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, int family); 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_reset(struct sock *sk, struct sk_buff *skb); 628 void tcp_fin(struct sock *sk); 629 void tcp_check_space(struct sock *sk); 630 void tcp_sack_compress_send_ack(struct sock *sk); 631 632 /* tcp_timer.c */ 633 void tcp_init_xmit_timers(struct sock *); 634 static inline void tcp_clear_xmit_timers(struct sock *sk) 635 { 636 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 637 __sock_put(sk); 638 639 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 640 __sock_put(sk); 641 642 inet_csk_clear_xmit_timers(sk); 643 } 644 645 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 646 unsigned int tcp_current_mss(struct sock *sk); 647 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 648 649 /* Bound MSS / TSO packet size with the half of the window */ 650 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 651 { 652 int cutoff; 653 654 /* When peer uses tiny windows, there is no use in packetizing 655 * to sub-MSS pieces for the sake of SWS or making sure there 656 * are enough packets in the pipe for fast recovery. 657 * 658 * On the other hand, for extremely large MSS devices, handling 659 * smaller than MSS windows in this way does make sense. 660 */ 661 if (tp->max_window > TCP_MSS_DEFAULT) 662 cutoff = (tp->max_window >> 1); 663 else 664 cutoff = tp->max_window; 665 666 if (cutoff && pktsize > cutoff) 667 return max_t(int, cutoff, 68U - tp->tcp_header_len); 668 else 669 return pktsize; 670 } 671 672 /* tcp.c */ 673 void tcp_get_info(struct sock *, struct tcp_info *); 674 675 /* Read 'sendfile()'-style from a TCP socket */ 676 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 677 sk_read_actor_t recv_actor); 678 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 679 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 680 void tcp_read_done(struct sock *sk, size_t len); 681 682 void tcp_initialize_rcv_mss(struct sock *sk); 683 684 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 685 int tcp_mss_to_mtu(struct sock *sk, int mss); 686 void tcp_mtup_init(struct sock *sk); 687 688 static inline void tcp_bound_rto(const struct sock *sk) 689 { 690 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 691 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 692 } 693 694 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 695 { 696 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 697 } 698 699 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 700 { 701 /* mptcp hooks are only on the slow path */ 702 if (sk_is_mptcp((struct sock *)tp)) 703 return; 704 705 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 706 ntohl(TCP_FLAG_ACK) | 707 snd_wnd); 708 } 709 710 static inline void tcp_fast_path_on(struct tcp_sock *tp) 711 { 712 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 713 } 714 715 static inline void tcp_fast_path_check(struct sock *sk) 716 { 717 struct tcp_sock *tp = tcp_sk(sk); 718 719 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 720 tp->rcv_wnd && 721 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 722 !tp->urg_data) 723 tcp_fast_path_on(tp); 724 } 725 726 u32 tcp_delack_max(const struct sock *sk); 727 728 /* Compute the actual rto_min value */ 729 static inline u32 tcp_rto_min(struct sock *sk) 730 { 731 const struct dst_entry *dst = __sk_dst_get(sk); 732 u32 rto_min = inet_csk(sk)->icsk_rto_min; 733 734 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 735 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 736 return rto_min; 737 } 738 739 static inline u32 tcp_rto_min_us(struct sock *sk) 740 { 741 return jiffies_to_usecs(tcp_rto_min(sk)); 742 } 743 744 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 745 { 746 return dst_metric_locked(dst, RTAX_CC_ALGO); 747 } 748 749 /* Minimum RTT in usec. ~0 means not available. */ 750 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 751 { 752 return minmax_get(&tp->rtt_min); 753 } 754 755 /* Compute the actual receive window we are currently advertising. 756 * Rcv_nxt can be after the window if our peer push more data 757 * than the offered window. 758 */ 759 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 760 { 761 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 762 763 if (win < 0) 764 win = 0; 765 return (u32) win; 766 } 767 768 /* Choose a new window, without checks for shrinking, and without 769 * scaling applied to the result. The caller does these things 770 * if necessary. This is a "raw" window selection. 771 */ 772 u32 __tcp_select_window(struct sock *sk); 773 774 void tcp_send_window_probe(struct sock *sk); 775 776 /* TCP uses 32bit jiffies to save some space. 777 * Note that this is different from tcp_time_stamp, which 778 * historically has been the same until linux-4.13. 779 */ 780 #define tcp_jiffies32 ((u32)jiffies) 781 782 /* 783 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 784 * It is no longer tied to jiffies, but to 1 ms clock. 785 * Note: double check if you want to use tcp_jiffies32 instead of this. 786 */ 787 #define TCP_TS_HZ 1000 788 789 static inline u64 tcp_clock_ns(void) 790 { 791 return ktime_get_ns(); 792 } 793 794 static inline u64 tcp_clock_us(void) 795 { 796 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 797 } 798 799 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 800 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 801 { 802 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 803 } 804 805 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 806 static inline u64 tcp_ns_to_ts(u64 ns) 807 { 808 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 809 } 810 811 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 812 static inline u32 tcp_time_stamp_raw(void) 813 { 814 return tcp_ns_to_ts(tcp_clock_ns()); 815 } 816 817 void tcp_mstamp_refresh(struct tcp_sock *tp); 818 819 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 820 { 821 return max_t(s64, t1 - t0, 0); 822 } 823 824 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 825 { 826 return tcp_ns_to_ts(skb->skb_mstamp_ns); 827 } 828 829 /* provide the departure time in us unit */ 830 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 831 { 832 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 833 } 834 835 836 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 837 838 #define TCPHDR_FIN 0x01 839 #define TCPHDR_SYN 0x02 840 #define TCPHDR_RST 0x04 841 #define TCPHDR_PSH 0x08 842 #define TCPHDR_ACK 0x10 843 #define TCPHDR_URG 0x20 844 #define TCPHDR_ECE 0x40 845 #define TCPHDR_CWR 0x80 846 847 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 848 849 /* This is what the send packet queuing engine uses to pass 850 * TCP per-packet control information to the transmission code. 851 * We also store the host-order sequence numbers in here too. 852 * This is 44 bytes if IPV6 is enabled. 