1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP module. 8 * 9 * Version: @(#)tcp.h 1.0.5 05/23/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 */ 14 #ifndef _TCP_H 15 #define _TCP_H 16 17 #define FASTRETRANS_DEBUG 1 18 19 #include <linux/list.h> 20 #include <linux/tcp.h> 21 #include <linux/bug.h> 22 #include <linux/slab.h> 23 #include <linux/cache.h> 24 #include <linux/percpu.h> 25 #include <linux/skbuff.h> 26 #include <linux/kref.h> 27 #include <linux/ktime.h> 28 #include <linux/indirect_call_wrapper.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/inet_ecn.h> 41 #include <net/dst.h> 42 #include <net/mptcp.h> 43 44 #include <linux/seq_file.h> 45 #include <linux/memcontrol.h> 46 #include <linux/bpf-cgroup.h> 47 #include <linux/siphash.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 52 int tcp_orphan_count_sum(void); 53 54 void tcp_time_wait(struct sock *sk, int state, int timeo); 55 56 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 57 #define MAX_TCP_OPTION_SPACE 40 58 #define TCP_MIN_SND_MSS 48 59 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 60 61 /* 62 * Never offer a window over 32767 without using window scaling. Some 63 * poor stacks do signed 16bit maths! 64 */ 65 #define MAX_TCP_WINDOW 32767U 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The initial MTU to use for probing */ 71 #define TCP_BASE_MSS 1024 72 73 /* probing interval, default to 10 minutes as per RFC4821 */ 74 #define TCP_PROBE_INTERVAL 600 75 76 /* Specify interval when tcp mtu probing will stop */ 77 #define TCP_PROBE_THRESHOLD 8 78 79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 80 #define TCP_FASTRETRANS_THRESH 3 81 82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 83 #define TCP_MAX_QUICKACKS 16U 84 85 /* Maximal number of window scale according to RFC1323 */ 86 #define TCP_MAX_WSCALE 14U 87 88 /* urg_data states */ 89 #define TCP_URG_VALID 0x0100 90 #define TCP_URG_NOTYET 0x0200 91 #define TCP_URG_READ 0x0400 92 93 #define TCP_RETR1 3 /* 94 * This is how many retries it does before it 95 * tries to figure out if the gateway is 96 * down. Minimal RFC value is 3; it corresponds 97 * to ~3sec-8min depending on RTO. 98 */ 99 100 #define TCP_RETR2 15 /* 101 * This should take at least 102 * 90 minutes to time out. 103 * RFC1122 says that the limit is 100 sec. 104 * 15 is ~13-30min depending on RTO. 105 */ 106 107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 108 * when active opening a connection. 109 * RFC1122 says the minimum retry MUST 110 * be at least 180secs. Nevertheless 111 * this value is corresponding to 112 * 63secs of retransmission with the 113 * current initial RTO. 114 */ 115 116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 117 * when passive opening a connection. 118 * This is corresponding to 31secs of 119 * retransmission with the current 120 * initial RTO. 121 */ 122 123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 124 * state, about 60 seconds */ 125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 126 /* BSD style FIN_WAIT2 deadlock breaker. 127 * It used to be 3min, new value is 60sec, 128 * to combine FIN-WAIT-2 timeout with 129 * TIME-WAIT timer. 130 */ 131 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 132 133 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 134 #if HZ >= 100 135 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 136 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 137 #else 138 #define TCP_DELACK_MIN 4U 139 #define TCP_ATO_MIN 4U 140 #endif 141 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 142 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 143 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 144 145 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 146 147 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 148 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 149 * used as a fallback RTO for the 150 * initial data transmission if no 151 * valid RTT sample has been acquired, 152 * most likely due to retrans in 3WHS. 153 */ 154 155 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 156 * for local resources. 157 */ 158 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 159 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 160 #define TCP_KEEPALIVE_INTVL (75*HZ) 161 162 #define MAX_TCP_KEEPIDLE 32767 163 #define MAX_TCP_KEEPINTVL 32767 164 #define MAX_TCP_KEEPCNT 127 165 #define MAX_TCP_SYNCNT 127 166 167 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 168 169 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 170 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 171 * after this time. It should be equal 172 * (or greater than) TCP_TIMEWAIT_LEN 173 * to provide reliability equal to one 174 * provided by timewait state. 175 */ 176 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 177 * timestamps. It must be less than 178 * minimal timewait lifetime. 179 */ 180 /* 181 * TCP option 182 */ 183 184 #define TCPOPT_NOP 1 /* Padding */ 185 #define TCPOPT_EOL 0 /* End of options */ 186 #define TCPOPT_MSS 2 /* Segment size negotiating */ 187 #define TCPOPT_WINDOW 3 /* Window scaling */ 188 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 189 #define TCPOPT_SACK 5 /* SACK Block */ 190 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 191 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 192 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 193 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 194 #define TCPOPT_EXP 254 /* Experimental */ 195 /* Magic number to be after the option value for sharing TCP 196 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 197 */ 198 #define TCPOPT_FASTOPEN_MAGIC 0xF989 199 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 200 201 /* 202 * TCP option lengths 203 */ 204 205 #define TCPOLEN_MSS 4 206 #define TCPOLEN_WINDOW 3 207 #define TCPOLEN_SACK_PERM 2 208 #define TCPOLEN_TIMESTAMP 10 209 #define TCPOLEN_MD5SIG 18 210 #define TCPOLEN_FASTOPEN_BASE 2 211 #define TCPOLEN_EXP_FASTOPEN_BASE 4 212 #define TCPOLEN_EXP_SMC_BASE 6 213 214 /* But this is what stacks really send out. */ 215 #define TCPOLEN_TSTAMP_ALIGNED 12 216 #define TCPOLEN_WSCALE_ALIGNED 4 217 #define TCPOLEN_SACKPERM_ALIGNED 4 218 #define TCPOLEN_SACK_BASE 2 219 #define TCPOLEN_SACK_BASE_ALIGNED 4 220 #define TCPOLEN_SACK_PERBLOCK 8 221 #define TCPOLEN_MD5SIG_ALIGNED 20 222 #define TCPOLEN_MSS_ALIGNED 4 223 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 224 225 /* Flags in tp->nonagle */ 226 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 227 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 228 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 229 230 /* TCP thin-stream limits */ 231 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 232 233 /* TCP initial congestion window as per rfc6928 */ 234 #define TCP_INIT_CWND 10 235 236 /* Bit Flags for sysctl_tcp_fastopen */ 237 #define TFO_CLIENT_ENABLE 1 238 #define TFO_SERVER_ENABLE 2 239 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 240 241 /* Accept SYN data w/o any cookie option */ 242 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 243 244 /* Force enable TFO on all listeners, i.e., not requiring the 245 * TCP_FASTOPEN socket option. 246 */ 247 #define TFO_SERVER_WO_SOCKOPT1 0x400 248 249 250 /* sysctl variables for tcp */ 251 extern int sysctl_tcp_max_orphans; 252 extern long sysctl_tcp_mem[3]; 253 254 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 255 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 256 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 257 258 extern atomic_long_t tcp_memory_allocated; 259 extern struct percpu_counter tcp_sockets_allocated; 260 extern unsigned long tcp_memory_pressure; 261 262 /* optimized version of sk_under_memory_pressure() for TCP sockets */ tcp_under_memory_pressure(const struct sock * sk)263 static inline bool tcp_under_memory_pressure(const struct sock *sk) 264 { 265 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 266 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 267 return true; 268 269 return READ_ONCE(tcp_memory_pressure); 270 } 271 /* 272 * The next routines deal with comparing 32 bit unsigned ints 273 * and worry about wraparound (automatic with unsigned arithmetic). 274 */ 275 before(__u32 seq1,__u32 seq2)276 static inline bool before(__u32 seq1, __u32 seq2) 277 { 278 return (__s32)(seq1-seq2) < 0; 279 } 280 #define after(seq2, seq1) before(seq1, seq2) 281 282 /* is s2<=s1<=s3 ? */ between(__u32 seq1,__u32 seq2,__u32 seq3)283 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 284 { 285 return seq3 - seq2 >= seq1 - seq2; 286 } 287 tcp_out_of_memory(struct sock * sk)288 static inline bool tcp_out_of_memory(struct sock *sk) 289 { 290 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 291 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 292 return true; 293 return false; 294 } 295 296 void sk_forced_mem_schedule(struct sock *sk, int size); 297 298 bool tcp_check_oom(struct sock *sk, int shift); 299 300 301 extern struct proto tcp_prot; 302 303 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 304 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 305 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 306 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 307 308 void tcp_tasklet_init(void); 309 310 int tcp_v4_err(struct sk_buff *skb, u32); 311 312 void tcp_shutdown(struct sock *sk, int how); 313 314 int tcp_v4_early_demux(struct sk_buff *skb); 315 int tcp_v4_rcv(struct sk_buff *skb); 316 317 void tcp_remove_empty_skb(struct sock *sk); 318 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 319 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 320 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 321 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 322 int flags); 323 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 324 size_t size, int flags); 325 struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags, 326 struct page *page, int offset, size_t *size); 327 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 328 size_t size, int flags); 329 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 330 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 331 int size_goal); 332 void tcp_release_cb(struct sock *sk); 333 void tcp_wfree(struct sk_buff *skb); 334 void tcp_write_timer_handler(struct sock *sk); 335 void tcp_delack_timer_handler(struct sock *sk); 336 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 337 