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 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); 135 136 #if HZ >= 100 137 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 138 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 139 #else 140 #define TCP_DELACK_MIN 4U 141 #define TCP_ATO_MIN 4U 142 #endif 143 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 144 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 145 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 146 147 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 148 149 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 150 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 151 * used as a fallback RTO for the 152 * initial data transmission if no 153 * valid RTT sample has been acquired, 154 * most likely due to retrans in 3WHS. 155 */ 156 157 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 158 * for local resources. 159 */ 160 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 161 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 162 #define TCP_KEEPALIVE_INTVL (75*HZ) 163 164 #define MAX_TCP_KEEPIDLE 32767 165 #define MAX_TCP_KEEPINTVL 32767 166 #define MAX_TCP_KEEPCNT 127 167 #define MAX_TCP_SYNCNT 127 168 169 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds 170 * to avoid overflows. This assumes a clock smaller than 1 Mhz. 171 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz. 172 */ 173 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC) 174 175 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 176 * after this time. It should be equal 177 * (or greater than) TCP_TIMEWAIT_LEN 178 * to provide reliability equal to one 179 * provided by timewait state. 180 */ 181 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 182 * timestamps. It must be less than 183 * minimal timewait lifetime. 184 */ 185 /* 186 * TCP option 187 */ 188 189 #define TCPOPT_NOP 1 /* Padding */ 190 #define TCPOPT_EOL 0 /* End of options */ 191 #define TCPOPT_MSS 2 /* Segment size negotiating */ 192 #define TCPOPT_WINDOW 3 /* Window scaling */ 193 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 194 #define TCPOPT_SACK 5 /* SACK Block */ 195 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 196 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 197 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 198 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 199 #define TCPOPT_EXP 254 /* Experimental */ 200 /* Magic number to be after the option value for sharing TCP 201 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 202 */ 203 #define TCPOPT_FASTOPEN_MAGIC 0xF989 204 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 205 206 /* 207 * TCP option lengths 208 */ 209 210 #define TCPOLEN_MSS 4 211 #define TCPOLEN_WINDOW 3 212 #define TCPOLEN_SACK_PERM 2 213 #define TCPOLEN_TIMESTAMP 10 214 #define TCPOLEN_MD5SIG 18 215 #define TCPOLEN_FASTOPEN_BASE 2 216 #define TCPOLEN_EXP_FASTOPEN_BASE 4 217 #define TCPOLEN_EXP_SMC_BASE 6 218 219 /* But this is what stacks really send out. */ 220 #define TCPOLEN_TSTAMP_ALIGNED 12 221 #define TCPOLEN_WSCALE_ALIGNED 4 222 #define TCPOLEN_SACKPERM_ALIGNED 4 223 #define TCPOLEN_SACK_BASE 2 224 #define TCPOLEN_SACK_BASE_ALIGNED 4 225 #define TCPOLEN_SACK_PERBLOCK 8 226 #define TCPOLEN_MD5SIG_ALIGNED 20 227 #define TCPOLEN_MSS_ALIGNED 4 228 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 229 230 /* Flags in tp->nonagle */ 231 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 232 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 233 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 234 235 /* TCP thin-stream limits */ 236 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 237 238 /* TCP initial congestion window as per rfc6928 */ 239 #define TCP_INIT_CWND 10 240 241 /* Bit Flags for sysctl_tcp_fastopen */ 242 #define TFO_CLIENT_ENABLE 1 243 #define TFO_SERVER_ENABLE 2 244 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 245 246 /* Accept SYN data w/o any cookie option */ 247 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 248 249 /* Force enable TFO on all listeners, i.e., not requiring the 250 * TCP_FASTOPEN socket option. 251 */ 252 #define TFO_SERVER_WO_SOCKOPT1 0x400 253 254 255 /* sysctl variables for tcp */ 256 extern int sysctl_tcp_max_orphans; 257 extern long sysctl_tcp_mem[3]; 258 259 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 260 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 261 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 262 263 extern atomic_long_t tcp_memory_allocated; 264 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 265 266 extern struct percpu_counter tcp_sockets_allocated; 267 extern unsigned long tcp_memory_pressure; 268 269 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 270 static inline bool tcp_under_memory_pressure(const struct sock *sk) 271 { 272 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 273 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 274 return true; 275 276 return READ_ONCE(tcp_memory_pressure); 277 } 278 /* 279 * The next routines deal with comparing 32 bit unsigned ints 280 * and worry about wraparound (automatic with unsigned arithmetic). 281 */ 282 283 static inline bool before(__u32 seq1, __u32 seq2) 284 { 285 return (__s32)(seq1-seq2) < 0; 286 } 287 #define after(seq2, seq1) before(seq1, seq2) 288 289 /* is s2<=s1<=s3 ? */ 290 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 291 { 292 return seq3 - seq2 >= seq1 - seq2; 293 } 294 295 static inline bool tcp_out_of_memory(struct sock *sk) 296 { 297 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 298 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 299 return true; 300 return false; 301 } 302 303 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 304 { 305 sk_wmem_queued_add(sk, -skb->truesize); 306 if (!skb_zcopy_pure(skb)) 307 sk_mem_uncharge(sk, skb->truesize); 308 else 309 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 310 __kfree_skb(skb); 311 } 312 313 void sk_forced_mem_schedule(struct sock *sk, int size); 314 315 bool tcp_check_oom(struct sock *sk, int shift); 316 317 318 extern struct proto tcp_prot; 319 320 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 321 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 322 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 323 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 324 325 void tcp_tasklet_init(void); 326 327 int tcp_v4_err(struct sk_buff *skb, u32); 328 329 void tcp_shutdown(struct sock *sk, int how); 330 331 int tcp_v4_early_demux(struct sk_buff *skb); 332 int tcp_v4_rcv(struct sk_buff *skb); 333 334 void tcp_remove_empty_skb(struct sock *sk); 335 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 336 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 337 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 338 size_t size, struct ubuf_info *uarg); 339 void tcp_splice_eof(struct socket *sock); 340 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 341 int tcp_wmem_schedule(struct sock *sk, int copy); 342 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 343 int size_goal); 344 void tcp_release_cb(struct sock *sk); 345 void tcp_wfree(struct sk_buff *skb); 346 void tcp_write_timer_handler(struct sock *sk); 347 void tcp_delack_timer_handler(struct sock *sk); 348 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 349 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 350 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 351 void tcp_rcv_space_adjust(struct sock *sk); 352 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 353 void tcp_twsk_destructor(struct sock *sk); 354 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 355 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 356 struct pipe_inode_info *pipe, size_t len, 357 unsigned int flags); 358 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 359 bool force_schedule); 360 361 static inline void tcp_dec_quickack_mode(struct sock *sk) 362 { 363 struct inet_connection_sock *icsk = inet_csk(sk); 364 365 if (icsk->icsk_ack.quick) { 366 /* How many ACKs S/ACKing new data have we sent? */ 367 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 368 369 if (pkts >= icsk->icsk_ack.quick) { 370 icsk->icsk_ack.quick = 0; 371 /* Leaving quickack mode we deflate ATO. */ 372 icsk->icsk_ack.ato = TCP_ATO_MIN; 373 } else 374 icsk->icsk_ack.quick -= pkts; 375 } 376 } 377 378 #define TCP_ECN_OK 1 379 #define TCP_ECN_QUEUE_CWR 2 380 #define TCP_ECN_DEMAND_CWR 4 381 #define TCP_ECN_SEEN 8 382 383 enum tcp_tw_status { 384 TCP_TW_SUCCESS = 0, 385 TCP_TW_RST = 1, 386 TCP_TW_ACK = 2, 387 TCP_TW_SYN = 3 388 }; 389 390 391 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 392 struct sk_buff *skb, 393 const struct tcphdr *th); 394 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 395 struct request_sock *req, bool fastopen, 396 bool *lost_race); 397 int tcp_child_process(struct sock *parent, struct sock *child, 398 struct sk_buff *skb); 399 void tcp_enter_loss(struct sock *sk); 400 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 401 void tcp_clear_retrans(struct tcp_sock *tp); 402 void tcp_update_metrics(struct sock *sk); 403 void tcp_init_metrics(struct sock *sk); 404 void tcp_metrics_init(void); 405 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 406 void __tcp_close(struct sock *sk, long timeout); 407 void tcp_close(struct sock *sk, long timeout); 408 void tcp_init_sock(struct sock *sk); 409 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 410 __poll_t tcp_poll(struct file *file, struct socket *sock, 411 struct poll_table_struct *wait); 412 int do_tcp_getsockopt(struct sock *sk, int level, 413 int optname, sockptr_t optval, sockptr_t optlen); 414 int tcp_getsockopt(struct sock *sk, int level, int optname, 415 char __user *optval, int __user *optlen); 416 bool