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 */ 144 145 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 146 147 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 148 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 149 * used as a fallback RTO for the 150 * initial data transmission if no 151 * valid RTT sample has been acquired, 152 * most likely due to retrans in 3WHS. 153 */ 154 155 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 156 * for local resources. 157 */ 158 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 159 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 160 #define TCP_KEEPALIVE_INTVL (75*HZ) 161 162 #define MAX_TCP_KEEPIDLE 32767 163 #define MAX_TCP_KEEPINTVL 32767 164 #define MAX_TCP_KEEPCNT 127 165 #define MAX_TCP_SYNCNT 127 166 167 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 168 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 169 * after this time. It should be equal 170 * (or greater than) TCP_TIMEWAIT_LEN 171 * to provide reliability equal to one 172 * provided by timewait state. 173 */ 174 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 175 * timestamps. It must be less than 176 * minimal timewait lifetime. 177 */ 178 /* 179 * TCP option 180 */ 181 182 #define TCPOPT_NOP 1 /* Padding */ 183 #define TCPOPT_EOL 0 /* End of options */ 184 #define TCPOPT_MSS 2 /* Segment size negotiating */ 185 #define TCPOPT_WINDOW 3 /* Window scaling */ 186 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 187 #define TCPOPT_SACK 5 /* SACK Block */ 188 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 189 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 190 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 191 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 192 #define TCPOPT_EXP 254 /* Experimental */ 193 /* Magic number to be after the option value for sharing TCP 194 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 195 */ 196 #define TCPOPT_FASTOPEN_MAGIC 0xF989 197 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 198 199 /* 200 * TCP option lengths 201 */ 202 203 #define TCPOLEN_MSS 4 204 #define TCPOLEN_WINDOW 3 205 #define TCPOLEN_SACK_PERM 2 206 #define TCPOLEN_TIMESTAMP 10 207 #define TCPOLEN_MD5SIG 18 208 #define TCPOLEN_FASTOPEN_BASE 2 209 #define TCPOLEN_EXP_FASTOPEN_BASE 4 210 #define TCPOLEN_EXP_SMC_BASE 6 211 212 /* But this is what stacks really send out. */ 213 #define TCPOLEN_TSTAMP_ALIGNED 12 214 #define TCPOLEN_WSCALE_ALIGNED 4 215 #define TCPOLEN_SACKPERM_ALIGNED 4 216 #define TCPOLEN_SACK_BASE 2 217 #define TCPOLEN_SACK_BASE_ALIGNED 4 218 #define TCPOLEN_SACK_PERBLOCK 8 219 #define TCPOLEN_MD5SIG_ALIGNED 20 220 #define TCPOLEN_MSS_ALIGNED 4 221 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 222 223 /* Flags in tp->nonagle */ 224 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 225 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 226 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 227 228 /* TCP thin-stream limits */ 229 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 230 231 /* TCP initial congestion window as per rfc6928 */ 232 #define TCP_INIT_CWND 10 233 234 /* Bit Flags for sysctl_tcp_fastopen */ 235 #define TFO_CLIENT_ENABLE 1 236 #define TFO_SERVER_ENABLE 2 237 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 238 239 /* Accept SYN data w/o any cookie option */ 240 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 241 242 /* Force enable TFO on all listeners, i.e., not requiring the 243 * TCP_FASTOPEN socket option. 244 */ 245 #define TFO_SERVER_WO_SOCKOPT1 0x400 246 247 248 /* sysctl variables for tcp */ 249 extern int sysctl_tcp_max_orphans; 250 extern long sysctl_tcp_mem[3]; 251 252 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 253 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 254 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 255 256 extern atomic_long_t tcp_memory_allocated; 257 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 258 259 extern struct percpu_counter tcp_sockets_allocated; 260 extern unsigned long tcp_memory_pressure; 261 262 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 263 static inline bool tcp_under_memory_pressure(const struct sock *sk) 264 { 265 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 266 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 267 return true; 268 269 return READ_ONCE(tcp_memory_pressure); 270 } 271 /* 272 * The next routines deal with comparing 32 bit unsigned ints 273 * and worry about wraparound (automatic with unsigned arithmetic). 274 */ 275 276 static inline bool before(__u32 seq1, __u32 seq2) 277 { 278 return (__s32)(seq1-seq2) < 0; 279 } 280 #define after(seq2, seq1) before(seq1, seq2) 281 282 /* is s2<=s1<=s3 ? */ 283 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 284 { 285 return seq3 - seq2 >= seq1 - seq2; 286 } 287 288 static inline bool tcp_out_of_memory(struct sock *sk) 289 { 290 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 291 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 292 return true; 293 return false; 294 } 295 296 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 297 { 298 sk_wmem_queued_add(sk, -skb->truesize); 299 if (!skb_zcopy_pure(skb)) 300 sk_mem_uncharge(sk, skb->truesize); 301 else 302 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 303 __kfree_skb(skb); 304 } 305 306 void sk_forced_mem_schedule(struct sock *sk, int size); 307 308 bool tcp_check_oom(struct sock *sk, int shift); 309 310 311 extern struct proto tcp_prot; 312 313 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 314 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 315 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 316 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 317 318 void tcp_tasklet_init(void); 319 320 int tcp_v4_err(struct sk_buff *skb, u32); 321 322 void tcp_shutdown(struct sock *sk, int how); 323 324 int tcp_v4_early_demux(struct sk_buff *skb); 325 int tcp_v4_rcv(struct sk_buff *skb); 326 327 void tcp_remove_empty_skb(struct sock *sk); 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_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 331 size_t size, struct ubuf_info *uarg); 332 void tcp_splice_eof(struct socket *sock); 333 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 334 int tcp_wmem_schedule(struct sock *sk, int copy); 335 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 336 int size_goal); 337 void tcp_release_cb(struct sock *sk); 338 void tcp_wfree(struct sk_buff *skb); 339 void tcp_write_timer_handler(struct sock *sk); 340 void tcp_delack_timer_handler(struct sock *sk); 341 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 342 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 344 void tcp_rcv_space_adjust(struct sock *sk); 345 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 346 void tcp_twsk_destructor(struct sock *sk); 347 void tcp_twsk_purge(struct list_head *net_exit_list); 348 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 349 struct pipe_inode_info *pipe, size_t len, 350 unsigned int flags); 351 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 352 bool force_schedule); 353 354 static inline void tcp_dec_quickack_mode(struct sock *sk) 355 { 356 struct inet_connection_sock *icsk = inet_csk(sk); 357 358 if (icsk->icsk_ack.quick) { 359 /* How many ACKs S/ACKing new data have we sent? */ 360 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 361 362 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 do_tcp_getsockopt(struct sock *sk, int level, 406 int optname, sockptr_t optval, sockptr_t optlen); 407 int tcp_getsockopt(struct sock *sk, int level, int optname, 408 char __user *optval, int __user *optlen); 409 bool tcp_bpf_bypass_getsockopt(int level, int optname); 410 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 411 sockptr_t optval, unsigned int optlen); 412 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 413 unsigned int optlen); 414 void tcp_set_keepalive(struct sock *sk, int val); 415 void tcp_syn_ack_timeout(const struct request_sock *req); 416 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 417 int flags, int *addr_len); 418 int tcp_set_rcvlowat(struct sock *sk, int val); 419 int tcp_set_window_clamp(struct sock *sk, int val); 420 void tcp_update_recv_tstamps(struct sk_buff *skb, 421 struct scm_timestamping_internal *tss); 422 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 423 struct scm_timestamping_internal *tss); 424 void tcp_data_ready(struct sock *sk); 425 #ifdef CONFIG_MMU 426 int tcp_mmap(struct file *file, struct socket *sock, 427 struct vm_area_struct *vma); 428 #endif 429 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 430 struct tcp_options_received *opt_rx, 431 int estab, struct tcp_fastopen_cookie *foc); 432 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 433 434 /* 435 * BPF SKB-less helpers 436 */ 437 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 438 struct tcphdr *th, u32 *cookie); 439 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 440 struct tcphdr *th, u32 *cookie); 441 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 442 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 443 const struct tcp_request_sock_ops *af_ops, 444 struct sock *sk, struct tcphdr *th); 445 /* 446 * TCP v4 functions exported for the inet6 API 447 */ 448 449 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 450 void tcp_v4_mtu_reduced(struct sock *sk); 451 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 452 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 453 