32 tcp_reno_undo_cwnd(struct sock *sk); 1124 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1125 extern struct tcp_congestion_ops tcp_reno; 1126 1127 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1128 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1129 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1130 #ifdef CONFIG_INET 1131 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1132 #else 1133 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1134 { 1135 return NULL; 1136 } 1137 #endif 1138 1139 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1140 { 1141 const struct inet_connection_sock *icsk = inet_csk(sk); 1142 1143 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1144 } 1145 1146 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1147 { 1148 struct inet_connection_sock *icsk = inet_csk(sk); 1149 1150 if (icsk->icsk_ca_ops->set_state) 1151 icsk->icsk_ca_ops->set_state(sk, ca_state); 1152 icsk->icsk_ca_state = ca_state; 1153 } 1154 1155 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1156 { 1157 const struct inet_connection_sock *icsk = inet_csk(sk); 1158 1159 if (icsk->icsk_ca_ops->cwnd_event) 1160 icsk->icsk_ca_ops->cwnd_event(sk, event); 1161 } 1162 1163 /* From tcp_rate.c */ 1164 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1165 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1166 struct rate_sample *rs); 1167 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1168 bool is_sack_reneg, struct rate_sample *rs); 1169 void tcp_rate_check_app_limited(struct sock *sk); 1170 1171 /* These functions determine how the current flow behaves in respect of SACK 1172 * handling. SACK is negotiated with the peer, and therefore it can vary 1173 * between different flows. 1174 * 1175 * tcp_is_sack - SACK enabled 1176 * tcp_is_reno - No SACK 1177 */ 1178 static inline int tcp_is_sack(const struct tcp_sock *tp) 1179 { 1180 return likely(tp->rx_opt.sack_ok); 1181 } 1182 1183 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1184 { 1185 return !tcp_is_sack(tp); 1186 } 1187 1188 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1189 { 1190 return tp->sacked_out + tp->lost_out; 1191 } 1192 1193 /* This determines how many packets are "in the network" to the best 1194 * of our knowledge. In many cases it is conservative, but where 1195 * detailed information is available from the receiver (via SACK 1196 * blocks etc.) we can make more aggressive calculations. 1197 * 1198 * Use this for decisions involving congestion control, use just 1199 * tp->packets_out to determine if the send queue is empty or not. 1200 * 1201 * Read this equation as: 1202 * 1203 * "Packets sent once on transmission queue" MINUS 1204 * "Packets left network, but not honestly ACKed yet" PLUS 1205 * "Packets fast retransmitted" 1206 */ 1207 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1208 { 1209 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1210 } 1211 1212 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1213 1214 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1215 { 1216 return tp->snd_cwnd < tp->snd_ssthresh; 1217 } 1218 1219 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1220 { 1221 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1222 } 1223 1224 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1225 { 1226 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1227 (1 << inet_csk(sk)->icsk_ca_state); 1228 } 1229 1230 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1231 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1232 * ssthresh. 1233 */ 1234 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1235 { 1236 const struct tcp_sock *tp = tcp_sk(sk); 1237 1238 if (tcp_in_cwnd_reduction(sk)) 1239 return tp->snd_ssthresh; 1240 else 1241 return max(tp->snd_ssthresh, 1242 ((tp->snd_cwnd >> 1) + 1243 (tp->snd_cwnd >> 2))); 1244 } 1245 1246 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1247 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1248 1249 void tcp_enter_cwr(struct sock *sk); 1250 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1251 1252 /* The maximum number of MSS of available cwnd for which TSO defers 1253 * sending if not using sysctl_tcp_tso_win_divisor. 1254 */ 1255 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1256 { 1257 return 3; 1258 } 1259 1260 /* Returns end sequence number of the receiver's advertised window */ 1261 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1262 { 1263 return tp->snd_una + tp->snd_wnd; 1264 } 1265 1266 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1267 * flexible approach. The RFC suggests cwnd should not be raised unless 1268 * it was fully used previously. And that's exactly what we do in 1269 * congestion avoidance mode. But in slow start we allow cwnd to grow 1270 * as long as the application has used half the cwnd. 1271 * Example : 1272 * cwnd is 10 (IW10), but application sends 9 frames. 