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 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 147 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 148 * used as a fallback RTO for the 149 * initial data transmission if no 150 * valid RTT sample has been acquired, 151 * most likely due to retrans in 3WHS. 152 */ 153 154 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 155 * for local resources. 156 */ 157 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 158 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 159 #define TCP_KEEPALIVE_INTVL (75*HZ) 160 161 #define MAX_TCP_KEEPIDLE 32767 162 #define MAX_TCP_KEEPINTVL 32767 163 #define MAX_TCP_KEEPCNT 127 164 #define MAX_TCP_SYNCNT 127 165 166 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 167 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 168 * after this time. It should be equal 169 * (or greater than) TCP_TIMEWAIT_LEN 170 * to provide reliability equal to one 171 * provided by timewait state. 172 */ 173 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 174 * timestamps. It must be less than 175 * minimal timewait lifetime. 176 */ 177 /* 178 * TCP option 179 */ 180 181 #define TCPOPT_NOP 1 /* Padding */ 182 #define TCPOPT_EOL 0 /* End of options */ 183 #define TCPOPT_MSS 2 /* Segment size negotiating */ 184 #define TCPOPT_WINDOW 3 /* Window scaling */ 185 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 186 #define TCPOPT_SACK 5 /* SACK Block */ 187 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 188 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 189 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 190 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 191 #define TCPOPT_EXP 254 /* Experimental */ 192 /* Magic number to be after the option value for sharing TCP 193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 194 */ 195 #define TCPOPT_FASTOPEN_MAGIC 0xF989 196 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 197 198 /* 199 * TCP option lengths 200 */ 201 202 #define TCPOLEN_MSS 4 203 #define TCPOLEN_WINDOW 3 204 #define TCPOLEN_SACK_PERM 2 205 #define TCPOLEN_TIMESTAMP 10 206 #define TCPOLEN_MD5SIG 18 207 #define TCPOLEN_FASTOPEN_BASE 2 208 #define TCPOLEN_EXP_FASTOPEN_BASE 4 209 #define TCPOLEN_EXP_SMC_BASE 6 210 211 /* But this is what stacks really send out. */ 212 #define TCPOLEN_TSTAMP_ALIGNED 12 213 #define TCPOLEN_WSCALE_ALIGNED 4 214 #define TCPOLEN_SACKPERM_ALIGNED 4 215 #define TCPOLEN_SACK_BASE 2 216 #define TCPOLEN_SACK_BASE_ALIGNED 4 217 #define TCPOLEN_SACK_PERBLOCK 8 218 #define TCPOLEN_MD5SIG_ALIGNED 20 219 #define TCPOLEN_MSS_ALIGNED 4 220 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 221 222 /* Flags in tp->nonagle */ 223 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 224 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 225 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 226 227 /* TCP thin-stream limits */ 228 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 229 230 /* TCP initial congestion window as per rfc6928 */ 231 #define TCP_INIT_CWND 10 232 233 /* Bit Flags for sysctl_tcp_fastopen */ 234 #define TFO_CLIENT_ENABLE 1 235 #define TFO_SERVER_ENABLE 2 236 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 237 238 /* Accept SYN data w/o any cookie option */ 239 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 240 241 /* Force enable TFO on all listeners, i.e., not requiring the 242 * TCP_FASTOPEN socket option. 243 */ 244 #define TFO_SERVER_WO_SOCKOPT1 0x400 245 246 247 /* sysctl variables for tcp */ 248 extern int sysctl_tcp_max_orphans; 249 extern long sysctl_tcp_mem[3]; 250 251 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 252 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 253 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 254 255 extern atomic_long_t tcp_memory_allocated; 256 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 257 258 extern struct percpu_counter tcp_sockets_allocated; 259 extern unsigned long tcp_memory_pressure; 260 261 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 262 static inline bool tcp_under_memory_pressure(const struct sock *sk) 263 { 264 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 265 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 266 return true; 267 268 return READ_ONCE(tcp_memory_pressure); 269 } 270 /* 271 * The next routines deal with comparing 32 bit unsigned ints 272 * and worry about wraparound (automatic with unsigned arithmetic). 273 */ 274 275 static inline bool before(__u32 seq1, __u32 seq2) 276 { 277 return (__s32)(seq1-seq2) < 0; 278 } 279 #define after(seq2, seq1) before(seq1, seq2) 280 281 /* is s2<=s1<=s3 ? */ 282 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 283 { 284 return seq3 - seq2 >= seq1 - seq2; 285 } 286 287 static inline bool tcp_out_of_memory(struct sock *sk) 288 { 289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 291 return true; 292 return false; 293 } 294 295 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 296 { 297 sk_wmem_queued_add(sk, -skb->truesize); 298 if (!skb_zcopy_pure(skb)) 299 sk_mem_uncharge(sk, skb->truesize); 300 else 301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 302 __kfree_skb(skb); 303 } 304 305 void sk_forced_mem_schedule(struct sock *sk, int size); 306 307 bool tcp_check_oom(struct sock *sk, int shift); 308 309 310 extern struct proto tcp_prot; 311 312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 316 317 void tcp_tasklet_init(void); 318 319 int tcp_v4_err(struct sk_buff *skb, u32); 320 321 void tcp_shutdown(struct sock *sk, int how); 322 323 int tcp_v4_early_demux(struct sk_buff *skb); 324 int tcp_v4_rcv(struct sk_buff *skb); 325 326 void tcp_remove_empty_skb(struct sock *sk); 327 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 328 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 329 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 330 size_t size, struct ubuf_info *uarg); 331 void tcp_splice_eof(struct socket *sock); 332 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 333 int tcp_wmem_schedule(struct sock *sk, int copy); 334 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 335 int size_goal); 336 void tcp_release_cb(struct sock *sk); 337 void tcp_wfree(struct sk_buff *skb); 338 void tcp_write_timer_handler(struct sock *sk); 339 void tcp_delack_timer_handler(struct sock *sk); 340 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 341 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 342 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_space_adjust(struct sock *sk); 344 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 345 void tcp_twsk_destructor(struct sock *sk); 346 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 347 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 348 struct pipe_inode_info *pipe, size_t len, 349 unsigned int flags); 350 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 351 bool force_schedule); 352 353 static inline void tcp_dec_quickack_mode(struct sock *sk) 354 { 355 struct inet_connection_sock *icsk = inet_csk(sk); 356 357 if (icsk->icsk_ack.quick) { 358 /* How many ACKs S/ACKing new data have we sent? */ 359 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 360 361 if (pkts >= icsk->icsk_ack.quick) { 362 icsk->icsk_ack.quick = 0; 363 /* Leaving quickack mode we deflate ATO. */ 364 icsk->icsk_ack.ato = TCP_ATO_MIN; 365 } else 366 icsk->icsk_ack.quick -= pkts; 367 } 368 } 369 370 #define TCP_ECN_OK 1 371 #define TCP_ECN_QUEUE_CWR 2 372 #define TCP_ECN_DEMAND_CWR 4 373 #define TCP_ECN_SEEN 8 374 375 enum tcp_tw_status { 376 TCP_TW_SUCCESS = 0, 377 TCP_TW_RST = 1, 378 TCP_TW_ACK = 2, 379 TCP_TW_SYN = 3 380 }; 381 382 383 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 384 struct sk_buff *skb, 385 const struct tcphdr *th); 386 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 387 struct request_sock *req, bool fastopen, 388 bool *lost_race); 389 int tcp_child_process(struct sock *parent, struct sock *child, 390 struct sk_buff *skb); 391 void tcp_enter_loss(struct sock *sk); 392 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 393 void tcp_clear_retrans(struct tcp_sock *tp); 394 void tcp_update_metrics(struct sock *sk); 395 void tcp_init_metrics(struct sock *sk); 396 void tcp_metrics_init(void); 397 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 398 void __tcp_close(struct sock *sk, long timeout); 399 void tcp_close(struct sock *sk, long timeout); 400 void tcp_init_sock(struct sock *sk); 401 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 402 __poll_t tcp_poll(struct file *file, struct socket *sock, 403 struct poll_table_struct *wait); 404 int do_tcp_getsockopt(struct sock *sk, int level, 405 int optname, sockptr_t optval, sockptr_t optlen); 406 int tcp_getsockopt(struct sock *sk, int level, int optname, 407 char __user *optval, int __user *optlen); 408 bool tcp_bpf_bypass_getsockopt(int level, int optname); 409 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 410 sockptr_t optval, unsigned int optlen); 411 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 412 unsigned int optlen); 413 void tcp_set_keepalive(struct sock *sk, int val); 414 void tcp_syn_ack_timeout(const struct request_sock *req); 415 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 416 int flags, int *addr_len); 417 int tcp_set_rcvlowat(struct sock *sk, int val); 418 int tcp_set_window_clamp(struct sock *sk, int val); 419 void tcp_update_recv_tstamps(struct sk_buff *skb, 420 struct scm_timestamping_internal *tss); 421 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 422 struct scm_timestamping_internal *tss); 423 