853 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 854 */ 855 struct tcp_skb_cb { 856 __u32 seq; /* Starting sequence number */ 857 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 858 union { 859 /* Note : tcp_tw_isn is used in input path only 860 * (isn chosen by tcp_timewait_state_process()) 861 * 862 * tcp_gso_segs/size are used in write queue only, 863 * cf tcp_skb_pcount()/tcp_skb_mss() 864 */ 865 __u32 tcp_tw_isn; 866 struct { 867 u16 tcp_gso_segs; 868 u16 tcp_gso_size; 869 }; 870 }; 871 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 872 873 __u8 sacked; /* State flags for SACK. */ 874 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 875 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 876 #define TCPCB_LOST 0x04 /* SKB is lost */ 877 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 878 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 879 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 880 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 881 TCPCB_REPAIRED) 882 883 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 884 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 885 eor:1, /* Is skb MSG_EOR marked? */ 886 has_rxtstamp:1, /* SKB has a RX timestamp */ 887 unused:5; 888 __u32 ack_seq; /* Sequence number ACK'd */ 889 union { 890 struct { 891 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 892 /* There is space for up to 24 bytes */ 893 __u32 is_app_limited:1, /* cwnd not fully used? */ 894 delivered_ce:20, 895 unused:11; 896 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 897 __u32 delivered; 898 /* start of send pipeline phase */ 899 u64 first_tx_mstamp; 900 /* when we reached the "delivered" count */ 901 u64 delivered_mstamp; 902 } tx; /* only used for outgoing skbs */ 903 union { 904 struct inet_skb_parm h4; 905 #if IS_ENABLED(CONFIG_IPV6) 906 struct inet6_skb_parm h6; 907 #endif 908 } header; /* For incoming skbs */ 909 }; 910 }; 911 912 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 913 914 extern const struct inet_connection_sock_af_ops ipv4_specific; 915 916 #if IS_ENABLED(CONFIG_IPV6) 917 /* This is the variant of inet6_iif() that must be used by TCP, 918 * as TCP moves IP6CB into a different location in skb->cb[] 919 */ 920 static inline int tcp_v6_iif(const struct sk_buff *skb) 921 { 922 return TCP_SKB_CB(skb)->header.h6.iif; 923 } 924 925 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 926 { 927 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 928 929 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 930 } 931 932 /* TCP_SKB_CB reference means this can not be used from early demux */ 933 static inline int tcp_v6_sdif(const struct sk_buff *skb) 934 { 935 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 936 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 937 return TCP_SKB_CB(skb)->header.h6.iif; 938 #endif 939 return 0; 940 } 941 942 extern const struct inet_connection_sock_af_ops ipv6_specific; 943 944 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 945 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 946 void tcp_v6_early_demux(struct sk_buff *skb); 947 948 #endif 949 950 /* TCP_SKB_CB reference means this can not be used from early demux */ 951 static inline int tcp_v4_sdif(struct sk_buff *skb) 952 { 953 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 954 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 955 return TCP_SKB_CB(skb)->header.h4.iif; 956 #endif 957 return 0; 958 } 959 960 /* Due to TSO, an SKB can be composed of multiple actual 961 * packets. To keep these tracked properly, we use this. 962 */ 963 static inline int tcp_skb_pcount(const struct sk_buff *skb) 964 { 965 return TCP_SKB_CB(skb)->tcp_gso_segs; 966 } 967 968 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 969 { 970 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 971 } 972 973 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 974 { 975 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 976 } 977 978 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 979 static inline int tcp_skb_mss(const struct sk_buff *skb) 980 { 981 return TCP_SKB_CB(skb)->tcp_gso_size; 982 } 983 984 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 985 { 986 return likely(!TCP_SKB_CB(skb)->eor); 987 } 988 989 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 990 const struct sk_buff *from) 991 { 992 return likely(tcp_skb_can_collapse_to(to) && 993 mptcp_skb_can_collapse(to, from) && 994 skb_pure_zcopy_same(to, from)); 995 } 996 997 /* Events passed to congestion control interface */ 998 enum tcp_ca_event { 999 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1000 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1001 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1002 CA_EVENT_LOSS, /* loss timeout */ 1003 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1004 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1005 }; 1006 1007 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1008 enum tcp_ca_ack_event_flags { 1009 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1010 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1011 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1012 }; 1013 1014 /* 1015 * Interface for adding new TCP congestion control handlers 1016 */ 1017 #define TCP_CA_NAME_MAX 16 1018 #define TCP_CA_MAX 128 1019 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1020 1021 #define TCP_CA_UNSPEC 0 1022 1023 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1024 #define TCP_CONG_NON_RESTRICTED 0x1 1025 /* Requires ECN/ECT set on all packets */ 1026 #define TCP_CONG_NEEDS_ECN 0x2 1027 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1028 1029 union tcp_cc_info; 1030 1031 struct ack_sample { 1032 u32 pkts_acked; 1033 s32 rtt_us; 1034 u32 in_flight; 1035 }; 1036 1037 /* A rate sample measures the number of (original/retransmitted) data 1038 * packets delivered "delivered" over an interval of time "interval_us". 1039 * The tcp_rate.c code fills in the rate sample, and congestion 1040 * control modules that define a cong_control function to run at the end 1041 * of ACK processing can optionally chose to consult this sample when 1042 * setting cwnd and pacing rate. 1043 * A sample is invalid if "delivered" or "interval_us" is negative. 1044 */ 1045 struct rate_sample { 1046 u64 prior_mstamp; /* starting timestamp for interval */ 1047 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1048 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1049 s32 delivered; /* number of packets delivered over interval */ 1050 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1051 long interval_us; /* time for tp->delivered to incr "delivered" */ 1052 u32 snd_interval_us; /* snd interval for delivered packets */ 1053 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1054 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1055 int losses; /* number of packets marked lost upon ACK */ 1056 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1057 u32 prior_in_flight; /* in flight before this ACK */ 1058 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1059 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1060 bool is_retrans; /* is sample from retransmission? */ 1061 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1062 }; 1063 1064 struct tcp_congestion_ops { 1065 /* fast path fields are put first to fill one cache line */ 1066 1067 /* return slow start threshold (required) */ 1068 u32 (*ssthresh)(struct sock *sk); 1069 1070 /* do new cwnd calculation (required) */ 1071 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1072 1073 /* call before changing ca_state (optional) */ 1074 void (*set_state)(struct sock *sk, u8 new_state); 1075 1076 /* call when cwnd event occurs (optional) */ 1077 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1078 1079 /* call when ack arrives (optional) */ 1080 void (*in_ack_event)(struct sock *sk, u32 flags); 1081 1082 /* hook for packet ack accounting (optional) */ 1083 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1084 1085 /* override sysctl_tcp_min_tso_segs */ 1086 u32 (*min_tso_segs)(struct sock *sk); 1087 1088 /* call when packets are delivered to update cwnd and pacing rate, 1089 * after all the ca_state processing. (optional) 1090 */ 1091 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1092 1093 1094 /* new value of cwnd after loss (required) */ 1095 u32 (*undo_cwnd)(struct sock *sk); 1096 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1097 u32 (*sndbuf_expand)(struct sock *sk); 1098 1099 /* control/slow paths put last */ 1100 /* get info for inet_diag (optional) */ 1101 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1102 union tcp_cc_info *info); 1103 1104 char name[TCP_CA_NAME_MAX]; 1105 struct module *owner; 1106 struct list_head list; 1107 u32 key; 1108 u32 flags; 1109 1110 /* initialize private data (optional) */ 1111 void (*init)(struct sock *sk); 1112 /* cleanup private data (optional) */ 1113 void (*release)(struct sock *sk); 1114 } ____cacheline_aligned_in_smp; 1115 1116 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1117 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1118 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1119 struct tcp_congestion_ops *old_type); 1120 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1121 1122 void tcp_assign_congestion_control(struct sock *sk); 1123 void tcp_init_congestion_control(struct sock *sk); 1124 void tcp_cleanup_congestion_control(struct sock *sk); 1125 int tcp_set_default_congestion_control(struct net *net, const char *name); 1126 void tcp_get_default_congestion_control(struct net *net, char *name); 1127 void tcp_get_available_congestion_control(char *buf, size_t len); 1128 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1129 int tcp_set_allowed_congestion_control(char *allowed); 1130 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1131 bool cap_net_admin); 1132 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1133 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1134 1135 u32 tcp_reno_ssthresh(struct sock *sk); 1136 u32 tcp_reno_undo_cwnd(struct sock *sk); 1137 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1138 extern struct tcp_congestion_ops tcp_reno; 1139 1140 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1141 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1142 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1143 #ifdef CONFIG_INET 1144 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1145 #else 1146 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1147 { 1148 return NULL; 1149 } 1150 #endif 1151 1152 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1153 { 1154 const struct inet_connection_sock *icsk = inet_csk(sk); 1155 1156 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1157 } 1158 1159 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1160 { 1161 const struct inet_connection_sock *icsk = inet_csk(sk); 1162 1163 if (icsk->icsk_ca_ops->cwnd_event) 1164 icsk->icsk_ca_ops->cwnd_event(sk, event); 1165 } 1166 1167 /* From tcp_cong.c */ 1168 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1169 1170 /* From tcp_rate.c */ 1171 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1172 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1173 struct rate_sample *rs); 1174 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1175 bool is_sack_reneg, struct rate_sample *rs); 1176 void tcp_rate_check_app_limited(struct sock *sk); 1177 1178 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1179 { 1180 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1181 } 1182 1183 /* These functions determine how the current flow behaves in respect of SACK 1184 * handling. SACK is negotiated with the peer, and therefore it can vary 1185 * between different flows. 1186 * 1187 * tcp_is_sack - SACK enabled 1188 * tcp_is_reno - No SACK 1189 */ 1190 static inline int tcp_is_sack(const struct tcp_sock *tp) 1191 { 1192 return likely(tp->rx_opt.sack_ok); 1193 } 1194 1195 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1196 { 1197 return !tcp_is_sack(tp); 1198 } 1199 1200 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1201 { 1202 return tp->sacked_out + tp->lost_out; 1203 } 1204 1205 /* This determines how many packets are "in the network" to the best 1206 * of our knowledge. In many cases it is conservative, but where 1207 * detailed information is available from the receiver (via SACK 1208 * blocks etc.) we can make more aggressive calculations. 1209 * 1210 * Use this for decisions involving congestion control, use just 1211 * tp->packets_out to determine if the send queue is empty or not. 1212 * 1213 * Read this equation as: 1214 * 1215 * "Packets sent once on transmission queue" MINUS 1216 * "Packets left network, but not honestly ACKed yet" PLUS 1217 * "Packets fast retransmitted" 1218 */ 1219 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1220 { 1221 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1222 } 1223 1224 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1225 1226 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1227 { 1228 return tp->snd_cwnd; 1229 } 1230 1231 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1232 { 1233 WARN_ON_ONCE((int)val <= 0); 1234 tp->snd_cwnd = val; 1235 } 1236 1237 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1238 { 1239 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1240 } 1241 1242 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1243 { 1244 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1245 } 1246 1247 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1248 { 1249 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1250 (1 << inet_csk(sk)->icsk_ca_state); 1251 } 1252 1253 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1254 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1255 * ssthresh. 1256 */ 1257 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1258 { 1259 const struct tcp_sock *tp = tcp_sk(sk); 1260 1261 if (tcp_in_cwnd_reduction(sk)) 1262 return tp->snd_ssthresh; 1263 else 1264 return max(tp->snd_ssthresh, 1265 ((tcp_snd_cwnd(tp) >> 1) + 1266 (tcp_snd_cwnd(tp) >> 2))); 1267 } 1268 1269 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1270 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1271 1272 void tcp_enter_cwr(struct sock *sk); 1273 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1274 1275 /* The maximum number of MSS of available cwnd for which TSO defers 1276 * sending if not using sysctl_tcp_tso_win_divisor. 1277 */ 1278 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1279 { 1280 return 3; 1281 } 1282 1283 /* Returns end sequence number of the receiver's advertised window */ 1284 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1285 { 1286 return tp->snd_una + tp->snd_wnd; 1287 } 1288 1289 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1290 * flexible approach. The RFC suggests cwnd should not be raised unless 1291 * it was fully used previously. And that's exactly what we do in 1292 * congestion avoidance mode. But in slow start we allow cwnd to grow 1293 * as long as the application has used half the cwnd. 1294 * Example : 1295 * cwnd is 10 (IW10), but application sends 9 frames. 1296 * We allow cwnd to reach 18 when all frames are ACKed. 1297 * This check is safe because it's as aggressive as slow start which already 1298 * risks 100% overshoot. The advantage is that we discourage application to 1299 * either send more filler packets or data to artificially blow up the cwnd 1300 * usage, and allow application-limited process to probe bw more aggressively. 