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 338 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 339 void tcp_rcv_space_adjust(struct sock *sk); 340 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 341 void tcp_twsk_destructor(struct sock *sk); 342 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 343 struct pipe_inode_info *pipe, size_t len, 344 unsigned int flags); 345 tcp_dec_quickack_mode(struct sock * sk)346 static inline void tcp_dec_quickack_mode(struct sock *sk) 347 { 348 struct inet_connection_sock *icsk = inet_csk(sk); 349 350 if (icsk->icsk_ack.quick) { 351 /* How many ACKs S/ACKing new data have we sent? */ 352 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 353 354 if (pkts >= icsk->icsk_ack.quick) { 355 icsk->icsk_ack.quick = 0; 356 /* Leaving quickack mode we deflate ATO. */ 357 icsk->icsk_ack.ato = TCP_ATO_MIN; 358 } else 359 icsk->icsk_ack.quick -= pkts; 360 } 361 } 362 363 #define TCP_ECN_OK 1 364 #define TCP_ECN_QUEUE_CWR 2 365 #define TCP_ECN_DEMAND_CWR 4 366 #define TCP_ECN_SEEN 8 367 368 enum tcp_tw_status { 369 TCP_TW_SUCCESS = 0, 370 TCP_TW_RST = 1, 371 TCP_TW_ACK = 2, 372 TCP_TW_SYN = 3 373 }; 374 375 376 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 377 struct sk_buff *skb, 378 const struct tcphdr *th); 379 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 380 struct request_sock *req, bool fastopen, 381 bool *lost_race); 382 int tcp_child_process(struct sock *parent, struct sock *child, 383 struct sk_buff *skb); 384 void tcp_enter_loss(struct sock *sk); 385 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 386 void tcp_clear_retrans(struct tcp_sock *tp); 387 void tcp_update_metrics(struct sock *sk); 388 void tcp_init_metrics(struct sock *sk); 389 void tcp_metrics_init(void); 390 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 391 void __tcp_close(struct sock *sk, long timeout); 392 void tcp_close(struct sock *sk, long timeout); 393 void tcp_init_sock(struct sock *sk); 394 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 395 __poll_t tcp_poll(struct file *file, struct socket *sock, 396 struct poll_table_struct *wait); 397 int tcp_getsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, int __user *optlen); 399 bool tcp_bpf_bypass_getsockopt(int level, int optname); 400 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 401 unsigned int optlen); 402 void tcp_set_keepalive(struct sock *sk, int val); 403 void tcp_syn_ack_timeout(const struct request_sock *req); 404 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 405 int flags, int *addr_len); 406 int tcp_set_rcvlowat(struct sock *sk, int val); 407 int tcp_set_window_clamp(struct sock *sk, int val); 408 void tcp_update_recv_tstamps(struct sk_buff *skb, 409 struct scm_timestamping_internal *tss); 410 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 411 struct scm_timestamping_internal *tss); 412 void tcp_data_ready(struct sock *sk); 413 #ifdef CONFIG_MMU 414 int tcp_mmap(struct file *file, struct socket *sock, 415 struct vm_area_struct *vma); 416 #endif 417 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 418 struct tcp_options_received *opt_rx, 419 int estab, struct tcp_fastopen_cookie *foc); 420 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 421 422 /* 423 * BPF SKB-less helpers 424 */ 425 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 426 struct tcphdr *th, u32 *cookie); 427 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 428 struct tcphdr *th, u32 *cookie); 429 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 430 const struct tcp_request_sock_ops *af_ops, 431 struct sock *sk, struct tcphdr *th); 432 /* 433 * TCP v4 functions exported for the inet6 API 434 */ 435 436 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 437 void tcp_v4_mtu_reduced(struct sock *sk); 438 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 439 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 440 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 441 struct sock *tcp_create_openreq_child(const struct sock *sk, 442 struct request_sock *req, 443 struct sk_buff *skb); 444 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 445 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 446 struct request_sock *req, 447 struct dst_entry *dst, 448 struct request_sock *req_unhash, 449 bool *own_req); 450 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 451 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 452 int tcp_connect(struct sock *sk); 453 enum tcp_synack_type { 454 TCP_SYNACK_NORMAL, 455 TCP_SYNACK_FASTOPEN, 456 TCP_SYNACK_COOKIE, 457 }; 458 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 459 struct request_sock *req, 460 struct tcp_fastopen_cookie *foc, 461 enum tcp_synack_type synack_type, 462 struct sk_buff *syn_skb); 463 int tcp_disconnect(struct sock *sk, int flags); 464 465 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 466 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 467 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 468 469 /* From syncookies.c */ 470 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 471 struct request_sock *req, 472 struct dst_entry *dst, u32 tsoff); 473 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 474 u32 cookie); 475 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 476 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 477 const struct tcp_request_sock_ops *af_ops, 478 struct sock *sk, struct sk_buff *skb); 479 #ifdef CONFIG_SYN_COOKIES 480 481 /* Syncookies use a monotonic timer which increments every 60 seconds. 482 * This counter is used both as a hash input and partially encoded into 483 * the cookie value. A cookie is only validated further if the delta 484 * between the current counter value and the encoded one is less than this, 485 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 486 * the counter advances immediately after a cookie is generated). 487 */ 488 #define MAX_SYNCOOKIE_AGE 2 489 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 490 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 491 492 /* syncookies: remember time of last synqueue overflow 493 * But do not dirty this field too often (once per second is enough) 494 * It is racy as we do not hold a lock, but race is very minor. 495 */ tcp_synq_overflow(const struct sock * sk)496 static inline void tcp_synq_overflow(const struct sock *sk) 497 { 498 unsigned int last_overflow; 499 unsigned int now = jiffies; 500 501 if (sk->sk_reuseport) { 502 struct sock_reuseport *reuse; 503 504 reuse = rcu_dereference(sk->sk_reuseport_cb); 505 if (likely(reuse)) { 506 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 507 if (!time_between32(now, last_overflow, 508 last_overflow + HZ)) 509 WRITE_ONCE(reuse->synq_overflow_ts, now); 510 return; 511 } 512 } 513 514 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 515 if (!time_between32(now, last_overflow, last_overflow + HZ)) 516 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 517 } 518 519 /* syncookies: no recent synqueue overflow on this listening socket? */ tcp_synq_no_recent_overflow(const struct sock * sk)520 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 521 { 522 unsigned int last_overflow; 523 unsigned int now = jiffies; 524 525 if (sk->sk_reuseport) { 526 struct sock_reuseport *reuse; 527 528 reuse = rcu_dereference(sk->sk_reuseport_cb); 529 if (likely(reuse)) { 530 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 531 return !time_between32(now, last_overflow - HZ, 532 last_overflow + 533 TCP_SYNCOOKIE_VALID); 534 } 535 } 536 537 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 538 539 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 540 * then we're under synflood. However, we have to use 541 * 'last_overflow - HZ' as lower bound. That's because a concurrent 542 * tcp_synq_overflow() could update .ts_recent_stamp after we read 543 * jiffies but before we store .ts_recent_stamp into last_overflow, 544 * which could lead to rejecting a valid syncookie. 545 */ 546 return !time_between32(now, last_overflow - HZ, 547 last_overflow + TCP_SYNCOOKIE_VALID); 548 } 549 tcp_cookie_time(void)550 static inline u32 tcp_cookie_time(void) 551 { 552 u64 val = get_jiffies_64(); 553 554 do_div(val, TCP_SYNCOOKIE_PERIOD); 555 return val; 556 } 557 558 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 559 u16 *mssp); 560 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 561 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 562 bool cookie_timestamp_decode(const struct net *net, 563 struct tcp_options_received *opt); 564 bool cookie_ecn_ok(const struct tcp_options_received *opt, 565 const struct net *net, const struct dst_entry *dst); 566 567 /* From net/ipv6/syncookies.c */ 568 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 569 u32 cookie); 570 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 571 572 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 573 const struct tcphdr *th, u16 *mssp); 574 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 575 #endif 576 /* tcp_output.c */ 577 578 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 579 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 580 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 581 int nonagle); 582 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 583 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 584 void tcp_retransmit_timer(struct sock *sk); 585 void tcp_xmit_retransmit_queue(struct sock *); 586 void tcp_simple_retransmit(struct sock *); 587 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 588 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 589 enum tcp_queue { 590 TCP_FRAG_IN_WRITE_QUEUE, 591 TCP_FRAG_IN_RTX_QUEUE, 592 }; 593 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 594 struct sk_buff *skb, u32 len, 595 unsigned int mss_now, gfp_t gfp); 596 597 void tcp_send_probe0(struct sock *); 598 void tcp_send_partial(struct sock *); 599 int tcp_write_wakeup(struct sock *, int mib); 600 void tcp_send_fin(struct sock *sk); 601 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 602 int tcp_send_synack(struct sock *); 603 void tcp_push_one(struct sock *, unsigned int mss_now); 604 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 605 void tcp_send_ack(struct sock *sk); 606 void tcp_send_delayed_ack(struct sock *sk); 607 void tcp_send_loss_probe(struct sock *sk); 608 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 609 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 610 const struct sk_buff *next_skb); 611 612 /* tcp_input.c */ 613 void tcp_rearm_rto(struct sock *sk); 614 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 615 void tcp_reset(struct sock *sk, struct sk_buff *skb); 616 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 617 void tcp_fin(struct sock *sk); 618 void tcp_check_space(struct sock *sk); 619 620 /* tcp_timer.c */ 621 void tcp_init_xmit_timers(struct sock *); tcp_clear_xmit_timers(struct sock * sk)622 static inline void tcp_clear_xmit_timers(struct sock *sk) 623 { 624 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 625 __sock_put(sk); 626 627 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 628 __sock_put(sk); 629 630 inet_csk_clear_xmit_timers(sk); 631 } 632 633 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 634 unsigned int tcp_current_mss(struct sock *sk); 635 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 636 637 /* Bound MSS / TSO packet size with the half of the window */ tcp_bound_to_half_wnd(struct tcp_sock * tp,int pktsize)638 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 639 { 640 int cutoff; 641 642 /* When peer uses tiny windows, there is no use in packetizing 643 * to sub-MSS pieces for the sake of SWS or making sure there 644 * are enough packets in the pipe for fast recovery. 