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 454 struct sock *tcp_create_openreq_child(const struct sock *sk, 455 struct request_sock *req, 456 struct sk_buff *skb); 457 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 458 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 459 struct request_sock *req, 460 struct dst_entry *dst, 461 struct request_sock *req_unhash, 462 bool *own_req); 463 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 464 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 465 int tcp_connect(struct sock *sk); 466 enum tcp_synack_type { 467 TCP_SYNACK_NORMAL, 468 TCP_SYNACK_FASTOPEN, 469 TCP_SYNACK_COOKIE, 470 }; 471 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 472 struct request_sock *req, 473 struct tcp_fastopen_cookie *foc, 474 enum tcp_synack_type synack_type, 475 struct sk_buff *syn_skb); 476 int tcp_disconnect(struct sock *sk, int flags); 477 478 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 479 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 480 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 481 482 /* From syncookies.c */ 483 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 484 struct request_sock *req, 485 struct dst_entry *dst, u32 tsoff); 486 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 487 u32 cookie); 488 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 489 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 490 const struct tcp_request_sock_ops *af_ops, 491 struct sock *sk, struct sk_buff *skb); 492 #ifdef CONFIG_SYN_COOKIES 493 494 /* Syncookies use a monotonic timer which increments every 60 seconds. 495 * This counter is used both as a hash input and partially encoded into 496 * the cookie value. A cookie is only validated further if the delta 497 * between the current counter value and the encoded one is less than this, 498 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 499 * the counter advances immediately after a cookie is generated). 500 */ 501 #define MAX_SYNCOOKIE_AGE 2 502 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 503 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 504 505 /* syncookies: remember time of last synqueue overflow 506 * But do not dirty this field too often (once per second is enough) 507 * It is racy as we do not hold a lock, but race is very minor. 508 */ 509 static inline void tcp_synq_overflow(const struct sock *sk) 510 { 511 unsigned int last_overflow; 512 unsigned int now = jiffies; 513 514 if (sk->sk_reuseport) { 515 struct sock_reuseport *reuse; 516 517 reuse = rcu_dereference(sk->sk_reuseport_cb); 518 if (likely(reuse)) { 519 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 520 if (!time_between32(now, last_overflow, 521 last_overflow + HZ)) 522 WRITE_ONCE(reuse->synq_overflow_ts, now); 523 return; 524 } 525 } 526 527 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 528 if (!time_between32(now, last_overflow, last_overflow + HZ)) 529 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 530 } 531 532 /* syncookies: no recent synqueue overflow on this listening socket? */ 533 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 534 { 535 unsigned int last_overflow; 536 unsigned int now = jiffies; 537 538 if (sk->sk_reuseport) { 539 struct sock_reuseport *reuse; 540 541 reuse = rcu_dereference(sk->sk_reuseport_cb); 542 if (likely(reuse)) { 543 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 544 return !time_between32(now, last_overflow - HZ, 545 last_overflow + 546 TCP_SYNCOOKIE_VALID); 547 } 548 } 549 550 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 551 552 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 553 * then we're under synflood. However, we have to use 554 * 'last_overflow - HZ' as lower bound. That's because a concurrent 555 * tcp_synq_overflow() could update .ts_recent_stamp after we read 556 * jiffies but before we store .ts_recent_stamp into last_overflow, 557 * which could lead to rejecting a valid syncookie. 558 */ 559 return !time_between32(now, last_overflow - HZ, 560 last_overflow + TCP_SYNCOOKIE_VALID); 561 } 562 563 static inline u32 tcp_cookie_time(void) 564 { 565 u64 val = get_jiffies_64(); 566 567 do_div(val, TCP_SYNCOOKIE_PERIOD); 568 return val; 569 } 570 571 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 572 u16 *mssp); 573 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 574 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 575 bool cookie_timestamp_decode(const struct net *net, 576 struct tcp_options_received *opt); 577 bool cookie_ecn_ok(const struct tcp_options_received *opt, 578 const struct net *net, const struct dst_entry *dst); 579 580 /* From net/ipv6/syncookies.c */ 581 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 582 u32 cookie); 583 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 584 585 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 586 const struct tcphdr *th, u16 *mssp); 587 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 588 #endif 589 /* tcp_output.c */ 590 591 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 592 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 593 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 594 int nonagle); 595 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 596 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 597 void tcp_retransmit_timer(struct sock *sk); 598 void tcp_xmit_retransmit_queue(struct sock *); 599 void tcp_simple_retransmit(struct sock *); 600 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 601 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 602 enum tcp_queue { 603 TCP_FRAG_IN_WRITE_QUEUE, 604 TCP_FRAG_IN_RTX_QUEUE, 605 }; 606 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 607 struct sk_buff *skb, u32 len, 608 unsigned int mss_now, gfp_t gfp); 609 610 void tcp_send_probe0(struct sock *); 611 int tcp_write_wakeup(struct sock *, int mib); 612 void tcp_send_fin(struct sock *sk); 613 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 614 int tcp_send_synack(struct sock *); 615 void tcp_push_one(struct sock *, unsigned int mss_now); 616 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 617 void tcp_send_ack(struct sock *sk); 618 void tcp_send_delayed_ack(struct sock *sk); 619 void tcp_send_loss_probe(struct sock *sk); 620 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 621 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 622 const struct sk_buff *next_skb); 623 624 /* tcp_input.c */ 625 void tcp_rearm_rto(struct sock *sk); 626 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 627 void tcp_done_with_error(struct sock *sk, int err); 628 void tcp_reset(struct sock *sk, struct sk_buff *skb); 629 void tcp_fin(struct sock *sk); 630 void tcp_check_space(struct sock *sk); 631 void tcp_sack_compress_send_ack(struct sock *sk); 632 633 /* tcp_timer.c */ 634 void tcp_init_xmit_timers(struct sock *); 635 static inline void tcp_clear_xmit_timers(struct sock *sk) 636 { 637 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 638 __sock_put(sk); 639 640 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 641 __sock_put(sk); 642 643 inet_csk_clear_xmit_timers(sk); 644 } 645 646 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 647 unsigned int tcp_current_mss(struct sock *sk); 648 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 649 650 /* Bound MSS / TSO packet size with the half of the window */ 651 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 652 { 653 int cutoff; 654 655 /* When peer uses tiny windows, there is no use in packetizing 656 * to sub-MSS pieces for the sake of SWS or making sure there 657 * are enough packets in the pipe for fast recovery. 658 * 659 * On the other hand, for extremely large MSS devices, handling 660 * smaller than MSS windows in this way does make sense. 661 */ 662 if (tp->max_window > TCP_MSS_DEFAULT) 663 cutoff = (tp->max_window >> 1); 664 else 665 cutoff = tp->max_window; 666 667 if (cutoff && pktsize > cutoff) 668 return max_t(int, cutoff, 68U - tp->tcp_header_len); 669 else 670 return pktsize; 671 } 672 673 /* tcp.c */ 674 void tcp_get_info(struct sock *, struct tcp_info *); 675 676 /* Read 'sendfile()'-style from a TCP socket */ 677 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 678 sk_read_actor_t recv_actor); 679 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 680 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 681 void tcp_read_done(struct sock *sk, size_t len); 682 683 void tcp_initialize_rcv_mss(struct sock *sk); 684 685 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 686 int tcp_mss_to_mtu(struct sock *sk, int mss); 687 void tcp_mtup_init(struct sock *sk); 688 689 static inline void tcp_bound_rto(const struct sock *sk) 690 { 691 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 692 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 693 } 694 695 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 696 { 697 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 698 } 699 700 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 701 { 702 /* mptcp hooks are only on the slow path */ 703 if (sk_is_mptcp((struct sock *)tp)) 704 return; 705 706 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 707 ntohl(TCP_FLAG_ACK) | 708 snd_wnd); 709 } 710 711 static inline void tcp_fast_path_on(struct tcp_sock *tp) 712 { 713 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 714 } 715 716 static inline void tcp_fast_path_check(struct sock *sk) 717 { 718 struct tcp_sock *tp = tcp_sk(sk); 719 720 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 721 tp->rcv_wnd && 722 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 723 !tp->urg_data) 724 tcp_fast_path_on(tp); 725 } 726 727 u32 tcp_delack_max(const struct sock *sk); 728 729 /* Compute the actual rto_min value */ 730 static inline u32 tcp_rto_min(struct sock *sk) 731 { 732 const struct dst_entry *dst = __sk_dst_get(sk); 733 u32 rto_min = inet_csk(sk)->icsk_rto_min; 734 735 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 736 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 737 return rto_min; 738 } 739 740 static inline u32 tcp_rto_min_us(struct sock *sk) 741 { 742 return jiffies_to_usecs(tcp_rto_min(sk)); 743 } 744 745 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 746 { 747 return dst_metric_locked(dst, RTAX_CC_ALGO); 748 } 749 750 /* Minimum RTT in usec. ~0 means not available. */ 751 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 752 { 753 return minmax_get(&tp->rtt_min); 754 } 755 756 /* Compute the actual receive window we are currently advertising. 