1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP module. 8 * 9 * Version: @(#)tcp.h 1.0.5 05/23/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 */ 14 #ifndef _TCP_H 15 #define _TCP_H 16 17 #define FASTRETRANS_DEBUG 1 18 19 #include <linux/list.h> 20 #include <linux/tcp.h> 21 #include <linux/bug.h> 22 #include <linux/slab.h> 23 #include <linux/cache.h> 24 #include <linux/percpu.h> 25 #include <linux/skbuff.h> 26 #include <linux/kref.h> 27 #include <linux/ktime.h> 28 #include <linux/indirect_call_wrapper.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/inet_ecn.h> 41 #include <net/dst.h> 42 #include <net/mptcp.h> 43 44 #include <linux/seq_file.h> 45 #include <linux/memcontrol.h> 46 #include <linux/bpf-cgroup.h> 47 #include <linux/siphash.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 52 int tcp_orphan_count_sum(void); 53 54 void tcp_time_wait(struct sock *sk, int state, int timeo); 55 56 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 57 #define MAX_TCP_OPTION_SPACE 40 58 #define TCP_MIN_SND_MSS 48 59 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 60 61 /* 62 * Never offer a window over 32767 without using window scaling. Some 63 * poor stacks do signed 16bit maths! 64 */ 65 #define MAX_TCP_WINDOW 32767U 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The initial MTU to use for probing */ 71 #define TCP_BASE_MSS 1024 72 73 /* probing interval, default to 10 minutes as per RFC4821 */ 74 #define TCP_PROBE_INTERVAL 600 75 76 /* Specify interval when tcp mtu probing will stop */ 77 #define TCP_PROBE_THRESHOLD 8 78 79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 80 #define TCP_FASTRETRANS_THRESH 3 81 82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 83 #define TCP_MAX_QUICKACKS 16U 84 85 /* Maximal number of window scale according to RFC1323 */ 86 #define TCP_MAX_WSCALE 14U 87 88 /* urg_data states */ 89 #define TCP_URG_VALID 0x0100 90 #define TCP_URG_NOTYET 0x0200 91 #define TCP_URG_READ 0x0400 92 93 #define TCP_RETR1 3 /* 94 * This is how many retries it does before it 95 * tries to figure out if the gateway is 96 * down. Minimal RFC value is 3; it corresponds 97 * to ~3sec-8min depending on RTO. 98 */ 99 100 #define TCP_RETR2 15 /* 101 * This should take at least 102 * 90 minutes to time out. 103 * RFC1122 says that the limit is 100 sec. 104 * 15 is ~13-30min depending on RTO. 105 */ 106 107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 108 * when active opening a connection. 109 * RFC1122 says the minimum retry MUST 110 * be at least 180secs. Nevertheless 111 * this value is corresponding to 112 * 63secs of retransmission with the 113 * current initial RTO. 114 */ 115 116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 117 * when passive opening a connection. 118 * This is corresponding to 31secs of 119 * retransmission with the current 120 * initial RTO. 121 */ 122 123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 124 * state, about 60 seconds */ 125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 126 /* BSD style FIN_WAIT2 deadlock breaker. 127 * It used to be 3min, new value is 60sec, 128 * to combine FIN-WAIT-2 timeout with 129 * TIME-WAIT timer. 130 */ 131 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 132 133 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 134 #if HZ >= 100 135 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 136 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 137 #else 138 #define TCP_DELACK_MIN 4U 139 #define TCP_ATO_MIN 4U 140 #endif 141 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 142 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 143 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 144 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 145 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 146 * used as a fallback RTO for the 147 * initial data transmission if no 148 * valid RTT sample has been acquired, 149 * most likely due to retrans in 3WHS. 150 */ 151 152 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 153 * for local resources. 154 */ 155 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 156 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 157 #define TCP_KEEPALIVE_INTVL (75*HZ) 158 159 #define MAX_TCP_KEEPIDLE 32767 160 #define MAX_TCP_KEEPINTVL 32767 161 #define MAX_TCP_KEEPCNT 127 162 #define MAX_TCP_SYNCNT 127 163 164 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 165 166 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 167 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 168 * after this time. It should be equal 169 * (or greater than) TCP_TIMEWAIT_LEN 170 * to provide reliability equal to one 171 * provided by timewait state. 172 */ 173 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 174 * timestamps. It must be less than 175 * minimal timewait lifetime. 