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 #include <net/xfrm.h> 45 46 #include <linux/seq_file.h> 47 #include <linux/memcontrol.h> 48 #include <linux/bpf-cgroup.h> 49 #include <linux/siphash.h> 50 51 extern struct inet_hashinfo tcp_hashinfo; 52 53 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 54 int tcp_orphan_count_sum(void); 55 56 DECLARE_PER_CPU(u32, tcp_tw_isn); 57 58 void tcp_time_wait(struct sock *sk, int state, int timeo); 59 60 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 61 #define MAX_TCP_OPTION_SPACE 40 62 #define TCP_MIN_SND_MSS 48 63 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 64 65 /* 66 * Never offer a window over 32767 without using window scaling. Some 67 * poor stacks do signed 16bit maths! 68 */ 69 #define MAX_TCP_WINDOW 32767U 70 71 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 72 #define TCP_MIN_MSS 88U 73 74 /* The initial MTU to use for probing */ 75 #define TCP_BASE_MSS 1024 76 77 /* probing interval, default to 10 minutes as per RFC4821 */ 78 #define TCP_PROBE_INTERVAL 600 79 80 /* Specify interval when tcp mtu probing will stop */ 81 #define TCP_PROBE_THRESHOLD 8 82 83 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 84 #define TCP_FASTRETRANS_THRESH 3 85 86 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 87 #define TCP_MAX_QUICKACKS 16U 88 89 /* Maximal number of window scale according to RFC1323 */ 90 #define TCP_MAX_WSCALE 14U 91 92 /* urg_data states */ 93 #define TCP_URG_VALID 0x0100 94 #define TCP_URG_NOTYET 0x0200 95 #define TCP_URG_READ 0x0400 96 97 #define TCP_RETR1 3 /* 98 * This is how many retries it does before it 99 * tries to figure out if the gateway is 100 * down. Minimal RFC value is 3; it corresponds 101 * to ~3sec-8min depending on RTO. 102 */ 103 104 #define TCP_RETR2 15 /* 105 * This should take at least 106 * 90 minutes to time out. 107 * RFC1122 says that the limit is 100 sec. 108 * 15 is ~13-30min depending on RTO. 109 */ 110 111 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 112 * when active opening a connection. 113 * RFC1122 says the minimum retry MUST 114 * be at least 180secs. Nevertheless 115 * this value is corresponding to 116 * 63secs of retransmission with the 117 * current initial RTO. 118 */ 119 120 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 121 * when passive opening a connection. 122 * This is corresponding to 31secs of 123 * retransmission with the current 124 * initial RTO. 125 */ 126 127 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 128 * state, about 60 seconds */ 129 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 130 /* BSD style FIN_WAIT2 deadlock breaker. 131 * It used to be 3min, new value is 60sec, 132 * to combine FIN-WAIT-2 timeout with 133 * TIME-WAIT timer. 134 */ 135 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 136 137 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 138 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); 139 140 #if HZ >= 100 141 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 142 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 143 #else 144 #define TCP_DELACK_MIN 4U 145 #define TCP_ATO_MIN 4U 146 #endif 147 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 148 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 149 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 150 151 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 152 153 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 154 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 155 * used as a fallback RTO for the 156 * initial data transmission if no 157 * valid RTT sample has been acquired, 158 * most likely due to retrans in 3WHS. 159 */ 160 161 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 162 * for local resources. 163 */ 164 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 165 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 166 #define TCP_KEEPALIVE_INTVL (75*HZ) 167 168 #define MAX_TCP_KEEPIDLE 32767 169 #define MAX_TCP_KEEPINTVL 32767 170 #define MAX_TCP_KEEPCNT 127 171 #define MAX_TCP_SYNCNT 127 172 173 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds 174 * to avoid overflows. This assumes a clock smaller than 1 Mhz. 175 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz. 176 */ 177 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC) 178 179 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 180 * after this time. It should be equal 181 * (or greater than) TCP_TIMEWAIT_LEN 182 * to provide reliability equal to one 183 * provided by timewait state. 184 */ 185 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 186 * timestamps. It must be less than 187 * minimal timewait lifetime. 188 */ 189 /* 190 * TCP option 191 */ 192 193 #define TCPOPT_NOP 1 /* Padding */ 194 #define TCPOPT_EOL 0 /* End of options */ 195 #define TCPOPT_MSS 2 /* Segment size negotiating */ 196 #define TCPOPT_WINDOW 3 /* Window scaling */ 197 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 198 #define TCPOPT_SACK 5 /* SACK Block */ 199 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 200 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 201 #define TCPOPT_AO 29 /* Authentication Option (RFC5925) */ 202 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 203 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 204 #define TCPOPT_EXP 254 /* Experimental */ 205 /* Magic number to be after the option value for sharing TCP 206 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 207 */ 208 #define TCPOPT_FASTOPEN_MAGIC 0xF989 209 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 210 211 /* 212 * TCP option lengths 213 */ 214 215 #define TCPOLEN_MSS 4 216 #define TCPOLEN_WINDOW 3 217 #define TCPOLEN_SACK_PERM 2 218 #define TCPOLEN_TIMESTAMP 10 219 #define TCPOLEN_MD5SIG 18 220 #define TCPOLEN_FASTOPEN_BASE 2 221 #define TCPOLEN_EXP_FASTOPEN_BASE 4 222 #define TCPOLEN_EXP_SMC_BASE 6 223 224 /* But this is what stacks really send out. */ 225 #define TCPOLEN_TSTAMP_ALIGNED 12 226 #define TCPOLEN_WSCALE_ALIGNED 4 227 #define TCPOLEN_SACKPERM_ALIGNED 4 228 #define TCPOLEN_SACK_BASE 2 229 #define TCPOLEN_SACK_BASE_ALIGNED 4 230 #define TCPOLEN_SACK_PERBLOCK 8 231 #define TCPOLEN_MD5SIG_ALIGNED 20 232 #define TCPOLEN_MSS_ALIGNED 4 233 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 234 235 /* Flags in tp->nonagle */ 236 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 237 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 238 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 239 240 /* TCP thin-stream limits */ 241 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 242 243 /* TCP initial congestion window as per rfc6928 */ 244 #define TCP_INIT_CWND 10 245 246 /* Bit Flags for sysctl_tcp_fastopen */ 247 #define TFO_CLIENT_ENABLE 1 248 #define TFO_SERVER_ENABLE 2 249 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 250 251 /* Accept SYN data w/o any cookie option */ 252 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 253 254 /* Force enable TFO on all listeners, i.e., not requiring the 255 * TCP_FASTOPEN socket option. 256 */ 257 #define TFO_SERVER_WO_SOCKOPT1 0x400 258 259 260 /* sysctl variables for tcp */ 261 extern int sysctl_tcp_max_orphans; 262 extern long sysctl_tcp_mem[3]; 263 264 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 265 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 266 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 267 268 extern atomic_long_t tcp_memory_allocated; 269 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 270 271 extern struct percpu_counter tcp_sockets_allocated; 272 extern unsigned long tcp_memory_pressure; 273 274 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 275 static inline bool tcp_under_memory_pressure(const struct sock *sk) 276 { 277 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 278 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 279 return true; 280 281 return READ_ONCE(tcp_memory_pressure); 282 } 283 /* 284 * The next routines deal with comparing 32 bit unsigned ints 285 * and worry about wraparound (automatic with unsigned arithmetic). 286 */ 287 288 static inline bool before(__u32 seq1, __u32 seq2) 289 { 290 return (__s32)(seq1-seq2) < 0; 291 } 292 #define after(seq2, seq1) before(seq1, seq2) 293 294 /* is s2<=s1<=s3 ? */ 295 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 296 { 297 return seq3 - seq2 >= seq1 - seq2; 298 } 299 300 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 301 { 302 sk_wmem_queued_add(sk, -skb->truesize); 303 if (!skb_zcopy_pure(skb)) 304 sk_mem_uncharge(sk, skb->truesize); 305 else 306 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 307 __kfree_skb(skb); 308 } 309 310 void sk_forced_mem_schedule(struct sock *sk, int size); 311 312 bool tcp_check_oom(const struct sock *sk, int shift); 313 314 315 extern struct proto tcp_prot; 316 317 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 318 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 319 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 320 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 321 322 void tcp_tasklet_init(void); 323 324 int tcp_v4_err(struct sk_buff *skb, u32); 325 326 void tcp_shutdown(struct sock *sk, int how); 327 328 int tcp_v4_early_demux(struct sk_buff *skb); 329 int tcp_v4_rcv(struct sk_buff *skb); 330 331 void tcp_remove_empty_skb(struct sock *sk); 332 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 333 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 334 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 335 size_t size, struct ubuf_info *uarg); 336 void tcp_splice_eof(struct socket *sock); 337 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 338 int tcp_wmem_schedule(struct sock *sk, int copy); 339 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 340 int size_goal); 341 void tcp_release_cb(struct sock *sk); 342 void tcp_wfree(struct sk_buff *skb); 343 void tcp_write_timer_handler(struct sock *sk); 344 void tcp_delack_timer_handler(struct sock *sk); 345 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 346 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 347 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 348 void tcp_rcv_space_adjust(struct sock *sk); 349 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 350 void tcp_twsk_destructor(struct sock *sk); 351 void tcp_twsk_purge(struct list_head *net_exit_list); 352 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 353 struct pipe_inode_info *pipe, size_t len, 354 unsigned int flags); 355 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 356 bool force_schedule); 357 358 static inline void tcp_dec_quickack_mode(struct sock *sk) 359 { 360 struct inet_connection_sock *icsk = inet_csk(sk); 361 362 if (icsk->icsk_ack.quick) { 363 /* How many ACKs S/ACKing new data have we sent? */ 364 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 365 366 if (pkts >= icsk->icsk_ack.quick) { 367 icsk->icsk_ack.quick = 0; 368 /* Leaving quickack mode we deflate ATO. */ 369 icsk->icsk_ack.ato = TCP_ATO_MIN; 370 } else 371 icsk->icsk_ack.quick -= pkts; 372 } 373 } 374 375 #define TCP_ECN_OK 1 376 #define TCP_ECN_QUEUE_CWR 2 377 #define TCP_ECN_DEMAND_CWR 4 378 #define TCP_ECN_SEEN 8 379 