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