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 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 257 258 extern struct percpu_counter tcp_sockets_allocated; 259 extern unsigned long tcp_memory_pressure; 260 261 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 262 static inline bool tcp_under_memory_pressure(const struct sock *sk) 263 { 264 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 265 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 266 return true; 267 268 return READ_ONCE(tcp_memory_pressure); 269 } 270 /* 271 * The next routines deal with comparing 32 bit unsigned ints 272 * and worry about wraparound (automatic with unsigned arithmetic). 273 */ 274 275 static inline bool before(__u32 seq1, __u32 seq2) 276 { 277 return (__s32)(seq1-seq2) < 0; 278 } 279 #define after(seq2, seq1) before(seq1, seq2) 280 281 /* is s2<=s1<=s3 ? */ 282 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 283 { 284 return seq3 - seq2 >= seq1 - seq2; 285 } 286 287 static inline bool tcp_out_of_memory(struct sock *sk) 288 { 289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 291 return true; 292 return false; 293 } 294 295 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 296 { 297 sk_wmem_queued_add(sk, -skb->truesize); 298 if (!skb_zcopy_pure(skb)) 299 sk_mem_uncharge(sk, skb->truesize); 300 else 301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 302 __kfree_skb(skb); 303 } 304 305 void sk_forced_mem_schedule(struct sock *sk, int size); 306 307 bool tcp_check_oom(struct sock *sk, int shift); 308 309 310 extern struct proto tcp_prot; 311 312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 316 317 void tcp_tasklet_init(void); 318 319 int tcp_v4_err(struct sk_buff *skb, u32); 320 321 void tcp_shutdown(struct sock *sk, int how); 322 323 int tcp_v4_early_demux(struct sk_buff *skb); 324 int tcp_v4_rcv(struct sk_buff *skb); 325 326 void tcp_remove_empty_skb(struct sock *sk); 327 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 328 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 329 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 330 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 331 int flags); 332 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 333 size_t size, int flags); 334 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 335 size_t size, int flags); 336 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 337 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 338 int size_goal); 339 void tcp_release_cb(struct sock *sk); 340 void tcp_wfree(struct sk_buff *skb); 341 void tcp_write_timer_handler(struct sock *sk); 342 void tcp_delack_timer_handler(struct sock *sk); 343 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 344 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 345 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 346 void tcp_rcv_space_adjust(struct sock *sk); 347 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 348 void tcp_twsk_destructor(struct sock *sk); 349 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 350 struct pipe_inode_info *pipe, size_t len, 351 unsigned int flags); 352 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 353 bool force_schedule); 354 355 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 356 static inline void tcp_dec_quickack_mode(struct sock *sk, 357 const unsigned int pkts) 358 { 359 struct inet_connection_sock *icsk = inet_csk(sk); 360 361 if (icsk->icsk_ack.quick) { 362 if (pkts >= icsk->icsk_ack.quick) { 363 icsk->icsk_ack.quick = 0; 364 /* Leaving quickack mode we deflate ATO. */ 365 icsk->icsk_ack.ato = TCP_ATO_MIN; 366 } else 367 icsk->icsk_ack.quick -= pkts; 368 } 369 } 370 371 #define TCP_ECN_OK 1 372 #define TCP_ECN_QUEUE_CWR 2 373 #define TCP_ECN_DEMAND_CWR 4 374 #define TCP_ECN_SEEN 8 375 376 enum tcp_tw_status { 377 TCP_TW_SUCCESS = 0, 378 TCP_TW_RST = 1, 379 TCP_TW_ACK = 2, 380 TCP_TW_SYN = 3 381 }; 382 383 384 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 385 struct sk_buff *skb, 386 const struct tcphdr *th); 387 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 388 struct request_sock *req, bool fastopen, 389 bool *lost_race); 390 int tcp_child_process(struct sock *parent, struct sock *child, 391 struct sk_buff *skb); 392 void tcp_enter_loss(struct sock *sk); 393 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 394 void tcp_clear_retrans(struct tcp_sock *tp); 395 void tcp_update_metrics(struct sock *sk); 396 void tcp_init_metrics(struct sock *sk); 397 void tcp_metrics_init(void); 398 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 399 void __tcp_close(struct sock *sk, long timeout); 400 void tcp_close(struct sock *sk, long timeout); 401 void tcp_init_sock(struct sock *sk); 402 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 403 __poll_t tcp_poll(struct file *file, struct socket *sock, 404 struct poll_table_struct *wait); 405 int tcp_getsockopt(struct sock *sk, int level, int optname, 406 char __user *optval, int __user *optlen); 407 bool tcp_bpf_bypass_getsockopt(int level, int optname); 408 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 409 unsigned int optlen); 410 void tcp_set_keepalive(struct sock *sk, int val); 411 void tcp_syn_ack_timeout(const struct request_sock *req); 412 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 413 int flags, int *addr_len); 414 int tcp_set_rcvlowat(struct sock *sk, int val); 415 int tcp_set_window_clamp(struct sock *sk, int val); 416 void tcp_update_recv_tstamps(struct sk_buff *skb, 417 struct scm_timestamping_internal *tss); 418 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 419 struct scm_timestamping_internal *tss); 420 void tcp_data_ready(struct sock *sk); 421 #ifdef CONFIG_MMU 422 int tcp_mmap(struct file *file, struct socket *sock, 423 struct vm_area_struct *vma); 424 #endif 425 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 426 struct tcp_options_received *opt_rx, 427 int estab, struct tcp_fastopen_cookie *foc); 428 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 429 430 /* 431 * BPF SKB-less helpers 432 */ 433 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 434 struct tcphdr *th, u32 *cookie); 435 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 436 struct tcphdr *th, u32 *cookie); 437 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 438 const struct tcp_request_sock_ops *af_ops, 439 struct sock *sk, struct tcphdr *th); 440 /* 441 * TCP v4 functions exported for the inet6 API 442 */ 443 444 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 445 void tcp_v4_mtu_reduced(struct sock *sk); 446 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 447 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 448 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 449 struct sock *tcp_create_openreq_child(const struct sock *sk, 450 struct request_sock *req, 451 struct sk_buff *skb); 452 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 453 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 454 struct request_sock *req, 455 struct dst_entry *dst, 456 struct request_sock *req_unhash, 457 bool *own_req); 458 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 459 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 460 int tcp_connect(struct sock *sk); 461 enum tcp_synack_type { 462 TCP_SYNACK_NORMAL, 463 TCP_SYNACK_FASTOPEN, 464 TCP_SYNACK_COOKIE, 465 }; 466 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 467 struct request_sock *req, 468 struct tcp_fastopen_cookie *foc, 469 enum tcp_synack_type synack_type, 470 struct sk_buff *syn_skb); 471 int tcp_disconnect(struct sock *sk, int flags); 472 473 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 474 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 475 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 476 477 /* From syncookies.c */ 478 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 479 struct request_sock *req, 480 struct dst_entry *dst, u32 tsoff); 481 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 482 u32 cookie); 483 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 484 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 485 const struct tcp_request_sock_ops *af_ops, 486 struct sock *sk, struct sk_buff *skb); 487 #ifdef CONFIG_SYN_COOKIES 488 489 /* Syncookies use a monotonic timer which increments every 60 seconds. 490 * This counter is used both as a hash input and partially encoded into 491 * the cookie value. A cookie is only validated further if the delta 492 * between the current counter value and the encoded one is less than this, 493 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 494 * the counter advances immediately after a cookie is generated). 