380 enum tcp_tw_status { 381 TCP_TW_SUCCESS = 0, 382 TCP_TW_RST = 1, 383 TCP_TW_ACK = 2, 384 TCP_TW_SYN = 3 385 }; 386 387 388 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 389 struct sk_buff *skb, 390 const struct tcphdr *th, 391 u32 *tw_isn); 392 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 393 struct request_sock *req, bool fastopen, 394 bool *lost_race); 395 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child, 396 struct sk_buff *skb); 397 void tcp_enter_loss(struct sock *sk); 398 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 399 void tcp_clear_retrans(struct tcp_sock *tp); 400 void tcp_update_metrics(struct sock *sk); 401 void tcp_init_metrics(struct sock *sk); 402 void tcp_metrics_init(void); 403 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 404 void __tcp_close(struct sock *sk, long timeout); 405 void tcp_close(struct sock *sk, long timeout); 406 void tcp_init_sock(struct sock *sk); 407 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 408 __poll_t tcp_poll(struct file *file, struct socket *sock, 409 struct poll_table_struct *wait); 410 int do_tcp_getsockopt(struct sock *sk, int level, 411 int optname, sockptr_t optval, sockptr_t optlen); 412 int tcp_getsockopt(struct sock *sk, int level, int optname, 413 char __user *optval, int __user *optlen); 414 bool tcp_bpf_bypass_getsockopt(int level, int optname); 415 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 416 sockptr_t optval, unsigned int optlen); 417 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 418 unsigned int optlen); 419 void tcp_set_keepalive(struct sock *sk, int val); 420 void tcp_syn_ack_timeout(const struct request_sock *req); 421 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 422 int flags, int *addr_len); 423 int tcp_set_rcvlowat(struct sock *sk, int val); 424 int tcp_set_window_clamp(struct sock *sk, int val); 425 void tcp_update_recv_tstamps(struct sk_buff *skb, 426 struct scm_timestamping_internal *tss); 427 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 428 struct scm_timestamping_internal *tss); 429 void tcp_data_ready(struct sock *sk); 430 #ifdef CONFIG_MMU 431 int tcp_mmap(struct file *file, struct socket *sock, 432 struct vm_area_struct *vma); 433 #endif 434 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 435 struct tcp_options_received *opt_rx, 436 int estab, struct tcp_fastopen_cookie *foc); 437 438 /* 439 * BPF SKB-less helpers 440 */ 441 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 442 struct tcphdr *th, u32 *cookie); 443 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 444 struct tcphdr *th, u32 *cookie); 445 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 446 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 447 const struct tcp_request_sock_ops *af_ops, 448 struct sock *sk, struct tcphdr *th); 449 /* 450 * TCP v4 functions exported for the inet6 API 451 */ 452 453 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 454 void tcp_v4_mtu_reduced(struct sock *sk); 455 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 456 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 457 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 458 struct sock *tcp_create_openreq_child(const struct sock *sk, 459 struct request_sock *req, 460 struct sk_buff *skb); 461 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 462 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 463 struct request_sock *req, 464 struct dst_entry *dst, 465 struct request_sock *req_unhash, 466 bool *own_req); 467 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 468 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 469 int tcp_connect(struct sock *sk); 470 enum tcp_synack_type { 471 TCP_SYNACK_NORMAL, 472 TCP_SYNACK_FASTOPEN, 473 TCP_SYNACK_COOKIE, 474 }; 475 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 476 struct request_sock *req, 477 struct tcp_fastopen_cookie *foc, 478 enum tcp_synack_type synack_type, 479 struct sk_buff *syn_skb); 480 int tcp_disconnect(struct sock *sk, int flags); 481 482 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 483 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 484 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 485 486 /* From syncookies.c */ 487 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 488 struct request_sock *req, 489 struct dst_entry *dst); 490 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th); 491 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 492 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 493 struct sock *sk, struct sk_buff *skb, 494 struct tcp_options_received *tcp_opt, 495 int mss, u32 tsoff); 496 497 #if IS_ENABLED(CONFIG_BPF) 498 struct bpf_tcp_req_attrs { 499 u32 rcv_tsval; 500 u32 rcv_tsecr; 501 u16 mss; 502 u8 rcv_wscale; 503 u8 snd_wscale; 504 u8 ecn_ok; 505 u8 wscale_ok; 506 u8 sack_ok; 507 u8 tstamp_ok; 508 u8 usec_ts_ok; 509 u8 reserved[3]; 510 }; 511 #endif 512 513 #ifdef CONFIG_SYN_COOKIES 514 515 /* Syncookies use a monotonic timer which increments every 60 seconds. 516 * This counter is used both as a hash input and partially encoded into 517 * the cookie value. A cookie is only validated further if the delta 518 * between the current counter value and the encoded one is less than this, 519 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 520 * the counter advances immediately after a cookie is generated). 521 */ 522 #define MAX_SYNCOOKIE_AGE 2 523 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 524 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 525 526 /* syncookies: remember time of last synqueue overflow 527 * But do not dirty this field too often (once per second is enough) 528 * It is racy as we do not hold a lock, but race is very minor. 529 */ 530 static inline void tcp_synq_overflow(const struct sock *sk) 531 { 532 unsigned int last_overflow; 533 unsigned int now = jiffies; 534 535 if (sk->sk_reuseport) { 536 struct sock_reuseport *reuse; 537 538 reuse = rcu_dereference(sk->sk_reuseport_cb); 539 if (likely(reuse)) { 540 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 541 if (!time_between32(now, last_overflow, 542 last_overflow + HZ)) 543 WRITE_ONCE(reuse->synq_overflow_ts, now); 544 return; 545 } 546 } 547 548 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 549 if (!time_between32(now, last_overflow, last_overflow + HZ)) 550 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 551 } 552 553 /* syncookies: no recent synqueue overflow on this listening socket? */ 554 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 555 { 556 unsigned int last_overflow; 557 unsigned int now = jiffies; 558 559 if (sk->sk_reuseport) { 560 struct sock_reuseport *reuse; 561 562 reuse = rcu_dereference(sk->sk_reuseport_cb); 563 if (likely(reuse)) { 564 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 565 return !time_between32(now, last_overflow - HZ, 566 last_overflow + 567 TCP_SYNCOOKIE_VALID); 568 } 569 } 570 571 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 572 573 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 574 * then we're under synflood. However, we have to use 575 * 'last_overflow - HZ' as lower bound. That's because a concurrent 576 * tcp_synq_overflow() could update .ts_recent_stamp after we read 577 * jiffies but before we store .ts_recent_stamp into last_overflow, 578 * which could lead to rejecting a valid syncookie. 579 */ 580 return !time_between32(now, last_overflow - HZ, 581 last_overflow + TCP_SYNCOOKIE_VALID); 582 } 583 584 static inline u32 tcp_cookie_time(void) 585 { 586 u64 val = get_jiffies_64(); 587 588 do_div(val, TCP_SYNCOOKIE_PERIOD); 589 return val; 590 } 591 592 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */ 593 static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val) 594 { 595 if (usec_ts) 596 return div_u64(val, NSEC_PER_USEC); 597 598 return div_u64(val, NSEC_PER_MSEC); 599 } 600 601 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 602 u16 *mssp); 603 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 604 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 605 bool cookie_timestamp_decode(const struct net *net, 606 struct tcp_options_received *opt); 607 608 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst) 609 { 610 return READ_ONCE(net->ipv4.sysctl_tcp_ecn) || 611 dst_feature(dst, RTAX_FEATURE_ECN); 612 } 613 614 #if IS_ENABLED(CONFIG_BPF) 615 static inline bool cookie_bpf_ok(struct sk_buff *skb) 616 { 617 return skb->sk; 618 } 619 620 struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb); 621 #else 622 static inline bool cookie_bpf_ok(struct sk_buff *skb) 623 { 624 return false; 625 } 626 627 static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk, 628 struct sk_buff *skb) 629 { 630 return NULL; 631 } 632 #endif 633 634 /* From net/ipv6/syncookies.c */ 635 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th); 636 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 637 638 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 639 const struct tcphdr *th, u16 *mssp); 640 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 641 #endif 642 /* tcp_output.c */ 643 644 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 645 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 646 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 647 int nonagle); 648 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 649 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 650 void tcp_retransmit_timer(struct sock *sk); 651 void tcp_xmit_retransmit_queue(struct sock *); 652 void tcp_simple_retransmit(struct sock *); 653 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 654 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 655 enum tcp_queue { 656 TCP_FRAG_IN_WRITE_QUEUE, 657 TCP_FRAG_IN_RTX_QUEUE, 658 }; 659 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 660 struct sk_buff *skb, u32 len, 661 unsigned int mss_now, gfp_t gfp); 662 663 void tcp_send_probe0(struct sock *); 664 int tcp_write_wakeup(struct sock *, int mib); 665 void tcp_send_fin(struct sock *sk); 666 void tcp_send_active_reset(struct sock *sk, gfp_t priority, 667 enum sk_rst_reason reason); 668 int tcp_send_synack(struct sock *); 669 void tcp_push_one(struct sock *, unsigned int mss_now); 670 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 671 void tcp_send_ack(struct sock *sk); 672 void tcp_send_delayed_ack(struct sock *sk); 673 void tcp_send_loss_probe(struct sock *sk); 674 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 675 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 676 const struct sk_buff *next_skb); 677 678 /* tcp_input.c */ 679 void tcp_rearm_rto(struct sock *sk); 680 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 681 void tcp_done_with_error(struct sock *sk, int err); 682 void tcp_reset(struct sock *sk, struct sk_buff *skb); 683 void tcp_fin(struct sock *sk); 684 void tcp_check_space(struct sock *sk); 685 void tcp_sack_compress_send_ack(struct sock *sk); 686 687 static inline void tcp_cleanup_skb(struct sk_buff *skb) 688 { 689 skb_dst_drop(skb); 690 secpath_reset(skb); 691 } 692 693 static inline void tcp_add_receive_queue(struct sock *sk, struct sk_buff *skb) 694 { 695 DEBUG_NET_WARN_ON_ONCE(skb_dst(skb)); 696 DEBUG_NET_WARN_ON_ONCE(secpath_exists(skb)); 697 __skb_queue_tail(&sk->sk_receive_queue, skb); 698 } 699 700 /* tcp_timer.c */ 701 void tcp_init_xmit_timers(struct sock *); 702 static inline void tcp_clear_xmit_timers(struct sock *sk) 703 { 704 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 705 __sock_put(sk); 706 707 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 708 __sock_put(sk); 709 710 inet_csk_clear_xmit_timers(sk); 711 } 712 713 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 714 unsigned int tcp_current_mss(struct sock *sk); 715 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 716 717 /* Bound MSS / TSO packet size with the half of the window */ 718 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 719 { 720 int cutoff; 721 722 /* When peer uses tiny windows, there is no use in packetizing 723 * to sub-MSS pieces for the sake of SWS or making sure there 724 * are enough packets in the pipe for fast recovery. 