void tcp_data_ready(struct sock *sk); 424 #ifdef CONFIG_MMU 425 int tcp_mmap(struct file *file, struct socket *sock, 426 struct vm_area_struct *vma); 427 #endif 428 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 429 struct tcp_options_received *opt_rx, 430 int estab, struct tcp_fastopen_cookie *foc); 431 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 432 433 /* 434 * BPF SKB-less helpers 435 */ 436 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 437 struct tcphdr *th, u32 *cookie); 438 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 439 struct tcphdr *th, u32 *cookie); 440 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 441 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 442 const struct tcp_request_sock_ops *af_ops, 443 struct sock *sk, struct tcphdr *th); 444 /* 445 * TCP v4 functions exported for the inet6 API 446 */ 447 448 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 449 void tcp_v4_mtu_reduced(struct sock *sk); 450 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 451 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 452 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 453 struct sock *tcp_create_openreq_child(const struct sock *sk, 454 struct request_sock *req, 455 struct sk_buff *skb); 456 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 457 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 458 struct request_sock *req, 459 struct dst_entry *dst, 460 struct request_sock *req_unhash, 461 bool *own_req); 462 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 463 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 464 int tcp_connect(struct sock *sk); 465 enum tcp_synack_type { 466 TCP_SYNACK_NORMAL, 467 TCP_SYNACK_FASTOPEN, 468 TCP_SYNACK_COOKIE, 469 }; 470 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 471 struct request_sock *req, 472 struct tcp_fastopen_cookie *foc, 473 enum tcp_synack_type synack_type, 474 struct sk_buff *syn_skb); 475 int tcp_disconnect(struct sock *sk, int flags); 476 477 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 478 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 479 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 480 481 /* From syncookies.c */ 482 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 483 struct request_sock *req, 484 struct dst_entry *dst, u32 tsoff); 485 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 486 u32 cookie); 487 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 488 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 489 const struct tcp_request_sock_ops *af_ops, 490 struct sock *sk, struct sk_buff *skb); 491 #ifdef CONFIG_SYN_COOKIES 492 493 /* Syncookies use a monotonic timer which increments every 60 seconds. 494 * This counter is used both as a hash input and partially encoded into 495 * the cookie value. A cookie is only validated further if the delta 496 * between the current counter value and the encoded one is less than this, 497 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 498 * the counter advances immediately after a cookie is generated). 499 */ 500 #define MAX_SYNCOOKIE_AGE 2 501 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 502 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 503 504 /* syncookies: remember time of last synqueue overflow 505 * But do not dirty this field too often (once per second is enough) 506 * It is racy as we do not hold a lock, but race is very minor. 507 */ 508 static inline void tcp_synq_overflow(const struct sock *sk) 509 { 510 unsigned int last_overflow; 511 unsigned int now = jiffies; 512 513 if (sk->sk_reuseport) { 514 struct sock_reuseport *reuse; 515 516 reuse = rcu_dereference(sk->sk_reuseport_cb); 517 if (likely(reuse)) { 518 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 519 if (!time_between32(now, last_overflow, 520 last_overflow + HZ)) 521 WRITE_ONCE(reuse->synq_overflow_ts, now); 522 return; 523 } 524 } 525 526 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 527 if (!time_between32(now, last_overflow, last_overflow + HZ)) 528 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 529 } 530 531 /* syncookies: no recent synqueue overflow on this listening socket? */ 532 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 533 { 534 unsigned int last_overflow; 535 unsigned int now = jiffies; 536 537 if (sk->sk_reuseport) { 538 struct sock_reuseport *reuse; 539 540 reuse = rcu_dereference(sk->sk_reuseport_cb); 541 if (likely(reuse)) { 542 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 543 return !time_between32(now, last_overflow - HZ, 544 last_overflow + 545 TCP_SYNCOOKIE_VALID); 546 } 547 } 548 549 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 550 551 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 552 * then we're under synflood. However, we have to use 553 * 'last_overflow - HZ' as lower bound. That's because a concurrent 554 * tcp_synq_overflow() could update .ts_recent_stamp after we read 555 * jiffies but before we store .ts_recent_stamp into last_overflow, 556 * which could lead to rejecting a valid syncookie. 557 */ 558 return !time_between32(now, last_overflow - HZ, 559 last_overflow + TCP_SYNCOOKIE_VALID); 560 } 561 562 static inline u32 tcp_cookie_time(void) 563 { 564 u64 val = get_jiffies_64(); 565 566 do_div(val, TCP_SYNCOOKIE_PERIOD); 567 return val; 568 } 569 570 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 571 u16 *mssp); 572 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 573 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 574 bool cookie_timestamp_decode(const struct net *net, 575 struct tcp_options_received *opt); 576 bool cookie_ecn_ok(const struct tcp_options_received *opt, 577 const struct net *net, const struct dst_entry *dst); 578 579 /* From net/ipv6/syncookies.c */ 580 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 581 u32 cookie); 582 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 583 584 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 585 const struct tcphdr *th, u16 *mssp); 586 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 587 #endif 588 /* tcp_output.c */ 589 590 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 591 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 592 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 593 int nonagle); 594 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 595 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 596 void tcp_retransmit_timer(struct sock *sk); 597 void tcp_xmit_retransmit_queue(struct sock *); 598 void tcp_simple_retransmit(struct sock *); 599 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 600 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 601 enum tcp_queue { 602 TCP_FRAG_IN_WRITE_QUEUE, 603 TCP_FRAG_IN_RTX_QUEUE, 604 }; 605 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 606 struct sk_buff *skb, u32 len, 607 unsigned int mss_now, gfp_t gfp); 608 609 void tcp_send_probe0(struct sock *); 610 int tcp_write_wakeup(struct sock *, int mib); 611 void tcp_send_fin(struct sock *sk); 612 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 613 int tcp_send_synack(struct sock *); 614 void tcp_push_one(struct sock *, unsigned int mss_now); 615 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 616 void tcp_send_ack(struct sock *sk); 617 void tcp_send_delayed_ack(struct sock *sk); 618 void tcp_send_loss_probe(struct sock *sk); 619 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 620 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 621 const struct sk_buff *next_skb); 622 623 /* tcp_input.c */ 624 void tcp_rearm_rto(struct sock *sk); 625 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 626 void tcp_reset(struct sock *sk, struct sk_buff *skb); 627 void tcp_fin(struct sock *sk); 628 void tcp_check_space(struct sock *sk); 629 void tcp_sack_compress_send_ack(struct sock *sk); 630 631 /* tcp_timer.c */ 632 void tcp_init_xmit_timers(struct sock *); 633 static inline void tcp_clear_xmit_timers(struct sock *sk) 634 { 635 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 636 __sock_put(sk); 637 638 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 639 __sock_put(sk); 640 641 inet_csk_clear_xmit_timers(sk); 642 } 643 644 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 645 unsigned int tcp_current_mss(struct sock *sk); 646 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 647 648 /* Bound MSS / TSO packet size with the half of the window */ 649 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 650 { 651 int cutoff; 652 653 /* When peer uses tiny windows, there is no use in packetizing 654 * to sub-MSS pieces for the sake of SWS or making sure there 655 * are enough packets in the pipe for fast recovery. 656 * 657 * On the other hand, for extremely large MSS devices, handling 658 * smaller than MSS windows in this way does make sense. 