495 */ 496 #define MAX_SYNCOOKIE_AGE 2 497 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 498 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 499 500 /* syncookies: remember time of last synqueue overflow 501 * But do not dirty this field too often (once per second is enough) 502 * It is racy as we do not hold a lock, but race is very minor. 503 */ 504 static inline void tcp_synq_overflow(const struct sock *sk) 505 { 506 unsigned int last_overflow; 507 unsigned int now = jiffies; 508 509 if (sk->sk_reuseport) { 510 struct sock_reuseport *reuse; 511 512 reuse = rcu_dereference(sk->sk_reuseport_cb); 513 if (likely(reuse)) { 514 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 515 if (!time_between32(now, last_overflow, 516 last_overflow + HZ)) 517 WRITE_ONCE(reuse->synq_overflow_ts, now); 518 return; 519 } 520 } 521 522 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 523 if (!time_between32(now, last_overflow, last_overflow + HZ)) 524 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 525 } 526 527 /* syncookies: no recent synqueue overflow on this listening socket? */ 528 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 529 { 530 unsigned int last_overflow; 531 unsigned int now = jiffies; 532 533 if (sk->sk_reuseport) { 534 struct sock_reuseport *reuse; 535 536 reuse = rcu_dereference(sk->sk_reuseport_cb); 537 if (likely(reuse)) { 538 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 539 return !time_between32(now, last_overflow - HZ, 540 last_overflow + 541 TCP_SYNCOOKIE_VALID); 542 } 543 } 544 545 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 546 547 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 548 * then we're under synflood. However, we have to use 549 * 'last_overflow - HZ' as lower bound. That's because a concurrent 550 * tcp_synq_overflow() could update .ts_recent_stamp after we read 551 * jiffies but before we store .ts_recent_stamp into last_overflow, 552 * which could lead to rejecting a valid syncookie. 553 */ 554 return !time_between32(now, last_overflow - HZ, 555 last_overflow + TCP_SYNCOOKIE_VALID); 556 } 557 558 static inline u32 tcp_cookie_time(void) 559 { 560 u64 val = get_jiffies_64(); 561 562 do_div(val, TCP_SYNCOOKIE_PERIOD); 563 return val; 564 } 565 566 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 567 u16 *mssp); 568 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 569 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 570 bool cookie_timestamp_decode(const struct net *net, 571 struct tcp_options_received *opt); 572 bool cookie_ecn_ok(const struct tcp_options_received *opt, 573 const struct net *net, const struct dst_entry *dst); 574 575 /* From net/ipv6/syncookies.c */ 576 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 577 u32 cookie); 578 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 579 580 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 581 const struct tcphdr *th, u16 *mssp); 582 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 583 #endif 584 /* tcp_output.c */ 585 586 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 587 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 588 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 589 int nonagle); 590 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 591 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 592 void tcp_retransmit_timer(struct sock *sk); 593 void tcp_xmit_retransmit_queue(struct sock *); 594 void tcp_simple_retransmit(struct sock *); 595 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 596 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 597 enum tcp_queue { 598 TCP_FRAG_IN_WRITE_QUEUE, 599 TCP_FRAG_IN_RTX_QUEUE, 600 }; 601 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 602 struct sk_buff *skb, u32 len, 603 unsigned int mss_now, gfp_t gfp); 604 605 void tcp_send_probe0(struct sock *); 606 void tcp_send_partial(struct sock *); 607 int tcp_write_wakeup(struct sock *, int mib); 608 void tcp_send_fin(struct sock *sk); 609 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 610 int tcp_send_synack(struct sock *); 611 void tcp_push_one(struct sock *, unsigned int mss_now); 612 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 613 void tcp_send_ack(struct sock *sk); 614 void tcp_send_delayed_ack(struct sock *sk); 615 void tcp_send_loss_probe(struct sock *sk); 616 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 617 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 618 const struct sk_buff *next_skb); 619 620 /* tcp_input.c */ 621 void tcp_rearm_rto(struct sock *sk); 622 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 623 void tcp_reset(struct sock *sk, struct sk_buff *skb); 624 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 625 void tcp_fin(struct sock *sk); 626 void tcp_check_space(struct sock *sk); 627 628 /* tcp_timer.c */ 629 void tcp_init_xmit_timers(struct sock *); 630 static inline void tcp_clear_xmit_timers(struct sock *sk) 631 { 632 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 633 __sock_put(sk); 634 635 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 636 __sock_put(sk); 637 638 inet_csk_clear_xmit_timers(sk); 639 } 640 641 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 642 unsigned int tcp_current_mss(struct sock *sk); 643 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 644 645 /* Bound MSS / TSO packet size with the half of the window */ 646 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 647 { 648 int cutoff; 649 650 /* When peer uses tiny windows, there is no use in packetizing 651 * to sub-MSS pieces for the sake of SWS or making sure there 652 * are enough packets in the pipe for fast recovery. 653 * 654 * On the other hand, for extremely large MSS devices, handling 655 * smaller than MSS windows in this way does make sense. 656 */ 657 if (tp->max_window > TCP_MSS_DEFAULT) 658 cutoff = (tp->max_window >> 1); 659 else 660 cutoff = tp->max_window; 661 662 if (cutoff && pktsize > cutoff) 663 return max_t(int, cutoff, 68U - tp->tcp_header_len); 664 else 665 return pktsize; 666 } 667 668 /* tcp.c */ 669 void tcp_get_info(struct sock *, struct tcp_info *); 670 671 /* Read 'sendfile()'-style from a TCP socket */ 672 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 673 sk_read_actor_t recv_actor); 674 675 void tcp_initialize_rcv_mss(struct sock *sk); 676 677 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 678 int tcp_mss_to_mtu(struct sock *sk, int mss); 679 void tcp_mtup_init(struct sock *sk); 680 681 static inline void tcp_bound_rto(const struct sock *sk) 682 { 683 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 684 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 685 } 686 687 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 688 { 689 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 690 } 691 692 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 693 { 694 /* mptcp hooks are only on the slow path */ 695 if (sk_is_mptcp((struct sock *)tp)) 696 return; 697 698 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 699 ntohl(TCP_FLAG_ACK) | 700 snd_wnd); 701 } 702 703 static inline void tcp_fast_path_on(struct tcp_sock *tp) 704 { 705 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 706 } 707 708 static inline void tcp_fast_path_check(struct sock *sk) 709 { 710 struct tcp_sock *tp = tcp_sk(sk); 711 712 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 713 tp->rcv_wnd && 714 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 715 !tp->urg_data) 716 tcp_fast_path_on(tp); 717 } 718 719 /* Compute the actual rto_min value */ 720 static inline u32 tcp_rto_min(struct sock *sk) 721 { 722 const struct dst_entry *dst = __sk_dst_get(sk); 723 u32 rto_min = inet_csk(sk)->icsk_rto_min; 724 725 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 726 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 727 return rto_min; 728 } 729 730 static inline u32 tcp_rto_min_us(struct sock *sk) 731 { 732 return jiffies_to_usecs(tcp_rto_min(sk)); 733 } 734 735 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 736 { 737 return dst_metric_locked(dst, RTAX_CC_ALGO); 738 } 739 740 /* Minimum RTT in usec. ~0 means not available. */ 741 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 742 { 743 return minmax_get(&tp->rtt_min); 744 } 745 746 /* Compute the actual receive window we are currently advertising. 747 * Rcv_nxt can be after the window if our peer push more data 748 * than the offered window. 749 */ 750 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 751 { 752 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 753 754 if (win < 0) 755 win = 0; 756 return (u32) win; 757 } 758 759 /* Choose a new window, without checks for shrinking, and without 760 * scaling applied to the result. The caller does these things 761 * if necessary. This is a "raw" window selection. 