725 * 726 * On the other hand, for extremely large MSS devices, handling 727 * smaller than MSS windows in this way does make sense. 728 */ 729 if (tp->max_window > TCP_MSS_DEFAULT) 730 cutoff = (tp->max_window >> 1); 731 else 732 cutoff = tp->max_window; 733 734 if (cutoff && pktsize > cutoff) 735 return max_t(int, cutoff, 68U - tp->tcp_header_len); 736 else 737 return pktsize; 738 } 739 740 /* tcp.c */ 741 void tcp_get_info(struct sock *, struct tcp_info *); 742 743 /* Read 'sendfile()'-style from a TCP socket */ 744 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 745 sk_read_actor_t recv_actor); 746 int tcp_read_sock_noack(struct sock *sk, read_descriptor_t *desc, 747 sk_read_actor_t recv_actor, bool noack, 748 u32 *copied_seq); 749 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 750 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 751 void tcp_read_done(struct sock *sk, size_t len); 752 753 void tcp_initialize_rcv_mss(struct sock *sk); 754 755 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 756 int tcp_mss_to_mtu(struct sock *sk, int mss); 757 void tcp_mtup_init(struct sock *sk); 758 759 static inline void tcp_bound_rto(struct sock *sk) 760 { 761 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 762 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 763 } 764 765 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 766 { 767 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 768 } 769 770 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 771 { 772 /* mptcp hooks are only on the slow path */ 773 if (sk_is_mptcp((struct sock *)tp)) 774 return; 775 776 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 777 ntohl(TCP_FLAG_ACK) | 778 snd_wnd); 779 } 780 781 static inline void tcp_fast_path_on(struct tcp_sock *tp) 782 { 783 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 784 } 785 786 static inline void tcp_fast_path_check(struct sock *sk) 787 { 788 struct tcp_sock *tp = tcp_sk(sk); 789 790 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 791 tp->rcv_wnd && 792 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 793 !tp->urg_data) 794 tcp_fast_path_on(tp); 795 } 796 797 u32 tcp_delack_max(const struct sock *sk); 798 799 /* Compute the actual rto_min value */ 800 static inline u32 tcp_rto_min(const struct sock *sk) 801 { 802 const struct dst_entry *dst = __sk_dst_get(sk); 803 u32 rto_min = inet_csk(sk)->icsk_rto_min; 804 805 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 806 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 807 return rto_min; 808 } 809 810 static inline u32 tcp_rto_min_us(const struct sock *sk) 811 { 812 return jiffies_to_usecs(tcp_rto_min(sk)); 813 } 814 815 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 816 { 817 return dst_metric_locked(dst, RTAX_CC_ALGO); 818 } 819 820 /* Minimum RTT in usec. ~0 means not available. */ 821 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 822 { 823 return minmax_get(&tp->rtt_min); 824 } 825 826 /* Compute the actual receive window we are currently advertising. 827 * Rcv_nxt can be after the window if our peer push more data 828 * than the offered window. 829 */ 830 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 831 { 832 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 833 834 if (win < 0) 835 win = 0; 836 return (u32) win; 837 } 838 839 /* Choose a new window, without checks for shrinking, and without 840 * scaling applied to the result. The caller does these things 841 * if necessary. This is a "raw" window selection. 842 */ 843 u32 __tcp_select_window(struct sock *sk); 844 845 void tcp_send_window_probe(struct sock *sk); 846 847 /* TCP uses 32bit jiffies to save some space. 848 * Note that this is different from tcp_time_stamp, which 849 * historically has been the same until linux-4.13. 850 */ 851 #define tcp_jiffies32 ((u32)jiffies) 852 853 /* 854 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 855 * It is no longer tied to jiffies, but to 1 ms clock. 856 * Note: double check if you want to use tcp_jiffies32 instead of this. 857 */ 858 #define TCP_TS_HZ 1000 859 860 static inline u64 tcp_clock_ns(void) 861 { 862 return ktime_get_ns(); 863 } 864 865 static inline u64 tcp_clock_us(void) 866 { 867 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 868 } 869 870 static inline u64 tcp_clock_ms(void) 871 { 872 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 873 } 874 875 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 876 * or usec resolution. Each socket carries a flag to select one or other 877 * resolution, as the route attribute could change anytime. 878 * Each flow must stick to initial resolution. 879 */ 880 static inline u32 tcp_clock_ts(bool usec_ts) 881 { 882 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 883 } 884 885 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 886 { 887 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 888 } 889 890 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 891 { 892 if (tp->tcp_usec_ts) 893 return tp->tcp_mstamp; 894 return tcp_time_stamp_ms(tp); 895 } 896 897 void tcp_mstamp_refresh(struct tcp_sock *tp); 898 899 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 900 { 901 return max_t(s64, t1 - t0, 0); 902 } 903 904 /* provide the departure time in us unit */ 905 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 906 { 907 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 908 } 909 910 /* Provide skb TSval in usec or ms unit */ 911 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 912 { 913 if (usec_ts) 914 return tcp_skb_timestamp_us(skb); 915 916 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 917 } 918 919 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 920 { 921 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 922 } 923 924 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 925 { 926 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 927 } 928 929 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 930 931 #define TCPHDR_FIN 0x01 932 #define TCPHDR_SYN 0x02 933 #define TCPHDR_RST 0x04 934 #define TCPHDR_PSH 0x08 935 #define TCPHDR_ACK 0x10 936 #define TCPHDR_URG 0x20 937 #define TCPHDR_ECE 0x40 938 #define TCPHDR_CWR 0x80 939 940 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 941 942 /* State flags for sacked in struct tcp_skb_cb */ 943 enum tcp_skb_cb_sacked_flags { 944 TCPCB_SACKED_ACKED = (1 << 0), /* SKB ACK'd by a SACK block */ 945 TCPCB_SACKED_RETRANS = (1 << 1), /* SKB retransmitted */ 946 TCPCB_LOST = (1 << 2), /* SKB is lost */ 947 TCPCB_TAGBITS = (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS | 948 TCPCB_LOST), /* All tag bits */ 949 TCPCB_REPAIRED = (1 << 4), /* SKB repaired (no skb_mstamp_ns) */ 950 TCPCB_EVER_RETRANS = (1 << 7), /* Ever retransmitted frame */ 951 TCPCB_RETRANS = (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS | 952 TCPCB_REPAIRED), 953 }; 954 955 /* This is what the send packet queuing engine uses to pass 956 * TCP per-packet control information to the transmission code. 957 * We also store the host-order sequence numbers in here too. 958 * This is 44 bytes if IPV6 is enabled. 959 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 960 */ 961 struct tcp_skb_cb { 962 __u32 seq; /* Starting sequence number */ 963 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 964 union { 965 /* Note : 966 * tcp_gso_segs/size are used in write queue only, 967 * cf tcp_skb_pcount()/tcp_skb_mss() 968 */ 969 struct { 970 u16 tcp_gso_segs; 971 u16 tcp_gso_size; 972 }; 973 }; 974 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 975 976 __u8 sacked; /* State flags for SACK. */ 977 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 978 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 979 eor:1, /* Is skb MSG_EOR marked? */ 980 has_rxtstamp:1, /* SKB has a RX timestamp */ 981 unused:5; 982 __u32 ack_seq; /* Sequence number ACK'd */ 983 union { 984 struct { 985 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 986 /* There is space for up to 24 bytes */ 987 __u32 is_app_limited:1, /* cwnd not fully used? */ 988 delivered_ce:20, 989 unused:11; 990 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 991 __u32 delivered; 992 /* start of send pipeline phase */ 993 u64 first_tx_mstamp; 994 /* when we reached the "delivered" count */ 995 u64 delivered_mstamp; 996 } tx; /* only used for outgoing skbs */ 997 union { 998 struct inet_skb_parm h4; 999 #if IS_ENABLED(CONFIG_IPV6) 1000 struct inet6_skb_parm h6; 1001 #endif 1002 } header; /* For incoming skbs */ 1003 }; 1004 }; 1005 1006 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 1007 1008 extern const struct inet_connection_sock_af_ops ipv4_specific; 1009 1010 #if IS_ENABLED(CONFIG_IPV6) 1011 /* This is the variant of inet6_iif() that must be used by TCP, 1012 * as TCP moves IP6CB into a different location in skb->cb[] 1013 */ 1014 static inline int tcp_v6_iif(const struct sk_buff *skb) 1015 { 1016 return TCP_SKB_CB(skb)->header.h6.iif; 1017 } 1018 1019 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 1020 { 1021 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 1022 1023 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 1024 } 1025 1026 /* TCP_SKB_CB reference means this can not be used from early demux */ 1027 static inline int tcp_v6_sdif(const struct sk_buff *skb) 1028 { 1029 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1030 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 1031 return TCP_SKB_CB(skb)->header.h6.iif; 1032 #endif 1033 return 0; 1034 } 1035 1036 extern const struct inet_connection_sock_af_ops ipv6_specific; 1037 1038 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 1039 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 1040 void tcp_v6_early_demux(struct sk_buff *skb); 1041 1042 #endif 1043 1044 /* TCP_SKB_CB reference means this can not be used from early demux */ 1045 static inline int tcp_v4_sdif(struct sk_buff *skb) 1046 { 1047 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1048 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 1049 return TCP_SKB_CB(skb)->header.h4.iif; 1050 #endif 1051 return 0; 1052 } 1053 1054 /* Due to TSO, an SKB can be composed of multiple actual 1055 * packets. To keep these tracked properly, we use this. 1056 */ 1057 static inline int tcp_skb_pcount(const