176 */ 177 /* 178 * TCP option 179 */ 180 181 #define TCPOPT_NOP 1 /* Padding */ 182 #define TCPOPT_EOL 0 /* End of options */ 183 #define TCPOPT_MSS 2 /* Segment size negotiating */ 184 #define TCPOPT_WINDOW 3 /* Window scaling */ 185 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 186 #define TCPOPT_SACK 5 /* SACK Block */ 187 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 188 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 189 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 190 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 191 #define TCPOPT_EXP 254 /* Experimental */ 192 /* Magic number to be after the option value for sharing TCP 193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 194 */ 195 #define TCPOPT_FASTOPEN_MAGIC 0xF989 196 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 197 198 /* 199 * TCP option lengths 200 */ 201 202 #define TCPOLEN_MSS 4 203 #define TCPOLEN_WINDOW 3 204 #define TCPOLEN_SACK_PERM 2 205 #define TCPOLEN_TIMESTAMP 10 206 #define TCPOLEN_MD5SIG 18 207 #define TCPOLEN_FASTOPEN_BASE 2 208 #define TCPOLEN_EXP_FASTOPEN_BASE 4 209 #define TCPOLEN_EXP_SMC_BASE 6 210 211 /* But this is what stacks really send out. */ 212 #define TCPOLEN_TSTAMP_ALIGNED 12 213 #define TCPOLEN_WSCALE_ALIGNED 4 214 #define TCPOLEN_SACKPERM_ALIGNED 4 215 #define TCPOLEN_SACK_BASE 2 216 #define TCPOLEN_SACK_BASE_ALIGNED 4 217 #define TCPOLEN_SACK_PERBLOCK 8 218 #define TCPOLEN_MD5SIG_ALIGNED 20 219 #define TCPOLEN_MSS_ALIGNED 4 220 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 221 222 /* Flags in tp->nonagle */ 223 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 224 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 225 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 226 227 /* TCP thin-stream limits */ 228 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 229 230 /* TCP initial congestion window as per rfc6928 */ 231 #define TCP_INIT_CWND 10 232 233 /* Bit Flags for sysctl_tcp_fastopen */ 234 #define TFO_CLIENT_ENABLE 1 235 #define TFO_SERVER_ENABLE 2 236 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 237 238 /* Accept SYN data w/o any cookie option */ 239 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 240 241 /* Force enable TFO on all listeners, i.e., not requiring the 242 * TCP_FASTOPEN socket option. 243 */ 244 #define TFO_SERVER_WO_SOCKOPT1 0x400 245 246 247 /* sysctl variables for tcp */ 248 extern int sysctl_tcp_max_orphans; 249 extern long sysctl_tcp_mem[3]; 250 251 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 252 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 253 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 254 255 extern atomic_long_t tcp_memory_allocated; 256 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 257 258 extern struct percpu_counter tcp_sockets_allocated; 259 extern unsigned long tcp_memory_pressure; 260 261 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 262 static inline bool tcp_under_memory_pressure(const struct sock *sk) 263 { 264 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 265 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 266 return true; 267 268 return READ_ONCE(tcp_memory_pressure); 269 } 270 /* 271 * The next routines deal with comparing 32 bit unsigned ints 272 * and worry about wraparound (automatic with unsigned arithmetic). 273 */ 274 275 static inline bool before(__u32 seq1, __u32 seq2) 276 { 277 return (__s32)(seq1-seq2) < 0; 278 } 279 #define after(seq2, seq1) before(seq1, seq2) 280 281 /* is s2<=s1<=s3 ? */ 282 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 283 { 284 return seq3 - seq2 >= seq1 - seq2; 285 } 286 287 static inline bool tcp_out_of_memory(struct sock *sk) 288 { 289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 291 return true; 292 return false; 293 } 294 295 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 296 { 297 sk_wmem_queued_add(sk, -skb->truesize); 298 if (!skb_zcopy_pure(skb)) 299 sk_mem_uncharge(sk, skb->truesize); 300 else 301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 302 __kfree_skb(skb); 303 } 304 305 void sk_forced_mem_schedule(struct sock *sk, int size); 306 307 bool tcp_check_oom(struct sock *sk, int shift); 308 309 310 extern struct proto tcp_prot; 311 312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 316 317 void tcp_tasklet_init(void); 318 319 int tcp_v4_err(struct sk_buff *skb, u32); 320 321 void tcp_shutdown(struct sock *sk, int how); 322 323 int tcp_v4_early_demux(struct sk_buff *skb); 324 int tcp_v4_rcv(struct sk_buff *skb); 325 326 void tcp_remove_empty_skb(struct sock *sk); 327 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 328 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 329 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 330 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 331 int flags); 332 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 333 size_t