762 */ 763 u32 __tcp_select_window(struct sock *sk); 764 765 void tcp_send_window_probe(struct sock *sk); 766 767 /* TCP uses 32bit jiffies to save some space. 768 * Note that this is different from tcp_time_stamp, which 769 * historically has been the same until linux-4.13. 770 */ 771 #define tcp_jiffies32 ((u32)jiffies) 772 773 /* 774 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 775 * It is no longer tied to jiffies, but to 1 ms clock. 776 * Note: double check if you want to use tcp_jiffies32 instead of this. 777 */ 778 #define TCP_TS_HZ 1000 779 780 static inline u64 tcp_clock_ns(void) 781 { 782 return ktime_get_ns(); 783 } 784 785 static inline u64 tcp_clock_us(void) 786 { 787 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 788 } 789 790 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 791 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 792 { 793 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 794 } 795 796 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 797 static inline u32 tcp_ns_to_ts(u64 ns) 798 { 799 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 800 } 801 802 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 803 static inline u32 tcp_time_stamp_raw(void) 804 { 805 return tcp_ns_to_ts(tcp_clock_ns()); 806 } 807 808 void tcp_mstamp_refresh(struct tcp_sock *tp); 809 810 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 811 { 812 return max_t(s64, t1 - t0, 0); 813 } 814 815 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 816 { 817 return tcp_ns_to_ts(skb->skb_mstamp_ns); 818 } 819 820 /* provide the departure time in us unit */ 821 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 822 { 823 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 824 } 825 826 827 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 828 829 #define TCPHDR_FIN 0x01 830 #define TCPHDR_SYN 0x02 831 #define TCPHDR_RST 0x04 832 #define TCPHDR_PSH 0x08 833 #define TCPHDR_ACK 0x10 834 #define TCPHDR_URG 0x20 835 #define TCPHDR_ECE 0x40 836 #define TCPHDR_CWR 0x80 837 838 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 839 840 /* This is what the send packet queuing engine uses to pass 841 * TCP per-packet control information to the transmission code. 842 * We also store the host-order sequence numbers in here too. 843 * This is 44 bytes if IPV6 is enabled. 844 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 845 */ 846 struct tcp_skb_cb { 847 __u32 seq; /* Starting sequence number */ 848 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 849 union { 850 /* Note : tcp_tw_isn is used in input path only 851 * (isn chosen by tcp_timewait_state_process()) 852 * 853 * tcp_gso_segs/size are used in write queue only, 854 * cf tcp_skb_pcount()/tcp_skb_mss() 855 */ 856 __u32 tcp_tw_isn; 857 struct { 858 u16 tcp_gso_segs; 859 u16 tcp_gso_size; 860 }; 861 }; 862 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 863 864 __u8 sacked; /* State flags for SACK. */ 865 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 866 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 867 #define TCPCB_LOST 0x04 /* SKB is lost */ 868 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 869 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 870 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 871 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 872 TCPCB_REPAIRED) 873 874 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 875 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 876 eor:1, /* Is skb MSG_EOR marked? */ 877 has_rxtstamp:1, /* SKB has a RX timestamp */ 878 unused:5; 879 __u32 ack_seq; /* Sequence number ACK'd */ 880 union { 881 struct { 882 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 883 /* There is space for up to 24 bytes */ 884 __u32 is_app_limited:1, /* cwnd not fully used? */ 885 delivered_ce:20, 886 unused:11; 887 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 888 __u32 delivered; 889 /* start of send pipeline phase */ 890 u64 first_tx_mstamp; 891 /* when we reached the "delivered" count */ 892 u64 delivered_mstamp; 893 } tx; /* only used for outgoing skbs */ 894 union { 895 struct inet_skb_parm h4; 896 #if IS_ENABLED(CONFIG_IPV6) 897 struct inet6_skb_parm h6; 898 #endif 899 } header; /* For incoming skbs */ 900 }; 901 }; 902 903 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 904 905 extern const struct inet_connection_sock_af_ops ipv4_specific; 906 907 #if IS_ENABLED(CONFIG_IPV6) 908 /* This is the variant of inet6_iif() that must be used by TCP, 909 * as TCP moves IP6CB into a different location in skb->cb[] 910 */ 911 static inline int tcp_v6_iif(const struct sk_buff *skb) 912 { 913 return TCP_SKB_CB(skb)->header.h6.iif; 914 } 915 916 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 917 { 918 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 919 920 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 921 } 922 923 /* TCP_SKB_CB reference means this can not be used from early demux */ 924 static inline int tcp_v6_sdif(const struct sk_buff *skb) 925 { 926 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 927 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 928 return TCP_SKB_CB(skb)->header.h6.iif; 929 #endif 930 return 0; 931 } 932 933 extern const struct inet_connection_sock_af_ops ipv6_specific; 934 935 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 936 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 937 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb)); 938 939 #endif 940 941 /* TCP_SKB_CB reference means this can not be used from early demux */ 942 static inline int tcp_v4_sdif(struct sk_buff *skb) 943 { 944 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 945 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 946 return TCP_SKB_CB(skb)->header.h4.iif; 947 #endif 948 return 0; 949 } 950 951 /* Due to TSO, an SKB can be composed of multiple actual 952 * packets. To keep these tracked properly, we use this. 953 */ 954 static inline int tcp_skb_pcount(const struct sk_buff *skb) 955 { 956 return TCP_SKB_CB(skb)->tcp_gso_segs; 957 } 958 959 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 960 { 961 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 962 } 963 964 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 965 { 966 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 967 } 968 969 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 970 static inline int tcp_skb_mss(const struct sk_buff *skb) 971 { 972 return TCP_SKB_CB(skb)->tcp_gso_size; 973 } 974 975 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 976 { 977 return likely(!TCP_SKB_CB(skb)->eor); 978 } 979 980 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 981 const struct sk_buff *from) 982 { 983 return likely(tcp_skb_can_collapse_to(to) && 984 mptcp_skb_can_collapse(to, from) && 985 skb_pure_zcopy_same(to, from)); 986 } 987 988 /* Events passed to congestion control interface */ 989 enum tcp_ca_event { 990 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 991 CA_EVENT_CWND_RESTART, /* congestion window restart */ 992 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 993 CA_EVENT_LOSS, /* loss timeout */ 994 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 995 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 996 }; 997 998 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 999 enum tcp_ca_ack_event_flags { 1000 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1001 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1002 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1003 }; 1004 1005 /* 1006 * Interface for adding new TCP congestion control handlers 1007 */ 1008 #define TCP_CA_NAME_MAX 16 1009 #define TCP_CA_MAX 128 1010 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1011 1012 #define TCP_CA_UNSPEC 0 1013 1014 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1015 #define TCP_CONG_NON_RESTRICTED 0x1 1016 /* Requires ECN/ECT set on all packets */ 1017 #define TCP_CONG_NEEDS_ECN 0x2 1018 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1019 1020 union tcp_cc_info; 1021 1022 struct ack_sample { 1023 u32 pkts_acked; 1024 s32 rtt_us; 1025 u32 in_flight; 1026 }; 1027 1028 /* A rate sample measures the number of (original/retransmitted) data 1029 * packets delivered "delivered" over an interval of time "interval_us". 1030 * The tcp_rate.c code fills in the rate sample, and congestion 1031 * control modules that define a cong_control function to run at the end 1032 * of ACK processing can optionally chose to consult this sample when 1033 * setting cwnd and pacing rate. 1034 * A sample is invalid if "delivered" or "interval_us" is negative. 