659 */ 660 if (tp->max_window > TCP_MSS_DEFAULT) 661 cutoff = (tp->max_window >> 1); 662 else 663 cutoff = tp->max_window; 664 665 if (cutoff && pktsize > cutoff) 666 return max_t(int, cutoff, 68U - tp->tcp_header_len); 667 else 668 return pktsize; 669 } 670 671 /* tcp.c */ 672 void tcp_get_info(struct sock *, struct tcp_info *); 673 674 /* Read 'sendfile()'-style from a TCP socket */ 675 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 676 sk_read_actor_t recv_actor); 677 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 678 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 679 void tcp_read_done(struct sock *sk, size_t len); 680 681 void tcp_initialize_rcv_mss(struct sock *sk); 682 683 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 684 int tcp_mss_to_mtu(struct sock *sk, int mss); 685 void tcp_mtup_init(struct sock *sk); 686 687 static inline void tcp_bound_rto(const struct sock *sk) 688 { 689 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 690 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 691 } 692 693 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 694 { 695 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 696 } 697 698 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 699 { 700 /* mptcp hooks are only on the slow path */ 701 if (sk_is_mptcp((struct sock *)tp)) 702 return; 703 704 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 705 ntohl(TCP_FLAG_ACK) | 706 snd_wnd); 707 } 708 709 static inline void tcp_fast_path_on(struct tcp_sock *tp) 710 { 711 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 712 } 713 714 static inline void tcp_fast_path_check(struct sock *sk) 715 { 716 struct tcp_sock *tp = tcp_sk(sk); 717 718 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 719 tp->rcv_wnd && 720 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 721 !tp->urg_data) 722 tcp_fast_path_on(tp); 723 } 724 725 u32 tcp_delack_max(const struct sock *sk); 726 727 /* Compute the actual rto_min value */ 728 static inline u32 tcp_rto_min(const struct sock *sk) 729 { 730 const struct dst_entry *dst = __sk_dst_get(sk); 731 u32 rto_min = inet_csk(sk)->icsk_rto_min; 732 733 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 734 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 735 return rto_min; 736 } 737 738 static inline u32 tcp_rto_min_us(const struct sock *sk) 739 { 740 return jiffies_to_usecs(tcp_rto_min(sk)); 741 } 742 743 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 744 { 745 return dst_metric_locked(dst, RTAX_CC_ALGO); 746 } 747 748 /* Minimum RTT in usec. ~0 means not available. */ 749 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 750 { 751 return minmax_get(&tp->rtt_min); 752 } 753 754 /* Compute the actual receive window we are currently advertising. 755 * Rcv_nxt can be after the window if our peer push more data 756 * than the offered window. 757 */ 758 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 759 { 760 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 761 762 if (win < 0) 763 win = 0; 764 return (u32) win; 765 } 766 767 /* Choose a new window, without checks for shrinking, and without 768 * scaling applied to the result. The caller does these things 769 * if necessary. This is a "raw" window selection. 770 */ 771 u32 __tcp_select_window(struct sock *sk); 772 773 void tcp_send_window_probe(struct sock *sk); 774 775 /* TCP uses 32bit jiffies to save some space. 776 * Note that this is different from tcp_time_stamp, which 777 * historically has been the same until linux-4.13. 778 */ 779 #define tcp_jiffies32 ((u32)jiffies) 780 781 /* 782 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 783 * It is no longer tied to jiffies, but to 1 ms clock. 784 * Note: double check if you want to use tcp_jiffies32 instead of this. 785 */ 786 #define TCP_TS_HZ 1000 787 788 static inline u64 tcp_clock_ns(void) 789 { 790 return ktime_get_ns(); 791 } 792 793 static inline u64 tcp_clock_us(void) 794 { 795 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 796 } 797 798 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 799 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 800 { 801 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 802 } 803 804 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 805 static inline u32 tcp_ns_to_ts(u64 ns) 806 { 807 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 808 } 809 810 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 811 static inline u32 tcp_time_stamp_raw(void) 812 { 813 return tcp_ns_to_ts(tcp_clock_ns()); 814 } 815 816 void tcp_mstamp_refresh(struct tcp_sock *tp); 817 818 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 819 { 820 return max_t(s64, t1 - t0, 0); 821 } 822 823 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 824 { 825 return tcp_ns_to_ts(skb->skb_mstamp_ns); 826 } 827 828 /* provide the departure time in us unit */ 829 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 830 { 831 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 832 } 833 834 835 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 836 837 #define TCPHDR_FIN 0x01 838 #define TCPHDR_SYN 0x02 839 #define TCPHDR_RST 0x04 840 #define TCPHDR_PSH 0x08 841 #define TCPHDR_ACK 0x10 842 #define TCPHDR_URG 0x20 843 #define TCPHDR_ECE 0x40 844 #define TCPHDR_CWR 0x80 845 846 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 847 848 /* This is what the send packet queuing engine uses to pass 849 * TCP per-packet control information to the transmission code. 850 * We also store the host-order sequence numbers in here too. 851 * This is 44 bytes if IPV6 is enabled. 852 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 853 */ 854 struct tcp_skb_cb { 855 __u32 seq; /* Starting sequence number */ 856 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 857 union { 858 /* Note : tcp_tw_isn is used in input path only 859 * (isn chosen by tcp_timewait_state_process()) 860 * 861 * tcp_gso_segs/size are used in write queue only, 862 * cf tcp_skb_pcount()/tcp_skb_mss() 863 */ 864 __u32 tcp_tw_isn; 865 struct { 866 u16 tcp_gso_segs; 867 u16 tcp_gso_size; 868 }; 869 }; 870 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 871 872 __u8 sacked; /* State flags for SACK. */ 873 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 874 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 875 #define TCPCB_LOST 0x04 /* SKB is lost */ 876 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 877 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 878 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 879 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 880 TCPCB_REPAIRED) 881 882 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 883 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 884 eor:1, /* Is skb MSG_EOR marked? */ 885 has_rxtstamp:1, /* SKB has a RX timestamp */ 886 unused:5; 887 __u32 ack_seq; /* Sequence number ACK'd */ 888 union { 889 struct { 890 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 891 /* There is space for up to 24 bytes */ 892 __u32 is_app_limited:1, /* cwnd not fully used? */ 893 delivered_ce:20, 894 unused:11; 895 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 896 __u32 delivered; 897 /* start of send pipeline phase */ 898 u64 first_tx_mstamp; 899 /* when we reached the "delivered" count */ 900 u64 delivered_mstamp; 901 } tx; /* only used for outgoing skbs */ 902 union { 903 struct inet_skb_parm h4; 904 #if IS_ENABLED(CONFIG_IPV6) 905 struct inet6_skb_parm h6; 906 #endif 907 } header; /* For incoming skbs */ 908 }; 909 }; 910 911 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 912 913 extern const struct inet_connection_sock_af_ops ipv4_specific; 914 915 #if IS_ENABLED(CONFIG_IPV6) 916 /* This is the variant of inet6_iif() that must be used by TCP, 917 * as TCP moves IP6CB into a different location in skb->cb[] 918 */ 919 static inline int tcp_v6_iif(const struct sk_buff *skb) 920 { 921 return TCP_SKB_CB(skb)->header.h6.iif; 922 } 923 924 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 925 { 926 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 927 928 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 929 } 930 931 /* TCP_SKB_CB reference means this can not be used from early demux */ 932 static inline int tcp_v6_sdif(const struct sk_buff *skb) 933 { 934 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 935 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 936 return TCP_SKB_CB(skb)->header.h6.iif; 937 #endif 938 return 0; 939 } 940 941 extern const struct inet_connection_sock_af_ops ipv6_specific; 942 943 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 944 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 945 void tcp_v6_early_demux(struct sk_buff *skb); 946 947 #endif 948 949 /* TCP_SKB_CB reference means this can not be used from early demux */ 950 static inline int tcp_v4_sdif(struct sk_buff *skb) 951 { 952 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 953 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 954 return TCP_SKB_CB(skb)->header.h4.iif; 955 #endif 956 return 0; 957 } 958 959 /* Due to TSO, an SKB can be composed of multiple actual 960 * packets. To keep these tracked properly, we use this. 961 */ 962 static inline int tcp_skb_pcount(const struct sk_buff *skb) 963 { 964 return TCP_SKB_CB(skb)->tcp_gso_segs; 965 } 966 967 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 968 { 969 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 970 } 971 972 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 973 { 974 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 975 } 976 977 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 978 static inline int tcp_skb_mss(const struct sk_buff *skb) 979 { 980 return TCP_SKB_CB(skb)->tcp_gso_size; 981 } 982 983 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 984 { 985 return likely(!TCP_SKB_CB(skb)->eor); 986 } 987 988 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 989 const struct sk_buff *from) 990 { 991 return likely(tcp_skb_can_collapse_to(to) && 992 mptcp_skb_can_collapse(to, from) && 993 skb_pure_zcopy_same(to, from)); 994 } 995 996 /* Events passed to congestion control interface */ 997 enum tcp_ca_event { 998 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 999 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1000 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1001 CA_EVENT_LOSS, /* loss timeout */ 1002 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1003 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1004 }; 1005 1006 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1007 enum tcp_ca_ack_event_flags { 1008 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1009 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1010 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1011 }; 1012 1013 /* 1014 * Interface for adding new TCP congestion control handlers 1015 */ 1016 #define TCP_CA_NAME_MAX 16 1017 #define TCP_CA_MAX 128 1018 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1019 1020 #define TCP_CA_UNSPEC 0 1021 1022 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1023 #define TCP_CONG_NON_RESTRICTED 0x1 1024 /* Requires ECN/ECT set on all packets */ 1025 #define TCP_CONG_NEEDS_ECN 0x2 1026 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1027 1028 union tcp_cc_info; 1029 1030 struct ack_sample { 1031 u32 pkts_acked; 1032 s32 rtt_us; 1033 u32 in_flight; 1034 }; 1035 1036 /* A rate sample measures the number of (original/retransmitted) data 1037 * packets delivered "delivered" over an interval of time "interval_us". 