645 * 646 * On the other hand, for extremely large MSS devices, handling 647 * smaller than MSS windows in this way does make sense. 648 */ 649 if (tp->max_window > TCP_MSS_DEFAULT) 650 cutoff = (tp->max_window >> 1); 651 else 652 cutoff = tp->max_window; 653 654 if (cutoff && pktsize > cutoff) 655 return max_t(int, cutoff, 68U - tp->tcp_header_len); 656 else 657 return pktsize; 658 } 659 660 /* tcp.c */ 661 void tcp_get_info(struct sock *, struct tcp_info *); 662 663 /* Read 'sendfile()'-style from a TCP socket */ 664 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 665 sk_read_actor_t recv_actor); 666 667 void tcp_initialize_rcv_mss(struct sock *sk); 668 669 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 670 int tcp_mss_to_mtu(struct sock *sk, int mss); 671 void tcp_mtup_init(struct sock *sk); 672 tcp_bound_rto(const struct sock * sk)673 static inline void tcp_bound_rto(const struct sock *sk) 674 { 675 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 676 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 677 } 678 __tcp_set_rto(const struct tcp_sock * tp)679 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 680 { 681 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 682 } 683 __tcp_fast_path_on(struct tcp_sock * tp,u32 snd_wnd)684 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 685 { 686 /* mptcp hooks are only on the slow path */ 687 if (sk_is_mptcp((struct sock *)tp)) 688 return; 689 690 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 691 ntohl(TCP_FLAG_ACK) | 692 snd_wnd); 693 } 694 tcp_fast_path_on(struct tcp_sock * tp)695 static inline void tcp_fast_path_on(struct tcp_sock *tp) 696 { 697 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 698 } 699 tcp_fast_path_check(struct sock * sk)700 static inline void tcp_fast_path_check(struct sock *sk) 701 { 702 struct tcp_sock *tp = tcp_sk(sk); 703 704 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 705 tp->rcv_wnd && 706 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 707 !tp->urg_data) 708 tcp_fast_path_on(tp); 709 } 710 711 /* Compute the actual rto_min value */ tcp_rto_min(struct sock * sk)712 static inline u32 tcp_rto_min(struct sock *sk) 713 { 714 const struct dst_entry *dst = __sk_dst_get(sk); 715 u32 rto_min = inet_csk(sk)->icsk_rto_min; 716 717 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 718 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 719 return rto_min; 720 } 721 tcp_rto_min_us(struct sock * sk)722 static inline u32 tcp_rto_min_us(struct sock *sk) 723 { 724 return jiffies_to_usecs(tcp_rto_min(sk)); 725 } 726 tcp_ca_dst_locked(const struct dst_entry * dst)727 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 728 { 729 return dst_metric_locked(dst, RTAX_CC_ALGO); 730 } 731 732 /* Minimum RTT in usec. ~0 means not available. */ tcp_min_rtt(const struct tcp_sock * tp)733 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 734 { 735 return minmax_get(&tp->rtt_min); 736 } 737 738 /* Compute the actual receive window we are currently advertising. 739 * Rcv_nxt can be after the window if our peer push more data 740 * than the offered window. 741 */ tcp_receive_window(const struct tcp_sock * tp)742 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 743 { 744 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 745 746 if (win < 0) 747 win = 0; 748 return (u32) win; 749 } 750 751 /* Choose a new window, without checks for shrinking, and without 752 * scaling applied to the result. The caller does these things 753 * if necessary. This is a "raw" window selection. 754 */ 755 u32 __tcp_select_window(struct sock *sk); 756 757 void tcp_send_window_probe(struct sock *sk); 758 759 /* TCP uses 32bit jiffies to save some space. 760 * Note that this is different from tcp_time_stamp, which 761 * historically has been the same until linux-4.13. 762 */ 763 #define tcp_jiffies32 ((u32)jiffies) 764 765 /* 766 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 767 * It is no longer tied to jiffies, but to 1 ms clock. 768 * Note: double check if you want to use tcp_jiffies32 instead of this. 769 */ 770 #define TCP_TS_HZ 1000 771 tcp_clock_ns(void)772 static inline u64 tcp_clock_ns(void) 773 { 774 return ktime_get_ns(); 775 } 776 tcp_clock_us(void)777 static inline u64 tcp_clock_us(void) 778 { 779 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 780 } 781 782 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ tcp_time_stamp(const struct tcp_sock * tp)783 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 784 { 785 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 786 } 787 788 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ tcp_ns_to_ts(u64 ns)789 static inline u64 tcp_ns_to_ts(u64 ns) 790 { 791 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 792 } 793 794 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ tcp_time_stamp_raw(void)795 static inline u32 tcp_time_stamp_raw(void) 796 { 797 return tcp_ns_to_ts(tcp_clock_ns()); 798 } 799 800 void tcp_mstamp_refresh(struct tcp_sock *tp); 801 tcp_stamp_us_delta(u64 t1,u64 t0)802 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 803 { 804 return max_t(s64, t1 - t0, 0); 805 } 806 tcp_skb_timestamp(const struct sk_buff * skb)807 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 808 { 809 return tcp_ns_to_ts(skb->skb_mstamp_ns); 810 } 811 812 /* provide the departure time in us unit */ tcp_skb_timestamp_us(const struct sk_buff * skb)813 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 814 { 815 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 816 } 817 818 819 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 820 821 #define TCPHDR_FIN 0x01 822 #define TCPHDR_SYN 0x02 823 #define TCPHDR_RST 0x04 824 #define TCPHDR_PSH 0x08 825 #define TCPHDR_ACK 0x10 826 #define TCPHDR_URG 0x20 827 #define TCPHDR_ECE 0x40 828 #define TCPHDR_CWR 0x80 829 830 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 831 832 /* This is what the send packet queuing engine uses to pass 833 * TCP per-packet control information to the transmission code. 834 * We also store the host-order sequence numbers in here too. 835 * This is 44 bytes if IPV6 is enabled. 836 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 837 */ 838 struct tcp_skb_cb { 839 __u32 seq; /* Starting sequence number */ 840 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 841 union { 842 /* Note : tcp_tw_isn is used in input path only 843 * (isn chosen by tcp_timewait_state_process()) 844 * 845 * tcp_gso_segs/size are used in write queue only, 846 * cf tcp_skb_pcount()/tcp_skb_mss() 847 */ 848 __u32 tcp_tw_isn; 849 struct { 850 u16 tcp_gso_segs; 851 u16 tcp_gso_size; 852 }; 853 }; 854 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 855 856 __u8 sacked; /* State flags for SACK. */ 857 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 858 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 859 #define TCPCB_LOST 0x04 /* SKB is lost */ 860 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 861 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 862 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 863 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 864 TCPCB_REPAIRED) 865 866 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 867 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 868 eor:1, /* Is skb MSG_EOR marked? */ 869 has_rxtstamp:1, /* SKB has a RX timestamp */ 870 unused:5; 871 __u32 ack_seq; /* Sequence number ACK'd */ 872 union { 873 struct { 874 /* There is space for up to 24 bytes */ 875 __u32 in_flight:30,/* Bytes in flight at transmit */ 876 is_app_limited:1, /* cwnd not fully used? */ 877 unused:1; 878 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 879 __u32 delivered; 880 /* start of send pipeline phase */ 881 u64 first_tx_mstamp; 882 /* when we reached the "delivered" count */ 883 u64 delivered_mstamp; 884 } tx; /* only used for outgoing skbs */ 885 union { 886 struct inet_skb_parm h4; 887 #if IS_ENABLED(CONFIG_IPV6) 888 struct inet6_skb_parm h6; 889 #endif 890 } header; /* For incoming skbs */ 891 }; 892 }; 893 894 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 895 896 extern const struct inet_connection_sock_af_ops ipv4_specific; 897 898 #if IS_ENABLED(CONFIG_IPV6) 899 /* This is the variant of inet6_iif() that must be used by TCP, 900 * as TCP moves IP6CB into a different location in skb->cb[] 901 */ tcp_v6_iif(const struct sk_buff * skb)902 static inline int tcp_v6_iif(const struct sk_buff *skb) 903 { 904 return TCP_SKB_CB(skb)->header.h6.iif; 905 } 906 tcp_v6_iif_l3_slave(const struct sk_buff * skb)907 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 908 { 909 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 910 911 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 912 } 913 914 /* TCP_SKB_CB reference means this can not be used from early demux */ tcp_v6_sdif(const struct sk_buff * skb)915 static inline int tcp_v6_sdif(const struct sk_buff *skb) 916 { 917 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 918 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 919 return TCP_SKB_CB(skb)->header.h6.iif; 920 #endif 921 return 0; 922 } 923 924 extern const struct inet_connection_sock_af_ops ipv6_specific; 925 926 