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 extern struct percpu_counter tcp_orphan_count; 52 void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 55 #define MAX_TCP_OPTION_SPACE 40 56 #define TCP_MIN_SND_MSS 48 57 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 58 59 /* 60 * Never offer a window over 32767 without using window scaling. Some 61 * poor stacks do signed 16bit maths! 62 */ 63 #define MAX_TCP_WINDOW 32767U 64 65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 66 #define TCP_MIN_MSS 88U 67 68 /* The initial MTU to use for probing */ 69 #define TCP_BASE_MSS 1024 70 71 /* probing interval, default to 10 minutes as per RFC4821 */ 72 #define TCP_PROBE_INTERVAL 600 73 74 /* Specify interval when tcp mtu probing will stop */ 75 #define TCP_PROBE_THRESHOLD 8 76 77 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 78 #define TCP_FASTRETRANS_THRESH 3 79 80 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 81 #define TCP_MAX_QUICKACKS 16U 82 83 /* Maximal number of window scale according to RFC1323 */ 84 #define TCP_MAX_WSCALE 14U 85 86 /* urg_data states */ 87 #define TCP_URG_VALID 0x0100 88 #define TCP_URG_NOTYET 0x0200 89 #define TCP_URG_READ 0x0400 90 91 #define TCP_RETR1 3 /* 92 * This is how many retries it does before it 93 * tries to figure out if the gateway is 94 * down. Minimal RFC value is 3; it corresponds 95 * to ~3sec-8min depending on RTO. 96 */ 97 98 #define TCP_RETR2 15 /* 99 * This should take at least 100 * 90 minutes to time out. 101 * RFC1122 says that the limit is 100 sec. 102 * 15 is ~13-30min depending on RTO. 103 */ 104 105 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 106 * when active opening a connection. 107 * RFC1122 says the minimum retry MUST 108 * be at least 180secs. Nevertheless 109 * this value is corresponding to 110 * 63secs of retransmission with the 111 * current initial RTO. 112 */ 113 114 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 115 * when passive opening a connection. 116 * This is corresponding to 31secs of 117 * retransmission with the current 118 * initial RTO. 119 */ 120 121 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 122 * state, about 60 seconds */ 123 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 124 /* BSD style FIN_WAIT2 deadlock breaker. 125 * It used to be 3min, new value is 60sec, 126 * to combine FIN-WAIT-2 timeout with 127 * TIME-WAIT timer. 128 */ 129 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 130 131 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 132 #if HZ >= 100 133 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 134 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 135 #else 136 #define TCP_DELACK_MIN 4U 137 #define TCP_ATO_MIN 4U 138 #endif 139 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 140 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 141 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 142 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 143 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 144 * used as a fallback RTO for the 145 * initial data transmission if no 146 * valid RTT sample has been acquired, 147 * most likely due to retrans in 3WHS. 148 */ 149 150 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 151 * for local resources. 152 */ 153 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 154 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 155 #define TCP_KEEPALIVE_INTVL (75*HZ) 156 157 #define MAX_TCP_KEEPIDLE 32767 158 #define MAX_TCP_KEEPINTVL 32767 159 #define MAX_TCP_KEEPCNT 127 160 #define MAX_TCP_SYNCNT 127 161 162 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 163 164 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 165 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 166 * after this time. It should be equal 167 * (or greater than) TCP_TIMEWAIT_LEN 168 * to provide reliability equal to one 169 * provided by timewait state. 170 */ 171 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 172 * timestamps. It must be less than 173 * minimal timewait lifetime. 174 */ 175 /* 176 * TCP option 177 */ 178 179 #define TCPOPT_NOP 1 /* Padding */ 180 #define TCPOPT_EOL 0 /* End of options */ 181 #define TCPOPT_MSS 2 /* Segment size negotiating */ 182 #define TCPOPT_WINDOW 3 /* Window scaling */ 183 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 184 #define TCPOPT_SACK 5 /* SACK Block */ 185 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 186 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 187 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 189 #define TCPOPT_EXP 254 /* Experimental */ 190 /* Magic number to be after the option value for sharing TCP 191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 192 */ 193 #define TCPOPT_FASTOPEN_MAGIC 0xF989 194 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 195 196 /* 197 * TCP option lengths 198 */ 199 200 #define TCPOLEN_MSS 4 201 #define TCPOLEN_WINDOW 3 202 #define TCPOLEN_SACK_PERM 2 203 #define TCPOLEN_TIMESTAMP 10 204 #define TCPOLEN_MD5SIG 18 205 #define TCPOLEN_FASTOPEN_BASE 2 206 #define TCPOLEN_EXP_FASTOPEN_BASE 4 207 #define TCPOLEN_EXP_SMC_BASE 6 208 209 /* But this is what stacks really send out. */ 210 #define TCPOLEN_TSTAMP_ALIGNED 12 211 #define TCPOLEN_WSCALE_ALIGNED 4 212 #define TCPOLEN_SACKPERM_ALIGNED 4 213 #define TCPOLEN_SACK_BASE 2 214 #define TCPOLEN_SACK_BASE_ALIGNED 4 215 #define TCPOLEN_SACK_PERBLOCK 8 216 #define TCPOLEN_MD5SIG_ALIGNED 20 217 #define TCPOLEN_MSS_ALIGNED 4 218 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 219 220 /* Flags in tp->nonagle */ 221 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 222 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 223 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 224 225 /* TCP thin-stream limits */ 226 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 227 228 /* TCP initial congestion window as per rfc6928 */ 229 #define TCP_INIT_CWND 10 230 231 /* Bit Flags for sysctl_tcp_fastopen */ 232 #define TFO_CLIENT_ENABLE 1 233 #define TFO_SERVER_ENABLE 2 234 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 235 236 /* Accept SYN data w/o any cookie option */ 237 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 238 239 /* Force enable TFO on all listeners, i.e., not requiring the 240 * TCP_FASTOPEN socket option. 241 */ 242 #define TFO_SERVER_WO_SOCKOPT1 0x400 243 244 245 /* sysctl variables for tcp */ 246 extern int sysctl_tcp_max_orphans; 247 extern long sysctl_tcp_mem[3]; 248 249 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 250 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 251 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 252 253 extern atomic_long_t tcp_memory_allocated; 254 extern struct percpu_counter tcp_sockets_allocated; 255 extern unsigned long tcp_memory_pressure; 256 257 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 258 static inline bool tcp_under_memory_pressure(const struct sock *sk) 259 { 260 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 261 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 262 return true; 263 264 return READ_ONCE(tcp_memory_pressure); 265 } 266 /* 267 * The next routines deal with comparing 32 bit unsigned ints 268 * and worry about wraparound (automatic with unsigned arithmetic). 269 */ 270 271 static inline bool before(__u32 seq1, __u32 seq2) 272 { 273 return (__s32)(seq1-seq2) < 0; 274 } 275 #define after(seq2, seq1) before(seq1, seq2) 276 277 /* is s2<=s1<=s3 ? */ 278 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 279 { 280 return seq3 - seq2 >= seq1 - seq2; 281 } 282 283 static inline bool tcp_out_of_memory(struct sock *sk) 284 { 285 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 286 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 287 return true; 288 return false; 289 } 290 291 void sk_forced_mem_schedule(struct sock *sk, int size); 292 293 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 294 { 295 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 296 int orphans = percpu_counter_read_positive(ocp); 297 298 if (orphans << shift > sysctl_tcp_max_orphans) { 299 orphans = percpu_counter_sum_positive(ocp); 300 if (orphans << shift > sysctl_tcp_max_orphans) 301 return true; 302 } 303 return false; 304 } 305 306 bool tcp_check_oom(struct sock *sk, int shift); 307 308 309 extern struct proto tcp_prot; 310 311 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 312 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 315 316 void tcp_tasklet_init(void); 317 318 int tcp_v4_err(struct sk_buff *skb, u32); 319 320 void tcp_shutdown(struct sock *sk, int how); 321 322 int tcp_v4_early_demux(struct sk_buff *skb); 323 int tcp_v4_rcv(struct sk_buff *skb); 324 325 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 326 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 