1035 */ 1036 struct rate_sample { 1037 u64 prior_mstamp; /* starting timestamp for interval */ 1038 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1039 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1040 s32 delivered; /* number of packets delivered over interval */ 1041 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1042 long interval_us; /* time for tp->delivered to incr "delivered" */ 1043 u32 snd_interval_us; /* snd interval for delivered packets */ 1044 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1045 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1046 int losses; /* number of packets marked lost upon ACK */ 1047 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1048 u32 prior_in_flight; /* in flight before this ACK */ 1049 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1050 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1051 bool is_retrans; /* is sample from retransmission? */ 1052 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1053 }; 1054 1055 struct tcp_congestion_ops { 1056 /* fast path fields are put first to fill one cache line */ 1057 1058 /* return slow start threshold (required) */ 1059 u32 (*ssthresh)(struct sock *sk); 1060 1061 /* do new cwnd calculation (required) */ 1062 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1063 1064 /* call before changing ca_state (optional) */ 1065 void (*set_state)(struct sock *sk, u8 new_state); 1066 1067 /* call when cwnd event occurs (optional) */ 1068 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1069 1070 /* call when ack arrives (optional) */ 1071 void (*in_ack_event)(struct sock *sk, u32 flags); 1072 1073 /* hook for packet ack accounting (optional) */ 1074 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1075 1076 /* override sysctl_tcp_min_tso_segs */ 1077 u32 (*min_tso_segs)(struct sock *sk); 1078 1079 /* call when packets are delivered to update cwnd and pacing rate, 1080 * after all the ca_state processing. (optional) 1081 */ 1082 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1083 1084 1085 /* new value of cwnd after loss (required) */ 1086 u32 (*undo_cwnd)(struct sock *sk); 1087 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1088 u32 (*sndbuf_expand)(struct sock *sk); 1089 1090 /* control/slow paths put last */ 1091 /* get info for inet_diag (optional) */ 1092 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1093 union tcp_cc_info *info); 1094 1095 char name[TCP_CA_NAME_MAX]; 1096 struct module *owner; 1097 struct list_head list; 1098 u32 key; 1099 u32 flags; 1100 1101 /* initialize private data (optional) */ 1102 void (*init)(struct sock *sk); 1103 /* cleanup private data (optional) */ 1104 void (*release)(struct sock *sk); 1105 } ____cacheline_aligned_in_smp; 1106 1107 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1108 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1109 1110 void tcp_assign_congestion_control(struct sock *sk); 1111 void tcp_init_congestion_control(struct sock *sk); 1112 void tcp_cleanup_congestion_control(struct sock *sk); 1113 int tcp_set_default_congestion_control(struct net *net, const char *name); 1114 void tcp_get_default_congestion_control(struct net *net, char *name); 1115 void tcp_get_available_congestion_control(char *buf, size_t len); 1116 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1117 int tcp_set_allowed_congestion_control(char *allowed); 1118 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1119 bool cap_net_admin); 1120 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1121 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1122 1123 u32 tcp_reno_ssthresh(struct sock *sk); 1124 u32 tcp_reno_undo_cwnd(struct sock *sk); 1125 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1126 extern struct tcp_congestion_ops tcp_reno; 1127 1128 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1129 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1130 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1131 #ifdef CONFIG_INET 1132 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1133 #else 1134 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1135 { 1136 return NULL; 1137 } 1138 #endif 1139 1140 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1141 { 1142 const struct inet_connection_sock *icsk = inet_csk(sk); 1143 1144 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 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_cong.c */ 1156 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1157 1158 /* From tcp_rate.c */ 1159 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1160 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1161 struct rate_sample *rs); 1162 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1163 bool is_sack_reneg, struct rate_sample *rs); 1164 void tcp_rate_check_app_limited(struct sock *sk); 1165 1166 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1167 { 1168 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1169 } 1170 1171 /* These functions determine how the current flow behaves in respect of SACK 1172 * handling. SACK is negotiated with the peer, and therefore it can vary 1173 * between different flows. 1174 * 1175 * tcp_is_sack - SACK enabled 1176 * tcp_is_reno - No SACK 1177 */ 1178 static inline int tcp_is_sack(const struct tcp_sock *tp) 1179 { 1180 return likely(tp->rx_opt.sack_ok); 1181 } 1182 1183 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1184 { 1185 return !tcp_is_sack(tp); 1186 } 1187 1188 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1189 { 1190 return tp->sacked_out + tp->lost_out; 1191 } 1192 1193 /* This determines how many packets are "in the network" to the best 1194 * of our knowledge. In many cases it is conservative, but where 1195 * detailed information is available from the receiver (via SACK 1196 * blocks etc.) we can make more aggressive calculations. 1197 * 1198 * Use this for decisions involving congestion control, use just 1199 * tp->packets_out to determine if the send queue is empty or not. 1200 * 1201 * Read this equation as: 1202 * 1203 * "Packets sent once on transmission queue" MINUS 1204 * "Packets left network, but not honestly ACKed yet" PLUS 1205 * "Packets fast retransmitted" 1206 */ 1207 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1208 { 1209 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1210 } 1211 1212 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1213 1214 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1215 { 1216 return tp->snd_cwnd; 1217 } 1218 1219 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1220 { 1221 WARN_ON_ONCE((int)val <= 0); 1222 tp->snd_cwnd = val; 1223 } 1224 1225 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1226 { 1227 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1228 } 1229 1230 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1231 { 1232 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1233 } 1234 1235 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1236 { 1237 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1238 (1 << inet_csk(sk)->icsk_ca_state); 1239 } 1240 1241 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1242 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1243 * ssthresh. 1244 */ 1245 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1246 { 1247 const struct tcp_sock *tp = tcp_sk(sk); 1248 1249 if (tcp_in_cwnd_reduction(sk)) 1250 return tp->snd_ssthresh; 1251 else 1252 return max(tp->snd_ssthresh, 1253 ((tcp_snd_cwnd(tp) >> 1) + 1254 (tcp_snd_cwnd(tp) >> 2))); 1255 } 1256 1257 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1258 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1259 1260 void tcp_enter_cwr(struct sock *sk); 1261 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1262 1263 /* The maximum number of MSS of available cwnd for which TSO defers 1264 * sending if not using sysctl_tcp_tso_win_divisor. 1265 */ 1266 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1267 { 1268 return 3; 1269 } 1270 1271 /* Returns end sequence number of the receiver's advertised window */ 1272 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1273 { 1274 return tp->snd_una + tp->snd_wnd; 1275 } 1276 1277 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1278 * flexible approach. The RFC suggests cwnd should not be raised unless 1279 * it was fully used previously. And that's exactly what we do in 1280 * congestion avoidance mode. But in slow start we allow cwnd to grow 1281 * as long as the application has used half the cwnd. 1282 * Example : 1283 * cwnd is 10 (IW10), but application sends 9 frames. 1284 * We allow cwnd to reach 18 when all frames are ACKed. 