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/tcp_ao.h> 41 #include <net/inet_ecn.h> 42 #include <net/dst.h> 43 #include <net/mptcp.h> 44 45 #include <linux/seq_file.h> 46 #include <linux/memcontrol.h> 47 #include <linux/bpf-cgroup.h> 48 #include <linux/siphash.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 53 int tcp_orphan_count_sum(void); 54 55 void tcp_time_wait(struct sock *sk, int state, int timeo); 56 57 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 58 #define MAX_TCP_OPTION_SPACE 40 59 #define TCP_MIN_SND_MSS 48 60 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 61 62 /* 63 * Never offer a window over 32767 without using window scaling. Some 64 * poor stacks do signed 16bit maths! 65 */ 66 #define MAX_TCP_WINDOW 32767U 67 68 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 69 #define TCP_MIN_MSS 88U 70 71 /* The initial MTU to use for probing */ 72 #define TCP_BASE_MSS 1024 73 74 /* probing interval, default to 10 minutes as per RFC4821 */ 75 #define TCP_PROBE_INTERVAL 600 76 77 /* Specify interval when tcp mtu probing will stop */ 78 #define TCP_PROBE_THRESHOLD 8 79 80 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 81 #define TCP_FASTRETRANS_THRESH 3 82 83 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 84 #define TCP_MAX_QUICKACKS 16U 85 86 /* Maximal number of window scale according to RFC1323 */ 87 #define TCP_MAX_WSCALE 14U 88 89 /* urg_data states */ 90 #define TCP_URG_VALID 0x0100 91 #define TCP_URG_NOTYET 0x0200 92 #define TCP_URG_READ 0x0400 93 94 #define TCP_RETR1 3 /* 95 * This is how many retries it does before it 96 * tries to figure out if the gateway is 97 * down. Minimal RFC value is 3; it corresponds 98 * to ~3sec-8min depending on RTO. 99 */ 100 101 #define TCP_RETR2 15 /* 102 * This should take at least 103 * 90 minutes to time out. 104 * RFC1122 says that the limit is 100 sec. 105 * 15 is ~13-30min depending on RTO. 106 */ 107 108 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 109 * when active opening a connection. 110 * RFC1122 says the minimum retry MUST 111 * be at least 180secs. Nevertheless 112 * this value is corresponding to 113 * 63secs of retransmission with the 114 * current initial RTO. 115 */ 116 117 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 118 * when passive opening a connection. 119 * This is corresponding to 31secs of 120 * retransmission with the current 121 * initial RTO. 122 */ 123 124 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 125 * state, about 60 seconds */ 126 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 127 /* BSD style FIN_WAIT2 deadlock breaker. 128 * It used to be 3min, new value is 60sec, 129 * to combine FIN-WAIT-2 timeout with 130 * TIME-WAIT timer. 131 */ 132 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 133 134 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 135 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); 136 137 #if HZ >= 100 138 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 139 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 140 #else 141 #define TCP_DELACK_MIN 4U 142 #define TCP_ATO_MIN 4U 143 #endif 144 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 145 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 146 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 147 148 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 149 150 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 151 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 152 * used as a fallback RTO for the 153 * initial data transmission if no 154 * valid RTT sample has been acquired, 155 * most likely due to retrans in 3WHS. 156 */ 157 158 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 159 * for local resources. 160 */ 161 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 162 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 163 #define TCP_KEEPALIVE_INTVL (75*HZ) 164 165 #define MAX_TCP_KEEPIDLE 32767 166 #define MAX_TCP_KEEPINTVL 32767 167 #define MAX_TCP_KEEPCNT 127 168 #define MAX_TCP_SYNCNT 127 169 170 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds 171 * to avoid overflows. This assumes a clock smaller than 1 Mhz. 172 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz. 173 */ 174 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC) 175 176 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 177 * after this time. It should be equal 178 * (or greater than) TCP_TIMEWAIT_LEN 179 * to provide reliability equal to one 180 * provided by timewait state. 181 */ 182 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 183 * timestamps. It must be less than 184 * minimal timewait lifetime. 185 */ 186 /* 187 * TCP option 188 */ 189 190 #define TCPOPT_NOP 1 /* Padding */ 191 #define TCPOPT_EOL 0 /* End of options */ 192 #define TCPOPT_MSS 2 /* Segment size negotiating */ 193 #define TCPOPT_WINDOW 3 /* Window scaling */ 194 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 195 #define TCPOPT_SACK 5 /* SACK Block */ 196 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 197 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 198 #define TCPOPT_AO 29 /* Authentication Option (RFC5925) */ 199 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 200 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 201 #define TCPOPT_EXP 254 /* Experimental */ 202 /* Magic number to be after the option value for sharing TCP 203 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 204 */ 205 #define TCPOPT_FASTOPEN_MAGIC 0xF989 206 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 207 208 /* 209 * TCP option lengths 210 */ 211 212 #define TCPOLEN_MSS 4 213 #define TCPOLEN_WINDOW 3 214 #define TCPOLEN_SACK_PERM 2 215 #define TCPOLEN_TIMESTAMP 10 216 #define TCPOLEN_MD5SIG 18 217 #define TCPOLEN_FASTOPEN_BASE 2 218 #define TCPOLEN_EXP_FASTOPEN_BASE 4 219 #define TCPOLEN_EXP_SMC_BASE 6 220 221 /* But this is what stacks really send out. */ 222 #define TCPOLEN_TSTAMP_ALIGNED 12 223 #define TCPOLEN_WSCALE_ALIGNED 4 224 #define TCPOLEN_SACKPERM_ALIGNED 4 225 #define TCPOLEN_SACK_BASE 2 226 #define TCPOLEN_SACK_BASE_ALIGNED 4 227 #define TCPOLEN_SACK_PERBLOCK 8 228 #define TCPOLEN_MD5SIG_ALIGNED 20 229 #define TCPOLEN_MSS_ALIGNED 4 230 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 231 232 /* Flags in tp->nonagle */ 233 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 234 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 235 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 236 237 /* TCP thin-stream limits */ 238 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 239 240 /* TCP initial congestion window as per rfc6928 */ 241 #define TCP_INIT_CWND 10 242 243 /* Bit Flags for sysctl_tcp_fastopen */ 244 #define TFO_CLIENT_ENABLE 1 245 #define TFO_SERVER_ENABLE 2 246 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 247 248 /* Accept SYN data w/o any cookie option */ 249 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 250 251 /* Force enable TFO on all listeners, i.e., not requiring the 252 * TCP_FASTOPEN socket option. 253 */ 254 #define TFO_SERVER_WO_SOCKOPT1 0x400 255 256 257 /* sysctl variables for tcp */ 258 extern int sysctl_tcp_max_orphans; 259 extern long sysctl_tcp_mem[3]; 260 261 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 262 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 263 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 264 265 extern atomic_long_t tcp_memory_allocated; 266 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 267 268 extern struct percpu_counter tcp_sockets_allocated; 269 extern unsigned long tcp_memory_pressure; 270 271 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 272 static inline bool tcp_under_memory_pressure(const struct sock *sk) 273 { 274 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 275 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 276 return true; 277 278 return READ_ONCE(tcp_memory_pressure); 279 } 280 /* 281 * The next routines deal with comparing 32 bit unsigned ints 282 * and worry about wraparound (automatic with unsigned arithmetic). 283 */ 284 285 static inline bool before(__u32 seq1, __u32 seq2) 286 { 287 return (__s32)(seq1-seq2) < 0; 288 } 289 #define after(seq2, seq1) before(seq1, seq2) 290 291 /* is s2<=s1<=s3 ? */ 292 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 293 { 294 return seq3 - seq2 >= seq1 - seq2; 295 } 296 297 static inline bool tcp_out_of_memory(struct sock *sk) 298 { 299 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 300 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 301 return true; 302 return false; 303 } 304 305 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 306 { 307 sk_wmem_queued_add(sk, -skb->truesize); 308 if (!skb_zcopy_pure(skb)) 309 sk_mem_uncharge(sk, skb->truesize); 310 else 311 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 312 __kfree_skb(skb); 313 } 314 315 void sk_forced_mem_schedule(struct sock *sk, int size); 316 317 bool tcp_check_oom(struct sock *sk, int shift); 318 319 320 extern struct proto tcp_prot; 321 322 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 323 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 324 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 325 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 326 327 void tcp_tasklet_init(void); 328 329 int tcp_v4_err(struct sk_buff *skb, u32); 330 331 void tcp_shutdown(struct sock *sk, int how); 332 333 int tcp_v4_early_demux(struct sk_buff *skb); 334 int tcp_v4_rcv(struct sk_buff *skb); 335 336 void tcp_remove_empty_skb(struct sock *sk); 337 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 338 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 339 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 340 size_t size, struct ubuf_info *uarg); 341 void tcp_splice_eof(struct socket *sock); 342 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 343 int tcp_wmem_schedule(struct sock *sk, int copy); 344 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 345 int size_goal); 346 void tcp_release_cb(struct sock *sk); 347 void tcp_wfree(struct sk_buff *skb); 348 void tcp_write_timer_handler(struct sock *sk); 349 void tcp_delack_timer_handler(struct sock *sk); 350 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 351 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 352 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 353 void tcp_rcv_space_adjust(struct sock *sk); 354 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 355 void tcp_twsk_destructor(struct sock *sk); 356 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 357 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 358 struct pipe_inode_info *pipe, size_t len, 359 unsigned int flags); 360 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 361 bool force_schedule); 362 363 static inline void tcp_dec_quickack_mode(struct sock *sk) 364 { 365 struct inet_connection_sock *icsk = inet_csk(sk); 366 367 if (icsk->icsk_ack.quick) { 368 /* How many ACKs S/ACKing new data have we sent? */ 369 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 370 371 if (pkts >= icsk->icsk_ack.quick) { 