1273 * We allow cwnd to reach 18 when all frames are ACKed. 1274 * This check is safe because it's as aggressive as slow start which already 1275 * risks 100% overshoot. The advantage is that we discourage application to 1276 * either send more filler packets or data to artificially blow up the cwnd 1277 * usage, and allow application-limited process to probe bw more aggressively. 1278 */ 1279 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1280 { 1281 const struct tcp_sock *tp = tcp_sk(sk); 1282 1283 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1284 if (tcp_in_slow_start(tp)) 1285 return tp->snd_cwnd < 2 * tp->max_packets_out; 1286 1287 return tp->is_cwnd_limited; 1288 } 1289 1290 /* BBR congestion control needs pacing. 1291 * Same remark for SO_MAX_PACING_RATE. 1292 * sch_fq packet scheduler is efficiently handling pacing, 1293 * but is not always installed/used. 1294 * Return true if TCP stack should pace packets itself. 1295 */ 1296 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1297 { 1298 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1299 } 1300 1301 /* Estimates in how many jiffies next packet for this flow can be sent. 1302 * Scheduling a retransmit timer too early would be silly. 1303 */ 1304 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1305 { 1306 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1307 1308 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1309 } 1310 1311 static inline void tcp_reset_xmit_timer(struct sock *sk, 1312 const int what, 1313 unsigned long when, 1314 const unsigned long max_when) 1315 { 1316 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1317 max_when); 1318 } 1319 1320 /* Something is really bad, we could not queue an additional packet, 1321 * because qdisc is full or receiver sent a 0 window, or we are paced. 1322 * We do not want to add fuel to the fire, or abort too early, 1323 * so make sure the timer we arm now is at least 200ms in the future, 1324 * regardless of current icsk_rto value (as it could be ~2ms) 1325 */ 1326 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1327 { 1328 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1329 } 1330 1331 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1332 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1333 unsigned long max_when) 1334 { 1335 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1336 1337 return (unsigned long)min_t(u64, when, max_when); 1338 } 1339 1340 static inline void tcp_check_probe_timer(struct sock *sk) 1341 { 1342 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1343 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1344 tcp_probe0_base(sk), TCP_RTO_MAX); 1345 } 1346 1347 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1348 { 1349 tp->snd_wl1 = seq; 1350 } 1351 1352 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1353 { 1354 tp->snd_wl1 = seq; 1355 } 1356 1357 /* 1358 * Calculate(/check) TCP checksum 1359 */ 1360 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1361 __be32 daddr, __wsum base) 1362 { 1363 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1364 } 1365 1366 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1367 { 1368 return !skb_csum_unnecessary(skb) && 1369 __skb_checksum_complete(skb); 1370 } 1371 1372 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1373 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1374 void tcp_set_state(struct sock *sk, int state); 1375 void tcp_done(struct sock *sk); 1376 int tcp_abort(struct sock *sk, int err); 1377 1378 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1379 { 1380 rx_opt->dsack = 0; 1381 rx_opt->num_sacks = 0; 1382 } 1383 1384 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1385 1386 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1387 { 1388 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1389 struct tcp_sock *tp = tcp_sk(sk); 1390 s32 delta; 1391 1392 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1393 ca_ops->cong_control) 1394 return; 1395 delta = tcp_jiffies32 - tp->lsndtime; 1396 if (delta > inet_csk(sk)->icsk_rto) 1397 tcp_cwnd_restart(sk, delta); 1398 } 1399 1400 /* Determine a window scaling and initial window to offer. */ 1401 void tcp_select_initial_window(const struct sock *sk, int __space, 1402 __u32 mss, __u32 *rcv_wnd, 1403 __u32 *window_clamp, int wscale_ok, 1404 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1405 1406 static inline int tcp_win_from_space(const struct sock *sk, int space) 1407 { 1408 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1409 1410 return tcp_adv_win_scale <= 0 ? 