1301 */ 1302 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1303 { 1304 const struct tcp_sock *tp = tcp_sk(sk); 1305 1306 if (tp->is_cwnd_limited) 1307 return true; 1308 1309 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1310 if (tcp_in_slow_start(tp)) 1311 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1312 1313 return false; 1314 } 1315 1316 /* BBR congestion control needs pacing. 1317 * Same remark for SO_MAX_PACING_RATE. 1318 * sch_fq packet scheduler is efficiently handling pacing, 1319 * but is not always installed/used. 1320 * Return true if TCP stack should pace packets itself. 1321 */ 1322 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1323 { 1324 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1325 } 1326 1327 /* Estimates in how many jiffies next packet for this flow can be sent. 1328 * Scheduling a retransmit timer too early would be silly. 1329 */ 1330 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1331 { 1332 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1333 1334 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1335 } 1336 1337 static inline void tcp_reset_xmit_timer(struct sock *sk, 1338 const int what, 1339 unsigned long when, 1340 const unsigned long max_when) 1341 { 1342 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1343 max_when); 1344 } 1345 1346 /* Something is really bad, we could not queue an additional packet, 1347 * because qdisc is full or receiver sent a 0 window, or we are paced. 1348 * We do not want to add fuel to the fire, or abort too early, 1349 * so make sure the timer we arm now is at least 200ms in the future, 1350 * regardless of current icsk_rto value (as it could be ~2ms) 1351 */ 1352 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1353 { 1354 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1355 } 1356 1357 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1358 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1359 unsigned long max_when) 1360 { 1361 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1362 inet_csk(sk)->icsk_backoff); 1363 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1364 1365 return (unsigned long)min_t(u64, when, max_when); 1366 } 1367 1368 static inline void tcp_check_probe_timer(struct sock *sk) 1369 { 1370 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1371 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1372 tcp_probe0_base(sk), TCP_RTO_MAX); 1373 } 1374 1375 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1376 { 1377 tp->snd_wl1 = seq; 1378 } 1379 1380 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1381 { 1382 tp->snd_wl1 = seq; 1383 } 1384 1385 /* 1386 * Calculate(/check) TCP checksum 1387 */ 1388 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1389 __be32 daddr, __wsum base) 1390 { 1391 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1392 } 1393 1394 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1395 { 1396 return !skb_csum_unnecessary(skb) && 1397 __skb_checksum_complete(skb); 1398 } 1399 1400 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1401 enum skb_drop_reason *reason); 1402 1403 1404 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1405 void tcp_set_state(struct sock *sk, int state); 1406 void tcp_done(struct sock *sk); 1407 int tcp_abort(struct sock *sk, int err); 1408 1409 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1410 { 1411 rx_opt->dsack = 0; 1412 rx_opt->num_sacks = 0; 1413 } 1414 1415 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1416 1417 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1418 { 1419 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1420 struct tcp_sock *tp = tcp_sk(sk); 1421 s32 delta; 1422 1423 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1424 tp->packets_out || ca_ops->cong_control) 1425 return; 1426 delta = tcp_jiffies32 - tp->lsndtime; 1427 if (delta > inet_csk(sk)->icsk_rto) 1428 tcp_cwnd_restart(sk, delta); 1429 } 1430 1431 /* Determine a window scaling and initial window to offer. */ 1432 void tcp_select_initial_window(const struct sock *sk, int __space, 1433 __u32 mss, __u32 *rcv_wnd, 1434 __u32 *window_clamp, int wscale_ok, 1435 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1436 1437 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1438 { 1439 s64 scaled_space = (s64)space * scaling_ratio; 1440 1441 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1442 } 1443 1444 static inline int tcp_win_from_space(const struct sock *sk, int space) 1445 { 1446 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1447 } 1448 1449 /* inverse of __tcp_win_from_space() */ 1450 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1451 { 1452 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1453 1454 do_div(val, scaling_ratio); 1455 return val; 1456 } 1457 1458 static inline int tcp_space_from_win(const struct sock *sk, int win) 1459 { 1460 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1461 } 1462 1463 static inline void tcp_scaling_ratio_init(struct sock *sk) 1464 { 1465 /* Assume a conservative default of 1200 bytes of payload per 4K page. 1466 * This may be adjusted later in tcp_measure_rcv_mss(). 1467 */ 1468 tcp_sk(sk)->scaling_ratio = (1200 << TCP_RMEM_TO_WIN_SCALE) / 1469 SKB_TRUESIZE(4096); 1470 } 1471 1472 /* Note: caller must be prepared to deal with negative returns */ 1473 static inline int tcp_space(const struct sock *sk) 1474 { 1475 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1476 READ_ONCE(sk->sk_backlog.len) - 1477 atomic_read(&sk->sk_rmem_alloc)); 1478 } 1479 1480 static inline int tcp_full_space(const struct sock *sk) 1481 { 1482 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1483 } 1484 1485 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh) 1486 { 1487 int unused_mem = sk_unused_reserved_mem(sk); 1488 struct tcp_sock *tp = tcp_sk(sk); 1489 1490 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh); 1491 if (unused_mem) 1492 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1493 tcp_win_from_space(sk, unused_mem)); 1494 } 1495 1496 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1497 { 1498 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss); 1499 } 1500 1501 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1502 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1503 1504 1505 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1506 * If 87.5 % (7/8) of the space has been consumed, we want to override 1507 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1508 * len/truesize ratio. 1509 */ 1510 static inline bool tcp_rmem_pressure(const struct sock *sk) 1511 { 1512 int rcvbuf, threshold; 1513 1514 if (tcp_under_memory_pressure(sk)) 1515 return true; 1516 1517 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1518 threshold = rcvbuf - (rcvbuf >> 3); 1519 1520 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1521 } 1522 1523 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1524 { 1525 const struct tcp_sock *tp = tcp_sk(sk); 1526 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1527 1528 if (avail <= 0) 1529 return false; 1530 1531 return (avail >= target) || tcp_rmem_pressure(sk) || 1532 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1533 } 1534 1535 extern void tcp_openreq_init_rwin(struct request_sock *req, 1536 const struct sock *sk_listener, 1537 const struct dst_entry *dst); 1538 1539 void tcp_enter_memory_pressure(struct sock *sk); 1540 void tcp_leave_memory_pressure(struct sock *sk); 1541 1542 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1543 { 1544 struct net *net = sock_net((struct sock *)tp); 1545 int val; 1546 1547 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1548 * and do_tcp_setsockopt(). 