757 * Rcv_nxt can be after the window if our peer push more data 758 * than the offered window. 759 */ 760 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 761 { 762 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 763 764 if (win < 0) 765 win = 0; 766 return (u32) win; 767 } 768 769 /* Choose a new window, without checks for shrinking, and without 770 * scaling applied to the result. The caller does these things 771 * if necessary. This is a "raw" window selection. 772 */ 773 u32 __tcp_select_window(struct sock *sk); 774 775 void tcp_send_window_probe(struct sock *sk); 776 777 /* TCP uses 32bit jiffies to save some space. 778 * Note that this is different from tcp_time_stamp, which 779 * historically has been the same until linux-4.13. 780 */ 781 #define tcp_jiffies32 ((u32)jiffies) 782 783 /* 784 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 785 * It is no longer tied to jiffies, but to 1 ms clock. 786 * Note: double check if you want to use tcp_jiffies32 instead of this. 787 */ 788 #define TCP_TS_HZ 1000 789 790 static inline u64 tcp_clock_ns(void) 791 { 792 return ktime_get_ns(); 793 } 794 795 static inline u64 tcp_clock_us(void) 796 { 797 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 798 } 799 800 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 801 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 802 { 803 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 804 } 805 806 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 807 static inline u64 tcp_ns_to_ts(u64 ns) 808 { 809 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 810 } 811 812 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 813 static inline u32 tcp_time_stamp_raw(void) 814 { 815 return tcp_ns_to_ts(tcp_clock_ns()); 816 } 817 818 void tcp_mstamp_refresh(struct tcp_sock *tp); 819 820 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 821 { 822 return max_t(s64, t1 - t0, 0); 823 } 824 825 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 826 { 827 return tcp_ns_to_ts(skb->skb_mstamp_ns); 828 } 829 830 /* provide the departure time in us unit */ 831 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 832 { 833 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 834 } 835 836 837 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 838 839 #define TCPHDR_FIN 0x01 840 #define TCPHDR_SYN 0x02 841 #define TCPHDR_RST 0x04 842 #define TCPHDR_PSH 0x08 843 #define TCPHDR_ACK 0x10 844 #define TCPHDR_URG 0x20 845 #define TCPHDR_ECE 0x40 846 #define TCPHDR_CWR 0x80 847 848 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 849 850 /* This is what the send packet queuing engine uses to pass 851 * TCP per-packet control information to the transmission code. 852 * We also store the host-order sequence numbers in here too. 853 * This is 44 bytes if IPV6 is enabled. 854 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 855 */ 856 struct tcp_skb_cb { 857 __u32 seq; /* Starting sequence number */ 858 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 859 union { 860 /* Note : tcp_tw_isn is used in input path only 861 * (isn chosen by tcp_timewait_state_process()) 862 * 863 * tcp_gso_segs/size are used in write queue only, 864 * cf tcp_skb_pcount()/tcp_skb_mss() 865 */ 866 __u32 tcp_tw_isn; 867 struct { 868 u16 tcp_gso_segs; 869 u16 tcp_gso_size; 870 }; 871 }; 872 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 873 874 __u8 sacked; /* State flags for SACK. */ 875 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 876 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 877 #define TCPCB_LOST 0x04 /* SKB is lost */ 878 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 879 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 880 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 881 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 882 TCPCB_REPAIRED) 883 884 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 885 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 886 eor:1, /* Is skb MSG_EOR marked? */ 887 has_rxtstamp:1, /* SKB has a RX timestamp */ 888 unused:5; 889 __u32 ack_seq; /* Sequence number ACK'd */ 890 union { 891 struct { 892 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 893 /* There is space for up to 24 bytes */ 894 __u32 is_app_limited:1, /* cwnd not fully used? */ 895 delivered_ce:20, 896 unused:11; 897 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 898 __u32 delivered; 899 /* start of send pipeline phase */ 900 u64 first_tx_mstamp; 901 /* when we reached the "delivered" count */ 902 u64 delivered_mstamp; 903 } tx; /* only used for outgoing skbs */ 904 union { 905 struct inet_skb_parm h4; 906 #if IS_ENABLED(CONFIG_IPV6) 907 struct inet6_skb_parm h6; 908 #endif 909 } header; /* For incoming skbs */ 910 }; 911 }; 912 913 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 914 915 extern const struct inet_connection_sock_af_ops ipv4_specific; 916 917 #if IS_ENABLED(CONFIG_IPV6) 918 /* This is the variant of inet6_iif() that must be used by TCP, 919 * as TCP moves IP6CB into a different location in skb->cb[] 920 */ 921 static inline int tcp_v6_iif(const struct sk_buff *skb) 922 { 923 return TCP_SKB_CB(skb)->header.h6.iif; 924 } 925 926 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 927 { 928 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 929 930 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 931 } 932 933 /* TCP_SKB_CB reference means this can not be used from early demux */ 934 static inline int tcp_v6_sdif(const struct sk_buff *skb) 935 { 936 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 937 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 938 return TCP_SKB_CB(skb)->header.h6.iif; 939 #endif 940 return 0; 941 } 942 943 extern const struct inet_connection_sock_af_ops ipv6_specific; 944 945 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 946 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 947 void tcp_v6_early_demux(struct sk_buff *skb); 948 949 #endif 950 951 /* TCP_SKB_CB reference means this can not be used from early demux */ 952 static inline int tcp_v4_sdif(struct sk_buff *skb) 953 { 954 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 955 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 956 return TCP_SKB_CB(skb)->header.h4.iif; 957 #endif 958 return 0; 959 } 960 961 /* Due to TSO, an SKB can be composed of multiple actual 962 * packets. To keep these tracked properly, we use this. 963 */ 964 static inline int tcp_skb_pcount(const struct sk_buff *skb) 965 { 966 return TCP_SKB_CB(skb)->tcp_gso_segs; 967 } 968 969 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 970 { 971 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 972 } 973 974 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 975 { 976 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 977 } 978 979 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 980 static inline int tcp_skb_mss(const struct sk_buff *skb) 981 { 982 return TCP_SKB_CB(skb)->tcp_gso_size; 983 } 984 985 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 986 { 987 return likely(!TCP_SKB_CB(skb)->eor); 988 } 989 990 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 991 const struct sk_buff *from) 992 { 993 return likely(tcp_skb_can_collapse_to(to) && 994 mptcp_skb_can_collapse(to, from) && 995 skb_pure_zcopy_same(to, from)); 996 } 997 998 /* Events passed to congestion control interface */ 999 enum tcp_ca_event { 1000 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1001 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1002 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1003 CA_EVENT_LOSS, /* loss timeout */ 1004 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1005 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1006 }; 1007 1008 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1009 enum tcp_ca_ack_event_flags { 1010 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1011 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1012 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1013 }; 1014 1015 /* 1016 * Interface for adding new TCP congestion control handlers 1017 */ 1018 #define TCP_CA_NAME_MAX 16 1019 #define TCP_CA_MAX 128 1020 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1021 1022 #define TCP_CA_UNSPEC 0 1023 1024 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1025 #define TCP_CONG_NON_RESTRICTED 0x1 1026 /* Requires ECN/ECT set on all packets */ 1027 #define TCP_CONG_NEEDS_ECN 0x2 1028 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1029 1030 union tcp_cc_info; 1031 1032 struct ack_sample { 1033 u32 pkts_acked; 1034 s32 rtt_us; 1035 u32 in_flight; 1036 }; 1037 1038 /* A rate sample measures the number of (original/retransmitted) data 1039 * packets delivered "delivered" over an interval of time "interval_us". 