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 927 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 928 void tcp_v6_early_demux(struct sk_buff *skb); 929 930 #endif 931 932 /* TCP_SKB_CB reference means this can not be used from early demux */ tcp_v4_sdif(struct sk_buff * skb)933 static inline int tcp_v4_sdif(struct sk_buff *skb) 934 { 935 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 936 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 937 return TCP_SKB_CB(skb)->header.h4.iif; 938 #endif 939 return 0; 940 } 941 942 /* Due to TSO, an SKB can be composed of multiple actual 943 * packets. To keep these tracked properly, we use this. 944 */ tcp_skb_pcount(const struct sk_buff * skb)945 static inline int tcp_skb_pcount(const struct sk_buff *skb) 946 { 947 return TCP_SKB_CB(skb)->tcp_gso_segs; 948 } 949 tcp_skb_pcount_set(struct sk_buff * skb,int segs)950 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 951 { 952 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 953 } 954 tcp_skb_pcount_add(struct sk_buff * skb,int segs)955 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 956 { 957 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 958 } 959 960 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ tcp_skb_mss(const struct sk_buff * skb)961 static inline int tcp_skb_mss(const struct sk_buff *skb) 962 { 963 return TCP_SKB_CB(skb)->tcp_gso_size; 964 } 965 tcp_skb_can_collapse_to(const struct sk_buff * skb)966 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 967 { 968 return likely(!TCP_SKB_CB(skb)->eor); 969 } 970 tcp_skb_can_collapse(const struct sk_buff * to,const struct sk_buff * from)971 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 972 const struct sk_buff *from) 973 { 974 return likely(tcp_skb_can_collapse_to(to) && 975 mptcp_skb_can_collapse(to, from)); 976 } 977 978 /* Events passed to congestion control interface */ 979 enum tcp_ca_event { 980 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 981 CA_EVENT_CWND_RESTART, /* congestion window restart */ 982 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 983 CA_EVENT_LOSS, /* loss timeout */ 984 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 985 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 986 }; 987 988 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 989 enum tcp_ca_ack_event_flags { 990 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 991 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 992 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 993 }; 994 995 /* 996 * Interface for adding new TCP congestion control handlers 997 */ 998 #define TCP_CA_NAME_MAX 16 999 #define TCP_CA_MAX 128 1000 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1001 1002 #define TCP_CA_UNSPEC 0 1003 1004 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1005 #define TCP_CONG_NON_RESTRICTED 0x1 1006 /* Requires ECN/ECT set on all packets */ 1007 #define TCP_CONG_NEEDS_ECN 0x2 1008 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1009 1010 union tcp_cc_info; 1011 1012 struct ack_sample { 1013 u32 pkts_acked; 1014 s32 rtt_us; 1015 u32 in_flight; 1016 }; 1017 1018 /* A rate sample measures the number of (original/retransmitted) data 1019 * packets delivered "delivered" over an interval of time "interval_us". 1020 * The tcp_rate.c code fills in the rate sample, and congestion 1021 * control modules that define a cong_control function to run at the end 1022 * of ACK processing can optionally chose to consult this sample when 1023 * setting cwnd and pacing rate. 1024 * A sample is invalid if "delivered" or "interval_us" is negative. 1025 */ 1026 struct rate_sample { 1027 u64 prior_mstamp; /* starting timestamp for interval */ 1028 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1029 s32 delivered; /* number of packets delivered over interval */ 1030 long interval_us; /* time for tp->delivered to incr "delivered" */ 1031 u32 snd_interval_us; /* snd interval for delivered packets */ 1032 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1033 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1034 int losses; /* number of packets marked lost upon ACK */ 1035 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1036 u32 prior_in_flight; /* in flight before this ACK */ 1037 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1038 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1039 bool is_retrans; /* is sample from retransmission? */ 1040 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1041 }; 1042 1043 struct tcp_congestion_ops { 1044 /* fast path fields are put first to fill one cache line */ 1045 1046 /* return slow start threshold (required) */ 1047 u32 (*ssthresh)(struct sock *sk); 1048 1049 /* do new cwnd calculation (required) */ 1050 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1051 1052 /* call before changing ca_state (optional) */ 1053 void (*set_state)(struct sock *sk, u8 new_state); 1054 1055 /* call when cwnd event occurs (optional) */ 1056 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1057 1058 /* call when ack arrives (optional) */ 1059 void (*in_ack_event)(struct sock *sk, u32 flags); 1060 1061 /* hook for packet ack accounting (optional) */ 1062 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1063 1064 /* override sysctl_tcp_min_tso_segs */ 1065 u32 (*min_tso_segs)(struct sock *sk); 1066 1067 /* call when packets are delivered to update cwnd and pacing rate, 1068 * after all the ca_state processing. (optional) 1069 */ 1070 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1071 1072 1073 /* new value of cwnd after loss (required) */ 1074 u32 (*undo_cwnd)(struct sock *sk); 1075 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1076 u32 (*sndbuf_expand)(struct sock *sk); 1077 1078 /* control/slow paths put last */ 1079 /* get info for inet_diag (optional) */ 1080 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1081 union tcp_cc_info *info); 1082 1083 char name[TCP_CA_NAME_MAX]; 1084 struct module *owner; 1085 struct list_head list; 1086 u32 key; 1087 u32 flags; 1088 1089 /* initialize private data (optional) */ 1090 void (*init)(struct sock *sk); 1091 /* cleanup private data (optional) */ 1092 void (*release)(struct sock *sk); 1093 } ____cacheline_aligned_in_smp; 1094 1095 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1096 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1097 1098 void tcp_assign_congestion_control(struct sock *sk); 1099 void tcp_init_congestion_control(struct sock *sk); 1100 void tcp_cleanup_congestion_control(struct sock *sk); 1101 int tcp_set_default_congestion_control(struct net *net, const char *name); 1102 void tcp_get_default_congestion_control(struct net *net, char *name); 1103 void tcp_get_available_congestion_control(char *buf, size_t len); 1104 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1105 int tcp_set_allowed_congestion_control(char *allowed); 1106 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1107 bool cap_net_admin); 1108 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1109 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1110 1111 u32 tcp_reno_ssthresh(struct sock *sk); 1112 u32 tcp_reno_undo_cwnd(struct sock *sk); 1113 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1114 extern struct tcp_congestion_ops tcp_reno; 1115 1116 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1117 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1118 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1119 #ifdef CONFIG_INET 1120 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1121 #else tcp_ca_get_name_by_key(u32 key,char * buffer)1122 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1123 { 1124 return NULL; 1125 } 1126 #endif 1127 tcp_ca_needs_ecn(const struct sock * sk)1128 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1129 { 1130 const struct inet_connection_sock *icsk = inet_csk(sk); 1131 1132 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1133 } 1134 tcp_set_ca_state(struct sock * sk,const u8 ca_state)1135 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1136 { 1137 struct inet_connection_sock *icsk = inet_csk(sk); 1138 1139 if (icsk->icsk_ca_ops->set_state) 1140 icsk->icsk_ca_ops->set_state(sk, ca_state); 1141 icsk->icsk_ca_state = ca_state; 1142 } 1143 tcp_ca_event(struct sock * sk,const enum tcp_ca_event event)1144 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1145 { 1146 const struct inet_connection_sock *icsk = inet_csk(sk); 1147 1148 if (icsk->icsk_ca_ops->cwnd_event) 1149 icsk->icsk_ca_ops->cwnd_event(sk, event); 1150 } 1151 1152 /* From tcp_rate.c */ 1153 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1154 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1155 struct rate_sample *rs); 1156 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1157 bool is_sack_reneg, struct rate_sample *rs); 1158 void tcp_rate_check_app_limited(struct sock *sk); 1159 tcp_skb_sent_after(u64 t1,u64 t2,u32 seq1,u32 seq2)1160 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1161 { 1162 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1163 } 1164 1165 /* These functions determine how the current flow behaves in respect of SACK 1166 * handling. SACK is negotiated with the peer, and therefore it can vary 1167 * between different flows. 1168 * 1169 * tcp_is_sack - SACK enabled 1170 * tcp_is_reno - No SACK 1171 */ tcp_is_sack(const struct tcp_sock * tp)1172 static inline int tcp_is_sack(const struct tcp_sock *tp) 1173 { 1174 return likely(tp->rx_opt.sack_ok); 1175 } 1176 tcp_is_reno(const struct tcp_sock * tp)1177 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1178 { 1179 return !tcp_is_sack(tp); 1180 } 1181 tcp_left_out(const struct tcp_sock * tp)1182 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1183 { 1184 return tp->sacked_out + tp->lost_out; 1185 } 1186 1187 /* This determines how many packets are "in the network" to the best 1188 * of our knowledge. In many cases it is conservative, but where 1189 * detailed information is available from the receiver (via SACK 1190 * blocks etc.) we can make more aggressive calculations. 