struct sk_buff *skb) 1058 { 1059 return TCP_SKB_CB(skb)->tcp_gso_segs; 1060 } 1061 1062 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 1063 { 1064 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1065 } 1066 1067 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1068 { 1069 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1070 } 1071 1072 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1073 static inline int tcp_skb_mss(const struct sk_buff *skb) 1074 { 1075 return TCP_SKB_CB(skb)->tcp_gso_size; 1076 } 1077 1078 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1079 { 1080 return likely(!TCP_SKB_CB(skb)->eor); 1081 } 1082 1083 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1084 const struct sk_buff *from) 1085 { 1086 /* skb_cmp_decrypted() not needed, use tcp_write_collapse_fence() */ 1087 return likely(tcp_skb_can_collapse_to(to) && 1088 mptcp_skb_can_collapse(to, from) && 1089 skb_pure_zcopy_same(to, from) && 1090 skb_frags_readable(to) == skb_frags_readable(from)); 1091 } 1092 1093 static inline bool tcp_skb_can_collapse_rx(const struct sk_buff *to, 1094 const struct sk_buff *from) 1095 { 1096 return likely(mptcp_skb_can_collapse(to, from) && 1097 !skb_cmp_decrypted(to, from)); 1098 } 1099 1100 /* Events passed to congestion control interface */ 1101 enum tcp_ca_event { 1102 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1103 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1104 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1105 CA_EVENT_LOSS, /* loss timeout */ 1106 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1107 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1108 }; 1109 1110 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1111 enum tcp_ca_ack_event_flags { 1112 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1113 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1114 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1115 }; 1116 1117 /* 1118 * Interface for adding new TCP congestion control handlers 1119 */ 1120 #define TCP_CA_NAME_MAX 16 1121 #define TCP_CA_MAX 128 1122 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1123 1124 #define TCP_CA_UNSPEC 0 1125 1126 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1127 #define TCP_CONG_NON_RESTRICTED 0x1 1128 /* Requires ECN/ECT set on all packets */ 1129 #define TCP_CONG_NEEDS_ECN 0x2 1130 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1131 1132 union tcp_cc_info; 1133 1134 struct ack_sample { 1135 u32 pkts_acked; 1136 s32 rtt_us; 1137 u32 in_flight; 1138 }; 1139 1140 /* A rate sample measures the number of (original/retransmitted) data 1141 * packets delivered "delivered" over an interval of time "interval_us". 1142 * The tcp_rate.c code fills in the rate sample, and congestion 1143 * control modules that define a cong_control function to run at the end 1144 * of ACK processing can optionally chose to consult this sample when 1145 * setting cwnd and pacing rate. 1146 * A sample is invalid if "delivered" or "interval_us" is negative. 1147 */ 1148 struct rate_sample { 1149 u64 prior_mstamp; /* starting timestamp for interval */ 1150 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1151 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1152 s32 delivered; /* number of packets delivered over interval */ 1153 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1154 long interval_us; /* time for tp->delivered to incr "delivered" */ 1155 u32 snd_interval_us; /* snd interval for delivered packets */ 1156 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1157 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1158 int losses; /* number of packets marked lost upon ACK */ 1159 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1160 u32 prior_in_flight; /* in flight before this ACK */ 1161 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1162 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1163 bool is_retrans; /* is sample from retransmission? */ 1164 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1165 }; 1166 1167 struct tcp_congestion_ops { 1168 /* fast path fields are put first to fill one cache line */ 1169 1170 /* return slow start threshold (required) */ 1171 u32 (*ssthresh)(struct sock *sk); 1172 1173 /* do new cwnd calculation (required) */ 1174 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1175 1176 /* call before changing ca_state (optional) */ 1177 void (*set_state)(struct sock *sk, u8 new_state); 1178 1179 /* call when cwnd event occurs (optional) */ 1180 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1181 1182 /* call when ack arrives (optional) */ 1183 void (*in_ack_event)(struct sock *sk, u32 flags); 1184 1185 /* hook for packet ack accounting (optional) */ 1186 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1187 1188 /* override sysctl_tcp_min_tso_segs */ 1189 u32 (*min_tso_segs)(struct sock *sk); 1190 1191 /* call when packets are delivered to update cwnd and pacing rate, 1192 * after all the ca_state processing. (optional) 1193 */ 1194 void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs); 1195 1196 1197 /* new value of cwnd after loss (required) */ 1198 u32 (*undo_cwnd)(struct sock *sk); 1199 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1200 u32 (*sndbuf_expand)(struct sock *sk); 1201 1202 /* control/slow paths put last */ 1203 /* get info for inet_diag (optional) */ 1204 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1205 union tcp_cc_info *info); 1206 1207 char name[TCP_CA_NAME_MAX]; 1208 struct module *owner; 1209 struct list_head list; 1210 u32 key; 1211 u32 flags; 1212 1213 /* initialize private data (optional) */ 1214 void (*init)(struct sock *sk); 1215 /* cleanup private data (optional) */ 1216 void (*release)(struct sock *sk); 1217 } ____cacheline_aligned_in_smp; 1218 1219 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1220 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1221 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1222 struct tcp_congestion_ops *old_type); 1223 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1224 1225 void tcp_assign_congestion_control(struct sock *sk); 1226 void tcp_init_congestion_control(struct sock *sk); 1227 void tcp_cleanup_congestion_control(struct sock *sk); 1228 int tcp_set_default_congestion_control(struct net *net, const char *name); 1229 void tcp_get_default_congestion_control(struct net *net, char *name); 1230 void tcp_get_available_congestion_control(char *buf, size_t len); 1231 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1232 int tcp_set_allowed_congestion_control(char *allowed); 1233 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1234 bool cap_net_admin); 1235 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1236 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1237 1238 u32 tcp_reno_ssthresh(struct sock *sk); 1239 u32 tcp_reno_undo_cwnd(struct sock *sk); 1240 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1241 extern struct tcp_congestion_ops tcp_reno; 1242 1243 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1244 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1245 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1246 #ifdef CONFIG_INET 1247 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1248 #else 1249 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1250 { 1251 return NULL; 1252 } 1253 #endif 1254 1255 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1256 { 1257 const struct inet_connection_sock *icsk = inet_csk(sk); 1258 1259 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1260 } 1261 1262 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1263 { 1264 const struct inet_connection_sock *icsk = inet_csk(sk); 1265 1266 if (icsk->icsk_ca_ops->cwnd_event) 1267 icsk->icsk_ca_ops->cwnd_event(sk, event); 1268 } 1269 1270 /* From tcp_cong.c */ 1271 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1272 1273 /* From tcp_rate.c */ 1274 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1275 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1276 struct rate_sample *rs); 1277 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1278 bool is_sack_reneg, struct rate_sample *rs); 1279 void tcp_rate_check_app_limited(struct sock *sk); 1280 1281 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1282 { 1283 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1284 } 1285 1286 /* These functions determine how the current flow behaves in respect of SACK 1287 * handling. SACK is negotiated with the peer, and therefore it can vary 1288 * between different flows. 1289 * 1290 * tcp_is_sack - SACK enabled 1291 * tcp_is_reno - No SACK 1292 */ 1293 static inline int tcp_is_sack(const struct tcp_sock *tp) 1294 { 1295 return likely(tp->rx_opt.sack_ok); 1296 } 1297 1298 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1299 { 1300 return !tcp_is_sack(tp); 1301 } 1302 1303 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1304 { 1305 return tp->sacked_out + tp->lost_out; 1306 } 1307 1308 /* This determines how many packets are "in the network" to the best 1309 * of our knowledge. In many cases it is conservative, but where 1310 * detailed information is available from the receiver (via SACK 1311 * blocks etc.) we can make more aggressive calculations. 1312 * 1313 * Use this for decisions involving congestion control, use just 1314 * tp->packets_out to determine if the send queue is empty or not. 1315 * 1316 * Read this equation as: 1317 * 1318 * "Packets sent once on transmission queue" MINUS 1319 * "Packets left network, but not honestly ACKed yet" PLUS 1320 * "Packets fast retransmitted" 1321 */ 1322 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1323 { 1324 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1325 } 1326 1327 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1328 1329 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1330 { 1331 return tp->snd_cwnd; 1332 } 1333 1334 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1335 { 1336 WARN_ON_ONCE((int)val <= 0); 1337 tp->snd_cwnd = val; 1338 } 1339 1340 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1341 { 1342 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1343 } 1344 1345 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1346 { 1347 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1348 } 1349 1350 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1351 { 1352 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1353 (1 << inet_csk(sk)->icsk_ca_state); 1354 } 1355 1356 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1357 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1358 * ssthresh. 1359 */ 1360 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1361 { 1362 const struct tcp_sock *tp = tcp_sk(sk); 1363 1364 if (tcp_in_cwnd_reduction(sk)) 1365 return tp->snd_ssthresh; 1366 else 1367 return max(tp->snd_ssthresh, 1368 ((tcp_snd_cwnd(tp) >> 1) + 1369 (tcp_snd_cwnd(tp) >> 2))); 1370 } 1371 1372 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1373 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1374 1375 void tcp_enter_cwr(struct sock *sk); 1376 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1377 1378 /* The maximum number of MSS of available cwnd for which TSO defers 1379 * sending if not using sysctl_tcp_tso_win_divisor. 