1040 * The tcp_rate.c code fills in the rate sample, and congestion 1041 * control modules that define a cong_control function to run at the end 1042 * of ACK processing can optionally chose to consult this sample when 1043 * setting cwnd and pacing rate. 1044 * A sample is invalid if "delivered" or "interval_us" is negative. 1045 */ 1046 struct rate_sample { 1047 u64 prior_mstamp; /* starting timestamp for interval */ 1048 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1049 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1050 s32 delivered; /* number of packets delivered over interval */ 1051 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1052 long interval_us; /* time for tp->delivered to incr "delivered" */ 1053 u32 snd_interval_us; /* snd interval for delivered packets */ 1054 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1055 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1056 int losses; /* number of packets marked lost upon ACK */ 1057 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1058 u32 prior_in_flight; /* in flight before this ACK */ 1059 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1060 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1061 bool is_retrans; /* is sample from retransmission? */ 1062 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1063 }; 1064 1065 struct tcp_congestion_ops { 1066 /* fast path fields are put first to fill one cache line */ 1067 1068 /* return slow start threshold (required) */ 1069 u32 (*ssthresh)(struct sock *sk); 1070 1071 /* do new cwnd calculation (required) */ 1072 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1073 1074 /* call before changing ca_state (optional) */ 1075 void (*set_state)(struct sock *sk, u8 new_state); 1076 1077 /* call when cwnd event occurs (optional) */ 1078 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1079 1080 /* call when ack arrives (optional) */ 1081 void (*in_ack_event)(struct sock *sk, u32 flags); 1082 1083 /* hook for packet ack accounting (optional) */ 1084 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1085 1086 /* override sysctl_tcp_min_tso_segs */ 1087 u32 (*min_tso_segs)(struct sock *sk); 1088 1089 /* call when packets are delivered to update cwnd and pacing rate, 1090 * after all the ca_state processing. (optional) 1091 */ 1092 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1093 1094 1095 /* new value of cwnd after loss (required) */ 1096 u32 (*undo_cwnd)(struct sock *sk); 1097 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1098 u32 (*sndbuf_expand)(struct sock *sk); 1099 1100 /* control/slow paths put last */ 1101 /* get info for inet_diag (optional) */ 1102 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1103 union tcp_cc_info *info); 1104 1105 char name[TCP_CA_NAME_MAX]; 1106 struct module *owner; 1107 struct list_head list; 1108 u32 key; 1109 u32 flags; 1110 1111 /* initialize private data (optional) */ 1112 void (*init)(struct sock *sk); 1113 /* cleanup private data (optional) */ 1114 void (*release)(struct sock *sk); 1115 } ____cacheline_aligned_in_smp; 1116 1117 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1118 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1119 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1120 struct tcp_congestion_ops *old_type); 1121 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1122 1123 void tcp_assign_congestion_control(struct sock *sk); 1124 void tcp_init_congestion_control(struct sock *sk); 1125 void tcp_cleanup_congestion_control(struct sock *sk); 1126 int tcp_set_default_congestion_control(struct net *net, const char *name); 1127 void tcp_get_default_congestion_control(struct net *net, char *name); 1128 void tcp_get_available_congestion_control(char *buf, size_t len); 1129 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1130 int tcp_set_allowed_congestion_control(char *allowed); 1131 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1132 bool cap_net_admin); 1133 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1134 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1135 1136 u32 tcp_reno_ssthresh(struct sock *sk); 1137 u32 tcp_reno_undo_cwnd(struct sock *sk); 1138 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1139 extern struct tcp_congestion_ops tcp_reno; 1140 1141 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1142 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1143 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1144 #ifdef CONFIG_INET 1145 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1146 #else 1147 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1148 { 1149 return NULL; 1150 } 1151 #endif 1152 1153 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1154 { 1155 const struct inet_connection_sock *icsk = inet_csk(sk); 1156 1157 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1158 } 1159 1160 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1161 { 1162 const struct inet_connection_sock *icsk = inet_csk(sk); 1163 1164 if (icsk->icsk_ca_ops->cwnd_event) 1165 icsk->icsk_ca_ops->cwnd_event(sk, event); 1166 } 1167 1168 /* From tcp_cong.c */ 1169 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1170 1171 /* From tcp_rate.c */ 1172 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1173 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1174 struct rate_sample *rs); 1175 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1176 bool is_sack_reneg, struct rate_sample *rs); 1177 void tcp_rate_check_app_limited(struct sock *sk); 1178 1179 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1180 { 1181 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1182 } 1183 1184 /* These functions determine how the current flow behaves in respect of SACK 1185 * handling. SACK is negotiated with the peer, and therefore it can vary 1186 * between different flows. 1187 * 1188 * tcp_is_sack - SACK enabled 1189 * tcp_is_reno - No SACK 1190 */ 1191 static inline int tcp_is_sack(const struct tcp_sock *tp) 1192 { 1193 return likely(tp->rx_opt.sack_ok); 1194 } 1195 1196 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1197 { 1198 return !tcp_is_sack(tp); 1199 } 1200 1201 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1202 { 1203 return tp->sacked_out + tp->lost_out; 1204 } 1205 1206 /* This determines how many packets are "in the network" to the best 1207 * of our knowledge. In many cases it is conservative, but where 1208 * detailed information is available from the receiver (via SACK 1209 * blocks etc.) we can make more aggressive calculations. 1210 * 1211 * Use this for decisions involving congestion control, use just 1212 * tp->packets_out to determine if the send queue is empty or not. 1213 * 1214 * Read this equation as: 1215 * 1216 * "Packets sent once on transmission queue" MINUS 1217 * "Packets left network, but not honestly ACKed yet" PLUS 1218 * "Packets fast retransmitted" 1219 */ 1220 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1221 { 1222 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1223 } 1224 1225 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1226 1227 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1228 { 1229 return tp->snd_cwnd; 1230 } 1231 1232 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1233 { 1234 WARN_ON_ONCE((int)val <= 0); 1235 tp->snd_cwnd = val; 1236 } 1237 1238 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1239 { 1240 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1241 } 1242 1243 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1244 { 1245 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1246 } 1247 1248 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1249 { 1250 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1251 (1 << inet_csk(sk)->icsk_ca_state); 1252 } 1253 1254 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1255 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1256 * ssthresh. 1257 */ 1258 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1259 { 1260 const struct tcp_sock *tp = tcp_sk(sk); 1261 1262 if (tcp_in_cwnd_reduction(sk)) 1263 return tp->snd_ssthresh; 1264 else 1265 return max(tp->snd_ssthresh, 1266 ((tcp_snd_cwnd(tp) >> 1) + 1267 (tcp_snd_cwnd(tp) >> 2))); 1268 } 1269 1270 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1271 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1272 1273 void tcp_enter_cwr(struct sock *sk); 1274 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1275 1276 /* The maximum number of MSS of available cwnd for which TSO defers 1277 * sending if not using sysctl_tcp_tso_win_divisor. 1278 */ 1279 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1280 { 1281 return 3; 1282 } 1283 1284 /* Returns end sequence number of the receiver's advertised window */ 1285 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1286 { 1287 return tp->snd_una + tp->snd_wnd; 1288 } 1289 1290 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1291 * flexible approach. The RFC suggests cwnd should not be raised unless 1292 * it was fully used previously. And that's exactly what we do in 1293 * congestion avoidance mode. But in slow start we allow cwnd to grow 1294 * as long as the application has used half the cwnd. 1295 * Example : 1296 * cwnd is 10 (IW10), but application sends 9 frames. 1297 * We allow cwnd to reach 18 when all frames are ACKed. 