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, struct sk_buff *skb); 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, sockptr_t optval, 403 unsigned int optlen); 404 void tcp_set_keepalive(struct sock *sk, int val); 405 void tcp_syn_ack_timeout(const struct request_sock *req); 406 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 407 int flags, int *addr_len); 408 int tcp_set_rcvlowat(struct sock *sk, int val); 409 void tcp_data_ready(struct sock *sk); 410 #ifdef CONFIG_MMU 411 int tcp_mmap(struct file *file, struct socket *sock, 412 struct vm_area_struct *vma); 413 #endif 414 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 415 struct tcp_options_received *opt_rx, 416 int estab, struct tcp_fastopen_cookie *foc); 417 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 418 419 /* 420 * BPF SKB-less helpers 421 */ 422 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 423 struct tcphdr *th, u32 *cookie); 424 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 425 struct tcphdr *th, u32 *cookie); 426 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 427 const struct tcp_request_sock_ops *af_ops, 428 struct sock *sk, struct tcphdr *th); 429 /* 430 * TCP v4 functions exported for the inet6 API 431 */ 432 433 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 434 void tcp_v4_mtu_reduced(struct sock *sk); 435 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 436 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 437 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 438 struct sock *tcp_create_openreq_child(const struct sock *sk, 439 struct request_sock *req, 440 struct sk_buff *skb); 441 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 442 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 443 struct request_sock *req, 444 struct dst_entry *dst, 445 struct request_sock *req_unhash, 446 bool *own_req); 447 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 448 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 449 int tcp_connect(struct sock *sk); 450 enum tcp_synack_type { 451 TCP_SYNACK_NORMAL, 452 TCP_SYNACK_FASTOPEN, 453 TCP_SYNACK_COOKIE, 454 }; 455 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 456 struct request_sock *req, 457 struct tcp_fastopen_cookie *foc, 458 enum tcp_synack_type synack_type, 459 struct sk_buff *syn_skb); 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 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 474 struct sock *sk, struct sk_buff *skb); 475 #ifdef CONFIG_SYN_COOKIES 476 477 /* Syncookies use a monotonic timer which increments every 60 seconds. 478 * This counter is used both as a hash input and partially encoded into 479 * the cookie value. A cookie is only validated further if the delta 480 * between the current counter value and the encoded one is less than this, 481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 482 * the counter advances immediately after a cookie is generated). 483 */ 484 #define MAX_SYNCOOKIE_AGE 2 485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 487 488 /* syncookies: remember time of last synqueue overflow 489 * But do not dirty this field too often (once per second is enough) 490 * It is racy as we do not hold a lock, but race is very minor. 491 */ 492 static inline void tcp_synq_overflow(const struct sock *sk) 493 { 494 unsigned int last_overflow; 495 unsigned int now = jiffies; 496 497 if (sk->sk_reuseport) { 498 struct sock_reuseport *reuse; 499 500 reuse = rcu_dereference(sk->sk_reuseport_cb); 501 if (likely(reuse)) { 502 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 503 if (!time_between32(now, last_overflow, 504 last_overflow + HZ)) 505 WRITE_ONCE(reuse->synq_overflow_ts, now); 506 return; 507 } 508 } 509 510 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 511 if (!time_between32(now, last_overflow, last_overflow + HZ)) 512 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 513 } 514 515 /* syncookies: no recent synqueue overflow on this listening socket? */ 516 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 517 { 518 unsigned int last_overflow; 519 unsigned int now = jiffies; 520 521 if (sk->sk_reuseport) { 522 struct sock_reuseport *reuse; 523 524 reuse = rcu_dereference(sk->sk_reuseport_cb); 525 if (likely(reuse)) { 526 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 527 return !time_between32(now, last_overflow - HZ, 528 last_overflow + 529 TCP_SYNCOOKIE_VALID); 530 } 531 } 532 533 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 534 535 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 536 * then we're under synflood. However, we have to use 537 * 'last_overflow - HZ' as lower bound. That's because a concurrent 538 * tcp_synq_overflow() could update .ts_recent_stamp after we read 539 * jiffies but before we store .ts_recent_stamp into last_overflow, 540 * which could lead to rejecting a valid syncookie. 541 */ 542 return !time_between32(now, last_overflow - HZ, 543 last_overflow + TCP_SYNCOOKIE_VALID); 544 } 545 546 static inline u32 tcp_cookie_time(void) 547 { 548 u64 val = get_jiffies_64(); 549 550 do_div(val, TCP_SYNCOOKIE_PERIOD); 551 return val; 552 } 553 554 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 555 u16 *mssp); 556 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 557 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 558 bool cookie_timestamp_decode(const struct net *net, 559 struct tcp_options_received *opt); 560 bool cookie_ecn_ok(const struct tcp_options_received *opt, 561 const struct net *net, const struct dst_entry *dst); 562 563 /* From net/ipv6/syncookies.c */ 564 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 565 u32 cookie); 566 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 567 568 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 569 const struct tcphdr *th, u16 *mssp); 570 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 571 #endif 572 /* tcp_output.c */ 573 574 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 575 int nonagle); 576 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 577 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 578 void tcp_retransmit_timer(struct sock *sk); 579 void tcp_xmit_retransmit_queue(struct sock *); 580 void tcp_simple_retransmit(struct sock *); 581 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 582 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 583 enum tcp_queue { 584 TCP_FRAG_IN_WRITE_QUEUE, 585 TCP_FRAG_IN_RTX_QUEUE, 586 }; 587 