1411 (space>>(-tcp_adv_win_scale)) : 1412 space - (space>>tcp_adv_win_scale); 1413 } 1414 1415 /* Note: caller must be prepared to deal with negative returns */ 1416 static inline int tcp_space(const struct sock *sk) 1417 { 1418 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1419 READ_ONCE(sk->sk_backlog.len) - 1420 atomic_read(&sk->sk_rmem_alloc)); 1421 } 1422 1423 static inline int tcp_full_space(const struct sock *sk) 1424 { 1425 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1426 } 1427 1428 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1429 * If 87.5 % (7/8) of the space has been consumed, we want to override 1430 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1431 * len/truesize ratio. 1432 */ 1433 static inline bool tcp_rmem_pressure(const struct sock *sk) 1434 { 1435 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1436 int threshold = rcvbuf - (rcvbuf >> 3); 1437 1438 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1439 } 1440 1441 extern void tcp_openreq_init_rwin(struct request_sock *req, 1442 const struct sock *sk_listener, 1443 const struct dst_entry *dst); 1444 1445 void tcp_enter_memory_pressure(struct sock *sk); 1446 void tcp_leave_memory_pressure(struct sock *sk); 1447 1448 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1449 { 1450 struct net *net = sock_net((struct sock *)tp); 1451 1452 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1453 } 1454 1455 static inline int keepalive_time_when(const struct tcp_sock *tp) 1456 { 1457 struct net *net = sock_net((struct sock *)tp); 1458 1459 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1460 } 1461 1462 static inline int keepalive_probes(const struct tcp_sock *tp) 1463 { 1464 struct net *net = sock_net((struct sock *)tp); 1465 1466 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1467 } 1468 1469 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1470 { 1471 const struct inet_connection_sock *icsk = &tp->inet_conn; 1472 1473 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1474 tcp_jiffies32 - tp->rcv_tstamp); 1475 } 1476 1477 static inline int tcp_fin_time(const struct sock *sk) 1478 { 1479 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1480 const int rto = inet_csk(sk)->icsk_rto; 1481 1482 if (fin_timeout < (rto << 2) - (rto >> 1)) 1483 fin_timeout = (rto << 2) - (rto >> 1); 1484 1485 return fin_timeout; 1486 } 1487 1488 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1489 int paws_win) 1490 { 1491 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1492 return true; 1493 if (unlikely(!time_before32(ktime_get_seconds(), 1494 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1495 return true; 1496 /* 1497 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1498 * then following tcp messages have valid values. Ignore 0 value, 1499 * or else 'negative' tsval might forbid us to accept their packets. 1500 */ 1501 if (!rx_opt->ts_recent) 1502 return true; 1503 return false; 1504 } 1505 1506 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1507 int rst) 1508 { 1509 if (tcp_paws_check(rx_opt, 0)) 1510 return false; 1511 1512 /* RST segments are not recommended to carry timestamp, 1513 and, if they do, it is recommended to ignore PAWS because 1514 "their cleanup function should take precedence over timestamps." 1515 Certainly, it is mistake. It is necessary to understand the reasons 1516 of this constraint to relax it: if peer reboots, clock may go 1517 out-of-sync and half-open connections will not be reset. 1518 Actually, the problem would be not existing if all 1519 the implementations followed draft about maintaining clock 1520 via reboots. Linux-2.2 DOES NOT! 1521 1522 However, we can relax time bounds for RST segments to MSL. 1523 */ 1524 if (rst && !time_before32(ktime_get_seconds(), 1525 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1526 return false; 1527 return true; 1528 } 1529 1530 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1531 int mib_idx, u32 *last_oow_ack_time); 1532 1533 static inline void tcp_mib_init(struct net *net) 1534 { 1535 /* See RFC 2012 */ 1536 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1537 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1538 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1539 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1540 } 1541 1542 /* from STCP */ 1543 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1544 { 1545 tp->lost_skb_hint = NULL; 1546 } 1547 1548 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1549 { 1550 tcp_clear_retrans_hints_partial(tp); 1551 tp->retransmit_skb_hint = NULL; 1552 } 1553 1554 union tcp_md5_addr { 1555 struct in_addr a4; 1556 #if IS_ENABLED(CONFIG_IPV6) 1557 struct in6_addr a6; 1558 #endif 1559 }; 1560 1561 /* - key database */ 1562 struct tcp_md5sig_key { 1563 struct hlist_node node; 1564 u8 keylen; 1565 u8 family; /* AF_INET or AF_INET6 */ 1566 u8 prefixlen; 1567 union tcp_md5_addr addr; 1568 int l3index; /* set