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