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 588 struct sk_buff *skb, u32 len, 589 unsigned int mss_now, gfp_t gfp); 590 591 void tcp_send_probe0(struct sock *); 592 void tcp_send_partial(struct sock *); 593 int tcp_write_wakeup(struct sock *, int mib); 594 void tcp_send_fin(struct sock *sk); 595 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 596 int tcp_send_synack(struct sock *); 597 void tcp_push_one(struct sock *, unsigned int mss_now); 598 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 599 void tcp_send_ack(struct sock *sk); 600 void tcp_send_delayed_ack(struct sock *sk); 601 void tcp_send_loss_probe(struct sock *sk); 602 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 603 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 604 const struct sk_buff *next_skb); 605 606 /* tcp_input.c */ 607 void tcp_rearm_rto(struct sock *sk); 608 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 609 void tcp_reset(struct sock *sk); 610 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 611 void tcp_fin(struct sock *sk); 612 613 /* tcp_timer.c */ 614 void tcp_init_xmit_timers(struct sock *); 615 static inline void tcp_clear_xmit_timers(struct sock *sk) 616 { 617 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 618 __sock_put(sk); 619 620 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 621 __sock_put(sk); 622 623 inet_csk_clear_xmit_timers(sk); 624 } 625 626 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 627 unsigned int tcp_current_mss(struct sock *sk); 628 629 /* Bound MSS / TSO packet size with the half of the window */ 630 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 631 { 632 int cutoff; 633 634 /* When peer uses tiny windows, there is no use in packetizing 635 * to sub-MSS pieces for the sake of SWS or making sure there 636 * are enough packets in the pipe for fast recovery. 637 * 638 * On the other hand, for extremely large MSS devices, handling 639 * smaller than MSS windows in this way does make sense. 640 */ 641 if (tp->max_window > TCP_MSS_DEFAULT) 642 cutoff = (tp->max_window >> 1); 643 else 644 cutoff = tp->max_window; 645 646 if (cutoff && pktsize > cutoff) 647 return max_t(int, cutoff, 68U - tp->tcp_header_len); 648 else 649 return pktsize; 650 } 651 652 /* tcp.c */ 653 void tcp_get_info(struct sock *, struct tcp_info *); 654 655 /* Read 'sendfile()'-style from a TCP socket */ 656 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 657 sk_read_actor_t recv_actor); 658 659 void tcp_initialize_rcv_mss(struct sock *sk); 660 661 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 662 int tcp_mss_to_mtu(struct sock *sk, int mss); 663 void tcp_mtup_init(struct sock *sk); 664 665 static inline void tcp_bound_rto(const struct sock *sk) 666 { 667 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 668 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 669 } 670 671 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 672 { 673 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 674 } 675 676 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 677 { 678 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 679 ntohl(TCP_FLAG_ACK) | 680 snd_wnd); 681 } 682 683 static inline void tcp_fast_path_on(struct tcp_sock *tp) 684 { 685 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 686 } 687 688 static inline void tcp_fast_path_check(struct sock *sk) 689 { 690 struct tcp_sock *tp = tcp_sk(sk); 691 692 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 693 tp->rcv_wnd && 694 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 695 !tp->urg_data) 696 tcp_fast_path_on(tp); 697 } 698 699 /* Compute the actual rto_min value */ 700 static inline u32 tcp_rto_min(struct sock *sk) 701 { 702 const struct dst_entry *dst = __sk_dst_get(sk); 703 u32 rto_min = inet_csk(sk)->icsk_rto_min; 704 705 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 706 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 707 return rto_min; 708 } 709 710 static inline u32 tcp_rto_min_us(struct sock *sk) 711 { 712 return jiffies_to_usecs(tcp_rto_min(sk)); 713 } 714 715 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 716 { 717 return dst_metric_locked(dst, RTAX_CC_ALGO); 718 } 719 720 /* Minimum RTT in usec. ~0 means not available. */ 721 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 722 { 723 return minmax_get(&tp->rtt_min); 724 } 725 726 /* Compute the actual receive window we are currently advertising. 727 * Rcv_nxt can be after the window if our peer push more data 728 * than the offered window. 729 */ 730 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 731 { 732 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 733 734 if (win < 0) 735 win = 0; 736 return (u32) win; 737 } 738 739 /* Choose a new window, without checks for shrinking, and without 740 * scaling applied to the result. The caller does these things 741 * if necessary. This is a "raw" window selection. 742 */ 743 u32 __tcp_select_window(struct sock *sk); 744 745 void tcp_send_window_probe(struct sock *sk); 746 747 /* TCP uses 32bit jiffies to save some space. 748 * Note that this is different from tcp_time_stamp, which 749 * historically has been the same until linux-4.13. 750 */ 751 #define tcp_jiffies32 ((u32)jiffies) 752 753 /* 754 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 755 * It is no longer tied to jiffies, but to 1 ms clock. 756 * Note: double check if you want to use tcp_jiffies32 instead of this. 757 */ 758 #define TCP_TS_HZ 1000 759 760 static inline u64 tcp_clock_ns(void) 761 { 762 return ktime_get_ns(); 763 } 764 765 static inline u64 tcp_clock_us(void) 766 { 767 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 768 } 769 770 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 771 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 772 { 773 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 774 } 775 776 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 777 static inline u32 tcp_ns_to_ts(u64 ns) 778 { 779 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 780 } 781 782 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 783 static inline u32 tcp_time_stamp_raw(void) 784 { 785 return tcp_ns_to_ts(tcp_clock_ns()); 786 } 787 788 void tcp_mstamp_refresh(struct tcp_sock *tp); 789 790 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 791 { 792 return max_t(s64, t1 - t0, 0); 793 } 794 795 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 796 { 797 return tcp_ns_to_ts(skb->skb_mstamp_ns); 798 } 799 800 /* provide the departure time in us unit */ 801 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 802 { 803 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 804 } 805 806 807 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 808 809 #define TCPHDR_FIN 0x01 810 #define TCPHDR_SYN 0x02 811 #define TCPHDR_RST 0x04 812 #define TCPHDR_PSH 0x08 813 #define TCPHDR_ACK 0x10 814 #define TCPHDR_URG 0x20 815 #define TCPHDR_ECE 0x40 816 #define TCPHDR_CWR 0x80 817 818 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 819 820 /* This is what the send packet queuing engine uses to pass 821 * TCP per-packet control information to the transmission code. 822 * We also store the host-order sequence numbers in here too. 823 * This is 44 bytes if IPV6 is enabled. 