if key added with L3 scope */ 1569 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1570 struct rcu_head rcu; 1571 }; 1572 1573 /* - sock block */ 1574 struct tcp_md5sig_info { 1575 struct hlist_head head; 1576 struct rcu_head rcu; 1577 }; 1578 1579 /* - pseudo header */ 1580 struct tcp4_pseudohdr { 1581 __be32 saddr; 1582 __be32 daddr; 1583 __u8 pad; 1584 __u8 protocol; 1585 __be16 len; 1586 }; 1587 1588 struct tcp6_pseudohdr { 1589 struct in6_addr saddr; 1590 struct in6_addr daddr; 1591 __be32 len; 1592 __be32 protocol; /* including padding */ 1593 }; 1594 1595 union tcp_md5sum_block { 1596 struct tcp4_pseudohdr ip4; 1597 #if IS_ENABLED(CONFIG_IPV6) 1598 struct tcp6_pseudohdr ip6; 1599 #endif 1600 }; 1601 1602 /* - pool: digest algorithm, hash description and scratch buffer */ 1603 struct tcp_md5sig_pool { 1604 struct ahash_request *md5_req; 1605 void *scratch; 1606 }; 1607 1608 /* - functions */ 1609 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1610 const struct sock *sk, const struct sk_buff *skb); 1611 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1612 int family, u8 prefixlen, int l3index, 1613 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1614 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1615 int family, u8 prefixlen, int l3index); 1616 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1617 const struct sock *addr_sk); 1618 1619 #ifdef CONFIG_TCP_MD5SIG 1620 #include <linux/jump_label.h> 1621 extern struct static_key_false tcp_md5_needed; 1622 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1623 const union tcp_md5_addr *addr, 1624 int family); 1625 static inline struct tcp_md5sig_key * 1626 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1627 const union tcp_md5_addr *addr, int family) 1628 { 1629 if (!static_branch_unlikely(&tcp_md5_needed)) 1630 return NULL; 1631 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1632 } 1633 1634 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1635 #else 1636 static inline struct tcp_md5sig_key * 1637 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1638 const union tcp_md5_addr *addr, int family) 1639 { 1640 return NULL; 1641 } 1642 #define tcp_twsk_md5_key(twsk) NULL 1643 #endif 1644 1645 bool tcp_alloc_md5sig_pool(void); 1646 1647 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1648 static inline void tcp_put_md5sig_pool(void) 1649 { 1650 local_bh_enable(); 1651 } 1652 1653 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1654 unsigned int header_len); 1655 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1656 const struct tcp_md5sig_key *key); 1657 1658 /* From tcp_fastopen.c */ 1659 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1660 struct tcp_fastopen_cookie *cookie); 1661 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1662 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1663 u16 try_exp); 1664 struct tcp_fastopen_request { 1665 /* Fast Open cookie. Size 0 means a cookie request */ 1666 struct tcp_fastopen_cookie cookie; 1667 struct msghdr *data; /* data in MSG_FASTOPEN */ 1668 size_t size; 1669 int copied; /* queued in tcp_connect() */ 1670 struct ubuf_info *uarg; 1671 }; 1672 void tcp_free_fastopen_req(struct tcp_sock *tp); 1673 void tcp_fastopen_destroy_cipher(struct sock *sk); 1674 void tcp_fastopen_ctx_destroy(struct net *net); 1675 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1676 void *primary_key, void *backup_key); 1677 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1678 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1679 struct request_sock *req, 1680 struct tcp_fastopen_cookie *foc, 1681 const struct dst_entry *dst); 1682 void tcp_fastopen_init_key_once(struct net *net); 1683 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1684 struct tcp_fastopen_cookie *cookie); 1685 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1686 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1687 #define TCP_FASTOPEN_KEY_MAX 2 1688 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1689 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1690 1691 /* Fastopen key context */ 1692 struct tcp_fastopen_context { 1693 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1694 int num; 1695 struct rcu_head rcu; 1696 }; 1697 1698 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1699 void tcp_fastopen_active_disable(struct sock *sk); 1700 bool tcp_fastopen_active_should_disable(struct sock *sk); 1701 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1702 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1703 1704 /* Caller needs to wrap with rcu_read_(un)lock() */ 1705 static inline 1706 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1707 { 1708 struct tcp_fastopen_context *ctx; 1709 1710 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1711 if (!ctx) 1712 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1713 return ctx; 1714 } 1715 1716 static inline 1717 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1718 const struct tcp_fastopen_cookie *orig) 1719 { 1720 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1721 orig->len == foc->len && 1722 !memcmp(orig->val, foc->val, foc->len)) 1723 return true; 1724 return false; 1725 } 1726 1727 static inline 1728 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1729 { 1730 return ctx->num; 1731 } 1732 1733 /* Latencies incurred by various limits for a sender. They are 1734 * chronograph-like stats that are mutually exclusive. 