372 icsk->icsk_ack.quick = 0; 373 /* Leaving quickack mode we deflate ATO. */ 374 icsk->icsk_ack.ato = TCP_ATO_MIN; 375 } else 376 icsk->icsk_ack.quick -= pkts; 377 } 378 } 379 380 #define TCP_ECN_OK 1 381 #define TCP_ECN_QUEUE_CWR 2 382 #define TCP_ECN_DEMAND_CWR 4 383 #define TCP_ECN_SEEN 8 384 385 enum tcp_tw_status { 386 TCP_TW_SUCCESS = 0, 387 TCP_TW_RST = 1, 388 TCP_TW_ACK = 2, 389 TCP_TW_SYN = 3 390 }; 391 392 393 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 394 struct sk_buff *skb, 395 const struct tcphdr *th); 396 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 397 struct request_sock *req, bool fastopen, 398 bool *lost_race); 399 int tcp_child_process(struct sock *parent, struct sock *child, 400 struct sk_buff *skb); 401 void tcp_enter_loss(struct sock *sk); 402 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 403 void tcp_clear_retrans(struct tcp_sock *tp); 404 void tcp_update_metrics(struct sock *sk); 405 void tcp_init_metrics(struct sock *sk); 406 void tcp_metrics_init(void); 407 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 408 void __tcp_close(struct sock *sk, long timeout); 409 void tcp_close(struct sock *sk, long timeout); 410 void tcp_init_sock(struct sock *sk); 411 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 412 __poll_t tcp_poll(struct file *file, struct socket *sock, 413 struct poll_table_struct *wait); 414 int do_tcp_getsockopt(struct sock *sk, int level, 415 int optname, sockptr_t optval, sockptr_t optlen); 416 int tcp_getsockopt(struct sock *sk, int level, int optname, 417 char __user *optval, int __user *optlen); 418 bool tcp_bpf_bypass_getsockopt(int level, int optname); 419 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 420 sockptr_t optval, unsigned int optlen); 421 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 422 unsigned int optlen); 423 void tcp_set_keepalive(struct sock *sk, int val); 424 void tcp_syn_ack_timeout(const struct request_sock *req); 425 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 426 int flags, int *addr_len); 427 int tcp_set_rcvlowat(struct sock *sk, int val); 428 int tcp_set_window_clamp(struct sock *sk, int val); 429 void tcp_update_recv_tstamps(struct sk_buff *skb, 430 struct scm_timestamping_internal *tss); 431 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 432 struct scm_timestamping_internal *tss); 433 void tcp_data_ready(struct sock *sk); 434 #ifdef CONFIG_MMU 435 int tcp_mmap(struct file *file, struct socket *sock, 436 struct vm_area_struct *vma); 437 #endif 438 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 439 struct tcp_options_received *opt_rx, 440 int estab, struct tcp_fastopen_cookie *foc); 441 442 /* 443 * BPF SKB-less helpers 444 */ 445 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 446 struct tcphdr *th, u32 *cookie); 447 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 448 struct tcphdr *th, u32 *cookie); 449 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 450 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 451 const struct tcp_request_sock_ops *af_ops, 452 struct sock *sk, struct tcphdr *th); 453 /* 454 * TCP v4 functions exported for the inet6 API 455 */ 456 457 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 458 void tcp_v4_mtu_reduced(struct sock *sk); 459 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 460 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 461 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 462 struct sock *tcp_create_openreq_child(const struct sock *sk, 463 struct request_sock *req, 464 struct sk_buff *skb); 465 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 466 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 467 struct request_sock *req, 468 struct dst_entry *dst, 469 struct request_sock *req_unhash, 470 bool *own_req); 471 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 472 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 473 int tcp_connect(struct sock *sk); 474 enum tcp_synack_type { 475 TCP_SYNACK_NORMAL, 476 TCP_SYNACK_FASTOPEN, 477 TCP_SYNACK_COOKIE, 478 }; 479 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 480 struct request_sock *req, 481 struct tcp_fastopen_cookie *foc, 482 enum tcp_synack_type synack_type, 483 struct sk_buff *syn_skb); 484 int tcp_disconnect(struct sock *sk, int flags); 485 486 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 487 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 488 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 489 490 /* From syncookies.c */ 491 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 492 struct request_sock *req, 493 struct dst_entry *dst, u32 tsoff); 494 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 495 u32 cookie); 496 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 497 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 498 const struct tcp_request_sock_ops *af_ops, 499 struct sock *sk, struct sk_buff *skb); 500 #ifdef CONFIG_SYN_COOKIES 501 502 /* Syncookies use a monotonic timer which increments every 60 seconds. 503 * This counter is used both as a hash input and partially encoded into 504 * the cookie value. A cookie is only validated further if the delta 505 * between the current counter value and the encoded one is less than this, 506 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 507 * the counter advances immediately after a cookie is generated). 508 */ 509 #define MAX_SYNCOOKIE_AGE 2 510 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 511 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 512 513 /* syncookies: remember time of last synqueue overflow 514 * But do not dirty this field too often (once per second is enough) 515 * It is racy as we do not hold a lock, but race is very minor. 516 */ 517 static inline void tcp_synq_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 if (!time_between32(now, last_overflow, 529 last_overflow + HZ)) 530 WRITE_ONCE(reuse->synq_overflow_ts, now); 531 return; 532 } 533 } 534 535 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 536 if (!time_between32(now, last_overflow, last_overflow + HZ)) 537 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 538 } 539 540 /* syncookies: no recent synqueue overflow on this listening socket? */ 541 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 542 { 543 unsigned int last_overflow; 544 unsigned int now = jiffies; 545 546 if (sk->sk_reuseport) { 547 struct sock_reuseport *reuse; 548 549 reuse = rcu_dereference(sk->sk_reuseport_cb); 550 if (likely(reuse)) { 551 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 552 return !time_between32(now, last_overflow - HZ, 553 last_overflow + 554 TCP_SYNCOOKIE_VALID); 555 } 556 } 557 558 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 559 560 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 561 * then we're under synflood. However, we have to use 562 * 'last_overflow - HZ' as lower bound. That's because a concurrent 563 * tcp_synq_overflow() could update .ts_recent_stamp after we read 564 * jiffies but before we store .ts_recent_stamp into last_overflow, 565 * which could lead to rejecting a valid syncookie. 566 */ 567 return !time_between32(now, last_overflow - HZ, 568 last_overflow + TCP_SYNCOOKIE_VALID); 569 } 570 571 static inline u32 tcp_cookie_time(void) 572 { 573 u64 val = get_jiffies_64(); 574 575 do_div(val, TCP_SYNCOOKIE_PERIOD); 576 return val; 577 } 578 579 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 580 u16 *mssp); 581 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 582 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 583 bool cookie_timestamp_decode(const struct net *net, 584 struct tcp_options_received *opt); 585 bool cookie_ecn_ok(const struct tcp_options_received *opt, 586 const struct net *net, const struct dst_entry *dst); 587 588 /* From net/ipv6/syncookies.c */ 589 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 590 u32 cookie); 591 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 592 593 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 594 const struct tcphdr *th, u16 *mssp); 595 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 596 #endif 597 /* tcp_output.c */ 598 599 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 600 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 601 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 602 int nonagle); 603 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 604 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 605 void tcp_retransmit_timer(struct sock *sk); 606 void tcp_xmit_retransmit_queue(struct sock *); 607 void tcp_simple_retransmit(struct sock *); 608 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 609 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 610 enum tcp_queue { 611 TCP_FRAG_IN_WRITE_QUEUE, 612 TCP_FRAG_IN_RTX_QUEUE, 613 }; 614 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 615 struct sk_buff *skb, u32 len, 616 unsigned int mss_now, gfp_t gfp); 617 618 void tcp_send_probe0(struct sock *); 619 int tcp_write_wakeup(struct sock *, int mib); 620 void tcp_send_fin(struct sock *sk); 621 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 622 int tcp_send_synack(struct sock *); 623 void tcp_push_one(struct sock *, unsigned int mss_now); 624 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 625 void tcp_send_ack(struct sock *sk); 626 void tcp_send_delayed_ack(struct sock *sk); 627 void tcp_send_loss_probe(struct sock *sk); 628 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 629 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 630 const struct sk_buff *next_skb); 631 632 /* tcp_input.c */ 633 void tcp_rearm_rto(struct sock *sk); 634 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 635 void tcp_reset(struct sock *sk, struct sk_buff *skb); 636 void tcp_fin(struct sock *sk); 637 void tcp_check_space(struct sock *sk); 638 void tcp_sack_compress_send_ack(struct sock *sk); 639 640 /* tcp_timer.c */ 641 void tcp_init_xmit_timers(struct sock *); 642 static inline void tcp_clear_xmit_timers(struct sock *sk) 643 { 644 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 645 __sock_put(sk); 646 647 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 648 __sock_put(sk); 649 650 inet_csk_clear_xmit_timers(sk); 651 } 652 653 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 654 unsigned int tcp_current_mss(struct sock *sk); 655 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 656 657 /* Bound MSS / TSO packet size with the half of the window */ 658 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 659 { 660 int cutoff; 661 662 /* When peer uses tiny windows, there is no use in packetizing 663 * to sub-MSS pieces for the sake of SWS or making sure there 664 * are enough packets in the pipe for fast recovery. 665 * 666 * On the other hand, for extremely large MSS devices, handling 667 * smaller than MSS windows in this way does make sense. 