tcp_bpf_bypass_getsockopt(int level, int optname); 417 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 418 sockptr_t optval, unsigned int optlen); 419 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 420 unsigned int optlen); 421 void tcp_set_keepalive(struct sock *sk, int val); 422 void tcp_syn_ack_timeout(const struct request_sock *req); 423 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 424 int flags, int *addr_len); 425 int tcp_set_rcvlowat(struct sock *sk, int val); 426 int tcp_set_window_clamp(struct sock *sk, int val); 427 void tcp_update_recv_tstamps(struct sk_buff *skb, 428 struct scm_timestamping_internal *tss); 429 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 430 struct scm_timestamping_internal *tss); 431 void tcp_data_ready(struct sock *sk); 432 #ifdef CONFIG_MMU 433 int tcp_mmap(struct file *file, struct socket *sock, 434 struct vm_area_struct *vma); 435 #endif 436 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 437 struct tcp_options_received *opt_rx, 438 int estab, struct tcp_fastopen_cookie *foc); 439 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 440 441 /* 442 * BPF SKB-less helpers 443 */ 444 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 445 struct tcphdr *th, u32 *cookie); 446 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 447 struct tcphdr *th, u32 *cookie); 448 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 449 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 450 const struct tcp_request_sock_ops *af_ops, 451 struct sock *sk, struct tcphdr *th); 452 /* 453 * TCP v4 functions exported for the inet6 API 454 */ 455 456 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 457 void tcp_v4_mtu_reduced(struct sock *sk); 458 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 459 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 460 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 461 struct sock *tcp_create_openreq_child(const struct sock *sk, 462 struct request_sock *req, 463 struct sk_buff *skb); 464 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 465 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 466 struct request_sock *req, 467 struct dst_entry *dst, 468 struct request_sock *req_unhash, 469 bool *own_req); 470 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 471 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 472 int tcp_connect(struct sock *sk); 473 enum tcp_synack_type { 474 TCP_SYNACK_NORMAL, 475 TCP_SYNACK_FASTOPEN, 476 TCP_SYNACK_COOKIE, 477 }; 478 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 479 struct request_sock *req, 480 struct tcp_fastopen_cookie *foc, 481 enum tcp_synack_type synack_type, 482 struct sk_buff *syn_skb); 483 int tcp_disconnect(struct sock *sk, int flags); 484 485 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 486 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 487 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 488 489 /* From syncookies.c */ 490 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 491 struct request_sock *req, 492 struct dst_entry *dst, u32 tsoff); 493 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 494 u32 cookie); 495 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 496 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 497 const struct tcp_request_sock_ops *af_ops, 498 struct sock *sk, struct sk_buff *skb); 499 #ifdef CONFIG_SYN_COOKIES 500 501 /* Syncookies use a monotonic timer which increments every 60 seconds. 502 * This counter is used both as a hash input and partially encoded into 503 * the cookie value. A cookie is only validated further if the delta 504 * between the current counter value and the encoded one is less than this, 505 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 506 * the counter advances immediately after a cookie is generated). 507 */ 508 #define MAX_SYNCOOKIE_AGE 2 509 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 510 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 511 512 /* syncookies: remember time of last synqueue overflow 513 * But do not dirty this field too often (once per second is enough) 514 * It is racy as we do not hold a lock, but race is very minor. 515 */ 516 static inline void tcp_synq_overflow(const struct sock *sk) 517 { 518 unsigned int last_overflow; 519 unsigned int now = jiffies; 520 521 if (sk->sk_reuseport) { 522 struct sock_reuseport *reuse; 523 524 reuse = rcu_dereference(sk->sk_reuseport_cb); 525 if (likely(reuse)) { 526 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 527 if (!time_between32(now, last_overflow, 528 last_overflow + HZ)) 529 WRITE_ONCE(reuse->synq_overflow_ts, now); 530 return; 531 } 532 } 533 534 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 535 if (!time_between32(now, last_overflow, last_overflow + HZ)) 536 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 537 } 538 539 /* syncookies: no recent synqueue overflow on this listening socket? */ 540 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 541 { 542 unsigned int last_overflow; 543 unsigned int now = jiffies; 544 545 if (sk->sk_reuseport) { 546 struct sock_reuseport *reuse; 547 548 reuse = rcu_dereference(sk->sk_reuseport_cb); 549 if (likely(reuse)) { 550 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 551 return !time_between32(now, last_overflow - HZ, 552 last_overflow + 553 TCP_SYNCOOKIE_VALID); 554 } 555 } 556 557 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 558 559 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 560 * then we're under synflood. However, we have to use 561 * 'last_overflow - HZ' as lower bound. That's because a concurrent 562 * tcp_synq_overflow() could update .ts_recent_stamp after we read 563 * jiffies but before we store .ts_recent_stamp into last_overflow, 564 * which could lead to rejecting a valid syncookie. 565 */ 566 return !time_between32(now, last_overflow - HZ, 567 last_overflow + TCP_SYNCOOKIE_VALID); 568 } 569 570 static inline u32 tcp_cookie_time(void) 571 { 572 u64 val = get_jiffies_64(); 573 574 do_div(val, TCP_SYNCOOKIE_PERIOD); 575 return val; 576 } 577 578 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 579 u16 *mssp); 580 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 581 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 582 bool cookie_timestamp_decode(const struct net *net, 583 struct tcp_options_received *opt); 584 bool cookie_ecn_ok(const struct tcp_options_received *opt, 585 const struct net *net, const struct dst_entry *dst); 586 587 /* From net/ipv6/syncookies.c */ 588 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 589 u32 cookie); 590 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 591 592 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 593 const struct tcphdr *th, u16 *mssp); 594 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 595 #endif 596 /* tcp_output.c */ 597 598 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 599 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 600 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 601 int nonagle); 602 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 603 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 604 void tcp_retransmit_timer(struct sock *sk); 605 void tcp_xmit_retransmit_queue(struct sock *); 606 void tcp_simple_retransmit(struct sock *); 607 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 608 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 609 enum tcp_queue { 610 TCP_FRAG_IN_WRITE_QUEUE, 611 TCP_FRAG_IN_RTX_QUEUE, 612 }; 613 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 614 struct sk_buff *skb, u32 len, 615 unsigned int mss_now, gfp_t gfp); 616 617 void tcp_send_probe0(struct sock *); 618 int tcp_write_wakeup(struct sock *, int mib); 619 void tcp_send_fin(struct sock *sk); 620 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 621 int tcp_send_synack(struct sock *); 622 void tcp_push_one(struct sock *, unsigned int mss_now); 623 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 624 void tcp_send_ack(struct sock *sk); 625 void tcp_send_delayed_ack(struct sock *sk); 626 void tcp_send_loss_probe(struct sock *sk); 627 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 628 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 629 const struct sk_buff *next_skb); 630 631 /* tcp_input.c */ 632 void tcp_rearm_rto(struct sock *sk); 633 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 634 void tcp_reset(struct sock *sk, struct sk_buff *skb); 635 void tcp_fin(struct sock *sk); 636 void tcp_check_space(struct sock *sk); 637 void tcp_sack_compress_send_ack(struct sock *sk); 638 639 /* tcp_timer.c */ 640 void tcp_init_xmit_timers(struct sock *); 641 static inline void tcp_clear_xmit_timers(struct sock *sk) 642 { 643 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 644 __sock_put(sk); 645 646 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 647 __sock_put(sk); 648 649 inet_csk_clear_xmit_timers(sk); 650 } 651 652 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 653 unsigned int tcp_current_mss(struct sock *sk); 654 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 655 656 /* Bound MSS / TSO packet size with the half of the window */ 657 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 658 { 659 int cutoff; 660 661 /* When peer uses tiny windows, there is no use in packetizing 662 * to sub-MSS pieces for the sake of SWS or making sure there 663 * are enough packets in the pipe for fast recovery. 664 * 665 * On the other hand, for extremely large MSS devices, handling 666 * smaller than MSS windows in this way does make sense. 