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); 494 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th); 495 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 496 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 497 struct sock *sk, struct sk_buff *skb, 498 struct tcp_options_received *tcp_opt, 499 int mss, u32 tsoff); 500 501 #ifdef CONFIG_SYN_COOKIES 502 503 /* Syncookies use a monotonic timer which increments every 60 seconds. 504 * This counter is used both as a hash input and partially encoded into 505 * the cookie value. A cookie is only validated further if the delta 506 * between the current counter value and the encoded one is less than this, 507 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 508 * the counter advances immediately after a cookie is generated). 509 */ 510 #define MAX_SYNCOOKIE_AGE 2 511 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 512 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 513 514 /* syncookies: remember time of last synqueue overflow 515 * But do not dirty this field too often (once per second is enough) 516 * It is racy as we do not hold a lock, but race is very minor. 517 */ 518 static inline void tcp_synq_overflow(const struct sock *sk) 519 { 520 unsigned int last_overflow; 521 unsigned int now = jiffies; 522 523 if (sk->sk_reuseport) { 524 struct sock_reuseport *reuse; 525 526 reuse = rcu_dereference(sk->sk_reuseport_cb); 527 if (likely(reuse)) { 528 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 529 if (!time_between32(now, last_overflow, 530 last_overflow + HZ)) 531 WRITE_ONCE(reuse->synq_overflow_ts, now); 532 return; 533 } 534 } 535 536 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 537 if (!time_between32(now, last_overflow, last_overflow + HZ)) 538 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 539 } 540 541 /* syncookies: no recent synqueue overflow on this listening socket? */ 542 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 543 { 544 unsigned int last_overflow; 545 unsigned int now = jiffies; 546 547 if (sk->sk_reuseport) { 548 struct sock_reuseport *reuse; 549 550 reuse = rcu_dereference(sk->sk_reuseport_cb); 551 if (likely(reuse)) { 552 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 553 return !time_between32(now, last_overflow - HZ, 554 last_overflow + 555 TCP_SYNCOOKIE_VALID); 556 } 557 } 558 559 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 560 561 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 562 * then we're under synflood. However, we have to use 563 * 'last_overflow - HZ' as lower bound. That's because a concurrent 564 * tcp_synq_overflow() could update .ts_recent_stamp after we read 565 * jiffies but before we store .ts_recent_stamp into last_overflow, 566 * which could lead to rejecting a valid syncookie. 567 */ 568 return !time_between32(now, last_overflow - HZ, 569 last_overflow + TCP_SYNCOOKIE_VALID); 570 } 571 572 static inline u32 tcp_cookie_time(void) 573 { 574 u64 val = get_jiffies_64(); 575 576 do_div(val, TCP_SYNCOOKIE_PERIOD); 577 return val; 578 } 579 580 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 581 u16 *mssp); 582 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 583 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 584 bool cookie_timestamp_decode(const struct net *net, 585 struct tcp_options_received *opt); 586 587 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst) 588 { 589 return READ_ONCE(net->ipv4.sysctl_tcp_ecn) || 590 dst_feature(dst, RTAX_FEATURE_ECN); 591 } 592 593 /* From net/ipv6/syncookies.c */ 594 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th); 595 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 596 597 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 598 const struct tcphdr *th, u16 *mssp); 599 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 600 #endif 601 /* tcp_output.c */ 602 603 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 604 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 605 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 606 int nonagle); 607 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 608 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 609 void tcp_retransmit_timer(struct sock *sk); 610 void tcp_xmit_retransmit_queue(struct sock *); 611 void tcp_simple_retransmit(struct sock *); 612 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 613 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 614 enum tcp_queue { 615 TCP_FRAG_IN_WRITE_QUEUE, 616 TCP_FRAG_IN_RTX_QUEUE, 617 }; 618 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 619 struct sk_buff *skb, u32 len, 620 unsigned int mss_now, gfp_t gfp); 621 622 void tcp_send_probe0(struct sock *); 623 int tcp_write_wakeup(struct sock *, int mib); 624 void tcp_send_fin(struct sock *sk); 625 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 626 int tcp_send_synack(struct sock *); 627 void tcp_push_one(struct sock *, unsigned int mss_now); 628 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 629 void tcp_send_ack(struct sock *sk); 630 void tcp_send_delayed_ack(struct sock *sk); 631 void tcp_send_loss_probe(struct sock *sk); 632 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 633 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 634 const struct sk_buff *next_skb); 635 636 /* tcp_input.c */ 637 void tcp_rearm_rto(struct sock *sk); 638 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 639 void tcp_reset(struct sock *sk, struct sk_buff *skb); 640 void tcp_fin(struct sock *sk); 641 void tcp_check_space(struct sock *sk); 642 void tcp_sack_compress_send_ack(struct sock *sk); 643 644 /* tcp_timer.c */ 645 void tcp_init_xmit_timers(struct sock *); 646 static inline void tcp_clear_xmit_timers(struct sock *sk) 647 { 648 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 649 __sock_put(sk); 650 651 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 652 __sock_put(sk); 653 654 inet_csk_clear_xmit_timers(sk); 655 } 656 657 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 658 unsigned int tcp_current_mss(struct sock *sk); 659 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 660 661 /* Bound MSS / TSO packet size with the half of the window */ 662 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 663 { 664 int cutoff; 665 666 /* When peer uses tiny windows, there is no use in packetizing 667 * to sub-MSS pieces for the sake of SWS or making sure there 668 * are enough packets in the pipe for fast recovery. 669 * 670 * On the other hand, for extremely large MSS devices, handling 671 * smaller than MSS windows in this way does make sense. 672 */ 673 if (tp->max_window > TCP_MSS_DEFAULT) 674 cutoff = (tp->max_window >> 1); 675 else 676 cutoff = tp->max_window; 677 678 if (cutoff && pktsize > cutoff) 679 return max_t(int, cutoff, 68U - tp->tcp_header_len); 680 else 681 return pktsize; 682 } 683 684 /* tcp.c */ 685 void tcp_get_info(struct sock *, struct tcp_info *); 686 687 /* Read 'sendfile()'-style from a TCP socket */ 688 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 689 sk_read_actor_t recv_actor); 690 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 691 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 692 void tcp_read_done(struct sock *sk, size_t len); 693 694 void tcp_initialize_rcv_mss(struct sock *sk); 695 696 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 697 int tcp_mss_to_mtu(struct sock *sk, int mss); 698 void tcp_mtup_init(struct sock *sk); 699 700 static inline void tcp_bound_rto(const struct sock *sk) 701 { 702 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 703 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 704 } 705 706 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 707 { 708 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 709 } 710 711 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 712 { 713 /* mptcp hooks are only on the slow path */ 714 if (sk_is_mptcp((struct sock *)tp)) 715 return; 716 717 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 718 ntohl(TCP_FLAG_ACK) | 719 snd_wnd); 720 } 721 722 static inline void tcp_fast_path_on(struct tcp_sock *tp) 723 { 724 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 725 } 726 727 static inline void tcp_fast_path_check(struct sock *sk) 728 { 729 struct tcp_sock *tp = tcp_sk(sk); 730 731 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 732 tp->rcv_wnd && 733 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 734 !tp->urg_data) 735 tcp_fast_path_on(tp); 736 } 737 738 u32 tcp_delack_max(const struct sock *sk); 739 740 /* Compute the actual rto_min value */ 741 static inline u32 tcp_rto_min(const struct sock *sk) 742 { 743 const struct dst_entry *dst = __sk_dst_get(sk); 744 u32 rto_min = inet_csk(sk)->icsk_rto_min; 745 746 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 747 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 748 return rto_min; 749 } 750 751 static inline u32 tcp_rto_min_us(const struct sock *sk) 752 { 753 return jiffies_to_usecs(tcp_rto_min(sk)); 754 } 755 756 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 757 { 758 return dst_metric_locked(dst, RTAX_CC_ALGO); 759 } 760 761 /* Minimum RTT in usec. ~0 means not available. */ 762 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 763 { 764 return minmax_get(&tp->rtt_min); 765 } 766 767 /* Compute the actual receive window we are currently advertising. 