1298 * This check is safe because it's as aggressive as slow start which already 1299 * risks 100% overshoot. The advantage is that we discourage application to 1300 * either send more filler packets or data to artificially blow up the cwnd 1301 * usage, and allow application-limited process to probe bw more aggressively. 1302 */ 1303 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1304 { 1305 const struct tcp_sock *tp = tcp_sk(sk); 1306 1307 if (tp->is_cwnd_limited) 1308 return true; 1309 1310 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1311 if (tcp_in_slow_start(tp)) 1312 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1313 1314 return false; 1315 } 1316 1317 /* BBR congestion control needs pacing. 1318 * Same remark for SO_MAX_PACING_RATE. 1319 * sch_fq packet scheduler is efficiently handling pacing, 1320 * but is not always installed/used. 1321 * Return true if TCP stack should pace packets itself. 1322 */ 1323 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1324 { 1325 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1326 } 1327 1328 /* Estimates in how many jiffies next packet for this flow can be sent. 1329 * Scheduling a retransmit timer too early would be silly. 1330 */ 1331 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1332 { 1333 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1334 1335 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1336 } 1337 1338 static inline void tcp_reset_xmit_timer(struct sock *sk, 1339 const int what, 1340 unsigned long when, 1341 const unsigned long max_when) 1342 { 1343 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1344 max_when); 1345 } 1346 1347 /* Something is really bad, we could not queue an additional packet, 1348 * because qdisc is full or receiver sent a 0 window, or we are paced. 1349 * We do not want to add fuel to the fire, or abort too early, 1350 * so make sure the timer we arm now is at least 200ms in the future, 1351 * regardless of current icsk_rto value (as it could be ~2ms) 1352 */ 1353 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1354 { 1355 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1356 } 1357 1358 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1359 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1360 unsigned long max_when) 1361 { 1362 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1363 inet_csk(sk)->icsk_backoff); 1364 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1365 1366 return (unsigned long)min_t(u64, when, max_when); 1367 } 1368 1369 static inline void tcp_check_probe_timer(struct sock *sk) 1370 { 1371 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1372 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1373 tcp_probe0_base(sk), TCP_RTO_MAX); 1374 } 1375 1376 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1377 { 1378 tp->snd_wl1 = seq; 1379 } 1380 1381 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1382 { 1383 tp->snd_wl1 = seq; 1384 } 1385 1386 /* 1387 * Calculate(/check) TCP checksum 1388 */ 1389 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1390 __be32 daddr, __wsum base) 1391 { 1392 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1393 } 1394 1395 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1396 { 1397 return !skb_csum_unnecessary(skb) && 1398 __skb_checksum_complete(skb); 1399 } 1400 1401 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1402 enum skb_drop_reason *reason); 1403 1404 1405 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1406 void tcp_set_state(struct sock *sk, int state); 1407 void tcp_done(struct sock *sk); 1408 int tcp_abort(struct sock *sk, int err); 1409 1410 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1411 { 1412 rx_opt->dsack = 0; 1413 rx_opt->num_sacks = 0; 1414 } 1415 1416 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1417 1418 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1419 { 1420 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1421 struct tcp_sock *tp = tcp_sk(sk); 1422 s32 delta; 1423 1424 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1425 tp->packets_out || ca_ops->cong_control) 1426 return; 1427 delta = tcp_jiffies32 - tp->lsndtime; 1428 if (delta > inet_csk(sk)->icsk_rto) 1429 tcp_cwnd_restart(sk, delta); 1430 } 1431 1432 /* Determine a window scaling and initial window to offer. */ 1433 void tcp_select_initial_window(const struct sock *sk, int __space, 1434 __u32 mss, __u32 *rcv_wnd, 1435 __u32 *window_clamp, int wscale_ok, 1436 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1437 1438 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1439 { 1440 s64 scaled_space = (s64)space * scaling_ratio; 1441 1442 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1443 } 1444 1445 static inline int tcp_win_from_space(const struct sock *sk, int space) 1446 { 1447 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1448 } 1449 1450 /* inverse of __tcp_win_from_space() */ 1451 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1452 { 1453 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1454 1455 do_div(val, scaling_ratio); 1456 return val; 1457 } 1458 1459 static inline int tcp_space_from_win(const struct sock *sk, int win) 1460 { 1461 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1462 } 1463 1464 /* Assume a 50% default for skb->len/skb->truesize ratio. 1465 * This may be adjusted later in tcp_measure_rcv_mss(). 1466 */ 1467 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1)) 1468 1469 static inline void tcp_scaling_ratio_init(struct sock *sk) 1470 { 1471 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1472 } 1473 1474 /* Note: caller must be prepared to deal with negative returns */ 1475 static inline int tcp_space(const struct sock *sk) 1476 { 1477 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1478 READ_ONCE(sk->sk_backlog.len) - 1479 atomic_read(&sk->sk_rmem_alloc)); 1480 } 1481 1482 static inline int tcp_full_space(const struct sock *sk) 1483 { 1484 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1485 } 1486 1487 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh) 1488 { 1489 int unused_mem = sk_unused_reserved_mem(sk); 1490 struct tcp_sock *tp = tcp_sk(sk); 1491 1492 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh); 1493 if (unused_mem) 1494 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1495 tcp_win_from_space(sk, unused_mem)); 1496 } 1497 1498 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1499 { 1500 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss); 1501 } 1502 1503 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1504 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1505 1506 1507 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1508 * If 87.5 % (7/8) of the space has been consumed, we want to override 1509 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1510 * len/truesize ratio. 1511 */ 1512 static inline bool tcp_rmem_pressure(const struct sock *sk) 1513 { 1514 int rcvbuf, threshold; 1515 1516 if (tcp_under_memory_pressure(sk)) 1517 return true; 1518 1519 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1520 threshold = rcvbuf - (rcvbuf >> 3); 1521 1522 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1523 } 1524 1525 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1526 { 1527 const struct tcp_sock *tp = tcp_sk(sk); 1528 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1529 1530 if (avail <= 0) 1531 return false; 1532 1533 return (avail >= target) || tcp_rmem_pressure(sk) || 1534 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1535 } 1536 1537 extern void tcp_openreq_init_rwin(struct request_sock *req, 1538 const struct sock *sk_listener, 1539 const struct dst_entry *dst); 1540 1541 void tcp_enter_memory_pressure(struct sock *sk); 1542 void tcp_leave_memory_pressure(struct sock *sk); 1543 1544 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1545 { 1546 struct net *net = sock_net((struct sock *)tp); 1547 int val; 1548 1549 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1550 * and do_tcp_setsockopt(). 1551 */ 1552 val = READ_ONCE(tp->keepalive_intvl); 1553 1554 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1555 } 1556 1557 static inline int keepalive_time_when(const struct tcp_sock *tp) 1558 { 1559 struct net *net = sock_net((struct sock *)tp); 1560 int val; 1561 1562 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1563 val = READ_ONCE(tp->keepalive_time); 1564 1565 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1566 } 1567 1568 static inline int keepalive_probes(const struct tcp_sock *tp) 1569 { 1570 struct net *net = sock_net((struct sock *)tp); 1571 int val; 1572 1573 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1574 * and do_tcp_setsockopt(). 1575 */ 1576 val = READ_ONCE(tp->keepalive_probes); 1577 1578 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1579 } 1580 1581 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1582 { 1583 const struct inet_connection_sock *icsk = &tp->inet_conn; 1584 1585 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1586 tcp_jiffies32 - tp->rcv_tstamp); 1587 } 1588 1589 static inline int tcp_fin_time(const struct sock *sk) 1590 { 1591 int fin_timeout = tcp_sk(sk)->linger2 ? : 1592 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1593 const int rto = inet_csk(sk)->icsk_rto; 1594 1595 if (fin_timeout < (rto << 2) - (rto >> 1)) 1596 fin_timeout = (rto << 2) - (rto >> 1); 1597 1598 return fin_timeout; 1599 } 1600 1601 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1602 int paws_win) 1603 { 1604 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1605 return true; 1606 if (unlikely(!time_before32(ktime_get_seconds(), 1607 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1608 return true; 1609 /* 1610 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1611 * then following tcp messages have valid values. Ignore 0 value, 1612 * or else 'negative' tsval might forbid us to accept their packets. 