824 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 825 */ 826 struct tcp_skb_cb { 827 __u32 seq; /* Starting sequence number */ 828 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 829 union { 830 /* Note : tcp_tw_isn is used in input path only 831 * (isn chosen by tcp_timewait_state_process()) 832 * 833 * tcp_gso_segs/size are used in write queue only, 834 * cf tcp_skb_pcount()/tcp_skb_mss() 835 */ 836 __u32 tcp_tw_isn; 837 struct { 838 u16 tcp_gso_segs; 839 u16 tcp_gso_size; 840 }; 841 }; 842 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 843 844 __u8 sacked; /* State flags for SACK. */ 845 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 846 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 847 #define TCPCB_LOST 0x04 /* SKB is lost */ 848 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 849 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 850 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 851 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 852 TCPCB_REPAIRED) 853 854 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 855 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 856 eor:1, /* Is skb MSG_EOR marked? */ 857 has_rxtstamp:1, /* SKB has a RX timestamp */ 858 unused:5; 859 __u32 ack_seq; /* Sequence number ACK'd */ 860 union { 861 struct { 862 /* There is space for up to 24 bytes */ 863 __u32 in_flight:30,/* Bytes in flight at transmit */ 864 is_app_limited:1, /* cwnd not fully used? */ 865 unused:1; 866 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 867 __u32 delivered; 868 /* start of send pipeline phase */ 869 u64 first_tx_mstamp; 870 /* when we reached the "delivered" count */ 871 u64 delivered_mstamp; 872 } tx; /* only used for outgoing skbs */ 873 union { 874 struct inet_skb_parm h4; 875 #if IS_ENABLED(CONFIG_IPV6) 876 struct inet6_skb_parm h6; 877 #endif 878 } header; /* For incoming skbs */ 879 struct { 880 __u32 flags; 881 struct sock *sk_redir; 882 void *data_end; 883 } bpf; 884 }; 885 }; 886 887 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 888 889 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 890 { 891 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 892 } 893 894 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 895 { 896 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 897 } 898 899 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 900 { 901 return TCP_SKB_CB(skb)->bpf.sk_redir; 902 } 903 904 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 905 { 906 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 907 } 908 909 extern const struct inet_connection_sock_af_ops ipv4_specific; 910 911 #if IS_ENABLED(CONFIG_IPV6) 912 /* This is the variant of inet6_iif() that must be used by TCP, 913 * as TCP moves IP6CB into a different location in skb->cb[] 914 */ 915 static inline int tcp_v6_iif(const struct sk_buff *skb) 916 { 917 return TCP_SKB_CB(skb)->header.h6.iif; 918 } 919 920 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 921 { 922 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 923 924 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 925 } 926 927 /* TCP_SKB_CB reference means this can not be used from early demux */ 928 static inline int tcp_v6_sdif(const struct sk_buff *skb) 929 { 930 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 931 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 932 return TCP_SKB_CB(skb)->header.h6.iif; 933 #endif 934 return 0; 935 } 936 937 extern const struct inet_connection_sock_af_ops ipv6_specific; 938 939 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 940 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 941 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb)); 942 943 #endif 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 /* Estimates in how many jiffies next packet for this flow can be sent. 1291 * Scheduling a retransmit timer too early would be silly. 1292 */ 1293 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1294 { 1295 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1296 1297 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1298 } 1299 1300 static inline void tcp_reset_xmit_timer(struct sock *sk, 1301 const int what, 1302 unsigned long when, 1303 const unsigned long max_when) 1304 { 1305 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1306 max_when); 1307 } 1308 1309 /* Something is really bad, we could not queue an additional packet, 1310 * because qdisc is full or receiver sent a 0 window, or we are paced. 1311 * We do not want to add fuel to the fire, or abort too early, 1312 * so make sure the timer we arm now is at least 200ms in the future, 1313 * regardless of current icsk_rto value (as it could be ~2ms) 1314 */ 1315 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1316 { 1317 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1318 } 1319 1320 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1321 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1322 unsigned long max_when) 1323 { 1324 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1325 1326 return (unsigned long)min_t(u64, when, max_when); 1327 } 1328 1329 static inline void tcp_check_probe_timer(struct sock *sk) 1330 { 1331 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1332 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1333 tcp_probe0_base(sk), TCP_RTO_MAX); 1334 } 1335 1336 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1337 { 1338 tp->snd_wl1 = seq; 1339 } 1340 1341 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1342 { 1343 tp->snd_wl1 = seq; 1344 } 1345 1346 /* 1347 * Calculate(/check) TCP checksum 1348 */ 1349 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1350 __be32 daddr, __wsum base) 1351 { 1352 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1353 } 1354 1355 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1356 { 1357 return !skb_csum_unnecessary(skb) && 1358 __skb_checksum_complete(skb); 1359 } 1360 1361 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1362 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1363 void tcp_set_state(struct sock *sk, int state); 1364 void tcp_done(struct sock *sk); 1365 int tcp_abort(struct sock *sk, int err); 1366 1367 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1368 { 1369 rx_opt->dsack = 0; 1370 rx_opt->num_sacks = 0; 1371 } 1372 1373 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1374 1375 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1376 { 1377 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1378 struct tcp_sock *tp = tcp_sk(sk); 1379 s32 delta; 1380 1381 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1382 ca_ops->cong_control) 1383 return; 1384 delta = tcp_jiffies32 - tp->lsndtime; 1385 if (delta > inet_csk(sk)->icsk_rto) 1386 tcp_cwnd_restart(sk, delta); 1387 } 1388 1389 /* Determine a window scaling and initial window to offer. */ 1390 void tcp_select_initial_window(const struct sock *sk, int __space, 1391 __u32 mss, __u32 *rcv_wnd, 1392 __u32 *window_clamp, int wscale_ok, 1393 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1394 1395 static inline int tcp_win_from_space(const struct sock *sk, int space) 1396 { 1397 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1398 1399 return tcp_adv_win_scale <= 0 ? 1400 (space>>(-tcp_adv_win_scale)) : 1401 space - (space>>tcp_adv_win_scale); 1402 } 1403 1404 /* Note: caller must be prepared to deal with negative returns */ 1405 static inline int tcp_space(const struct sock *sk) 1406 { 1407 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1408 READ_ONCE(sk->sk_backlog.len) - 1409 atomic_read(&sk->sk_rmem_alloc)); 1410 } 1411 1412 static inline int tcp_full_space(const struct sock *sk) 1413 { 1414 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1415 } 1416 1417 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1418 * If 87.5 % (7/8) of the space has been consumed, we want to override 1419 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1420 * len/truesize ratio. 1421 */ 1422 static inline bool tcp_rmem_pressure(const struct sock *sk) 1423 { 1424 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1425 int threshold = rcvbuf - (rcvbuf >> 3); 1426 1427 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1428 } 1429 1430 extern void tcp_openreq_init_rwin(struct request_sock *req, 1431 const struct sock *sk_listener, 1432 const struct dst_entry *dst); 1433 1434 void tcp_enter_memory_pressure(struct sock *sk); 1435 void tcp_leave_memory_pressure(struct sock *sk); 1436 1437 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1438 { 1439 struct net *net = sock_net((struct sock *)tp); 1440 1441 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1442 } 1443 1444 static inline int keepalive_time_when(const struct tcp_sock *tp) 1445 { 1446 struct net *net = sock_net((struct sock *)tp); 1447 1448 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1449 } 1450 1451 static inline int keepalive_probes(const struct tcp_sock *tp) 1452 { 1453 struct net *net = sock_net((struct sock *)tp); 1454 1455 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1456 } 1457 1458 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1459 { 1460 const struct inet_connection_sock *icsk = &tp->inet_conn; 1461 1462 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1463 tcp_jiffies32 - tp->rcv_tstamp); 1464 } 1465 1466 static inline int tcp_fin_time(const struct sock *sk) 1467 { 1468 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1469 const int rto = inet_csk(sk)->icsk_rto; 1470 1471 if (fin_timeout < (rto << 2) - (rto >> 1)) 1472 fin_timeout = (rto << 2) - (rto >> 1); 1473 1474 return fin_timeout; 1475 } 1476 1477 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1478 int paws_win) 1479 { 1480 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1481 return true; 1482 if (unlikely(!time_before32(ktime_get_seconds(), 1483 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1484 return true; 1485 /* 1486 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1487 * then following tcp messages have valid values. Ignore 0 value, 1488 * or else 'negative' tsval might forbid us to accept their packets. 1489 */ 1490 if (!rx_opt->ts_recent) 1491 return true; 1492 return false; 1493 } 1494 1495 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1496 int rst) 1497 { 1498 if (tcp_paws_check(rx_opt, 0)) 1499 return false; 1500 1501 /* RST segments are not recommended to carry timestamp, 1502 and, if they do, it is recommended to ignore PAWS because 1503 "their cleanup function should take precedence over timestamps." 1504 Certainly, it is mistake. It is necessary to understand the reasons 1505 of this constraint to relax it: if peer reboots, clock may go 1506 out-of-sync and half-open connections will not be reset. 1507 Actually, the problem would be not existing if all 1508 the implementations followed draft about maintaining clock 1509 via reboots. Linux-2.2 DOES NOT! 1510 1511 However, we can relax time bounds for RST segments to MSL. 1512 */ 1513 if (rst && !time_before32(ktime_get_seconds(), 1514 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1515 return false; 1516 return true; 1517 } 1518 1519 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1520 int mib_idx, u32 *last_oow_ack_time); 1521 1522 static inline void tcp_mib_init(struct net *net) 1523 { 1524 /* See RFC 2012 */ 1525 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1526 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1527 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1528 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1529 } 1530 1531 /* from STCP */ 1532 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1533 { 1534 tp->lost_skb_hint = NULL; 1535 } 1536 1537 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1538 { 1539 tcp_clear_retrans_hints_partial(tp); 1540 tp->retransmit_skb_hint = NULL; 1541 } 1542 1543 union tcp_md5_addr { 1544 struct in_addr a4; 1545 #if IS_ENABLED(CONFIG_IPV6) 1546 struct in6_addr a6; 1547 #endif 1548 }; 1549 1550 /* - key database */ 1551 struct tcp_md5sig_key { 1552 struct hlist_node node; 1553 u8 keylen; 1554 u8 family; /* AF_INET