1549 */ 1550 val = READ_ONCE(tp->keepalive_intvl); 1551 1552 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1553 } 1554 1555 static inline int keepalive_time_when(const struct tcp_sock *tp) 1556 { 1557 struct net *net = sock_net((struct sock *)tp); 1558 int val; 1559 1560 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1561 val = READ_ONCE(tp->keepalive_time); 1562 1563 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1564 } 1565 1566 static inline int keepalive_probes(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_keepcnt() 1572 * and do_tcp_setsockopt(). 1573 */ 1574 val = READ_ONCE(tp->keepalive_probes); 1575 1576 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1577 } 1578 1579 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1580 { 1581 const struct inet_connection_sock *icsk = &tp->inet_conn; 1582 1583 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1584 tcp_jiffies32 - tp->rcv_tstamp); 1585 } 1586 1587 static inline int tcp_fin_time(const struct sock *sk) 1588 { 1589 int fin_timeout = tcp_sk(sk)->linger2 ? : 1590 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1591 const int rto = inet_csk(sk)->icsk_rto; 1592 1593 if (fin_timeout < (rto << 2) - (rto >> 1)) 1594 fin_timeout = (rto << 2) - (rto >> 1); 1595 1596 return fin_timeout; 1597 } 1598 1599 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1600 int paws_win) 1601 { 1602 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1603 return true; 1604 if (unlikely(!time_before32(ktime_get_seconds(), 1605 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1606 return true; 1607 /* 1608 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1609 * then following tcp messages have valid values. Ignore 0 value, 1610 * or else 'negative' tsval might forbid us to accept their packets. 1611 */ 1612 if (!rx_opt->ts_recent) 1613 return true; 1614 return false; 1615 } 1616 1617 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1618 int rst) 1619 { 1620 if (tcp_paws_check(rx_opt, 0)) 1621 return false; 1622 1623 /* RST segments are not recommended to carry timestamp, 1624 and, if they do, it is recommended to ignore PAWS because 1625 "their cleanup function should take precedence over timestamps." 1626 Certainly, it is mistake. It is necessary to understand the reasons 1627 of this constraint to relax it: if peer reboots, clock may go 1628 out-of-sync and half-open connections will not be reset. 1629 Actually, the problem would be not existing if all 1630 the implementations followed draft about maintaining clock 1631 via reboots. Linux-2.2 DOES NOT! 1632 1633 However, we can relax time bounds for RST segments to MSL. 1634 */ 1635 if (rst && !time_before32(ktime_get_seconds(), 1636 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1637 return false; 1638 return true; 1639 } 1640 1641 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1642 int mib_idx, u32 *last_oow_ack_time); 1643 1644 static inline void tcp_mib_init(struct net *net) 1645 { 1646 /* See RFC 2012 */ 1647 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1648 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1649 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1650 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1651 } 1652 1653 /* from STCP */ 1654 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1655 { 1656 tp->lost_skb_hint = NULL; 1657 } 1658 1659 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1660 { 1661 tcp_clear_retrans_hints_partial(tp); 1662 tp->retransmit_skb_hint = NULL; 1663 } 1664 1665 union tcp_md5_addr { 1666 struct in_addr a4; 1667 #if IS_ENABLED(CONFIG_IPV6) 1668 struct in6_addr a6; 1669 #endif 1670 }; 1671 1672 /* - key database */ 1673 struct tcp_md5sig_key { 1674 struct hlist_node node; 1675 u8 keylen; 1676 u8 family; /* AF_INET or AF_INET6 */ 1677 u8 prefixlen; 1678 u8 flags; 1679 union tcp_md5_addr addr; 1680 int l3index; /* set if key added with L3 scope */ 1681 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1682 struct rcu_head rcu; 1683 }; 1684 1685 /* - sock block */ 1686 struct tcp_md5sig_info { 1687 struct hlist_head head; 1688 struct rcu_head rcu; 1689 }; 1690 1691 /* - pseudo header */ 1692 struct tcp4_pseudohdr { 1693 __be32 saddr; 1694 __be32 daddr; 1695 __u8 pad; 1696 __u8 protocol; 1697 __be16 len; 1698 }; 1699 1700 struct tcp6_pseudohdr { 1701 struct in6_addr saddr; 1702 struct in6_addr daddr; 1703 __be32 len; 1704 __be32 protocol; /* including padding */ 1705 }; 1706 1707 union tcp_md5sum_block { 1708 struct tcp4_pseudohdr ip4; 1709 #if IS_ENABLED(CONFIG_IPV6) 1710 struct tcp6_pseudohdr ip6; 1711 #endif 1712 }; 1713 1714 /* - pool: digest algorithm, hash description and scratch buffer */ 1715 struct tcp_md5sig_pool { 1716 struct ahash_request *md5_req; 1717 void *scratch; 1718 }; 1719 1720 /* - functions */ 1721 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1722 const struct sock *sk, const struct sk_buff *skb); 1723 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1724 int family, u8 prefixlen, int l3index, u8 flags, 1725 const u8 *newkey, u8 newkeylen); 1726 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1727 int family, u8 prefixlen, int l3index, 1728 struct tcp_md5sig_key *key); 1729 1730 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1731 int family, u8 prefixlen, int l3index, u8 flags); 1732 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1733 const struct sock *addr_sk); 1734 1735 #ifdef CONFIG_TCP_MD5SIG 1736 #include <linux/jump_label.h> 1737 extern struct static_key_false_deferred tcp_md5_needed; 1738 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1739 const union tcp_md5_addr *addr, 1740 int family); 1741 static inline struct tcp_md5sig_key * 1742 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1743 const union tcp_md5_addr *addr, int family) 1744 { 1745 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1746 return NULL; 1747 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1748 } 1749 1750 enum skb_drop_reason 1751 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1752 const void *saddr, const void *daddr, 1753 int family, int dif, int sdif); 1754 1755 1756 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1757 #else 1758 static inline struct tcp_md5sig_key * 1759 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1760 const union tcp_md5_addr *addr, int family) 1761 { 1762 return NULL; 1763 } 1764 1765 static inline enum skb_drop_reason 1766 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1767 const void *saddr, const void *daddr, 1768 int family, int dif, int sdif) 1769 { 1770 return SKB_NOT_DROPPED_YET; 1771 } 1772 #define tcp_twsk_md5_key(twsk) NULL 1773 #endif 1774 1775 bool tcp_alloc_md5sig_pool(void); 1776 1777 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1778 static inline void tcp_put_md5sig_pool(void) 1779 { 1780 local_bh_enable(); 1781 } 1782 1783 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1784 unsigned int header_len); 1785 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1786 const struct tcp_md5sig_key *key); 1787 1788 /* From tcp_fastopen.c */ 1789 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1790 struct tcp_fastopen_cookie *cookie); 1791 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1792 