1191 * 1192 * Use this for decisions involving congestion control, use just 1193 * tp->packets_out to determine if the send queue is empty or not. 1194 * 1195 * Read this equation as: 1196 * 1197 * "Packets sent once on transmission queue" MINUS 1198 * "Packets left network, but not honestly ACKed yet" PLUS 1199 * "Packets fast retransmitted" 1200 */ tcp_packets_in_flight(const struct tcp_sock * tp)1201 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1202 { 1203 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1204 } 1205 1206 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1207 tcp_snd_cwnd(const struct tcp_sock * tp)1208 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1209 { 1210 return tp->snd_cwnd; 1211 } 1212 tcp_snd_cwnd_set(struct tcp_sock * tp,u32 val)1213 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1214 { 1215 WARN_ON_ONCE((int)val <= 0); 1216 tp->snd_cwnd = val; 1217 } 1218 tcp_in_slow_start(const struct tcp_sock * tp)1219 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1220 { 1221 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1222 } 1223 tcp_in_initial_slowstart(const struct tcp_sock * tp)1224 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1225 { 1226 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1227 } 1228 tcp_in_cwnd_reduction(const struct sock * sk)1229 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1230 { 1231 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1232 (1 << inet_csk(sk)->icsk_ca_state); 1233 } 1234 1235 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1236 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1237 * ssthresh. 1238 */ tcp_current_ssthresh(const struct sock * sk)1239 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1240 { 1241 const struct tcp_sock *tp = tcp_sk(sk); 1242 1243 if (tcp_in_cwnd_reduction(sk)) 1244 return tp->snd_ssthresh; 1245 else 1246 return max(tp->snd_ssthresh, 1247 ((tcp_snd_cwnd(tp) >> 1) + 1248 (tcp_snd_cwnd(tp) >> 2))); 1249 } 1250 1251 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1252 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1253 1254 void tcp_enter_cwr(struct sock *sk); 1255 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1256 1257 /* The maximum number of MSS of available cwnd for which TSO defers 1258 * sending if not using sysctl_tcp_tso_win_divisor. 1259 */ tcp_max_tso_deferred_mss(const struct tcp_sock * tp)1260 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1261 { 1262 return 3; 1263 } 1264 1265 /* Returns end sequence number of the receiver's advertised window */ tcp_wnd_end(const struct tcp_sock * tp)1266 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1267 { 1268 return tp->snd_una + tp->snd_wnd; 1269 } 1270 1271 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1272 * flexible approach. The RFC suggests cwnd should not be raised unless 1273 * it was fully used previously. And that's exactly what we do in 1274 * congestion avoidance mode. But in slow start we allow cwnd to grow 1275 * as long as the application has used half the cwnd. 1276 * Example : 1277 * cwnd is 10 (IW10), but application sends 9 frames. 1278 * We allow cwnd to reach 18 when all frames are ACKed. 1279 * This check is safe because it's as aggressive as slow start which already 1280 * risks 100% overshoot. The advantage is that we discourage application to 1281 * either send more filler packets or data to artificially blow up the cwnd 1282 * usage, and allow application-limited process to probe bw more aggressively. 1283 */ tcp_is_cwnd_limited(const struct sock * sk)1284 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1285 { 1286 const struct tcp_sock *tp = tcp_sk(sk); 1287 1288 if (tp->is_cwnd_limited) 1289 return true; 1290 1291 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1292 if (tcp_in_slow_start(tp)) 1293 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1294 1295 return false; 1296 } 1297 1298 /* BBR congestion control needs pacing. 1299 * Same remark for SO_MAX_PACING_RATE. 1300 * sch_fq packet scheduler is efficiently handling pacing, 1301 * but is not always installed/used. 1302 * Return true if TCP stack should pace packets itself. 1303 */ tcp_needs_internal_pacing(const struct sock * sk)1304 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1305 { 1306 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1307 } 1308 1309 /* Estimates in how many jiffies next packet for this flow can be sent. 1310 * Scheduling a retransmit timer too early would be silly. 1311 */ tcp_pacing_delay(const struct sock * sk)1312 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1313 { 1314 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1315 1316 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1317 } 1318 tcp_reset_xmit_timer(struct sock * sk,const int what,unsigned long when,const unsigned long max_when)1319 static inline void tcp_reset_xmit_timer(struct sock *sk, 1320 const int what, 1321 unsigned long when, 1322 const unsigned long max_when) 1323 { 1324 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1325 max_when); 1326 } 1327 1328 /* Something is really bad, we could not queue an additional packet, 1329 * because qdisc is full or receiver sent a 0 window, or we are paced. 1330 * We do not want to add fuel to the fire, or abort too early, 1331 * so make sure the timer we arm now is at least 200ms in the future, 1332 * regardless of current icsk_rto value (as it could be ~2ms) 1333 */ tcp_probe0_base(const struct sock * sk)1334 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1335 { 1336 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1337 } 1338 1339 /* Variant of inet_csk_rto_backoff() used for zero window probes */ tcp_probe0_when(const struct sock * sk,unsigned long max_when)1340 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1341 unsigned long max_when) 1342 { 1343 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1344 inet_csk(sk)->icsk_backoff); 1345 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1346 1347 return (unsigned long)min_t(u64, when, max_when); 1348 } 1349 tcp_check_probe_timer(struct sock * sk)1350 static inline void tcp_check_probe_timer(struct sock *sk) 1351 { 1352 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1353 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1354 tcp_probe0_base(sk), TCP_RTO_MAX); 1355 } 1356 tcp_init_wl(struct tcp_sock * tp,u32 seq)1357 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1358 { 1359 tp->snd_wl1 = seq; 1360 } 1361 tcp_update_wl(struct tcp_sock * tp,u32 seq)1362 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1363 { 1364 tp->snd_wl1 = seq; 1365 } 1366 1367 /* 1368 * Calculate(/check) TCP checksum 1369 */ tcp_v4_check(int len,__be32 saddr,__be32 daddr,__wsum base)1370 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1371 __be32 daddr, __wsum base) 1372 { 1373 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1374 } 1375 tcp_checksum_complete(struct sk_buff * skb)1376 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1377 { 1378 return !skb_csum_unnecessary(skb) && 1379 __skb_checksum_complete(skb); 1380 } 1381 1382 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1383 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1384 void tcp_set_state(struct sock *sk, int state); 1385 void tcp_done(struct sock *sk); 1386 int tcp_abort(struct sock *sk, int err); 1387 tcp_sack_reset(struct tcp_options_received * rx_opt)1388 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1389 { 1390 rx_opt->dsack = 0; 1391 rx_opt->num_sacks = 0; 1392 } 1393 1394 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1395 tcp_slow_start_after_idle_check(struct sock * sk)1396 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1397 { 1398 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1399 struct tcp_sock *tp = tcp_sk(sk); 1400 s32 delta; 1401 1402 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1403 tp->packets_out || ca_ops->cong_control) 1404 return; 1405 delta = tcp_jiffies32 - tp->lsndtime; 1406 if (delta > inet_csk(sk)->icsk_rto) 1407 tcp_cwnd_restart(sk, delta); 1408 } 1409 1410 /* Determine a window scaling and initial window to offer. */ 1411 void tcp_select_initial_window(const struct sock *sk, int __space, 1412 __u32 mss, __u32 *rcv_wnd, 1413 __u32 *window_clamp, int wscale_ok, 1414 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1415 tcp_win_from_space(const struct sock * sk,int space)1416 static inline int tcp_win_from_space(const struct sock *sk, int space) 1417 { 1418 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale); 1419 1420 return tcp_adv_win_scale <= 0 ? 1421 (space>>(-tcp_adv_win_scale)) : 1422 space - (space>>tcp_adv_win_scale); 1423 } 1424 1425 /* Note: caller must be prepared to deal with negative returns */ tcp_space(const struct sock * sk)1426 static inline int tcp_space(const struct sock *sk) 1427 { 1428 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1429 READ_ONCE(sk->sk_backlog.len) - 1430 atomic_read(&sk->sk_rmem_alloc)); 1431 } 1432 tcp_full_space(const struct sock * sk)1433 static inline int tcp_full_space(const struct sock *sk) 1434 { 1435 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1436 } 1437 1438 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1439 1440 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1441 * If 87.5 % (7/8) of the space has been consumed, we want to override 1442 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1443 * len/truesize ratio. 1444 */ tcp_rmem_pressure(const struct sock * sk)1445 static inline bool tcp_rmem_pressure(const struct sock *sk) 1446 { 1447 int rcvbuf, threshold; 1448 1449 if (tcp_under_memory_pressure(sk)) 1450 return true; 1451 1452 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1453 threshold = rcvbuf - (rcvbuf >> 3); 1454 1455 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1456 } 1457 tcp_epollin_ready(const struct sock * sk,int target)1458 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1459 { 1460 const struct tcp_sock *tp = tcp_sk(sk); 1461 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1462 1463 if (avail <= 0) 1464 return false; 1465 1466 return (avail >= target) || tcp_rmem_pressure(sk) || 1467 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1468 } 1469 1470 extern void tcp_openreq_init_rwin(struct request_sock *req, 1471 const struct sock *sk_listener, 1472 const struct dst_entry *dst); 1473 1474 void tcp_enter_memory_pressure(struct sock *sk); 1475 void tcp_leave_memory_pressure(struct sock *sk); 1476 keepalive_intvl_when(const struct tcp_sock * tp)1477 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1478 { 1479 struct net *net = sock_net((struct sock *)tp); 1480 int val; 1481 1482 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1483 * and do_tcp_setsockopt(). 