size, int flags); 334 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 335 size_t size, int flags); 336 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 337 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 338 int size_goal); 339 void tcp_release_cb(struct sock *sk); 340 void tcp_wfree(struct sk_buff *skb); 341 void tcp_write_timer_handler(struct sock *sk); 342 void tcp_delack_timer_handler(struct sock *sk); 343 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 344 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 345 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 346 void tcp_rcv_space_adjust(struct sock *sk); 347 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 348 void tcp_twsk_destructor(struct sock *sk); 349 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 350 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 351 struct pipe_inode_info *pipe, size_t len, 352 unsigned int flags); 353 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 354 bool force_schedule); 355 356 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 357 static inline void tcp_dec_quickack_mode(struct sock *sk, 358 const unsigned int pkts) 359 { 360 struct inet_connection_sock *icsk = inet_csk(sk); 361 362 if (icsk->icsk_ack.quick) { 363 if (pkts >= icsk->icsk_ack.quick) { 364 icsk->icsk_ack.quick = 0; 365 /* Leaving quickack mode we deflate ATO. */ 366 icsk->icsk_ack.ato = TCP_ATO_MIN; 367 } else 368 icsk->icsk_ack.quick -= pkts; 369 } 370 } 371 372 #define TCP_ECN_OK 1 373 #define TCP_ECN_QUEUE_CWR 2 374 #define TCP_ECN_DEMAND_CWR 4 375 #define TCP_ECN_SEEN 8 376 377 enum tcp_tw_status { 378 TCP_TW_SUCCESS = 0, 379 TCP_TW_RST = 1, 380 TCP_TW_ACK = 2, 381 TCP_TW_SYN = 3 382 }; 383 384 385 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 386 struct sk_buff *skb, 387 const struct tcphdr *th); 388 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 389 struct request_sock *req, bool fastopen, 390 bool *lost_race); 391 int tcp_child_process(struct sock *parent, struct sock *child, 392 struct sk_buff *skb); 393 void tcp_enter_loss(struct sock *sk); 394 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 395 void tcp_clear_retrans(struct tcp_sock *tp); 396 void tcp_update_metrics(struct sock *sk); 397 void tcp_init_metrics(struct sock *sk); 398 void tcp_metrics_init(void); 399 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 400 void __tcp_close(struct sock *sk, long timeout); 401 void tcp_close(struct sock *sk, long timeout); 402 void tcp_init_sock(struct sock *sk); 403 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 404 __poll_t tcp_poll(struct file *file, struct socket *sock, 405 struct poll_table_struct *wait); 406 int do_tcp_getsockopt(struct sock *sk, int level, 407 int optname, sockptr_t optval, sockptr_t optlen); 408 int tcp_getsockopt(struct sock *sk, int level, int optname, 409 char __user *optval, int __user *optlen); 410 bool tcp_bpf_bypass_getsockopt(int level, int optname); 411 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 412 sockptr_t optval, unsigned int optlen); 413 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 414 unsigned int optlen); 415 void tcp_set_keepalive(struct sock *sk, int val); 416 void tcp_syn_ack_timeout(const struct request_sock *req); 417 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 418 int flags, int *addr_len); 419 int tcp_set_rcvlowat(struct sock *sk, int val); 420 int tcp_set_window_clamp(struct sock *sk, int val); 421 void tcp_update_recv_tstamps(struct sk_buff *skb, 422 struct scm_timestamping_internal *tss); 423 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 424 struct scm_timestamping_internal *tss); 425 void tcp_data_ready(struct sock *sk); 426 #ifdef CONFIG_MMU 427 int tcp_mmap(struct file *file, struct socket *sock, 428 struct vm_area_struct *vma); 429 #endif 430 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 431 struct tcp_options_received *opt_rx, 432 int estab, struct tcp_fastopen_cookie *foc); 433 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 434 435 /* 436 * BPF SKB-less helpers 437 */ 438 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 439 struct tcphdr *th, u32 *cookie); 440 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 441 struct tcphdr *th, u32 *cookie); 442 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 443 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 444 const struct tcp_request_sock_ops *af_ops, 445 struct sock *sk, struct tcphdr *th); 446 /* 447 * TCP v4 functions exported for the inet6 API 448 */ 449 450 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 451 void tcp_v4_mtu_reduced(struct sock *sk); 452 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 453 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 454 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 455 struct sock *tcp_create_openreq_child(const struct