668 */ 669 if (tp->max_window > TCP_MSS_DEFAULT) 670 cutoff = (tp->max_window >> 1); 671 else 672 cutoff = tp->max_window; 673 674 if (cutoff && pktsize > cutoff) 675 return max_t(int, cutoff, 68U - tp->tcp_header_len); 676 else 677 return pktsize; 678 } 679 680 /* tcp.c */ 681 void tcp_get_info(struct sock *, struct tcp_info *); 682 683 /* Read 'sendfile()'-style from a TCP socket */ 684 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 685 sk_read_actor_t recv_actor); 686 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 687 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 688 void tcp_read_done(struct sock *sk, size_t len); 689 690 void tcp_initialize_rcv_mss(struct sock *sk); 691 692 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 693 int tcp_mss_to_mtu(struct sock *sk, int mss); 694 void tcp_mtup_init(struct sock *sk); 695 696 static inline void tcp_bound_rto(const struct sock *sk) 697 { 698 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 699 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 700 } 701 702 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 703 { 704 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 705 } 706 707 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 708 { 709 /* mptcp hooks are only on the slow path */ 710 if (sk_is_mptcp((struct sock *)tp)) 711 return; 712 713 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 714 ntohl(TCP_FLAG_ACK) | 715 snd_wnd); 716 } 717 718 static inline void tcp_fast_path_on(struct tcp_sock *tp) 719 { 720 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 721 } 722 723 static inline void tcp_fast_path_check(struct sock *sk) 724 { 725 struct tcp_sock *tp = tcp_sk(sk); 726 727 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 728 tp->rcv_wnd && 729 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 730 !tp->urg_data) 731 tcp_fast_path_on(tp); 732 } 733 734 u32 tcp_delack_max(const struct sock *sk); 735 736 /* Compute the actual rto_min value */ 737 static inline u32 tcp_rto_min(const struct sock *sk) 738 { 739 const struct dst_entry *dst = __sk_dst_get(sk); 740 u32 rto_min = inet_csk(sk)->icsk_rto_min; 741 742 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 743 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 744 return rto_min; 745 } 746 747 static inline u32 tcp_rto_min_us(const struct sock *sk) 748 { 749 return jiffies_to_usecs(tcp_rto_min(sk)); 750 } 751 752 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 753 { 754 return dst_metric_locked(dst, RTAX_CC_ALGO); 755 } 756 757 /* Minimum RTT in usec. ~0 means not available. */ 758 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 759 { 760 return minmax_get(&tp->rtt_min); 761 } 762 763 /* Compute the actual receive window we are currently advertising. 764 * Rcv_nxt can be after the window if our peer push more data 765 * than the offered window. 766 */ 767 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 768 { 769 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 770 771 if (win < 0) 772 win = 0; 773 return (u32) win; 774 } 775 776 /* Choose a new window, without checks for shrinking, and without 777 * scaling applied to the result. The caller does these things 778 * if necessary. This is a "raw" window selection. 779 */ 780 u32 __tcp_select_window(struct sock *sk); 781 782 void tcp_send_window_probe(struct sock *sk); 783 784 /* TCP uses 32bit jiffies to save some space. 785 * Note that this is different from tcp_time_stamp, which 786 * historically has been the same until linux-4.13. 787 */ 788 #define tcp_jiffies32 ((u32)jiffies) 789 790 /* 791 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 792 * It is no longer tied to jiffies, but to 1 ms clock. 793 * Note: double check if you want to use tcp_jiffies32 instead of this. 794 */ 795 #define TCP_TS_HZ 1000 796 797 static inline u64 tcp_clock_ns(void) 798 { 799 return ktime_get_ns(); 800 } 801 802 static inline u64 tcp_clock_us(void) 803 { 804 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 805 } 806 807 static inline u64 tcp_clock_ms(void) 808 { 809 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 810 } 811 812 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 813 * or usec resolution. Each socket carries a flag to select one or other 814 * resolution, as the route attribute could change anytime. 815 * Each flow must stick to initial resolution. 816 */ 817 static inline u32 tcp_clock_ts(bool usec_ts) 818 { 819 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 820 } 821 822 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 823 { 824 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 825 } 826 827 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 828 { 829 if (tp->tcp_usec_ts) 830 return tp->tcp_mstamp; 831 return tcp_time_stamp_ms(tp); 832 } 833 834 void tcp_mstamp_refresh(struct tcp_sock *tp); 835 836 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 837 { 838 return max_t(s64, t1 - t0, 0); 839 } 840 841 /* provide the departure time in us unit */ 842 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 843 { 844 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 845 } 846 847 /* Provide skb TSval in usec or ms unit */ 848 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 849 { 850 if (usec_ts) 851 return tcp_skb_timestamp_us(skb); 852 853 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 854 } 855 856 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 857 { 858 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 859 } 860 861 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 862 { 863 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 864 } 865 866 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 867 868 #define TCPHDR_FIN 0x01 869 #define TCPHDR_SYN 0x02 870 #define TCPHDR_RST 0x04 871 #define TCPHDR_PSH 0x08 872 #define TCPHDR_ACK 0x10 873 #define TCPHDR_URG 0x20 874 #define TCPHDR_ECE 0x40 875 #define TCPHDR_CWR 0x80 876 877 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 878 879 /* This is what the send packet queuing engine uses to pass 880 * TCP per-packet control information to the transmission code. 881 * We also store the host-order sequence numbers in here too. 882 * This is 44 bytes if IPV6 is enabled. 883 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 884 */ 885 struct tcp_skb_cb { 886 __u32 seq; /* Starting sequence number */ 887 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 888 union { 889 /* Note : tcp_tw_isn is used in input path only 890 * (isn chosen by tcp_timewait_state_process()) 891 * 892 * tcp_gso_segs/size are used in write queue only, 893 * cf tcp_skb_pcount()/tcp_skb_mss() 894 */ 895 __u32 tcp_tw_isn; 896 struct { 897 u16 tcp_gso_segs; 898 u16 tcp_gso_size; 899 }; 900 }; 901 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 902 903 __u8 sacked; /* State flags for SACK. */ 904 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 905 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 906 #define TCPCB_LOST 0x04 /* SKB is lost */ 907 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 908 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 909 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 910 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 911 TCPCB_REPAIRED) 912 913 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 914 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 915 eor:1, /* Is skb MSG_EOR marked? */ 916 has_rxtstamp:1, /* SKB has a RX timestamp */ 917 unused:5; 918 __u32 ack_seq; /* Sequence number ACK'd */ 919 union { 920 struct { 921 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 922 /* There is space for up to 24 bytes */ 923 __u32 is_app_limited:1, /* cwnd not fully used? */ 924 delivered_ce:20, 925 unused:11; 926 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 927 __u32 delivered; 928 /* start of send pipeline phase */ 929 u64 first_tx_mstamp; 930 /* when we reached the "delivered" count */ 931 u64 delivered_mstamp; 932 } tx; /* only used for outgoing skbs */ 933 union { 934 struct inet_skb_parm h4; 935 #if IS_ENABLED(CONFIG_IPV6) 936 struct inet6_skb_parm h6; 937 #endif 938 } header; /* For incoming skbs */ 939 }; 940 }; 941 942 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 943 944 extern const struct inet_connection_sock_af_ops ipv4_specific; 945 946 #if IS_ENABLED(CONFIG_IPV6) 947 /* This is the variant of inet6_iif() that must be used by TCP, 948 * as TCP moves IP6CB into a different location in skb->cb[] 949 */ 950 static inline int tcp_v6_iif(const struct sk_buff *skb) 951 { 952 return TCP_SKB_CB(skb)->header.h6.iif; 953 } 954 955 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 956 { 957 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 958 959 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 960 } 961 962 /* TCP_SKB_CB reference means this can not be used from early demux */ 963 static inline int tcp_v6_sdif(const struct sk_buff *skb) 964 { 965 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 966 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 967 return TCP_SKB_CB(skb)->header.h6.iif; 968 #endif 969 return 0; 970 } 971 972 extern const struct inet_connection_sock_af_ops ipv6_specific; 973 974 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 975 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 976 void tcp_v6_early_demux(struct sk_buff *skb); 977 978 #endif 979 980 /* TCP_SKB_CB reference means this can not be used from early demux */ 981 static inline int tcp_v4_sdif(struct sk_buff *skb) 982 { 983 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 984 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 985 return TCP_SKB_CB(skb)->header.h4.iif; 986 #endif 987 return 0; 988 } 989 990 /* Due to TSO, an SKB can be composed of multiple actual 991 * packets. To keep these tracked properly, we use this. 992 */ 993 static inline int tcp_skb_pcount(const struct sk_buff *skb) 994 { 995 return TCP_SKB_CB(skb)->tcp_gso_segs; 996 } 997 998 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 999 { 1000 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1001 } 1002 1003 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1004 { 1005 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1006 } 1007 1008 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1009 static inline int tcp_skb_mss(const struct sk_buff *skb) 1010 { 1011 return TCP_SKB_CB(skb)->tcp_gso_size; 1012 } 1013 1014 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1015 { 1016 return likely(!TCP_SKB_CB(skb)->eor); 1017 } 1018 1019 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1020 const struct sk_buff *from) 1021 { 1022 return likely(tcp_skb_can_collapse_to(to) && 1023 mptcp_skb_can_collapse(to, from) && 1024 skb_pure_zcopy_same(to, from)); 1025 } 1026 1027 /* Events passed to congestion control interface */ 1028 enum tcp_ca_event { 1029 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1030 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1031 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1032 CA_EVENT_LOSS, /* loss timeout */ 1033 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1034 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1035 }; 1036 1037 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1038 enum tcp_ca_ack_event_flags { 1039 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1040 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1041 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1042 }; 1043 1044 /* 1045 * Interface for adding new TCP congestion control handlers 1046 */ 1047 #define TCP_CA_NAME_MAX 16 1048 #define TCP_CA_MAX 128 1049 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1050 1051 #define TCP_CA_UNSPEC 0 1052 1053 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1054 #define TCP_CONG_NON_RESTRICTED 0x1 1055 /* Requires ECN/ECT set on all packets */ 1056 #define TCP_CONG_NEEDS_ECN 0x2 1057 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1058 1059 union tcp_cc_info; 1060 1061 struct ack_sample { 1062 u32 pkts_acked; 1063 s32 rtt_us; 1064 u32 in_flight; 1065 }; 1066 1067 /* A rate sample measures the number of (original/retransmitted) data 1068 * packets delivered "delivered" over an interval of time "interval_us". 