1484 */ 1485 val = READ_ONCE(tp->keepalive_intvl); 1486 1487 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1488 } 1489 keepalive_time_when(const struct tcp_sock * tp)1490 static inline int keepalive_time_when(const struct tcp_sock *tp) 1491 { 1492 struct net *net = sock_net((struct sock *)tp); 1493 int val; 1494 1495 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1496 val = READ_ONCE(tp->keepalive_time); 1497 1498 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1499 } 1500 keepalive_probes(const struct tcp_sock * tp)1501 static inline int keepalive_probes(const struct tcp_sock *tp) 1502 { 1503 struct net *net = sock_net((struct sock *)tp); 1504 int val; 1505 1506 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1507 * and do_tcp_setsockopt(). 1508 */ 1509 val = READ_ONCE(tp->keepalive_probes); 1510 1511 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1512 } 1513 keepalive_time_elapsed(const struct tcp_sock * tp)1514 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1515 { 1516 const struct inet_connection_sock *icsk = &tp->inet_conn; 1517 1518 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1519 tcp_jiffies32 - tp->rcv_tstamp); 1520 } 1521 tcp_fin_time(const struct sock * sk)1522 static inline int tcp_fin_time(const struct sock *sk) 1523 { 1524 int fin_timeout = tcp_sk(sk)->linger2 ? : 1525 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1526 const int rto = inet_csk(sk)->icsk_rto; 1527 1528 if (fin_timeout < (rto << 2) - (rto >> 1)) 1529 fin_timeout = (rto << 2) - (rto >> 1); 1530 1531 return fin_timeout; 1532 } 1533 tcp_paws_check(const struct tcp_options_received * rx_opt,int paws_win)1534 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1535 int paws_win) 1536 { 1537 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1538 return true; 1539 if (unlikely(!time_before32(ktime_get_seconds(), 1540 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1541 return true; 1542 /* 1543 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1544 * then following tcp messages have valid values. Ignore 0 value, 1545 * or else 'negative' tsval might forbid us to accept their packets. 1546 */ 1547 if (!rx_opt->ts_recent) 1548 return true; 1549 return false; 1550 } 1551 tcp_paws_reject(const struct tcp_options_received * rx_opt,int rst)1552 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1553 int rst) 1554 { 1555 if (tcp_paws_check(rx_opt, 0)) 1556 return false; 1557 1558 /* RST segments are not recommended to carry timestamp, 1559 and, if they do, it is recommended to ignore PAWS because 1560 "their cleanup function should take precedence over timestamps." 1561 Certainly, it is mistake. It is necessary to understand the reasons 1562 of this constraint to relax it: if peer reboots, clock may go 1563 out-of-sync and half-open connections will not be reset. 1564 Actually, the problem would be not existing if all 1565 the implementations followed draft about maintaining clock 1566 via reboots. Linux-2.2 DOES NOT! 1567 1568 However, we can relax time bounds for RST segments to MSL. 1569 */ 1570 if (rst && !time_before32(ktime_get_seconds(), 1571 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1572 return false; 1573 return true; 1574 } 1575 1576 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1577 int mib_idx, u32 *last_oow_ack_time); 1578 tcp_mib_init(struct net * net)1579 static inline void tcp_mib_init(struct net *net) 1580 { 1581 /* See RFC 2012 */ 1582 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1583 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1584 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1585 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1586 } 1587 1588 /* from STCP */ tcp_clear_retrans_hints_partial(struct tcp_sock * tp)1589 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1590 { 1591 tp->lost_skb_hint = NULL; 1592 } 1593 tcp_clear_all_retrans_hints(struct tcp_sock * tp)1594 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1595 { 1596 tcp_clear_retrans_hints_partial(tp); 1597 tp->retransmit_skb_hint = NULL; 1598 } 1599 1600 union tcp_md5_addr { 1601 struct in_addr a4; 1602 #if IS_ENABLED(CONFIG_IPV6) 1603 struct in6_addr a6; 1604 #endif 1605 }; 1606 1607 /* - key database */ 1608 struct tcp_md5sig_key { 1609 struct hlist_node node; 1610 u8 keylen; 1611 u8 family; /* AF_INET or AF_INET6 */ 1612 u8 prefixlen; 1613 u8 flags; 1614 union tcp_md5_addr addr; 1615 int l3index; /* set if key added with L3 scope */ 1616 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1617 struct rcu_head rcu; 1618 }; 1619 1620 /* - sock block */ 1621 struct tcp_md5sig_info { 1622 struct hlist_head head; 1623 struct rcu_head rcu; 1624 }; 1625 1626 /* - pseudo header */ 1627 struct tcp4_pseudohdr { 1628 __be32 saddr; 1629 __be32 daddr; 1630 __u8 pad; 1631 __u8 protocol; 1632 __be16 len; 1633 }; 1634 1635 struct tcp6_pseudohdr { 1636 struct in6_addr saddr; 1637 struct in6_addr daddr; 1638 __be32 len; 1639 __be32 protocol; /* including padding */ 1640 }; 1641 1642 union tcp_md5sum_block { 1643 struct tcp4_pseudohdr ip4; 1644 #if IS_ENABLED(CONFIG_IPV6) 1645 struct tcp6_pseudohdr ip6; 1646 #endif 1647 }; 1648 1649 /* - pool: digest algorithm, hash description and scratch buffer */ 1650 struct tcp_md5sig_pool { 1651 struct ahash_request *md5_req; 1652 void *scratch; 1653 }; 1654 1655 /* - functions */ 1656 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1657 const struct sock *sk, const struct sk_buff *skb); 1658 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1659 int family, u8 prefixlen, int l3index, u8 flags, 1660 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1661 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1662 int family, u8 prefixlen, int l3index, u8 flags); 1663 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1664 const struct sock *addr_sk); 1665 1666 #ifdef CONFIG_TCP_MD5SIG 1667 #include <linux/jump_label.h> 1668 extern struct static_key_false tcp_md5_needed; 1669 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1670 const union tcp_md5_addr *addr, 1671 int family); 1672 static inline struct tcp_md5sig_key * tcp_md5_do_lookup(const struct sock * sk,int l3index,const union tcp_md5_addr * addr,int family)1673 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1674 const union tcp_md5_addr *addr, int family) 1675 { 1676 if (!static_branch_unlikely(&tcp_md5_needed)) 1677 return NULL; 1678 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1679 } 1680 1681 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1682 #else 1683 static inline struct tcp_md5sig_key * tcp_md5_do_lookup(const struct sock * sk,int l3index,const union tcp_md5_addr * addr,int family)1684 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1685 const union tcp_md5_addr *addr, int family) 1686 { 1687 return NULL; 1688 } 1689 #define tcp_twsk_md5_key(twsk) NULL 1690 #endif 1691 1692 bool tcp_alloc_md5sig_pool(void); 1693 1694 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); tcp_put_md5sig_pool(void)1695 static inline void tcp_put_md5sig_pool(void) 1696 { 1697 local_bh_enable(); 1698 } 1699 1700 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1701 unsigned int header_len); 1702 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1703 const struct tcp_md5sig_key *key); 1704 1705 /* From tcp_fastopen.c */ 1706 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1707 struct tcp_fastopen_cookie *cookie); 1708 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1709 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1710 u16 try_exp); 1711 struct tcp_fastopen_request { 1712 /* Fast Open cookie. Size 0 means a cookie request */ 1713 struct tcp_fastopen_cookie cookie; 1714 struct msghdr *data; /* data in MSG_FASTOPEN */ 1715 size_t size; 1716 int copied; /* queued in tcp_connect() */ 1717 struct ubuf_info *uarg; 1718 }; 1719 void tcp_free_fastopen_req(struct tcp_sock *tp); 1720 void tcp_fastopen_destroy_cipher(struct sock *sk); 1721 void tcp_fastopen_ctx_destroy(struct net *net); 1722 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1723 void *primary_key, void *backup_key); 1724 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1725 u64 *key); 1726 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1727 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1728 struct request_sock *req, 1729 struct tcp_fastopen_cookie *foc, 1730 const struct dst_entry *dst); 1731 void tcp_fastopen_init_key_once(struct net *net); 1732 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1733 struct tcp_fastopen_cookie *cookie); 1734 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1735 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1736 #define TCP_FASTOPEN_KEY_MAX 2 1737 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1738 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1739 1740 /* Fastopen key context */ 1741 struct tcp_fastopen_context { 1742 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1743 int num; 1744 struct rcu_head rcu; 1745 }; 1746 1747 void tcp_fastopen_active_disable(struct sock *sk); 1748 bool tcp_fastopen_active_should_disable(struct sock *sk); 1749 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1750 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1751 1752 /* Caller needs to wrap with rcu_read_(un)lock() */ 1753 static inline tcp_fastopen_get_ctx(const struct sock * sk)1754 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1755 { 1756 struct tcp_fastopen_context *ctx; 1757 1758 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1759 if (!ctx) 1760 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1761 return ctx; 1762 } 1763 1764 static inline tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie * foc,const struct tcp_fastopen_cookie * orig)1765 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1766 const struct tcp_fastopen_cookie *orig) 1767 { 1768 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1769 orig->len == foc->len && 1770 !memcmp(orig->val, foc->val, foc->len)) 1771 return true; 1772 return false; 1773 } 1774 1775 static inline tcp_fastopen_context_len(const struct tcp_fastopen_context * ctx)1776 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1777 { 1778 return ctx->num; 1779 } 1780 1781 /* Latencies incurred by various limits for a sender. They are 1782 * chronograph-like stats that are mutually exclusive. 