1735 */ 1736 enum tcp_chrono { 1737 TCP_CHRONO_UNSPEC, 1738 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1739 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1740 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1741 __TCP_CHRONO_MAX, 1742 }; 1743 1744 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1745 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1746 1747 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1748 * the same memory storage than skb->destructor/_skb_refdst 1749 */ 1750 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1751 { 1752 skb->destructor = NULL; 1753 skb->_skb_refdst = 0UL; 1754 } 1755 1756 #define tcp_skb_tsorted_save(skb) { \ 1757 unsigned long _save = skb->_skb_refdst; \ 1758 skb->_skb_refdst = 0UL; 1759 1760 #define tcp_skb_tsorted_restore(skb) \ 1761 skb->_skb_refdst = _save; \ 1762 } 1763 1764 void tcp_write_queue_purge(struct sock *sk); 1765 1766 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1767 { 1768 return skb_rb_first(&sk->tcp_rtx_queue); 1769 } 1770 1771 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1772 { 1773 return skb_rb_last(&sk->tcp_rtx_queue); 1774 } 1775 1776 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1777 { 1778 return skb_peek(&sk->sk_write_queue); 1779 } 1780 1781 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1782 { 1783 return skb_peek_tail(&sk->sk_write_queue); 1784 } 1785 1786 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1787 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1788 1789 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1790 { 1791 return skb_peek(&sk->sk_write_queue); 1792 } 1793 1794 static inline bool tcp_skb_is_last(const struct sock *sk, 1795 const struct sk_buff *skb) 1796 { 1797 return skb_queue_is_last(&sk->sk_write_queue, skb); 1798 } 1799 1800 /** 1801 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1802 * @sk: socket 1803 * 1804 * Since the write queue can have a temporary empty skb in it, 1805 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1806 */ 1807 static inline bool tcp_write_queue_empty(const struct sock *sk) 1808 { 1809 const struct tcp_sock *tp = tcp_sk(sk); 1810 1811 return tp->write_seq == tp->snd_nxt; 1812 } 1813 1814 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1815 { 1816 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1817 } 1818 1819 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1820 { 1821 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1822 } 1823 1824 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1825 { 1826 __skb_queue_tail(&sk->sk_write_queue, skb); 1827 1828 /* Queue it, remembering where we must start sending. */ 1829 if (sk->sk_write_queue.next == skb) 1830 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1831 } 1832 1833 /* Insert new before skb on the write queue of sk. */ 1834 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1835 struct sk_buff *skb, 1836 struct sock *sk) 1837 { 1838 __skb_queue_before(&sk->sk_write_queue, skb, new); 1839 } 1840 1841 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1842 { 1843 tcp_skb_tsorted_anchor_cleanup(skb); 1844 __skb_unlink(skb, &sk->sk_write_queue); 1845 } 1846 1847 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1848 1849 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1850 { 1851 tcp_skb_tsorted_anchor_cleanup(skb); 1852 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1853 } 1854 1855 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1856 { 1857 list_del(&skb->tcp_tsorted_anchor); 1858 tcp_rtx_queue_unlink(skb, sk); 1859 sk_wmem_free_skb(sk, skb); 1860 } 1861 1862 static inline void tcp_push_pending_frames(struct sock *sk) 1863 { 1864 if (tcp_send_head(sk)) { 1865 struct tcp_sock *tp = tcp_sk(sk); 1866 1867 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1868 } 1869 } 1870 1871 /* Start sequence of the skb just after the highest skb with SACKed 1872 * bit, valid only if sacked_out > 0 or when the caller has ensured 1873 * validity by itself. 