1285 * This check is safe because it's as aggressive as slow start which already 1286 * risks 100% overshoot. The advantage is that we discourage application to 1287 * either send more filler packets or data to artificially blow up the cwnd 1288 * usage, and allow application-limited process to probe bw more aggressively. 1289 */ 1290 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1291 { 1292 const struct tcp_sock *tp = tcp_sk(sk); 1293 1294 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1295 if (tcp_in_slow_start(tp)) 1296 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1297 1298 return tp->is_cwnd_limited; 1299 } 1300 1301 /* BBR congestion control needs pacing. 1302 * Same remark for SO_MAX_PACING_RATE. 1303 * sch_fq packet scheduler is efficiently handling pacing, 1304 * but is not always installed/used. 1305 * Return true if TCP stack should pace packets itself. 1306 */ 1307 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1308 { 1309 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1310 } 1311 1312 /* Estimates in how many jiffies next packet for this flow can be sent. 1313 * Scheduling a retransmit timer too early would be silly. 1314 */ 1315 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1316 { 1317 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1318 1319 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1320 } 1321 1322 static inline void tcp_reset_xmit_timer(struct sock *sk, 1323 const int what, 1324 unsigned long when, 1325 const unsigned long max_when) 1326 { 1327 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1328 max_when); 1329 } 1330 1331 /* Something is really bad, we could not queue an additional packet, 1332 * because qdisc is full or receiver sent a 0 window, or we are paced. 1333 * We do not want to add fuel to the fire, or abort too early, 1334 * so make sure the timer we arm now is at least 200ms in the future, 1335 * regardless of current icsk_rto value (as it could be ~2ms) 1336 */ 1337 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1338 { 1339 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1340 } 1341 1342 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1343 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1344 unsigned long max_when) 1345 { 1346 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1347 inet_csk(sk)->icsk_backoff); 1348 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1349 1350 return (unsigned long)min_t(u64, when, max_when); 1351 } 1352 1353 static inline void tcp_check_probe_timer(struct sock *sk) 1354 { 1355 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1356 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1357 tcp_probe0_base(sk), TCP_RTO_MAX); 1358 } 1359 1360 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1361 { 1362 tp->snd_wl1 = seq; 1363 } 1364 1365 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1366 { 1367 tp->snd_wl1 = seq; 1368 } 1369 1370 /* 1371 * Calculate(/check) TCP checksum 1372 */ 1373 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1374 __be32 daddr, __wsum base) 1375 { 1376 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1377 } 1378 1379 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1380 { 1381 return !skb_csum_unnecessary(skb) && 1382 __skb_checksum_complete(skb); 1383 } 1384 1385 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1386 enum skb_drop_reason *reason); 1387 1388 1389 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1390 void tcp_set_state(struct sock *sk, int state); 1391 void tcp_done(struct sock *sk); 1392 int tcp_abort(struct sock *sk, int err); 1393 1394 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1395 { 1396 rx_opt->dsack = 0; 1397 rx_opt->num_sacks = 0; 1398 } 1399 1400 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1401 1402 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1403 { 1404 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1405 struct tcp_sock *tp = tcp_sk(sk); 1406 s32 delta; 1407 1408 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1409 ca_ops->cong_control) 1410 return; 1411 delta = tcp_jiffies32 - tp->lsndtime; 1412 if (delta > inet_csk(sk)->icsk_rto) 1413 tcp_cwnd_restart(sk, delta); 1414 } 1415 1416 /* Determine a window scaling and initial window to offer. */ 1417 void tcp_select_initial_window(const struct sock *sk, int __space, 1418 __u32 mss, __u32 *rcv_wnd, 1419 __u32 *window_clamp, int wscale_ok, 1420 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1421 1422 static inline int tcp_win_from_space(const struct sock *sk, int space) 1423 { 1424 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1425 1426 return tcp_adv_win_scale <= 0 ? 1427 (space>>(-tcp_adv_win_scale)) : 1428 space - (space>>tcp_adv_win_scale); 1429 } 1430 1431 /* Note: caller must be prepared to deal with negative returns */ 1432 static inline int tcp_space(const struct sock *sk) 1433 { 1434 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1435 READ_ONCE(sk->sk_backlog.len) - 1436 atomic_read(&sk->sk_rmem_alloc)); 1437 } 1438 1439 static inline int tcp_full_space(const struct sock *sk) 1440 { 1441 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1442 } 1443 1444 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1445 { 1446 int unused_mem = sk_unused_reserved_mem(sk); 1447 struct tcp_sock *tp = tcp_sk(sk); 1448 1449 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1450 if (unused_mem) 1451 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1452 tcp_win_from_space(sk, unused_mem)); 1453 } 1454 1455 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1456 1457 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1458 * If 87.5 % (7/8) of the space has been consumed, we want to override 1459 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1460 * len/truesize ratio. 1461 */ 1462 static inline bool tcp_rmem_pressure(const struct sock *sk) 1463 { 1464 int rcvbuf, threshold; 1465 1466 if (tcp_under_memory_pressure(sk)) 1467 return true; 1468 1469 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1470 threshold = rcvbuf - (rcvbuf >> 3); 1471 1472 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1473 } 1474 1475 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1476 { 1477 const struct tcp_sock *tp = tcp_sk(sk); 1478 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1479 1480 if (avail <= 0) 1481 return false; 1482 1483 return (avail >= target) || tcp_rmem_pressure(sk) || 1484 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1485 } 1486 1487 extern void tcp_openreq_init_rwin(struct request_sock *req, 1488 const struct sock *sk_listener, 1489 const struct dst_entry *dst); 1490 1491 void tcp_enter_memory_pressure(struct sock *sk); 1492 void tcp_leave_memory_pressure(struct sock *sk); 1493 1494 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1495 { 1496 struct net *net = sock_net((struct sock *)tp); 1497 1498 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1499 } 1500 1501 static inline int keepalive_time_when(const struct tcp_sock *tp) 1502 { 1503 struct net *net = sock_net((struct sock *)tp); 1504 1505 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1506 } 1507 1508 static inline int keepalive_probes(const struct tcp_sock *tp) 1509 { 1510 struct net *net = sock_net((struct sock *)tp); 1511 1512 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1513 } 1514 1515 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1516 { 1517 const struct inet_connection_sock *icsk = &tp->inet_conn; 1518 1519 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1520 tcp_jiffies32 - tp->rcv_tstamp); 1521 } 1522 1523 static inline int tcp_fin_time(const struct sock *sk) 1524 { 1525 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1526 const int rto = inet_csk(sk)->icsk_rto; 1527 1528 if (fin_timeout < (rto << 2) - (rto >> 1)) 1529 fin_timeout = (rto << 2) - (rto >> 1); 1530 1531 return fin_timeout; 1532 } 1533 1534 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1535 int paws_win) 1536 { 1537 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1538 return true; 1539 if (unlikely(!time_before32(ktime_get_seconds(), 1540 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1541 return true; 1542 /* 1543 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1544 * then following tcp messages have valid values. Ignore 0 value, 1545 * or else 'negative' tsval might forbid us to accept their packets. 1546 */ 1547 if (!rx_opt->ts_recent) 1548 return true; 1549 return false; 1550 } 1551 1552 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1553 int rst) 1554 { 1555 if (tcp_paws_check(rx_opt, 0)) 1556 return false; 1557 1558 /* RST segments are not recommended to carry timestamp, 1559 and, if they do, it is recommended to ignore PAWS because 1560 "their cleanup function should take precedence over timestamps." 