1069 * The tcp_rate.c code fills in the rate sample, and congestion 1070 * control modules that define a cong_control function to run at the end 1071 * of ACK processing can optionally chose to consult this sample when 1072 * setting cwnd and pacing rate. 1073 * A sample is invalid if "delivered" or "interval_us" is negative. 1074 */ 1075 struct rate_sample { 1076 u64 prior_mstamp; /* starting timestamp for interval */ 1077 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1078 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1079 s32 delivered; /* number of packets delivered over interval */ 1080 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1081 long interval_us; /* time for tp->delivered to incr "delivered" */ 1082 u32 snd_interval_us; /* snd interval for delivered packets */ 1083 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1084 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1085 int losses; /* number of packets marked lost upon ACK */ 1086 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1087 u32 prior_in_flight; /* in flight before this ACK */ 1088 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1089 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1090 bool is_retrans; /* is sample from retransmission? */ 1091 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1092 }; 1093 1094 struct tcp_congestion_ops { 1095 /* fast path fields are put first to fill one cache line */ 1096 1097 /* return slow start threshold (required) */ 1098 u32 (*ssthresh)(struct sock *sk); 1099 1100 /* do new cwnd calculation (required) */ 1101 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1102 1103 /* call before changing ca_state (optional) */ 1104 void (*set_state)(struct sock *sk, u8 new_state); 1105 1106 /* call when cwnd event occurs (optional) */ 1107 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1108 1109 /* call when ack arrives (optional) */ 1110 void (*in_ack_event)(struct sock *sk, u32 flags); 1111 1112 /* hook for packet ack accounting (optional) */ 1113 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1114 1115 /* override sysctl_tcp_min_tso_segs */ 1116 u32 (*min_tso_segs)(struct sock *sk); 1117 1118 /* call when packets are delivered to update cwnd and pacing rate, 1119 * after all the ca_state processing. (optional) 1120 */ 1121 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1122 1123 1124 /* new value of cwnd after loss (required) */ 1125 u32 (*undo_cwnd)(struct sock *sk); 1126 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1127 u32 (*sndbuf_expand)(struct sock *sk); 1128 1129 /* control/slow paths put last */ 1130 /* get info for inet_diag (optional) */ 1131 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1132 union tcp_cc_info *info); 1133 1134 char name[TCP_CA_NAME_MAX]; 1135 struct module *owner; 1136 struct list_head list; 1137 u32 key; 1138 u32 flags; 1139 1140 /* initialize private data (optional) */ 1141 void (*init)(struct sock *sk); 1142 /* cleanup private data (optional) */ 1143 void (*release)(struct sock *sk); 1144 } ____cacheline_aligned_in_smp; 1145 1146 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1147 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1148 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1149 struct tcp_congestion_ops *old_type); 1150 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1151 1152 void tcp_assign_congestion_control(struct sock *sk); 1153 void tcp_init_congestion_control(struct sock *sk); 1154 void tcp_cleanup_congestion_control(struct sock *sk); 1155 int tcp_set_default_congestion_control(struct net *net, const char *name); 1156 void tcp_get_default_congestion_control(struct net *net, char *name); 1157 void tcp_get_available_congestion_control(char *buf, size_t len); 1158 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1159 int tcp_set_allowed_congestion_control(char *allowed); 1160 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1161 bool cap_net_admin); 1162 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1163 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1164 1165 u32 tcp_reno_ssthresh(struct sock *sk); 1166 u32 tcp_reno_undo_cwnd(struct sock *sk); 1167 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1168 extern struct tcp_congestion_ops tcp_reno; 1169 1170 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1171 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1172 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1173 #ifdef CONFIG_INET 1174 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1175 #else 1176 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1177 { 1178 return NULL; 1179 } 1180 #endif 1181 1182 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1183 { 1184 const struct inet_connection_sock *icsk = inet_csk(sk); 1185 1186 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1187 } 1188 1189 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1190 { 1191 const struct inet_connection_sock *icsk = inet_csk(sk); 1192 1193 if (icsk->icsk_ca_ops->cwnd_event) 1194 icsk->icsk_ca_ops->cwnd_event(sk, event); 1195 } 1196 1197 /* From tcp_cong.c */ 1198 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1199 1200 /* From tcp_rate.c */ 1201 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1202 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1203 struct rate_sample *rs); 1204 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1205 bool is_sack_reneg, struct rate_sample *rs); 1206 void tcp_rate_check_app_limited(struct sock *sk); 1207 1208 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1209 { 1210 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1211 } 1212 1213 /* These functions determine how the current flow behaves in respect of SACK 1214 * handling. SACK is negotiated with the peer, and therefore it can vary 1215 * between different flows. 1216 * 1217 * tcp_is_sack - SACK enabled 1218 * tcp_is_reno - No SACK 1219 */ 1220 static inline int tcp_is_sack(const struct tcp_sock *tp) 1221 { 1222 return likely(tp->rx_opt.sack_ok); 1223 } 1224 1225 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1226 { 1227 return !tcp_is_sack(tp); 1228 } 1229 1230 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1231 { 1232 return tp->sacked_out + tp->lost_out; 1233 } 1234 1235 /* This determines how many packets are "in the network" to the best 1236 * of our knowledge. In many cases it is conservative, but where 1237 * detailed information is available from the receiver (via SACK 1238 * blocks etc.) we can make more aggressive calculations. 1239 * 1240 * Use this for decisions involving congestion control, use just 1241 * tp->packets_out to determine if the send queue is empty or not. 1242 * 1243 * Read this equation as: 1244 * 1245 * "Packets sent once on transmission queue" MINUS 1246 * "Packets left network, but not honestly ACKed yet" PLUS 1247 * "Packets fast retransmitted" 1248 */ 1249 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1250 { 1251 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1252 } 1253 1254 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1255 1256 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1257 { 1258 return tp->snd_cwnd; 1259 } 1260 1261 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1262 { 1263 WARN_ON_ONCE((int)val <= 0); 1264 tp->snd_cwnd = val; 1265 } 1266 1267 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1268 { 1269 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1270 } 1271 1272 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1273 { 1274 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1275 } 1276 1277 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1278 { 1279 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1280 (1 << inet_csk(sk)->icsk_ca_state); 1281 } 1282 1283 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1284 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1285 * ssthresh. 1286 */ 1287 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1288 { 1289 const struct tcp_sock *tp = tcp_sk(sk); 1290 1291 if (tcp_in_cwnd_reduction(sk)) 1292 return tp->snd_ssthresh; 1293 else 1294 return max(tp->snd_ssthresh, 1295 ((tcp_snd_cwnd(tp) >> 1) + 1296 (tcp_snd_cwnd(tp) >> 2))); 1297 } 1298 1299 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1300 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1301 1302 void tcp_enter_cwr(struct sock *sk); 1303 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1304 1305 /* The maximum number of MSS of available cwnd for which TSO defers 1306 * sending if not using sysctl_tcp_tso_win_divisor. 1307 */ 1308 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1309 { 1310 return 3; 1311 } 1312 1313 /* Returns end sequence number of the receiver's advertised window */ 1314 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1315 { 1316 return tp->snd_una + tp->snd_wnd; 1317 } 1318 1319 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1320 * flexible approach. The RFC suggests cwnd should not be raised unless 1321 * it was fully used previously. And that's exactly what we do in 1322 * congestion avoidance mode. But in slow start we allow cwnd to grow 1323 * as long as the application has used half the cwnd. 1324 * Example : 1325 * cwnd is 10 (IW10), but application sends 9 frames. 1326 * We allow cwnd to reach 18 when all frames are ACKed. 1327 * This check is safe because it's as aggressive as slow start which already 1328 * risks 100% overshoot. The advantage is that we discourage application to 1329 * either send more filler packets or data to artificially blow up the cwnd 1330 * usage, and allow application-limited process to probe bw more aggressively. 1331 */ 1332 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1333 { 1334 const struct tcp_sock *tp = tcp_sk(sk); 1335 1336 if (tp->is_cwnd_limited) 1337 return true; 1338 1339 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1340 if (tcp_in_slow_start(tp)) 1341 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1342 1343 return false; 1344 } 1345 1346 /* BBR congestion control needs pacing. 1347 * Same remark for SO_MAX_PACING_RATE. 1348 * sch_fq packet scheduler is efficiently handling pacing, 1349 * but is not always installed/used. 1350 * Return true if TCP stack should pace packets itself. 