sock *sk, 456 struct request_sock *req, 457 struct sk_buff *skb); 458 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 459 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 460 struct request_sock *req, 461 struct dst_entry *dst, 462 struct request_sock *req_unhash, 463 bool *own_req); 464 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 465 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 466 int tcp_connect(struct sock *sk); 467 enum tcp_synack_type { 468 TCP_SYNACK_NORMAL, 469 TCP_SYNACK_FASTOPEN, 470 TCP_SYNACK_COOKIE, 471 }; 472 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 473 struct request_sock *req, 474 struct tcp_fastopen_cookie *foc, 475 enum tcp_synack_type synack_type, 476 struct sk_buff *syn_skb); 477 int tcp_disconnect(struct sock *sk, int flags); 478 479 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 480 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 481 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 482 483 /* From syncookies.c */ 484 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 485 struct request_sock *req, 486 struct dst_entry *dst, u32 tsoff); 487 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 488 u32 cookie); 489 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 490 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 491 const struct tcp_request_sock_ops *af_ops, 492 struct sock *sk, struct sk_buff *skb); 493 #ifdef CONFIG_SYN_COOKIES 494 495 /* Syncookies use a monotonic timer which increments every 60 seconds. 496 * This counter is used both as a hash input and partially encoded into 497 * the cookie value. A cookie is only validated further if the delta 498 * between the current counter value and the encoded one is less than this, 499 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 500 * the counter advances immediately after a cookie is generated). 501 */ 502 #define MAX_SYNCOOKIE_AGE 2 503 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 504 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 505 506 /* syncookies: remember time of last synqueue overflow 507 * But do not dirty this field too often (once per second is enough) 508 * It is racy as we do not hold a lock, but race is very minor. 509 */ 510 static inline void tcp_synq_overflow(const struct sock *sk) 511 { 512 unsigned int last_overflow; 513 unsigned int now = jiffies; 514 515 if (sk->sk_reuseport) { 516 struct sock_reuseport *reuse; 517 518 reuse = rcu_dereference(sk->sk_reuseport_cb); 519 if (likely(reuse)) { 520 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 521 if (!time_between32(now, last_overflow, 522 last_overflow + HZ)) 523 WRITE_ONCE(reuse->synq_overflow_ts, now); 524 return; 525 } 526 } 527 528 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 529 if (!time_between32(now, last_overflow, last_overflow + HZ)) 530 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 531 } 532 533 /* syncookies: no recent synqueue overflow on this listening socket? */ 534 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 535 { 536 unsigned int last_overflow; 537 unsigned int now = jiffies; 538 539 if (sk->sk_reuseport) { 540 struct sock_reuseport *reuse; 541 542 reuse = rcu_dereference(sk->sk_reuseport_cb); 543 if (likely(reuse)) { 544 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 545 return !time_between32(now, last_overflow - HZ, 546 last_overflow + 547 TCP_SYNCOOKIE_VALID); 548 } 549 } 550 551 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 552 553 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 554 * then we're under synflood. However, we have to use 555 * 'last_overflow - HZ' as lower bound. That's because a concurrent 556 * tcp_synq_overflow() could update .ts_recent_stamp after we read 557 * jiffies but before we store .ts_recent_stamp into last_overflow, 558 * which could lead to rejecting a valid syncookie. 559 */ 560 return !time_between32(now, last_overflow - HZ, 561 last_overflow + TCP_SYNCOOKIE_VALID); 562 } 563 564 static inline u32 tcp_cookie_time(void) 565 { 566 u64 val = get_jiffies_64(); 567 568 do_div(val, TCP_SYNCOOKIE_PERIOD); 569 return val; 570 } 571 572 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 573 u16 *mssp); 574 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 575 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 576 bool cookie_timestamp_decode(const struct net *net, 577 struct tcp_options_received *opt); 578 bool cookie_ecn_ok(const struct tcp_options_received *opt, 579 const struct net *net, const struct dst_entry *dst); 580 581 /* From net/ipv6/syncookies.c */ 582 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 583 u32 cookie); 584 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 585 586 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 587 const struct tcphdr *th, u16 *mssp); 588 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 