or AF_INET6 */ 1555 u8 prefixlen; 1556 union tcp_md5_addr addr; 1557 int l3index; /* set if key added with L3 scope */ 1558 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1559 struct rcu_head rcu; 1560 }; 1561 1562 /* - sock block */ 1563 struct tcp_md5sig_info { 1564 struct hlist_head head; 1565 struct rcu_head rcu; 1566 }; 1567 1568 /* - pseudo header */ 1569 struct tcp4_pseudohdr { 1570 __be32 saddr; 1571 __be32 daddr; 1572 __u8 pad; 1573 __u8 protocol; 1574 __be16 len; 1575 }; 1576 1577 struct tcp6_pseudohdr { 1578 struct in6_addr saddr; 1579 struct in6_addr daddr; 1580 __be32 len; 1581 __be32 protocol; /* including padding */ 1582 }; 1583 1584 union tcp_md5sum_block { 1585 struct tcp4_pseudohdr ip4; 1586 #if IS_ENABLED(CONFIG_IPV6) 1587 struct tcp6_pseudohdr ip6; 1588 #endif 1589 }; 1590 1591 /* - pool: digest algorithm, hash description and scratch buffer */ 1592 struct tcp_md5sig_pool { 1593 struct ahash_request *md5_req; 1594 void *scratch; 1595 }; 1596 1597 /* - functions */ 1598 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1599 const struct sock *sk, const struct sk_buff *skb); 1600 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1601 int family, u8 prefixlen, int l3index, 1602 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1603 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1604 int family, u8 prefixlen, int l3index); 1605 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1606 const struct sock *addr_sk); 1607 1608 #ifdef CONFIG_TCP_MD5SIG 1609 #include <linux/jump_label.h> 1610 extern struct static_key_false tcp_md5_needed; 1611 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1612 const union tcp_md5_addr *addr, 1613 int family); 1614 static inline struct tcp_md5sig_key * 1615 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1616 const union tcp_md5_addr *addr, int family) 1617 { 1618 if (!static_branch_unlikely(&tcp_md5_needed)) 1619 return NULL; 1620 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1621 } 1622 1623 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1624 #else 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 return NULL; 1630 } 1631 #define tcp_twsk_md5_key(twsk) NULL 1632 #endif 1633 1634 bool tcp_alloc_md5sig_pool(void); 1635 1636 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1637 static inline void tcp_put_md5sig_pool(void) 1638 { 1639 local_bh_enable(); 1640 } 1641 1642 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1643 unsigned int header_len); 1644 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1645 const struct tcp_md5sig_key *key); 1646 1647 /* From tcp_fastopen.c */ 1648 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1649 struct tcp_fastopen_cookie *cookie); 1650 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1651 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1652 u16 try_exp); 1653 struct tcp_fastopen_request { 1654 /* Fast Open cookie. Size 0 means a cookie request */ 1655 struct tcp_fastopen_cookie cookie; 1656 struct msghdr *data; /* data in MSG_FASTOPEN */ 1657 size_t size; 1658 int copied; /* queued in tcp_connect() */ 1659 struct ubuf_info *uarg; 1660 }; 1661 void tcp_free_fastopen_req(struct tcp_sock *tp); 1662 void tcp_fastopen_destroy_cipher(struct sock *sk); 1663 void tcp_fastopen_ctx_destroy(struct net *net); 1664 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1665 void *primary_key, void *backup_key); 1666 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1667 u64 *key); 1668 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1669 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1670 struct request_sock *req, 1671 struct tcp_fastopen_cookie *foc, 1672 const struct dst_entry *dst); 1673 void tcp_fastopen_init_key_once(struct net *net); 1674 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1675 struct tcp_fastopen_cookie *cookie); 1676 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1677 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1678 #define TCP_FASTOPEN_KEY_MAX 2 1679 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1680 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1681 1682 /* Fastopen key context */ 1683 struct tcp_fastopen_context { 1684 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1685 int num; 1686 struct rcu_head rcu; 1687 }; 1688 1689 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1690 void tcp_fastopen_active_disable(struct sock *sk); 1691 bool tcp_fastopen_active_should_disable(struct sock *sk); 1692 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1693 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1694 1695 /* Caller needs to wrap with rcu_read_(un)lock() */ 1696 static inline 1697 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1698 { 1699 struct tcp_fastopen_context *ctx; 1700 1701 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1702 if (!ctx) 1703 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1704 return ctx; 1705 } 1706 1707 static inline 1708 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1709 const struct tcp_fastopen_cookie *orig) 1710 { 1711 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1712 orig->len == foc->len && 1713 !memcmp(orig->val, foc->val, foc->len)) 1714 return true; 1715 return false; 1716 } 1717 1718 static inline 1719 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1720 { 1721 return ctx->num; 1722 } 1723 1724 /* Latencies incurred by various limits for a sender. They are 1725 * chronograph-like stats that are mutually exclusive. 