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1793 u16 try_exp); 1794 struct tcp_fastopen_request { 1795 /* Fast Open cookie. Size 0 means a cookie request */ 1796 struct tcp_fastopen_cookie cookie; 1797 struct msghdr *data; /* data in MSG_FASTOPEN */ 1798 size_t size; 1799 int copied; /* queued in tcp_connect() */ 1800 struct ubuf_info *uarg; 1801 }; 1802 void tcp_free_fastopen_req(struct tcp_sock *tp); 1803 void tcp_fastopen_destroy_cipher(struct sock *sk); 1804 void tcp_fastopen_ctx_destroy(struct net *net); 1805 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1806 void *primary_key, void *backup_key); 1807 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1808 u64 *key); 1809 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1810 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1811 struct request_sock *req, 1812 struct tcp_fastopen_cookie *foc, 1813 const struct dst_entry *dst); 1814 void tcp_fastopen_init_key_once(struct net *net); 1815 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1816 struct tcp_fastopen_cookie *cookie); 1817 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1818 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1819 #define TCP_FASTOPEN_KEY_MAX 2 1820 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1821 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1822 1823 /* Fastopen key context */ 1824 struct tcp_fastopen_context { 1825 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1826 int num; 1827 struct rcu_head rcu; 1828 }; 1829 1830 void tcp_fastopen_active_disable(struct sock *sk); 1831 bool tcp_fastopen_active_should_disable(struct sock *sk); 1832 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1833 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1834 1835 /* Caller needs to wrap with rcu_read_(un)lock() */ 1836 static inline 1837 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1838 { 1839 struct tcp_fastopen_context *ctx; 1840 1841 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1842 if (!ctx) 1843 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1844 return ctx; 1845 } 1846 1847 static inline 1848 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1849 const struct tcp_fastopen_cookie *orig) 1850 { 1851 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1852 orig->len == foc->len && 1853 !memcmp(orig->val, foc->val, foc->len)) 1854 return true; 1855 return false; 1856 } 1857 1858 static inline 1859 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1860 { 1861 return ctx->num; 1862 } 1863 1864 /* Latencies incurred by various limits for a sender. They are 1865 * chronograph-like stats that are mutually exclusive. 1866 */ 1867 enum tcp_chrono { 1868 TCP_CHRONO_UNSPEC, 1869 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1870 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1871 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1872 __TCP_CHRONO_MAX, 1873 }; 1874 1875 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1876 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1877 1878 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1879 * the same memory storage than skb->destructor/_skb_refdst 1880 */ 1881 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1882 { 1883 skb->destructor = NULL; 1884 skb->_skb_refdst = 0UL; 1885 } 1886 1887 #define tcp_skb_tsorted_save(skb) { \ 1888 unsigned long _save = skb->_skb_refdst; \ 1889 skb->_skb_refdst = 0UL; 1890 1891 #define tcp_skb_tsorted_restore(skb) \ 1892 skb->_skb_refdst = _save; \ 1893 } 1894 1895 void tcp_write_queue_purge(struct sock *sk); 1896 1897 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1898 { 1899 return skb_rb_first(&sk->tcp_rtx_queue); 1900 } 1901 1902 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1903 { 1904 return skb_rb_last(&sk->tcp_rtx_queue); 1905 } 1906 1907 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1908 { 1909 return skb_peek_tail(&sk->sk_write_queue); 1910 } 1911 1912 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1913 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1914 1915 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1916 { 1917 return skb_peek(&sk->sk_write_queue); 1918 } 1919 1920 static inline bool tcp_skb_is_last(const struct sock *sk, 1921 const struct sk_buff *skb) 1922 { 1923 return skb_queue_is_last(&sk->sk_write_queue, skb); 1924 } 1925 1926 /** 1927 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1928 * @sk: socket 1929 * 1930 * Since the write queue can have a temporary empty skb in it, 1931 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1932 */ 1933 static inline bool tcp_write_queue_empty(const struct sock *sk) 1934 { 1935 const struct tcp_sock *tp = tcp_sk(sk); 1936 1937 return tp->write_seq == tp->snd_nxt; 1938 } 1939 1940 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1941 { 1942 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1943 } 1944 1945 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1946 { 1947 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1948 } 1949 1950 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1951 { 1952 __skb_queue_tail(&sk->sk_write_queue, skb); 1953 1954 /* Queue it, remembering where we must start sending. */ 1955 if (sk->sk_write_queue.next == skb) 1956 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1957 } 1958 1959 /* Insert new before skb on the write queue of sk. */ 1960 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1961 struct sk_buff *skb, 1962 struct sock *sk) 1963 { 1964 __skb_queue_before(&sk->sk_write_queue, skb, new); 1965 } 1966 1967 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1968 { 1969 tcp_skb_tsorted_anchor_cleanup(skb); 1970 __skb_unlink(skb, &sk->sk_write_queue); 1971 } 1972 1973 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1974 1975 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1976 { 1977 tcp_skb_tsorted_anchor_cleanup(skb); 1978 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1979 } 1980 1981 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1982 { 1983 list_del(&skb->tcp_tsorted_anchor); 1984 tcp_rtx_queue_unlink(skb, sk); 1985 tcp_wmem_free_skb(sk, skb); 1986 } 1987 1988 static inline void tcp_push_pending_frames(struct sock *sk) 1989 { 1990 if (tcp_send_head(sk)) { 1991 struct tcp_sock *tp = tcp_sk(sk); 1992 1993 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1994 } 1995 } 1996 1997 /* Start sequence of the skb just after the highest skb with SACKed 1998 * bit, valid only if sacked_out > 0 or when the caller has ensured 1999 * validity by itself. 2000 */ 2001 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2002 { 2003 if (!tp->sacked_out) 2004 return tp->snd_una; 2005 2006 if (tp->highest_sack == NULL) 2007 return tp->snd_nxt; 2008 2009 return TCP_SKB_CB(tp->highest_sack)->seq; 2010 } 2011 2012 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2013 { 2014 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2015 } 2016 2017 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2018 { 2019 return tcp_sk(sk)->highest_sack; 2020 } 2021 2022 static inline void tcp_highest_sack_reset(struct sock *sk) 2023 { 2024 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2025 } 2026 2027 /* Called when old skb is about to be deleted and replaced by new skb */ 2028 static inline void tcp_highest_sack_replace(struct sock *sk, 2029 struct sk_buff *old, 2030 struct sk_buff *new) 2031 { 2032 if (old == tcp_highest_sack(sk)) 2033 tcp_sk(sk)->highest_sack = new; 2034 } 2035 2036 /* This helper checks if socket has IP_TRANSPARENT set */ 2037 static inline bool inet_sk_transparent(const struct sock *sk) 2038 { 2039 switch (sk->sk_state) { 2040 case TCP_TIME_WAIT: 2041 return inet_twsk(sk)->tw_transparent; 2042 case TCP_NEW_SYN_RECV: 2043 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2044 } 2045 return inet_test_bit(TRANSPARENT, sk); 2046 } 2047 2048 /* Determines whether this is a thin stream (which may suffer from 2049 * increased latency). Used to trigger latency-reducing mechanisms. 