1874 */ 1875 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1876 { 1877 if (!tp->sacked_out) 1878 return tp->snd_una; 1879 1880 if (tp->highest_sack == NULL) 1881 return tp->snd_nxt; 1882 1883 return TCP_SKB_CB(tp->highest_sack)->seq; 1884 } 1885 1886 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1887 { 1888 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1889 } 1890 1891 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1892 { 1893 return tcp_sk(sk)->highest_sack; 1894 } 1895 1896 static inline void tcp_highest_sack_reset(struct sock *sk) 1897 { 1898 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1899 } 1900 1901 /* Called when old skb is about to be deleted and replaced by new skb */ 1902 static inline void tcp_highest_sack_replace(struct sock *sk, 1903 struct sk_buff *old, 1904 struct sk_buff *new) 1905 { 1906 if (old == tcp_highest_sack(sk)) 1907 tcp_sk(sk)->highest_sack = new; 1908 } 1909 1910 /* This helper checks if socket has IP_TRANSPARENT set */ 1911 static inline bool inet_sk_transparent(const struct sock *sk) 1912 { 1913 switch (sk->sk_state) { 1914 case TCP_TIME_WAIT: 1915 return inet_twsk(sk)->tw_transparent; 1916 case TCP_NEW_SYN_RECV: 1917 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1918 } 1919 return inet_sk(sk)->transparent; 1920 } 1921 1922 /* Determines whether this is a thin stream (which may suffer from 1923 * increased latency). Used to trigger latency-reducing mechanisms. 1924 */ 1925 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1926 { 1927 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1928 } 1929 1930 /* /proc */ 1931 enum tcp_seq_states { 1932 TCP_SEQ_STATE_LISTENING, 1933 TCP_SEQ_STATE_ESTABLISHED, 1934 }; 1935 1936 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1937 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1938 void tcp_seq_stop(struct seq_file *seq, void *v); 1939 1940 struct tcp_seq_afinfo { 1941 sa_family_t family; 1942 }; 1943 1944 struct tcp_iter_state { 1945 struct seq_net_private p; 1946 enum tcp_seq_states state; 1947 struct sock *syn_wait_sk; 1948 struct tcp_seq_afinfo *bpf_seq_afinfo; 1949 int bucket, offset, sbucket, num; 1950 loff_t last_pos; 1951 }; 1952 1953 extern struct request_sock_ops tcp_request_sock_ops; 1954 extern struct request_sock_ops tcp6_request_sock_ops; 1955 1956 void tcp_v4_destroy_sock(struct sock *sk); 1957 1958 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1959 netdev_features_t features); 1960 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1961 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 1962 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 1963 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 1964 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 1965 int tcp_gro_complete(struct sk_buff *skb); 1966 1967 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1968 1969 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1970 { 1971 struct net *net = sock_net((struct sock *)tp); 1972 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1973 } 1974 1975 /* @wake is one when sk_stream_write_space() calls us. 1976 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1977 * This mimics the strategy used in sock_def_write_space(). 1978 */ 1979 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1980 { 1981 const struct tcp_sock *tp = tcp_sk(sk); 1982 u32 notsent_bytes = READ_ONCE(tp->write_seq) - 1983 READ_ONCE(tp->snd_nxt); 1984 1985 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1986 } 1987 1988 #ifdef CONFIG_PROC_FS 1989 int tcp4_proc_init(void); 1990 void tcp4_proc_exit(void); 1991 #endif 1992 1993 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1994 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1995 const struct tcp_request_sock_ops *af_ops, 1996 struct sock *sk, struct sk_buff *skb); 1997 1998 /* TCP af-specific functions */ 1999 struct tcp_sock_af_ops { 2000 #ifdef CONFIG_TCP_MD5SIG 2001 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2002 const struct sock *addr_sk); 2003 int (*calc_md5_hash)(char *location, 2004 const struct tcp_md5sig_key *md5, 2005 const struct sock *sk, 2006 const struct sk_buff *skb); 2007 int (*md5_parse)(struct sock *sk, 2008 int optname, 2009 char __user *optval, 2010 int optlen); 2011 #endif 2012 }; 2013 2014 struct tcp_request_sock_ops { 2015 u16 mss_clamp; 2016 #ifdef CONFIG_TCP_MD5SIG 2017 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2018 const struct sock *addr_sk); 2019 int (*calc_md5_hash) (char *location, 2020 const struct tcp_md5sig_key *md5, 2021 const struct sock *sk, 2022 const struct sk_buff *skb); 2023 #endif 2024 void (*init_req)(struct request_sock *req, 2025 const struct sock *sk_listener, 2026 struct sk_buff *skb); 2027 #ifdef CONFIG_SYN_COOKIES 2028 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2029 __u16 *mss); 2030 #endif 2031 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 2032 const struct request_sock *req); 2033 u32 (*init_seq)(const struct sk_buff *skb); 2034 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2035 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2036 struct flowi *fl, struct request_sock *req, 2037 struct tcp_fastopen_cookie *foc, 2038 enum tcp_synack_type synack_type); 2039 }; 2040 2041 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2042 #if IS_ENABLED(CONFIG_IPV6) 2043 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2044 #endif 2045 2046 #ifdef CONFIG_SYN_COOKIES 2047 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2048 const struct sock *sk, struct sk_buff *skb, 2049 __u16 *mss) 2050 { 2051 tcp_synq_overflow(sk); 2052 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2053 return ops->cookie_init_seq(skb, mss); 2054 } 2055 #else 2056 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2057 const struct sock *sk, struct sk_buff *skb, 2058 __u16 *mss) 2059 { 2060 return 0; 2061 } 2062 #endif 2063 2064 int tcpv4_offload_init(void); 2065 2066 void tcp_v4_init(void); 2067 void tcp_init(void); 2068 2069 /* tcp_recovery.c */ 2070 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2071 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2072 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2073 u32 reo_wnd); 2074 extern void tcp_rack_mark_lost(struct sock *sk); 2075 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2076 u64 xmit_time); 2077 extern void tcp_rack_reo_timeout(struct sock *sk); 2078 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2079 2080 /* At how many usecs into the future should the RTO fire? */ 2081 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2082 { 2083 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2084 u32 rto = inet_csk(sk)->icsk_rto; 2085 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2086 2087 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2088 } 2089 2090 /* 2091 * Save and compile IPv4 options, return a pointer to it 2092 */ 2093 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2094 struct sk_buff *skb) 2095 { 2096 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2097 struct ip_options_rcu *dopt = NULL; 2098 2099 if (opt->optlen) { 2100 int opt_size = sizeof(*dopt) + opt->optlen; 2101 2102 dopt = kmalloc(opt_size, GFP_ATOMIC); 2103 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2104 kfree(dopt); 2105 dopt = NULL; 2106 } 2107 } 2108 return dopt; 2109 } 2110 2111 /* locally generated TCP pure ACKs have skb->truesize == 2 2112 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2113 * This is much faster than dissecting the packet to find out. 2114 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2115 */ 2116 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2117 { 2118 return skb->truesize == 2; 2119 } 2120 2121 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2122 { 2123 skb->truesize = 2; 2124 } 2125 2126 static inline int tcp_inq(struct sock *sk) 2127 { 2128 struct tcp_sock *tp = tcp_sk(sk); 2129 int answ; 2130 2131 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2132 answ = 0; 2133 } else if (sock_flag(sk, SOCK_URGINLINE) || 2134 !tp->urg_data || 2135 before(tp->urg_seq, tp->copied_seq) || 2136 !before(tp->urg_seq, tp->rcv_nxt)) { 2137 2138 answ = tp->rcv_nxt - tp->copied_seq; 2139 2140 /* Subtract 1, if FIN was received */ 2141 if (answ && sock_flag(sk, SOCK_DONE)) 2142 answ--; 2143 } else { 2144 answ = tp->urg_seq - tp->copied_seq; 2145 } 2146 2147 return answ; 2148 } 2149 2150 int tcp_peek_len(struct socket *sock); 2151 2152 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2153 { 2154 u16 segs_in; 2155 2156 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2157 tp->segs_in += segs_in; 2158 if (skb->len > tcp_hdrlen(skb)) 2159 tp->data_segs_in += segs_in; 2160 } 2161 2162 /* 2163 * TCP listen path runs lockless. 2164 * We forced "struct sock" to be const qualified to make sure 2165 * we don't modify one of its field by mistake. 2166 * Here, we increment sk_drops which is an atomic_t, so we can safely 2167 * make sock writable again. 2168 */ 2169 static inline void tcp_listendrop(const struct sock *sk) 2170 { 2171 atomic_inc(&((struct sock *)sk)->sk_drops); 2172 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2173 } 2174 2175 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2176 2177 /* 2178 * Interface for adding Upper Level Protocols over TCP 2179 */ 2180 2181 #define TCP_ULP_NAME_MAX 16 2182 #define TCP_ULP_MAX 128 2183 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2184 2185 struct tcp_ulp_ops { 2186 struct list_head list; 2187 2188 /* initialize ulp */ 2189 int (*init)(struct sock *sk); 2190 /* update ulp */ 2191 void (*update)(struct sock *sk, struct proto *p, 2192 void (*write_space)(struct sock *sk)); 2193 /* cleanup ulp */ 2194 void (*release)(struct sock *sk); 2195 /* diagnostic */ 2196 