1726 */ 1727 enum tcp_chrono { 1728 TCP_CHRONO_UNSPEC, 1729 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1730 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1731 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1732 __TCP_CHRONO_MAX, 1733 }; 1734 1735 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1736 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1737 1738 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1739 * the same memory storage than skb->destructor/_skb_refdst 1740 */ 1741 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1742 { 1743 skb->destructor = NULL; 1744 skb->_skb_refdst = 0UL; 1745 } 1746 1747 #define tcp_skb_tsorted_save(skb) { \ 1748 unsigned long _save = skb->_skb_refdst; \ 1749 skb->_skb_refdst = 0UL; 1750 1751 #define tcp_skb_tsorted_restore(skb) \ 1752 skb->_skb_refdst = _save; \ 1753 } 1754 1755 void tcp_write_queue_purge(struct sock *sk); 1756 1757 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1758 { 1759 return skb_rb_first(&sk->tcp_rtx_queue); 1760 } 1761 1762 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1763 { 1764 return skb_rb_last(&sk->tcp_rtx_queue); 1765 } 1766 1767 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1768 { 1769 return skb_peek(&sk->sk_write_queue); 1770 } 1771 1772 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1773 { 1774 return skb_peek_tail(&sk->sk_write_queue); 1775 } 1776 1777 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1778 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1779 1780 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1781 { 1782 return skb_peek(&sk->sk_write_queue); 1783 } 1784 1785 static inline bool tcp_skb_is_last(const struct sock *sk, 1786 const struct sk_buff *skb) 1787 { 1788 return skb_queue_is_last(&sk->sk_write_queue, skb); 1789 } 1790 1791 /** 1792 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1793 * @sk: socket 1794 * 1795 * Since the write queue can have a temporary empty skb in it, 1796 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1797 */ 1798 static inline bool tcp_write_queue_empty(const struct sock *sk) 1799 { 1800 const struct tcp_sock *tp = tcp_sk(sk); 1801 1802 return tp->write_seq == tp->snd_nxt; 1803 } 1804 1805 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1806 { 1807 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1808 } 1809 1810 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1811 { 1812 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1813 } 1814 1815 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1816 { 1817 __skb_queue_tail(&sk->sk_write_queue, skb); 1818 1819 /* Queue it, remembering where we must start sending. */ 1820 if (sk->sk_write_queue.next == skb) 1821 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1822 } 1823 1824 /* Insert new before skb on the write queue of sk. */ 1825 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1826 struct sk_buff *skb, 1827 struct sock *sk) 1828 { 1829 __skb_queue_before(&sk->sk_write_queue, skb, new); 1830 } 1831 1832 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1833 { 1834 tcp_skb_tsorted_anchor_cleanup(skb); 1835 __skb_unlink(skb, &sk->sk_write_queue); 1836 } 1837 1838 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1839 1840 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1841 { 1842 tcp_skb_tsorted_anchor_cleanup(skb); 1843 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1844 } 1845 1846 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1847 { 1848 list_del(&skb->tcp_tsorted_anchor); 1849 tcp_rtx_queue_unlink(skb, sk); 1850 sk_wmem_free_skb(sk, skb); 1851 } 1852 1853 static inline void tcp_push_pending_frames(struct sock *sk) 1854 { 1855 if (tcp_send_head(sk)) { 1856 struct tcp_sock *tp = tcp_sk(sk); 1857 1858 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1859 } 1860 } 1861 1862 /* Start sequence of the skb just after the highest skb with SACKed 1863 * bit, valid only if sacked_out > 0 or when the caller has ensured 1864 * validity by itself. 1865 */ 1866 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1867 { 1868 if (!tp->sacked_out) 1869 return tp->snd_una; 1870 1871 if (tp->highest_sack == NULL) 1872 return tp->snd_nxt; 1873 1874 return TCP_SKB_CB(tp->highest_sack)->seq; 1875 } 1876 1877 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1878 { 1879 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1880 } 1881 1882 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1883 { 1884 return tcp_sk(sk)->highest_sack; 1885 } 1886 1887 static inline void tcp_highest_sack_reset(struct sock *sk) 1888 { 1889 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1890 } 1891 1892 /* Called when old skb is about to be deleted and replaced by new skb */ 1893 static inline void tcp_highest_sack_replace(struct sock *sk, 1894 struct sk_buff *old, 1895 struct sk_buff *new) 1896 { 1897 if (old == tcp_highest_sack(sk)) 1898 tcp_sk(sk)->highest_sack = new; 1899 } 1900 1901 /* This helper checks if socket has IP_TRANSPARENT set */ 1902 static inline bool inet_sk_transparent(const struct sock *sk) 1903 { 1904 switch (sk->sk_state) { 1905 case TCP_TIME_WAIT: 1906 return inet_twsk(sk)->tw_transparent; 1907 case TCP_NEW_SYN_RECV: 1908 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1909 } 1910 return inet_sk(sk)->transparent; 1911 } 1912 1913 /* Determines whether this is a thin stream (which may suffer from 1914 * increased latency). Used to trigger latency-reducing mechanisms. 1915 */ 1916 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1917 { 1918 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1919 } 1920 1921 /* /proc */ 1922 enum tcp_seq_states { 1923 TCP_SEQ_STATE_LISTENING, 1924 TCP_SEQ_STATE_ESTABLISHED, 1925 }; 1926 1927 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1928 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1929 void tcp_seq_stop(struct seq_file *seq, void *v); 1930 1931 struct tcp_seq_afinfo { 1932 sa_family_t family; 1933 }; 1934 1935 struct tcp_iter_state { 1936 struct seq_net_private p; 1937 enum tcp_seq_states state; 1938 struct sock *syn_wait_sk; 1939 struct tcp_seq_afinfo *bpf_seq_afinfo; 1940 int bucket, offset, sbucket, num; 1941 loff_t last_pos; 1942 }; 1943 1944 extern struct request_sock_ops tcp_request_sock_ops; 1945 extern struct request_sock_ops tcp6_request_sock_ops; 1946 1947 void tcp_v4_destroy_sock(struct sock *sk); 1948 1949 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1950 netdev_features_t features); 1951 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1952 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 1953 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 1954 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 1955 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 1956 int tcp_gro_complete(struct sk_buff *skb); 1957 1958 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1959 1960 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1961 { 1962 struct net *net = sock_net((struct sock *)tp); 1963 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1964 } 1965 1966 /* @wake is one when sk_stream_write_space() calls us. 1967 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1968 * This mimics the strategy used in sock_def_write_space(). 