589 #endif 590 /* tcp_output.c */ 591 592 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 593 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 594 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 595 int nonagle); 596 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 597 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 598 void tcp_retransmit_timer(struct sock *sk); 599 void tcp_xmit_retransmit_queue(struct sock *); 600 void tcp_simple_retransmit(struct sock *); 601 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 602 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 603 enum tcp_queue { 604 TCP_FRAG_IN_WRITE_QUEUE, 605 TCP_FRAG_IN_RTX_QUEUE, 606 }; 607 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 608 struct sk_buff *skb, u32 len, 609 unsigned int mss_now, gfp_t gfp); 610 611 void tcp_send_probe0(struct sock *); 612 void tcp_send_partial(struct sock *); 613 int tcp_write_wakeup(struct sock *, int mib); 614 void tcp_send_fin(struct sock *sk); 615 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 616 int tcp_send_synack(struct sock *); 617 void tcp_push_one(struct sock *, unsigned int mss_now); 618 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 619 void tcp_send_ack(struct sock *sk); 620 void tcp_send_delayed_ack(struct sock *sk); 621 void tcp_send_loss_probe(struct sock *sk); 622 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 623 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 624 const struct sk_buff *next_skb); 625 626 /* tcp_input.c */ 627 void tcp_rearm_rto(struct sock *sk); 628 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 629 void tcp_reset(struct sock *sk, struct sk_buff *skb); 630 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 631 void tcp_fin(struct sock *sk); 632 void tcp_check_space(struct sock *sk); 633 634 /* tcp_timer.c */ 635 void tcp_init_xmit_timers(struct sock *); 636 static inline void tcp_clear_xmit_timers(struct sock *sk) 637 { 638 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 639 __sock_put(sk); 640 641 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 642 __sock_put(sk); 643 644 inet_csk_clear_xmit_timers(sk); 645 } 646 647 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 648 unsigned int tcp_current_mss(struct sock *sk); 649 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 650 651 /* Bound MSS / TSO packet size with the half of the window */ 652 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 653 { 654 int cutoff; 655 656 /* When peer uses tiny windows, there is no use in packetizing 657 * to sub-MSS pieces for the sake of SWS or making sure there 658 * are enough packets in the pipe for fast recovery. 659 * 660 * On the other hand, for extremely large MSS devices, handling 661 * smaller than MSS windows in this way does make sense. 662 */ 663 if (tp->max_window > TCP_MSS_DEFAULT) 664 cutoff = (tp->max_window >> 1); 665 else 666 cutoff = tp->max_window; 667 668 if (cutoff && pktsize > cutoff) 669 return max_t(int, cutoff, 68U - tp->tcp_header_len); 670 else 671 return pktsize; 672 } 673 674 /* tcp.c */ 675 void tcp_get_info(struct sock *, struct tcp_info *); 676 677 /* Read 'sendfile()'-style from a TCP socket */ 678 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 679 sk_read_actor_t recv_actor); 680 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 681 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 682 void tcp_read_done(struct sock *sk, size_t len); 683 684 void tcp_initialize_rcv_mss(struct sock *sk); 685 686 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 687 int tcp_mss_to_mtu(struct sock *sk, int mss); 688 void tcp_mtup_init(struct sock *sk); 689 690 static inline void tcp_bound_rto(const struct sock *sk) 691 { 692 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 693 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 694 } 695 696 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 697 { 698 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 699 } 700 701 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 702 { 703 /* mptcp hooks are only on the slow path */ 704 if (sk_is_mptcp((struct sock *)tp)) 705 return; 706 707 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 708 ntohl(TCP_FLAG_ACK) | 709 snd_wnd); 710 } 711 712 static inline void tcp_fast_path_on(struct tcp_sock *tp) 713 { 714 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 715 } 716 717 static inline void tcp_fast_path_check(struct sock *sk) 718 { 719 struct tcp_sock *tp = tcp_sk(sk); 720 721 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 722 tp->rcv_wnd && 723 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 724 !tp->urg_data) 725 tcp_fast_path_on(tp); 726 } 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 u32 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 1119 void tcp_assign_congestion_control(struct sock *sk); 1120 void