1380 */ 1381 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1382 { 1383 return 3; 1384 } 1385 1386 /* Returns end sequence number of the receiver's advertised window */ 1387 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1388 { 1389 return tp->snd_una + tp->snd_wnd; 1390 } 1391 1392 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1393 * flexible approach. The RFC suggests cwnd should not be raised unless 1394 * it was fully used previously. And that's exactly what we do in 1395 * congestion avoidance mode. But in slow start we allow cwnd to grow 1396 * as long as the application has used half the cwnd. 1397 * Example : 1398 * cwnd is 10 (IW10), but application sends 9 frames. 1399 * We allow cwnd to reach 18 when all frames are ACKed. 1400 * This check is safe because it's as aggressive as slow start which already 1401 * risks 100% overshoot. The advantage is that we discourage application to 1402 * either send more filler packets or data to artificially blow up the cwnd 1403 * usage, and allow application-limited process to probe bw more aggressively. 1404 */ 1405 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1406 { 1407 const struct tcp_sock *tp = tcp_sk(sk); 1408 1409 if (tp->is_cwnd_limited) 1410 return true; 1411 1412 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1413 if (tcp_in_slow_start(tp)) 1414 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1415 1416 return false; 1417 } 1418 1419 /* BBR congestion control needs pacing. 1420 * Same remark for SO_MAX_PACING_RATE. 1421 * sch_fq packet scheduler is efficiently handling pacing, 1422 * but is not always installed/used. 1423 * Return true if TCP stack should pace packets itself. 1424 */ 1425 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1426 { 1427 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1428 } 1429 1430 /* Estimates in how many jiffies next packet for this flow can be sent. 1431 * Scheduling a retransmit timer too early would be silly. 1432 */ 1433 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1434 { 1435 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1436 1437 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1438 } 1439 1440 static inline void tcp_reset_xmit_timer(struct sock *sk, 1441 const int what, 1442 unsigned long when, 1443 const unsigned long max_when) 1444 { 1445 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1446 max_when); 1447 } 1448 1449 /* Something is really bad, we could not queue an additional packet, 1450 * because qdisc is full or receiver sent a 0 window, or we are paced. 1451 * We do not want to add fuel to the fire, or abort too early, 1452 * so make sure the timer we arm now is at least 200ms in the future, 1453 * regardless of current icsk_rto value (as it could be ~2ms) 1454 */ 1455 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1456 { 1457 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1458 } 1459 1460 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1461 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1462 unsigned long max_when) 1463 { 1464 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1465 inet_csk(sk)->icsk_backoff); 1466 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1467 1468 return (unsigned long)min_t(u64, when, max_when); 1469 } 1470 1471 static inline void tcp_check_probe_timer(struct sock *sk) 1472 { 1473 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1474 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1475 tcp_probe0_base(sk), TCP_RTO_MAX); 1476 } 1477 1478 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1479 { 1480 tp->snd_wl1 = seq; 1481 } 1482 1483 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1484 { 1485 tp->snd_wl1 = seq; 1486 } 1487 1488 /* 1489 * Calculate(/check) TCP checksum 1490 */ 1491 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1492 __be32 daddr, __wsum base) 1493 { 1494 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1495 } 1496 1497 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1498 { 1499 return !skb_csum_unnecessary(skb) && 1500 __skb_checksum_complete(skb); 1501 } 1502 1503 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1504 enum skb_drop_reason *reason); 1505 1506 1507 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1508 void tcp_set_state(struct sock *sk, int state); 1509 void tcp_done(struct sock *sk); 1510 int tcp_abort(struct sock *sk, int err); 1511 1512 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1513 { 1514 rx_opt->dsack = 0; 1515 rx_opt->num_sacks = 0; 1516 } 1517 1518 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1519 1520 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1521 { 1522 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1523 struct tcp_sock *tp = tcp_sk(sk); 1524 s32 delta; 1525 1526 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1527 tp->packets_out || ca_ops->cong_control) 1528 return; 1529 delta = tcp_jiffies32 - tp->lsndtime; 1530 if (delta > inet_csk(sk)->icsk_rto) 1531 tcp_cwnd_restart(sk, delta); 1532 } 1533 1534 /* Determine a window scaling and initial window to offer. */ 1535 void tcp_select_initial_window(const struct sock *sk, int __space, 1536 __u32 mss, __u32 *rcv_wnd, 1537 __u32 *window_clamp, int wscale_ok, 1538 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1539 1540 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1541 { 1542 s64 scaled_space = (s64)space * scaling_ratio; 1543 1544 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1545 } 1546 1547 static inline int tcp_win_from_space(const struct sock *sk, int space) 1548 { 1549 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1550 } 1551 1552 /* inverse of __tcp_win_from_space() */ 1553 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1554 { 1555 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1556 1557 do_div(val, scaling_ratio); 1558 return val; 1559 } 1560 1561 static inline int tcp_space_from_win(const struct sock *sk, int win) 1562 { 1563 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1564 } 1565 1566 /* Assume a 50% default for skb->len/skb->truesize ratio. 1567 * This may be adjusted later in tcp_measure_rcv_mss(). 1568 */ 1569 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1)) 1570 1571 static inline void tcp_scaling_ratio_init(struct sock *sk) 1572 { 1573 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1574 } 1575 1576 /* Note: caller must be prepared to deal with negative returns */ 1577 static inline int tcp_space(const struct sock *sk) 1578 { 1579 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1580 READ_ONCE(sk->sk_backlog.len) - 1581 atomic_read(&sk->sk_rmem_alloc)); 1582 } 1583 1584 static inline int tcp_full_space(const struct sock *sk) 1585 { 1586 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1587 } 1588 1589 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh) 1590 { 1591 int unused_mem = sk_unused_reserved_mem(sk); 1592 struct tcp_sock *tp = tcp_sk(sk); 1593 1594 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh); 1595 if (unused_mem) 1596 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1597 tcp_win_from_space(sk, unused_mem)); 1598 } 1599 1600 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1601 { 1602 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss); 1603 } 1604 1605 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1606 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1607 1608 1609 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1610 * If 87.5 % (7/8) of the space has been consumed, we want to override 1611 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1612 * len/truesize ratio. 1613 */ 1614 static inline bool tcp_rmem_pressure(const struct sock *sk) 1615 { 1616 int rcvbuf, threshold; 1617 1618 if (tcp_under_memory_pressure(sk)) 1619 return true; 1620 1621 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1622 threshold = rcvbuf - (rcvbuf >> 3); 1623 1624 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1625 } 1626 1627 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1628 { 1629 const struct tcp_sock *tp = tcp_sk(sk); 1630 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1631 1632 if (avail <= 0) 1633 return false; 1634 1635 return (avail >= target) || tcp_rmem_pressure(sk) || 1636 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1637 } 1638 1639 extern void tcp_openreq_init_rwin(struct request_sock *req, 1640 const struct sock *sk_listener, 1641 const struct dst_entry *dst); 1642 1643 void tcp_enter_memory_pressure(struct sock *sk); 1644 void tcp_leave_memory_pressure(struct sock *sk); 1645 1646 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1647 { 1648 struct net *net = sock_net((struct sock *)tp); 1649 int val; 1650 1651 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1652 * and do_tcp_setsockopt(). 1653 */ 1654 val = READ_ONCE(tp->keepalive_intvl); 1655 1656 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1657 } 1658 1659 static inline int keepalive_time_when(const struct tcp_sock *tp) 1660 { 1661 struct net *net = sock_net((struct sock *)tp); 1662 int val; 1663 1664 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1665 val = READ_ONCE(tp->keepalive_time); 1666 1667 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1668 } 1669 1670 static inline int keepalive_probes(const struct tcp_sock *tp) 1671 { 1672 struct net *net = sock_net((struct sock *)tp); 1673 int val; 1674 1675 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1676 * and do_tcp_setsockopt(). 1677 */ 1678 val = READ_ONCE(tp->keepalive_probes); 1679 1680 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1681 } 1682 1683 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1684 { 1685 const struct inet_connection_sock *icsk = &tp->inet_conn; 1686 1687 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1688 tcp_jiffies32 - tp->rcv_tstamp); 1689 } 1690 1691 static inline int tcp_fin_time(const struct sock *sk) 1692 { 1693 int fin_timeout = tcp_sk(sk)->linger2 ? : 1694 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1695 const int rto = inet_csk(sk)->icsk_rto; 1696 1697 if (fin_timeout < (rto << 2) - (rto >> 1)) 1698 fin_timeout = (rto << 2) - (rto >> 1); 1699 1700 return fin_timeout; 1701 } 1702 1703 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1704 int paws_win) 1705 { 1706 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1707 return true; 1708 if (unlikely(!time_before32(ktime_get_seconds(), 1709 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1710 return true; 1711 /* 1712 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1713 * then following tcp messages have valid values. Ignore 0 value, 1714 * or else 'negative' tsval might forbid us to accept their packets. 1715 */ 1716 if (!rx_opt->ts_recent) 1717 return true; 1718 return false; 1719 } 1720 1721 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1722 int rst) 1723 { 1724 if (tcp_paws_check(rx_opt, 0)) 1725 return false; 1726 1727 /* RST segments are not recommended to carry timestamp, 1728 and, if they do, it is recommended to ignore PAWS because 1729 "their cleanup function should take precedence over timestamps." 