768 * Rcv_nxt can be after the window if our peer push more data 769 * than the offered window. 770 */ 771 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 772 { 773 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 774 775 if (win < 0) 776 win = 0; 777 return (u32) win; 778 } 779 780 /* Choose a new window, without checks for shrinking, and without 781 * scaling applied to the result. The caller does these things 782 * if necessary. This is a "raw" window selection. 783 */ 784 u32 __tcp_select_window(struct sock *sk); 785 786 void tcp_send_window_probe(struct sock *sk); 787 788 /* TCP uses 32bit jiffies to save some space. 789 * Note that this is different from tcp_time_stamp, which 790 * historically has been the same until linux-4.13. 791 */ 792 #define tcp_jiffies32 ((u32)jiffies) 793 794 /* 795 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 796 * It is no longer tied to jiffies, but to 1 ms clock. 797 * Note: double check if you want to use tcp_jiffies32 instead of this. 798 */ 799 #define TCP_TS_HZ 1000 800 801 static inline u64 tcp_clock_ns(void) 802 { 803 return ktime_get_ns(); 804 } 805 806 static inline u64 tcp_clock_us(void) 807 { 808 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 809 } 810 811 static inline u64 tcp_clock_ms(void) 812 { 813 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 814 } 815 816 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 817 * or usec resolution. Each socket carries a flag to select one or other 818 * resolution, as the route attribute could change anytime. 819 * Each flow must stick to initial resolution. 820 */ 821 static inline u32 tcp_clock_ts(bool usec_ts) 822 { 823 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 824 } 825 826 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 827 { 828 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 829 } 830 831 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 832 { 833 if (tp->tcp_usec_ts) 834 return tp->tcp_mstamp; 835 return tcp_time_stamp_ms(tp); 836 } 837 838 void tcp_mstamp_refresh(struct tcp_sock *tp); 839 840 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 841 { 842 return max_t(s64, t1 - t0, 0); 843 } 844 845 /* provide the departure time in us unit */ 846 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 847 { 848 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 849 } 850 851 /* Provide skb TSval in usec or ms unit */ 852 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 853 { 854 if (usec_ts) 855 return tcp_skb_timestamp_us(skb); 856 857 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 858 } 859 860 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 861 { 862 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 863 } 864 865 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 866 { 867 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 868 } 869 870 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 871 872 #define TCPHDR_FIN 0x01 873 #define TCPHDR_SYN 0x02 874 #define TCPHDR_RST 0x04 875 #define TCPHDR_PSH 0x08 876 #define TCPHDR_ACK 0x10 877 #define TCPHDR_URG 0x20 878 #define TCPHDR_ECE 0x40 879 #define TCPHDR_CWR 0x80 880 881 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 882 883 /* This is what the send packet queuing engine uses to pass 884 * TCP per-packet control information to the transmission code. 885 * We also store the host-order sequence numbers in here too. 886 * This is 44 bytes if IPV6 is enabled. 887 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 888 */ 889 struct tcp_skb_cb { 890 __u32 seq; /* Starting sequence number */ 891 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 892 union { 893 /* Note : tcp_tw_isn is used in input path only 894 * (isn chosen by tcp_timewait_state_process()) 895 * 896 * tcp_gso_segs/size are used in write queue only, 897 * cf tcp_skb_pcount()/tcp_skb_mss() 898 */ 899 __u32 tcp_tw_isn; 900 struct { 901 u16 tcp_gso_segs; 902 u16 tcp_gso_size; 903 }; 904 }; 905 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 906 907 __u8 sacked; /* State flags for SACK. */ 908 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 909 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 910 #define TCPCB_LOST 0x04 /* SKB is lost */ 911 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 912 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 913 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 914 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 915 TCPCB_REPAIRED) 916 917 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 918 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 919 eor:1, /* Is skb MSG_EOR marked? */ 920 has_rxtstamp:1, /* SKB has a RX timestamp */ 921 unused:5; 922 __u32 ack_seq; /* Sequence number ACK'd */ 923 union { 924 struct { 925 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 926 /* There is space for up to 24 bytes */ 927 __u32 is_app_limited:1, /* cwnd not fully used? */ 928 delivered_ce:20, 929 unused:11; 930 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 931 __u32 delivered; 932 /* start of send pipeline phase */ 933 u64 first_tx_mstamp; 934 /* when we reached the "delivered" count */ 935 u64 delivered_mstamp; 936 } tx; /* only used for outgoing skbs */ 937 union { 938 struct inet_skb_parm h4; 939 #if IS_ENABLED(CONFIG_IPV6) 940 struct inet6_skb_parm h6; 941 #endif 942 } header; /* For incoming skbs */ 943 }; 944 }; 945 946 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 947 948 extern const struct inet_connection_sock_af_ops ipv4_specific; 949 950 #if IS_ENABLED(CONFIG_IPV6) 951 /* This is the variant of inet6_iif() that must be used by TCP, 952 * as TCP moves IP6CB into a different location in skb->cb[] 953 */ 954 static inline int tcp_v6_iif(const struct sk_buff *skb) 955 { 956 return TCP_SKB_CB(skb)->header.h6.iif; 957 } 958 959 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 960 { 961 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 962 963 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 964 } 965 966 /* TCP_SKB_CB reference means this can not be used from early demux */ 967 static inline int tcp_v6_sdif(const struct sk_buff *skb) 968 { 969 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 970 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 971 return TCP_SKB_CB(skb)->header.h6.iif; 972 #endif 973 return 0; 974 } 975 976 extern const struct inet_connection_sock_af_ops ipv6_specific; 977 978 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 979 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 980 void tcp_v6_early_demux(struct sk_buff *skb); 981 982 #endif 983 984 /* TCP_SKB_CB reference means this can not be used from early demux */ 985 static inline int tcp_v4_sdif(struct sk_buff *skb) 986 { 987 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 988 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 989 return TCP_SKB_CB(skb)->header.h4.iif; 990 #endif 991 return 0; 992 } 993 994 /* Due to TSO, an SKB can be composed of multiple actual 995 * packets. To keep these tracked properly, we use this. 996 */ 997 static inline int tcp_skb_pcount(const struct sk_buff *skb) 998 { 999 return TCP_SKB_CB(skb)->tcp_gso_segs; 1000 } 1001 1002 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 1003 { 1004 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1005 } 1006 1007 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1008 { 1009 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1010 } 1011 1012 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1013 static inline int tcp_skb_mss(const struct sk_buff *skb) 1014 { 1015 return TCP_SKB_CB(skb)->tcp_gso_size; 1016 } 1017 1018 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1019 { 1020 return likely(!TCP_SKB_CB(skb)->eor); 1021 } 1022 1023 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1024 const struct sk_buff *from) 1025 { 1026 return likely(tcp_skb_can_collapse_to(to) && 1027 mptcp_skb_can_collapse(to, from) && 1028 skb_pure_zcopy_same(to, from)); 1029 } 1030 1031 /* Events passed to congestion control interface */ 1032 enum tcp_ca_event { 1033 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1034 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1035 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1036 CA_EVENT_LOSS, /* loss timeout */ 1037 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1038 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1039 }; 1040 1041 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1042 enum tcp_ca_ack_event_flags { 1043 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1044 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1045 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1046 }; 1047 1048 /* 1049 * Interface for adding new TCP congestion control handlers 1050 */ 1051 #define TCP_CA_NAME_MAX 16 1052 #define TCP_CA_MAX 128 1053 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1054 1055 #define TCP_CA_UNSPEC 0 1056 1057 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1058 #define TCP_CONG_NON_RESTRICTED 0x1 1059 /* Requires ECN/ECT set on all packets */ 1060 #define TCP_CONG_NEEDS_ECN 0x2 1061 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1062 1063 union tcp_cc_info; 1064 1065 struct ack_sample { 1066 u32 pkts_acked; 1067 s32 rtt_us; 1068 u32 in_flight; 1069 }; 1070 1071 /* A rate sample measures the number of (original/retransmitted) data 1072 * packets delivered "delivered" over an interval of time "interval_us". 