327 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 328 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 329 int flags); 330 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 331 size_t size, int flags); 332 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 333 size_t size, int flags); 334 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 335 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 336 int size_goal); 337 void tcp_release_cb(struct sock *sk); 338 void tcp_wfree(struct sk_buff *skb); 339 void tcp_write_timer_handler(struct sock *sk); 340 void tcp_delack_timer_handler(struct sock *sk); 341 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 342 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 344 void tcp_rcv_space_adjust(struct sock *sk); 345 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 346 void tcp_twsk_destructor(struct sock *sk); 347 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 348 struct pipe_inode_info *pipe, size_t len, 349 unsigned int flags); 350 351 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 352 static inline void tcp_dec_quickack_mode(struct sock *sk, 353 const unsigned int pkts) 354 { 355 struct inet_connection_sock *icsk = inet_csk(sk); 356 357 if (icsk->icsk_ack.quick) { 358 if (pkts >= icsk->icsk_ack.quick) { 359 icsk->icsk_ack.quick = 0; 360 /* Leaving quickack mode we deflate ATO. */ 361 icsk->icsk_ack.ato = TCP_ATO_MIN; 362 } else 363 icsk->icsk_ack.quick -= pkts; 364 } 365 } 366 367 #define TCP_ECN_OK 1 368 #define TCP_ECN_QUEUE_CWR 2 369 #define TCP_ECN_DEMAND_CWR 4 370 #define TCP_ECN_SEEN 8 371 372 enum tcp_tw_status { 373 TCP_TW_SUCCESS = 0, 374 TCP_TW_RST = 1, 375 TCP_TW_ACK = 2, 376 TCP_TW_SYN = 3 377 }; 378 379 380 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 381 struct sk_buff *skb, 382 const struct tcphdr *th); 383 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 384 struct request_sock *req, bool fastopen, 385 bool *lost_race); 386 int tcp_child_process(struct sock *parent, struct sock *child, 387 struct sk_buff *skb); 388 void tcp_enter_loss(struct sock *sk); 389 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 390 void tcp_clear_retrans(struct tcp_sock *tp); 391 void tcp_update_metrics(struct sock *sk); 392 void tcp_init_metrics(struct sock *sk); 393 void tcp_metrics_init(void); 394 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 395 void tcp_close(struct sock *sk, long timeout); 396 void tcp_init_sock(struct sock *sk); 397 void tcp_init_transfer(struct sock *sk, int bpf_op); 398 __poll_t tcp_poll(struct file *file, struct socket *sock, 399 struct poll_table_struct *wait); 400 int tcp_getsockopt(struct sock *sk, int level, int optname, 401 char __user *optval, int __user *optlen); 402 int tcp_setsockopt(struct sock *sk, int level, int optname, 403 char __user *optval, unsigned int optlen); 404 void tcp_set_keepalive(struct sock *sk, int val); 405 void tcp_syn_ack_timeout(const struct request_sock *req); 406 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 407 int flags, int *addr_len); 408 int tcp_set_rcvlowat(struct sock *sk, int val); 409 void tcp_data_ready(struct sock *sk); 410 #ifdef CONFIG_MMU 411 int tcp_mmap(struct file *file, struct socket *sock, 412 struct vm_area_struct *vma); 413 #endif 414 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 415 struct tcp_options_received *opt_rx, 416 int estab, struct tcp_fastopen_cookie *foc); 417 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 418 419 /* 420 * BPF SKB-less helpers 421 */ 422 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 423 struct tcphdr *th, u32 *cookie); 424 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 425 struct tcphdr *th, u32 *cookie); 426 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 427 const struct tcp_request_sock_ops *af_ops, 428 struct sock *sk, struct tcphdr *th); 429 /* 430 * TCP v4 functions exported for the inet6 API 431 */ 432 433 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 434 void tcp_v4_mtu_reduced(struct sock *sk); 435 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 436 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 437 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 438 struct sock *tcp_create_openreq_child(const struct sock *sk, 439 struct request_sock *req, 440 struct sk_buff *skb); 441 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 442 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 443 struct request_sock *req, 444 struct dst_entry *dst, 445 struct request_sock *req_unhash, 446 bool *own_req); 447 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 448 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 449 int tcp_connect(struct sock *sk); 450 enum tcp_synack_type { 451 TCP_SYNACK_NORMAL, 452 TCP_SYNACK_FASTOPEN, 453 TCP_SYNACK_COOKIE, 454 }; 455 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 456 struct request_sock *req, 457 struct tcp_fastopen_cookie *foc, 458 enum tcp_synack_type synack_type); 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 #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_push_pending_frames(struct sock *sk, unsigned int cur_mss, 572 int nonagle); 573 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 574 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 575 void tcp_retransmit_timer(struct sock *sk); 576 void tcp_xmit_retransmit_queue(struct sock *); 577 void tcp_simple_retransmit(struct sock *); 578 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 579 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 580 enum tcp_queue { 581 TCP_FRAG_IN_WRITE_QUEUE, 582 TCP_FRAG_IN_RTX_QUEUE, 583 }; 584 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 585 struct sk_buff *skb, u32 len, 586 unsigned int mss_now, gfp_t gfp); 587 588 void tcp_send_probe0(struct sock *); 589 void tcp_send_partial(struct sock *); 590 int tcp_write_wakeup(struct sock *, int mib); 591 void tcp_send_fin(struct sock *sk); 592 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 593 int tcp_send_synack(struct sock *); 594 void tcp_push_one(struct sock *, unsigned int mss_now); 595 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 596 void tcp_send_ack(struct sock *sk); 597 void tcp_send_delayed_ack(struct sock *sk); 598 void tcp_send_loss_probe(struct sock *sk); 599 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 600 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 601 const struct sk_buff *next_skb); 602 603 /* tcp_input.c */ 604 void tcp_rearm_rto(struct sock *sk); 605 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 606 void tcp_reset(struct sock *sk); 607 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 608 void tcp_fin(struct sock *sk); 609 610 /* tcp_timer.c */ 611 void tcp_init_xmit_timers(struct sock *); 612 static inline void tcp_clear_xmit_timers(struct sock *sk) 613 { 614 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 615 __sock_put(sk); 616 617 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 618 __sock_put(sk); 619 620 inet_csk_clear_xmit_timers(sk); 621 } 622 623 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 624 unsigned int tcp_current_mss(struct sock *sk); 625 626 /* Bound MSS / TSO packet size with the half of the window */ 627 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 628 { 629 int cutoff; 630 631 /* When peer uses tiny windows, there is no use in packetizing 632 * to sub-MSS pieces for the sake of SWS or making sure there 633 * are enough packets in the pipe for fast recovery. 634 * 635 * On the other hand, for extremely large MSS devices, handling 636 * smaller than MSS windows in this way does make sense. 