2050 */ 2051 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2052 { 2053 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2054 } 2055 2056 /* /proc */ 2057 enum tcp_seq_states { 2058 TCP_SEQ_STATE_LISTENING, 2059 TCP_SEQ_STATE_ESTABLISHED, 2060 }; 2061 2062 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2063 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2064 void tcp_seq_stop(struct seq_file *seq, void *v); 2065 2066 struct tcp_seq_afinfo { 2067 sa_family_t family; 2068 }; 2069 2070 struct tcp_iter_state { 2071 struct seq_net_private p; 2072 enum tcp_seq_states state; 2073 struct sock *syn_wait_sk; 2074 int bucket, offset, sbucket, num; 2075 loff_t last_pos; 2076 }; 2077 2078 extern struct request_sock_ops tcp_request_sock_ops; 2079 extern struct request_sock_ops tcp6_request_sock_ops; 2080 2081 void tcp_v4_destroy_sock(struct sock *sk); 2082 2083 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2084 netdev_features_t features); 2085 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2086 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2087 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2088 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2089 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2090 void tcp_gro_complete(struct sk_buff *skb); 2091 2092 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2093 2094 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2095 { 2096 struct net *net = sock_net((struct sock *)tp); 2097 u32 val; 2098 2099 val = READ_ONCE(tp->notsent_lowat); 2100 2101 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2102 } 2103 2104 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2105 2106 #ifdef CONFIG_PROC_FS 2107 int tcp4_proc_init(void); 2108 void tcp4_proc_exit(void); 2109 #endif 2110 2111 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2112 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2113 const struct tcp_request_sock_ops *af_ops, 2114 struct sock *sk, struct sk_buff *skb); 2115 2116 /* TCP af-specific functions */ 2117 struct tcp_sock_af_ops { 2118 #ifdef CONFIG_TCP_MD5SIG 2119 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2120 const struct sock *addr_sk); 2121 int (*calc_md5_hash)(char *location, 2122 const struct tcp_md5sig_key *md5, 2123 const struct sock *sk, 2124 const struct sk_buff *skb); 2125 int (*md5_parse)(struct sock *sk, 2126 int optname, 2127 sockptr_t optval, 2128 int optlen); 2129 #endif 2130 }; 2131 2132 struct tcp_request_sock_ops { 2133 u16 mss_clamp; 2134 #ifdef CONFIG_TCP_MD5SIG 2135 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2136 const struct sock *addr_sk); 2137 int (*calc_md5_hash) (char *location, 2138 const struct tcp_md5sig_key *md5, 2139 const struct sock *sk, 2140 const struct sk_buff *skb); 2141 #endif 2142 #ifdef CONFIG_SYN_COOKIES 2143 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2144 __u16 *mss); 2145 #endif 2146 struct dst_entry *(*route_req)(const struct sock *sk, 2147 struct sk_buff *skb, 2148 struct flowi *fl, 2149 struct request_sock *req); 2150 u32 (*init_seq)(const struct sk_buff *skb); 2151 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2152 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2153 struct flowi *fl, struct request_sock *req, 2154 struct tcp_fastopen_cookie *foc, 2155 enum tcp_synack_type synack_type, 2156 struct sk_buff *syn_skb); 2157 }; 2158 2159 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2160 #if IS_ENABLED(CONFIG_IPV6) 2161 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2162 #endif 2163 2164 #ifdef CONFIG_SYN_COOKIES 2165 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2166 const struct sock *sk, struct sk_buff *skb, 2167 __u16 *mss) 2168 { 2169 tcp_synq_overflow(sk); 2170 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2171 return ops->cookie_init_seq(skb, mss); 2172 } 2173 #else 2174 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2175 const struct sock *sk, struct sk_buff *skb, 2176 __u16 *mss) 2177 { 2178 return 0; 2179 } 2180 #endif 2181 2182 int tcpv4_offload_init(void); 2183 2184 void tcp_v4_init(void); 2185 void tcp_init(void); 2186 2187 /* tcp_recovery.c */ 2188 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2189 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2190 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2191 u32 reo_wnd); 2192 extern bool tcp_rack_mark_lost(struct sock *sk); 2193 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2194 u64 xmit_time); 2195 extern void tcp_rack_reo_timeout(struct sock *sk); 2196 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2197 2198 /* tcp_plb.c */ 2199 2200 /* 2201 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2202 * expects cong_ratio which represents fraction of traffic that experienced 2203 * congestion over a single RTT. In order to avoid floating point operations, 2204 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2205 */ 2206 #define TCP_PLB_SCALE 8 2207 2208 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2209 struct tcp_plb_state { 2210 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2211 unused:3; 2212 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2213 }; 2214 2215 static inline void tcp_plb_init(const struct sock *sk, 2216 struct tcp_plb_state *plb) 2217 { 2218 plb->consec_cong_rounds = 0; 2219 plb->pause_until = 0; 2220 } 2221 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2222 const int cong_ratio); 2223 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2224 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2225 2226 /* At how many usecs into the future should the RTO fire? */ 2227 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2228 { 2229 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2230 u32 rto = inet_csk(sk)->icsk_rto; 2231 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2232 2233 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2234 } 2235 2236 /* 2237 * Save and compile IPv4 options, return a pointer to it 2238 */ 2239 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2240 struct sk_buff *skb) 2241 { 2242 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2243 struct ip_options_rcu *dopt = NULL; 2244 2245 if (opt->optlen) { 2246 int opt_size = sizeof(*dopt) + opt->optlen; 2247 2248 dopt = kmalloc(opt_size, GFP_ATOMIC); 2249 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2250 kfree(dopt); 2251 dopt = NULL; 2252 } 2253 } 2254 return dopt; 2255 } 2256 2257 /* locally generated TCP pure ACKs have skb->truesize == 2 2258 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2259 * This is much faster than dissecting the packet to find out. 2260 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2261 */ 2262 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2263 { 2264 return skb->truesize == 2; 2265 } 2266 2267 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2268 { 2269 skb->truesize = 2; 2270 } 2271 2272 static inline int tcp_inq(struct sock *sk) 2273 { 2274 struct tcp_sock *tp = tcp_sk(sk); 2275 int answ; 2276 2277 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2278 answ = 0; 2279 } else if (sock_flag(sk, SOCK_URGINLINE) || 2280 !tp->urg_data || 2281 before(tp->urg_seq, tp->copied_seq) || 2282 !before(tp->urg_seq, tp->rcv_nxt)) { 2283 2284 answ = tp->rcv_nxt - tp->copied_seq; 2285 2286 /* Subtract 1, if FIN was received */ 2287 if (answ && sock_flag(sk, SOCK_DONE)) 2288 answ--; 2289 } else { 2290 answ = tp->urg_seq - tp->copied_seq; 2291 } 2292 2293 return answ; 2294 } 2295 2296 int tcp_peek_len(struct socket *sock); 2297 2298 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2299 { 2300 u16 segs_in; 2301 2302 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2303 2304 /* We update these fields while other threads might 2305 * read them from tcp_get_info() 2306 */ 2307 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2308 if (skb->len > tcp_hdrlen(skb)) 2309 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2310 } 2311 2312 /* 2313 * TCP listen path runs lockless. 