1351 */ 1352 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1353 { 1354 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1355 } 1356 1357 /* Estimates in how many jiffies next packet for this flow can be sent. 1358 * Scheduling a retransmit timer too early would be silly. 1359 */ 1360 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1361 { 1362 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1363 1364 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1365 } 1366 1367 static inline void tcp_reset_xmit_timer(struct sock *sk, 1368 const int what, 1369 unsigned long when, 1370 const unsigned long max_when) 1371 { 1372 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1373 max_when); 1374 } 1375 1376 /* Something is really bad, we could not queue an additional packet, 1377 * because qdisc is full or receiver sent a 0 window, or we are paced. 1378 * We do not want to add fuel to the fire, or abort too early, 1379 * so make sure the timer we arm now is at least 200ms in the future, 1380 * regardless of current icsk_rto value (as it could be ~2ms) 1381 */ 1382 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1383 { 1384 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1385 } 1386 1387 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1388 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1389 unsigned long max_when) 1390 { 1391 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1392 inet_csk(sk)->icsk_backoff); 1393 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1394 1395 return (unsigned long)min_t(u64, when, max_when); 1396 } 1397 1398 static inline void tcp_check_probe_timer(struct sock *sk) 1399 { 1400 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1401 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1402 tcp_probe0_base(sk), TCP_RTO_MAX); 1403 } 1404 1405 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1406 { 1407 tp->snd_wl1 = seq; 1408 } 1409 1410 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1411 { 1412 tp->snd_wl1 = seq; 1413 } 1414 1415 /* 1416 * Calculate(/check) TCP checksum 1417 */ 1418 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1419 __be32 daddr, __wsum base) 1420 { 1421 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1422 } 1423 1424 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1425 { 1426 return !skb_csum_unnecessary(skb) && 1427 __skb_checksum_complete(skb); 1428 } 1429 1430 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1431 enum skb_drop_reason *reason); 1432 1433 1434 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1435 void tcp_set_state(struct sock *sk, int state); 1436 void tcp_done(struct sock *sk); 1437 int tcp_abort(struct sock *sk, int err); 1438 1439 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1440 { 1441 rx_opt->dsack = 0; 1442 rx_opt->num_sacks = 0; 1443 } 1444 1445 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1446 1447 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1448 { 1449 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1450 struct tcp_sock *tp = tcp_sk(sk); 1451 s32 delta; 1452 1453 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1454 tp->packets_out || ca_ops->cong_control) 1455 return; 1456 delta = tcp_jiffies32 - tp->lsndtime; 1457 if (delta > inet_csk(sk)->icsk_rto) 1458 tcp_cwnd_restart(sk, delta); 1459 } 1460 1461 /* Determine a window scaling and initial window to offer. */ 1462 void tcp_select_initial_window(const struct sock *sk, int __space, 1463 __u32 mss, __u32 *rcv_wnd, 1464 __u32 *window_clamp, int wscale_ok, 1465 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1466 1467 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1468 { 1469 s64 scaled_space = (s64)space * scaling_ratio; 1470 1471 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1472 } 1473 1474 static inline int tcp_win_from_space(const struct sock *sk, int space) 1475 { 1476 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1477 } 1478 1479 /* inverse of __tcp_win_from_space() */ 1480 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1481 { 1482 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1483 1484 do_div(val, scaling_ratio); 1485 return val; 1486 } 1487 1488 static inline int tcp_space_from_win(const struct sock *sk, int win) 1489 { 1490 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1491 } 1492 1493 /* Assume a conservative default of 1200 bytes of payload per 4K page. 1494 * This may be adjusted later in tcp_measure_rcv_mss(). 1495 */ 1496 #define TCP_DEFAULT_SCALING_RATIO ((1200 << TCP_RMEM_TO_WIN_SCALE) / \ 1497 SKB_TRUESIZE(4096)) 1498 1499 static inline void tcp_scaling_ratio_init(struct sock *sk) 1500 { 1501 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1502 } 1503 1504 /* Note: caller must be prepared to deal with negative returns */ 1505 static inline int tcp_space(const struct sock *sk) 1506 { 1507 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1508 READ_ONCE(sk->sk_backlog.len) - 1509 atomic_read(&sk->sk_rmem_alloc)); 1510 } 1511 1512 static inline int tcp_full_space(const struct sock *sk) 1513 { 1514 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1515 } 1516 1517 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1518 { 1519 int unused_mem = sk_unused_reserved_mem(sk); 1520 struct tcp_sock *tp = tcp_sk(sk); 1521 1522 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1523 if (unused_mem) 1524 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1525 tcp_win_from_space(sk, unused_mem)); 1526 } 1527 1528 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1529 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1530 1531 1532 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1533 * If 87.5 % (7/8) of the space has been consumed, we want to override 1534 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1535 * len/truesize ratio. 1536 */ 1537 static inline bool tcp_rmem_pressure(const struct sock *sk) 1538 { 1539 int rcvbuf, threshold; 1540 1541 if (tcp_under_memory_pressure(sk)) 1542 return true; 1543 1544 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1545 threshold = rcvbuf - (rcvbuf >> 3); 1546 1547 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1548 } 1549 1550 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1551 { 1552 const struct tcp_sock *tp = tcp_sk(sk); 1553 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1554 1555 if (avail <= 0) 1556 return false; 1557 1558 return (avail >= target) || tcp_rmem_pressure(sk) || 1559 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1560 } 1561 1562 extern void tcp_openreq_init_rwin(struct request_sock *req, 1563 const struct sock *sk_listener, 1564 const struct dst_entry *dst); 1565 1566 void tcp_enter_memory_pressure(struct sock *sk); 1567 void tcp_leave_memory_pressure(struct sock *sk); 1568 1569 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1570 { 1571 struct net *net = sock_net((struct sock *)tp); 1572 int val; 1573 1574 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1575 * and do_tcp_setsockopt(). 1576 */ 1577 val = READ_ONCE(tp->keepalive_intvl); 1578 1579 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1580 } 1581 1582 static inline int keepalive_time_when(const struct tcp_sock *tp) 1583 { 1584 struct net *net = sock_net((struct sock *)tp); 1585 int val; 1586 1587 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1588 val = READ_ONCE(tp->keepalive_time); 1589 1590 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1591 } 1592 1593 static inline int keepalive_probes(const struct tcp_sock *tp) 1594 { 1595 struct net *net = sock_net((struct sock *)tp); 1596 int val; 1597 1598 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1599 * and do_tcp_setsockopt(). 1600 */ 1601 val = READ_ONCE(tp->keepalive_probes); 1602 1603 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1604 } 1605 1606 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1607 { 1608 const struct inet_connection_sock *icsk = &tp->inet_conn; 1609 1610 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1611 tcp_jiffies32 - tp->rcv_tstamp); 1612 } 1613 1614 static inline int tcp_fin_time(const struct sock *sk) 1615 { 1616 int fin_timeout = tcp_sk(sk)->linger2 ? : 1617 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1618 const int rto = inet_csk(sk)->icsk_rto; 1619 1620 if (fin_timeout < (rto << 2) - (rto >> 1)) 1621 fin_timeout = (rto << 2) - (rto >> 1); 1622 1623 return fin_timeout; 1624 } 1625 1626 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1627 int paws_win) 1628 { 1629 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1630 return true; 1631 if (unlikely(!time_before32(ktime_get_seconds(), 1632 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1633 return true; 1634 /* 1635 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1636 * then following tcp messages have valid values. Ignore 0 value, 1637 * or else 'negative' tsval might forbid us to accept their packets. 1638 */ 1639 if (!rx_opt->ts_recent) 1640 return true; 1641 return false; 1642 } 1643 1644 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1645 int rst) 1646 { 1647 if (tcp_paws_check(rx_opt, 0)) 1648 return false; 1649 1650 /* RST segments are not recommended to carry timestamp, 1651 and, if they do, it is recommended to ignore PAWS because 1652 "their cleanup function should take precedence over timestamps." 1653 Certainly, it is mistake. It is necessary to understand the reasons 1654 of this constraint to relax it: if peer reboots, clock may go 1655 out-of-sync and half-open connections will not be reset. 