637 */ 638 if (tp->max_window > TCP_MSS_DEFAULT) 639 cutoff = (tp->max_window >> 1); 640 else 641 cutoff = tp->max_window; 642 643 if (cutoff && pktsize > cutoff) 644 return max_t(int, cutoff, 68U - tp->tcp_header_len); 645 else 646 return pktsize; 647 } 648 649 /* tcp.c */ 650 void tcp_get_info(struct sock *, struct tcp_info *); 651 652 /* Read 'sendfile()'-style from a TCP socket */ 653 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 654 sk_read_actor_t recv_actor); 655 656 void tcp_initialize_rcv_mss(struct sock *sk); 657 658 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 659 int tcp_mss_to_mtu(struct sock *sk, int mss); 660 void tcp_mtup_init(struct sock *sk); 661 662 static inline void tcp_bound_rto(const struct sock *sk) 663 { 664 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 665 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 666 } 667 668 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 669 { 670 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 671 } 672 673 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 674 { 675 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 676 ntohl(TCP_FLAG_ACK) | 677 snd_wnd); 678 } 679 680 static inline void tcp_fast_path_on(struct tcp_sock *tp) 681 { 682 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 683 } 684 685 static inline void tcp_fast_path_check(struct sock *sk) 686 { 687 struct tcp_sock *tp = tcp_sk(sk); 688 689 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 690 tp->rcv_wnd && 691 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 692 !tp->urg_data) 693 tcp_fast_path_on(tp); 694 } 695 696 /* Compute the actual rto_min value */ 697 static inline u32 tcp_rto_min(struct sock *sk) 698 { 699 const struct dst_entry *dst = __sk_dst_get(sk); 700 u32 rto_min = TCP_RTO_MIN; 701 702 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 703 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 704 return rto_min; 705 } 706 707 static inline u32 tcp_rto_min_us(struct sock *sk) 708 { 709 return jiffies_to_usecs(tcp_rto_min(sk)); 710 } 711 712 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 713 { 714 return dst_metric_locked(dst, RTAX_CC_ALGO); 715 } 716 717 /* Minimum RTT in usec. ~0 means not available. */ 718 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 719 { 720 return minmax_get(&tp->rtt_min); 721 } 722 723 /* Compute the actual receive window we are currently advertising. 724 * Rcv_nxt can be after the window if our peer push more data 725 * than the offered window. 726 */ 727 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 728 { 729 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 730 731 if (win < 0) 732 win = 0; 733 return (u32) win; 734 } 735 736 /* Choose a new window, without checks for shrinking, and without 737 * scaling applied to the result. The caller does these things 738 * if necessary. This is a "raw" window selection. 739 */ 740 u32 __tcp_select_window(struct sock *sk); 741 742 void tcp_send_window_probe(struct sock *sk); 743 744 /* TCP uses 32bit jiffies to save some space. 745 * Note that this is different from tcp_time_stamp, which 746 * historically has been the same until linux-4.13. 747 */ 748 #define tcp_jiffies32 ((u32)jiffies) 749 750 /* 751 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 752 * It is no longer tied to jiffies, but to 1 ms clock. 753 * Note: double check if you want to use tcp_jiffies32 instead of this. 754 */ 755 #define TCP_TS_HZ 1000 756 757 static inline u64 tcp_clock_ns(void) 758 { 759 return ktime_get_ns(); 760 } 761 762 static inline u64 tcp_clock_us(void) 763 { 764 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 765 } 766 767 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 768 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 769 { 770 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 771 } 772 773 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 774 static inline u32 tcp_ns_to_ts(u64 ns) 775 { 776 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 777 } 778 779 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 780 static inline u32 tcp_time_stamp_raw(void) 781 { 782 return tcp_ns_to_ts(tcp_clock_ns()); 783 } 784 785 void tcp_mstamp_refresh(struct tcp_sock *tp); 786 787 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 788 { 789 return max_t(s64, t1 - t0, 0); 790 } 791 792 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 793 { 794 return tcp_ns_to_ts(skb->skb_mstamp_ns); 795 } 796 797 /* provide the departure time in us unit */ 798 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 799 { 800 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 801 } 802 803 804 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 805 806 #define TCPHDR_FIN 0x01 807 #define TCPHDR_SYN 0x02 808 #define TCPHDR_RST 0x04 809 #define TCPHDR_PSH 0x08 810 #define TCPHDR_ACK 0x10 811 #define TCPHDR_URG 0x20 812 #define TCPHDR_ECE 0x40 813 #define TCPHDR_CWR 0x80 814 815 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 816 817 /* This is what the send packet queuing engine uses to pass 818 * TCP per-packet control information to the transmission code. 819 * We also store the host-order sequence numbers in here too. 820 * This is 44 bytes if IPV6 is enabled. 821 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 822 */ 823 struct tcp_skb_cb { 824 __u32 seq; /* Starting sequence number */ 825 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 826 union { 827 /* Note : tcp_tw_isn is used in input path only 828 * (isn chosen by tcp_timewait_state_process()) 829 * 830 * tcp_gso_segs/size are used in write queue only, 831 * cf tcp_skb_pcount()/tcp_skb_mss() 832 */ 833 __u32 tcp_tw_isn; 834 struct { 835 u16 tcp_gso_segs; 836 u16 tcp_gso_size; 837 }; 838 }; 839 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 840 841 __u8 sacked; /* State flags for SACK. */ 842 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 843 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 844 #define TCPCB_LOST 0x04 /* SKB is lost */ 845 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 846 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 847 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 848 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 849 TCPCB_REPAIRED) 850 851 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 852 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 853 eor:1, /* Is skb MSG_EOR marked? */ 854 has_rxtstamp:1, /* SKB has a RX timestamp */ 855 unused:5; 856 __u32 ack_seq; /* Sequence number ACK'd */ 857 union { 858 struct { 859 /* There is space for up to 24 bytes */ 860 __u32 in_flight:30,/* Bytes in flight at transmit */ 861 is_app_limited:1, /* cwnd not fully used? */ 862 unused:1; 863 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 864 __u32 delivered; 865 /* start of send pipeline phase */ 866 u64 first_tx_mstamp; 867 /* when we reached the "delivered" count */ 868 u64 delivered_mstamp; 869 } tx; /* only used for outgoing skbs */ 870 union { 871 struct inet_skb_parm h4; 872 #if IS_ENABLED(CONFIG_IPV6) 873 struct inet6_skb_parm h6; 874 #endif 875 } header; /* For incoming skbs */ 876 struct { 877 __u32 flags; 878 struct sock *sk_redir; 879 void *data_end; 880 } bpf; 881 }; 882 }; 883 884 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 885 886 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) 887 { 888 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); 889 } 890 891 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) 892 { 893 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; 894 } 895 896 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) 897 { 898 return TCP_SKB_CB(skb)->bpf.sk_redir; 899 } 900 901 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) 902 { 903 TCP_SKB_CB(skb)->bpf.sk_redir = NULL; 904 } 905 906 extern const struct inet_connection_sock_af_ops ipv4_specific; 907 908 #if IS_ENABLED(CONFIG_IPV6) 909 /* This is the variant of inet6_iif() that must be used by TCP, 910 * as TCP moves IP6CB into a different location in skb->cb[] 911 */ 912 static inline int tcp_v6_iif(const struct sk_buff *skb) 913 { 914 return TCP_SKB_CB(skb)->header.h6.iif; 915 } 916 917 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 918 { 919 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 920 921 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 922 } 923 924 /* TCP_SKB_CB reference means this can not be used from early demux */ 925 static inline int tcp_v6_sdif(const struct sk_buff *skb) 926 { 927 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 928 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 929 return TCP_SKB_CB(skb)->header.h6.iif; 930 #endif 931 return 0; 932 } 933 934 extern const struct inet_connection_sock_af_ops ipv6_specific; 935 936 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 937 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 938 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb)); 939 940 #endif 941 942 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 943 { 944 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 945 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 946 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 947 return true; 948 #endif 949 return false; 950 } 951 952 /* TCP_SKB_CB reference means this can not be used from early demux */ 953 static inline int tcp_v4_sdif(struct sk_buff *skb) 954 { 955 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 956 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 957 return TCP_SKB_CB(skb)->header.h4.iif; 958 #endif 959 return 0; 960 } 961 962 /* Due to TSO, an SKB can be composed of multiple actual 963 * packets. To keep these tracked properly, we use this. 964 */ 965 static inline int tcp_skb_pcount(const struct sk_buff *skb) 966 { 967 return TCP_SKB_CB(skb)->tcp_gso_segs; 968 } 969 970 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 971 { 972 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 973 } 974 975 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 