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