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2197 size_t (*get_info_size)(const struct sock *sk); 2198 /* clone ulp */ 2199 void (*clone)(const struct request_sock *req, struct sock *newsk, 2200 const gfp_t priority); 2201 2202 char name[TCP_ULP_NAME_MAX]; 2203 struct module *owner; 2204 }; 2205 int tcp_register_ulp(struct tcp_ulp_ops *type); 2206 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2207 int tcp_set_ulp(struct sock *sk, const char *name); 2208 void tcp_get_available_ulp(char *buf, size_t len); 2209 void tcp_cleanup_ulp(struct sock *sk); 2210 void tcp_update_ulp(struct sock *sk, struct proto *p, 2211 void (*write_space)(struct sock *sk)); 2212 2213 #define MODULE_ALIAS_TCP_ULP(name) \ 2214 __MODULE_INFO(alias, alias_userspace, name); \ 2215 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2216 2217 struct sk_msg; 2218 struct sk_psock; 2219 2220 #ifdef CONFIG_BPF_STREAM_PARSER 2221 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2222 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2223 #else 2224 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2225 { 2226 } 2227 #endif /* CONFIG_BPF_STREAM_PARSER */ 2228 2229 #ifdef CONFIG_NET_SOCK_MSG 2230 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2231 int flags); 2232 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2233 struct msghdr *msg, int len, int flags); 2234 #endif /* CONFIG_NET_SOCK_MSG */ 2235 2236 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2237 * is < 0, then the BPF op failed (for example if the loaded BPF 2238 * program does not support the chosen operation or there is no BPF 2239 * program loaded). 2240 */ 2241 #ifdef CONFIG_BPF 2242 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2243 { 2244 struct bpf_sock_ops_kern sock_ops; 2245 int ret; 2246 2247 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2248 if (sk_fullsock(sk)) { 2249 sock_ops.is_fullsock = 1; 2250 sock_owned_by_me(sk); 2251 } 2252 2253 sock_ops.sk = sk; 2254 sock_ops.op = op; 2255 if (nargs > 0) 2256 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2257 2258 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2259 if (ret == 0) 2260 ret = sock_ops.reply; 2261 else 2262 ret = -1; 2263 return ret; 2264 } 2265 2266 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2267 { 2268 u32 args[2] = {arg1, arg2}; 2269 2270 return tcp_call_bpf(sk, op, 2, args); 2271 } 2272 2273 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2274 u32 arg3) 2275 { 2276 u32 args[3] = {arg1, arg2, arg3}; 2277 2278 return tcp_call_bpf(sk, op, 3, args); 2279 } 2280 2281 #else 2282 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2283 { 2284 return -EPERM; 2285 } 2286 2287 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2288 { 2289 return -EPERM; 2290 } 2291 2292 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2293 u32 arg3) 2294 { 2295 return -EPERM; 2296 } 2297 2298 #endif 2299 2300 static inline u32 tcp_timeout_init(struct sock *sk) 2301 { 2302 int timeout; 2303 2304 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2305 2306 if (timeout <= 0) 2307 timeout = TCP_TIMEOUT_INIT; 2308 return timeout; 2309 } 2310 2311 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2312 { 2313 int rwnd; 2314 2315 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2316 2317 if (rwnd < 0) 2318 rwnd = 0; 2319 return rwnd; 2320 } 2321 2322 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2323 { 2324 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2325 } 2326 2327 static inline void tcp_bpf_rtt(struct sock *sk) 2328 { 2329 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2330 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2331 } 2332 2333 #if IS_ENABLED(CONFIG_SMC) 2334 extern struct static_key_false tcp_have_smc; 2335 #endif 2336 2337 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2338 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2339 void (*cad)(struct sock *sk, u32 ack_seq)); 2340 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2341 void clean_acked_data_flush(void); 2342 #endif 2343 2344 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2345 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2346 const struct tcp_sock *tp) 2347 { 2348 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2349 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2350 } 2351 2352 /* Compute Earliest Departure Time for some control packets 2353 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2354 */ 2355 static inline u64 tcp_transmit_time(const struct sock *sk) 2356 { 2357 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2358 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2359 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2360 2361 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2362 } 2363 return 0; 2364 } 2365 2366 #endif /* _TCP_H */ 2367