667 */ 668 if (tp->max_window > TCP_MSS_DEFAULT) 669 cutoff = (tp->max_window >> 1); 670 else 671 cutoff = tp->max_window; 672 673 if (cutoff && pktsize > cutoff) 674 return max_t(int, cutoff, 68U - tp->tcp_header_len); 675 else 676 return pktsize; 677 } 678 679 /* tcp.c */ 680 void tcp_get_info(struct sock *, struct tcp_info *); 681 682 /* Read 'sendfile()'-style from a TCP socket */ 683 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 684 sk_read_actor_t recv_actor); 685 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 686 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 687 void tcp_read_done(struct sock *sk, size_t len); 688 689 void tcp_initialize_rcv_mss(struct sock *sk); 690 691 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 692 int tcp_mss_to_mtu(struct sock *sk, int mss); 693 void tcp_mtup_init(struct sock *sk); 694 695 static inline void tcp_bound_rto(const struct sock *sk) 696 { 697 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 698 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 699 } 700 701 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 702 { 703 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 704 } 705 706 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 707 { 708 /* mptcp hooks are only on the slow path */ 709 if (sk_is_mptcp((struct sock *)tp)) 710 return; 711 712 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 713 ntohl(TCP_FLAG_ACK) | 714 snd_wnd); 715 } 716 717 static inline void tcp_fast_path_on(struct tcp_sock *tp) 718 { 719 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 720 } 721 722 static inline void tcp_fast_path_check(struct sock *sk) 723 { 724 struct tcp_sock *tp = tcp_sk(sk); 725 726 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 727 tp->rcv_wnd && 728 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 729 !tp->urg_data) 730 tcp_fast_path_on(tp); 731 } 732 733 u32 tcp_delack_max(const struct sock *sk); 734 735 /* Compute the actual rto_min value */ 736 static inline u32 tcp_rto_min(const struct sock *sk) 737 { 738 const struct dst_entry *dst = __sk_dst_get(sk); 739 u32 rto_min = inet_csk(sk)->icsk_rto_min; 740 741 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 742 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 743 return rto_min; 744 } 745 746 static inline u32 tcp_rto_min_us(const struct sock *sk) 747 { 748 return jiffies_to_usecs(tcp_rto_min(sk)); 749 } 750 751 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 752 { 753 return dst_metric_locked(dst, RTAX_CC_ALGO); 754 } 755 756 /* Minimum RTT in usec. ~0 means not available. */ 757 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 758 { 759 return minmax_get(&tp->rtt_min); 760 } 761 762 /* Compute the actual receive window we are currently advertising. 763 * Rcv_nxt can be after the window if our peer push more data 764 * than the offered window. 765 */ 766 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 767 { 768 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 769 770 if (win < 0) 771 win = 0; 772 return (u32) win; 773 } 774 775 /* Choose a new window, without checks for shrinking, and without 776 * scaling applied to the result. The caller does these things 777 * if necessary. This is a "raw" window selection. 778 */ 779 u32 __tcp_select_window(struct sock *sk); 780 781 void tcp_send_window_probe(struct sock *sk); 782 783 /* TCP uses 32bit jiffies to save some space. 784 * Note that this is different from tcp_time_stamp, which 785 * historically has been the same until linux-4.13. 786 */ 787 #define tcp_jiffies32 ((u32)jiffies) 788 789 /* 790 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 791 * It is no longer tied to jiffies, but to 1 ms clock. 792 * Note: double check if you want to use tcp_jiffies32 instead of this. 793 */ 794 #define TCP_TS_HZ 1000 795 796 static inline u64 tcp_clock_ns(void) 797 { 798 return ktime_get_ns(); 799 } 800 801 static inline u64 tcp_clock_us(void) 802 { 803 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 804 } 805 806 static inline u64 tcp_clock_ms(void) 807 { 808 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 809 } 810 811 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 812 * or usec resolution. Each socket carries a flag to select one or other 813 * resolution, as the route attribute could change anytime. 814 * Each flow must stick to initial resolution. 815 */ 816 static inline u32 tcp_clock_ts(bool usec_ts) 817 { 818 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 819 } 820 821 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 822 { 823 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 824 } 825 826 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 827 { 828 if (tp->tcp_usec_ts) 829 return tp->tcp_mstamp; 830 return tcp_time_stamp_ms(tp); 831 } 832 833 void tcp_mstamp_refresh(struct tcp_sock *tp); 834 835 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 836 { 837 return max_t(s64, t1 - t0, 0); 838 } 839 840 /* provide the departure time in us unit */ 841 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 842 { 843 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 844 } 845 846 /* Provide skb TSval in usec or ms unit */ 847 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 848 { 849 if (usec_ts) 850 return tcp_skb_timestamp_us(skb); 851 852 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 853 } 854 855 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 856 { 857 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 858 } 859 860 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 861 { 862 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 863 } 864 865 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 866 867 #define TCPHDR_FIN 0x01 868 #define TCPHDR_SYN 0x02 869 #define TCPHDR_RST 0x04 870 #define TCPHDR_PSH 0x08 871 #define TCPHDR_ACK 0x10 872 #define TCPHDR_URG 0x20 873 #define TCPHDR_ECE 0x40 874 #define TCPHDR_CWR 0x80 875 876 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 877 878 /* This is what the send packet queuing engine uses to pass 879 * TCP per-packet control information to the transmission code. 880 * We also store the host-order sequence numbers in here too. 881 * This is 44 bytes if IPV6 is enabled. 882 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 883 */ 884 struct tcp_skb_cb { 885 __u32 seq; /* Starting sequence number */ 886 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 887 union { 888 /* Note : tcp_tw_isn is used in input path only 889 * (isn chosen by tcp_timewait_state_process()) 890 * 891 * tcp_gso_segs/size are used in write queue only, 892 * cf tcp_skb_pcount()/tcp_skb_mss() 893 */ 894 __u32 tcp_tw_isn; 895 struct { 896 u16 tcp_gso_segs; 897 u16 tcp_gso_size; 898 }; 899 }; 900 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 901 902 __u8 sacked; /* State flags for SACK. */ 903 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 904 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 905 #define TCPCB_LOST 0x04 /* SKB is lost */ 906 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 907 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 908 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 909 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 910 TCPCB_REPAIRED) 911 912 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 913 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 914 eor:1, /* Is skb MSG_EOR marked? */ 915 has_rxtstamp:1, /* SKB has a RX timestamp */ 916 unused:5; 917 __u32 ack_seq; /* Sequence number ACK'd */ 918 union { 919 struct { 920 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 921 /* There is space for up to 24 bytes */ 922 __u32 is_app_limited:1, /* cwnd not fully used? */ 923 delivered_ce:20, 924 unused:11; 925 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 926 __u32 delivered; 927 /* start of send pipeline phase */ 928 u64 first_tx_mstamp; 929 /* when we reached the "delivered" count */ 930 u64 delivered_mstamp; 931 } tx; /* only used for outgoing skbs */ 932 union { 933 struct inet_skb_parm h4; 934 #if IS_ENABLED(CONFIG_IPV6) 935 struct inet6_skb_parm h6; 936 #endif 937 } header; /* For incoming skbs */ 938 }; 939 }; 940 941 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 942 943 extern const struct inet_connection_sock_af_ops ipv4_specific; 944 945 #if IS_ENABLED(CONFIG_IPV6) 946 /* This is the variant of inet6_iif() that must be used by TCP, 947 * as TCP moves IP6CB into a different location in skb->cb[] 948 */ 949 static inline int tcp_v6_iif(const struct sk_buff *skb) 950 { 951 return TCP_SKB_CB(skb)->header.h6.iif; 952 } 953 954 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 955 { 956 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 957 958 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 959 } 960 961 /* TCP_SKB_CB reference means this can not be used from early demux */ 962 static inline int tcp_v6_sdif(const struct sk_buff *skb) 963 { 964 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 965 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 966 return TCP_SKB_CB(skb)->header.h6.iif; 967 #endif 968 return 0; 969 } 970 971 extern const struct inet_connection_sock_af_ops ipv6_specific; 972 973 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 974 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 975 void tcp_v6_early_demux(struct sk_buff *skb); 976 977 #endif 978 979 /* TCP_SKB_CB reference means this can not be used from early demux */ 980 static inline int tcp_v4_sdif(struct sk_buff *skb) 981 { 982 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 983 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 984 return TCP_SKB_CB(skb)->header.h4.iif; 985 #endif 986 return 0; 987 } 988 989 /* Due to TSO, an SKB can be composed of multiple actual 990 * packets. To keep these tracked properly, we use this. 991 */ 992 static inline int tcp_skb_pcount(const struct sk_buff *skb) 993 { 994 return TCP_SKB_CB(skb)->tcp_gso_segs; 995 } 996 997 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 998 { 999 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1000 } 1001 1002 