1561 Certainly, it is mistake. It is necessary to understand the reasons 1562 of this constraint to relax it: if peer reboots, clock may go 1563 out-of-sync and half-open connections will not be reset. 1564 Actually, the problem would be not existing if all 1565 the implementations followed draft about maintaining clock 1566 via reboots. Linux-2.2 DOES NOT! 1567 1568 However, we can relax time bounds for RST segments to MSL. 1569 */ 1570 if (rst && !time_before32(ktime_get_seconds(), 1571 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1572 return false; 1573 return true; 1574 } 1575 1576 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1577 int mib_idx, u32 *last_oow_ack_time); 1578 1579 static inline void tcp_mib_init(struct net *net) 1580 { 1581 /* See RFC 2012 */ 1582 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1583 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1584 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1585 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1586 } 1587 1588 /* from STCP */ 1589 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1590 { 1591 tp->lost_skb_hint = NULL; 1592 } 1593 1594 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1595 { 1596 tcp_clear_retrans_hints_partial(tp); 1597 tp->retransmit_skb_hint = NULL; 1598 } 1599 1600 union tcp_md5_addr { 1601 struct in_addr a4; 1602 #if IS_ENABLED(CONFIG_IPV6) 1603 struct in6_addr a6; 1604 #endif 1605 }; 1606 1607 /* - key database */ 1608 struct tcp_md5sig_key { 1609 struct hlist_node node; 1610 u8 keylen; 1611 u8 family; /* AF_INET or AF_INET6 */ 1612 u8 prefixlen; 1613 u8 flags; 1614 union tcp_md5_addr addr; 1615 int l3index; /* set if key added with L3 scope */ 1616 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1617 struct rcu_head rcu; 1618 }; 1619 1620 /* - sock block */ 1621 struct tcp_md5sig_info { 1622 struct hlist_head head; 1623 struct rcu_head rcu; 1624 }; 1625 1626 /* - pseudo header */ 1627 struct tcp4_pseudohdr { 1628 __be32 saddr; 1629 __be32 daddr; 1630 __u8 pad; 1631 __u8 protocol; 1632 __be16 len; 1633 }; 1634 1635 struct tcp6_pseudohdr { 1636 struct in6_addr saddr; 1637 struct in6_addr daddr; 1638 __be32 len; 1639 __be32 protocol; /* including padding */ 1640 }; 1641 1642 union tcp_md5sum_block { 1643 struct tcp4_pseudohdr ip4; 1644 #if IS_ENABLED(CONFIG_IPV6) 1645 struct tcp6_pseudohdr ip6; 1646 #endif 1647 }; 1648 1649 /* - pool: digest algorithm, hash description and scratch buffer */ 1650 struct tcp_md5sig_pool { 1651 struct ahash_request *md5_req; 1652 void *scratch; 1653 }; 1654 1655 /* - functions */ 1656 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1657 const struct sock *sk, const struct sk_buff *skb); 1658 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1659 int family, u8 prefixlen, int l3index, u8 flags, 1660 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1661 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1662 int family, u8 prefixlen, int l3index, u8 flags); 1663 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1664 const struct sock *addr_sk); 1665 1666 #ifdef CONFIG_TCP_MD5SIG 1667 #include <linux/jump_label.h> 1668 extern struct static_key_false tcp_md5_needed; 1669 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1670 const union tcp_md5_addr *addr, 1671 int family); 1672 static inline struct tcp_md5sig_key * 1673 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1674 const union tcp_md5_addr *addr, int family) 1675 { 1676 if (!static_branch_unlikely(&tcp_md5_needed)) 1677 return NULL; 1678 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1679 } 1680 1681 enum skb_drop_reason 1682 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1683 const void *saddr, const void *daddr, 1684 int family, int dif, int sdif); 1685 1686 1687 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1688 #else 1689 static inline struct tcp_md5sig_key * 1690 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1691 const union tcp_md5_addr *addr, int family) 1692 { 1693 return NULL; 1694 } 1695 1696 static inline enum skb_drop_reason 1697 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1698 const void *saddr, const void *daddr, 1699 int family, int dif, int sdif) 1700 { 1701 return SKB_NOT_DROPPED_YET; 1702 } 1703 #define tcp_twsk_md5_key(twsk) NULL 1704 #endif 1705 1706 bool tcp_alloc_md5sig_pool(void); 1707 1708 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1709 static inline void tcp_put_md5sig_pool(void) 1710 { 1711 local_bh_enable(); 1712 } 1713 1714 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1715 unsigned int header_len); 1716 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1717 const struct tcp_md5sig_key *key); 1718 1719 /* From tcp_fastopen.c */ 1720 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1721 struct tcp_fastopen_cookie *cookie); 1722 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1723 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1724 u16 try_exp); 1725 struct tcp_fastopen_request { 1726 /* Fast Open cookie. Size 0 means a cookie request */ 1727 struct tcp_fastopen_cookie cookie; 1728 struct msghdr *data; /* data in MSG_FASTOPEN */ 1729 size_t size; 1730 int copied; /* queued in tcp_connect() */ 1731 struct ubuf_info *uarg; 1732 }; 1733 void tcp_free_fastopen_req(struct tcp_sock *tp); 1734 void tcp_fastopen_destroy_cipher(struct sock *sk); 1735 void tcp_fastopen_ctx_destroy(struct net *net); 1736 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1737 void *primary_key, void *backup_key); 1738 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1739 u64 *key); 1740 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1741 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1742 struct request_sock *req, 1743 struct tcp_fastopen_cookie *foc, 1744 const struct dst_entry *dst); 1745 void tcp_fastopen_init_key_once(struct net *net); 1746 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1747 struct tcp_fastopen_cookie *cookie); 1748 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1749 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1750 #define TCP_FASTOPEN_KEY_MAX 2 1751 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1752 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1753 1754 /* Fastopen key context */ 1755 struct tcp_fastopen_context { 1756 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1757 int num; 1758 struct rcu_head rcu; 1759 }; 1760 1761 void tcp_fastopen_active_disable(struct sock *sk); 1762 bool tcp_fastopen_active_should_disable(struct sock *sk); 1763 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1764 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1765 1766 /* Caller needs to wrap with rcu_read_(un)lock() */ 1767 static inline 1768 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1769 { 1770 struct tcp_fastopen_context *ctx; 1771 1772 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1773 if (!ctx) 1774 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1775 return ctx; 1776 } 1777 1778 static inline 1779 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1780 const struct tcp_fastopen_cookie *orig) 1781 { 1782 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1783 orig->len == foc->len && 1784 !memcmp(orig->val, foc->val, foc->len)) 1785 return true; 1786 return false; 1787 } 1788 1789 static inline 1790 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1791 { 1792 return ctx->num; 1793 } 1794 1795 /* Latencies incurred by various limits for a sender. They are 1796 * chronograph-like stats that are mutually exclusive. 