1073 * The tcp_rate.c code fills in the rate sample, and congestion 1074 * control modules that define a cong_control function to run at the end 1075 * of ACK processing can optionally chose to consult this sample when 1076 * setting cwnd and pacing rate. 1077 * A sample is invalid if "delivered" or "interval_us" is negative. 1078 */ 1079 struct rate_sample { 1080 u64 prior_mstamp; /* starting timestamp for interval */ 1081 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1082 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1083 s32 delivered; /* number of packets delivered over interval */ 1084 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1085 long interval_us; /* time for tp->delivered to incr "delivered" */ 1086 u32 snd_interval_us; /* snd interval for delivered packets */ 1087 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1088 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1089 int losses; /* number of packets marked lost upon ACK */ 1090 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1091 u32 prior_in_flight; /* in flight before this ACK */ 1092 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1093 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1094 bool is_retrans; /* is sample from retransmission? */ 1095 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1096 }; 1097 1098 struct tcp_congestion_ops { 1099 /* fast path fields are put first to fill one cache line */ 1100 1101 /* return slow start threshold (required) */ 1102 u32 (*ssthresh)(struct sock *sk); 1103 1104 /* do new cwnd calculation (required) */ 1105 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1106 1107 /* call before changing ca_state (optional) */ 1108 void (*set_state)(struct sock *sk, u8 new_state); 1109 1110 /* call when cwnd event occurs (optional) */ 1111 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1112 1113 /* call when ack arrives (optional) */ 1114 void (*in_ack_event)(struct sock *sk, u32 flags); 1115 1116 /* hook for packet ack accounting (optional) */ 1117 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1118 1119 /* override sysctl_tcp_min_tso_segs */ 1120 u32 (*min_tso_segs)(struct sock *sk); 1121 1122 /* call when packets are delivered to update cwnd and pacing rate, 1123 * after all the ca_state processing. (optional) 1124 */ 1125 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1126 1127 1128 /* new value of cwnd after loss (required) */ 1129 u32 (*undo_cwnd)(struct sock *sk); 1130 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1131 u32 (*sndbuf_expand)(struct sock *sk); 1132 1133 /* control/slow paths put last */ 1134 /* get info for inet_diag (optional) */ 1135 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1136 union tcp_cc_info *info); 1137 1138 char name[TCP_CA_NAME_MAX]; 1139 struct module *owner; 1140 struct list_head list; 1141 u32 key; 1142 u32 flags; 1143 1144 /* initialize private data (optional) */ 1145 void (*init)(struct sock *sk); 1146 /* cleanup private data (optional) */ 1147 void (*release)(struct sock *sk); 1148 } ____cacheline_aligned_in_smp; 1149 1150 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1151 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1152 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1153 struct tcp_congestion_ops *old_type); 1154 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1155 1156 void tcp_assign_congestion_control(struct sock *sk); 1157 void tcp_init_congestion_control(struct sock *sk); 1158 void tcp_cleanup_congestion_control(struct sock *sk); 1159 int tcp_set_default_congestion_control(struct net *net, const char *name); 1160 void tcp_get_default_congestion_control(struct net *net, char *name); 1161 void tcp_get_available_congestion_control(char *buf, size_t len); 1162 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1163 int tcp_set_allowed_congestion_control(char *allowed); 1164 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1165 bool cap_net_admin); 1166 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1167 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1168 1169 u32 tcp_reno_ssthresh(struct sock *sk); 1170 u32 tcp_reno_undo_cwnd(struct sock *sk); 1171 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1172 extern struct tcp_congestion_ops tcp_reno; 1173 1174 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1175 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1176 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1177 #ifdef CONFIG_INET 1178 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1179 #else 1180 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1181 { 1182 return NULL; 1183 } 1184 #endif 1185 1186 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1187 { 1188 const struct inet_connection_sock *icsk = inet_csk(sk); 1189 1190 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1191 } 1192 1193 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1194 { 1195 const struct inet_connection_sock *icsk = inet_csk(sk); 1196 1197 if (icsk->icsk_ca_ops->cwnd_event) 1198 icsk->icsk_ca_ops->cwnd_event(sk, event); 1199 } 1200 1201 /* From tcp_cong.c */ 1202 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1203 1204 /* From tcp_rate.c */ 1205 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1206 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1207 struct rate_sample *rs); 1208 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1209 bool is_sack_reneg, struct rate_sample *rs); 1210 void tcp_rate_check_app_limited(struct sock *sk); 1211 1212 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1213 { 1214 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1215 } 1216 1217 /* These functions determine how the current flow behaves in respect of SACK 1218 * handling. SACK is negotiated with the peer, and therefore it can vary 1219 * between different flows. 1220 * 1221 * tcp_is_sack - SACK enabled 1222 * tcp_is_reno - No SACK 1223 */ 1224 static inline int tcp_is_sack(const struct tcp_sock *tp) 1225 { 1226 return likely(tp->rx_opt.sack_ok); 1227 } 1228 1229 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1230 { 1231 return !tcp_is_sack(tp); 1232 } 1233 1234 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1235 { 1236 return tp->sacked_out + tp->lost_out; 1237 } 1238 1239 /* This determines how many packets are "in the network" to the best 1240 * of our knowledge. In many cases it is conservative, but where 1241 * detailed information is available from the receiver (via SACK 1242 * blocks etc.) we can make more aggressive calculations. 1243 * 1244 * Use this for decisions involving congestion control, use just 1245 * tp->packets_out to determine if the send queue is empty or not. 1246 * 1247 * Read this equation as: 1248 * 1249 * "Packets sent once on transmission queue" MINUS 1250 * "Packets left network, but not honestly ACKed yet" PLUS 1251 * "Packets fast retransmitted" 1252 */ 1253 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1254 { 1255 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1256 } 1257 1258 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1259 1260 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1261 { 1262 return tp->snd_cwnd; 1263 } 1264 1265 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1266 { 1267 WARN_ON_ONCE((int)val <= 0); 1268 tp->snd_cwnd = val; 1269 } 1270 1271 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1272 { 1273 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1274 } 1275 1276 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1277 { 1278 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1279 } 1280 1281 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1282 { 1283 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1284 (1 << inet_csk(sk)->icsk_ca_state); 1285 } 1286 1287 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1288 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1289 * ssthresh. 1290 */ 1291 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1292 { 1293 const struct tcp_sock *tp = tcp_sk(sk); 1294 1295 if (tcp_in_cwnd_reduction(sk)) 1296 return tp->snd_ssthresh; 1297 else 1298 return max(tp->snd_ssthresh, 1299 ((tcp_snd_cwnd(tp) >> 1) + 1300 (tcp_snd_cwnd(tp) >> 2))); 1301 } 1302 1303 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1304 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1305 1306 void tcp_enter_cwr(struct sock *sk); 1307 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1308 1309 /* The maximum number of MSS of available cwnd for which TSO defers 1310 * sending if not using sysctl_tcp_tso_win_divisor. 