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1003 { 1004 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1005 } 1006 1007 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1008 static inline int tcp_skb_mss(const struct sk_buff *skb) 1009 { 1010 return TCP_SKB_CB(skb)->tcp_gso_size; 1011 } 1012 1013 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1014 { 1015 return likely(!TCP_SKB_CB(skb)->eor); 1016 } 1017 1018 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1019 const struct sk_buff *from) 1020 { 1021 return likely(tcp_skb_can_collapse_to(to) && 1022 mptcp_skb_can_collapse(to, from) && 1023 skb_pure_zcopy_same(to, from)); 1024 } 1025 1026 /* Events passed to congestion control interface */ 1027 enum tcp_ca_event { 1028 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1029 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1030 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1031 CA_EVENT_LOSS, /* loss timeout */ 1032 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1033 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1034 }; 1035 1036 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1037 enum tcp_ca_ack_event_flags { 1038 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1039 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1040 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1041 }; 1042 1043 /* 1044 * Interface for adding new TCP congestion control handlers 1045 */ 1046 #define TCP_CA_NAME_MAX 16 1047 #define TCP_CA_MAX 128 1048 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1049 1050 #define TCP_CA_UNSPEC 0 1051 1052 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1053 #define TCP_CONG_NON_RESTRICTED 0x1 1054 /* Requires ECN/ECT set on all packets */ 1055 #define TCP_CONG_NEEDS_ECN 0x2 1056 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1057 1058 union tcp_cc_info; 1059 1060 struct ack_sample { 1061 u32 pkts_acked; 1062 s32 rtt_us; 1063 u32 in_flight; 1064 }; 1065 1066 /* A rate sample measures the number of (original/retransmitted) data 1067 * packets delivered "delivered" over an interval of time "interval_us". 1068 * The tcp_rate.c code fills in the rate sample, and congestion 1069 * control modules that define a cong_control function to run at the end 1070 * of ACK processing can optionally chose to consult this sample when 1071 * setting cwnd and pacing rate. 1072 * A sample is invalid if "delivered" or "interval_us" is negative. 1073 */ 1074 struct rate_sample { 1075 u64 prior_mstamp; /* starting timestamp for interval */ 1076 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1077 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1078 s32 delivered; /* number of packets delivered over interval */ 1079 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1080 long interval_us; /* time for tp->delivered to incr "delivered" */ 1081 u32 snd_interval_us; /* snd interval for delivered packets */ 1082 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1083 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1084 int losses; /* number of packets marked lost upon ACK */ 1085 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1086 u32 prior_in_flight; /* in flight before this ACK */ 1087 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1088 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1089 bool is_retrans; /* is sample from retransmission? */ 1090 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1091 }; 1092 1093 struct tcp_congestion_ops { 1094 /* fast path fields are put first to fill one cache line */ 1095 1096 /* return slow start threshold (required) */ 1097 u32 (*ssthresh)(struct sock *sk); 1098 1099 /* do new cwnd calculation (required) */ 1100 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1101 1102 /* call before changing ca_state (optional) */ 1103 void (*set_state)(struct sock *sk, u8 new_state); 1104 1105 /* call when cwnd event occurs (optional) */ 1106 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1107 1108 /* call when ack arrives (optional) */ 1109 void (*in_ack_event)(struct sock *sk, u32 flags); 1110 1111 /* hook for packet ack accounting (optional) */ 1112 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1113 1114 /* override sysctl_tcp_min_tso_segs */ 1115 u32 (*min_tso_segs)(struct sock *sk); 1116 1117 /* call when packets are delivered to update cwnd and pacing rate, 1118 * after all the ca_state processing. (optional) 1119 */ 1120 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1121 1122 1123 /* new value of cwnd after loss (required) */ 1124 u32 (*undo_cwnd)(struct sock *sk); 1125 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1126 u32 (*sndbuf_expand)(struct sock *sk); 1127 1128 /* control/slow paths put last */ 1129 /* get info for inet_diag (optional) */ 1130 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1131 union tcp_cc_info *info); 1132 1133 char name[TCP_CA_NAME_MAX]; 1134 struct module *owner; 1135 struct list_head list; 1136 u32 key; 1137 u32 flags; 1138 1139 /* initialize private data (optional) */ 1140 void (*init)(struct sock *sk); 1141 /* cleanup private data (optional) */ 1142 void (*release)(struct sock *sk); 1143 } ____cacheline_aligned_in_smp; 1144 1145 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1146 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1147 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1148 struct tcp_congestion_ops *old_type); 1149 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1150 1151 void tcp_assign_congestion_control(struct sock *sk); 1152 void tcp_init_congestion_control(struct sock *sk); 1153 void tcp_cleanup_congestion_control(struct sock *sk); 1154 int tcp_set_default_congestion_control(struct net *net, const char *name); 1155 void tcp_get_default_congestion_control(struct net *net, char *name); 1156 void tcp_get_available_congestion_control(char *buf, size_t len); 1157 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1158 int tcp_set_allowed_congestion_control(char *allowed); 1159 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1160 bool cap_net_admin); 1161 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1162 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1163 1164 u32 tcp_reno_ssthresh(struct sock *sk); 1165 u32 tcp_reno_undo_cwnd(struct sock *sk); 1166 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1167 extern struct tcp_congestion_ops tcp_reno; 1168 1169 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1170 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1171 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1172 #ifdef CONFIG_INET 1173 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1174 #else 1175 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1176 { 1177 return NULL; 1178 } 1179 #endif 1180 1181 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1182 { 1183 const struct inet_connection_sock *icsk = inet_csk(sk); 1184 1185 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1186 } 1187 1188 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1189 { 1190 const struct inet_connection_sock *icsk = inet_csk(sk); 1191 1192 if (icsk->icsk_ca_ops->cwnd_event) 1193 icsk->icsk_ca_ops->cwnd_event(sk, event); 1194 } 1195 1196 /* From tcp_cong.c */ 1197 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1198 1199 /* From tcp_rate.c */ 1200 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1201 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1202 struct rate_sample *rs); 1203 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1204 bool is_sack_reneg, struct rate_sample *rs); 1205 void tcp_rate_check_app_limited(struct sock *sk); 1206 1207 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1208 { 1209 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1210 } 1211 1212 /* These functions determine how the current flow behaves in respect of SACK 1213 * handling. SACK is negotiated with the peer, and therefore it can vary 1214 * between different flows. 1215 * 1216 * tcp_is_sack - SACK enabled 1217 * tcp_is_reno - No SACK 1218 */ 1219 static inline int tcp_is_sack(const struct tcp_sock *tp) 1220 { 1221 return likely(tp->rx_opt.sack_ok); 1222 } 1223 1224 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1225 { 1226 return !tcp_is_sack(tp); 1227 } 1228 1229 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1230 { 1231 return tp->sacked_out + tp->lost_out; 1232 } 1233 1234 /* This determines how many packets are "in the network" to the best 1235 * of our knowledge. In many cases it is conservative, but where 1236 * detailed information is available from the receiver (via SACK 1237 * blocks etc.) we can make more aggressive calculations. 1238 * 1239 * Use this for decisions involving congestion control, use just 1240 * tp->packets_out to determine if the send queue is empty or not. 1241 * 1242 * Read this equation as: 1243 * 1244 * "Packets sent once on transmission queue" MINUS 1245 * "Packets left network, but not honestly ACKed yet" PLUS 1246 * "Packets fast retransmitted" 1247 */ 1248 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1249 { 1250 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1251 } 1252 1253 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1254 1255 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1256 { 1257 return tp->snd_cwnd; 1258 } 1259 1260 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1261 { 1262 WARN_ON_ONCE((int)val <= 0); 1263 tp->snd_cwnd = val; 1264 } 1265 1266 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1267 { 1268 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1269 } 1270 1271 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1272 { 1273 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1274 } 1275 1276 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1277 { 1278 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1279 (1 << inet_csk(sk)->icsk_ca_state); 1280 } 1281 1282 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1283 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1284 * ssthresh. 