1969 */ 1970 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1971 { 1972 const struct tcp_sock *tp = tcp_sk(sk); 1973 u32 notsent_bytes = READ_ONCE(tp->write_seq) - 1974 READ_ONCE(tp->snd_nxt); 1975 1976 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1977 } 1978 1979 #ifdef CONFIG_PROC_FS 1980 int tcp4_proc_init(void); 1981 void tcp4_proc_exit(void); 1982 #endif 1983 1984 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1985 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1986 const struct tcp_request_sock_ops *af_ops, 1987 struct sock *sk, struct sk_buff *skb); 1988 1989 /* TCP af-specific functions */ 1990 struct tcp_sock_af_ops { 1991 #ifdef CONFIG_TCP_MD5SIG 1992 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1993 const struct sock *addr_sk); 1994 int (*calc_md5_hash)(char *location, 1995 const struct tcp_md5sig_key *md5, 1996 const struct sock *sk, 1997 const struct sk_buff *skb); 1998 int (*md5_parse)(struct sock *sk, 1999 int optname, 2000 sockptr_t optval, 2001 int optlen); 2002 #endif 2003 }; 2004 2005 struct tcp_request_sock_ops { 2006 u16 mss_clamp; 2007 #ifdef CONFIG_TCP_MD5SIG 2008 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2009 const struct sock *addr_sk); 2010 int (*calc_md5_hash) (char *location, 2011 const struct tcp_md5sig_key *md5, 2012 const struct sock *sk, 2013 const struct sk_buff *skb); 2014 #endif 2015 void (*init_req)(struct request_sock *req, 2016 const struct sock *sk_listener, 2017 struct sk_buff *skb); 2018 #ifdef CONFIG_SYN_COOKIES 2019 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2020 __u16 *mss); 2021 #endif 2022 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 2023 const struct request_sock *req); 2024 u32 (*init_seq)(const struct sk_buff *skb); 2025 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2026 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2027 struct flowi *fl, struct request_sock *req, 2028 struct tcp_fastopen_cookie *foc, 2029 enum tcp_synack_type synack_type, 2030 struct sk_buff *syn_skb); 2031 }; 2032 2033 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2034 #if IS_ENABLED(CONFIG_IPV6) 2035 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2036 #endif 2037 2038 #ifdef CONFIG_SYN_COOKIES 2039 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2040 const struct sock *sk, struct sk_buff *skb, 2041 __u16 *mss) 2042 { 2043 tcp_synq_overflow(sk); 2044 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2045 return ops->cookie_init_seq(skb, mss); 2046 } 2047 #else 2048 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2049 const struct sock *sk, struct sk_buff *skb, 2050 __u16 *mss) 2051 { 2052 return 0; 2053 } 2054 #endif 2055 2056 int tcpv4_offload_init(void); 2057 2058 void tcp_v4_init(void); 2059 void tcp_init(void); 2060 2061 /* tcp_recovery.c */ 2062 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2063 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2064 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2065 u32 reo_wnd); 2066 extern void tcp_rack_mark_lost(struct sock *sk); 2067 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2068 u64 xmit_time); 2069 extern void tcp_rack_reo_timeout(struct sock *sk); 2070 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2071 2072 /* At how many usecs into the future should the RTO fire? */ 2073 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2074 { 2075 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2076 u32 rto = inet_csk(sk)->icsk_rto; 2077 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2078 2079 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2080 } 2081 2082 /* 2083 * Save and compile IPv4 options, return a pointer to it 2084 */ 2085 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2086 struct sk_buff *skb) 2087 { 2088 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2089 struct ip_options_rcu *dopt = NULL; 2090 2091 if (opt->optlen) { 2092 int opt_size = sizeof(*dopt) + opt->optlen; 2093 2094 dopt = kmalloc(opt_size, GFP_ATOMIC); 2095 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2096 kfree(dopt); 2097 dopt = NULL; 2098 } 2099 } 2100 return dopt; 2101 } 2102 2103 /* locally generated TCP pure ACKs have skb->truesize == 2 2104 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2105 * This is much faster than dissecting the packet to find out. 2106 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2107 */ 2108 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2109 { 2110 return skb->truesize == 2; 2111 } 2112 2113 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2114 { 2115 skb->truesize = 2; 2116 } 2117 2118 static inline int tcp_inq(struct sock *sk) 2119 { 2120 struct tcp_sock *tp = tcp_sk(sk); 2121 int answ; 2122 2123 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2124 answ = 0; 2125 } else if (sock_flag(sk, SOCK_URGINLINE) || 2126 !tp->urg_data || 2127 before(tp->urg_seq, tp->copied_seq) || 2128 !before(tp->urg_seq, tp->rcv_nxt)) { 2129 2130 answ = tp->rcv_nxt - tp->copied_seq; 2131 2132 /* Subtract 1, if FIN was received */ 2133 if (answ && sock_flag(sk, SOCK_DONE)) 2134 answ--; 2135 } else { 2136 answ = tp->urg_seq - tp->copied_seq; 2137 } 2138 2139 return answ; 2140 } 2141 2142 int tcp_peek_len(struct socket *sock); 2143 2144 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2145 { 2146 u16 segs_in; 2147 2148 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2149 tp->segs_in += segs_in; 2150 if (skb->len > tcp_hdrlen(skb)) 2151 tp->data_segs_in += segs_in; 2152 } 2153 2154 /* 2155 * TCP listen path runs lockless. 2156 * We forced "struct sock" to be const qualified to make sure 2157 * we don't modify one of its field by mistake. 2158 * Here, we increment sk_drops which is an atomic_t, so we can safely 2159 * make sock writable again. 