1311 */ 1312 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1313 { 1314 return 3; 1315 } 1316 1317 /* Returns end sequence number of the receiver's advertised window */ 1318 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1319 { 1320 return tp->snd_una + tp->snd_wnd; 1321 } 1322 1323 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1324 * flexible approach. The RFC suggests cwnd should not be raised unless 1325 * it was fully used previously. And that's exactly what we do in 1326 * congestion avoidance mode. But in slow start we allow cwnd to grow 1327 * as long as the application has used half the cwnd. 1328 * Example : 1329 * cwnd is 10 (IW10), but application sends 9 frames. 1330 * We allow cwnd to reach 18 when all frames are ACKed. 1331 * This check is safe because it's as aggressive as slow start which already 1332 * risks 100% overshoot. The advantage is that we discourage application to 1333 * either send more filler packets or data to artificially blow up the cwnd 1334 * usage, and allow application-limited process to probe bw more aggressively. 1335 */ 1336 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1337 { 1338 const struct tcp_sock *tp = tcp_sk(sk); 1339 1340 if (tp->is_cwnd_limited) 1341 return true; 1342 1343 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1344 if (tcp_in_slow_start(tp)) 1345 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1346 1347 return false; 1348 } 1349 1350 /* BBR congestion control needs pacing. 1351 * Same remark for SO_MAX_PACING_RATE. 1352 * sch_fq packet scheduler is efficiently handling pacing, 1353 * but is not always installed/used. 1354 * Return true if TCP stack should pace packets itself. 1355 */ 1356 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1357 { 1358 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1359 } 1360 1361 /* Estimates in how many jiffies next packet for this flow can be sent. 1362 * Scheduling a retransmit timer too early would be silly. 1363 */ 1364 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1365 { 1366 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1367 1368 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1369 } 1370 1371 static inline void tcp_reset_xmit_timer(struct sock *sk, 1372 const int what, 1373 unsigned long when, 1374 const unsigned long max_when) 1375 { 1376 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1377 max_when); 1378 } 1379 1380 /* Something is really bad, we could not queue an additional packet, 1381 * because qdisc is full or receiver sent a 0 window, or we are paced. 1382 * We do not want to add fuel to the fire, or abort too early, 1383 * so make sure the timer we arm now is at least 200ms in the future, 1384 * regardless of current icsk_rto value (as it could be ~2ms) 1385 */ 1386 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1387 { 1388 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1389 } 1390 1391 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1392 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1393 unsigned long max_when) 1394 { 1395 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1396 inet_csk(sk)->icsk_backoff); 1397 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1398 1399 return (unsigned long)min_t(u64, when, max_when); 1400 } 1401 1402 static inline void tcp_check_probe_timer(struct sock *sk) 1403 { 1404 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1405 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1406 tcp_probe0_base(sk), TCP_RTO_MAX); 1407 } 1408 1409 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1410 { 1411 tp->snd_wl1 = seq; 1412 } 1413 1414 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1415 { 1416 tp->snd_wl1 = seq; 1417 } 1418 1419 /* 1420 * Calculate(/check) TCP checksum 1421 */ 1422 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1423 __be32 daddr, __wsum base) 1424 { 1425 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1426 } 1427 1428 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1429 { 1430 return !skb_csum_unnecessary(skb) && 1431 __skb_checksum_complete(skb); 1432 } 1433 1434 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1435 enum skb_drop_reason *reason); 1436 1437 1438 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1439 void tcp_set_state(struct sock *sk, int state); 1440 void tcp_done(struct sock *sk); 1441 int tcp_abort(struct sock *sk, int err); 1442 1443 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1444 { 1445 rx_opt->dsack = 0; 1446 rx_opt->num_sacks = 0; 1447 } 1448 1449 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1450 1451 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1452 { 1453 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1454 struct tcp_sock *tp = tcp_sk(sk); 1455 s32 delta; 1456 1457 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1458 tp->packets_out || ca_ops->cong_control) 1459 return; 1460 delta = tcp_jiffies32 - tp->lsndtime; 1461 if (delta > inet_csk(sk)->icsk_rto) 1462 tcp_cwnd_restart(sk, delta); 1463 } 1464 1465 /* Determine a window scaling and initial window to offer. */ 1466 void tcp_select_initial_window(const struct sock *sk, int __space, 1467 __u32 mss, __u32 *rcv_wnd, 1468 __u32 *window_clamp, int wscale_ok, 1469 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1470 1471 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1472 { 1473 s64 scaled_space = (s64)space * scaling_ratio; 1474 1475 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1476 } 1477 1478 static inline int tcp_win_from_space(const struct sock *sk, int space) 1479 { 1480 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1481 } 1482 1483 /* inverse of __tcp_win_from_space() */ 1484 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1485 { 1486 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1487 1488 do_div(val, scaling_ratio); 1489 return val; 1490 } 1491 1492 static inline int tcp_space_from_win(const struct sock *sk, int win) 1493 { 1494 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1495 } 1496 1497 /* Assume a conservative default of 1200 bytes of payload per 4K page. 1498 * This may be adjusted later in tcp_measure_rcv_mss(). 1499 */ 1500 #define TCP_DEFAULT_SCALING_RATIO ((1200 << TCP_RMEM_TO_WIN_SCALE) / \ 1501 SKB_TRUESIZE(4096)) 1502 1503 static inline void tcp_scaling_ratio_init(struct sock *sk) 1504 { 1505 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1506 } 1507 1508 /* Note: caller must be prepared to deal with negative returns */ 1509 static inline int tcp_space(const struct sock *sk) 1510 { 1511 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1512 READ_ONCE(sk->sk_backlog.len) - 1513 atomic_read(&sk->sk_rmem_alloc)); 1514 } 1515 1516 static inline int tcp_full_space(const struct sock *sk) 1517 { 1518 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1519 } 1520 1521 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1522 { 1523 int unused_mem = sk_unused_reserved_mem(sk); 1524 struct tcp_sock *tp = tcp_sk(sk); 1525 1526 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1527 if (unused_mem) 1528 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1529 tcp_win_from_space(sk, unused_mem)); 1530 } 1531 1532 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1533 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1534 1535 1536 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1537 * If 87.5 % (7/8) of the space has been consumed, we want to override 1538 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1539 * len/truesize ratio. 1540 */ 1541 static inline bool tcp_rmem_pressure(const struct sock *sk) 1542 { 1543 int rcvbuf, threshold; 1544 1545 if (tcp_under_memory_pressure(sk)) 1546 return true; 1547 1548 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1549 threshold = rcvbuf - (rcvbuf >> 3); 1550 1551 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1552 } 1553 1554 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1555 { 1556 const struct tcp_sock *tp = tcp_sk(sk); 1557 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1558 1559 if (avail <= 0) 1560 return false; 1561 1562 return (avail >= target) || tcp_rmem_pressure(sk) || 1563 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1564 } 1565 1566 extern void tcp_openreq_init_rwin(struct request_sock *req, 1567 const struct sock *sk_listener, 1568 const struct dst_entry *dst); 1569 1570 void tcp_enter_memory_pressure(struct sock *sk); 1571 void tcp_leave_memory_pressure(struct sock *sk); 1572 1573 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1574 { 1575 struct net *net = sock_net((struct sock *)tp); 1576 int val; 1577 1578 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1579 * and do_tcp_setsockopt(). 1580 */ 1581 val = READ_ONCE(tp->keepalive_intvl); 1582 1583 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1584 } 1585 1586 static inline int keepalive_time_when(const struct tcp_sock *tp) 1587 { 1588 struct net *net = sock_net((struct sock *)tp); 1589 int val; 1590 1591 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1592 val = READ_ONCE(tp->keepalive_time); 1593 1594 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1595 } 1596 1597 static inline int keepalive_probes(const struct tcp_sock *tp) 1598 { 1599 struct net *net = sock_net((struct sock *)tp); 1600 int val; 1601 1602 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1603 * and do_tcp_setsockopt(). 