1038 * The tcp_rate.c code fills in the rate sample, and congestion 1039 * control modules that define a cong_control function to run at the end 1040 * of ACK processing can optionally chose to consult this sample when 1041 * setting cwnd and pacing rate. 1042 * A sample is invalid if "delivered" or "interval_us" is negative. 1043 */ 1044 struct rate_sample { 1045 u64 prior_mstamp; /* starting timestamp for interval */ 1046 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1047 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1048 s32 delivered; /* number of packets delivered over interval */ 1049 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1050 long interval_us; /* time for tp->delivered to incr "delivered" */ 1051 u32 snd_interval_us; /* snd interval for delivered packets */ 1052 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1053 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1054 int losses; /* number of packets marked lost upon ACK */ 1055 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1056 u32 prior_in_flight; /* in flight before this ACK */ 1057 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1058 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1059 bool is_retrans; /* is sample from retransmission? */ 1060 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1061 }; 1062 1063 struct tcp_congestion_ops { 1064 /* fast path fields are put first to fill one cache line */ 1065 1066 /* return slow start threshold (required) */ 1067 u32 (*ssthresh)(struct sock *sk); 1068 1069 /* do new cwnd calculation (required) */ 1070 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1071 1072 /* call before changing ca_state (optional) */ 1073 void (*set_state)(struct sock *sk, u8 new_state); 1074 1075 /* call when cwnd event occurs (optional) */ 1076 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1077 1078 /* call when ack arrives (optional) */ 1079 void (*in_ack_event)(struct sock *sk, u32 flags); 1080 1081 /* hook for packet ack accounting (optional) */ 1082 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1083 1084 /* override sysctl_tcp_min_tso_segs */ 1085 u32 (*min_tso_segs)(struct sock *sk); 1086 1087 /* call when packets are delivered to update cwnd and pacing rate, 1088 * after all the ca_state processing. (optional) 1089 */ 1090 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1091 1092 1093 /* new value of cwnd after loss (required) */ 1094 u32 (*undo_cwnd)(struct sock *sk); 1095 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1096 u32 (*sndbuf_expand)(struct sock *sk); 1097 1098 /* control/slow paths put last */ 1099 /* get info for inet_diag (optional) */ 1100 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1101 union tcp_cc_info *info); 1102 1103 char name[TCP_CA_NAME_MAX]; 1104 struct module *owner; 1105 struct list_head list; 1106 u32 key; 1107 u32 flags; 1108 1109 /* initialize private data (optional) */ 1110 void (*init)(struct sock *sk); 1111 /* cleanup private data (optional) */ 1112 void (*release)(struct sock *sk); 1113 } ____cacheline_aligned_in_smp; 1114 1115 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1116 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1117 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1118 struct tcp_congestion_ops *old_type); 1119 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1120 1121 void tcp_assign_congestion_control(struct sock *sk); 1122 void tcp_init_congestion_control(struct sock *sk); 1123 void tcp_cleanup_congestion_control(struct sock *sk); 1124 int tcp_set_default_congestion_control(struct net *net, const char *name); 1125 void tcp_get_default_congestion_control(struct net *net, char *name); 1126 void tcp_get_available_congestion_control(char *buf, size_t len); 1127 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1128 int tcp_set_allowed_congestion_control(char *allowed); 1129 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1130 bool cap_net_admin); 1131 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1132 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1133 1134 u32 tcp_reno_ssthresh(struct sock *sk); 1135 u32 tcp_reno_undo_cwnd(struct sock *sk); 1136 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1137 extern struct tcp_congestion_ops tcp_reno; 1138 1139 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1140 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1141 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1142 #ifdef CONFIG_INET 1143 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1144 #else 1145 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1146 { 1147 return NULL; 1148 } 1149 #endif 1150 1151 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1152 { 1153 const struct inet_connection_sock *icsk = inet_csk(sk); 1154 1155 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1156 } 1157 1158 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1159 { 1160 const struct inet_connection_sock *icsk = inet_csk(sk); 1161 1162 if (icsk->icsk_ca_ops->cwnd_event) 1163 icsk->icsk_ca_ops->cwnd_event(sk, event); 1164 } 1165 1166 /* From tcp_cong.c */ 1167 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1168 1169 /* From tcp_rate.c */ 1170 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1171 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1172 struct rate_sample *rs); 1173 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1174 bool is_sack_reneg, struct rate_sample *rs); 1175 void tcp_rate_check_app_limited(struct sock *sk); 1176 1177 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1178 { 1179 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1180 } 1181 1182 /* These functions determine how the current flow behaves in respect of SACK 1183 * handling. SACK is negotiated with the peer, and therefore it can vary 1184 * between different flows. 1185 * 1186 * tcp_is_sack - SACK enabled 1187 * tcp_is_reno - No SACK 1188 */ 1189 static inline int tcp_is_sack(const struct tcp_sock *tp) 1190 { 1191 return likely(tp->rx_opt.sack_ok); 1192 } 1193 1194 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1195 { 1196 return !tcp_is_sack(tp); 1197 } 1198 1199 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1200 { 1201 return tp->sacked_out + tp->lost_out; 1202 } 1203 1204 /* This determines how many packets are "in the network" to the best 1205 * of our knowledge. In many cases it is conservative, but where 1206 * detailed information is available from the receiver (via SACK 1207 * blocks etc.) we can make more aggressive calculations. 1208 * 1209 * Use this for decisions involving congestion control, use just 1210 * tp->packets_out to determine if the send queue is empty or not. 1211 * 1212 * Read this equation as: 1213 * 1214 * "Packets sent once on transmission queue" MINUS 1215 * "Packets left network, but not honestly ACKed yet" PLUS 1216 * "Packets fast retransmitted" 1217 */ 1218 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1219 { 1220 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1221 } 1222 1223 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1224 1225 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1226 { 1227 return tp->snd_cwnd; 1228 } 1229 1230 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1231 { 1232 WARN_ON_ONCE((int)val <= 0); 1233 tp->snd_cwnd = val; 1234 } 1235 1236 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1237 { 1238 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1239 } 1240 1241 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1242 { 1243 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1244 } 1245 1246 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1247 { 1248 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1249 (1 << inet_csk(sk)->icsk_ca_state); 1250 } 1251 1252 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1253 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1254 * ssthresh. 1255 */ 1256 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1257 { 1258 const struct tcp_sock *tp = tcp_sk(sk); 1259 1260 if (tcp_in_cwnd_reduction(sk)) 1261 return tp->snd_ssthresh; 1262 else 1263 return max(tp->snd_ssthresh, 1264 ((tcp_snd_cwnd(tp) >> 1) + 1265 (tcp_snd_cwnd(tp) >> 2))); 1266 } 1267 1268 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1269 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1270 1271 void tcp_enter_cwr(struct sock *sk); 1272 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1273 1274 /* The maximum number of MSS of available cwnd for which TSO defers 1275 * sending if not using sysctl_tcp_tso_win_divisor. 