1797 */ 1798 enum tcp_chrono { 1799 TCP_CHRONO_UNSPEC, 1800 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1801 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1802 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1803 __TCP_CHRONO_MAX, 1804 }; 1805 1806 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1807 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1808 1809 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1810 * the same memory storage than skb->destructor/_skb_refdst 1811 */ 1812 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1813 { 1814 skb->destructor = NULL; 1815 skb->_skb_refdst = 0UL; 1816 } 1817 1818 #define tcp_skb_tsorted_save(skb) { \ 1819 unsigned long _save = skb->_skb_refdst; \ 1820 skb->_skb_refdst = 0UL; 1821 1822 #define tcp_skb_tsorted_restore(skb) \ 1823 skb->_skb_refdst = _save; \ 1824 } 1825 1826 void tcp_write_queue_purge(struct sock *sk); 1827 1828 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1829 { 1830 return skb_rb_first(&sk->tcp_rtx_queue); 1831 } 1832 1833 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1834 { 1835 return skb_rb_last(&sk->tcp_rtx_queue); 1836 } 1837 1838 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1839 { 1840 return skb_peek_tail(&sk->sk_write_queue); 1841 } 1842 1843 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1844 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1845 1846 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1847 { 1848 return skb_peek(&sk->sk_write_queue); 1849 } 1850 1851 static inline bool tcp_skb_is_last(const struct sock *sk, 1852 const struct sk_buff *skb) 1853 { 1854 return skb_queue_is_last(&sk->sk_write_queue, skb); 1855 } 1856 1857 /** 1858 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1859 * @sk: socket 1860 * 1861 * Since the write queue can have a temporary empty skb in it, 1862 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1863 */ 1864 static inline bool tcp_write_queue_empty(const struct sock *sk) 1865 { 1866 const struct tcp_sock *tp = tcp_sk(sk); 1867 1868 return tp->write_seq == tp->snd_nxt; 1869 } 1870 1871 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1872 { 1873 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1874 } 1875 1876 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1877 { 1878 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1879 } 1880 1881 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1882 { 1883 __skb_queue_tail(&sk->sk_write_queue, skb); 1884 1885 /* Queue it, remembering where we must start sending. */ 1886 if (sk->sk_write_queue.next == skb) 1887 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1888 } 1889 1890 /* Insert new before skb on the write queue of sk. */ 1891 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1892 struct sk_buff *skb, 1893 struct sock *sk) 1894 { 1895 __skb_queue_before(&sk->sk_write_queue, skb, new); 1896 } 1897 1898 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1899 { 1900 tcp_skb_tsorted_anchor_cleanup(skb); 1901 __skb_unlink(skb, &sk->sk_write_queue); 1902 } 1903 1904 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1905 1906 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1907 { 1908 tcp_skb_tsorted_anchor_cleanup(skb); 1909 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1910 } 1911 1912 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1913 { 1914 list_del(&skb->tcp_tsorted_anchor); 1915 tcp_rtx_queue_unlink(skb, sk); 1916 tcp_wmem_free_skb(sk, skb); 1917 } 1918 1919 static inline void tcp_push_pending_frames(struct sock *sk) 1920 { 1921 if (tcp_send_head(sk)) { 1922 struct tcp_sock *tp = tcp_sk(sk); 1923 1924 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1925 } 1926 } 1927 1928 /* Start sequence of the skb just after the highest skb with SACKed 1929 * bit, valid only if sacked_out > 0 or when the caller has ensured 1930 * validity by itself. 1931 */ 1932 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1933 { 1934 if (!tp->sacked_out) 1935 return tp->snd_una; 1936 1937 if (tp->highest_sack == NULL) 1938 return tp->snd_nxt; 1939 1940 return TCP_SKB_CB(tp->highest_sack)->seq; 1941 } 1942 1943 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1944 { 1945 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1946 } 1947 1948 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1949 { 1950 return tcp_sk(sk)->highest_sack; 1951 } 1952 1953 static inline void tcp_highest_sack_reset(struct sock *sk) 1954 { 1955 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1956 } 1957 1958 /* Called when old skb is about to be deleted and replaced by new skb */ 1959 static inline void tcp_highest_sack_replace(struct sock *sk, 1960 struct sk_buff *old, 1961 struct sk_buff *new) 1962 { 1963 if (old == tcp_highest_sack(sk)) 1964 tcp_sk(sk)->highest_sack = new; 1965 } 1966 1967 /* This helper checks if socket has IP_TRANSPARENT set */ 1968 static inline bool inet_sk_transparent(const struct sock *sk) 1969 { 1970 switch (sk->sk_state) { 1971 case TCP_TIME_WAIT: 1972 return inet_twsk(sk)->tw_transparent; 1973 case TCP_NEW_SYN_RECV: 1974 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1975 } 1976 return inet_sk(sk)->transparent; 1977 } 1978 1979 /* Determines whether this is a thin stream (which may suffer from 1980 * increased latency). Used to trigger latency-reducing mechanisms. 1981 */ 1982 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1983 { 1984 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1985 } 1986 1987 /* /proc */ 1988 enum tcp_seq_states { 1989 TCP_SEQ_STATE_LISTENING, 1990 TCP_SEQ_STATE_ESTABLISHED, 1991 }; 1992 1993 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1994 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 1995 void tcp_seq_stop(struct seq_file *seq, void *v); 1996 1997 struct tcp_seq_afinfo { 1998 sa_family_t family; 1999 }; 2000 2001 struct tcp_iter_state { 2002 struct seq_net_private p; 2003 enum tcp_seq_states state; 2004 struct sock *syn_wait_sk; 2005 int bucket, offset, sbucket, num; 2006 loff_t last_pos; 2007 }; 2008 2009 extern struct request_sock_ops tcp_request_sock_ops; 2010 extern struct request_sock_ops tcp6_request_sock_ops; 2011 2012 void tcp_v4_destroy_sock(struct sock *sk); 2013 2014 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2015 netdev_features_t features); 2016 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2017 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2018 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2019 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2020 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2021 int tcp_gro_complete(struct sk_buff *skb); 2022 2023 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2024 2025 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2026 { 2027 struct net *net = sock_net((struct sock *)tp); 2028 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 2029 } 2030 2031 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2032 2033 #ifdef CONFIG_PROC_FS 2034 int tcp4_proc_init(void); 2035 void tcp4_proc_exit(void); 2036 #endif 2037 2038 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2039 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2040 const struct tcp_request_sock_ops *af_ops, 2041 struct sock *sk, struct sk_buff *skb); 2042 2043 /* TCP af-specific functions */ 2044 struct tcp_sock_af_ops { 2045 #ifdef CONFIG_TCP_MD5SIG 2046 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2047 const struct sock *addr_sk); 2048 int (*calc_md5_hash)(char *location, 2049 const struct tcp_md5sig_key *md5, 2050 const struct sock *sk, 2051 const struct sk_buff *skb); 2052 int (*md5_parse)(struct sock *sk, 2053 int optname, 2054 sockptr_t optval, 2055 int optlen); 2056 #endif 2057 }; 2058 2059 struct tcp_request_sock_ops { 2060 u16 mss_clamp; 2061 #ifdef CONFIG_TCP_MD5SIG 2062 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2063 const struct sock *addr_sk); 2064 int (*calc_md5_hash) (char *location, 2065 const struct tcp_md5sig_key *md5, 2066 const struct sock *sk, 2067 const struct sk_buff *skb); 2068 #endif 2069 #ifdef CONFIG_SYN_COOKIES 2070 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2071 __u16 *mss); 2072 #endif 2073 struct dst_entry *(*route_req)(const struct sock *sk, 2074 struct sk_buff *skb, 2075 struct flowi *fl, 2076 struct request_sock *req); 2077 u32 (*init_seq)(const struct sk_buff *skb); 2078 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2079 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2080 struct flowi *fl, struct request_sock *req, 2081 struct tcp_fastopen_cookie *foc, 2082 enum tcp_synack_type synack_type, 2083 struct sk_buff *syn_skb); 2084 }; 2085 2086 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2087 #if IS_ENABLED(CONFIG_IPV6) 2088 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2089 #endif 2090 2091 #ifdef CONFIG_SYN_COOKIES 2092 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2093 const struct sock *sk, struct sk_buff *skb, 2094 __u16 *mss) 2095 { 2096 tcp_synq_overflow(sk); 2097 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2098 return ops->cookie_init_seq(skb, mss); 2099 } 2100 #else 2101 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2102 const struct sock *sk, struct sk_buff *skb, 2103 __u16 *mss) 2104 { 2105 return 0; 2106 } 2107 #endif 2108 2109 int tcpv4_offload_init(void); 2110 2111 void tcp_v4_init(void); 2112 void tcp_init(void); 2113 2114 /* tcp_recovery.c */ 2115 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2116 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2117 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2118 u32 reo_wnd); 2119 extern bool tcp_rack_mark_lost(struct sock *sk); 2120 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2121 u64 xmit_time); 2122 extern void tcp_rack_reo_timeout(struct sock *sk); 2123 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2124 2125 /* At