1604 */ 1605 val = READ_ONCE(tp->keepalive_probes); 1606 1607 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1608 } 1609 1610 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1611 { 1612 const struct inet_connection_sock *icsk = &tp->inet_conn; 1613 1614 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1615 tcp_jiffies32 - tp->rcv_tstamp); 1616 } 1617 1618 static inline int tcp_fin_time(const struct sock *sk) 1619 { 1620 int fin_timeout = tcp_sk(sk)->linger2 ? : 1621 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1622 const int rto = inet_csk(sk)->icsk_rto; 1623 1624 if (fin_timeout < (rto << 2) - (rto >> 1)) 1625 fin_timeout = (rto << 2) - (rto >> 1); 1626 1627 return fin_timeout; 1628 } 1629 1630 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1631 int paws_win) 1632 { 1633 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1634 return true; 1635 if (unlikely(!time_before32(ktime_get_seconds(), 1636 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1637 return true; 1638 /* 1639 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1640 * then following tcp messages have valid values. Ignore 0 value, 1641 * or else 'negative' tsval might forbid us to accept their packets. 1642 */ 1643 if (!rx_opt->ts_recent) 1644 return true; 1645 return false; 1646 } 1647 1648 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1649 int rst) 1650 { 1651 if (tcp_paws_check(rx_opt, 0)) 1652 return false; 1653 1654 /* RST segments are not recommended to carry timestamp, 1655 and, if they do, it is recommended to ignore PAWS because 1656 "their cleanup function should take precedence over timestamps." 1657 Certainly, it is mistake. It is necessary to understand the reasons 1658 of this constraint to relax it: if peer reboots, clock may go 1659 out-of-sync and half-open connections will not be reset. 1660 Actually, the problem would be not existing if all 1661 the implementations followed draft about maintaining clock 1662 via reboots. Linux-2.2 DOES NOT! 1663 1664 However, we can relax time bounds for RST segments to MSL. 1665 */ 1666 if (rst && !time_before32(ktime_get_seconds(), 1667 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1668 return false; 1669 return true; 1670 } 1671 1672 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1673 int mib_idx, u32 *last_oow_ack_time); 1674 1675 static inline void tcp_mib_init(struct net *net) 1676 { 1677 /* See RFC 2012 */ 1678 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1679 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1680 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1681 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1682 } 1683 1684 /* from STCP */ 1685 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1686 { 1687 tp->lost_skb_hint = NULL; 1688 } 1689 1690 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1691 { 1692 tcp_clear_retrans_hints_partial(tp); 1693 tp->retransmit_skb_hint = NULL; 1694 } 1695 1696 #define tcp_md5_addr tcp_ao_addr 1697 1698 /* - key database */ 1699 struct tcp_md5sig_key { 1700 struct hlist_node node; 1701 u8 keylen; 1702 u8 family; /* AF_INET or AF_INET6 */ 1703 u8 prefixlen; 1704 u8 flags; 1705 union tcp_md5_addr addr; 1706 int l3index; /* set if key added with L3 scope */ 1707 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1708 struct rcu_head rcu; 1709 }; 1710 1711 /* - sock block */ 1712 struct tcp_md5sig_info { 1713 struct hlist_head head; 1714 struct rcu_head rcu; 1715 }; 1716 1717 /* - pseudo header */ 1718 struct tcp4_pseudohdr { 1719 __be32 saddr; 1720 __be32 daddr; 1721 __u8 pad; 1722 __u8 protocol; 1723 __be16 len; 1724 }; 1725 1726 struct tcp6_pseudohdr { 1727 struct in6_addr saddr; 1728 struct in6_addr daddr; 1729 __be32 len; 1730 __be32 protocol; /* including padding */ 1731 }; 1732 1733 union tcp_md5sum_block { 1734 struct tcp4_pseudohdr ip4; 1735 #if IS_ENABLED(CONFIG_IPV6) 1736 struct tcp6_pseudohdr ip6; 1737 #endif 1738 }; 1739 1740 /* 1741 * struct tcp_sigpool - per-CPU pool of ahash_requests 1742 * @scratch: per-CPU temporary area, that can be used between 1743 * tcp_sigpool_start() and tcp_sigpool_end() to perform 1744 * crypto request 1745 * @req: pre-allocated ahash request 1746 */ 1747 struct tcp_sigpool { 1748 void *scratch; 1749 struct ahash_request *req; 1750 }; 1751 1752 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size); 1753 void tcp_sigpool_get(unsigned int id); 1754 void tcp_sigpool_release(unsigned int id); 1755 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp, 1756 const struct sk_buff *skb, 1757 unsigned int header_len); 1758 1759 /** 1760 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash 1761 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash() 1762 * @c: returned tcp_sigpool for usage (uninitialized on failure) 1763 * 1764 * Returns 0 on success, error otherwise. 1765 */ 1766 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c); 1767 /** 1768 * tcp_sigpool_end - enable bh and stop using tcp_sigpool 1769 * @c: tcp_sigpool context that was returned by tcp_sigpool_start() 1770 */ 1771 void tcp_sigpool_end(struct tcp_sigpool *c); 1772 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len); 1773 /* - functions */ 1774 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1775 const struct sock *sk, const struct sk_buff *skb); 1776 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1777 int family, u8 prefixlen, int l3index, u8 flags, 1778 const u8 *newkey, u8 newkeylen); 1779 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1780 int family, u8 prefixlen, int l3index, 1781 struct tcp_md5sig_key *key); 1782 1783 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1784 int family, u8 prefixlen, int l3index, u8 flags); 1785 void tcp_clear_md5_list(struct sock *sk); 1786 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1787 const struct sock *addr_sk); 1788 1789 #ifdef CONFIG_TCP_MD5SIG 1790 #include <linux/jump_label.h> 1791 extern struct static_key_false_deferred tcp_md5_needed; 1792 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1793 const union tcp_md5_addr *addr, 1794 int family, bool any_l3index); 1795 static inline struct tcp_md5sig_key * 1796 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1797 const union tcp_md5_addr *addr, int family) 1798 { 1799 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1800 return NULL; 1801 return __tcp_md5_do_lookup(sk, l3index, addr, family, false); 1802 } 1803 1804 static inline struct tcp_md5sig_key * 1805 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1806 const union tcp_md5_addr *addr, int family) 1807 { 1808 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1809 return NULL; 1810 return __tcp_md5_do_lookup(sk, 0, addr, family, true); 1811 } 1812 1813 enum skb_drop_reason 1814 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1815 const void *saddr, const void *daddr, 1816 int family, int l3index, const __u8 *hash_location); 1817 1818 1819 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1820 #else 1821 static inline struct tcp_md5sig_key * 1822 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1823 const union tcp_md5_addr *addr, int family) 1824 { 1825 return NULL; 1826 } 1827 1828 static inline struct tcp_md5sig_key * 1829 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1830 const union tcp_md5_addr *addr, int family) 1831 { 1832 return NULL; 1833 } 1834 1835 static inline enum skb_drop_reason 1836 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1837 const void *saddr, const void *daddr, 1838 int family, int l3index, const __u8 *hash_location) 1839 { 1840 return SKB_NOT_DROPPED_YET; 1841 } 1842 #define tcp_twsk_md5_key(twsk) NULL 1843 #endif 1844 1845 int tcp_md5_alloc_sigpool(void); 1846 void tcp_md5_release_sigpool(void); 1847 void tcp_md5_add_sigpool(void); 1848 extern int tcp_md5_sigpool_id; 1849 1850 int tcp_md5_hash_key(struct tcp_sigpool *hp, 1851 const struct tcp_md5sig_key *key); 1852 1853 /* From tcp_fastopen.c */ 1854 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1855 struct tcp_fastopen_cookie *cookie); 1856 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1857 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1858 u16 try_exp); 1859 struct tcp_fastopen_request { 1860 /* Fast Open cookie. Size 0 means a cookie request */ 1861 struct tcp_fastopen_cookie cookie; 1862 struct msghdr *data; /* data in MSG_FASTOPEN */ 1863 size_t size; 1864 int copied; /* queued in tcp_connect() */ 1865 struct ubuf_info *uarg; 1866 }; 1867 void tcp_free_fastopen_req(struct tcp_sock *tp); 1868 void tcp_fastopen_destroy_cipher(struct sock *sk); 1869 void tcp_fastopen_ctx_destroy(struct net *net); 1870 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1871 void *primary_key, void *backup_key); 1872 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1873 u64 *key); 1874 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1875 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1876 struct request_sock *req, 1877 struct tcp_fastopen_cookie *foc, 1878 const struct dst_entry *dst); 1879 void tcp_fastopen_init_key_once(struct net *net); 