tcp_init_congestion_control(struct sock *sk); 1121 void tcp_cleanup_congestion_control(struct sock *sk); 1122 int tcp_set_default_congestion_control(struct net *net, const char *name); 1123 void tcp_get_default_congestion_control(struct net *net, char *name); 1124 void tcp_get_available_congestion_control(char *buf, size_t len); 1125 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1126 int tcp_set_allowed_congestion_control(char *allowed); 1127 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1128 bool cap_net_admin); 1129 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1130 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1131 1132 u32 tcp_reno_ssthresh(struct sock *sk); 1133 u32 tcp_reno_undo_cwnd(struct sock *sk); 1134 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1135 extern struct tcp_congestion_ops tcp_reno; 1136 1137 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1138 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1139 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1140 #ifdef CONFIG_INET 1141 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1142 #else 1143 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1144 { 1145 return NULL; 1146 } 1147 #endif 1148 1149 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1150 { 1151 const struct inet_connection_sock *icsk = inet_csk(sk); 1152 1153 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1154 } 1155 1156 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1157 { 1158 const struct inet_connection_sock *icsk = inet_csk(sk); 1159 1160 if (icsk->icsk_ca_ops->cwnd_event) 1161 icsk->icsk_ca_ops->cwnd_event(sk, event); 1162 } 1163 1164 /* From tcp_cong.c */ 1165 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1166 1167 /* From tcp_rate.c */ 1168 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1169 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1170 struct rate_sample *rs); 1171 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1172 bool is_sack_reneg, struct rate_sample *rs); 1173 void tcp_rate_check_app_limited(struct sock *sk); 1174 1175 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1176 { 1177 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1178 } 1179 1180 /* These functions determine how the current flow behaves in respect of SACK 1181 * handling. SACK is negotiated with the peer, and therefore it can vary 1182 * between different flows. 1183 * 1184 * tcp_is_sack - SACK enabled 1185 * tcp_is_reno - No SACK 1186 */ 1187 static inline int tcp_is_sack(const struct tcp_sock *tp) 1188 { 1189 return likely(tp->rx_opt.sack_ok); 1190 } 1191 1192 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1193 { 1194 return !tcp_is_sack(tp); 1195 } 1196 1197 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1198 { 1199 return tp->sacked_out + tp->lost_out; 1200 } 1201 1202 /* This determines how many packets are "in the network" to the best 1203 * of our knowledge. In many cases it is conservative, but where 1204 * detailed information is available from the receiver (via SACK 1205 * blocks etc.) we can make more aggressive calculations. 1206 * 1207 * Use this for decisions involving congestion control, use just 1208 * tp->packets_out to determine if the send queue is empty or not. 1209 * 1210 * Read this equation as: 1211 * 1212 * "Packets sent once on transmission queue" MINUS 1213 * "Packets left network, but not honestly ACKed yet" PLUS 1214 * "Packets fast retransmitted" 1215 */ 1216 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1217 { 1218 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1219 } 1220 1221 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1222 1223 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1224 { 1225 return tp->snd_cwnd; 1226 } 1227 1228 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1229 { 1230 WARN_ON_ONCE((int)val <= 0); 1231 tp->snd_cwnd = val; 1232 } 1233 1234 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1235 { 1236 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1237 } 1238 1239 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1240 { 1241 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1242 } 1243 1244 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1245 { 1246 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1247 (1 << inet_csk(sk)->icsk_ca_state); 1248 } 1249 1250 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1251 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1252 * ssthresh. 1253 */ 1254 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1255 { 1256 const struct tcp_sock *tp = tcp_sk(sk); 1257 1258 if (tcp_in_cwnd_reduction(sk)) 1259 return tp->snd_ssthresh; 1260 else 1261 return max(tp->snd_ssthresh, 1262 ((tcp_snd_cwnd(tp) >> 1) + 1263 (tcp_snd_cwnd(tp) >> 2))); 1264 } 1265 1266 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1267 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1268 1269 void tcp_enter_cwr(struct sock *sk); 1270 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1271 1272 /* The maximum number of MSS of available cwnd for which TSO defers 1273 * sending if not using sysctl_tcp_tso_win_divisor. 