1285 */ 1286 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1287 { 1288 const struct tcp_sock *tp = tcp_sk(sk); 1289 1290 if (tcp_in_cwnd_reduction(sk)) 1291 return tp->snd_ssthresh; 1292 else 1293 return max(tp->snd_ssthresh, 1294 ((tcp_snd_cwnd(tp) >> 1) + 1295 (tcp_snd_cwnd(tp) >> 2))); 1296 } 1297 1298 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1299 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1300 1301 void tcp_enter_cwr(struct sock *sk); 1302 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1303 1304 /* The maximum number of MSS of available cwnd for which TSO defers 1305 * sending if not using sysctl_tcp_tso_win_divisor. 1306 */ 1307 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1308 { 1309 return 3; 1310 } 1311 1312 /* Returns end sequence number of the receiver's advertised window */ 1313 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1314 { 1315 return tp->snd_una + tp->snd_wnd; 1316 } 1317 1318 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1319 * flexible approach. The RFC suggests cwnd should not be raised unless 1320 * it was fully used previously. And that's exactly what we do in 1321 * congestion avoidance mode. But in slow start we allow cwnd to grow 1322 * as long as the application has used half the cwnd. 1323 * Example : 1324 * cwnd is 10 (IW10), but application sends 9 frames. 1325 * We allow cwnd to reach 18 when all frames are ACKed. 1326 * This check is safe because it's as aggressive as slow start which already 1327 * risks 100% overshoot. The advantage is that we discourage application to 1328 * either send more filler packets or data to artificially blow up the cwnd 1329 * usage, and allow application-limited process to probe bw more aggressively. 1330 */ 1331 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1332 { 1333 const struct tcp_sock *tp = tcp_sk(sk); 1334 1335 if (tp->is_cwnd_limited) 1336 return true; 1337 1338 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1339 if (tcp_in_slow_start(tp)) 1340 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1341 1342 return false; 1343 } 1344 1345 /* BBR congestion control needs pacing. 1346 * Same remark for SO_MAX_PACING_RATE. 1347 * sch_fq packet scheduler is efficiently handling pacing, 1348 * but is not always installed/used. 1349 * Return true if TCP stack should pace packets itself. 1350 */ 1351 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1352 { 1353 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1354 } 1355 1356 /* Estimates in how many jiffies next packet for this flow can be sent. 1357 * Scheduling a retransmit timer too early would be silly. 1358 */ 1359 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1360 { 1361 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1362 1363 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1364 } 1365 1366 static inline void tcp_reset_xmit_timer(struct sock *sk, 1367 const int what, 1368 unsigned long when, 1369 const unsigned long max_when) 1370 { 1371 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1372 max_when); 1373 } 1374 1375 /* Something is really bad, we could not queue an additional packet, 1376 * because qdisc is full or receiver sent a 0 window, or we are paced. 1377 * We do not want to add fuel to the fire, or abort too early, 1378 * so make sure the timer we arm now is at least 200ms in the future, 1379 * regardless of current icsk_rto value (as it could be ~2ms) 1380 */ 1381 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1382 { 1383 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1384 } 1385 1386 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1387 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1388 unsigned long max_when) 1389 { 1390 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1391 inet_csk(sk)->icsk_backoff); 1392 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1393 1394 return (unsigned long)min_t(u64, when, max_when); 1395 } 1396 1397 static inline void tcp_check_probe_timer(struct sock *sk) 1398 { 1399 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1400 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1401 tcp_probe0_base(sk), TCP_RTO_MAX); 1402 } 1403 1404 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1405 { 1406 tp->snd_wl1 = seq; 1407 } 1408 1409 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1410 { 1411 tp->snd_wl1 = seq; 1412 } 1413 1414 /* 1415 * Calculate(/check) TCP checksum 1416 */ 1417 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1418 __be32 daddr, __wsum base) 1419 { 1420 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1421 } 1422 1423 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1424 { 1425 return !skb_csum_unnecessary(skb) && 1426 __skb_checksum_complete(skb); 1427 } 1428 1429 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1430 enum skb_drop_reason *reason); 1431 1432 1433 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1434 void tcp_set_state(struct sock *sk, int state); 1435 void tcp_done(struct sock *sk); 1436 int tcp_abort(struct sock *sk, int err); 1437 1438 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1439 { 1440 rx_opt->dsack = 0; 1441 rx_opt->num_sacks = 0; 1442 } 1443 1444 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1445 1446 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1447 { 1448 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1449 struct tcp_sock *tp = tcp_sk(sk); 1450 s32 delta; 1451 1452 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1453 tp->packets_out || ca_ops->cong_control) 1454 return; 1455 delta = tcp_jiffies32 - tp->lsndtime; 1456 if (delta > inet_csk(sk)->icsk_rto) 1457 tcp_cwnd_restart(sk, delta); 1458 } 1459 1460 /* Determine a window scaling and initial window to offer. */ 1461 void tcp_select_initial_window(const struct sock *sk, int __space, 1462 __u32 mss, __u32 *rcv_wnd, 1463 __u32 *window_clamp, int wscale_ok, 1464 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1465 1466 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1467 { 1468 s64 scaled_space = (s64)space * scaling_ratio; 1469 1470 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1471 } 1472 1473 static inline int tcp_win_from_space(const struct sock *sk, int space) 1474 { 1475 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1476 } 1477 1478 /* inverse of __tcp_win_from_space() */ 1479 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1480 { 1481 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1482 1483 do_div(val, scaling_ratio); 1484 return val; 1485 } 1486 1487 static inline int tcp_space_from_win(const struct sock *sk, int win) 1488 { 1489 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1490 } 1491 1492 static inline void tcp_scaling_ratio_init(struct sock *sk) 1493 { 1494 /* Assume a conservative default of 1200 bytes of payload per 4K page. 1495 * This may be adjusted later in tcp_measure_rcv_mss(). 1496 */ 1497 tcp_sk(sk)->scaling_ratio = (1200 << TCP_RMEM_TO_WIN_SCALE) / 1498 SKB_TRUESIZE(4096); 1499 } 1500 1501 /* Note: caller must be prepared to deal with negative returns */ 1502 static inline int tcp_space(const struct sock *sk) 1503 { 1504 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1505 READ_ONCE(sk->sk_backlog.len) - 1506 atomic_read(&sk->sk_rmem_alloc)); 1507 } 1508 1509 static inline int tcp_full_space(const struct sock *sk) 1510 { 1511 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1512 } 1513 1514 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1515 { 1516 int unused_mem = sk_unused_reserved_mem(sk); 1517 struct tcp_sock *tp = tcp_sk(sk); 1518 1519 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1520 if (unused_mem) 1521 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1522 tcp_win_from_space(sk, unused_mem)); 1523 } 1524 1525 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1526 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1527 1528 1529 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1530 * If 87.5 % (7/8) of the space has been consumed, we want to override 1531 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1532 * len/truesize ratio. 1533 */ 1534 static inline bool tcp_rmem_pressure(const struct sock *sk) 1535 { 1536 int rcvbuf, threshold; 1537 1538 if (tcp_under_memory_pressure(sk)) 1539 return true; 1540 1541 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1542 threshold = rcvbuf - (rcvbuf >> 3); 1543 1544 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1545 } 1546 1547 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1548 { 1549 const struct tcp_sock *tp = tcp_sk(sk); 1550 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1551 1552 if (avail <= 0) 1553 return false; 1554 1555 return (avail >= target) || tcp_rmem_pressure(sk) || 1556 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1557 } 1558 1559 extern void tcp_openreq_init_rwin(struct request_sock *req, 1560 const struct sock *sk_listener, 1561 const struct dst_entry *dst); 1562 1563 void tcp_enter_memory_pressure(struct sock *sk); 1564 void tcp_leave_memory_pressure(struct sock *sk); 1565 1566 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1567 { 1568 struct net *net = sock_net((struct sock *)tp); 1569 int val; 1570 1571 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1572 * and do_tcp_setsockopt(). 1573 */ 1574 val = READ_ONCE(tp->keepalive_intvl); 1575 1576 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1577 } 1578 1579 static inline int keepalive_time_when(const struct tcp_sock *tp) 1580 { 1581 struct net *net = sock_net((struct sock *)tp); 1582 int val; 1583 1584 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1585 val = READ_ONCE(tp->keepalive_time); 1586 1587 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1588 } 1589 1590 static inline int keepalive_probes(const struct tcp_sock *tp) 1591 { 1592 struct net *net = sock_net((struct sock *)tp); 1593 int val; 1594 1595 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1596 * and do_tcp_setsockopt(). 