976 { 977 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 978 } 979 980 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 981 static inline int tcp_skb_mss(const struct sk_buff *skb) 982 { 983 return TCP_SKB_CB(skb)->tcp_gso_size; 984 } 985 986 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 987 { 988 return likely(!TCP_SKB_CB(skb)->eor); 989 } 990 991 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 992 const struct sk_buff *from) 993 { 994 return likely(tcp_skb_can_collapse_to(to) && 995 mptcp_skb_can_collapse(to, from)); 996 } 997 998 /* Events passed to congestion control interface */ 999 enum tcp_ca_event { 1000 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1001 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1002 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1003 CA_EVENT_LOSS, /* loss timeout */ 1004 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1005 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1006 }; 1007 1008 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1009 enum tcp_ca_ack_event_flags { 1010 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1011 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1012 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1013 }; 1014 1015 /* 1016 * Interface for adding new TCP congestion control handlers 1017 */ 1018 #define TCP_CA_NAME_MAX 16 1019 #define TCP_CA_MAX 128 1020 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1021 1022 #define TCP_CA_UNSPEC 0 1023 1024 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1025 #define TCP_CONG_NON_RESTRICTED 0x1 1026 /* Requires ECN/ECT set on all packets */ 1027 #define TCP_CONG_NEEDS_ECN 0x2 1028 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1029 1030 union tcp_cc_info; 1031 1032 struct ack_sample { 1033 u32 pkts_acked; 1034 s32 rtt_us; 1035 u32 in_flight; 1036 }; 1037 1038 /* A rate sample measures the number of (original/retransmitted) data 1039 * packets delivered "delivered" over an interval of time "interval_us". 1040 * The tcp_rate.c code fills in the rate sample, and congestion 1041 * control modules that define a cong_control function to run at the end 1042 * of ACK processing can optionally chose to consult this sample when 1043 * setting cwnd and pacing rate. 1044 * A sample is invalid if "delivered" or "interval_us" is negative. 1045 */ 1046 struct rate_sample { 1047 u64 prior_mstamp; /* starting timestamp for interval */ 1048 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1049 s32 delivered; /* number of packets delivered over interval */ 1050 long interval_us; /* time for tp->delivered to incr "delivered" */ 1051 u32 snd_interval_us; /* snd interval for delivered packets */ 1052 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1053 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1054 int losses; /* number of packets marked lost upon ACK */ 1055 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1056 u32 prior_in_flight; /* in flight before this ACK */ 1057 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1058 bool is_retrans; /* is sample from retransmission? */ 1059 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1060 }; 1061 1062 struct tcp_congestion_ops { 1063 struct list_head list; 1064 u32 key; 1065 u32 flags; 1066 1067 /* initialize private data (optional) */ 1068 void (*init)(struct sock *sk); 1069 /* cleanup private data (optional) */ 1070 void (*release)(struct sock *sk); 1071 1072 /* return slow start threshold (required) */ 1073 u32 (*ssthresh)(struct sock *sk); 1074 /* do new cwnd calculation (required) */ 1075 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1076 /* call before changing ca_state (optional) */ 1077 void (*set_state)(struct sock *sk, u8 new_state); 1078 /* call when cwnd event occurs (optional) */ 1079 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1080 /* call when ack arrives (optional) */ 1081 void (*in_ack_event)(struct sock *sk, u32 flags); 1082 /* new value of cwnd after loss (required) */ 1083 u32 (*undo_cwnd)(struct sock *sk); 1084 /* hook for packet ack accounting (optional) */ 1085 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1086 /* override sysctl_tcp_min_tso_segs */ 1087 u32 (*min_tso_segs)(struct sock *sk); 1088 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1089 u32 (*sndbuf_expand)(struct sock *sk); 1090 /* call when packets are delivered to update cwnd and pacing rate, 1091 * after all the ca_state processing. (optional) 1092 */ 1093 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1094 /* get info for inet_diag (optional) */ 1095 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1096 union tcp_cc_info *info); 1097 1098 char name[TCP_CA_NAME_MAX]; 1099 struct module *owner; 1100 }; 1101 1102 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1103 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1104 1105 void tcp_assign_congestion_control(struct sock *sk); 1106 void tcp_init_congestion_control(struct sock *sk); 1107 void tcp_cleanup_congestion_control(struct sock *sk); 1108 int tcp_set_default_congestion_control(struct net *net, const char *name); 1109 void tcp_get_default_congestion_control(struct net *net, char *name); 1110 void tcp_get_available_congestion_control(char *buf, size_t len); 1111 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1112 int tcp_set_allowed_congestion_control(char *allowed); 1113 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1114 bool reinit, bool cap_net_admin); 1115 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1116 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1117 1118 u32 tcp_reno_ssthresh(struct sock *sk); 1119 u32 tcp_reno_undo_cwnd(struct sock *sk); 1120 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1121 extern struct tcp_congestion_ops tcp_reno; 1122 1123 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1124 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1125 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1126 #ifdef CONFIG_INET 1127 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1128 #else 1129 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1130 { 1131 return NULL; 1132 } 1133 #endif 1134 1135 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1136 { 1137 const struct inet_connection_sock *icsk = inet_csk(sk); 1138 1139 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1140 } 1141 1142 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1143 { 1144 struct inet_connection_sock *icsk = inet_csk(sk); 1145 1146 if (icsk->icsk_ca_ops->set_state) 1147 icsk->icsk_ca_ops->set_state(sk, ca_state); 1148 icsk->icsk_ca_state = ca_state; 1149 } 1150 1151 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1152 { 1153 const struct inet_connection_sock *icsk = inet_csk(sk); 1154 1155 if (icsk->icsk_ca_ops->cwnd_event) 1156 icsk->icsk_ca_ops->cwnd_event(sk, event); 1157 } 1158 1159 /* From tcp_rate.c */ 1160 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1161 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1162 struct rate_sample *rs); 1163 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1164 bool is_sack_reneg, struct rate_sample *rs); 1165 void tcp_rate_check_app_limited(struct sock *sk); 1166 1167 /* These functions determine how the current flow behaves in respect of SACK 1168 * handling. SACK is negotiated with the peer, and therefore it can vary 1169 * between different flows. 1170 * 1171 * tcp_is_sack - SACK enabled 1172 * tcp_is_reno - No SACK 1173 */ 1174 static inline int tcp_is_sack(const struct tcp_sock *tp) 1175 { 1176 return likely(tp->rx_opt.sack_ok); 1177 } 1178 1179 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1180 { 1181 return !tcp_is_sack(tp); 1182 } 1183 1184 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1185 { 1186 return tp->sacked_out + tp->lost_out; 1187 } 1188 1189 /* This determines how many packets are "in the network" to the best 1190 * of our knowledge. In many cases it is conservative, but where 1191 * detailed information is available from the receiver (via SACK 1192 * blocks etc.) we can make more aggressive calculations. 1193 * 1194 * Use this for decisions involving congestion control, use just 1195 * tp->packets_out to determine if the send queue is empty or not. 1196 * 1197 * Read this equation as: 1198 * 1199 * "Packets sent once on transmission queue" MINUS 1200 * "Packets left network, but not honestly ACKed yet" PLUS 1201 * "Packets fast retransmitted" 1202 */ 1203 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1204 { 1205 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1206 } 1207 1208 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1209 1210 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1211 { 1212 return tp->snd_cwnd < tp->snd_ssthresh; 1213 } 1214 1215 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1216 { 1217 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1218 } 1219 1220 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1221 { 1222 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1223 (1 << inet_csk(sk)->icsk_ca_state); 1224 } 1225 1226 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1227 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1228 * ssthresh. 