1783 */ 1784 enum tcp_chrono { 1785 TCP_CHRONO_UNSPEC, 1786 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1787 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1788 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1789 __TCP_CHRONO_MAX, 1790 }; 1791 1792 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1793 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1794 1795 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1796 * the same memory storage than skb->destructor/_skb_refdst 1797 */ tcp_skb_tsorted_anchor_cleanup(struct sk_buff * skb)1798 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1799 { 1800 skb->destructor = NULL; 1801 skb->_skb_refdst = 0UL; 1802 } 1803 1804 #define tcp_skb_tsorted_save(skb) { \ 1805 unsigned long _save = skb->_skb_refdst; \ 1806 skb->_skb_refdst = 0UL; 1807 1808 #define tcp_skb_tsorted_restore(skb) \ 1809 skb->_skb_refdst = _save; \ 1810 } 1811 1812 void tcp_write_queue_purge(struct sock *sk); 1813 tcp_rtx_queue_head(const struct sock * sk)1814 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1815 { 1816 return skb_rb_first(&sk->tcp_rtx_queue); 1817 } 1818 tcp_rtx_queue_tail(const struct sock * sk)1819 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1820 { 1821 return skb_rb_last(&sk->tcp_rtx_queue); 1822 } 1823 tcp_write_queue_head(const struct sock * sk)1824 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1825 { 1826 return skb_peek(&sk->sk_write_queue); 1827 } 1828 tcp_write_queue_tail(const struct sock * sk)1829 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1830 { 1831 return skb_peek_tail(&sk->sk_write_queue); 1832 } 1833 1834 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1835 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1836 tcp_send_head(const struct sock * sk)1837 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1838 { 1839 return skb_peek(&sk->sk_write_queue); 1840 } 1841 tcp_skb_is_last(const struct sock * sk,const struct sk_buff * skb)1842 static inline bool tcp_skb_is_last(const struct sock *sk, 1843 const struct sk_buff *skb) 1844 { 1845 return skb_queue_is_last(&sk->sk_write_queue, skb); 1846 } 1847 1848 /** 1849 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1850 * @sk: socket 1851 * 1852 * Since the write queue can have a temporary empty skb in it, 1853 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1854 */ tcp_write_queue_empty(const struct sock * sk)1855 static inline bool tcp_write_queue_empty(const struct sock *sk) 1856 { 1857 const struct tcp_sock *tp = tcp_sk(sk); 1858 1859 return tp->write_seq == tp->snd_nxt; 1860 } 1861 tcp_rtx_queue_empty(const struct sock * sk)1862 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1863 { 1864 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1865 } 1866 tcp_rtx_and_write_queues_empty(const struct sock * sk)1867 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1868 { 1869 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1870 } 1871 tcp_add_write_queue_tail(struct sock * sk,struct sk_buff * skb)1872 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1873 { 1874 __skb_queue_tail(&sk->sk_write_queue, skb); 1875 1876 /* Queue it, remembering where we must start sending. */ 1877 if (sk->sk_write_queue.next == skb) 1878 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1879 } 1880 1881 /* Insert new before skb on the write queue of sk. */ tcp_insert_write_queue_before(struct sk_buff * new,struct sk_buff * skb,struct sock * sk)1882 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1883 struct sk_buff *skb, 1884 struct sock *sk) 1885 { 1886 __skb_queue_before(&sk->sk_write_queue, skb, new); 1887 } 1888 tcp_unlink_write_queue(struct sk_buff * skb,struct sock * sk)1889 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1890 { 1891 tcp_skb_tsorted_anchor_cleanup(skb); 1892 __skb_unlink(skb, &sk->sk_write_queue); 1893 } 1894 1895 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1896 tcp_rtx_queue_unlink(struct sk_buff * skb,struct sock * sk)1897 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1898 { 1899 tcp_skb_tsorted_anchor_cleanup(skb); 1900 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1901 } 1902 tcp_rtx_queue_unlink_and_free(struct sk_buff * skb,struct sock * sk)1903 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1904 { 1905 list_del(&skb->tcp_tsorted_anchor); 1906 tcp_rtx_queue_unlink(skb, sk); 1907 sk_wmem_free_skb(sk, skb); 1908 } 1909 tcp_push_pending_frames(struct sock * sk)1910 static inline void tcp_push_pending_frames(struct sock *sk) 1911 { 1912 if (tcp_send_head(sk)) { 1913 struct tcp_sock *tp = tcp_sk(sk); 1914 1915 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1916 } 1917 } 1918 1919 /* Start sequence of the skb just after the highest skb with SACKed 1920 * bit, valid only if sacked_out > 0 or when the caller has ensured 1921 * validity by itself. 1922 */ tcp_highest_sack_seq(struct tcp_sock * tp)1923 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1924 { 1925 if (!tp->sacked_out) 1926 return tp->snd_una; 1927 1928 if (tp->highest_sack == NULL) 1929 return tp->snd_nxt; 1930 1931 return TCP_SKB_CB(tp->highest_sack)->seq; 1932 } 1933 tcp_advance_highest_sack(struct sock * sk,struct sk_buff * skb)1934 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1935 { 1936 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1937 } 1938 tcp_highest_sack(struct sock * sk)1939 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1940 { 1941 return tcp_sk(sk)->highest_sack; 1942 } 1943 tcp_highest_sack_reset(struct sock * sk)1944 static inline void tcp_highest_sack_reset(struct sock *sk) 1945 { 1946 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1947 } 1948 1949 /* Called when old skb is about to be deleted and replaced by new skb */ tcp_highest_sack_replace(struct sock * sk,struct sk_buff * old,struct sk_buff * new)1950 static inline void tcp_highest_sack_replace(struct sock *sk, 1951 struct sk_buff *old, 1952 struct sk_buff *new) 1953 { 1954 if (old == tcp_highest_sack(sk)) 1955 tcp_sk(sk)->highest_sack = new; 1956 } 1957 1958 /* This helper checks if socket has IP_TRANSPARENT set */ inet_sk_transparent(const struct sock * sk)1959 static inline bool inet_sk_transparent(const struct sock *sk) 1960 { 1961 switch (sk->sk_state) { 1962 case TCP_TIME_WAIT: 1963 return inet_twsk(sk)->tw_transparent; 1964 case TCP_NEW_SYN_RECV: 1965 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1966 } 1967 return inet_sk(sk)->transparent; 1968 } 1969 1970 /* Determines whether this is a thin stream (which may suffer from 1971 * increased latency). Used to trigger latency-reducing mechanisms. 1972 */ tcp_stream_is_thin(struct tcp_sock * tp)1973 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1974 { 1975 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1976 } 1977 1978 /* /proc */ 1979 enum tcp_seq_states { 1980 TCP_SEQ_STATE_LISTENING, 1981 TCP_SEQ_STATE_ESTABLISHED, 1982 }; 1983 1984 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1985 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1986 void tcp_seq_stop(struct seq_file *seq, void *v); 1987 1988 struct tcp_seq_afinfo { 1989 sa_family_t family; 1990 }; 1991 1992 struct tcp_iter_state { 1993 struct seq_net_private p; 1994 enum tcp_seq_states state; 1995 struct sock *syn_wait_sk; 1996 int bucket, offset, sbucket, num; 1997 loff_t last_pos; 1998 }; 1999 2000 extern struct request_sock_ops tcp_request_sock_ops; 2001 extern struct request_sock_ops tcp6_request_sock_ops; 2002 2003 void tcp_v4_destroy_sock(struct sock *sk); 2004 2005 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2006 netdev_features_t features); 2007 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2008 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2009 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2010 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2011 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2012 int tcp_gro_complete(struct sk_buff *skb); 2013 2014 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2015 tcp_notsent_lowat(const struct tcp_sock * tp)2016 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2017 { 2018 struct net *net = sock_net((struct sock *)tp); 2019 u32 val; 2020 2021 val = READ_ONCE(tp->notsent_lowat); 2022 2023 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2024 } 2025 2026 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2027 2028 #ifdef CONFIG_PROC_FS 2029 int tcp4_proc_init(void); 2030 void tcp4_proc_exit(void); 2031 #endif 2032 2033 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2034 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2035 const struct tcp_request_sock_ops *af_ops, 2036 struct sock *sk, struct sk_buff *skb); 2037 2038 /* TCP af-specific functions */ 2039 struct tcp_sock_af_ops { 2040 #ifdef CONFIG_TCP_MD5SIG 2041 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2042 const struct sock *addr_sk); 2043 int (*calc_md5_hash)(char *location, 2044 const struct tcp_md5sig_key *md5, 2045 const struct sock *sk, 2046 const struct sk_buff *skb); 2047 int (*md5_parse)(struct sock *sk, 2048 int optname, 2049 sockptr_t optval, 2050 int optlen); 2051 #endif 2052 }; 2053 2054 struct tcp_request_sock_ops { 2055 u16 mss_clamp; 2056 #ifdef CONFIG_TCP_MD5SIG 2057 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2058 const struct sock *addr_sk); 2059 int (*calc_md5_hash) (char *location, 2060 const struct tcp_md5sig_key *md5, 2061 const struct sock *sk, 2062 const struct sk_buff *skb); 2063 #endif 2064 #ifdef CONFIG_SYN_COOKIES 2065 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2066 __u16 *mss); 2067 #endif 2068 struct dst_entry *(*route_req)(const struct sock *sk, 2069 struct sk_buff *skb, 2070 struct flowi *fl, 2071 struct request_sock *req); 2072 u32 (*init_seq)(const struct sk_buff *skb); 2073 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2074 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2075 struct flowi *fl, struct request_sock *req, 2076 struct tcp_fastopen_cookie *foc, 2077 enum tcp_synack_type synack_type, 2078 struct sk_buff *syn_skb); 2079 }; 2080 2081 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2082 #if IS_ENABLED(CONFIG_IPV6) 2083 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2084 #endif 2085 2086 #ifdef CONFIG_SYN_COOKIES cookie_init_sequence(const struct tcp_request_sock_ops * ops,const struct sock * sk,struct sk_buff * skb,__u16 * mss)2087 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2088 const struct sock *sk, struct sk_buff *skb, 2089 __u16 *mss) 2090 { 2091 tcp_synq_overflow(sk); 2092 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2093 return ops->cookie_init_seq(skb, mss); 2094 } 2095 #else cookie_init_sequence(const struct tcp_request_sock_ops * ops,const struct sock * sk,struct sk_buff * skb,__u16 * mss)2096 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2097 const struct sock *sk, struct sk_buff *skb, 2098 __u16 *mss) 2099 { 2100 return 0; 2101 } 2102 #endif 2103 2104 int tcpv4_offload_init(void); 2105 2106 void tcp_v4_init(void); 2107 void tcp_init(void); 2108 2109 /* tcp_recovery.c */ 2110 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2111 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2112 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2113 u32 reo_wnd); 2114 extern bool tcp_rack_mark_lost(struct sock *sk); 2115 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2116 u64 xmit_time); 2117 extern void tcp_rack_reo_timeout(struct sock *sk); 2118 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2119 2120 /* At how many usecs into the future should the RTO fire? */ tcp_rto_delta_us(const struct sock * sk)2121 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2122 { 2123 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2124 u32 rto = inet_csk(sk)->icsk_rto; 2125 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2126 2127 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2128 } 2129 2130 /* 2131 * Save and compile IPv4 options, return a pointer to it 2132 */ tcp_v4_save_options(struct net * net,struct sk_buff * skb)2133 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2134 struct sk_buff *skb) 2135 { 2136 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2137 struct ip_options_rcu *dopt = NULL; 2138 2139 if (opt->optlen) { 2140 int opt_size = sizeof(*dopt) + opt->optlen; 2141 2142 dopt = kmalloc(opt_size, GFP_ATOMIC); 2143 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2144 kfree(dopt); 2145 dopt = NULL; 2146 } 2147 } 2148 return dopt; 2149 } 2150 2151 /* locally generated TCP pure ACKs have skb->truesize == 2 2152 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2153 * This is much faster than dissecting the packet to find out. 2154 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2155 */ skb_is_tcp_pure_ack(const struct sk_buff * skb)2156 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2157 { 2158 return skb->truesize == 2; 2159 } 2160 skb_set_tcp_pure_ack(struct sk_buff * skb)2161 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2162 { 2163 skb->truesize = 2; 2164 } 2165 tcp_inq(struct sock * sk)2166 static inline int tcp_inq(struct sock *sk) 2167 { 2168 struct tcp_sock *tp = tcp_sk(sk); 2169 int answ; 2170 2171 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2172 answ = 0; 2173 } else if (sock_flag(sk, SOCK_URGINLINE) || 2174 !tp->urg_data || 2175 before(tp->urg_seq, tp->copied_seq) || 2176 !before(tp->urg_seq, tp->rcv_nxt)) { 2177 2178 answ = tp->rcv_nxt - tp->copied_seq; 2179 2180 /* Subtract 1, if FIN was received */ 2181 if (answ && sock_flag(sk, SOCK_DONE)) 2182 answ--; 2183 } else { 2184 answ = tp->urg_seq - tp->copied_seq; 2185 } 2186 2187 return answ; 2188 } 2189 2190 int tcp_peek_len(struct socket *sock); 2191 tcp_segs_in(struct tcp_sock * tp,const struct sk_buff * skb)2192 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2193 { 2194 u16 segs_in; 2195 2196 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2197 tp->segs_in += segs_in; 2198 if (skb->len > tcp_hdrlen(skb)) 2199 tp->data_segs_in += segs_in; 2200 } 2201 2202 /* 2203 * TCP listen path runs lockless. 2204 * We forced "struct sock" to be const qualified to make sure 2205 * we don't modify one of its field by mistake. 2206 * Here, we increment sk_drops which is an atomic_t, so we can safely 2207 * make sock writable again. 2208 */ tcp_listendrop(const struct sock * sk)2209 static inline void tcp_listendrop(const struct sock *sk) 2210 { 2211 atomic_inc(&((struct sock *)sk)->sk_drops); 2212 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2213 } 2214 2215 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2216 2217 /* 2218 * Interface for adding Upper Level Protocols over TCP 2219 */ 2220 2221 #define TCP_ULP_NAME_MAX 16 2222 #define TCP_ULP_MAX 128 2223 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2224 2225 struct tcp_ulp_ops { 2226 struct list_head list; 2227 2228 /* initialize ulp */ 2229 int (*init)(struct sock *sk); 2230 /* update ulp */ 2231 void (*update)(struct sock *sk, struct proto *p, 2232 void (*write_space)(struct sock *sk)); 2233 /* cleanup ulp */ 2234 void (*release)(struct sock *sk); 2235 /* diagnostic */ 2236 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2237 size_t (*get_info_size)(const struct sock *sk); 2238 /* clone ulp */ 2239 void (*clone)(const struct request_sock *req, struct sock *newsk, 2240 const gfp_t priority); 2241 2242 char name[TCP_ULP_NAME_MAX]; 2243 struct module *owner; 2244 }; 2245 int tcp_register_ulp(struct tcp_ulp_ops *type); 2246 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2247 int tcp_set_ulp(struct sock *sk, const char *name); 2248 void tcp_get_available_ulp(char *buf, size_t len); 2249 void tcp_cleanup_ulp(struct sock *sk); 2250 void tcp_update_ulp(struct sock *sk, struct proto *p, 2251 void (*write_space)(struct sock *sk)); 2252 2253 #define MODULE_ALIAS_TCP_ULP(name) \ 2254 __MODULE_INFO(alias, alias_userspace, name); \ 2255 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2256 2257 #ifdef CONFIG_NET_SOCK_MSG 2258 struct sk_msg; 2259 struct sk_psock; 2260 2261 #ifdef CONFIG_BPF_SYSCALL 2262 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2263 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2264 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2265 #endif /* CONFIG_BPF_SYSCALL */ 2266 2267 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2268 int flags); 2269 #endif /* CONFIG_NET_SOCK_MSG */ 2270 2271 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) tcp_bpf_clone(const struct sock * sk,struct sock * newsk)2272 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2273 { 2274 } 2275 #endif 2276 2277 #ifdef CONFIG_CGROUP_BPF bpf_skops_init_skb(struct bpf_sock_ops_kern * skops,struct sk_buff * skb,unsigned int end_offset)2278 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2279 struct sk_buff *skb, 2280 unsigned int end_offset) 2281 { 2282 skops->skb = skb; 2283 skops->skb_data_end = skb->data + end_offset; 2284 } 2285 #else bpf_skops_init_skb(struct bpf_sock_ops_kern * skops,struct sk_buff * skb,unsigned int end_offset)2286 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2287 struct sk_buff *skb, 2288 unsigned int end_offset) 2289 { 2290 } 2291 #endif 2292 2293 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2294 * is < 0, then the BPF op failed (for example if the loaded BPF 2295 * program does not support the chosen operation or there is no BPF 2296 * program loaded). 2297 */ 2298 #ifdef CONFIG_BPF tcp_call_bpf(struct sock * sk,int op,u32 nargs,u32 * args)2299 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2300 { 2301 struct bpf_sock_ops_kern sock_ops; 2302 int ret; 2303 2304 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2305 if (sk_fullsock(sk)) { 2306 sock_ops.is_fullsock = 1; 2307 sock_owned_by_me(sk); 2308 } 2309 2310 sock_ops.sk = sk; 2311 sock_ops.op = op; 2312 if (nargs > 0) 2313 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2314 2315 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2316 if (ret == 0) 2317 ret = sock_ops.reply; 2318 else 2319 ret = -1; 2320 return ret; 2321 } 2322 tcp_call_bpf_2arg(struct sock * sk,int op,u32 arg1,u32 arg2)2323 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2324 { 2325 u32 args[2] = {arg1, arg2}; 2326 2327 return tcp_call_bpf(sk, op, 2, args); 2328 } 2329 tcp_call_bpf_3arg(struct sock * sk,int op,u32 arg1,u32 arg2,u32 arg3)2330 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2331 u32 arg3) 2332 { 2333 u32 args[3] = {arg1, arg2, arg3}; 2334 2335 return tcp_call_bpf(sk, op, 3, args); 2336 } 2337 2338 #else tcp_call_bpf(struct sock * sk,int op,u32 nargs,u32 * args)2339 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2340 { 2341 return -EPERM; 2342 } 2343 tcp_call_bpf_2arg(struct sock * sk,int op,u32 arg1,u32 arg2)2344 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2345 { 2346 return -EPERM; 2347 } 2348 tcp_call_bpf_3arg(struct sock * sk,int op,u32 arg1,u32 arg2,u32 arg3)2349 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2350 u32 arg3) 2351 { 2352 return -EPERM; 2353 } 2354 2355 #endif 2356 tcp_timeout_init(struct sock * sk)2357 static inline u32 tcp_timeout_init(struct sock *sk) 2358 { 2359 int timeout; 2360 2361 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2362 2363 if (timeout <= 0) 2364 timeout = TCP_TIMEOUT_INIT; 2365 return timeout; 2366 } 2367 tcp_rwnd_init_bpf(struct sock * sk)2368 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2369 { 2370 int rwnd; 2371 2372 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2373 2374 if (rwnd < 0) 2375 rwnd = 0; 2376 return rwnd; 2377 } 2378 tcp_bpf_ca_needs_ecn(struct sock * sk)2379 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2380 { 2381 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2382 } 2383 tcp_bpf_rtt(struct sock * sk)2384 static inline void tcp_bpf_rtt(struct sock *sk) 2385 { 2386 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2387 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2388 } 2389 2390 #if IS_ENABLED(CONFIG_SMC) 2391 extern struct static_key_false tcp_have_smc; 2392 #endif 2393 2394 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2395 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2396 void (*cad)(struct sock *sk, u32 ack_seq)); 2397 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2398 void clean_acked_data_flush(void); 2399 #endif 2400 2401 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); tcp_add_tx_delay(struct sk_buff * skb,const struct tcp_sock * tp)2402 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2403 const struct tcp_sock *tp) 2404 { 2405 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2406 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2407 } 2408 2409 /* Compute Earliest Departure Time for some control packets 2410 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2411 */ tcp_transmit_time(const struct sock * sk)2412 static inline u64 tcp_transmit_time(const struct sock *sk) 2413 { 2414 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2415 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2416 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2417 2418 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2419 } 2420 return 0; 2421 } 2422 2423 #endif /* _TCP_H */ 2424