1880 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1881 struct tcp_fastopen_cookie *cookie); 1882 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1883 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1884 #define TCP_FASTOPEN_KEY_MAX 2 1885 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1886 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1887 1888 /* Fastopen key context */ 1889 struct tcp_fastopen_context { 1890 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1891 int num; 1892 struct rcu_head rcu; 1893 }; 1894 1895 void tcp_fastopen_active_disable(struct sock *sk); 1896 bool tcp_fastopen_active_should_disable(struct sock *sk); 1897 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1898 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1899 1900 /* Caller needs to wrap with rcu_read_(un)lock() */ 1901 static inline 1902 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1903 { 1904 struct tcp_fastopen_context *ctx; 1905 1906 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1907 if (!ctx) 1908 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1909 return ctx; 1910 } 1911 1912 static inline 1913 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1914 const struct tcp_fastopen_cookie *orig) 1915 { 1916 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1917 orig->len == foc->len && 1918 !memcmp(orig->val, foc->val, foc->len)) 1919 return true; 1920 return false; 1921 } 1922 1923 static inline 1924 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1925 { 1926 return ctx->num; 1927 } 1928 1929 /* Latencies incurred by various limits for a sender. They are 1930 * chronograph-like stats that are mutually exclusive. 1931 */ 1932 enum tcp_chrono { 1933 TCP_CHRONO_UNSPEC, 1934 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1935 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1936 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1937 __TCP_CHRONO_MAX, 1938 }; 1939 1940 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1941 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1942 1943 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1944 * the same memory storage than skb->destructor/_skb_refdst 1945 */ 1946 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1947 { 1948 skb->destructor = NULL; 1949 skb->_skb_refdst = 0UL; 1950 } 1951 1952 #define tcp_skb_tsorted_save(skb) { \ 1953 unsigned long _save = skb->_skb_refdst; \ 1954 skb->_skb_refdst = 0UL; 1955 1956 #define tcp_skb_tsorted_restore(skb) \ 1957 skb->_skb_refdst = _save; \ 1958 } 1959 1960 void tcp_write_queue_purge(struct sock *sk); 1961 1962 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1963 { 1964 return skb_rb_first(&sk->tcp_rtx_queue); 1965 } 1966 1967 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1968 { 1969 return skb_rb_last(&sk->tcp_rtx_queue); 1970 } 1971 1972 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1973 { 1974 return skb_peek_tail(&sk->sk_write_queue); 1975 } 1976 1977 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1978 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1979 1980 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1981 { 1982 return skb_peek(&sk->sk_write_queue); 1983 } 1984 1985 static inline bool tcp_skb_is_last(const struct sock *sk, 1986 const struct sk_buff *skb) 1987 { 1988 return skb_queue_is_last(&sk->sk_write_queue, skb); 1989 } 1990 1991 /** 1992 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1993 * @sk: socket 1994 * 1995 * Since the write queue can have a temporary empty skb in it, 1996 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1997 */ 1998 static inline bool tcp_write_queue_empty(const struct sock *sk) 1999 { 2000 const struct tcp_sock *tp = tcp_sk(sk); 2001 2002 return tp->write_seq == tp->snd_nxt; 2003 } 2004 2005 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 2006 { 2007 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 2008 } 2009 2010 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 2011 { 2012 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 2013 } 2014 2015 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 2016 { 2017 __skb_queue_tail(&sk->sk_write_queue, skb); 2018 2019 /* Queue it, remembering where we must start sending. */ 2020 if (sk->sk_write_queue.next == skb) 2021 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 2022 } 2023 2024 /* Insert new before skb on the write queue of sk. */ 2025 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 2026 struct sk_buff *skb, 2027 struct sock *sk) 2028 { 2029 __skb_queue_before(&sk->sk_write_queue, skb, new); 2030 } 2031 2032 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 2033 { 2034 tcp_skb_tsorted_anchor_cleanup(skb); 2035 __skb_unlink(skb, &sk->sk_write_queue); 2036 } 2037 2038 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 2039 2040 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2041 { 2042 tcp_skb_tsorted_anchor_cleanup(skb); 2043 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2044 } 2045 2046 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2047 { 2048 list_del(&skb->tcp_tsorted_anchor); 2049 tcp_rtx_queue_unlink(skb, sk); 2050 tcp_wmem_free_skb(sk, skb); 2051 } 2052 2053 static inline void tcp_push_pending_frames(struct sock *sk) 2054 { 2055 if (tcp_send_head(sk)) { 2056 struct tcp_sock *tp = tcp_sk(sk); 2057 2058 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2059 } 2060 } 2061 2062 /* Start sequence of the skb just after the highest skb with SACKed 2063 * bit, valid only if sacked_out > 0 or when the caller has ensured 2064 * validity by itself. 2065 */ 2066 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2067 { 2068 if (!tp->sacked_out) 2069 return tp->snd_una; 2070 2071 if (tp->highest_sack == NULL) 2072 return tp->snd_nxt; 2073 2074 return TCP_SKB_CB(tp->highest_sack)->seq; 2075 } 2076 2077 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2078 { 2079 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2080 } 2081 2082 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2083 { 2084 return tcp_sk(sk)->highest_sack; 2085 } 2086 2087 static inline void tcp_highest_sack_reset(struct sock *sk) 2088 { 2089 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2090 } 2091 2092 /* Called when old skb is about to be deleted and replaced by new skb */ 2093 static inline void tcp_highest_sack_replace(struct sock *sk, 2094 struct sk_buff *old, 2095 struct sk_buff *new) 2096 { 2097 if (old == tcp_highest_sack(sk)) 2098 tcp_sk(sk)->highest_sack = new; 2099 } 2100 2101 /* This helper checks if socket has IP_TRANSPARENT set */ 2102 static inline bool inet_sk_transparent(const struct sock *sk) 2103 { 2104 switch (sk->sk_state) { 2105 case TCP_TIME_WAIT: 2106 return inet_twsk(sk)->tw_transparent; 2107 case TCP_NEW_SYN_RECV: 2108 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2109 } 2110 return inet_test_bit(TRANSPARENT, sk); 2111 } 2112 2113 /* Determines whether this is a thin stream (which may suffer from 2114 * increased latency). Used to trigger latency-reducing mechanisms. 2115 */ 2116 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2117 { 2118 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2119 } 2120 2121 /* /proc */ 2122 enum tcp_seq_states { 2123 TCP_SEQ_STATE_LISTENING, 2124 TCP_SEQ_STATE_ESTABLISHED, 2125 }; 2126 2127 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2128 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2129 void tcp_seq_stop(struct seq_file *seq, void *v); 2130 2131 struct tcp_seq_afinfo { 2132 sa_family_t family; 2133 }; 2134 2135 struct tcp_iter_state { 2136 struct seq_net_private p; 2137 enum tcp_seq_states state; 2138 struct sock *syn_wait_sk; 2139 int bucket, offset, sbucket, num; 2140 loff_t last_pos; 2141 }; 2142 2143 extern struct request_sock_ops tcp_request_sock_ops; 2144 extern struct request_sock_ops tcp6_request_sock_ops; 2145 2146 void tcp_v4_destroy_sock(struct sock *sk); 2147 2148 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2149 netdev_features_t features); 2150 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2151 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2152 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2153 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2154 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2155 #ifdef CONFIG_INET 2156 void tcp_gro_complete(struct sk_buff *skb); 2157 #else 2158 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2159 #endif 2160 2161 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2162 2163 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2164 { 2165 struct net *net = sock_net((struct sock *)tp); 2166 u32 val; 2167 2168 val = READ_ONCE(tp->notsent_lowat); 2169 2170 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2171 } 2172 2173 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2174 2175 #ifdef CONFIG_PROC_FS 2176 int tcp4_proc_init(void); 2177 void tcp4_proc_exit(void); 2178 #endif 2179 2180 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2181 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2182 const struct tcp_request_sock_ops *af_ops, 2183 struct sock *sk, struct sk_buff *skb); 2184 2185 /* TCP af-specific