1229 */ 1230 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1231 { 1232 const struct tcp_sock *tp = tcp_sk(sk); 1233 1234 if (tcp_in_cwnd_reduction(sk)) 1235 return tp->snd_ssthresh; 1236 else 1237 return max(tp->snd_ssthresh, 1238 ((tp->snd_cwnd >> 1) + 1239 (tp->snd_cwnd >> 2))); 1240 } 1241 1242 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1243 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1244 1245 void tcp_enter_cwr(struct sock *sk); 1246 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1247 1248 /* The maximum number of MSS of available cwnd for which TSO defers 1249 * sending if not using sysctl_tcp_tso_win_divisor. 1250 */ 1251 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1252 { 1253 return 3; 1254 } 1255 1256 /* Returns end sequence number of the receiver's advertised window */ 1257 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1258 { 1259 return tp->snd_una + tp->snd_wnd; 1260 } 1261 1262 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1263 * flexible approach. The RFC suggests cwnd should not be raised unless 1264 * it was fully used previously. And that's exactly what we do in 1265 * congestion avoidance mode. But in slow start we allow cwnd to grow 1266 * as long as the application has used half the cwnd. 1267 * Example : 1268 * cwnd is 10 (IW10), but application sends 9 frames. 1269 * We allow cwnd to reach 18 when all frames are ACKed. 1270 * This check is safe because it's as aggressive as slow start which already 1271 * risks 100% overshoot. The advantage is that we discourage application to 1272 * either send more filler packets or data to artificially blow up the cwnd 1273 * usage, and allow application-limited process to probe bw more aggressively. 1274 */ 1275 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1276 { 1277 const struct tcp_sock *tp = tcp_sk(sk); 1278 1279 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1280 if (tcp_in_slow_start(tp)) 1281 return tp->snd_cwnd < 2 * tp->max_packets_out; 1282 1283 return tp->is_cwnd_limited; 1284 } 1285 1286 /* BBR congestion control needs pacing. 1287 * Same remark for SO_MAX_PACING_RATE. 1288 * sch_fq packet scheduler is efficiently handling pacing, 1289 * but is not always installed/used. 1290 * Return true if TCP stack should pace packets itself. 1291 */ 1292 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1293 { 1294 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1295 } 1296 1297 /* Estimates in how many jiffies next packet for this flow can be sent. 1298 * Scheduling a retransmit timer too early would be silly. 1299 */ 1300 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1301 { 1302 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1303 1304 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1305 } 1306 1307 static inline void tcp_reset_xmit_timer(struct sock *sk, 1308 const int what, 1309 unsigned long when, 1310 const unsigned long max_when) 1311 { 1312 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1313 max_when); 1314 } 1315 1316 /* Something is really bad, we could not queue an additional packet, 1317 * because qdisc is full or receiver sent a 0 window, or we are paced. 1318 * We do not want to add fuel to the fire, or abort too early, 1319 * so make sure the timer we arm now is at least 200ms in the future, 1320 * regardless of current icsk_rto value (as it could be ~2ms) 1321 */ 1322 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1323 { 1324 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1325 } 1326 1327 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1328 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1329 unsigned long max_when) 1330 { 1331 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1332 1333 return (unsigned long)min_t(u64, when, max_when); 1334 } 1335 1336 static inline void tcp_check_probe_timer(struct sock *sk) 1337 { 1338 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1339 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1340 tcp_probe0_base(sk), TCP_RTO_MAX); 1341 } 1342 1343 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1344 { 1345 tp->snd_wl1 = seq; 1346 } 1347 1348 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1349 { 1350 tp->snd_wl1 = seq; 1351 } 1352 1353 /* 1354 * Calculate(/check) TCP checksum 1355 */ 1356 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1357 __be32 daddr, __wsum base) 1358 { 1359 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1360 } 1361 1362 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1363 { 1364 return !skb_csum_unnecessary(skb) && 1365 __skb_checksum_complete(skb); 1366 } 1367 1368 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1369 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1370 void tcp_set_state(struct sock *sk, int state); 1371 void tcp_done(struct sock *sk); 1372 int tcp_abort(struct sock *sk, int err); 1373 1374 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1375 { 1376 rx_opt->dsack = 0; 1377 rx_opt->num_sacks = 0; 1378 } 1379 1380 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1381 1382 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1383 { 1384 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1385 struct tcp_sock *tp = tcp_sk(sk); 1386 s32 delta; 1387 1388 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1389 ca_ops->cong_control) 1390 return; 1391 delta = tcp_jiffies32 - tp->lsndtime; 1392 if (delta > inet_csk(sk)->icsk_rto) 1393 tcp_cwnd_restart(sk, delta); 1394 } 1395 1396 /* Determine a window scaling and initial window to offer. */ 1397 void tcp_select_initial_window(const struct sock *sk, int __space, 1398 __u32 mss, __u32 *rcv_wnd, 1399 __u32 *window_clamp, int wscale_ok, 1400 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1401 1402 static inline int tcp_win_from_space(const struct sock *sk, int space) 1403 { 1404 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1405 1406 return tcp_adv_win_scale <= 0 ? 1407 (space>>(-tcp_adv_win_scale)) : 1408 space - (space>>tcp_adv_win_scale); 1409 } 1410 1411 /* Note: caller must be prepared to deal with negative returns */ 1412 static inline int tcp_space(const struct sock *sk) 1413 { 1414 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1415 READ_ONCE(sk->sk_backlog.len) - 1416 atomic_read(&sk->sk_rmem_alloc)); 1417 } 1418 1419 static inline int tcp_full_space(const struct sock *sk) 1420 { 1421 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1422 } 1423 1424 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1425 * If 87.5 % (7/8) of the space has been consumed, we want to override 1426 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1427 * len/truesize ratio. 1428 */ 1429 static inline bool tcp_rmem_pressure(const struct sock *sk) 1430 { 1431 int rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1432 int threshold = rcvbuf - (rcvbuf >> 3); 1433 1434 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1435 } 1436 1437 extern void tcp_openreq_init_rwin(struct request_sock *req, 1438 const struct sock *sk_listener, 1439 const struct dst_entry *dst); 1440 1441 void tcp_enter_memory_pressure(struct sock *sk); 1442 void tcp_leave_memory_pressure(struct sock *sk); 1443 1444 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1445 { 1446 struct net *net = sock_net((struct sock *)tp); 1447 1448 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1449 } 1450 1451 static inline int keepalive_time_when(const struct tcp_sock *tp) 1452 { 1453 struct net *net = sock_net((struct sock *)tp); 1454 1455 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1456 } 1457 1458 static inline int keepalive_probes(const struct tcp_sock *tp) 1459 { 1460 struct net *net = sock_net((struct sock *)tp); 1461 1462 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1463 } 1464 1465 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1466 { 1467 const struct inet_connection_sock *icsk = &tp->inet_conn; 1468 1469 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1470 tcp_jiffies32 - tp->rcv_tstamp); 1471 } 1472 1473 static inline int tcp_fin_time(const struct sock *sk) 1474 { 1475 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1476 const int rto = inet_csk(sk)->icsk_rto; 1477 1478 if (fin_timeout < (rto << 2) - (rto >> 1)) 1479 fin_timeout = (rto << 2) - (rto >> 1); 1480 1481 return fin_timeout; 1482 } 1483 1484 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1485 int paws_win) 1486 { 1487 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1488 return true; 1489 if (unlikely(!time_before32(ktime_get_seconds(), 1490 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1491 return true; 1492 /* 1493 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1494 * then following tcp messages have valid values. Ignore 0 value, 1495 * or else 'negative' tsval might forbid us to accept their packets. 