1613 */ 1614 if (!rx_opt->ts_recent) 1615 return true; 1616 return false; 1617 } 1618 1619 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1620 int rst) 1621 { 1622 if (tcp_paws_check(rx_opt, 0)) 1623 return false; 1624 1625 /* RST segments are not recommended to carry timestamp, 1626 and, if they do, it is recommended to ignore PAWS because 1627 "their cleanup function should take precedence over timestamps." 1628 Certainly, it is mistake. It is necessary to understand the reasons 1629 of this constraint to relax it: if peer reboots, clock may go 1630 out-of-sync and half-open connections will not be reset. 1631 Actually, the problem would be not existing if all 1632 the implementations followed draft about maintaining clock 1633 via reboots. Linux-2.2 DOES NOT! 1634 1635 However, we can relax time bounds for RST segments to MSL. 1636 */ 1637 if (rst && !time_before32(ktime_get_seconds(), 1638 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1639 return false; 1640 return true; 1641 } 1642 1643 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1644 int mib_idx, u32 *last_oow_ack_time); 1645 1646 static inline void tcp_mib_init(struct net *net) 1647 { 1648 /* See RFC 2012 */ 1649 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1650 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1651 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1652 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1653 } 1654 1655 /* from STCP */ 1656 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1657 { 1658 tp->lost_skb_hint = NULL; 1659 } 1660 1661 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1662 { 1663 tcp_clear_retrans_hints_partial(tp); 1664 tp->retransmit_skb_hint = NULL; 1665 } 1666 1667 union tcp_md5_addr { 1668 struct in_addr a4; 1669 #if IS_ENABLED(CONFIG_IPV6) 1670 struct in6_addr a6; 1671 #endif 1672 }; 1673 1674 /* - key database */ 1675 struct tcp_md5sig_key { 1676 struct hlist_node node; 1677 u8 keylen; 1678 u8 family; /* AF_INET or AF_INET6 */ 1679 u8 prefixlen; 1680 u8 flags; 1681 union tcp_md5_addr addr; 1682 int l3index; /* set if key added with L3 scope */ 1683 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1684 struct rcu_head rcu; 1685 }; 1686 1687 /* - sock block */ 1688 struct tcp_md5sig_info { 1689 struct hlist_head head; 1690 struct rcu_head rcu; 1691 }; 1692 1693 /* - pseudo header */ 1694 struct tcp4_pseudohdr { 1695 __be32 saddr; 1696 __be32 daddr; 1697 __u8 pad; 1698 __u8 protocol; 1699 __be16 len; 1700 }; 1701 1702 struct tcp6_pseudohdr { 1703 struct in6_addr saddr; 1704 struct in6_addr daddr; 1705 __be32 len; 1706 __be32 protocol; /* including padding */ 1707 }; 1708 1709 union tcp_md5sum_block { 1710 struct tcp4_pseudohdr ip4; 1711 #if IS_ENABLED(CONFIG_IPV6) 1712 struct tcp6_pseudohdr ip6; 1713 #endif 1714 }; 1715 1716 /* - pool: digest algorithm, hash description and scratch buffer */ 1717 struct tcp_md5sig_pool { 1718 struct ahash_request *md5_req; 1719 void *scratch; 1720 }; 1721 1722 /* - functions */ 1723 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1724 const struct sock *sk, const struct sk_buff *skb); 1725 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1726 int family, u8 prefixlen, int l3index, u8 flags, 1727 const u8 *newkey, u8 newkeylen); 1728 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1729 int family, u8 prefixlen, int l3index, 1730 struct tcp_md5sig_key *key); 1731 1732 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1733 int family, u8 prefixlen, int l3index, u8 flags); 1734 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1735 const struct sock *addr_sk); 1736 1737 #ifdef CONFIG_TCP_MD5SIG 1738 #include <linux/jump_label.h> 1739 extern struct static_key_false_deferred tcp_md5_needed; 1740 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1741 const union tcp_md5_addr *addr, 1742 int family); 1743 static inline struct tcp_md5sig_key * 1744 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1745 const union tcp_md5_addr *addr, int family) 1746 { 1747 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1748 return NULL; 1749 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1750 } 1751 1752 enum skb_drop_reason 1753 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1754 const void *saddr, const void *daddr, 1755 int family, int dif, int sdif); 1756 1757 1758 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1759 #else 1760 static inline struct tcp_md5sig_key * 1761 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1762 const union tcp_md5_addr *addr, int family) 1763 { 1764 return NULL; 1765 } 1766 1767 static inline enum skb_drop_reason 1768 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1769 const void *saddr, const void *daddr, 1770 int family, int dif, int sdif) 1771 { 1772 return SKB_NOT_DROPPED_YET; 1773 } 1774 #define tcp_twsk_md5_key(twsk) NULL 1775 #endif 1776 1777 bool tcp_alloc_md5sig_pool(void); 1778 1779 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1780 static inline void tcp_put_md5sig_pool(void) 1781 { 1782 local_bh_enable(); 1783 } 1784 1785 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1786 unsigned int header_len); 1787 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1788 const struct tcp_md5sig_key *key); 1789 1790 /* From tcp_fastopen.c */ 1791 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1792 struct tcp_fastopen_cookie *cookie); 1793 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1794 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1795 u16 try_exp); 1796 struct tcp_fastopen_request { 1797 /* Fast Open cookie. Size 0 means a cookie request */ 1798 struct tcp_fastopen_cookie cookie; 1799 struct msghdr *data; /* data in MSG_FASTOPEN */ 1800 size_t size; 1801 int copied; /* queued in tcp_connect() */ 1802 struct ubuf_info *uarg; 1803 }; 1804 void tcp_free_fastopen_req(struct tcp_sock *tp); 1805 void tcp_fastopen_destroy_cipher(struct sock *sk); 1806 void tcp_fastopen_ctx_destroy(struct net *net); 1807 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1808 void *primary_key, void *backup_key); 1809 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1810 u64 *key); 1811 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1812 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1813 struct request_sock *req, 1814 struct tcp_fastopen_cookie *foc, 1815 const struct dst_entry *dst); 1816 void tcp_fastopen_init_key_once(struct net *net); 1817 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1818 struct tcp_fastopen_cookie *cookie); 1819 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1820 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1821 #define TCP_FASTOPEN_KEY_MAX 2 1822 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1823 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1824 1825 /* Fastopen key context */ 1826 struct tcp_fastopen_context { 1827 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1828 int num; 1829 struct rcu_head rcu; 1830 }; 1831 1832 void tcp_fastopen_active_disable(struct sock *sk); 1833 bool tcp_fastopen_active_should_disable(struct sock *sk); 1834 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1835 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1836 1837 /* Caller needs to wrap with rcu_read_(un)lock() */ 1838 static inline 1839 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1840 { 1841 struct tcp_fastopen_context *ctx; 1842 1843 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1844 if (!ctx) 1845 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1846 return ctx; 1847 } 1848 1849 static inline 1850 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1851 const struct tcp_fastopen_cookie *orig) 1852 { 1853 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1854 orig->len == foc->len && 1855 !memcmp(orig->val, foc->val, foc->len)) 1856 return true; 1857 return false; 1858 } 1859 1860 static inline 1861 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1862 { 1863 return ctx->num; 1864 } 1865 1866 /* Latencies incurred by various limits for a sender. They are 1867 * chronograph-like stats that are mutually exclusive. 