functions */ 2186 struct tcp_sock_af_ops { 2187 #ifdef CONFIG_TCP_MD5SIG 2188 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2189 const struct sock *addr_sk); 2190 int (*calc_md5_hash)(char *location, 2191 const struct tcp_md5sig_key *md5, 2192 const struct sock *sk, 2193 const struct sk_buff *skb); 2194 int (*md5_parse)(struct sock *sk, 2195 int optname, 2196 sockptr_t optval, 2197 int optlen); 2198 #endif 2199 #ifdef CONFIG_TCP_AO 2200 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen); 2201 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2202 struct sock *addr_sk, 2203 int sndid, int rcvid); 2204 int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key, 2205 const struct sock *sk, 2206 __be32 sisn, __be32 disn, bool send); 2207 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao, 2208 const struct sock *sk, const struct sk_buff *skb, 2209 const u8 *tkey, int hash_offset, u32 sne); 2210 #endif 2211 }; 2212 2213 struct tcp_request_sock_ops { 2214 u16 mss_clamp; 2215 #ifdef CONFIG_TCP_MD5SIG 2216 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2217 const struct sock *addr_sk); 2218 int (*calc_md5_hash) (char *location, 2219 const struct tcp_md5sig_key *md5, 2220 const struct sock *sk, 2221 const struct sk_buff *skb); 2222 #endif 2223 #ifdef CONFIG_TCP_AO 2224 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2225 struct request_sock *req, 2226 int sndid, int rcvid); 2227 int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk); 2228 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt, 2229 struct request_sock *req, const struct sk_buff *skb, 2230 int hash_offset, u32 sne); 2231 #endif 2232 #ifdef CONFIG_SYN_COOKIES 2233 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2234 __u16 *mss); 2235 #endif 2236 struct dst_entry *(*route_req)(const struct sock *sk, 2237 struct sk_buff *skb, 2238 struct flowi *fl, 2239 struct request_sock *req); 2240 u32 (*init_seq)(const struct sk_buff *skb); 2241 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2242 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2243 struct flowi *fl, struct request_sock *req, 2244 struct tcp_fastopen_cookie *foc, 2245 enum tcp_synack_type synack_type, 2246 struct sk_buff *syn_skb); 2247 }; 2248 2249 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2250 #if IS_ENABLED(CONFIG_IPV6) 2251 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2252 #endif 2253 2254 #ifdef CONFIG_SYN_COOKIES 2255 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2256 const struct sock *sk, struct sk_buff *skb, 2257 __u16 *mss) 2258 { 2259 tcp_synq_overflow(sk); 2260 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2261 return ops->cookie_init_seq(skb, mss); 2262 } 2263 #else 2264 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2265 const struct sock *sk, struct sk_buff *skb, 2266 __u16 *mss) 2267 { 2268 return 0; 2269 } 2270 #endif 2271 2272 struct tcp_key { 2273 union { 2274 struct { 2275 struct tcp_ao_key *ao_key; 2276 char *traffic_key; 2277 u32 sne; 2278 u8 rcv_next; 2279 }; 2280 struct tcp_md5sig_key *md5_key; 2281 }; 2282 enum { 2283 TCP_KEY_NONE = 0, 2284 TCP_KEY_MD5, 2285 TCP_KEY_AO, 2286 } type; 2287 }; 2288 2289 static inline void tcp_get_current_key(const struct sock *sk, 2290 struct tcp_key *out) 2291 { 2292 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG) 2293 const struct tcp_sock *tp = tcp_sk(sk); 2294 #endif 2295 2296 #ifdef CONFIG_TCP_AO 2297 if (static_branch_unlikely(&tcp_ao_needed.key)) { 2298 struct tcp_ao_info *ao; 2299 2300 ao = rcu_dereference_protected(tp->ao_info, 2301 lockdep_sock_is_held(sk)); 2302 if (ao) { 2303 out->ao_key = READ_ONCE(ao->current_key); 2304 out->type = TCP_KEY_AO; 2305 return; 2306 } 2307 } 2308 #endif 2309 #ifdef CONFIG_TCP_MD5SIG 2310 if (static_branch_unlikely(&tcp_md5_needed.key) && 2311 rcu_access_pointer(tp->md5sig_info)) { 2312 out->md5_key = tp->af_specific->md5_lookup(sk, sk); 2313 if (out->md5_key) { 2314 out->type = TCP_KEY_MD5; 2315 return; 2316 } 2317 } 2318 #endif 2319 out->type = TCP_KEY_NONE; 2320 } 2321 2322 static inline bool tcp_key_is_md5(const struct tcp_key *key) 2323 { 2324 #ifdef CONFIG_TCP_MD5SIG 2325 if (static_branch_unlikely(&tcp_md5_needed.key) && 2326 key->type == TCP_KEY_MD5) 2327 return true; 2328 #endif 2329 return false; 2330 } 2331 2332 static inline bool tcp_key_is_ao(const struct tcp_key *key) 2333 { 2334 #ifdef CONFIG_TCP_AO 2335 if (static_branch_unlikely(&tcp_ao_needed.key) && 2336 key->type == TCP_KEY_AO) 2337 return true; 2338 #endif 2339 return false; 2340 } 2341 2342 int tcpv4_offload_init(void); 2343 2344 void tcp_v4_init(void); 2345 void tcp_init(void); 2346 2347 /* tcp_recovery.c */ 2348 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2349 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2350 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2351 u32 reo_wnd); 2352 extern bool tcp_rack_mark_lost(struct sock *sk); 2353 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2354 u64 xmit_time); 2355 extern void tcp_rack_reo_timeout(struct sock *sk); 2356 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2357 2358 /* tcp_plb.c */ 2359 2360 /* 2361 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2362 * expects cong_ratio which represents fraction of traffic that experienced 2363 * congestion over a single RTT. In order to avoid floating point operations, 2364 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2365 */ 2366 #define TCP_PLB_SCALE 8 2367 2368 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2369 struct tcp_plb_state { 2370 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2371 unused:3; 2372 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2373 }; 2374 2375 static inline void tcp_plb_init(const struct sock *sk, 2376 struct tcp_plb_state *plb) 2377 { 2378 plb->consec_cong_rounds = 0; 2379 plb->pause_until = 0; 2380 } 2381 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2382 const int cong_ratio); 2383 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2384 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2385 2386 /* At how many usecs into the future should the RTO fire? */ 2387 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2388 { 2389 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2390 u32 rto = inet_csk(sk)->icsk_rto; 2391 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2392 2393 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2394 } 2395 2396 /* 2397 * Save and compile IPv4 options, return a pointer to it 2398 */ 2399 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2400 struct sk_buff *skb) 2401 { 2402 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2403 struct ip_options_rcu *dopt = NULL; 2404 2405 if (opt->optlen) { 2406 int opt_size = sizeof(*dopt) + opt->optlen; 2407 2408 dopt = kmalloc(opt_size, GFP_ATOMIC); 2409 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2410 kfree(dopt); 2411 dopt = NULL; 2412 } 2413 } 2414 return dopt; 2415 } 2416 2417 /* locally generated TCP pure ACKs have skb->truesize == 2 2418 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2419 * This is much faster than dissecting the packet to find out. 2420 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2421 */ 2422 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2423 { 2424 return skb->truesize == 2; 2425 } 2426 2427 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2428 { 2429 skb->truesize = 2; 2430 } 2431 2432 static inline int tcp_inq(struct sock *sk) 2433 { 2434 struct tcp_sock *tp = tcp_sk(sk); 2435 int answ; 2436 2437 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2438 answ = 0; 2439 } else if (sock_flag(sk, SOCK_URGINLINE) || 2440 !tp->urg_data || 2441 before(tp->urg_seq, tp->copied_seq) || 2442 !before(tp->urg_seq, tp->rcv_nxt)) { 2443 2444 answ = tp->rcv_nxt - tp->copied_seq; 2445 2446 /* Subtract 1, if FIN was received */ 2447 if (answ && sock_flag(sk, SOCK_DONE)) 2448 answ--; 2449 } else { 2450 answ = tp->urg_seq - tp->copied_seq; 2451 } 2452 2453 return answ; 2454 } 2455 2456 int tcp_peek_len(struct socket *sock); 2457 2458 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2459 { 2460 u16 segs_in; 2461 2462 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2463 2464 /* We update these fields while other threads might 2465 * read them from tcp_get_info() 2466 */ 2467 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2468 if (skb->len > tcp_hdrlen(skb)) 2469 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2470 } 2471 2472 /* 2473 * TCP listen path runs lockless. 2474 * We forced "struct sock" to be const qualified to make sure 2475 * we don't modify one of its field by mistake. 2476 * Here, we increment sk_drops which is an atomic_t, so we can safely 2477 * make sock writable again. 