1496 */ 1497 if (!rx_opt->ts_recent) 1498 return true; 1499 return false; 1500 } 1501 1502 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1503 int rst) 1504 { 1505 if (tcp_paws_check(rx_opt, 0)) 1506 return false; 1507 1508 /* RST segments are not recommended to carry timestamp, 1509 and, if they do, it is recommended to ignore PAWS because 1510 "their cleanup function should take precedence over timestamps." 1511 Certainly, it is mistake. It is necessary to understand the reasons 1512 of this constraint to relax it: if peer reboots, clock may go 1513 out-of-sync and half-open connections will not be reset. 1514 Actually, the problem would be not existing if all 1515 the implementations followed draft about maintaining clock 1516 via reboots. Linux-2.2 DOES NOT! 1517 1518 However, we can relax time bounds for RST segments to MSL. 1519 */ 1520 if (rst && !time_before32(ktime_get_seconds(), 1521 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1522 return false; 1523 return true; 1524 } 1525 1526 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1527 int mib_idx, u32 *last_oow_ack_time); 1528 1529 static inline void tcp_mib_init(struct net *net) 1530 { 1531 /* See RFC 2012 */ 1532 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1533 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1534 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1535 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1536 } 1537 1538 /* from STCP */ 1539 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1540 { 1541 tp->lost_skb_hint = NULL; 1542 } 1543 1544 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1545 { 1546 tcp_clear_retrans_hints_partial(tp); 1547 tp->retransmit_skb_hint = NULL; 1548 } 1549 1550 union tcp_md5_addr { 1551 struct in_addr a4; 1552 #if IS_ENABLED(CONFIG_IPV6) 1553 struct in6_addr a6; 1554 #endif 1555 }; 1556 1557 /* - key database */ 1558 struct tcp_md5sig_key { 1559 struct hlist_node node; 1560 u8 keylen; 1561 u8 family; /* AF_INET or AF_INET6 */ 1562 u8 prefixlen; 1563 union tcp_md5_addr addr; 1564 int l3index; /* set if key added with L3 scope */ 1565 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1566 struct rcu_head rcu; 1567 }; 1568 1569 /* - sock block */ 1570 struct tcp_md5sig_info { 1571 struct hlist_head head; 1572 struct rcu_head rcu; 1573 }; 1574 1575 /* - pseudo header */ 1576 struct tcp4_pseudohdr { 1577 __be32 saddr; 1578 __be32 daddr; 1579 __u8 pad; 1580 __u8 protocol; 1581 __be16 len; 1582 }; 1583 1584 struct tcp6_pseudohdr { 1585 struct in6_addr saddr; 1586 struct in6_addr daddr; 1587 __be32 len; 1588 __be32 protocol; /* including padding */ 1589 }; 1590 1591 union tcp_md5sum_block { 1592 struct tcp4_pseudohdr ip4; 1593 #if IS_ENABLED(CONFIG_IPV6) 1594 struct tcp6_pseudohdr ip6; 1595 #endif 1596 }; 1597 1598 /* - pool: digest algorithm, hash description and scratch buffer */ 1599 struct tcp_md5sig_pool { 1600 struct ahash_request *md5_req; 1601 void *scratch; 1602 }; 1603 1604 /* - functions */ 1605 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1606 const struct sock *sk, const struct sk_buff *skb); 1607 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1608 int family, u8 prefixlen, int l3index, 1609 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1610 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1611 int family, u8 prefixlen, int l3index); 1612 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1613 const struct sock *addr_sk); 1614 1615 #ifdef CONFIG_TCP_MD5SIG 1616 #include <linux/jump_label.h> 1617 extern struct static_key_false tcp_md5_needed; 1618 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1619 const union tcp_md5_addr *addr, 1620 int family); 1621 static inline struct tcp_md5sig_key * 1622 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1623 const union tcp_md5_addr *addr, int family) 1624 { 1625 if (!static_branch_unlikely(&tcp_md5_needed)) 1626 return NULL; 1627 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1628 } 1629 1630 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1631 #else 1632 static inline struct tcp_md5sig_key * 1633 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1634 const union tcp_md5_addr *addr, int family) 1635 { 1636 return NULL; 1637 } 1638 #define tcp_twsk_md5_key(twsk) NULL 1639 #endif 1640 1641 bool tcp_alloc_md5sig_pool(void); 1642 1643 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1644 static inline void tcp_put_md5sig_pool(void) 1645 { 1646 local_bh_enable(); 1647 } 1648 1649 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1650 unsigned int header_len); 1651 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1652 const struct tcp_md5sig_key *key); 1653 1654 /* From tcp_fastopen.c */ 1655 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1656 struct tcp_fastopen_cookie *cookie); 1657 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1658 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1659 u16 try_exp); 1660 struct tcp_fastopen_request { 1661 /* Fast Open cookie. Size 0 means a cookie request */ 1662 struct tcp_fastopen_cookie cookie; 1663 struct msghdr *data; /* data in MSG_FASTOPEN */ 1664 size_t size; 1665 int copied; /* queued in tcp_connect() */ 1666 struct ubuf_info *uarg; 1667 }; 1668 void tcp_free_fastopen_req(struct tcp_sock *tp); 1669 void tcp_fastopen_destroy_cipher(struct sock *sk); 1670 void tcp_fastopen_ctx_destroy(struct net *net); 1671 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1672 void *primary_key, void *backup_key); 1673 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1674 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1675 struct request_sock *req, 1676 struct tcp_fastopen_cookie *foc, 1677 const struct dst_entry *dst); 1678 void tcp_fastopen_init_key_once(struct net *net); 1679 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1680 struct tcp_fastopen_cookie *cookie); 1681 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1682 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1683 #define TCP_FASTOPEN_KEY_MAX 2 1684 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1685 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1686 1687 /* Fastopen key context */ 1688 struct tcp_fastopen_context { 1689 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1690 int num; 1691 struct rcu_head rcu; 1692 }; 1693 1694 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1695 void tcp_fastopen_active_disable(struct sock *sk); 1696 bool tcp_fastopen_active_should_disable(struct sock *sk); 1697 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1698 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1699 1700 /* Caller needs to wrap with rcu_read_(un)lock() */ 1701 static inline 1702 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1703 { 1704 struct tcp_fastopen_context *ctx; 1705 1706 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1707 if (!ctx) 1708 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1709 return ctx; 1710 } 1711 1712 static inline 1713 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1714 const struct tcp_fastopen_cookie *orig) 1715 { 1716 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1717 orig->len == foc->len && 1718 !memcmp(orig->val, foc->val, foc->len)) 1719 return true; 1720 return false; 1721 } 1722 1723 static inline 1724 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1725 { 1726 return ctx->num; 1727 } 1728 1729 /* Latencies incurred by various limits for a sender. They are 1730 * chronograph-like stats that are mutually exclusive. 