1597 */ 1598 val = READ_ONCE(tp->keepalive_probes); 1599 1600 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1601 } 1602 1603 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1604 { 1605 const struct inet_connection_sock *icsk = &tp->inet_conn; 1606 1607 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1608 tcp_jiffies32 - tp->rcv_tstamp); 1609 } 1610 1611 static inline int tcp_fin_time(const struct sock *sk) 1612 { 1613 int fin_timeout = tcp_sk(sk)->linger2 ? : 1614 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1615 const int rto = inet_csk(sk)->icsk_rto; 1616 1617 if (fin_timeout < (rto << 2) - (rto >> 1)) 1618 fin_timeout = (rto << 2) - (rto >> 1); 1619 1620 return fin_timeout; 1621 } 1622 1623 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1624 int paws_win) 1625 { 1626 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1627 return true; 1628 if (unlikely(!time_before32(ktime_get_seconds(), 1629 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1630 return true; 1631 /* 1632 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1633 * then following tcp messages have valid values. Ignore 0 value, 1634 * or else 'negative' tsval might forbid us to accept their packets. 1635 */ 1636 if (!rx_opt->ts_recent) 1637 return true; 1638 return false; 1639 } 1640 1641 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1642 int rst) 1643 { 1644 if (tcp_paws_check(rx_opt, 0)) 1645 return false; 1646 1647 /* RST segments are not recommended to carry timestamp, 1648 and, if they do, it is recommended to ignore PAWS because 1649 "their cleanup function should take precedence over timestamps." 1650 Certainly, it is mistake. It is necessary to understand the reasons 1651 of this constraint to relax it: if peer reboots, clock may go 1652 out-of-sync and half-open connections will not be reset. 1653 Actually, the problem would be not existing if all 1654 the implementations followed draft about maintaining clock 1655 via reboots. Linux-2.2 DOES NOT! 1656 1657 However, we can relax time bounds for RST segments to MSL. 1658 */ 1659 if (rst && !time_before32(ktime_get_seconds(), 1660 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1661 return false; 1662 return true; 1663 } 1664 1665 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1666 int mib_idx, u32 *last_oow_ack_time); 1667 1668 static inline void tcp_mib_init(struct net *net) 1669 { 1670 /* See RFC 2012 */ 1671 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1672 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1673 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1674 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1675 } 1676 1677 /* from STCP */ 1678 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1679 { 1680 tp->lost_skb_hint = NULL; 1681 } 1682 1683 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1684 { 1685 tcp_clear_retrans_hints_partial(tp); 1686 tp->retransmit_skb_hint = NULL; 1687 } 1688 1689 union tcp_md5_addr { 1690 struct in_addr a4; 1691 #if IS_ENABLED(CONFIG_IPV6) 1692 struct in6_addr a6; 1693 #endif 1694 }; 1695 1696 /* - key database */ 1697 struct tcp_md5sig_key { 1698 struct hlist_node node; 1699 u8 keylen; 1700 u8 family; /* AF_INET or AF_INET6 */ 1701 u8 prefixlen; 1702 u8 flags; 1703 union tcp_md5_addr addr; 1704 int l3index; /* set if key added with L3 scope */ 1705 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1706 struct rcu_head rcu; 1707 }; 1708 1709 /* - sock block */ 1710 struct tcp_md5sig_info { 1711 struct hlist_head head; 1712 struct rcu_head rcu; 1713 }; 1714 1715 /* - pseudo header */ 1716 struct tcp4_pseudohdr { 1717 __be32 saddr; 1718 __be32 daddr; 1719 __u8 pad; 1720 __u8 protocol; 1721 __be16 len; 1722 }; 1723 1724 struct tcp6_pseudohdr { 1725 struct in6_addr saddr; 1726 struct in6_addr daddr; 1727 __be32 len; 1728 __be32 protocol; /* including padding */ 1729 }; 1730 1731 union tcp_md5sum_block { 1732 struct tcp4_pseudohdr ip4; 1733 #if IS_ENABLED(CONFIG_IPV6) 1734 struct tcp6_pseudohdr ip6; 1735 #endif 1736 }; 1737 1738 /* - pool: digest algorithm, hash description and scratch buffer */ 1739 struct tcp_md5sig_pool { 1740 struct ahash_request *md5_req; 1741 void *scratch; 1742 }; 1743 1744 /* - functions */ 1745 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1746 const struct sock *sk, const struct sk_buff *skb); 1747 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1748 int family, u8 prefixlen, int l3index, u8 flags, 1749 const u8 *newkey, u8 newkeylen); 1750 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1751 int family, u8 prefixlen, int l3index, 1752 struct tcp_md5sig_key *key); 1753 1754 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1755 int family, u8 prefixlen, int l3index, u8 flags); 1756 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1757 const struct sock *addr_sk); 1758 1759 #ifdef CONFIG_TCP_MD5SIG 1760 #include <linux/jump_label.h> 1761 extern struct static_key_false_deferred tcp_md5_needed; 1762 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1763 const union tcp_md5_addr *addr, 1764 int family); 1765 static inline struct tcp_md5sig_key * 1766 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1767 const union tcp_md5_addr *addr, int family) 1768 { 1769 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1770 return NULL; 1771 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1772 } 1773 1774 enum skb_drop_reason 1775 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1776 const void *saddr, const void *daddr, 1777 int family, int dif, int sdif); 1778 1779 1780 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1781 #else 1782 static inline struct tcp_md5sig_key * 1783 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1784 const union tcp_md5_addr *addr, int family) 1785 { 1786 return NULL; 1787 } 1788 1789 static inline enum skb_drop_reason 1790 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1791 const void *saddr, const void *daddr, 1792 int family, int dif, int sdif) 1793 { 1794 return SKB_NOT_DROPPED_YET; 1795 } 1796 #define tcp_twsk_md5_key(twsk) NULL 1797 #endif 1798 1799 bool tcp_alloc_md5sig_pool(void); 1800 1801 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1802 static inline void tcp_put_md5sig_pool(void) 1803 { 1804 local_bh_enable(); 1805 } 1806 1807 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1808 unsigned int header_len); 1809 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1810 const struct tcp_md5sig_key *key); 1811 1812 /* From tcp_fastopen.c */ 1813 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1814 struct tcp_fastopen_cookie *cookie); 1815 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1816 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1817 u16 try_exp); 1818 struct tcp_fastopen_request { 1819 /* Fast Open cookie. Size 0 means a cookie request */ 1820 struct tcp_fastopen_cookie cookie; 1821 struct msghdr *data; /* data in MSG_FASTOPEN */ 1822 size_t size; 1823 int copied; /* queued in tcp_connect() */ 1824 struct ubuf_info *uarg; 1825 }; 1826 void tcp_free_fastopen_req(struct tcp_sock *tp); 1827 void tcp_fastopen_destroy_cipher(struct sock *sk); 1828 void tcp_fastopen_ctx_destroy(struct net *net); 1829 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1830 void *primary_key, void *backup_key); 1831 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1832 u64 *key); 1833 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1834 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1835 struct request_sock *req, 1836 struct tcp_fastopen_cookie *foc, 1837 const struct dst_entry *dst); 1838 void tcp_fastopen_init_key_once(struct net *net); 1839 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1840 struct tcp_fastopen_cookie *cookie); 1841 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1842 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1843 #define TCP_FASTOPEN_KEY_MAX 2 1844 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1845 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1846 1847 /* Fastopen key context */ 1848 struct tcp_fastopen_context { 1849 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1850 int num; 1851 struct rcu_head rcu; 1852 }; 1853 1854 void tcp_fastopen_active_disable(struct sock *sk); 1855 bool tcp_fastopen_active_should_disable(struct sock *sk); 1856 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1857 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1858 1859 /* Caller needs to wrap with rcu_read_(un)lock() */ 1860 static inline 1861 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1862 { 1863 struct tcp_fastopen_context *ctx; 1864 1865 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1866 if (!ctx) 1867 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1868 return ctx; 1869 } 1870 1871 static inline 1872 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1873 const struct tcp_fastopen_cookie *orig) 1874 { 1875 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1876 orig->len == foc->len && 1877 !memcmp(orig->val, foc->val, foc->len)) 1878 return true; 1879 return false; 1880 } 1881 1882 static inline 1883 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1884 { 1885 return ctx->num; 1886 } 1887 1888 /* Latencies incurred by various limits for a sender. They are 1889 * chronograph-like stats that are mutually exclusive. 