2478 */ 2479 static inline void tcp_listendrop(const struct sock *sk) 2480 { 2481 atomic_inc(&((struct sock *)sk)->sk_drops); 2482 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2483 } 2484 2485 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2486 2487 /* 2488 * Interface for adding Upper Level Protocols over TCP 2489 */ 2490 2491 #define TCP_ULP_NAME_MAX 16 2492 #define TCP_ULP_MAX 128 2493 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2494 2495 struct tcp_ulp_ops { 2496 struct list_head list; 2497 2498 /* initialize ulp */ 2499 int (*init)(struct sock *sk); 2500 /* update ulp */ 2501 void (*update)(struct sock *sk, struct proto *p, 2502 void (*write_space)(struct sock *sk)); 2503 /* cleanup ulp */ 2504 void (*release)(struct sock *sk); 2505 /* diagnostic */ 2506 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2507 size_t (*get_info_size)(const struct sock *sk); 2508 /* clone ulp */ 2509 void (*clone)(const struct request_sock *req, struct sock *newsk, 2510 const gfp_t priority); 2511 2512 char name[TCP_ULP_NAME_MAX]; 2513 struct module *owner; 2514 }; 2515 int tcp_register_ulp(struct tcp_ulp_ops *type); 2516 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2517 int tcp_set_ulp(struct sock *sk, const char *name); 2518 void tcp_get_available_ulp(char *buf, size_t len); 2519 void tcp_cleanup_ulp(struct sock *sk); 2520 void tcp_update_ulp(struct sock *sk, struct proto *p, 2521 void (*write_space)(struct sock *sk)); 2522 2523 #define MODULE_ALIAS_TCP_ULP(name) \ 2524 __MODULE_INFO(alias, alias_userspace, name); \ 2525 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2526 2527 #ifdef CONFIG_NET_SOCK_MSG 2528 struct sk_msg; 2529 struct sk_psock; 2530 2531 #ifdef CONFIG_BPF_SYSCALL 2532 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2533 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2534 #endif /* CONFIG_BPF_SYSCALL */ 2535 2536 #ifdef CONFIG_INET 2537 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2538 #else 2539 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2540 { 2541 } 2542 #endif 2543 2544 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2545 struct sk_msg *msg, u32 bytes, int flags); 2546 #endif /* CONFIG_NET_SOCK_MSG */ 2547 2548 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2549 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2550 { 2551 } 2552 #endif 2553 2554 #ifdef CONFIG_CGROUP_BPF 2555 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2556 struct sk_buff *skb, 2557 unsigned int end_offset) 2558 { 2559 skops->skb = skb; 2560 skops->skb_data_end = skb->data + end_offset; 2561 } 2562 #else 2563 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2564 struct sk_buff *skb, 2565 unsigned int end_offset) 2566 { 2567 } 2568 #endif 2569 2570 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2571 * is < 0, then the BPF op failed (for example if the loaded BPF 2572 * program does not support the chosen operation or there is no BPF 2573 * program loaded). 2574 */ 2575 #ifdef CONFIG_BPF 2576 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2577 { 2578 struct bpf_sock_ops_kern sock_ops; 2579 int ret; 2580 2581 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2582 if (sk_fullsock(sk)) { 2583 sock_ops.is_fullsock = 1; 2584 sock_owned_by_me(sk); 2585 } 2586 2587 sock_ops.sk = sk; 2588 sock_ops.op = op; 2589 if (nargs > 0) 2590 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2591 2592 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2593 if (ret == 0) 2594 ret = sock_ops.reply; 2595 else 2596 ret = -1; 2597 return ret; 2598 } 2599 2600 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2601 { 2602 u32 args[2] = {arg1, arg2}; 2603 2604 return tcp_call_bpf(sk, op, 2, args); 2605 } 2606 2607 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2608 u32 arg3) 2609 { 2610 u32 args[3] = {arg1, arg2, arg3}; 2611 2612 return tcp_call_bpf(sk, op, 3, args); 2613 } 2614 2615 #else 2616 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2617 { 2618 return -EPERM; 2619 } 2620 2621 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2622 { 2623 return -EPERM; 2624 } 2625 2626 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2627 u32 arg3) 2628 { 2629 return -EPERM; 2630 } 2631 2632 #endif 2633 2634 static inline u32 tcp_timeout_init(struct sock *sk) 2635 { 2636 int timeout; 2637 2638 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2639 2640 if (timeout <= 0) 2641 timeout = TCP_TIMEOUT_INIT; 2642 return min_t(int, timeout, TCP_RTO_MAX); 2643 } 2644 2645 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2646 { 2647 int rwnd; 2648 2649 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2650 2651 if (rwnd < 0) 2652 rwnd = 0; 2653 return rwnd; 2654 } 2655 2656 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2657 { 2658 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2659 } 2660 2661 static inline void tcp_bpf_rtt(struct sock *sk) 2662 { 2663 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2664 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2665 } 2666 2667 #if IS_ENABLED(CONFIG_SMC) 2668 extern struct static_key_false tcp_have_smc; 2669 #endif 2670 2671 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2672 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2673 void (*cad)(struct sock *sk, u32 ack_seq)); 2674 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2675 void clean_acked_data_flush(void); 2676 #endif 2677 2678 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2679 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2680 const struct tcp_sock *tp) 2681 { 2682 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2683 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2684 } 2685 2686 /* Compute Earliest Departure Time for some control packets 2687 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2688 */ 2689 static inline u64 tcp_transmit_time(const struct sock *sk) 2690 { 2691 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2692 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2693 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2694 2695 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2696 } 2697 return 0; 2698 } 2699 2700 static inline int tcp_parse_auth_options(const struct tcphdr *th, 2701 const u8 **md5_hash, const struct tcp_ao_hdr **aoh) 2702 { 2703 const u8 *md5_tmp, *ao_tmp; 2704 int ret; 2705 2706 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp); 2707 if (ret) 2708 return ret; 2709 2710 if (md5_hash) 2711 *md5_hash = md5_tmp; 2712 2713 if (aoh) { 2714 if (!ao_tmp) 2715 *aoh = NULL; 2716 else 2717 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2); 2718 } 2719 2720 return 0; 2721 } 2722 2723 static inline bool tcp_ao_required(struct sock *sk, const void *saddr, 2724 int family, int l3index, bool stat_inc) 2725 { 2726 #ifdef CONFIG_TCP_AO 2727 struct tcp_ao_info *ao_info; 2728 struct tcp_ao_key *ao_key; 2729 2730 if (!static_branch_unlikely(&tcp_ao_needed.key)) 2731 return false; 2732 2733 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info, 2734 lockdep_sock_is_held(sk)); 2735 if (!ao_info) 2736 return false; 2737 2738 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1); 2739 if (ao_info->ao_required || ao_key) { 2740 if (stat_inc) { 2741 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED); 2742 atomic64_inc(&ao_info->counters.ao_required); 2743 } 2744 return true; 2745 } 2746 #endif 2747 return false; 2748 } 2749 2750 /* Called with rcu_read_lock() */ 2751 static inline enum skb_drop_reason 2752 tcp_inbound_hash(struct sock *sk, const struct request_sock *req, 2753 const struct sk_buff *skb, 2754 const void *saddr, const void *daddr, 2755 int family, int dif, int sdif) 2756 { 2757 const struct tcphdr *th = tcp_hdr(skb); 2758 const struct tcp_ao_hdr *aoh; 2759 const __u8 *md5_location; 2760 int l3index; 2761 2762 /* Invalid option or two times meet any of auth options */ 2763 if (tcp_parse_auth_options(th, &md5_location, &aoh)) { 2764 tcp_hash_fail("TCP segment has incorrect auth options set", 2765 family, skb, ""); 2766 return SKB_DROP_REASON_TCP_AUTH_HDR; 2767 } 2768 2769 if (req) { 2770 if (tcp_rsk_used_ao(req) != !!aoh) { 2771 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD); 2772 tcp_hash_fail("TCP connection can't start/end using TCP-AO", 2773 family, skb, "%s", 2774 !aoh ? "missing AO" : "AO signed"); 2775 return SKB_DROP_REASON_TCP_AOFAILURE; 2776 } 2777 } 2778 2779 /* sdif set, means packet ingressed via a device 2780 * in an L3 domain and dif is set to the l3mdev 2781 */ 2782 l3index = sdif ? dif : 0; 2783 2784 /* Fast path: unsigned segments */ 2785 if (likely(!md5_location && !aoh)) { 2786 /* Drop if there's TCP-MD5 or TCP-AO key with any rcvid/sndid 2787 * for the remote peer. On TCP-AO established connection 2788 * the last key is impossible to remove, so there's 2789 * always at least one current_key. 2790 */ 2791 if (tcp_ao_required(sk, saddr, family, l3index, true)) { 2792 tcp_hash_fail("AO hash is required, but not found", 2793 family, skb, "L3 index %d", l3index); 2794 return SKB_DROP_REASON_TCP_AONOTFOUND; 2795 } 2796 if (unlikely(tcp_md5_do_lookup(sk, l3index, saddr, family))) { 2797 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 2798 tcp_hash_fail("MD5 Hash not found", 2799 family, skb, "L3 index %d", l3index); 2800 return SKB_DROP_REASON_TCP_MD5NOTFOUND; 2801 } 2802 return SKB_NOT_DROPPED_YET; 2803 } 2804 2805 if (aoh) 2806 return tcp_inbound_ao_hash(sk, skb, family, req, l3index, aoh); 2807 2808 return tcp_inbound_md5_hash(sk, skb, saddr, daddr, family, 2809 l3index, md5_location); 2810 } 2811 2812 #endif /* _TCP_H */ 2813