1276 */ 1277 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1278 { 1279 return 3; 1280 } 1281 1282 /* Returns end sequence number of the receiver's advertised window */ 1283 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1284 { 1285 return tp->snd_una + tp->snd_wnd; 1286 } 1287 1288 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1289 * flexible approach. The RFC suggests cwnd should not be raised unless 1290 * it was fully used previously. And that's exactly what we do in 1291 * congestion avoidance mode. But in slow start we allow cwnd to grow 1292 * as long as the application has used half the cwnd. 1293 * Example : 1294 * cwnd is 10 (IW10), but application sends 9 frames. 1295 * We allow cwnd to reach 18 when all frames are ACKed. 1296 * This check is safe because it's as aggressive as slow start which already 1297 * risks 100% overshoot. The advantage is that we discourage application to 1298 * either send more filler packets or data to artificially blow up the cwnd 1299 * usage, and allow application-limited process to probe bw more aggressively. 1300 */ 1301 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1302 { 1303 const struct tcp_sock *tp = tcp_sk(sk); 1304 1305 if (tp->is_cwnd_limited) 1306 return true; 1307 1308 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1309 if (tcp_in_slow_start(tp)) 1310 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1311 1312 return false; 1313 } 1314 1315 /* BBR congestion control needs pacing. 1316 * Same remark for SO_MAX_PACING_RATE. 1317 * sch_fq packet scheduler is efficiently handling pacing, 1318 * but is not always installed/used. 1319 * Return true if TCP stack should pace packets itself. 1320 */ 1321 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1322 { 1323 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1324 } 1325 1326 /* Estimates in how many jiffies next packet for this flow can be sent. 1327 * Scheduling a retransmit timer too early would be silly. 1328 */ 1329 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1330 { 1331 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1332 1333 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1334 } 1335 1336 static inline void tcp_reset_xmit_timer(struct sock *sk, 1337 const int what, 1338 unsigned long when, 1339 const unsigned long max_when) 1340 { 1341 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1342 max_when); 1343 } 1344 1345 /* Something is really bad, we could not queue an additional packet, 1346 * because qdisc is full or receiver sent a 0 window, or we are paced. 1347 * We do not want to add fuel to the fire, or abort too early, 1348 * so make sure the timer we arm now is at least 200ms in the future, 1349 * regardless of current icsk_rto value (as it could be ~2ms) 1350 */ 1351 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1352 { 1353 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1354 } 1355 1356 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1357 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1358 unsigned long max_when) 1359 { 1360 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1361 inet_csk(sk)->icsk_backoff); 1362 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1363 1364 return (unsigned long)min_t(u64, when, max_when); 1365 } 1366 1367 static inline void tcp_check_probe_timer(struct sock *sk) 1368 { 1369 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1370 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1371 tcp_probe0_base(sk), TCP_RTO_MAX); 1372 } 1373 1374 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1375 { 1376 tp->snd_wl1 = seq; 1377 } 1378 1379 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1380 { 1381 tp->snd_wl1 = seq; 1382 } 1383 1384 /* 1385 * Calculate(/check) TCP checksum 1386 */ 1387 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1388 __be32 daddr, __wsum base) 1389 { 1390 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1391 } 1392 1393 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1394 { 1395 return !skb_csum_unnecessary(skb) && 1396 __skb_checksum_complete(skb); 1397 } 1398 1399 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1400 enum skb_drop_reason *reason); 1401 1402 1403 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1404 void tcp_set_state(struct sock *sk, int state); 1405 void tcp_done(struct sock *sk); 1406 int tcp_abort(struct sock *sk, int err); 1407 1408 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1409 { 1410 rx_opt->dsack = 0; 1411 rx_opt->num_sacks = 0; 1412 } 1413 1414 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1415 1416 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1417 { 1418 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1419 struct tcp_sock *tp = tcp_sk(sk); 1420 s32 delta; 1421 1422 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1423 tp->packets_out || ca_ops->cong_control) 1424 return; 1425 delta = tcp_jiffies32 - tp->lsndtime; 1426 if (delta > inet_csk(sk)->icsk_rto) 1427 tcp_cwnd_restart(sk, delta); 1428 } 1429 1430 /* Determine a window scaling and initial window to offer. */ 1431 void tcp_select_initial_window(const struct sock *sk, int __space, 1432 __u32 mss, __u32 *rcv_wnd, 1433 __u32 *window_clamp, int wscale_ok, 1434 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1435 1436 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1437 { 1438 s64 scaled_space = (s64)space * scaling_ratio; 1439 1440 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1441 } 1442 1443 static inline int tcp_win_from_space(const struct sock *sk, int space) 1444 { 1445 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1446 } 1447 1448 /* inverse of __tcp_win_from_space() */ 1449 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1450 { 1451 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1452 1453 do_div(val, scaling_ratio); 1454 return val; 1455 } 1456 1457 static inline int tcp_space_from_win(const struct sock *sk, int win) 1458 { 1459 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1460 } 1461 1462 static inline void tcp_scaling_ratio_init(struct sock *sk) 1463 { 1464 /* Assume a conservative default of 1200 bytes of payload per 4K page. 1465 * This may be adjusted later in tcp_measure_rcv_mss(). 1466 */ 1467 tcp_sk(sk)->scaling_ratio = (1200 << TCP_RMEM_TO_WIN_SCALE) / 1468 SKB_TRUESIZE(4096); 1469 } 1470 1471 /* Note: caller must be prepared to deal with negative returns */ 1472 static inline int tcp_space(const struct sock *sk) 1473 { 1474 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1475 READ_ONCE(sk->sk_backlog.len) - 1476 atomic_read(&sk->sk_rmem_alloc)); 1477 } 1478 1479 static inline int tcp_full_space(const struct sock *sk) 1480 { 1481 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1482 } 1483 1484 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1485 { 1486 int unused_mem = sk_unused_reserved_mem(sk); 1487 struct tcp_sock *tp = tcp_sk(sk); 1488 1489 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1490 if (unused_mem) 1491 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1492 tcp_win_from_space(sk, unused_mem)); 1493 } 1494 1495 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1496 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1497 1498 1499 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1500 * If 87.5 % (7/8) of the space has been consumed, we want to override 1501 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1502 * len/truesize ratio. 1503 */ 1504 static inline bool tcp_rmem_pressure(const struct sock *sk) 1505 { 1506 int rcvbuf, threshold; 1507 1508 if (tcp_under_memory_pressure(sk)) 1509 return true; 1510 1511 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1512 threshold = rcvbuf - (rcvbuf >> 3); 1513 1514 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1515 } 1516 1517 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1518 { 1519 const struct tcp_sock *tp = tcp_sk(sk); 1520 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1521 1522 if (avail <= 0) 1523 return false; 1524 1525 return (avail >= target) || tcp_rmem_pressure(sk) || 1526 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1527 } 1528 1529 extern void tcp_openreq_init_rwin(struct request_sock *req, 1530 const struct sock *sk_listener, 1531 const struct dst_entry *dst); 1532 1533 void tcp_enter_memory_pressure(struct sock *sk); 1534 void tcp_leave_memory_pressure(struct sock *sk); 1535 1536 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1537 { 1538 struct net *net = sock_net((struct sock *)tp); 1539 int val; 1540 1541 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1542 * and do_tcp_setsockopt(). 1543 */ 1544 val = READ_ONCE(tp->keepalive_intvl); 1545 1546 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1547 } 1548 1549 static inline int keepalive_time_when(const struct tcp_sock *tp) 1550 { 1551 struct net *net = sock_net((struct sock *)tp); 1552 int val; 1553 1554 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1555 val = READ_ONCE(tp->keepalive_time); 1556 1557 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1558 } 1559 1560 static inline int keepalive_probes(const struct tcp_sock *tp) 1561 { 1562 struct net *net = sock_net((struct sock *)tp); 1563 int val; 1564 1565 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1566 * and do_tcp_setsockopt(). 