2160 */ 2161 static inline void tcp_listendrop(const struct sock *sk) 2162 { 2163 atomic_inc(&((struct sock *)sk)->sk_drops); 2164 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2165 } 2166 2167 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2168 2169 /* 2170 * Interface for adding Upper Level Protocols over TCP 2171 */ 2172 2173 #define TCP_ULP_NAME_MAX 16 2174 #define TCP_ULP_MAX 128 2175 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2176 2177 struct tcp_ulp_ops { 2178 struct list_head list; 2179 2180 /* initialize ulp */ 2181 int (*init)(struct sock *sk); 2182 /* update ulp */ 2183 void (*update)(struct sock *sk, struct proto *p, 2184 void (*write_space)(struct sock *sk)); 2185 /* cleanup ulp */ 2186 void (*release)(struct sock *sk); 2187 /* diagnostic */ 2188 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2189 size_t (*get_info_size)(const struct sock *sk); 2190 /* clone ulp */ 2191 void (*clone)(const struct request_sock *req, struct sock *newsk, 2192 const gfp_t priority); 2193 2194 char name[TCP_ULP_NAME_MAX]; 2195 struct module *owner; 2196 }; 2197 int tcp_register_ulp(struct tcp_ulp_ops *type); 2198 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2199 int tcp_set_ulp(struct sock *sk, const char *name); 2200 void tcp_get_available_ulp(char *buf, size_t len); 2201 void tcp_cleanup_ulp(struct sock *sk); 2202 void tcp_update_ulp(struct sock *sk, struct proto *p, 2203 void (*write_space)(struct sock *sk)); 2204 2205 #define MODULE_ALIAS_TCP_ULP(name) \ 2206 __MODULE_INFO(alias, alias_userspace, name); \ 2207 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2208 2209 struct sk_msg; 2210 struct sk_psock; 2211 2212 #ifdef CONFIG_BPF_STREAM_PARSER 2213 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2214 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2215 #else 2216 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2217 { 2218 } 2219 #endif /* CONFIG_BPF_STREAM_PARSER */ 2220 2221 #ifdef CONFIG_NET_SOCK_MSG 2222 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2223 int flags); 2224 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2225 struct msghdr *msg, int len, int flags); 2226 #endif /* CONFIG_NET_SOCK_MSG */ 2227 2228 #ifdef CONFIG_CGROUP_BPF 2229 /* Copy the listen sk's HDR_OPT_CB flags to its child. 2230 * 2231 * During 3-Way-HandShake, the synack is usually sent from 2232 * the listen sk with the HDR_OPT_CB flags set so that 2233 * bpf-prog will be called to write the BPF hdr option. 2234 * 2235 * In fastopen, the child sk is used to send synack instead 2236 * of the listen sk. Thus, inheriting the HDR_OPT_CB flags 2237 * from the listen sk gives the bpf-prog a chance to write 2238 * BPF hdr option in the synack pkt during fastopen. 2239 * 2240 * Both fastopen and non-fastopen child will inherit the 2241 * HDR_OPT_CB flags to keep the bpf-prog having a consistent 2242 * behavior when deciding to clear this cb flags (or not) 2243 * during the PASSIVE_ESTABLISHED_CB. 2244 * 2245 * In the future, other cb flags could be inherited here also. 2246 */ 2247 static inline void bpf_skops_init_child(const struct sock *sk, 2248 struct sock *child) 2249 { 2250 tcp_sk(child)->bpf_sock_ops_cb_flags = 2251 tcp_sk(sk)->bpf_sock_ops_cb_flags & 2252 (BPF_SOCK_OPS_PARSE_ALL_HDR_OPT_CB_FLAG | 2253 BPF_SOCK_OPS_PARSE_UNKNOWN_HDR_OPT_CB_FLAG | 2254 BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG); 2255 } 2256 2257 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2258 struct sk_buff *skb, 2259 unsigned int end_offset) 2260 { 2261 skops->skb = skb; 2262 skops->skb_data_end = skb->data + end_offset; 2263 } 2264 #else 2265 static inline void bpf_skops_init_child(const struct sock *sk, 2266 struct sock *child) 2267 { 2268 } 2269 2270 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2271 struct sk_buff *skb, 2272 unsigned int end_offset) 2273 { 2274 } 2275 #endif 2276 2277 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2278 * is < 0, then the BPF op failed (for example if the loaded BPF 2279 * program does not support the chosen operation or there is no BPF 2280 * program loaded). 2281 */ 2282 #ifdef CONFIG_BPF 2283 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2284 { 2285 struct bpf_sock_ops_kern sock_ops; 2286 int ret; 2287 2288 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2289 if (sk_fullsock(sk)) { 2290 sock_ops.is_fullsock = 1; 2291 sock_owned_by_me(sk); 2292 } 2293 2294 sock_ops.sk = sk; 2295 sock_ops.op = op; 2296 if (nargs > 0) 2297 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2298 2299 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2300 if (ret == 0) 2301 ret = sock_ops.reply; 2302 else 2303 ret = -1; 2304 return ret; 2305 } 2306 2307 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2308 { 2309 u32 args[2] = {arg1, arg2}; 2310 2311 return tcp_call_bpf(sk, op, 2, args); 2312 } 2313 2314 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2315 u32 arg3) 2316 { 2317 u32 args[3] = {arg1, arg2, arg3}; 2318 2319 return tcp_call_bpf(sk, op, 3, args); 2320 } 2321 2322 #else 2323 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2324 { 2325 return -EPERM; 2326 } 2327 2328 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2329 { 2330 return -EPERM; 2331 } 2332 2333 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2334 u32 arg3) 2335 { 2336 return -EPERM; 2337 } 2338 2339 #endif 2340 2341 static inline u32 tcp_timeout_init(struct sock *sk) 2342 { 2343 int timeout; 2344 2345 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2346 2347 if (timeout <= 0) 2348 timeout = TCP_TIMEOUT_INIT; 2349 return timeout; 2350 } 2351 2352 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2353 { 2354 int rwnd; 2355 2356 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2357 2358 if (rwnd < 0) 2359 rwnd = 0; 2360 return rwnd; 2361 } 2362 2363 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2364 { 2365 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2366 } 2367 2368 static inline void tcp_bpf_rtt(struct sock *sk) 2369 { 2370 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2371 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2372 } 2373 2374 #if IS_ENABLED(CONFIG_SMC) 2375 extern struct static_key_false tcp_have_smc; 2376 #endif 2377 2378 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2379 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2380 void (*cad)(struct sock *sk, u32 ack_seq)); 2381 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2382 void clean_acked_data_flush(void); 2383 #endif 2384 2385 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2386 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2387 const struct tcp_sock *tp) 2388 { 2389 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2390 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2391 } 2392 2393 /* Compute Earliest Departure Time for some control packets 2394 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2395 */ 2396 static inline u64 tcp_transmit_time(const struct sock *sk) 2397 { 2398 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2399 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2400 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2401 2402 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2403 } 2404 return 0; 2405 } 2406 2407 #endif /* _TCP_H */ 2408