1656 Actually, the problem would be not existing if all 1657 the implementations followed draft about maintaining clock 1658 via reboots. Linux-2.2 DOES NOT! 1659 1660 However, we can relax time bounds for RST segments to MSL. 1661 */ 1662 if (rst && !time_before32(ktime_get_seconds(), 1663 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1664 return false; 1665 return true; 1666 } 1667 1668 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1669 int mib_idx, u32 *last_oow_ack_time); 1670 1671 static inline void tcp_mib_init(struct net *net) 1672 { 1673 /* See RFC 2012 */ 1674 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1675 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1676 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1677 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1678 } 1679 1680 /* from STCP */ 1681 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1682 { 1683 tp->lost_skb_hint = NULL; 1684 } 1685 1686 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1687 { 1688 tcp_clear_retrans_hints_partial(tp); 1689 tp->retransmit_skb_hint = NULL; 1690 } 1691 1692 #define tcp_md5_addr tcp_ao_addr 1693 1694 /* - key database */ 1695 struct tcp_md5sig_key { 1696 struct hlist_node node; 1697 u8 keylen; 1698 u8 family; /* AF_INET or AF_INET6 */ 1699 u8 prefixlen; 1700 u8 flags; 1701 union tcp_md5_addr addr; 1702 int l3index; /* set if key added with L3 scope */ 1703 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1704 struct rcu_head rcu; 1705 }; 1706 1707 /* - sock block */ 1708 struct tcp_md5sig_info { 1709 struct hlist_head head; 1710 struct rcu_head rcu; 1711 }; 1712 1713 /* - pseudo header */ 1714 struct tcp4_pseudohdr { 1715 __be32 saddr; 1716 __be32 daddr; 1717 __u8 pad; 1718 __u8 protocol; 1719 __be16 len; 1720 }; 1721 1722 struct tcp6_pseudohdr { 1723 struct in6_addr saddr; 1724 struct in6_addr daddr; 1725 __be32 len; 1726 __be32 protocol; /* including padding */ 1727 }; 1728 1729 union tcp_md5sum_block { 1730 struct tcp4_pseudohdr ip4; 1731 #if IS_ENABLED(CONFIG_IPV6) 1732 struct tcp6_pseudohdr ip6; 1733 #endif 1734 }; 1735 1736 /* 1737 * struct tcp_sigpool - per-CPU pool of ahash_requests 1738 * @scratch: per-CPU temporary area, that can be used between 1739 * tcp_sigpool_start() and tcp_sigpool_end() to perform 1740 * crypto request 1741 * @req: pre-allocated ahash request 1742 */ 1743 struct tcp_sigpool { 1744 void *scratch; 1745 struct ahash_request *req; 1746 }; 1747 1748 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size); 1749 void tcp_sigpool_get(unsigned int id); 1750 void tcp_sigpool_release(unsigned int id); 1751 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp, 1752 const struct sk_buff *skb, 1753 unsigned int header_len); 1754 1755 /** 1756 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash 1757 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash() 1758 * @c: returned tcp_sigpool for usage (uninitialized on failure) 1759 * 1760 * Returns 0 on success, error otherwise. 1761 */ 1762 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c); 1763 /** 1764 * tcp_sigpool_end - enable bh and stop using tcp_sigpool 1765 * @c: tcp_sigpool context that was returned by tcp_sigpool_start() 1766 */ 1767 void tcp_sigpool_end(struct tcp_sigpool *c); 1768 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len); 1769 /* - functions */ 1770 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1771 const struct sock *sk, const struct sk_buff *skb); 1772 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1773 int family, u8 prefixlen, int l3index, u8 flags, 1774 const u8 *newkey, u8 newkeylen); 1775 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1776 int family, u8 prefixlen, int l3index, 1777 struct tcp_md5sig_key *key); 1778 1779 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1780 int family, u8 prefixlen, int l3index, u8 flags); 1781 void tcp_clear_md5_list(struct sock *sk); 1782 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1783 const struct sock *addr_sk); 1784 1785 #ifdef CONFIG_TCP_MD5SIG 1786 #include <linux/jump_label.h> 1787 extern struct static_key_false_deferred tcp_md5_needed; 1788 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1789 const union tcp_md5_addr *addr, 1790 int family, bool any_l3index); 1791 static inline struct tcp_md5sig_key * 1792 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1793 const union tcp_md5_addr *addr, int family) 1794 { 1795 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1796 return NULL; 1797 return __tcp_md5_do_lookup(sk, l3index, addr, family, false); 1798 } 1799 1800 static inline struct tcp_md5sig_key * 1801 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1802 const union tcp_md5_addr *addr, int family) 1803 { 1804 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1805 return NULL; 1806 return __tcp_md5_do_lookup(sk, 0, addr, family, true); 1807 } 1808 1809 enum skb_drop_reason 1810 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1811 const void *saddr, const void *daddr, 1812 int family, int l3index, const __u8 *hash_location); 1813 1814 1815 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1816 #else 1817 static inline struct tcp_md5sig_key * 1818 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1819 const union tcp_md5_addr *addr, int family) 1820 { 1821 return NULL; 1822 } 1823 1824 static inline struct tcp_md5sig_key * 1825 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1826 const union tcp_md5_addr *addr, int family) 1827 { 1828 return NULL; 1829 } 1830 1831 static inline enum skb_drop_reason 1832 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1833 const void *saddr, const void *daddr, 1834 int family, int l3index, const __u8 *hash_location) 1835 { 1836 return SKB_NOT_DROPPED_YET; 1837 } 1838 #define tcp_twsk_md5_key(twsk) NULL 1839 #endif 1840 1841 int tcp_md5_alloc_sigpool(void); 1842 void tcp_md5_release_sigpool(void); 1843 void tcp_md5_add_sigpool(void); 1844 extern int tcp_md5_sigpool_id; 1845 1846 int tcp_md5_hash_key(struct tcp_sigpool *hp, 1847 const struct tcp_md5sig_key *key); 1848 1849 /* From tcp_fastopen.c */ 1850 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1851 struct tcp_fastopen_cookie *cookie); 1852 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1853 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1854 u16 try_exp); 1855 struct tcp_fastopen_request { 1856 /* Fast Open cookie. Size 0 means a cookie request */ 1857 struct tcp_fastopen_cookie cookie; 1858 struct msghdr *data; /* data in MSG_FASTOPEN */ 1859 size_t size; 1860 int copied; /* queued in tcp_connect() */ 1861 struct ubuf_info *uarg; 1862 }; 1863 void tcp_free_fastopen_req(struct tcp_sock *tp); 1864 void tcp_fastopen_destroy_cipher(struct sock *sk); 1865 void tcp_fastopen_ctx_destroy(struct net *net); 1866 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1867 void *primary_key, void *backup_key); 1868 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1869 u64 *key); 1870 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1871 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1872 struct request_sock *req, 1873 struct tcp_fastopen_cookie *foc, 1874 const struct dst_entry *dst); 1875 void tcp_fastopen_init_key_once(struct net *net); 1876 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1877 struct tcp_fastopen_cookie *cookie); 1878 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1879 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1880 #define TCP_FASTOPEN_KEY_MAX 2 1881 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1882 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1883 1884 /* Fastopen key context */ 1885 struct tcp_fastopen_context { 1886 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1887 int num; 1888 struct rcu_head rcu; 1889 }; 1890 1891 void tcp_fastopen_active_disable(struct sock *sk); 1892 bool tcp_fastopen_active_should_disable(struct sock *sk); 1893 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1894 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1895 1896 /* Caller needs to wrap with rcu_read_(un)lock() */ 1897 static inline 1898 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1899 { 1900 struct tcp_fastopen_context *ctx; 1901 1902 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1903 if (!ctx) 1904 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1905 return ctx; 1906 } 1907 1908 static inline 1909 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1910 const struct tcp_fastopen_cookie *orig) 1911 { 1912 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1913 orig->len == foc->len && 1914 !memcmp(orig->val, foc->val, foc->len)) 1915 return true; 1916 return false; 1917 } 1918 1919 static inline 1920 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1921 { 1922 return ctx->num; 1923 } 1924 1925 /* Latencies incurred by various limits for a sender. They are 1926 * chronograph-like stats that are mutually exclusive. 1927 */ 1928 enum tcp_chrono { 1929 TCP_CHRONO_UNSPEC, 1930 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1931 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1932 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1933 __TCP_CHRONO_MAX, 1934 }; 1935 1936 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1937 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1938 1939 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1940 * the same memory storage than skb->destructor/_skb_refdst 1941 */ 1942 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1943 { 1944 skb->destructor = NULL; 1945 skb->_skb_refdst = 0UL; 1946 } 1947 1948 #define tcp_skb_tsorted_save(skb) { \ 1949 unsigned long _save = skb->_skb_refdst; \ 1950 skb->_skb_refdst = 0UL; 1951 1952 #define tcp_skb_tsorted_restore(skb) \ 1953 skb->_skb_refdst = _save; \ 1954 } 1955 1956 void tcp_write_queue_purge(struct sock *sk); 1957 1958 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1959 { 1960 return skb_rb_first(&sk->tcp_rtx_queue); 1961 } 1962 1963 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1964 { 1965 return skb_rb_last(&sk->tcp_rtx_queue); 1966 } 1967 1968 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1969 { 1970 return skb_peek_tail(&sk->sk_write_queue); 1971 } 1972 1973 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1974 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1975 1976 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1977 { 1978 return skb_peek(&sk->sk_write_queue); 1979 } 1980 1981 static inline bool tcp_skb_is_last(const struct sock *sk, 1982 const struct sk_buff *skb) 1983 { 1984 return skb_queue_is_last(&sk->sk_write_queue, skb); 1985 } 1986 1987 /** 1988 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1989 * @sk: socket 1990 * 1991 * Since the write queue can have a temporary empty skb in it, 1992 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1993 */ 1994 static inline bool tcp_write_queue_empty(const struct sock *sk) 1995 { 1996 const struct tcp_sock *tp = tcp_sk(sk); 1997 1998 return tp->write_seq == tp->snd_nxt; 1999 } 2000 2001 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 2002 { 2003 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 2004 } 2005 2006 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 2007 { 2008 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 2009 } 2010 2011 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 2012 { 2013 __skb_queue_tail(&sk->sk_write_queue, skb); 2014 2015 /* Queue it, remembering where we must start sending. */ 2016 if (sk->sk_write_queue.next == skb) 2017 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 2018 } 2019 2020 /* Insert new before skb on the write queue of sk. */ 2021 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 2022 struct sk_buff *skb, 2023 struct sock *sk) 2024 { 2025 __skb_queue_before(&sk->sk_write_queue, skb, new); 2026 } 2027 2028 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 2029 { 2030 tcp_skb_tsorted_anchor_cleanup(skb); 2031 __skb_unlink(skb, &sk->sk_write_queue); 2032 } 2033 2034 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 2035 2036 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2037 { 2038 tcp_skb_tsorted_anchor_cleanup(skb); 2039 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2040 } 2041 2042 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2043 { 2044 list_del(&skb->tcp_tsorted_anchor); 2045 tcp_rtx_queue_unlink(skb, sk); 2046 tcp_wmem_free_skb(sk, skb); 2047 } 2048 2049 static inline void tcp_push_pending_frames(struct sock *sk) 2050 { 2051 if (tcp_send_head(sk)) { 2052 struct tcp_sock *tp = tcp_sk(sk); 2053 2054 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2055 } 2056 } 2057 2058 /* Start sequence of the skb just after the highest skb with SACKed 2059 * bit, valid only if sacked_out > 0 or when the caller has ensured 2060 * validity by itself. 