1567 */ 1568 val = READ_ONCE(tp->keepalive_probes); 1569 1570 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1571 } 1572 1573 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1574 { 1575 const struct inet_connection_sock *icsk = &tp->inet_conn; 1576 1577 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1578 tcp_jiffies32 - tp->rcv_tstamp); 1579 } 1580 1581 static inline int tcp_fin_time(const struct sock *sk) 1582 { 1583 int fin_timeout = tcp_sk(sk)->linger2 ? : 1584 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1585 const int rto = inet_csk(sk)->icsk_rto; 1586 1587 if (fin_timeout < (rto << 2) - (rto >> 1)) 1588 fin_timeout = (rto << 2) - (rto >> 1); 1589 1590 return fin_timeout; 1591 } 1592 1593 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1594 int paws_win) 1595 { 1596 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1597 return true; 1598 if (unlikely(!time_before32(ktime_get_seconds(), 1599 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1600 return true; 1601 /* 1602 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1603 * then following tcp messages have valid values. Ignore 0 value, 1604 * or else 'negative' tsval might forbid us to accept their packets. 1605 */ 1606 if (!rx_opt->ts_recent) 1607 return true; 1608 return false; 1609 } 1610 1611 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1612 int rst) 1613 { 1614 if (tcp_paws_check(rx_opt, 0)) 1615 return false; 1616 1617 /* RST segments are not recommended to carry timestamp, 1618 and, if they do, it is recommended to ignore PAWS because 1619 "their cleanup function should take precedence over timestamps." 1620 Certainly, it is mistake. It is necessary to understand the reasons 1621 of this constraint to relax it: if peer reboots, clock may go 1622 out-of-sync and half-open connections will not be reset. 1623 Actually, the problem would be not existing if all 1624 the implementations followed draft about maintaining clock 1625 via reboots. Linux-2.2 DOES NOT! 1626 1627 However, we can relax time bounds for RST segments to MSL. 1628 */ 1629 if (rst && !time_before32(ktime_get_seconds(), 1630 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1631 return false; 1632 return true; 1633 } 1634 1635 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1636 int mib_idx, u32 *last_oow_ack_time); 1637 1638 static inline void tcp_mib_init(struct net *net) 1639 { 1640 /* See RFC 2012 */ 1641 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1642 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1643 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1644 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1645 } 1646 1647 /* from STCP */ 1648 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1649 { 1650 tp->lost_skb_hint = NULL; 1651 } 1652 1653 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1654 { 1655 tcp_clear_retrans_hints_partial(tp); 1656 tp->retransmit_skb_hint = NULL; 1657 } 1658 1659 union tcp_md5_addr { 1660 struct in_addr a4; 1661 #if IS_ENABLED(CONFIG_IPV6) 1662 struct in6_addr a6; 1663 #endif 1664 }; 1665 1666 /* - key database */ 1667 struct tcp_md5sig_key { 1668 struct hlist_node node; 1669 u8 keylen; 1670 u8 family; /* AF_INET or AF_INET6 */ 1671 u8 prefixlen; 1672 u8 flags; 1673 union tcp_md5_addr addr; 1674 int l3index; /* set if key added with L3 scope */ 1675 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1676 struct rcu_head rcu; 1677 }; 1678 1679 /* - sock block */ 1680 struct tcp_md5sig_info { 1681 struct hlist_head head; 1682 struct rcu_head rcu; 1683 }; 1684 1685 /* - pseudo header */ 1686 struct tcp4_pseudohdr { 1687 __be32 saddr; 1688 __be32 daddr; 1689 __u8 pad; 1690 __u8 protocol; 1691 __be16 len; 1692 }; 1693 1694 struct tcp6_pseudohdr { 1695 struct in6_addr saddr; 1696 struct in6_addr daddr; 1697 __be32 len; 1698 __be32 protocol; /* including padding */ 1699 }; 1700 1701 union tcp_md5sum_block { 1702 struct tcp4_pseudohdr ip4; 1703 #if IS_ENABLED(CONFIG_IPV6) 1704 struct tcp6_pseudohdr ip6; 1705 #endif 1706 }; 1707 1708 /* - pool: digest algorithm, hash description and scratch buffer */ 1709 struct tcp_md5sig_pool { 1710 struct ahash_request *md5_req; 1711 void *scratch; 1712 }; 1713 1714 /* - functions */ 1715 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1716 const struct sock *sk, const struct sk_buff *skb); 1717 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1718 int family, u8 prefixlen, int l3index, u8 flags, 1719 const u8 *newkey, u8 newkeylen); 1720 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1721 int family, u8 prefixlen, int l3index, 1722 struct tcp_md5sig_key *key); 1723 1724 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1725 int family, u8 prefixlen, int l3index, u8 flags); 1726 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1727 const struct sock *addr_sk); 1728 1729 #ifdef CONFIG_TCP_MD5SIG 1730 #include <linux/jump_label.h> 1731 extern struct static_key_false_deferred tcp_md5_needed; 1732 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1733 const union tcp_md5_addr *addr, 1734 int family); 1735 static inline struct tcp_md5sig_key * 1736 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1737 const union tcp_md5_addr *addr, int family) 1738 { 1739 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1740 return NULL; 1741 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1742 } 1743 1744 enum skb_drop_reason 1745 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1746 const void *saddr, const void *daddr, 1747 int family, int dif, int sdif); 1748 1749 1750 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1751 #else 1752 static inline struct tcp_md5sig_key * 1753 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1754 const union tcp_md5_addr *addr, int family) 1755 { 1756 return NULL; 1757 } 1758 1759 static inline enum skb_drop_reason 1760 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1761 const void *saddr, const void *daddr, 1762 int family, int dif, int sdif) 1763 { 1764 return SKB_NOT_DROPPED_YET; 1765 } 1766 #define tcp_twsk_md5_key(twsk) NULL 1767 #endif 1768 1769 bool tcp_alloc_md5sig_pool(void); 1770 1771 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1772 static inline void tcp_put_md5sig_pool(void) 1773 { 1774 local_bh_enable(); 1775 } 1776 1777 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1778 unsigned int header_len); 1779 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1780 const struct tcp_md5sig_key *key); 1781 1782 /* From tcp_fastopen.c */ 1783 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1784 struct tcp_fastopen_cookie *cookie); 1785 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1786 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1787 u16 try_exp); 1788 struct tcp_fastopen_request { 1789 /* Fast Open cookie. Size 0 means a cookie request */ 1790 struct tcp_fastopen_cookie cookie; 1791 struct msghdr *data; /* data in MSG_FASTOPEN */ 1792 size_t size; 1793 int copied; /* queued in tcp_connect() */ 1794 struct ubuf_info *uarg; 1795 }; 1796 void tcp_free_fastopen_req(struct tcp_sock *tp); 1797 void tcp_fastopen_destroy_cipher(struct sock *sk); 1798 void tcp_fastopen_ctx_destroy(struct net *net); 1799 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1800 void *primary_key, void *backup_key); 1801 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1802 u64 *key); 1803 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1804 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1805 struct request_sock *req, 1806 struct tcp_fastopen_cookie *foc, 1807 const struct dst_entry *dst); 1808 void tcp_fastopen_init_key_once(struct net *net); 1809 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1810 struct tcp_fastopen_cookie *cookie); 1811 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1812 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1813 #define TCP_FASTOPEN_KEY_MAX 2 1814 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1815 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1816 1817 /* Fastopen key context */ 1818 struct tcp_fastopen_context { 1819 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1820 int num; 1821 struct rcu_head rcu; 1822 }; 1823 1824 void tcp_fastopen_active_disable(struct sock *sk); 1825 bool tcp_fastopen_active_should_disable(struct sock *sk); 1826 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1827 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1828 1829 /* Caller needs to wrap with rcu_read_(un)lock() */ 1830 static inline 1831 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1832 { 1833 struct tcp_fastopen_context *ctx; 1834 1835 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1836 if (!ctx) 1837 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1838 return ctx; 1839 } 1840 1841 static inline 1842 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1843 const struct tcp_fastopen_cookie *orig) 1844 { 1845 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1846 orig->len == foc->len && 1847 !memcmp(orig->val, foc->val, foc->len)) 1848 return true; 1849 return false; 1850 } 1851 1852 static inline 1853 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1854 { 1855 return ctx->num; 1856 } 1857 1858 /* Latencies incurred by various limits for a sender. They are 1859 * chronograph-like stats that are mutually exclusive. 