1868 */ 1869 enum tcp_chrono { 1870 TCP_CHRONO_UNSPEC, 1871 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1872 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1873 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1874 __TCP_CHRONO_MAX, 1875 }; 1876 1877 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1878 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1879 1880 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1881 * the same memory storage than skb->destructor/_skb_refdst 1882 */ 1883 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1884 { 1885 skb->destructor = NULL; 1886 skb->_skb_refdst = 0UL; 1887 } 1888 1889 #define tcp_skb_tsorted_save(skb) { \ 1890 unsigned long _save = skb->_skb_refdst; \ 1891 skb->_skb_refdst = 0UL; 1892 1893 #define tcp_skb_tsorted_restore(skb) \ 1894 skb->_skb_refdst = _save; \ 1895 } 1896 1897 void tcp_write_queue_purge(struct sock *sk); 1898 1899 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1900 { 1901 return skb_rb_first(&sk->tcp_rtx_queue); 1902 } 1903 1904 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1905 { 1906 return skb_rb_last(&sk->tcp_rtx_queue); 1907 } 1908 1909 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1910 { 1911 return skb_peek_tail(&sk->sk_write_queue); 1912 } 1913 1914 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1915 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1916 1917 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1918 { 1919 return skb_peek(&sk->sk_write_queue); 1920 } 1921 1922 static inline bool tcp_skb_is_last(const struct sock *sk, 1923 const struct sk_buff *skb) 1924 { 1925 return skb_queue_is_last(&sk->sk_write_queue, skb); 1926 } 1927 1928 /** 1929 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1930 * @sk: socket 1931 * 1932 * Since the write queue can have a temporary empty skb in it, 1933 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1934 */ 1935 static inline bool tcp_write_queue_empty(const struct sock *sk) 1936 { 1937 const struct tcp_sock *tp = tcp_sk(sk); 1938 1939 return tp->write_seq == tp->snd_nxt; 1940 } 1941 1942 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1943 { 1944 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1945 } 1946 1947 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1948 { 1949 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1950 } 1951 1952 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1953 { 1954 __skb_queue_tail(&sk->sk_write_queue, skb); 1955 1956 /* Queue it, remembering where we must start sending. */ 1957 if (sk->sk_write_queue.next == skb) 1958 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1959 } 1960 1961 /* Insert new before skb on the write queue of sk. */ 1962 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1963 struct sk_buff *skb, 1964 struct sock *sk) 1965 { 1966 __skb_queue_before(&sk->sk_write_queue, skb, new); 1967 } 1968 1969 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1970 { 1971 tcp_skb_tsorted_anchor_cleanup(skb); 1972 __skb_unlink(skb, &sk->sk_write_queue); 1973 } 1974 1975 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1976 1977 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1978 { 1979 tcp_skb_tsorted_anchor_cleanup(skb); 1980 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1981 } 1982 1983 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1984 { 1985 list_del(&skb->tcp_tsorted_anchor); 1986 tcp_rtx_queue_unlink(skb, sk); 1987 tcp_wmem_free_skb(sk, skb); 1988 } 1989 1990 static inline void tcp_push_pending_frames(struct sock *sk) 1991 { 1992 if (tcp_send_head(sk)) { 1993 struct tcp_sock *tp = tcp_sk(sk); 1994 1995 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1996 } 1997 } 1998 1999 /* Start sequence of the skb just after the highest skb with SACKed 2000 * bit, valid only if sacked_out > 0 or when the caller has ensured 2001 * validity by itself. 2002 */ 2003 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2004 { 2005 if (!tp->sacked_out) 2006 return tp->snd_una; 2007 2008 if (tp->highest_sack == NULL) 2009 return tp->snd_nxt; 2010 2011 return TCP_SKB_CB(tp->highest_sack)->seq; 2012 } 2013 2014 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2015 { 2016 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2017 } 2018 2019 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2020 { 2021 return tcp_sk(sk)->highest_sack; 2022 } 2023 2024 static inline void tcp_highest_sack_reset(struct sock *sk) 2025 { 2026 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2027 } 2028 2029 /* Called when old skb is about to be deleted and replaced by new skb */ 2030 static inline void tcp_highest_sack_replace(struct sock *sk, 2031 struct sk_buff *old, 2032 struct sk_buff *new) 2033 { 2034 if (old == tcp_highest_sack(sk)) 2035 tcp_sk(sk)->highest_sack = new; 2036 } 2037 2038 /* This helper checks if socket has IP_TRANSPARENT set */ 2039 static inline bool inet_sk_transparent(const struct sock *sk) 2040 { 2041 switch (sk->sk_state) { 2042 case TCP_TIME_WAIT: 2043 return inet_twsk(sk)->tw_transparent; 2044 case TCP_NEW_SYN_RECV: 2045 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2046 } 2047 return inet_test_bit(TRANSPARENT, sk); 2048 } 2049 2050 /* Determines whether this is a thin stream (which may suffer from 2051 * increased latency). Used to trigger latency-reducing mechanisms. 2052 */ 2053 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2054 { 2055 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2056 } 2057 2058 /* /proc */ 2059 enum tcp_seq_states { 2060 TCP_SEQ_STATE_LISTENING, 2061 TCP_SEQ_STATE_ESTABLISHED, 2062 }; 2063 2064 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2065 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2066 void tcp_seq_stop(struct seq_file *seq, void *v); 2067 2068 struct tcp_seq_afinfo { 2069 sa_family_t family; 2070 }; 2071 2072 struct tcp_iter_state { 2073 struct seq_net_private p; 2074 enum tcp_seq_states state; 2075 struct sock *syn_wait_sk; 2076 int bucket, offset, sbucket, num; 2077 loff_t last_pos; 2078 }; 2079 2080 extern struct request_sock_ops tcp_request_sock_ops; 2081 extern struct request_sock_ops tcp6_request_sock_ops; 2082 2083 void tcp_v4_destroy_sock(struct sock *sk); 2084 2085 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2086 netdev_features_t features); 2087 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2088 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2089 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2090 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2091 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2092 void tcp_gro_complete(struct sk_buff *skb); 2093 2094 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2095 2096 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2097 { 2098 struct net *net = sock_net((struct sock *)tp); 2099 u32 val; 2100 2101 val = READ_ONCE(tp->notsent_lowat); 2102 2103 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2104 } 2105 2106 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2107 2108 #ifdef CONFIG_PROC_FS 2109 int tcp4_proc_init(void); 2110 void tcp4_proc_exit(void); 2111 #endif 2112 2113 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2114 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2115 const struct tcp_request_sock_ops *af_ops, 2116 struct sock *sk, struct sk_buff *skb); 2117 2118 /* TCP af-specific functions */ 2119 struct tcp_sock_af_ops { 2120 #ifdef CONFIG_TCP_MD5SIG 2121 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2122 const struct sock *addr_sk); 2123 int (*calc_md5_hash)(char *location, 2124 const struct tcp_md5sig_key *md5, 2125 const struct sock *sk, 2126 const struct sk_buff *skb); 2127 int (*md5_parse)(struct sock *sk, 2128 int optname, 2129 sockptr_t optval, 2130 int optlen); 2131 #endif 2132 }; 2133 2134 struct tcp_request_sock_ops { 2135 u16 mss_clamp; 2136 #ifdef CONFIG_TCP_MD5SIG 2137 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2138 const struct sock *addr_sk); 2139 int (*calc_md5_hash) (char *location, 2140 const struct tcp_md5sig_key *md5, 2141 const struct sock *sk, 2142 const struct sk_buff *skb); 2143 #endif 2144 #ifdef CONFIG_SYN_COOKIES 2145 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2146 __u16 *mss); 2147 #endif 2148 struct dst_entry *(*route_req)(const struct sock *sk, 2149 struct sk_buff *skb, 2150 struct flowi *fl, 2151 struct request_sock *req); 2152 u32 (*init_seq)(const struct sk_buff *skb); 2153 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2154 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2155 struct flowi *fl, struct request_sock *req, 2156 struct tcp_fastopen_cookie *foc, 2157 enum tcp_synack_type synack_type, 2158 struct sk_buff *syn_skb); 2159 }; 2160 2161 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2162 #if IS_ENABLED(CONFIG_IPV6) 2163 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2164 #endif 2165 2166 #ifdef CONFIG_SYN_COOKIES 2167 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2168 const struct sock *sk, struct sk_buff *skb, 2169 __u16 *mss) 2170 { 2171 tcp_synq_overflow(sk); 2172 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2173 return ops->cookie_init_seq(skb, mss); 2174 } 2175 #else 2176 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2177 const struct sock *sk, struct sk_buff *skb, 2178 __u16 *mss) 2179 { 2180 return 0; 2181 } 2182 #endif 2183 2184 int tcpv4_offload_init(void); 2185 2186 void tcp_v4_init(void); 2187 void tcp_init(void); 2188 2189 /* tcp_recovery.c */ 2190 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2191 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2192 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2193 u32 reo_wnd); 2194 extern bool tcp_rack_mark_lost(struct sock *sk); 2195 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2196 u64 xmit_time); 2197 extern void tcp_rack_reo_timeout(struct sock *sk); 2198 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2199 2200 /* tcp_plb.c */ 2201 2202 /* 2203 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2204 * expects cong_ratio which represents fraction of traffic that experienced 2205 * congestion over a single RTT. In order to avoid floating point operations, 2206 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2207 */ 2208 #define TCP_PLB_SCALE 8 2209 2210 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2211 struct tcp_plb_state { 2212 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2213 unused:3; 2214 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2215 }; 2216 2217 static inline void tcp_plb_init(const struct sock *sk, 2218 struct tcp_plb_state *plb) 2219 { 2220 plb->consec_cong_rounds = 0; 2221 plb->pause_until = 0; 2222 } 2223 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2224 