2061 */ 2062 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2063 { 2064 if (!tp->sacked_out) 2065 return tp->snd_una; 2066 2067 if (tp->highest_sack == NULL) 2068 return tp->snd_nxt; 2069 2070 return TCP_SKB_CB(tp->highest_sack)->seq; 2071 } 2072 2073 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2074 { 2075 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2076 } 2077 2078 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2079 { 2080 return tcp_sk(sk)->highest_sack; 2081 } 2082 2083 static inline void tcp_highest_sack_reset(struct sock *sk) 2084 { 2085 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2086 } 2087 2088 /* Called when old skb is about to be deleted and replaced by new skb */ 2089 static inline void tcp_highest_sack_replace(struct sock *sk, 2090 struct sk_buff *old, 2091 struct sk_buff *new) 2092 { 2093 if (old == tcp_highest_sack(sk)) 2094 tcp_sk(sk)->highest_sack = new; 2095 } 2096 2097 /* This helper checks if socket has IP_TRANSPARENT set */ 2098 static inline bool inet_sk_transparent(const struct sock *sk) 2099 { 2100 switch (sk->sk_state) { 2101 case TCP_TIME_WAIT: 2102 return inet_twsk(sk)->tw_transparent; 2103 case TCP_NEW_SYN_RECV: 2104 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2105 } 2106 return inet_test_bit(TRANSPARENT, sk); 2107 } 2108 2109 /* Determines whether this is a thin stream (which may suffer from 2110 * increased latency). Used to trigger latency-reducing mechanisms. 2111 */ 2112 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2113 { 2114 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2115 } 2116 2117 /* /proc */ 2118 enum tcp_seq_states { 2119 TCP_SEQ_STATE_LISTENING, 2120 TCP_SEQ_STATE_ESTABLISHED, 2121 }; 2122 2123 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2124 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2125 void tcp_seq_stop(struct seq_file *seq, void *v); 2126 2127 struct tcp_seq_afinfo { 2128 sa_family_t family; 2129 }; 2130 2131 struct tcp_iter_state { 2132 struct seq_net_private p; 2133 enum tcp_seq_states state; 2134 struct sock *syn_wait_sk; 2135 int bucket, offset, sbucket, num; 2136 loff_t last_pos; 2137 }; 2138 2139 extern struct request_sock_ops tcp_request_sock_ops; 2140 extern struct request_sock_ops tcp6_request_sock_ops; 2141 2142 void tcp_v4_destroy_sock(struct sock *sk); 2143 2144 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2145 netdev_features_t features); 2146 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2147 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2148 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2149 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2150 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2151 #ifdef CONFIG_INET 2152 void tcp_gro_complete(struct sk_buff *skb); 2153 #else 2154 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2155 #endif 2156 2157 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2158 2159 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2160 { 2161 struct net *net = sock_net((struct sock *)tp); 2162 u32 val; 2163 2164 val = READ_ONCE(tp->notsent_lowat); 2165 2166 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2167 } 2168 2169 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2170 2171 #ifdef CONFIG_PROC_FS 2172 int tcp4_proc_init(void); 2173 void tcp4_proc_exit(void); 2174 #endif 2175 2176 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2177 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2178 const struct tcp_request_sock_ops *af_ops, 2179 struct sock *sk, struct sk_buff *skb); 2180 2181 /* TCP af-specific functions */ 2182 struct tcp_sock_af_ops { 2183 #ifdef CONFIG_TCP_MD5SIG 2184 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2185 const struct sock *addr_sk); 2186 int (*calc_md5_hash)(char *location, 2187 const struct tcp_md5sig_key *md5, 2188 const struct sock *sk, 2189 const struct sk_buff *skb); 2190 int (*md5_parse)(struct sock *sk, 2191 int optname, 2192 sockptr_t optval, 2193 int optlen); 2194 #endif 2195 #ifdef CONFIG_TCP_AO 2196 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen); 2197 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2198 struct sock *addr_sk, 2199 int sndid, int rcvid); 2200 int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key, 2201 const struct sock *sk, 2202 __be32 sisn, __be32 disn, bool send); 2203 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao, 2204 const struct sock *sk, const struct sk_buff *skb, 2205 const u8 *tkey, int hash_offset, u32 sne); 2206 #endif 2207 }; 2208 2209 struct tcp_request_sock_ops { 2210 u16 mss_clamp; 2211 #ifdef CONFIG_TCP_MD5SIG 2212 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2213 const struct sock *addr_sk); 2214 int (*calc_md5_hash) (char *location, 2215 const struct tcp_md5sig_key *md5, 2216 const struct sock *sk, 2217 const struct sk_buff *skb); 2218 #endif 2219 #ifdef CONFIG_TCP_AO 2220 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2221 struct request_sock *req, 2222 int sndid, int rcvid); 2223 int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk); 2224 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt, 2225 struct request_sock *req, const struct sk_buff *skb, 2226 int hash_offset, u32 sne); 2227 #endif 2228 #ifdef CONFIG_SYN_COOKIES 2229 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2230 __u16 *mss); 2231 #endif 2232 struct dst_entry *(*route_req)(const struct sock *sk, 2233 struct sk_buff *skb, 2234 struct flowi *fl, 2235 struct request_sock *req); 2236 u32 (*init_seq)(const struct sk_buff *skb); 2237 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2238 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2239 struct flowi *fl, struct request_sock *req, 2240 struct tcp_fastopen_cookie *foc, 2241 enum tcp_synack_type synack_type, 2242 struct sk_buff *syn_skb); 2243 }; 2244 2245 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2246 #if IS_ENABLED(CONFIG_IPV6) 2247 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2248 #endif 2249 2250 #ifdef CONFIG_SYN_COOKIES 2251 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2252 const struct sock *sk, struct sk_buff *skb, 2253 __u16 *mss) 2254 { 2255 tcp_synq_overflow(sk); 2256 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2257 return ops->cookie_init_seq(skb, mss); 2258 } 2259 #else 2260 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2261 const struct sock *sk, struct sk_buff *skb, 2262 __u16 *mss) 2263 { 2264 return 0; 2265 } 2266 #endif 2267 2268 struct tcp_key { 2269 union { 2270 struct { 2271 struct tcp_ao_key *ao_key; 2272 char *traffic_key; 2273 u32 sne; 2274 u8 rcv_next; 2275 }; 2276 struct tcp_md5sig_key *md5_key; 2277 }; 2278 enum { 2279 TCP_KEY_NONE = 0, 2280 TCP_KEY_MD5, 2281 TCP_KEY_AO, 2282 } type; 2283 }; 2284 2285 static inline void tcp_get_current_key(const struct sock *sk, 2286 struct tcp_key *out) 2287 { 2288 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG) 2289 const struct tcp_sock *tp = tcp_sk(sk); 2290 #endif 2291 2292 #ifdef CONFIG_TCP_AO 2293 if (static_branch_unlikely(&tcp_ao_needed.key)) { 2294 struct tcp_ao_info *ao; 2295 2296 ao = rcu_dereference_protected(tp->ao_info, 2297 lockdep_sock_is_held(sk)); 2298 if (ao) { 2299 out->ao_key = READ_ONCE(ao->current_key); 2300 out->type = TCP_KEY_AO; 2301 return; 2302 } 2303 } 2304 #endif 2305 #ifdef CONFIG_TCP_MD5SIG 2306 if (static_branch_unlikely(&tcp_md5_needed.key) && 2307 rcu_access_pointer(tp->md5sig_info)) { 2308 out->md5_key = tp->af_specific->md5_lookup(sk, sk); 2309 if (out->md5_key) { 2310 out->type = TCP_KEY_MD5; 2311 return; 2312 } 2313 } 2314 #endif 2315 out->type = TCP_KEY_NONE; 2316 } 2317 2318 static inline bool tcp_key_is_md5(const struct tcp_key *key) 2319 { 2320 #ifdef CONFIG_TCP_MD5SIG 2321 if (static_branch_unlikely(&tcp_md5_needed.key) && 2322 key->type == TCP_KEY_MD5) 2323 return true; 2324 #endif 2325 return false; 2326 } 2327 2328 static inline bool tcp_key_is_ao(const struct tcp_key *key) 2329 { 2330 #ifdef CONFIG_TCP_AO 2331 if (static_branch_unlikely(&tcp_ao_needed.key) && 2332 key->type == TCP_KEY_AO) 2333 return true; 2334 #endif 2335 return false; 2336 } 2337 2338 int tcpv4_offload_init(void); 2339 2340 void tcp_v4_init(void); 2341 void tcp_init(void); 2342 2343 /* tcp_recovery.c */ 2344 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2345 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2346 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2347 u32 reo_wnd); 2348 extern bool tcp_rack_mark_lost(struct sock *sk); 2349 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2350 u64 xmit_time); 2351 extern void tcp_rack_reo_timeout(struct sock *sk); 2352 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2353 2354 /* tcp_plb.c */ 2355 2356 /* 2357 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2358 * expects cong_ratio which represents fraction of traffic that experienced 2359 * congestion over a single RTT. In order to avoid floating point operations, 2360 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2361 */ 2362 #define TCP_PLB_SCALE 8 2363 2364 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2365 struct tcp_plb_state { 2366 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2367 unused:3; 2368 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2369 }; 2370 2371 static inline void tcp_plb_init(const struct sock *sk, 2372 struct tcp_plb_state *plb) 2373 { 2374 plb->consec_cong_rounds = 0; 2375 plb->pause_until = 0; 2376 } 2377 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2378 const int cong_ratio); 2379 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2380 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2381 2382 /* At how many usecs into the future should the RTO fire? */ 2383 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2384 { 2385 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2386 u32 rto = inet_csk(sk)->icsk_rto; 2387 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2388 2389 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2390 } 2391 2392 /* 2393 * Save and compile IPv4 options, return a pointer to it 2394 */ 2395 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2396 struct sk_buff *skb) 2397 { 2398 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2399 struct ip_options_rcu *dopt = NULL; 2400 2401 if (opt->optlen) { 2402 int opt_size = sizeof(*dopt) + opt->optlen; 2403 2404 dopt = kmalloc(opt_size, GFP_ATOMIC); 2405 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2406 kfree(dopt); 2407 dopt = NULL; 2408 } 2409 } 2410 return dopt; 2411 } 2412 2413 /* locally generated TCP pure ACKs have skb->truesize == 2 2414 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2415 * This is much faster than dissecting the packet to find out. 2416 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2417 */ 2418 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2419 { 2420 return skb->truesize == 2; 2421 } 2422 2423 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2424 { 2425 skb->truesize = 2; 2426 } 2427 2428 static inline int tcp_inq(struct sock *sk) 2429 { 2430 struct tcp_sock *tp = tcp_sk(sk); 2431 int answ; 2432 2433 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2434 answ = 0; 2435 } else if (sock_flag(sk, SOCK_URGINLINE) || 2436 !tp->urg_data || 2437 before(tp->urg_seq, tp->copied_seq) || 2438 !before(tp->urg_seq, tp->rcv_nxt)) { 2439 2440 answ = tp->rcv_nxt - tp->copied_seq; 2441 2442 /* Subtract 1, if FIN was received */ 2443 if (answ && sock_flag(sk, SOCK_DONE)) 2444 answ--; 2445 } else { 2446 answ = tp->urg_seq - tp->copied_seq; 2447 } 2448 2449 return answ; 2450 } 2451 2452 int tcp_peek_len(struct socket *sock); 2453 2454 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2455 { 2456 u16 segs_in; 2457 2458 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2459 2460 /* We update these fields while other threads might 2461 * read them from tcp_get_info() 2462 */ 2463 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2464 if (skb->len > tcp_hdrlen(skb)) 2465 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2466 } 2467 2468 /* 2469 * TCP listen path runs lockless. 