2314 * We forced "struct sock" to be const qualified to make sure 2315 * we don't modify one of its field by mistake. 2316 * Here, we increment sk_drops which is an atomic_t, so we can safely 2317 * make sock writable again. 2318 */ 2319 static inline void tcp_listendrop(const struct sock *sk) 2320 { 2321 atomic_inc(&((struct sock *)sk)->sk_drops); 2322 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2323 } 2324 2325 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2326 2327 /* 2328 * Interface for adding Upper Level Protocols over TCP 2329 */ 2330 2331 #define TCP_ULP_NAME_MAX 16 2332 #define TCP_ULP_MAX 128 2333 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2334 2335 struct tcp_ulp_ops { 2336 struct list_head list; 2337 2338 /* initialize ulp */ 2339 int (*init)(struct sock *sk); 2340 /* update ulp */ 2341 void (*update)(struct sock *sk, struct proto *p, 2342 void (*write_space)(struct sock *sk)); 2343 /* cleanup ulp */ 2344 void (*release)(struct sock *sk); 2345 /* diagnostic */ 2346 int (*get_info)(struct sock *sk, struct sk_buff *skb); 2347 size_t (*get_info_size)(const struct sock *sk); 2348 /* clone ulp */ 2349 void (*clone)(const struct request_sock *req, struct sock *newsk, 2350 const gfp_t priority); 2351 2352 char name[TCP_ULP_NAME_MAX]; 2353 struct module *owner; 2354 }; 2355 int tcp_register_ulp(struct tcp_ulp_ops *type); 2356 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2357 int tcp_set_ulp(struct sock *sk, const char *name); 2358 void tcp_get_available_ulp(char *buf, size_t len); 2359 void tcp_cleanup_ulp(struct sock *sk); 2360 void tcp_update_ulp(struct sock *sk, struct proto *p, 2361 void (*write_space)(struct sock *sk)); 2362 2363 #define MODULE_ALIAS_TCP_ULP(name) \ 2364 __MODULE_INFO(alias, alias_userspace, name); \ 2365 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2366 2367 #ifdef CONFIG_NET_SOCK_MSG 2368 struct sk_msg; 2369 struct sk_psock; 2370 2371 #ifdef CONFIG_BPF_SYSCALL 2372 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2373 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2374 #endif /* CONFIG_BPF_SYSCALL */ 2375 2376 #ifdef CONFIG_INET 2377 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2378 #else 2379 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2380 { 2381 } 2382 #endif 2383 2384 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2385 struct sk_msg *msg, u32 bytes, int flags); 2386 #endif /* CONFIG_NET_SOCK_MSG */ 2387 2388 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2389 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2390 { 2391 } 2392 #endif 2393 2394 #ifdef CONFIG_CGROUP_BPF 2395 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2396 struct sk_buff *skb, 2397 unsigned int end_offset) 2398 { 2399 skops->skb = skb; 2400 skops->skb_data_end = skb->data + end_offset; 2401 } 2402 #else 2403 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2404 struct sk_buff *skb, 2405 unsigned int end_offset) 2406 { 2407 } 2408 #endif 2409 2410 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2411 * is < 0, then the BPF op failed (for example if the loaded BPF 2412 * program does not support the chosen operation or there is no BPF 2413 * program loaded). 2414 */ 2415 #ifdef CONFIG_BPF 2416 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2417 { 2418 struct bpf_sock_ops_kern sock_ops; 2419 int ret; 2420 2421 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2422 if (sk_fullsock(sk)) { 2423 sock_ops.is_fullsock = 1; 2424 sock_owned_by_me(sk); 2425 } 2426 2427 sock_ops.sk = sk; 2428 sock_ops.op = op; 2429 if (nargs > 0) 2430 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2431 2432 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2433 if (ret == 0) 2434 ret = sock_ops.reply; 2435 else 2436 ret = -1; 2437 return ret; 2438 } 2439 2440 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2441 { 2442 u32 args[2] = {arg1, arg2}; 2443 2444 return tcp_call_bpf(sk, op, 2, args); 2445 } 2446 2447 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2448 u32 arg3) 2449 { 2450 u32 args[3] = {arg1, arg2, arg3}; 2451 2452 return tcp_call_bpf(sk, op, 3, args); 2453 } 2454 2455 #else 2456 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2457 { 2458 return -EPERM; 2459 } 2460 2461 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2462 { 2463 return -EPERM; 2464 } 2465 2466 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2467 u32 arg3) 2468 { 2469 return -EPERM; 2470 } 2471 2472 #endif 2473 2474 static inline u32 tcp_timeout_init(struct sock *sk) 2475 { 2476 int timeout; 2477 2478 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2479 2480 if (timeout <= 0) 2481 timeout = TCP_TIMEOUT_INIT; 2482 return min_t(int, timeout, TCP_RTO_MAX); 2483 } 2484 2485 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2486 { 2487 int rwnd; 2488 2489 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2490 2491 if (rwnd < 0) 2492 rwnd = 0; 2493 return rwnd; 2494 } 2495 2496 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2497 { 2498 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2499 } 2500 2501 static inline void tcp_bpf_rtt(struct sock *sk) 2502 { 2503 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2504 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2505 } 2506 2507 #if IS_ENABLED(CONFIG_SMC) 2508 extern struct static_key_false tcp_have_smc; 2509 #endif 2510 2511 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2512 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2513 void (*cad)(struct sock *sk, u32 ack_seq)); 2514 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2515 void clean_acked_data_flush(void); 2516 #endif 2517 2518 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2519 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2520 const struct tcp_sock *tp) 2521 { 2522 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2523 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2524 } 2525 2526 /* Compute Earliest Departure Time for some control packets 2527 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2528 */ 2529 static inline u64 tcp_transmit_time(const struct sock *sk) 2530 { 2531 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2532 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2533 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2534 2535 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2536 } 2537 return 0; 2538 } 2539 2540 #endif /* _TCP_H */ 2541