how many usecs into the future should the RTO fire? */ 2126 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2127 { 2128 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2129 u32 rto = inet_csk(sk)->icsk_rto; 2130 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2131 2132 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2133 } 2134 2135 /* 2136 * Save and compile IPv4 options, return a pointer to it 2137 */ 2138 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2139 struct sk_buff *skb) 2140 { 2141 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2142 struct ip_options_rcu *dopt = NULL; 2143 2144 if (opt->optlen) { 2145 int opt_size = sizeof(*dopt) + opt->optlen; 2146 2147 dopt = kmalloc(opt_size, GFP_ATOMIC); 2148 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2149 kfree(dopt); 2150 dopt = NULL; 2151 } 2152 } 2153 return dopt; 2154 } 2155 2156 /* locally generated TCP pure ACKs have skb->truesize == 2 2157 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2158 * This is much faster than dissecting the packet to find out. 2159 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2160 */ 2161 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2162 { 2163 return skb->truesize == 2; 2164 } 2165 2166 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2167 { 2168 skb->truesize = 2; 2169 } 2170 2171 static inline int tcp_inq(struct sock *sk) 2172 { 2173 struct tcp_sock *tp = tcp_sk(sk); 2174 int answ; 2175 2176 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2177 answ = 0; 2178 } else if (sock_flag(sk, SOCK_URGINLINE) || 2179 !tp->urg_data || 2180 before(tp->urg_seq, tp->copied_seq) || 2181 !before(tp->urg_seq, tp->rcv_nxt)) { 2182 2183 answ = tp->rcv_nxt - tp->copied_seq; 2184 2185 /* Subtract 1, if FIN was received */ 2186 if (answ && sock_flag(sk, SOCK_DONE)) 2187 answ--; 2188 } else { 2189 answ = tp->urg_seq - tp->copied_seq; 2190 } 2191 2192 return answ; 2193 } 2194 2195 int tcp_peek_len(struct socket *sock); 2196 2197 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2198 { 2199 u16 segs_in; 2200 2201 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2202 2203 /* We update these fields while other threads might 2204 * read them from tcp_get_info() 2205 */ 2206 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2207 if (skb->len > tcp_hdrlen(skb)) 2208 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2209 } 2210 2211 /* 2212 * TCP listen path runs lockless. 2213 * We forced "struct sock" to be const qualified to make sure 2214 * we don't modify one of its field by mistake. 2215 * Here, we increment sk_drops which is an atomic_t, so we can safely 2216 * make sock writable again. 2217 */ 2218 static inline void tcp_listendrop(const struct sock *sk) 2219 { 2220 atomic_inc(&((struct sock *)sk)->sk_drops); 2221 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2222 } 2223 2224 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2225 2226 /* 2227 * Interface for adding Upper Level Protocols over TCP 2228 */ 2229 2230 #define TCP_ULP_NAME_MAX 16 2231 #define TCP_ULP_MAX 128 2232 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2233 2234 struct tcp_ulp_ops { 2235 struct list_head list; 2236 2237 /* initialize ulp */ 2238 int (*init)(struct sock *sk); 2239 /* update ulp */ 2240 void (*update)(struct sock *sk, struct proto *p, 2241 void (*write_space)(struct sock *sk)); 2242 /* cleanup ulp */ 2243 void (*release)(struct sock *sk); 2244 /* diagnostic */ 2245 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2246 size_t (*get_info_size)(const struct sock *sk); 2247 /* clone ulp */ 2248 void (*clone)(const struct request_sock *req, struct sock *newsk, 2249 const gfp_t priority); 2250 2251 char name[TCP_ULP_NAME_MAX]; 2252 struct module *owner; 2253 }; 2254 int tcp_register_ulp(struct tcp_ulp_ops *type); 2255 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2256 int tcp_set_ulp(struct sock *sk, const char *name); 2257 void tcp_get_available_ulp(char *buf, size_t len); 2258 void tcp_cleanup_ulp(struct sock *sk); 2259 void tcp_update_ulp(struct sock *sk, struct proto *p, 2260 void (*write_space)(struct sock *sk)); 2261 2262 #define MODULE_ALIAS_TCP_ULP(name) \ 2263 __MODULE_INFO(alias, alias_userspace, name); \ 2264 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2265 2266 #ifdef CONFIG_NET_SOCK_MSG 2267 struct sk_msg; 2268 struct sk_psock; 2269 2270 #ifdef CONFIG_BPF_SYSCALL 2271 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2272 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2273 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2274 #endif /* CONFIG_BPF_SYSCALL */ 2275 2276 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2277 int flags); 2278 #endif /* CONFIG_NET_SOCK_MSG */ 2279 2280 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2281 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2282 { 2283 } 2284 #endif 2285 2286 #ifdef CONFIG_CGROUP_BPF 2287 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2288 struct sk_buff *skb, 2289 unsigned int end_offset) 2290 { 2291 skops->skb = skb; 2292 skops->skb_data_end = skb->data + end_offset; 2293 } 2294 #else 2295 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2296 struct sk_buff *skb, 2297 unsigned int end_offset) 2298 { 2299 } 2300 #endif 2301 2302 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2303 * is < 0, then the BPF op failed (for example if the loaded BPF 2304 * program does not support the chosen operation or there is no BPF 2305 * program loaded). 2306 */ 2307 #ifdef CONFIG_BPF 2308 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2309 { 2310 struct bpf_sock_ops_kern sock_ops; 2311 int ret; 2312 2313 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2314 if (sk_fullsock(sk)) { 2315 sock_ops.is_fullsock = 1; 2316 sock_owned_by_me(sk); 2317 } 2318 2319 sock_ops.sk = sk; 2320 sock_ops.op = op; 2321 if (nargs > 0) 2322 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2323 2324 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2325 if (ret == 0) 2326 ret = sock_ops.reply; 2327 else 2328 ret = -1; 2329 return ret; 2330 } 2331 2332 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2333 { 2334 u32 args[2] = {arg1, arg2}; 2335 2336 return tcp_call_bpf(sk, op, 2, args); 2337 } 2338 2339 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2340 u32 arg3) 2341 { 2342 u32 args[3] = {arg1, arg2, arg3}; 2343 2344 return tcp_call_bpf(sk, op, 3, args); 2345 } 2346 2347 #else 2348 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2349 { 2350 return -EPERM; 2351 } 2352 2353 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2354 { 2355 return -EPERM; 2356 } 2357 2358 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2359 u32 arg3) 2360 { 2361 return -EPERM; 2362 } 2363 2364 #endif 2365 2366 static inline u32 tcp_timeout_init(struct sock *sk) 2367 { 2368 int timeout; 2369 2370 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2371 2372 if (timeout <= 0) 2373 timeout = TCP_TIMEOUT_INIT; 2374 return min_t(int, timeout, TCP_RTO_MAX); 2375 } 2376 2377 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2378 { 2379 int rwnd; 2380 2381 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2382 2383 if (rwnd < 0) 2384 rwnd = 0; 2385 return rwnd; 2386 } 2387 2388 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2389 { 2390 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2391 } 2392 2393 static inline void tcp_bpf_rtt(struct sock *sk) 2394 { 2395 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2396 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2397 } 2398 2399 #if IS_ENABLED(CONFIG_SMC) 2400 extern struct static_key_false tcp_have_smc; 2401 #endif 2402 2403 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2404 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2405 void (*cad)(struct sock *sk, u32 ack_seq)); 2406 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2407 void clean_acked_data_flush(void); 2408 #endif 2409 2410 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2411 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2412 const struct tcp_sock *tp) 2413 { 2414 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2415 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2416 } 2417 2418 /* Compute Earliest Departure Time for some control packets 2419 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2420 */ 2421 static inline u64 tcp_transmit_time(const struct sock *sk) 2422 { 2423 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2424 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2425 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2426 2427 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2428 } 2429 return 0; 2430 } 2431 2432 #endif /* _TCP_H */ 2433