1890 */ 1891 enum tcp_chrono { 1892 TCP_CHRONO_UNSPEC, 1893 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1894 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1895 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1896 __TCP_CHRONO_MAX, 1897 }; 1898 1899 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1900 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1901 1902 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1903 * the same memory storage than skb->destructor/_skb_refdst 1904 */ 1905 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1906 { 1907 skb->destructor = NULL; 1908 skb->_skb_refdst = 0UL; 1909 } 1910 1911 #define tcp_skb_tsorted_save(skb) { \ 1912 unsigned long _save = skb->_skb_refdst; \ 1913 skb->_skb_refdst = 0UL; 1914 1915 #define tcp_skb_tsorted_restore(skb) \ 1916 skb->_skb_refdst = _save; \ 1917 } 1918 1919 void tcp_write_queue_purge(struct sock *sk); 1920 1921 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1922 { 1923 return skb_rb_first(&sk->tcp_rtx_queue); 1924 } 1925 1926 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1927 { 1928 return skb_rb_last(&sk->tcp_rtx_queue); 1929 } 1930 1931 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1932 { 1933 return skb_peek_tail(&sk->sk_write_queue); 1934 } 1935 1936 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1937 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1938 1939 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1940 { 1941 return skb_peek(&sk->sk_write_queue); 1942 } 1943 1944 static inline bool tcp_skb_is_last(const struct sock *sk, 1945 const struct sk_buff *skb) 1946 { 1947 return skb_queue_is_last(&sk->sk_write_queue, skb); 1948 } 1949 1950 /** 1951 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1952 * @sk: socket 1953 * 1954 * Since the write queue can have a temporary empty skb in it, 1955 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1956 */ 1957 static inline bool tcp_write_queue_empty(const struct sock *sk) 1958 { 1959 const struct tcp_sock *tp = tcp_sk(sk); 1960 1961 return tp->write_seq == tp->snd_nxt; 1962 } 1963 1964 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1965 { 1966 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1967 } 1968 1969 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1970 { 1971 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1972 } 1973 1974 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1975 { 1976 __skb_queue_tail(&sk->sk_write_queue, skb); 1977 1978 /* Queue it, remembering where we must start sending. */ 1979 if (sk->sk_write_queue.next == skb) 1980 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1981 } 1982 1983 /* Insert new before skb on the write queue of sk. */ 1984 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1985 struct sk_buff *skb, 1986 struct sock *sk) 1987 { 1988 __skb_queue_before(&sk->sk_write_queue, skb, new); 1989 } 1990 1991 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1992 { 1993 tcp_skb_tsorted_anchor_cleanup(skb); 1994 __skb_unlink(skb, &sk->sk_write_queue); 1995 } 1996 1997 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1998 1999 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2000 { 2001 tcp_skb_tsorted_anchor_cleanup(skb); 2002 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2003 } 2004 2005 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2006 { 2007 list_del(&skb->tcp_tsorted_anchor); 2008 tcp_rtx_queue_unlink(skb, sk); 2009 tcp_wmem_free_skb(sk, skb); 2010 } 2011 2012 static inline void tcp_push_pending_frames(struct sock *sk) 2013 { 2014 if (tcp_send_head(sk)) { 2015 struct tcp_sock *tp = tcp_sk(sk); 2016 2017 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2018 } 2019 } 2020 2021 /* Start sequence of the skb just after the highest skb with SACKed 2022 * bit, valid only if sacked_out > 0 or when the caller has ensured 2023 * validity by itself. 2024 */ 2025 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2026 { 2027 if (!tp->sacked_out) 2028 return tp->snd_una; 2029 2030 if (tp->highest_sack == NULL) 2031 return tp->snd_nxt; 2032 2033 return TCP_SKB_CB(tp->highest_sack)->seq; 2034 } 2035 2036 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2037 { 2038 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2039 } 2040 2041 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2042 { 2043 return tcp_sk(sk)->highest_sack; 2044 } 2045 2046 static inline void tcp_highest_sack_reset(struct sock *sk) 2047 { 2048 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2049 } 2050 2051 /* Called when old skb is about to be deleted and replaced by new skb */ 2052 static inline void tcp_highest_sack_replace(struct sock *sk, 2053 struct sk_buff *old, 2054 struct sk_buff *new) 2055 { 2056 if (old == tcp_highest_sack(sk)) 2057 tcp_sk(sk)->highest_sack = new; 2058 } 2059 2060 /* This helper checks if socket has IP_TRANSPARENT set */ 2061 static inline bool inet_sk_transparent(const struct sock *sk) 2062 { 2063 switch (sk->sk_state) { 2064 case TCP_TIME_WAIT: 2065 return inet_twsk(sk)->tw_transparent; 2066 case TCP_NEW_SYN_RECV: 2067 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2068 } 2069 return inet_test_bit(TRANSPARENT, sk); 2070 } 2071 2072 /* Determines whether this is a thin stream (which may suffer from 2073 * increased latency). Used to trigger latency-reducing mechanisms. 2074 */ 2075 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2076 { 2077 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2078 } 2079 2080 /* /proc */ 2081 enum tcp_seq_states { 2082 TCP_SEQ_STATE_LISTENING, 2083 TCP_SEQ_STATE_ESTABLISHED, 2084 }; 2085 2086 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2087 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2088 void tcp_seq_stop(struct seq_file *seq, void *v); 2089 2090 struct tcp_seq_afinfo { 2091 sa_family_t family; 2092 }; 2093 2094 struct tcp_iter_state { 2095 struct seq_net_private p; 2096 enum tcp_seq_states state; 2097 struct sock *syn_wait_sk; 2098 int bucket, offset, sbucket, num; 2099 loff_t last_pos; 2100 }; 2101 2102 extern struct request_sock_ops tcp_request_sock_ops; 2103 extern struct request_sock_ops tcp6_request_sock_ops; 2104 2105 void tcp_v4_destroy_sock(struct sock *sk); 2106 2107 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2108 netdev_features_t features); 2109 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2110 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2111 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2112 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2113 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2114 #ifdef CONFIG_INET 2115 void tcp_gro_complete(struct sk_buff *skb); 2116 #else 2117 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2118 #endif 2119 2120 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2121 2122 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2123 { 2124 struct net *net = sock_net((struct sock *)tp); 2125 u32 val; 2126 2127 val = READ_ONCE(tp->notsent_lowat); 2128 2129 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2130 } 2131 2132 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2133 2134 #ifdef CONFIG_PROC_FS 2135 int tcp4_proc_init(void); 2136 void tcp4_proc_exit(void); 2137 #endif 2138 2139 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2140 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2141 const struct tcp_request_sock_ops *af_ops, 2142 struct sock *sk, struct sk_buff *skb); 2143 2144 /* TCP af-specific functions */ 2145 struct tcp_sock_af_ops { 2146 #ifdef CONFIG_TCP_MD5SIG 2147 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2148 const struct sock *addr_sk); 2149 int (*calc_md5_hash)(char *location, 2150 const struct tcp_md5sig_key *md5, 2151 const struct sock *sk, 2152 const struct sk_buff *skb); 2153 int (*md5_parse)(struct sock *sk, 2154 int optname, 2155 sockptr_t optval, 2156 int optlen); 2157 #endif 2158 }; 2159 2160 struct tcp_request_sock_ops { 2161 u16 mss_clamp; 2162 #ifdef CONFIG_TCP_MD5SIG 2163 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2164 const struct sock *addr_sk); 2165 int (*calc_md5_hash) (char *location, 2166 const struct tcp_md5sig_key *md5, 2167 const struct sock *sk, 2168 const struct sk_buff *skb); 2169 #endif 2170 #ifdef CONFIG_SYN_COOKIES 2171 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2172 __u16 *mss); 2173 #endif 2174 struct dst_entry *(*route_req)(const struct sock *sk, 2175 struct sk_buff *skb, 2176 struct flowi *fl, 2177 struct request_sock *req); 2178 u32 (*init_seq)(const struct sk_buff *skb); 2179 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2180 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2181 struct flowi *fl, struct request_sock *req, 2182 struct tcp_fastopen_cookie *foc, 2183 enum tcp_synack_type synack_type, 2184 struct sk_buff *syn_skb); 2185 }; 2186 2187 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2188 #if IS_ENABLED(CONFIG_IPV6) 2189 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2190 #endif 2191 2192 #ifdef CONFIG_SYN_COOKIES 2193 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2194 const struct sock *sk, struct sk_buff *skb, 2195 __u16 *mss) 2196 { 2197 tcp_synq_overflow(sk); 2198 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2199 return ops->cookie_init_seq(skb, mss); 2200 } 2201 #else 2202 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2203 const struct sock *sk, struct sk_buff *skb, 2204 __u16 *mss) 2205 { 2206 return 0; 2207 } 2208 #endif 2209 2210 int tcpv4_offload_init(void); 2211 2212 void tcp_v4_init(void); 2213 void tcp_init(void); 2214 2215 /* tcp_recovery.c */ 2216 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2217 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2218 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2219 u32 reo_wnd); 2220 extern bool tcp_rack_mark_lost(struct sock *sk); 2221 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2222 u64 xmit_time); 2223 extern void tcp_rack_reo_timeout(struct sock *sk); 2224 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2225 