1860 */ 1861 enum tcp_chrono { 1862 TCP_CHRONO_UNSPEC, 1863 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1864 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1865 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1866 __TCP_CHRONO_MAX, 1867 }; 1868 1869 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1870 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1871 1872 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1873 * the same memory storage than skb->destructor/_skb_refdst 1874 */ 1875 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1876 { 1877 skb->destructor = NULL; 1878 skb->_skb_refdst = 0UL; 1879 } 1880 1881 #define tcp_skb_tsorted_save(skb) { \ 1882 unsigned long _save = skb->_skb_refdst; \ 1883 skb->_skb_refdst = 0UL; 1884 1885 #define tcp_skb_tsorted_restore(skb) \ 1886 skb->_skb_refdst = _save; \ 1887 } 1888 1889 void tcp_write_queue_purge(struct sock *sk); 1890 1891 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1892 { 1893 return skb_rb_first(&sk->tcp_rtx_queue); 1894 } 1895 1896 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1897 { 1898 return skb_rb_last(&sk->tcp_rtx_queue); 1899 } 1900 1901 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1902 { 1903 return skb_peek_tail(&sk->sk_write_queue); 1904 } 1905 1906 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1907 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1908 1909 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1910 { 1911 return skb_peek(&sk->sk_write_queue); 1912 } 1913 1914 static inline bool tcp_skb_is_last(const struct sock *sk, 1915 const struct sk_buff *skb) 1916 { 1917 return skb_queue_is_last(&sk->sk_write_queue, skb); 1918 } 1919 1920 /** 1921 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1922 * @sk: socket 1923 * 1924 * Since the write queue can have a temporary empty skb in it, 1925 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1926 */ 1927 static inline bool tcp_write_queue_empty(const struct sock *sk) 1928 { 1929 const struct tcp_sock *tp = tcp_sk(sk); 1930 1931 return tp->write_seq == tp->snd_nxt; 1932 } 1933 1934 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1935 { 1936 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1937 } 1938 1939 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1940 { 1941 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1942 } 1943 1944 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1945 { 1946 __skb_queue_tail(&sk->sk_write_queue, skb); 1947 1948 /* Queue it, remembering where we must start sending. */ 1949 if (sk->sk_write_queue.next == skb) 1950 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1951 } 1952 1953 /* Insert new before skb on the write queue of sk. */ 1954 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1955 struct sk_buff *skb, 1956 struct sock *sk) 1957 { 1958 __skb_queue_before(&sk->sk_write_queue, skb, new); 1959 } 1960 1961 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1962 { 1963 tcp_skb_tsorted_anchor_cleanup(skb); 1964 __skb_unlink(skb, &sk->sk_write_queue); 1965 } 1966 1967 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1968 1969 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1970 { 1971 tcp_skb_tsorted_anchor_cleanup(skb); 1972 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1973 } 1974 1975 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1976 { 1977 list_del(&skb->tcp_tsorted_anchor); 1978 tcp_rtx_queue_unlink(skb, sk); 1979 tcp_wmem_free_skb(sk, skb); 1980 } 1981 1982 static inline void tcp_push_pending_frames(struct sock *sk) 1983 { 1984 if (tcp_send_head(sk)) { 1985 struct tcp_sock *tp = tcp_sk(sk); 1986 1987 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1988 } 1989 } 1990 1991 /* Start sequence of the skb just after the highest skb with SACKed 1992 * bit, valid only if sacked_out > 0 or when the caller has ensured 1993 * validity by itself. 1994 */ 1995 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1996 { 1997 if (!tp->sacked_out) 1998 return tp->snd_una; 1999 2000 if (tp->highest_sack == NULL) 2001 return tp->snd_nxt; 2002 2003 return TCP_SKB_CB(tp->highest_sack)->seq; 2004 } 2005 2006 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2007 { 2008 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2009 } 2010 2011 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2012 { 2013 return tcp_sk(sk)->highest_sack; 2014 } 2015 2016 static inline void tcp_highest_sack_reset(struct sock *sk) 2017 { 2018 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2019 } 2020 2021 /* Called when old skb is about to be deleted and replaced by new skb */ 2022 static inline void tcp_highest_sack_replace(struct sock *sk, 2023 struct sk_buff *old, 2024 struct sk_buff *new) 2025 { 2026 if (old == tcp_highest_sack(sk)) 2027 tcp_sk(sk)->highest_sack = new; 2028 } 2029 2030 /* This helper checks if socket has IP_TRANSPARENT set */ 2031 static inline bool inet_sk_transparent(const struct sock *sk) 2032 { 2033 switch (sk->sk_state) { 2034 case TCP_TIME_WAIT: 2035 return inet_twsk(sk)->tw_transparent; 2036 case TCP_NEW_SYN_RECV: 2037 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2038 } 2039 return inet_test_bit(TRANSPARENT, sk); 2040 } 2041 2042 /* Determines whether this is a thin stream (which may suffer from 2043 * increased latency). Used to trigger latency-reducing mechanisms. 2044 */ 2045 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2046 { 2047 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2048 } 2049 2050 /* /proc */ 2051 enum tcp_seq_states { 2052 TCP_SEQ_STATE_LISTENING, 2053 TCP_SEQ_STATE_ESTABLISHED, 2054 }; 2055 2056 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2057 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2058 void tcp_seq_stop(struct seq_file *seq, void *v); 2059 2060 struct tcp_seq_afinfo { 2061 sa_family_t family; 2062 }; 2063 2064 struct tcp_iter_state { 2065 struct seq_net_private p; 2066 enum tcp_seq_states state; 2067 struct sock *syn_wait_sk; 2068 int bucket, offset, sbucket, num; 2069 loff_t last_pos; 2070 }; 2071 2072 extern struct request_sock_ops tcp_request_sock_ops; 2073 extern struct request_sock_ops tcp6_request_sock_ops; 2074 2075 void tcp_v4_destroy_sock(struct sock *sk); 2076 2077 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2078 netdev_features_t features); 2079 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2080 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2081 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2082 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2083 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2084 void tcp_gro_complete(struct sk_buff *skb); 2085 2086 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2087 2088 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2089 { 2090 struct net *net = sock_net((struct sock *)tp); 2091 u32 val; 2092 2093 val = READ_ONCE(tp->notsent_lowat); 2094 2095 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2096 } 2097 2098 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2099 2100 #ifdef CONFIG_PROC_FS 2101 int tcp4_proc_init(void); 2102 void tcp4_proc_exit(void); 2103 #endif 2104 2105 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2106 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2107 const struct tcp_request_sock_ops *af_ops, 2108 struct sock *sk, struct sk_buff *skb); 2109 2110 /* TCP af-specific functions */ 2111 struct tcp_sock_af_ops { 2112 #ifdef CONFIG_TCP_MD5SIG 2113 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2114 const struct sock *addr_sk); 2115 int (*calc_md5_hash)(char *location, 2116 const struct tcp_md5sig_key *md5, 2117 const struct sock *sk, 2118 const struct sk_buff *skb); 2119 int (*md5_parse)(struct sock *sk, 2120 int optname, 2121 sockptr_t optval, 2122 int optlen); 2123 #endif 2124 }; 2125 2126 struct tcp_request_sock_ops { 2127 u16 mss_clamp; 2128 #ifdef CONFIG_TCP_MD5SIG 2129 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2130 const struct sock *addr_sk); 2131 int (*calc_md5_hash) (char *location, 2132 const struct tcp_md5sig_key *md5, 2133 const struct sock *sk, 2134 const struct sk_buff *skb); 2135 #endif 2136 #ifdef CONFIG_SYN_COOKIES 2137 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2138 __u16 *mss); 2139 #endif 2140 struct dst_entry *(*route_req)(const struct sock *sk, 2141 struct sk_buff *skb, 2142 struct flowi *fl, 2143 struct request_sock *req); 2144 u32 (*init_seq)(const struct sk_buff *skb); 2145 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2146 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2147 struct flowi *fl, struct request_sock *req, 2148 struct tcp_fastopen_cookie *foc, 2149 enum tcp_synack_type synack_type, 2150 struct sk_buff *syn_skb); 2151 }; 2152 2153 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2154 #if IS_ENABLED(CONFIG_IPV6) 2155 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2156 #endif 2157 2158 #ifdef CONFIG_SYN_COOKIES 2159 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2160 const struct sock *sk, struct sk_buff *skb, 2161 __u16 *mss) 2162 { 2163 tcp_synq_overflow(sk); 2164 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2165 return ops->cookie_init_seq(skb, mss); 2166 } 2167 #else 2168 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2169 const struct sock *sk, struct sk_buff *skb, 2170 __u16 *mss) 2171 { 2172 return 0; 2173 } 2174 #endif 2175 2176 int tcpv4_offload_init(void); 2177 2178 void tcp_v4_init(void); 2179 void tcp_init(void); 2180 2181 /* tcp_recovery.c */ 2182 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2183 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2184 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2185 u32 reo_wnd); 2186 extern bool tcp_rack_mark_lost(struct sock *sk); 2187 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2188 u64 xmit_time); 2189 extern void tcp_rack_reo_timeout(struct sock *sk); 2190 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2191 2192 /* tcp_plb.c */ 2193 2194 /* 2195 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2196 * expects cong_ratio which represents fraction of traffic that experienced 2197 * congestion over a single RTT. In