1730 Certainly, it is mistake. It is necessary to understand the reasons 1731 of this constraint to relax it: if peer reboots, clock may go 1732 out-of-sync and half-open connections will not be reset. 1733 Actually, the problem would be not existing if all 1734 the implementations followed draft about maintaining clock 1735 via reboots. Linux-2.2 DOES NOT! 1736 1737 However, we can relax time bounds for RST segments to MSL. 1738 */ 1739 if (rst && !time_before32(ktime_get_seconds(), 1740 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1741 return false; 1742 return true; 1743 } 1744 1745 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1746 int mib_idx, u32 *last_oow_ack_time); 1747 1748 static inline void tcp_mib_init(struct net *net) 1749 { 1750 /* See RFC 2012 */ 1751 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1752 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1753 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1754 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1755 } 1756 1757 /* from STCP */ 1758 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1759 { 1760 tp->lost_skb_hint = NULL; 1761 } 1762 1763 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1764 { 1765 tcp_clear_retrans_hints_partial(tp); 1766 tp->retransmit_skb_hint = NULL; 1767 } 1768 1769 #define tcp_md5_addr tcp_ao_addr 1770 1771 /* - key database */ 1772 struct tcp_md5sig_key { 1773 struct hlist_node node; 1774 u8 keylen; 1775 u8 family; /* AF_INET or AF_INET6 */ 1776 u8 prefixlen; 1777 u8 flags; 1778 union tcp_md5_addr addr; 1779 int l3index; /* set if key added with L3 scope */ 1780 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1781 struct rcu_head rcu; 1782 }; 1783 1784 /* - sock block */ 1785 struct tcp_md5sig_info { 1786 struct hlist_head head; 1787 struct rcu_head rcu; 1788 }; 1789 1790 /* - pseudo header */ 1791 struct tcp4_pseudohdr { 1792 __be32 saddr; 1793 __be32 daddr; 1794 __u8 pad; 1795 __u8 protocol; 1796 __be16 len; 1797 }; 1798 1799 struct tcp6_pseudohdr { 1800 struct in6_addr saddr; 1801 struct in6_addr daddr; 1802 __be32 len; 1803 __be32 protocol; /* including padding */ 1804 }; 1805 1806 union tcp_md5sum_block { 1807 struct tcp4_pseudohdr ip4; 1808 #if IS_ENABLED(CONFIG_IPV6) 1809 struct tcp6_pseudohdr ip6; 1810 #endif 1811 }; 1812 1813 /* 1814 * struct tcp_sigpool - per-CPU pool of ahash_requests 1815 * @scratch: per-CPU temporary area, that can be used between 1816 * tcp_sigpool_start() and tcp_sigpool_end() to perform 1817 * crypto request 1818 * @req: pre-allocated ahash request 1819 */ 1820 struct tcp_sigpool { 1821 void *scratch; 1822 struct ahash_request *req; 1823 }; 1824 1825 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size); 1826 void tcp_sigpool_get(unsigned int id); 1827 void tcp_sigpool_release(unsigned int id); 1828 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp, 1829 const struct sk_buff *skb, 1830 unsigned int header_len); 1831 1832 /** 1833 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash 1834 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash() 1835 * @c: returned tcp_sigpool for usage (uninitialized on failure) 1836 * 1837 * Returns: 0 on success, error otherwise. 1838 */ 1839 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c); 1840 /** 1841 * tcp_sigpool_end - enable bh and stop using tcp_sigpool 1842 * @c: tcp_sigpool context that was returned by tcp_sigpool_start() 1843 */ 1844 void tcp_sigpool_end(struct tcp_sigpool *c); 1845 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len); 1846 /* - functions */ 1847 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1848 const struct sock *sk, const struct sk_buff *skb); 1849 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1850 int family, u8 prefixlen, int l3index, u8 flags, 1851 const u8 *newkey, u8 newkeylen); 1852 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1853 int family, u8 prefixlen, int l3index, 1854 struct tcp_md5sig_key *key); 1855 1856 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1857 int family, u8 prefixlen, int l3index, u8 flags); 1858 void tcp_clear_md5_list(struct sock *sk); 1859 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1860 const struct sock *addr_sk); 1861 1862 #ifdef CONFIG_TCP_MD5SIG 1863 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1864 const union tcp_md5_addr *addr, 1865 int family, bool any_l3index); 1866 static inline struct tcp_md5sig_key * 1867 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1868 const union tcp_md5_addr *addr, int family) 1869 { 1870 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1871 return NULL; 1872 return __tcp_md5_do_lookup(sk, l3index, addr, family, false); 1873 } 1874 1875 static inline struct tcp_md5sig_key * 1876 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1877 const union tcp_md5_addr *addr, int family) 1878 { 1879 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1880 return NULL; 1881 return __tcp_md5_do_lookup(sk, 0, addr, family, true); 1882 } 1883 1884 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1885 #else 1886 static inline struct tcp_md5sig_key * 1887 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1888 const union tcp_md5_addr *addr, int family) 1889 { 1890 return NULL; 1891 } 1892 1893 static inline struct tcp_md5sig_key * 1894 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1895 const union tcp_md5_addr *addr, int family) 1896 { 1897 return NULL; 1898 } 1899 1900 #define tcp_twsk_md5_key(twsk) NULL 1901 #endif 1902 1903 int tcp_md5_alloc_sigpool(void); 1904 void tcp_md5_release_sigpool(void); 1905 void tcp_md5_add_sigpool(void); 1906 extern int tcp_md5_sigpool_id; 1907 1908 int tcp_md5_hash_key(struct tcp_sigpool *hp, 1909 const struct tcp_md5sig_key *key); 1910 1911 /* From tcp_fastopen.c */ 1912 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1913 struct tcp_fastopen_cookie *cookie); 1914 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1915 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1916 u16 try_exp); 1917 struct tcp_fastopen_request { 1918 /* Fast Open cookie. Size 0 means a cookie request */ 1919 struct tcp_fastopen_cookie cookie; 1920 struct msghdr *data; /* data in MSG_FASTOPEN */ 1921 size_t size; 1922 int copied; /* queued in tcp_connect() */ 1923 struct ubuf_info *uarg; 1924 }; 1925 void tcp_free_fastopen_req(struct tcp_sock *tp); 1926 void tcp_fastopen_destroy_cipher(struct sock *sk); 1927 void tcp_fastopen_ctx_destroy(struct net *net); 1928 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1929 void *primary_key, void *backup_key); 1930 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1931 u64 *key); 1932 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1933 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1934 struct request_sock *req, 1935 struct tcp_fastopen_cookie *foc, 1936 const struct dst_entry *dst); 1937 void tcp_fastopen_init_key_once(struct net *net); 1938 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1939 struct tcp_fastopen_cookie *cookie); 1940 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1941 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1942 #define TCP_FASTOPEN_KEY_MAX 2 1943 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1944 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1945 1946 /* Fastopen key context */ 1947 struct tcp_fastopen_context { 1948 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1949 int num; 1950 struct rcu_head rcu; 1951 }; 1952 1953 void tcp_fastopen_active_disable(struct sock *sk); 1954 bool tcp_fastopen_active_should_disable(struct sock *sk); 1955 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1956 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1957 1958 /* Caller needs to wrap with rcu_read_(un)lock() */ 1959 static inline 1960 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1961 { 1962 struct tcp_fastopen_context *ctx; 1963 1964 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1965 if (!ctx) 1966 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1967 return ctx; 1968 } 1969 1970 static inline 1971 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1972 const struct tcp_fastopen_cookie *orig) 1973 { 1974 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1975 orig->len == foc->len && 1976 !memcmp(orig->val, foc->val, foc->len)) 1977 return true; 1978 return false; 1979 } 1980 1981 static inline 1982 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1983 { 1984 return ctx->num; 1985 } 1986 1987 /* Latencies incurred by various limits for a sender. They are 1988 * chronograph-like stats that are mutually exclusive. 1989 */ 1990 enum tcp_chrono { 1991 TCP_CHRONO_UNSPEC, 1992 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1993 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1994 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1995 __TCP_CHRONO_MAX, 1996 }; 1997 1998 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1999 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 2000 2001 /* This helper is needed, because skb->tcp_tsorted_anchor uses 2002 * the same memory storage than skb->destructor/_skb_refdst 2003 */ 2004 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 2005 { 2006 skb->destructor = NULL; 2007 skb->_skb_refdst = 0UL; 2008 } 2009 2010 #define tcp_skb_tsorted_save(skb) { \ 2011 unsigned long _save = skb->_skb_refdst; \ 2012 skb->_skb_refdst = 0UL; 2013 2014 #define tcp_skb_tsorted_restore(skb) \ 2015 skb->_skb_refdst = _save; \ 2016 } 2017 2018 void tcp_write_queue_purge(struct sock *sk); 2019 2020 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 2021 { 2022 return skb_rb_first(&sk->tcp_rtx_queue); 2023 } 2024 2025 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 2026 { 2027 return skb_rb_last(&sk->tcp_rtx_queue); 2028 } 2029 2030 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 2031 { 2032 return skb_peek_tail(&sk->sk_write_queue); 2033 } 2034 2035 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 2036 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 2037 2038 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 2039 { 2040 return skb_peek(&sk->sk_write_queue); 2041 } 2042 2043 static inline bool tcp_skb_is_last(const struct sock *sk, 2044 const struct sk_buff *skb) 2045 { 2046 return skb_queue_is_last(&sk->sk_write_queue, skb); 2047 } 2048 2049 /** 2050 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 2051 * @sk: socket 2052 * 2053 * Since the write queue can have a temporary empty skb in it, 2054 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 2055 */ 2056 static inline bool tcp_write_queue_empty(const struct sock *sk) 2057 { 2058 const struct tcp_sock *tp = tcp_sk(sk); 2059 2060 return tp->write_seq == tp->snd_nxt; 2061 } 2062 2063 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 2064 { 2065 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 2066 } 2067 2068 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 2069 { 2070 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 2071 } 2072 2073 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 2074 { 2075 __skb_queue_tail(&sk->sk_write_queue, skb); 2076 2077 /* Queue it, remembering where we must start sending. */ 2078 if (sk->sk_write_queue.next == skb) 2079 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 2080 } 2081 2082 /* Insert new before skb on the write queue of sk. */ 2083 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 2084 struct sk_buff *skb, 2085 struct sock *sk) 2086 { 2087 __skb_queue_before(&sk->sk_write_queue, skb, new); 2088 } 2089 2090 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 2091 { 2092 tcp_skb_tsorted_anchor_cleanup(skb); 2093 __skb_unlink(skb, &sk->sk_write_queue); 2094 } 2095 2096 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 2097 2098 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2099 { 2100 tcp_skb_tsorted_anchor_cleanup(skb); 2101 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2102 } 2103 2104 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2105 { 2106 list_del(&skb->tcp_tsorted_anchor); 2107 tcp_rtx_queue_unlink(skb, sk); 2108 tcp_wmem_free_skb(sk, skb); 2109 } 2110 2111 static inline void tcp_write_collapse_fence(struct sock *sk) 2112 { 2113 struct sk_buff *skb = tcp_write_queue_tail(sk); 2114 2115 if (skb) 2116 TCP_SKB_CB(skb)->eor = 1; 2117 } 2118 2119 static inline void tcp_push_pending_frames(struct sock *sk) 2120 { 2121 if (tcp_send_head(sk)) { 2122 struct tcp_sock *tp = tcp_sk(sk); 2123 2124 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2125 } 2126 } 2127 2128 /* Start sequence of the skb just after the highest skb with SACKed 2129 * bit, valid only if sacked_out > 0 or when the caller has ensured 2130 * validity by itself. 