2470 * We forced "struct sock" to be const qualified to make sure 2471 * we don't modify one of its field by mistake. 2472 * Here, we increment sk_drops which is an atomic_t, so we can safely 2473 * make sock writable again. 2474 */ 2475 static inline void tcp_listendrop(const struct sock *sk) 2476 { 2477 atomic_inc(&((struct sock *)sk)->sk_drops); 2478 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2479 } 2480 2481 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2482 2483 /* 2484 * Interface for adding Upper Level Protocols over TCP 2485 */ 2486 2487 #define TCP_ULP_NAME_MAX 16 2488 #define TCP_ULP_MAX 128 2489 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2490 2491 struct tcp_ulp_ops { 2492 struct list_head list; 2493 2494 /* initialize ulp */ 2495 int (*init)(struct sock *sk); 2496 /* update ulp */ 2497 void (*update)(struct sock *sk, struct proto *p, 2498 void (*write_space)(struct sock *sk)); 2499 /* cleanup ulp */ 2500 void (*release)(struct sock *sk); 2501 /* diagnostic */ 2502 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2503 size_t (*get_info_size)(const struct sock *sk); 2504 /* clone ulp */ 2505 void (*clone)(const struct request_sock *req, struct sock *newsk, 2506 const gfp_t priority); 2507 2508 char name[TCP_ULP_NAME_MAX]; 2509 struct module *owner; 2510 }; 2511 int tcp_register_ulp(struct tcp_ulp_ops *type); 2512 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2513 int tcp_set_ulp(struct sock *sk, const char *name); 2514 void tcp_get_available_ulp(char *buf, size_t len); 2515 void tcp_cleanup_ulp(struct sock *sk); 2516 void tcp_update_ulp(struct sock *sk, struct proto *p, 2517 void (*write_space)(struct sock *sk)); 2518 2519 #define MODULE_ALIAS_TCP_ULP(name) \ 2520 __MODULE_INFO(alias, alias_userspace, name); \ 2521 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2522 2523 #ifdef CONFIG_NET_SOCK_MSG 2524 struct sk_msg; 2525 struct sk_psock; 2526 2527 #ifdef CONFIG_BPF_SYSCALL 2528 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2529 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2530 #endif /* CONFIG_BPF_SYSCALL */ 2531 2532 #ifdef CONFIG_INET 2533 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2534 #else 2535 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2536 { 2537 } 2538 #endif 2539 2540 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2541 struct sk_msg *msg, u32 bytes, int flags); 2542 #endif /* CONFIG_NET_SOCK_MSG */ 2543 2544 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2545 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2546 { 2547 } 2548 #endif 2549 2550 #ifdef CONFIG_CGROUP_BPF 2551 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2552 struct sk_buff *skb, 2553 unsigned int end_offset) 2554 { 2555 skops->skb = skb; 2556 skops->skb_data_end = skb->data + end_offset; 2557 } 2558 #else 2559 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2560 struct sk_buff *skb, 2561 unsigned int end_offset) 2562 { 2563 } 2564 #endif 2565 2566 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2567 * is < 0, then the BPF op failed (for example if the loaded BPF 2568 * program does not support the chosen operation or there is no BPF 2569 * program loaded). 2570 */ 2571 #ifdef CONFIG_BPF 2572 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2573 { 2574 struct bpf_sock_ops_kern sock_ops; 2575 int ret; 2576 2577 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2578 if (sk_fullsock(sk)) { 2579 sock_ops.is_fullsock = 1; 2580 sock_owned_by_me(sk); 2581 } 2582 2583 sock_ops.sk = sk; 2584 sock_ops.op = op; 2585 if (nargs > 0) 2586 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2587 2588 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2589 if (ret == 0) 2590 ret = sock_ops.reply; 2591 else 2592 ret = -1; 2593 return ret; 2594 } 2595 2596 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2597 { 2598 u32 args[2] = {arg1, arg2}; 2599 2600 return tcp_call_bpf(sk, op, 2, args); 2601 } 2602 2603 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2604 u32 arg3) 2605 { 2606 u32 args[3] = {arg1, arg2, arg3}; 2607 2608 return tcp_call_bpf(sk, op, 3, args); 2609 } 2610 2611 #else 2612 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2613 { 2614 return -EPERM; 2615 } 2616 2617 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2618 { 2619 return -EPERM; 2620 } 2621 2622 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2623 u32 arg3) 2624 { 2625 return -EPERM; 2626 } 2627 2628 #endif 2629 2630 static inline u32 tcp_timeout_init(struct sock *sk) 2631 { 2632 int timeout; 2633 2634 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2635 2636 if (timeout <= 0) 2637 timeout = TCP_TIMEOUT_INIT; 2638 return min_t(int, timeout, TCP_RTO_MAX); 2639 } 2640 2641 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2642 { 2643 int rwnd; 2644 2645 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2646 2647 if (rwnd < 0) 2648 rwnd = 0; 2649 return rwnd; 2650 } 2651 2652 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2653 { 2654 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2655 } 2656 2657 static inline void tcp_bpf_rtt(struct sock *sk) 2658 { 2659 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2660 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2661 } 2662 2663 #if IS_ENABLED(CONFIG_SMC) 2664 extern struct static_key_false tcp_have_smc; 2665 #endif 2666 2667 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2668 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2669 void (*cad)(struct sock *sk, u32 ack_seq)); 2670 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2671 void clean_acked_data_flush(void); 2672 #endif 2673 2674 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2675 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2676 const struct tcp_sock *tp) 2677 { 2678 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2679 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2680 } 2681 2682 /* Compute Earliest Departure Time for some control packets 2683 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2684 */ 2685 static inline u64 tcp_transmit_time(const struct sock *sk) 2686 { 2687 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2688 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2689 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2690 2691 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2692 } 2693 return 0; 2694 } 2695 2696 static inline int tcp_parse_auth_options(const struct tcphdr *th, 2697 const u8 **md5_hash, const struct tcp_ao_hdr **aoh) 2698 { 2699 const u8 *md5_tmp, *ao_tmp; 2700 int ret; 2701 2702 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp); 2703 if (ret) 2704 return ret; 2705 2706 if (md5_hash) 2707 *md5_hash = md5_tmp; 2708 2709 if (aoh) { 2710 if (!ao_tmp) 2711 *aoh = NULL; 2712 else 2713 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2); 2714 } 2715 2716 return 0; 2717 } 2718 2719 static inline bool tcp_ao_required(struct sock *sk, const void *saddr, 2720 int family, int l3index, bool stat_inc) 2721 { 2722 #ifdef CONFIG_TCP_AO 2723 struct tcp_ao_info *ao_info; 2724 struct tcp_ao_key *ao_key; 2725 2726 if (!static_branch_unlikely(&tcp_ao_needed.key)) 2727 return false; 2728 2729 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info, 2730 lockdep_sock_is_held(sk)); 2731 if (!ao_info) 2732 return false; 2733 2734 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1); 2735 if (ao_info->ao_required || ao_key) { 2736 if (stat_inc) { 2737 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED); 2738 atomic64_inc(&ao_info->counters.ao_required); 2739 } 2740 return true; 2741 } 2742 #endif 2743 return false; 2744 } 2745 2746 /* Called with rcu_read_lock() */ 2747 static inline enum skb_drop_reason 2748 tcp_inbound_hash(struct sock *sk, const struct request_sock *req, 2749 const struct sk_buff *skb, 2750 const void *saddr, const void *daddr, 2751 int family, int dif, int sdif) 2752 { 2753 const struct tcphdr *th = tcp_hdr(skb); 2754 const struct tcp_ao_hdr *aoh; 2755 const __u8 *md5_location; 2756 int l3index; 2757 2758 /* Invalid option or two times meet any of auth options */ 2759 if (tcp_parse_auth_options(th, &md5_location, &aoh)) { 2760 tcp_hash_fail("TCP segment has incorrect auth options set", 2761 family, skb, ""); 2762 return SKB_DROP_REASON_TCP_AUTH_HDR; 2763 } 2764 2765 if (req) { 2766 if (tcp_rsk_used_ao(req) != !!aoh) { 2767 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD); 2768 tcp_hash_fail("TCP connection can't start/end using TCP-AO", 2769 family, skb, "%s", 2770 !aoh ? "missing AO" : "AO signed"); 2771 return SKB_DROP_REASON_TCP_AOFAILURE; 2772 } 2773 } 2774 2775 /* sdif set, means packet ingressed via a device 2776 * in an L3 domain and dif is set to the l3mdev 2777 */ 2778 l3index = sdif ? dif : 0; 2779 2780 /* Fast path: unsigned segments */ 2781 if (likely(!md5_location && !aoh)) { 2782 /* Drop if there's TCP-MD5 or TCP-AO key with any rcvid/sndid 2783 * for the remote peer. On TCP-AO established connection 2784 * the last key is impossible to remove, so there's 2785 * always at least one current_key. 2786 */ 2787 if (tcp_ao_required(sk, saddr, family, l3index, true)) { 2788 tcp_hash_fail("AO hash is required, but not found", 2789 family, skb, "L3 index %d", l3index); 2790 return SKB_DROP_REASON_TCP_AONOTFOUND; 2791 } 2792 if (unlikely(tcp_md5_do_lookup(sk, l3index, saddr, family))) { 2793 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 2794 tcp_hash_fail("MD5 Hash not found", 2795 family, skb, "L3 index %d", l3index); 2796 return SKB_DROP_REASON_TCP_MD5NOTFOUND; 2797 } 2798 return SKB_NOT_DROPPED_YET; 2799 } 2800 2801 if (aoh) 2802 return tcp_inbound_ao_hash(sk, skb, family, req, l3index, aoh); 2803 2804 return tcp_inbound_md5_hash(sk, skb, saddr, daddr, family, 2805 l3index, md5_location); 2806 } 2807 2808 #endif /* _TCP_H */ 2809