2226 /* tcp_plb.c */ 2227 2228 /* 2229 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2230 * expects cong_ratio which represents fraction of traffic that experienced 2231 * congestion over a single RTT. In order to avoid floating point operations, 2232 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2233 */ 2234 #define TCP_PLB_SCALE 8 2235 2236 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2237 struct tcp_plb_state { 2238 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2239 unused:3; 2240 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2241 }; 2242 2243 static inline void tcp_plb_init(const struct sock *sk, 2244 struct tcp_plb_state *plb) 2245 { 2246 plb->consec_cong_rounds = 0; 2247 plb->pause_until = 0; 2248 } 2249 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2250 const int cong_ratio); 2251 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2252 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2253 2254 /* At how many usecs into the future should the RTO fire? */ 2255 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2256 { 2257 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2258 u32 rto = inet_csk(sk)->icsk_rto; 2259 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2260 2261 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2262 } 2263 2264 /* 2265 * Save and compile IPv4 options, return a pointer to it 2266 */ 2267 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2268 struct sk_buff *skb) 2269 { 2270 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2271 struct ip_options_rcu *dopt = NULL; 2272 2273 if (opt->optlen) { 2274 int opt_size = sizeof(*dopt) + opt->optlen; 2275 2276 dopt = kmalloc(opt_size, GFP_ATOMIC); 2277 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2278 kfree(dopt); 2279 dopt = NULL; 2280 } 2281 } 2282 return dopt; 2283 } 2284 2285 /* locally generated TCP pure ACKs have skb->truesize == 2 2286 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2287 * This is much faster than dissecting the packet to find out. 2288 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2289 */ 2290 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2291 { 2292 return skb->truesize == 2; 2293 } 2294 2295 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2296 { 2297 skb->truesize = 2; 2298 } 2299 2300 static inline int tcp_inq(struct sock *sk) 2301 { 2302 struct tcp_sock *tp = tcp_sk(sk); 2303 int answ; 2304 2305 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2306 answ = 0; 2307 } else if (sock_flag(sk, SOCK_URGINLINE) || 2308 !tp->urg_data || 2309 before(tp->urg_seq, tp->copied_seq) || 2310 !before(tp->urg_seq, tp->rcv_nxt)) { 2311 2312 answ = tp->rcv_nxt - tp->copied_seq; 2313 2314 /* Subtract 1, if FIN was received */ 2315 if (answ && sock_flag(sk, SOCK_DONE)) 2316 answ--; 2317 } else { 2318 answ = tp->urg_seq - tp->copied_seq; 2319 } 2320 2321 return answ; 2322 } 2323 2324 int tcp_peek_len(struct socket *sock); 2325 2326 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2327 { 2328 u16 segs_in; 2329 2330 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2331 2332 /* We update these fields while other threads might 2333 * read them from tcp_get_info() 2334 */ 2335 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2336 if (skb->len > tcp_hdrlen(skb)) 2337 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2338 } 2339 2340 /* 2341 * TCP listen path runs lockless. 2342 * We forced "struct sock" to be const qualified to make sure 2343 * we don't modify one of its field by mistake. 2344 * Here, we increment sk_drops which is an atomic_t, so we can safely 2345 * make sock writable again. 2346 */ 2347 static inline void tcp_listendrop(const struct sock *sk) 2348 { 2349 atomic_inc(&((struct sock *)sk)->sk_drops); 2350 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2351 } 2352 2353 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2354 2355 /* 2356 * Interface for adding Upper Level Protocols over TCP 2357 */ 2358 2359 #define TCP_ULP_NAME_MAX 16 2360 #define TCP_ULP_MAX 128 2361 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2362 2363 struct tcp_ulp_ops { 2364 struct list_head list; 2365 2366 /* initialize ulp */ 2367 int (*init)(struct sock *sk); 2368 /* update ulp */ 2369 void (*update)(struct sock *sk, struct proto *p, 2370 void (*write_space)(struct sock *sk)); 2371 /* cleanup ulp */ 2372 void (*release)(struct sock *sk); 2373 /* diagnostic */ 2374 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2375 size_t (*get_info_size)(const struct sock *sk); 2376 /* clone ulp */ 2377 void (*clone)(const struct request_sock *req, struct sock *newsk, 2378 const gfp_t priority); 2379 2380 char name[TCP_ULP_NAME_MAX]; 2381 struct module *owner; 2382 }; 2383 int tcp_register_ulp(struct tcp_ulp_ops *type); 2384 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2385 int tcp_set_ulp(struct sock *sk, const char *name); 2386 void tcp_get_available_ulp(char *buf, size_t len); 2387 void tcp_cleanup_ulp(struct sock *sk); 2388 void tcp_update_ulp(struct sock *sk, struct proto *p, 2389 void (*write_space)(struct sock *sk)); 2390 2391 #define MODULE_ALIAS_TCP_ULP(name) \ 2392 __MODULE_INFO(alias, alias_userspace, name); \ 2393 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2394 2395 #ifdef CONFIG_NET_SOCK_MSG 2396 struct sk_msg; 2397 struct sk_psock; 2398 2399 #ifdef CONFIG_BPF_SYSCALL 2400 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2401 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2402 #endif /* CONFIG_BPF_SYSCALL */ 2403 2404 #ifdef CONFIG_INET 2405 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2406 #else 2407 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2408 { 2409 } 2410 #endif 2411 2412 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2413 struct sk_msg *msg, u32 bytes, int flags); 2414 #endif /* CONFIG_NET_SOCK_MSG */ 2415 2416 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2417 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2418 { 2419 } 2420 #endif 2421 2422 #ifdef CONFIG_CGROUP_BPF 2423 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2424 struct sk_buff *skb, 2425 unsigned int end_offset) 2426 { 2427 skops->skb = skb; 2428 skops->skb_data_end = skb->data + end_offset; 2429 } 2430 #else 2431 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2432 struct sk_buff *skb, 2433 unsigned int end_offset) 2434 { 2435 } 2436 #endif 2437 2438 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2439 * is < 0, then the BPF op failed (for example if the loaded BPF 2440 * program does not support the chosen operation or there is no BPF 2441 * program loaded). 2442 */ 2443 #ifdef CONFIG_BPF 2444 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2445 { 2446 struct bpf_sock_ops_kern sock_ops; 2447 int ret; 2448 2449 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2450 if (sk_fullsock(sk)) { 2451 sock_ops.is_fullsock = 1; 2452 sock_owned_by_me(sk); 2453 } 2454 2455 sock_ops.sk = sk; 2456 sock_ops.op = op; 2457 if (nargs > 0) 2458 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2459 2460 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2461 if (ret == 0) 2462 ret = sock_ops.reply; 2463 else 2464 ret = -1; 2465 return ret; 2466 } 2467 2468 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2469 { 2470 u32 args[2] = {arg1, arg2}; 2471 2472 return tcp_call_bpf(sk, op, 2, args); 2473 } 2474 2475 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2476 u32 arg3) 2477 { 2478 u32 args[3] = {arg1, arg2, arg3}; 2479 2480 return tcp_call_bpf(sk, op, 3, args); 2481 } 2482 2483 #else 2484 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2485 { 2486 return -EPERM; 2487 } 2488 2489 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2490 { 2491 return -EPERM; 2492 } 2493 2494 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2495 u32 arg3) 2496 { 2497 return -EPERM; 2498 } 2499 2500 #endif 2501 2502 static inline u32 tcp_timeout_init(struct sock *sk) 2503 { 2504 int timeout; 2505 2506 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2507 2508 if (timeout <= 0) 2509 timeout = TCP_TIMEOUT_INIT; 2510 return min_t(int, timeout, TCP_RTO_MAX); 2511 } 2512 2513 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2514 { 2515 int rwnd; 2516 2517 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2518 2519 if (rwnd < 0) 2520 rwnd = 0; 2521 return rwnd; 2522 } 2523 2524 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2525 { 2526 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2527 } 2528 2529 static inline void tcp_bpf_rtt(struct sock *sk) 2530 { 2531 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2532 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2533 } 2534 2535 #if IS_ENABLED(CONFIG_SMC) 2536 extern struct static_key_false tcp_have_smc; 2537 #endif 2538 2539 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2540 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2541 void (*cad)(struct sock *sk, u32 ack_seq)); 2542 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2543 void clean_acked_data_flush(void); 2544 #endif 2545 2546 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2547 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2548 const struct tcp_sock *tp) 2549 { 2550 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2551 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2552 } 2553 2554 /* Compute Earliest Departure Time for some control packets 2555 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2556 */ 2557 static inline u64 tcp_transmit_time(const struct sock *sk) 2558 { 2559 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2560 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2561 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2562 2563 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2564 } 2565 return 0; 2566 } 2567 2568 #endif /* _TCP_H */ 2569