2131 */ 2132 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2133 { 2134 if (!tp->sacked_out) 2135 return tp->snd_una; 2136 2137 if (tp->highest_sack == NULL) 2138 return tp->snd_nxt; 2139 2140 return TCP_SKB_CB(tp->highest_sack)->seq; 2141 } 2142 2143 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2144 { 2145 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2146 } 2147 2148 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2149 { 2150 return tcp_sk(sk)->highest_sack; 2151 } 2152 2153 static inline void tcp_highest_sack_reset(struct sock *sk) 2154 { 2155 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2156 } 2157 2158 /* Called when old skb is about to be deleted and replaced by new skb */ 2159 static inline void tcp_highest_sack_replace(struct sock *sk, 2160 struct sk_buff *old, 2161 struct sk_buff *new) 2162 { 2163 if (old == tcp_highest_sack(sk)) 2164 tcp_sk(sk)->highest_sack = new; 2165 } 2166 2167 /* This helper checks if socket has IP_TRANSPARENT set */ 2168 static inline bool inet_sk_transparent(const struct sock *sk) 2169 { 2170 switch (sk->sk_state) { 2171 case TCP_TIME_WAIT: 2172 return inet_twsk(sk)->tw_transparent; 2173 case TCP_NEW_SYN_RECV: 2174 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2175 } 2176 return inet_test_bit(TRANSPARENT, sk); 2177 } 2178 2179 /* Determines whether this is a thin stream (which may suffer from 2180 * increased latency). Used to trigger latency-reducing mechanisms. 2181 */ 2182 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2183 { 2184 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2185 } 2186 2187 /* /proc */ 2188 enum tcp_seq_states { 2189 TCP_SEQ_STATE_LISTENING, 2190 TCP_SEQ_STATE_ESTABLISHED, 2191 }; 2192 2193 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2194 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2195 void tcp_seq_stop(struct seq_file *seq, void *v); 2196 2197 struct tcp_seq_afinfo { 2198 sa_family_t family; 2199 }; 2200 2201 struct tcp_iter_state { 2202 struct seq_net_private p; 2203 enum tcp_seq_states state; 2204 struct sock *syn_wait_sk; 2205 int bucket, offset, sbucket, num; 2206 loff_t last_pos; 2207 }; 2208 2209 extern struct request_sock_ops tcp_request_sock_ops; 2210 extern struct request_sock_ops tcp6_request_sock_ops; 2211 2212 void tcp_v4_destroy_sock(struct sock *sk); 2213 2214 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2215 netdev_features_t features); 2216 struct tcphdr *tcp_gro_pull_header(struct sk_buff *skb); 2217 struct sk_buff *tcp_gro_lookup(struct list_head *head, struct tcphdr *th); 2218 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb, 2219 struct tcphdr *th); 2220 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2221 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2222 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2223 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2224 #ifdef CONFIG_INET 2225 void tcp_gro_complete(struct sk_buff *skb); 2226 #else 2227 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2228 #endif 2229 2230 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2231 2232 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2233 { 2234 struct net *net = sock_net((struct sock *)tp); 2235 u32 val; 2236 2237 val = READ_ONCE(tp->notsent_lowat); 2238 2239 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2240 } 2241 2242 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2243 2244 #ifdef CONFIG_PROC_FS 2245 int tcp4_proc_init(void); 2246 void tcp4_proc_exit(void); 2247 #endif 2248 2249 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2250 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2251 const struct tcp_request_sock_ops *af_ops, 2252 struct sock *sk, struct sk_buff *skb); 2253 2254 /* TCP af-specific functions */ 2255 struct tcp_sock_af_ops { 2256 #ifdef CONFIG_TCP_MD5SIG 2257 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2258 const struct sock *addr_sk); 2259 int (*calc_md5_hash)(char *location, 2260 const struct tcp_md5sig_key *md5, 2261 const struct sock *sk, 2262 const struct sk_buff *skb); 2263 int (*md5_parse)(struct sock *sk, 2264 int optname, 2265 sockptr_t optval, 2266 int optlen); 2267 #endif 2268 #ifdef CONFIG_TCP_AO 2269 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen); 2270 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2271 struct sock *addr_sk, 2272 int sndid, int rcvid); 2273 int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key, 2274 const struct sock *sk, 2275 __be32 sisn, __be32 disn, bool send); 2276 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao, 2277 const struct sock *sk, const struct sk_buff *skb, 2278 const u8 *tkey, int hash_offset, u32 sne); 2279 #endif 2280 }; 2281 2282 struct tcp_request_sock_ops { 2283 u16 mss_clamp; 2284 #ifdef CONFIG_TCP_MD5SIG 2285 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2286 const struct sock *addr_sk); 2287 int (*calc_md5_hash) (char *location, 2288 const struct tcp_md5sig_key *md5, 2289 const struct sock *sk, 2290 const struct sk_buff *skb); 2291 #endif 2292 #ifdef CONFIG_TCP_AO 2293 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2294 struct request_sock *req, 2295 int sndid, int rcvid); 2296 int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk); 2297 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt, 2298 struct request_sock *req, const struct sk_buff *skb, 2299 int hash_offset, u32 sne); 2300 #endif 2301 #ifdef CONFIG_SYN_COOKIES 2302 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2303 __u16 *mss); 2304 #endif 2305 struct dst_entry *(*route_req)(const struct sock *sk, 2306 struct sk_buff *skb, 2307 struct flowi *fl, 2308 struct request_sock *req, 2309 u32 tw_isn); 2310 u32 (*init_seq)(const struct sk_buff *skb); 2311 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2312 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2313 struct flowi *fl, struct request_sock *req, 2314 struct tcp_fastopen_cookie *foc, 2315 enum tcp_synack_type synack_type, 2316 struct sk_buff *syn_skb); 2317 }; 2318 2319 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2320 #if IS_ENABLED(CONFIG_IPV6) 2321 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2322 #endif 2323 2324 #ifdef CONFIG_SYN_COOKIES 2325 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2326 const struct sock *sk, struct sk_buff *skb, 2327 __u16 *mss) 2328 { 2329 tcp_synq_overflow(sk); 2330 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2331 return ops->cookie_init_seq(skb, mss); 2332 } 2333 #else 2334 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2335 const struct sock *sk, struct sk_buff *skb, 2336 __u16 *mss) 2337 { 2338 return 0; 2339 } 2340 #endif 2341 2342 struct tcp_key { 2343 union { 2344 struct { 2345 struct tcp_ao_key *ao_key; 2346 char *traffic_key; 2347 u32 sne; 2348 u8 rcv_next; 2349 }; 2350 struct tcp_md5sig_key *md5_key; 2351 }; 2352 enum { 2353 TCP_KEY_NONE = 0, 2354 TCP_KEY_MD5, 2355 TCP_KEY_AO, 2356 } type; 2357 }; 2358 2359 static inline void tcp_get_current_key(const struct sock *sk, 2360 struct tcp_key *out) 2361 { 2362 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG) 2363 const struct tcp_sock *tp = tcp_sk(sk); 2364 #endif 2365 2366 #ifdef CONFIG_TCP_AO 2367 if (static_branch_unlikely(&tcp_ao_needed.key)) { 2368 struct tcp_ao_info *ao; 2369 2370 ao = rcu_dereference_protected(tp->ao_info, 2371 lockdep_sock_is_held(sk)); 2372 if (ao) { 2373 out->ao_key = READ_ONCE(ao->current_key); 2374 out->type = TCP_KEY_AO; 2375 return; 2376 } 2377 } 2378 #endif 2379 #ifdef CONFIG_TCP_MD5SIG 2380 if (static_branch_unlikely(&tcp_md5_needed.key) && 2381 rcu_access_pointer(tp->md5sig_info)) { 2382 out->md5_key = tp->af_specific->md5_lookup(sk, sk); 2383 if (out->md5_key) { 2384 out->type = TCP_KEY_MD5; 2385 return; 2386 } 2387 } 2388 #endif 2389 out->type = TCP_KEY_NONE; 2390 } 2391 2392 static inline bool tcp_key_is_md5(const struct tcp_key *key) 2393 { 2394 if (static_branch_tcp_md5()) 2395 return key->type == TCP_KEY_MD5; 2396 return false; 2397 } 2398 2399 static inline bool tcp_key_is_ao(const struct tcp_key *key) 2400 { 2401 if (static_branch_tcp_ao()) 2402 return key->type == TCP_KEY_AO; 2403 return false; 2404 } 2405 2406 int tcpv4_offload_init(void); 2407 2408 void tcp_v4_init(void); 2409 void tcp_init(void); 2410 2411 /* tcp_recovery.c */ 2412 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2413 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2414 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2415 u32 reo_wnd); 2416 extern bool tcp_rack_mark_lost(struct sock *sk); 2417 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2418 u64 xmit_time); 2419 extern void tcp_rack_reo_timeout(struct sock *sk); 2420 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2421 2422 /* tcp_plb.c */ 2423 2424 /* 2425 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2426 * expects cong_ratio which represents fraction of traffic that experienced 2427 * congestion over a single RTT. In order to avoid floating point operations, 2428 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2429 */ 2430 #define TCP_PLB_SCALE 8 2431 2432 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2433 struct tcp_plb_state { 2434 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2435 unused:3; 2436 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2437 }; 2438 2439 static inline void tcp_plb_init(const struct sock *sk, 2440 struct tcp_plb_state *plb) 2441 { 2442 plb->consec_cong_rounds = 0; 2443 plb->pause_until = 0; 2444 } 2445 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2446 const int cong_ratio); 2447 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2448 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2449 2450 static inline void tcp_warn_once(const struct sock *sk, bool cond, const char *str) 