const int cong_ratio); 2225 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2226 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2227 2228 /* At how many usecs into the future should the RTO fire? */ 2229 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2230 { 2231 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2232 u32 rto = inet_csk(sk)->icsk_rto; 2233 2234 if (likely(skb)) { 2235 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2236 2237 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2238 } else { 2239 WARN_ONCE(1, 2240 "rtx queue emtpy: " 2241 "out:%u sacked:%u lost:%u retrans:%u " 2242 "tlp_high_seq:%u sk_state:%u ca_state:%u " 2243 "advmss:%u mss_cache:%u pmtu:%u\n", 2244 tcp_sk(sk)->packets_out, tcp_sk(sk)->sacked_out, 2245 tcp_sk(sk)->lost_out, tcp_sk(sk)->retrans_out, 2246 tcp_sk(sk)->tlp_high_seq, sk->sk_state, 2247 inet_csk(sk)->icsk_ca_state, 2248 tcp_sk(sk)->advmss, tcp_sk(sk)->mss_cache, 2249 inet_csk(sk)->icsk_pmtu_cookie); 2250 return jiffies_to_usecs(rto); 2251 } 2252 2253 } 2254 2255 /* 2256 * Save and compile IPv4 options, return a pointer to it 2257 */ 2258 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2259 struct sk_buff *skb) 2260 { 2261 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2262 struct ip_options_rcu *dopt = NULL; 2263 2264 if (opt->optlen) { 2265 int opt_size = sizeof(*dopt) + opt->optlen; 2266 2267 dopt = kmalloc(opt_size, GFP_ATOMIC); 2268 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2269 kfree(dopt); 2270 dopt = NULL; 2271 } 2272 } 2273 return dopt; 2274 } 2275 2276 /* locally generated TCP pure ACKs have skb->truesize == 2 2277 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2278 * This is much faster than dissecting the packet to find out. 2279 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2280 */ 2281 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2282 { 2283 return skb->truesize == 2; 2284 } 2285 2286 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2287 { 2288 skb->truesize = 2; 2289 } 2290 2291 static inline int tcp_inq(struct sock *sk) 2292 { 2293 struct tcp_sock *tp = tcp_sk(sk); 2294 int answ; 2295 2296 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2297 answ = 0; 2298 } else if (sock_flag(sk, SOCK_URGINLINE) || 2299 !tp->urg_data || 2300 before(tp->urg_seq, tp->copied_seq) || 2301 !before(tp->urg_seq, tp->rcv_nxt)) { 2302 2303 answ = tp->rcv_nxt - tp->copied_seq; 2304 2305 /* Subtract 1, if FIN was received */ 2306 if (answ && sock_flag(sk, SOCK_DONE)) 2307 answ--; 2308 } else { 2309 answ = tp->urg_seq - tp->copied_seq; 2310 } 2311 2312 return answ; 2313 } 2314 2315 int tcp_peek_len(struct socket *sock); 2316 2317 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2318 { 2319 u16 segs_in; 2320 2321 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2322 2323 /* We update these fields while other threads might 2324 * read them from tcp_get_info() 2325 */ 2326 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2327 if (skb->len > tcp_hdrlen(skb)) 2328 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2329 } 2330 2331 /* 2332 * TCP listen path runs lockless. 2333 * We forced "struct sock" to be const qualified to make sure 2334 * we don't modify one of its field by mistake. 2335 * Here, we increment sk_drops which is an atomic_t, so we can safely 2336 * make sock writable again. 2337 */ 2338 static inline void tcp_listendrop(const struct sock *sk) 2339 { 2340 atomic_inc(&((struct sock *)sk)->sk_drops); 2341 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2342 } 2343 2344 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2345 2346 /* 2347 * Interface for adding Upper Level Protocols over TCP 2348 */ 2349 2350 #define TCP_ULP_NAME_MAX 16 2351 #define TCP_ULP_MAX 128 2352 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2353 2354 struct tcp_ulp_ops { 2355 struct list_head list; 2356 2357 /* initialize ulp */ 2358 int (*init)(struct sock *sk); 2359 /* update ulp */ 2360 void (*update)(struct sock *sk, struct proto *p, 2361 void (*write_space)(struct sock *sk)); 2362 /* cleanup ulp */ 2363 void (*release)(struct sock *sk); 2364 /* diagnostic */ 2365 int (*get_info)(struct sock *sk, struct sk_buff *skb); 2366 size_t (*get_info_size)(const struct sock *sk); 2367 /* clone ulp */ 2368 void (*clone)(const struct request_sock *req, struct sock *newsk, 2369 const gfp_t priority); 2370 2371 char name[TCP_ULP_NAME_MAX]; 2372 struct module *owner; 2373 }; 2374 int tcp_register_ulp(struct tcp_ulp_ops *type); 2375 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2376 int tcp_set_ulp(struct sock *sk, const char *name); 2377 void tcp_get_available_ulp(char *buf, size_t len); 2378 void tcp_cleanup_ulp(struct sock *sk); 2379 void tcp_update_ulp(struct sock *sk, struct proto *p, 2380 void (*write_space)(struct sock *sk)); 2381 2382 #define MODULE_ALIAS_TCP_ULP(name) \ 2383 __MODULE_INFO(alias, alias_userspace, name); \ 2384 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2385 2386 #ifdef CONFIG_NET_SOCK_MSG 2387 struct sk_msg; 2388 struct sk_psock; 2389 2390 #ifdef CONFIG_BPF_SYSCALL 2391 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2392 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2393 #endif /* CONFIG_BPF_SYSCALL */ 2394 2395 #ifdef CONFIG_INET 2396 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2397 #else 2398 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2399 { 2400 } 2401 #endif 2402 2403 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2404 struct sk_msg *msg, u32 bytes, int flags); 2405 #endif /* CONFIG_NET_SOCK_MSG */ 2406 2407 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2408 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2409 { 2410 } 2411 #endif 2412 2413 #ifdef CONFIG_CGROUP_BPF 2414 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2415 struct sk_buff *skb, 2416 unsigned int end_offset) 2417 { 2418 skops->skb = skb; 2419 skops->skb_data_end = skb->data + end_offset; 2420 } 2421 #else 2422 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2423 struct sk_buff *skb, 2424 unsigned int end_offset) 2425 { 2426 } 2427 #endif 2428 2429 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2430 * is < 0, then the BPF op failed (for example if the loaded BPF 2431 * program does not support the chosen operation or there is no BPF 2432 * program loaded). 2433 */ 2434 #ifdef CONFIG_BPF 2435 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2436 { 2437 struct bpf_sock_ops_kern sock_ops; 2438 int ret; 2439 2440 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2441 if (sk_fullsock(sk)) { 2442 sock_ops.is_fullsock = 1; 2443 sock_owned_by_me(sk); 2444 } 2445 2446 sock_ops.sk = sk; 2447 sock_ops.op = op; 2448 if (nargs > 0) 2449 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2450 2451 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2452 if (ret == 0) 2453 ret = sock_ops.reply; 2454 else 2455 ret = -1; 2456 return ret; 2457 } 2458 2459 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2460 { 2461 u32 args[2] = {arg1, arg2}; 2462 2463 return tcp_call_bpf(sk, op, 2, args); 2464 } 2465 2466 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2467 u32 arg3) 2468 { 2469 u32 args[3] = {arg1, arg2, arg3}; 2470 2471 return tcp_call_bpf(sk, op, 3, args); 2472 } 2473 2474 #else 2475 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2476 { 2477 return -EPERM; 2478 } 2479 2480 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2481 { 2482 return -EPERM; 2483 } 2484 2485 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2486 u32 arg3) 2487 { 2488 return -EPERM; 2489 } 2490 2491 #endif 2492 2493 static inline u32 tcp_timeout_init(struct sock *sk) 2494 { 2495 int timeout; 2496 2497 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2498 2499 if (timeout <= 0) 2500 timeout = TCP_TIMEOUT_INIT; 2501 return min_t(int, timeout, TCP_RTO_MAX); 2502 } 2503 2504 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2505 { 2506 int rwnd; 2507 2508 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2509 2510 if (rwnd < 0) 2511 rwnd = 0; 2512 return rwnd; 2513 } 2514 2515 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2516 { 2517 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2518 } 2519 2520 static inline void tcp_bpf_rtt(struct sock *sk) 2521 { 2522 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2523 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2524 } 2525 2526 #if IS_ENABLED(CONFIG_SMC) 2527 extern struct static_key_false tcp_have_smc; 2528 #endif 2529 2530 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2531 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2532 void (*cad)(struct sock *sk, u32 ack_seq)); 2533 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2534 void clean_acked_data_flush(void); 2535 #endif 2536 2537 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2538 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2539 const struct tcp_sock *tp) 2540 { 2541 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2542 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2543 } 2544 2545 /* Compute Earliest Departure Time for some control packets 2546 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2547 */ 2548 static inline u64 tcp_transmit_time(const struct sock *sk) 2549 { 2550 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2551 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2552 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2553 2554 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2555 } 2556 return 0; 2557 } 2558 2559 #endif /* _TCP_H */ 2560