order to avoid floating point operations, 2198 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2199 */ 2200 #define TCP_PLB_SCALE 8 2201 2202 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2203 struct tcp_plb_state { 2204 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2205 unused:3; 2206 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2207 }; 2208 2209 static inline void tcp_plb_init(const struct sock *sk, 2210 struct tcp_plb_state *plb) 2211 { 2212 plb->consec_cong_rounds = 0; 2213 plb->pause_until = 0; 2214 } 2215 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2216 const int cong_ratio); 2217 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2218 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2219 2220 /* At how many usecs into the future should the RTO fire? */ 2221 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2222 { 2223 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2224 u32 rto = inet_csk(sk)->icsk_rto; 2225 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2226 2227 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2228 } 2229 2230 /* 2231 * Save and compile IPv4 options, return a pointer to it 2232 */ 2233 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2234 struct sk_buff *skb) 2235 { 2236 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2237 struct ip_options_rcu *dopt = NULL; 2238 2239 if (opt->optlen) { 2240 int opt_size = sizeof(*dopt) + opt->optlen; 2241 2242 dopt = kmalloc(opt_size, GFP_ATOMIC); 2243 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2244 kfree(dopt); 2245 dopt = NULL; 2246 } 2247 } 2248 return dopt; 2249 } 2250 2251 /* locally generated TCP pure ACKs have skb->truesize == 2 2252 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2253 * This is much faster than dissecting the packet to find out. 2254 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2255 */ 2256 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2257 { 2258 return skb->truesize == 2; 2259 } 2260 2261 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2262 { 2263 skb->truesize = 2; 2264 } 2265 2266 static inline int tcp_inq(struct sock *sk) 2267 { 2268 struct tcp_sock *tp = tcp_sk(sk); 2269 int answ; 2270 2271 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2272 answ = 0; 2273 } else if (sock_flag(sk, SOCK_URGINLINE) || 2274 !tp->urg_data || 2275 before(tp->urg_seq, tp->copied_seq) || 2276 !before(tp->urg_seq, tp->rcv_nxt)) { 2277 2278 answ = tp->rcv_nxt - tp->copied_seq; 2279 2280 /* Subtract 1, if FIN was received */ 2281 if (answ && sock_flag(sk, SOCK_DONE)) 2282 answ--; 2283 } else { 2284 answ = tp->urg_seq - tp->copied_seq; 2285 } 2286 2287 return answ; 2288 } 2289 2290 int tcp_peek_len(struct socket *sock); 2291 2292 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2293 { 2294 u16 segs_in; 2295 2296 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2297 2298 /* We update these fields while other threads might 2299 * read them from tcp_get_info() 2300 */ 2301 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2302 if (skb->len > tcp_hdrlen(skb)) 2303 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2304 } 2305 2306 /* 2307 * TCP listen path runs lockless. 2308 * We forced "struct sock" to be const qualified to make sure 2309 * we don't modify one of its field by mistake. 2310 * Here, we increment sk_drops which is an atomic_t, so we can safely 2311 * make sock writable again. 2312 */ 2313 static inline void tcp_listendrop(const struct sock *sk) 2314 { 2315 atomic_inc(&((struct sock *)sk)->sk_drops); 2316 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2317 } 2318 2319 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2320 2321 /* 2322 * Interface for adding Upper Level Protocols over TCP 2323 */ 2324 2325 #define TCP_ULP_NAME_MAX 16 2326 #define TCP_ULP_MAX 128 2327 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2328 2329 struct tcp_ulp_ops { 2330 struct list_head list; 2331 2332 /* initialize ulp */ 2333 int (*init)(struct sock *sk); 2334 /* update ulp */ 2335 void (*update)(struct sock *sk, struct proto *p, 2336 void (*write_space)(struct sock *sk)); 2337 /* cleanup ulp */ 2338 void (*release)(struct sock *sk); 2339 /* diagnostic */ 2340 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2341 size_t (*get_info_size)(const struct sock *sk); 2342 /* clone ulp */ 2343 void (*clone)(const struct request_sock *req, struct sock *newsk, 2344 const gfp_t priority); 2345 2346 char name[TCP_ULP_NAME_MAX]; 2347 struct module *owner; 2348 }; 2349 int tcp_register_ulp(struct tcp_ulp_ops *type); 2350 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2351 int tcp_set_ulp(struct sock *sk, const char *name); 2352 void tcp_get_available_ulp(char *buf, size_t len); 2353 void tcp_cleanup_ulp(struct sock *sk); 2354 void tcp_update_ulp(struct sock *sk, struct proto *p, 2355 void (*write_space)(struct sock *sk)); 2356 2357 #define MODULE_ALIAS_TCP_ULP(name) \ 2358 __MODULE_INFO(alias, alias_userspace, name); \ 2359 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2360 2361 #ifdef CONFIG_NET_SOCK_MSG 2362 struct sk_msg; 2363 struct sk_psock; 2364 2365 #ifdef CONFIG_BPF_SYSCALL 2366 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2367 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2368 #endif /* CONFIG_BPF_SYSCALL */ 2369 2370 #ifdef CONFIG_INET 2371 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2372 #else 2373 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2374 { 2375 } 2376 #endif 2377 2378 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2379 struct sk_msg *msg, u32 bytes, int flags); 2380 #endif /* CONFIG_NET_SOCK_MSG */ 2381 2382 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2383 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2384 { 2385 } 2386 #endif 2387 2388 #ifdef CONFIG_CGROUP_BPF 2389 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2390 struct sk_buff *skb, 2391 unsigned int end_offset) 2392 { 2393 skops->skb = skb; 2394 skops->skb_data_end = skb->data + end_offset; 2395 } 2396 #else 2397 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2398 struct sk_buff *skb, 2399 unsigned int end_offset) 2400 { 2401 } 2402 #endif 2403 2404 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2405 * is < 0, then the BPF op failed (for example if the loaded BPF 2406 * program does not support the chosen operation or there is no BPF 2407 * program loaded). 2408 */ 2409 #ifdef CONFIG_BPF 2410 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2411 { 2412 struct bpf_sock_ops_kern sock_ops; 2413 int ret; 2414 2415 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2416 if (sk_fullsock(sk)) { 2417 sock_ops.is_fullsock = 1; 2418 sock_owned_by_me(sk); 2419 } 2420 2421 sock_ops.sk = sk; 2422 sock_ops.op = op; 2423 if (nargs > 0) 2424 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2425 2426 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2427 if (ret == 0) 2428 ret = sock_ops.reply; 2429 else 2430 ret = -1; 2431 return ret; 2432 } 2433 2434 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2435 { 2436 u32 args[2] = {arg1, arg2}; 2437 2438 return tcp_call_bpf(sk, op, 2, args); 2439 } 2440 2441 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2442 u32 arg3) 2443 { 2444 u32 args[3] = {arg1, arg2, arg3}; 2445 2446 return tcp_call_bpf(sk, op, 3, args); 2447 } 2448 2449 #else 2450 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2451 { 2452 return -EPERM; 2453 } 2454 2455 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2456 { 2457 return -EPERM; 2458 } 2459 2460 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2461 u32 arg3) 2462 { 2463 return -EPERM; 2464 } 2465 2466 #endif 2467 2468 static inline u32 tcp_timeout_init(struct sock *sk) 2469 { 2470 int timeout; 2471 2472 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2473 2474 if (timeout <= 0) 2475 timeout = TCP_TIMEOUT_INIT; 2476 return min_t(int, timeout, TCP_RTO_MAX); 2477 } 2478 2479 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2480 { 2481 int rwnd; 2482 2483 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2484 2485 if (rwnd < 0) 2486 rwnd = 0; 2487 return rwnd; 2488 } 2489 2490 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2491 { 2492 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2493 } 2494 2495 static inline void tcp_bpf_rtt(struct sock *sk) 2496 { 2497 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2498 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2499 } 2500 2501 #if IS_ENABLED(CONFIG_SMC) 2502 extern struct static_key_false tcp_have_smc; 2503 #endif 2504 2505 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2506 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2507 void (*cad)(struct sock *sk, u32 ack_seq)); 2508 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2509 void clean_acked_data_flush(void); 2510 #endif 2511 2512 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2513 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2514 const struct tcp_sock *tp) 2515 { 2516 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2517 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2518 } 2519 2520 /* Compute Earliest Departure Time for some control packets 2521 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2522 */ 2523 static inline u64 tcp_transmit_time(const struct sock *sk) 2524 { 2525 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2526 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2527 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2528 2529 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2530 } 2531 return 0; 2532 } 2533 2534 #endif /* _TCP_H */ 2535