1731 */ 1732 enum tcp_chrono { 1733 TCP_CHRONO_UNSPEC, 1734 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1735 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1736 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1737 __TCP_CHRONO_MAX, 1738 }; 1739 1740 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1741 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1742 1743 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1744 * the same memory storage than skb->destructor/_skb_refdst 1745 */ 1746 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1747 { 1748 skb->destructor = NULL; 1749 skb->_skb_refdst = 0UL; 1750 } 1751 1752 #define tcp_skb_tsorted_save(skb) { \ 1753 unsigned long _save = skb->_skb_refdst; \ 1754 skb->_skb_refdst = 0UL; 1755 1756 #define tcp_skb_tsorted_restore(skb) \ 1757 skb->_skb_refdst = _save; \ 1758 } 1759 1760 void tcp_write_queue_purge(struct sock *sk); 1761 1762 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1763 { 1764 return skb_rb_first(&sk->tcp_rtx_queue); 1765 } 1766 1767 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1768 { 1769 return skb_rb_last(&sk->tcp_rtx_queue); 1770 } 1771 1772 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1773 { 1774 return skb_peek(&sk->sk_write_queue); 1775 } 1776 1777 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1778 { 1779 return skb_peek_tail(&sk->sk_write_queue); 1780 } 1781 1782 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1783 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1784 1785 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1786 { 1787 return skb_peek(&sk->sk_write_queue); 1788 } 1789 1790 static inline bool tcp_skb_is_last(const struct sock *sk, 1791 const struct sk_buff *skb) 1792 { 1793 return skb_queue_is_last(&sk->sk_write_queue, skb); 1794 } 1795 1796 /** 1797 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1798 * @sk: socket 1799 * 1800 * Since the write queue can have a temporary empty skb in it, 1801 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1802 */ 1803 static inline bool tcp_write_queue_empty(const struct sock *sk) 1804 { 1805 const struct tcp_sock *tp = tcp_sk(sk); 1806 1807 return tp->write_seq == tp->snd_nxt; 1808 } 1809 1810 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1811 { 1812 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1813 } 1814 1815 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1816 { 1817 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1818 } 1819 1820 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1821 { 1822 __skb_queue_tail(&sk->sk_write_queue, skb); 1823 1824 /* Queue it, remembering where we must start sending. */ 1825 if (sk->sk_write_queue.next == skb) 1826 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1827 } 1828 1829 /* Insert new before skb on the write queue of sk. */ 1830 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1831 struct sk_buff *skb, 1832 struct sock *sk) 1833 { 1834 __skb_queue_before(&sk->sk_write_queue, skb, new); 1835 } 1836 1837 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1838 { 1839 tcp_skb_tsorted_anchor_cleanup(skb); 1840 __skb_unlink(skb, &sk->sk_write_queue); 1841 } 1842 1843 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1844 1845 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1846 { 1847 tcp_skb_tsorted_anchor_cleanup(skb); 1848 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1849 } 1850 1851 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1852 { 1853 list_del(&skb->tcp_tsorted_anchor); 1854 tcp_rtx_queue_unlink(skb, sk); 1855 sk_wmem_free_skb(sk, skb); 1856 } 1857 1858 static inline void tcp_push_pending_frames(struct sock *sk) 1859 { 1860 if (tcp_send_head(sk)) { 1861 struct tcp_sock *tp = tcp_sk(sk); 1862 1863 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1864 } 1865 } 1866 1867 /* Start sequence of the skb just after the highest skb with SACKed 1868 * bit, valid only if sacked_out > 0 or when the caller has ensured 1869 * validity by itself. 1870 */ 1871 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1872 { 1873 if (!tp->sacked_out) 1874 return tp->snd_una; 1875 1876 if (tp->highest_sack == NULL) 1877 return tp->snd_nxt; 1878 1879 return TCP_SKB_CB(tp->highest_sack)->seq; 1880 } 1881 1882 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1883 { 1884 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1885 } 1886 1887 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1888 { 1889 return tcp_sk(sk)->highest_sack; 1890 } 1891 1892 static inline void tcp_highest_sack_reset(struct sock *sk) 1893 { 1894 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1895 } 1896 1897 /* Called when old skb is about to be deleted and replaced by new skb */ 1898 static inline void tcp_highest_sack_replace(struct sock *sk, 1899 struct sk_buff *old, 1900 struct sk_buff *new) 1901 { 1902 if (old == tcp_highest_sack(sk)) 1903 tcp_sk(sk)->highest_sack = new; 1904 } 1905 1906 /* This helper checks if socket has IP_TRANSPARENT set */ 1907 static inline bool inet_sk_transparent(const struct sock *sk) 1908 { 1909 switch (sk->sk_state) { 1910 case TCP_TIME_WAIT: 1911 return inet_twsk(sk)->tw_transparent; 1912 case TCP_NEW_SYN_RECV: 1913 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1914 } 1915 return inet_sk(sk)->transparent; 1916 } 1917 1918 /* Determines whether this is a thin stream (which may suffer from 1919 * increased latency). Used to trigger latency-reducing mechanisms. 1920 */ 1921 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1922 { 1923 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1924 } 1925 1926 /* /proc */ 1927 enum tcp_seq_states { 1928 TCP_SEQ_STATE_LISTENING, 1929 TCP_SEQ_STATE_ESTABLISHED, 1930 }; 1931 1932 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1933 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1934 void tcp_seq_stop(struct seq_file *seq, void *v); 1935 1936 struct tcp_seq_afinfo { 1937 sa_family_t family; 1938 }; 1939 1940 struct tcp_iter_state { 1941 struct seq_net_private p; 1942 enum tcp_seq_states state; 1943 struct sock *syn_wait_sk; 1944 struct tcp_seq_afinfo *bpf_seq_afinfo; 1945 int bucket, offset, sbucket, num; 1946 loff_t last_pos; 1947 }; 1948 1949 extern struct request_sock_ops tcp_request_sock_ops; 1950 extern struct request_sock_ops tcp6_request_sock_ops; 1951 1952 void tcp_v4_destroy_sock(struct sock *sk); 1953 1954 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1955 netdev_features_t features); 1956 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 1957 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 1958 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 1959 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 1960 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 1961 int tcp_gro_complete(struct sk_buff *skb); 1962 1963 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1964 1965 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1966 { 1967 struct net *net = sock_net((struct sock *)tp); 1968 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1969 } 1970 1971 /* @wake is one when sk_stream_write_space() calls us. 1972 * This sends EPOLLOUT only if notsent_bytes is half the limit. 1973 * This mimics the strategy used in sock_def_write_space(). 1974 */ 1975 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) 1976 { 1977 const struct tcp_sock *tp = tcp_sk(sk); 1978 u32 notsent_bytes = READ_ONCE(tp->write_seq) - 1979 READ_ONCE(tp->snd_nxt); 1980 1981 return (notsent_bytes << wake) < tcp_notsent_lowat(tp); 1982 } 1983 1984 #ifdef CONFIG_PROC_FS 1985 int tcp4_proc_init(void); 1986 void tcp4_proc_exit(void); 1987 #endif 1988 1989 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1990 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1991 const struct tcp_request_sock_ops *af_ops, 1992 struct sock *sk, struct sk_buff *skb); 1993 1994 /* TCP af-specific functions */ 1995 struct tcp_sock_af_ops { 1996 #ifdef CONFIG_TCP_MD5SIG 1997 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1998 const struct sock *addr_sk); 1999 int (*calc_md5_hash)(char *location, 2000 const struct tcp_md5sig_key *md5, 2001 const struct sock *sk, 2002 const struct sk_buff *skb); 2003 int (*md5_parse)(struct sock *sk, 2004 int optname, 2005 char __user *optval, 2006 int optlen); 2007 #endif 2008 }; 2009 2010 struct tcp_request_sock_ops { 2011 u16 mss_clamp; 2012 #ifdef CONFIG_TCP_MD5SIG 2013 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2014 const struct sock *addr_sk); 2015 int (*calc_md5_hash) (char *location, 2016 const struct tcp_md5sig_key *md5, 2017 const struct sock *sk, 2018 const struct sk_buff *skb); 2019 #endif 2020 void (*init_req)(struct request_sock *req, 2021 const struct sock *sk_listener, 2022 struct sk_buff *skb); 2023 #ifdef CONFIG_SYN_COOKIES 2024 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2025 __u16 *mss); 2026 #endif 2027 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 2028 const struct request_sock *req); 2029 u32 (*init_seq)(const struct sk_buff *skb); 2030 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2031 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2032 struct flowi *fl, struct request_sock *req, 2033 struct tcp_fastopen_cookie *foc, 2034 enum tcp_synack_type synack_type); 2035 }; 2036 2037 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2038 #if IS_ENABLED(CONFIG_IPV6) 2039 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2040 #endif 2041 2042 #ifdef CONFIG_SYN_COOKIES 2043 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2044 const struct sock *sk, struct sk_buff *skb, 2045 __u16 *mss) 2046 { 2047 tcp_synq_overflow(sk); 2048 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2049 return ops->cookie_init_seq(skb, mss); 2050 } 2051 #else 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 return 0; 2057 } 2058 #endif 2059 2060 int tcpv4_offload_init(void); 2061 2062 void tcp_v4_init(void); 2063 void tcp_init(void); 2064 2065 /* tcp_recovery.c */ 2066 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2067 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2068 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2069 u32 reo_wnd); 2070 extern void tcp_rack_mark_lost(struct sock *sk); 2071 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2072 u64 xmit_time); 2073 extern void tcp_rack_reo_timeout(struct sock *sk); 2074 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2075 2076 /* At how many usecs into the future should the RTO fire? */ 2077 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2078 { 2079 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2080 u32 rto = inet_csk(sk)->icsk_rto; 2081 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2082 2083 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2084 } 2085 2086 /* 2087 * Save and compile IPv4 options, return a pointer to it 2088 */ 2089 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2090 struct sk_buff *skb) 2091 { 2092 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2093 struct ip_options_rcu *dopt = NULL; 2094 2095 if (opt->optlen) { 2096 int opt_size = sizeof(*dopt) + opt->optlen; 2097 2098 dopt = kmalloc(opt_size, GFP_ATOMIC); 2099 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2100 kfree(dopt); 2101 dopt = NULL; 2102 } 2103 } 2104 return dopt; 2105 } 2106 2107 /* locally generated TCP pure ACKs have skb->truesize == 2 2108 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2109 * This is much faster than dissecting the packet to find out. 2110 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2111 */ 2112 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2113 { 2114 return skb->truesize == 2; 2115 } 2116 2117 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2118 { 2119 skb->truesize = 2; 2120 } 2121 2122 static inline int tcp_inq(struct sock *sk) 2123 { 2124 struct tcp_sock *tp = tcp_sk(sk); 2125 int answ; 2126 2127 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2128 answ = 0; 2129 } else if (sock_flag(sk, SOCK_URGINLINE) || 2130 !tp->urg_data || 2131 before(tp->urg_seq, tp->copied_seq) || 2132 !before(tp->urg_seq, tp->rcv_nxt)) { 2133 2134 answ = tp->rcv_nxt - tp->copied_seq; 2135 2136 /* Subtract 1, if FIN was received */ 2137 if (answ && sock_flag(sk, SOCK_DONE)) 2138 answ--; 2139 } else { 2140 answ = tp->urg_seq - tp->copied_seq; 2141 } 2142 2143 return answ; 2144 } 2145 2146 int tcp_peek_len(struct socket *sock); 2147 2148 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2149 { 2150 u16 segs_in; 2151 2152 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2153 tp->segs_in += segs_in; 2154 if (skb->len > tcp_hdrlen(skb)) 2155 tp->data_segs_in += segs_in; 2156 } 2157 2158 /* 2159 * TCP listen path runs lockless. 2160 * We forced "struct sock" to be const qualified to make sure 2161 * we don't modify one of its field by mistake. 2162 * Here, we increment sk_drops which is an atomic_t, so we can safely 2163 * make sock writable again. 2164 */ 2165 static inline void tcp_listendrop(const struct sock *sk) 2166 { 2167 atomic_inc(&((struct sock *)sk)->sk_drops); 2168 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2169 } 2170 2171 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2172 2173 /* 2174 * Interface for adding Upper Level Protocols over TCP 2175 */ 2176 2177 #define TCP_ULP_NAME_MAX 16 2178 #define TCP_ULP_MAX 128 2179 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2180 2181 struct tcp_ulp_ops { 2182 struct list_head list; 2183 2184 /* initialize ulp */ 2185 int (*init)(struct sock *sk); 2186 /* update ulp */ 2187 void (*update)(struct sock *sk, struct proto *p, 2188 void (*write_space)(struct sock *sk)); 2189 /* cleanup ulp */ 2190 void (*release)(struct sock *sk); 2191 /* diagnostic */ 2192 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2193 size_t (*get_info_size)(const struct sock *sk); 2194 /* clone ulp */ 2195 void (*clone)(const struct request_sock *req, struct sock *newsk, 2196 const gfp_t priority); 2197 2198 char name[TCP_ULP_NAME_MAX]; 2199 struct module *owner; 2200 }; 2201 int tcp_register_ulp(struct tcp_ulp_ops *type); 2202 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2203 int tcp_set_ulp(struct sock *sk, const char *name); 2204 void tcp_get_available_ulp(char *buf, size_t len); 2205 void tcp_cleanup_ulp(struct sock *sk); 2206 void tcp_update_ulp(struct sock *sk, struct proto *p, 2207 void (*write_space)(struct sock *sk)); 2208 2209 #define MODULE_ALIAS_TCP_ULP(name) \ 2210 __MODULE_INFO(alias, alias_userspace, name); \ 2211 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2212 2213 struct sk_msg; 2214 struct sk_psock; 2215 2216 #ifdef CONFIG_BPF_STREAM_PARSER 2217 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2218 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2219 #else 2220 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2221 { 2222 } 2223 #endif /* CONFIG_BPF_STREAM_PARSER */ 2224 2225 #ifdef CONFIG_NET_SOCK_MSG 2226 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2227 int flags); 2228 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, 2229 struct msghdr *msg, int len, int flags); 2230 #endif /* CONFIG_NET_SOCK_MSG */ 2231 2232 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2233 * is < 0, then the BPF op failed (for example if the loaded BPF 2234 * program does not support the chosen operation or there is no BPF 2235 * program loaded). 2236 */ 2237 #ifdef CONFIG_BPF 2238 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2239 { 2240 struct bpf_sock_ops_kern sock_ops; 2241 int ret; 2242 2243 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2244 if (sk_fullsock(sk)) { 2245 sock_ops.is_fullsock = 1; 2246 sock_owned_by_me(sk); 2247 } 2248 2249 sock_ops.sk = sk; 2250 sock_ops.op = op; 2251 if (nargs > 0) 2252 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2253 2254 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2255 if (ret == 0) 2256 ret = sock_ops.reply; 2257 else 2258 ret = -1; 2259 return ret; 2260 } 2261 2262 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2263 { 2264 u32 args[2] = {arg1, arg2}; 2265 2266 return tcp_call_bpf(sk, op, 2, args); 2267 } 2268 2269 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2270 u32 arg3) 2271 { 2272 u32 args[3] = {arg1, arg2, arg3}; 2273 2274 return tcp_call_bpf(sk, op, 3, args); 2275 } 2276 2277 #else 2278 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2279 { 2280 return -EPERM; 2281 } 2282 2283 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2284 { 2285 return -EPERM; 2286 } 2287 2288 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2289 u32 arg3) 2290 { 2291 return -EPERM; 2292 } 2293 2294 #endif 2295 2296 static inline u32 tcp_timeout_init(struct sock *sk) 2297 { 2298 int timeout; 2299 2300 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2301 2302 if (timeout <= 0) 2303 timeout = TCP_TIMEOUT_INIT; 2304 return timeout; 2305 } 2306 2307 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2308 { 2309 int rwnd; 2310 2311 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2312 2313 if (rwnd < 0) 2314 rwnd = 0; 2315 return rwnd; 2316 } 2317 2318 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2319 { 2320 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2321 } 2322 2323 static inline void tcp_bpf_rtt(struct sock *sk) 2324 { 2325 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2326 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2327 } 2328 2329 #if IS_ENABLED(CONFIG_SMC) 2330 extern struct static_key_false tcp_have_smc; 2331 #endif 2332 2333 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2334 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2335 void (*cad)(struct sock *sk, u32 ack_seq)); 2336 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2337 void clean_acked_data_flush(void); 2338 #endif 2339 2340 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2341 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2342 const struct tcp_sock *tp) 2343 { 2344 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2345 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2346 } 2347 2348 /* Compute Earliest Departure Time for some control packets 2349 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2350 */ 2351 static inline u64 tcp_transmit_time(const struct sock *sk) 2352 { 2353 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2354 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2355 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2356 2357 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2358 } 2359 return 0; 2360 } 2361 2362 #endif /* _TCP_H */ 2363