2451 { 2452 WARN_ONCE(cond, 2453 "%scwn:%u out:%u sacked:%u lost:%u retrans:%u tlp_high_seq:%u sk_state:%u ca_state:%u advmss:%u mss_cache:%u pmtu:%u\n", 2454 str, 2455 tcp_snd_cwnd(tcp_sk(sk)), 2456 tcp_sk(sk)->packets_out, tcp_sk(sk)->sacked_out, 2457 tcp_sk(sk)->lost_out, tcp_sk(sk)->retrans_out, 2458 tcp_sk(sk)->tlp_high_seq, sk->sk_state, 2459 inet_csk(sk)->icsk_ca_state, 2460 tcp_sk(sk)->advmss, tcp_sk(sk)->mss_cache, 2461 inet_csk(sk)->icsk_pmtu_cookie); 2462 } 2463 2464 /* At how many usecs into the future should the RTO fire? */ 2465 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2466 { 2467 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2468 u32 rto = inet_csk(sk)->icsk_rto; 2469 2470 if (likely(skb)) { 2471 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2472 2473 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2474 } else { 2475 tcp_warn_once(sk, 1, "rtx queue empty: "); 2476 return jiffies_to_usecs(rto); 2477 } 2478 2479 } 2480 2481 /* 2482 * Save and compile IPv4 options, return a pointer to it 2483 */ 2484 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2485 struct sk_buff *skb) 2486 { 2487 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2488 struct ip_options_rcu *dopt = NULL; 2489 2490 if (opt->optlen) { 2491 int opt_size = sizeof(*dopt) + opt->optlen; 2492 2493 dopt = kmalloc(opt_size, GFP_ATOMIC); 2494 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2495 kfree(dopt); 2496 dopt = NULL; 2497 } 2498 } 2499 return dopt; 2500 } 2501 2502 /* locally generated TCP pure ACKs have skb->truesize == 2 2503 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2504 * This is much faster than dissecting the packet to find out. 2505 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2506 */ 2507 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2508 { 2509 return skb->truesize == 2; 2510 } 2511 2512 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2513 { 2514 skb->truesize = 2; 2515 } 2516 2517 static inline int tcp_inq(struct sock *sk) 2518 { 2519 struct tcp_sock *tp = tcp_sk(sk); 2520 int answ; 2521 2522 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2523 answ = 0; 2524 } else if (sock_flag(sk, SOCK_URGINLINE) || 2525 !tp->urg_data || 2526 before(tp->urg_seq, tp->copied_seq) || 2527 !before(tp->urg_seq, tp->rcv_nxt)) { 2528 2529 answ = tp->rcv_nxt - tp->copied_seq; 2530 2531 /* Subtract 1, if FIN was received */ 2532 if (answ && sock_flag(sk, SOCK_DONE)) 2533 answ--; 2534 } else { 2535 answ = tp->urg_seq - tp->copied_seq; 2536 } 2537 2538 return answ; 2539 } 2540 2541 int tcp_peek_len(struct socket *sock); 2542 2543 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2544 { 2545 u16 segs_in; 2546 2547 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2548 2549 /* We update these fields while other threads might 2550 * read them from tcp_get_info() 2551 */ 2552 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2553 if (skb->len > tcp_hdrlen(skb)) 2554 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2555 } 2556 2557 /* 2558 * TCP listen path runs lockless. 2559 * We forced "struct sock" to be const qualified to make sure 2560 * we don't modify one of its field by mistake. 2561 * Here, we increment sk_drops which is an atomic_t, so we can safely 2562 * make sock writable again. 2563 */ 2564 static inline void tcp_listendrop(const struct sock *sk) 2565 { 2566 atomic_inc(&((struct sock *)sk)->sk_drops); 2567 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2568 } 2569 2570 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2571 2572 /* 2573 * Interface for adding Upper Level Protocols over TCP 2574 */ 2575 2576 #define TCP_ULP_NAME_MAX 16 2577 #define TCP_ULP_MAX 128 2578 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2579 2580 struct tcp_ulp_ops { 2581 struct list_head list; 2582 2583 /* initialize ulp */ 2584 int (*init)(struct sock *sk); 2585 /* update ulp */ 2586 void (*update)(struct sock *sk, struct proto *p, 2587 void (*write_space)(struct sock *sk)); 2588 /* cleanup ulp */ 2589 void (*release)(struct sock *sk); 2590 /* diagnostic */ 2591 int (*get_info)(struct sock *sk, struct sk_buff *skb); 2592 size_t (*get_info_size)(const struct sock *sk); 2593 /* clone ulp */ 2594 void (*clone)(const struct request_sock *req, struct sock *newsk, 2595 const gfp_t priority); 2596 2597 char name[TCP_ULP_NAME_MAX]; 2598 struct module *owner; 2599 }; 2600 int tcp_register_ulp(struct tcp_ulp_ops *type); 2601 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2602 int tcp_set_ulp(struct sock *sk, const char *name); 2603 void tcp_get_available_ulp(char *buf, size_t len); 2604 void tcp_cleanup_ulp(struct sock *sk); 2605 void tcp_update_ulp(struct sock *sk, struct proto *p, 2606 void (*write_space)(struct sock *sk)); 2607 2608 #define MODULE_ALIAS_TCP_ULP(name) \ 2609 __MODULE_INFO(alias, alias_userspace, name); \ 2610 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2611 2612 #ifdef CONFIG_NET_SOCK_MSG 2613 struct sk_msg; 2614 struct sk_psock; 2615 2616 #ifdef CONFIG_BPF_SYSCALL 2617 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2618 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2619 #ifdef CONFIG_BPF_STREAM_PARSER 2620 struct strparser; 2621 int tcp_bpf_strp_read_sock(struct strparser *strp, read_descriptor_t *desc, 2622 sk_read_actor_t recv_actor); 2623 #endif /* CONFIG_BPF_STREAM_PARSER */ 2624 #endif /* CONFIG_BPF_SYSCALL */ 2625 2626 #ifdef CONFIG_INET 2627 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2628 #else 2629 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2630 { 2631 } 2632 #endif 2633 2634 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2635 struct sk_msg *msg, u32 bytes, int flags); 2636 #endif /* CONFIG_NET_SOCK_MSG */ 2637 2638 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2639 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2640 { 2641 } 2642 #endif 2643 2644 #ifdef CONFIG_CGROUP_BPF 2645 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2646 struct sk_buff *skb, 2647 unsigned int end_offset) 2648 { 2649 skops->skb = skb; 2650 skops->skb_data_end = skb->data + end_offset; 2651 } 2652 #else 2653 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2654 struct sk_buff *skb, 2655 unsigned int end_offset) 2656 { 2657 } 2658 #endif 2659 2660 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2661 * is < 0, then the BPF op failed (for example if the loaded BPF 2662 * program does not support the chosen operation or there is no BPF 2663 * program loaded). 2664 */ 2665 #ifdef CONFIG_BPF 2666 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2667 { 2668 struct bpf_sock_ops_kern sock_ops; 2669 int ret; 2670 2671 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2672 if (sk_fullsock(sk)) { 2673 sock_ops.is_fullsock = 1; 2674 sock_owned_by_me(sk); 2675 } 2676 2677 sock_ops.sk = sk; 2678 sock_ops.op = op; 2679 if (nargs > 0) 2680 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2681 2682 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2683 if (ret == 0) 2684 ret = sock_ops.reply; 2685 else 2686 ret = -1; 2687 return ret; 2688 } 2689 2690 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2691 { 2692 u32 args[2] = {arg1, arg2}; 2693 2694 return tcp_call_bpf(sk, op, 2, args); 2695 } 2696 2697 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2698 u32 arg3) 2699 { 2700 u32 args[3] = {arg1, arg2, arg3}; 2701 2702 return tcp_call_bpf(sk, op, 3, args); 2703 } 2704 2705 #else 2706 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2707 { 2708 return -EPERM; 2709 } 2710 2711 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2712 { 2713 return -EPERM; 2714 } 2715 2716 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2717 u32 arg3) 2718 { 2719 return -EPERM; 2720 } 2721 2722 #endif 2723 2724 static inline u32 tcp_timeout_init(struct sock *sk) 2725 { 2726 int timeout; 2727 2728 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2729 2730 if (timeout <= 0) 2731 timeout = TCP_TIMEOUT_INIT; 2732 return min_t(int, timeout, TCP_RTO_MAX); 2733 } 2734 2735 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2736 { 2737 int rwnd; 2738 2739 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2740 2741 if (rwnd < 0) 2742 rwnd = 0; 2743 return rwnd; 2744 } 2745 2746 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2747 { 2748 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2749 } 2750 2751 static inline void tcp_bpf_rtt(struct sock *sk, long mrtt, u32 srtt) 2752 { 2753 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2754 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_RTT_CB, mrtt, srtt); 2755 } 2756 2757 #if IS_ENABLED(CONFIG_SMC) 2758 extern struct static_key_false tcp_have_smc; 2759 #endif 2760 2761 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2762 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2763 void (*cad)(struct sock *sk, u32 ack_seq)); 2764 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2765 void clean_acked_data_flush(void); 2766 #endif 2767 2768 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2769 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2770 const struct tcp_sock *tp) 2771 { 2772 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2773 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2774 } 2775 2776 /* Compute Earliest Departure Time for some control packets 2777 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2778 */ 2779 static inline u64 tcp_transmit_time(const struct sock *sk) 2780 { 2781 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2782 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2783 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2784 2785 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2786 } 2787 return 0; 2788 } 2789 2790 static inline int tcp_parse_auth_options(const struct tcphdr *th, 2791 const u8 **md5_hash, const struct tcp_ao_hdr **aoh) 2792 { 2793 const u8 *md5_tmp, *ao_tmp; 2794 int ret; 2795 2796 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp); 2797 if (ret) 2798 return ret; 2799 2800 if (md5_hash) 2801 *md5_hash = md5_tmp; 2802 2803 if (aoh) { 2804 if (!ao_tmp) 2805 *aoh = NULL; 2806 else 2807 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2); 2808 } 2809 2810 return 0; 2811 } 2812 2813 static inline bool tcp_ao_required(struct sock *sk, const void *saddr, 2814 int family, int l3index, bool stat_inc) 2815 { 2816 #ifdef CONFIG_TCP_AO 2817 struct tcp_ao_info *ao_info; 2818 struct tcp_ao_key *ao_key; 2819 2820 if (!static_branch_unlikely(&tcp_ao_needed.key)) 2821 return false; 2822 2823 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info, 2824 lockdep_sock_is_held(sk)); 2825 if (!ao_info) 2826 return false; 2827 2828 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1); 2829 if (ao_info->ao_required || ao_key) { 2830 if (stat_inc) { 2831 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED); 2832 atomic64_inc(&ao_info->counters.ao_required); 2833 } 2834 return true; 2835 } 2836 #endif 2837 return false; 2838 } 2839 2840 enum skb_drop_reason tcp_inbound_hash(struct sock *sk, 2841 const struct request_sock *req, const struct sk_buff *skb, 2842 const void *saddr, const void *daddr, 2843 int family, int dif, int sdif); 2844 2845 #endif /* _TCP_H */ 2846