1274 */ 1275 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1276 { 1277 return 3; 1278 } 1279 1280 /* Returns end sequence number of the receiver's advertised window */ 1281 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1282 { 1283 return tp->snd_una + tp->snd_wnd; 1284 } 1285 1286 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1287 * flexible approach. The RFC suggests cwnd should not be raised unless 1288 * it was fully used previously. And that's exactly what we do in 1289 * congestion avoidance mode. But in slow start we allow cwnd to grow 1290 * as long as the application has used half the cwnd. 1291 * Example : 1292 * cwnd is 10 (IW10), but application sends 9 frames. 1293 * We allow cwnd to reach 18 when all frames are ACKed. 1294 * This check is safe because it's as aggressive as slow start which already 1295 * risks 100% overshoot. The advantage is that we discourage application to 1296 * either send more filler packets or data to artificially blow up the cwnd 1297 * usage, and allow application-limited process to probe bw more aggressively. 1298 */ 1299 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1300 { 1301 const struct tcp_sock *tp = tcp_sk(sk); 1302 1303 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1304 if (tcp_in_slow_start(tp)) 1305 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1306 1307 return tp->is_cwnd_limited; 1308 } 1309 1310 /* BBR congestion control needs pacing. 1311 * Same remark for SO_MAX_PACING_RATE. 1312 * sch_fq packet scheduler is efficiently handling pacing, 1313 * but is not always installed/used. 1314 * Return true if TCP stack should pace packets itself. 1315 */ 1316 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1317 { 1318 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1319 } 1320 1321 /* Estimates in how many jiffies next packet for this flow can be sent. 1322 * Scheduling a retransmit timer too early would be silly. 1323 */ 1324 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1325 { 1326 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1327 1328 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1329 } 1330 1331 static inline void tcp_reset_xmit_timer(struct sock *sk, 1332 const int what, 1333 unsigned long when, 1334 const unsigned long max_when) 1335 { 1336 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1337 max_when); 1338 } 1339 1340 /* Something is really bad, we could not queue an additional packet, 1341 * because qdisc is full or receiver sent a 0 window, or we are paced. 1342 * We do not want to add fuel to the fire, or abort too early, 1343 * so make sure the timer we arm now is at least 200ms in the future, 1344 * regardless of current icsk_rto value (as it could be ~2ms) 1345 */ 1346 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1347 { 1348 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1349 } 1350 1351 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1352 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1353 unsigned long max_when) 1354 { 1355 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1356 inet_csk(sk)->icsk_backoff); 1357 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1358 1359 return (unsigned long)min_t(u64, when, max_when); 1360 } 1361 1362 static inline void tcp_check_probe_timer(struct sock *sk) 1363 { 1364 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1365 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1366 tcp_probe0_base(sk), TCP_RTO_MAX); 1367 } 1368 1369 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1370 { 1371 tp->snd_wl1 = seq; 1372 } 1373 1374 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1375 { 1376 tp->snd_wl1 = seq; 1377 } 1378 1379 /* 1380 * Calculate(/check) TCP checksum 1381 */ 1382 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1383 __be32 daddr, __wsum base) 1384 { 1385 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1386 } 1387 1388 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1389 { 1390 return !skb_csum_unnecessary(skb) && 1391 __skb_checksum_complete(skb); 1392 } 1393 1394 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1395 enum skb_drop_reason *reason); 1396 1397 1398 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1399 void tcp_set_state(struct sock *sk, int state); 1400 void tcp_done(struct sock *sk); 1401 int tcp_abort(struct sock *sk, int err); 1402 1403 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1404 { 1405 rx_opt->dsack = 0; 1406 rx_opt->num_sacks = 0; 1407 } 1408 1409 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1410 1411 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1412 { 1413 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1414 struct tcp_sock *tp = tcp_sk(sk); 1415 s32 delta; 1416 1417 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1418 tp->packets_out || ca_ops->cong_control) 1419 return; 1420 delta = tcp_jiffies32 - tp->lsndtime; 1421 if (delta > inet_csk(sk)->icsk_rto) 1422 tcp_cwnd_restart(sk, delta); 1423 } 1424 1425 /* Determine a window scaling and initial window to offer. */ 1426 void tcp_select_initial_window(const struct sock *sk, int __space, 1427 __u32 mss, __u32 *rcv_wnd, 1428 __u32 *window_clamp, int wscale_ok, 1429 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1430 1431 static inline int tcp_win_from_space(const struct sock *sk, int space) 1432 { 1433 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale); 1434 1435 return tcp_adv_win_scale <= 0 ? 1436 (space>>(-tcp_adv_win_scale)) : 1437 space - (space>>tcp_adv_win_scale); 1438 } 1439 1440 /* Note: caller must be prepared to deal with negative returns */ 1441 static inline int tcp_space(const struct sock *sk) 1442 { 1443 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1444 READ_ONCE(sk->sk_backlog.len) - 1445 atomic_read(&sk->sk_rmem_alloc)); 1446 } 1447 1448 static inline int tcp_full_space(const struct sock *sk) 1449 { 1450 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1451 } 1452 1453 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1454 { 1455 int unused_mem = sk_unused_reserved_mem(sk); 1456 struct tcp_sock *tp = tcp_sk(sk); 1457 1458 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1459 if (unused_mem) 1460 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1461 tcp_win_from_space(sk, unused_mem)); 1462 } 1463 1464 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1465 1466 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1467 * If 87.5 % (7/8) of the space has been consumed, we want to override 1468 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1469 * len/truesize ratio. 1470 */ 1471 static inline bool tcp_rmem_pressure(const struct sock *sk) 1472 { 1473 int rcvbuf, threshold; 1474 1475 if (tcp_under_memory_pressure(sk)) 1476 return true; 1477 1478 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1479 threshold = rcvbuf - (rcvbuf >> 3); 1480 1481 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1482 } 1483 1484 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1485 { 1486 const struct tcp_sock *tp = tcp_sk(sk); 1487 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1488 1489 if (avail <= 0) 1490 return false; 1491 1492 return (avail >= target) || tcp_rmem_pressure(sk) || 1493 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1494 } 1495 1496 extern void tcp_openreq_init_rwin(struct request_sock *req, 1497 const struct sock *sk_listener, 1498 const struct dst_entry *dst); 1499 1500 void tcp_enter_memory_pressure(struct sock *sk); 1501 void tcp_leave_memory_pressure(struct sock *sk); 1502 1503 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1504 { 1505 struct net *net = sock_net((struct sock *)tp); 1506 1507 return tp->keepalive_intvl ? : 1508 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1509 } 1510 1511 static inline int keepalive_time_when(const struct tcp_sock *tp) 1512 { 1513 struct net *net = sock_net((struct sock *)tp); 1514 1515 return tp->keepalive_time ? : 1516 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1517 } 1518 1519 static inline int keepalive_probes(const struct tcp_sock *tp) 1520 { 1521 struct net *net = sock_net((struct sock *)tp); 1522 1523 return tp->keepalive_probes ? : 1524 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1525 } 1526 1527 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1528 { 1529 const struct inet_connection_sock *icsk = &tp->inet_conn; 1530 1531 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1532 tcp_jiffies32 - tp->rcv_tstamp); 1533 } 1534 1535 static inline int tcp_fin_time(const struct sock *sk) 1536 { 1537 int fin_timeout = tcp_sk(sk)->linger2 ? : 1538 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1539 const int rto = inet_csk(sk)->icsk_rto; 1540 1541 if (fin_timeout < (rto << 2) - (rto >> 1)) 1542 fin_timeout = (rto << 2) - (rto >> 1); 1543 1544 return fin_timeout; 1545 } 1546 1547 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1548 int paws_win) 1549 { 1550 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1551 return true; 1552 if (unlikely(!time_before32(ktime_get_seconds(), 1553 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1554 return true; 1555 /* 1556 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1557 * then following tcp messages have valid values. Ignore 0 value, 1558 * or else 'negative' tsval might forbid us to accept their packets. 1559 */ 1560 if (!rx_opt->ts_recent) 1561 return true; 1562 return false; 1563 } 1564 1565 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1566 int rst) 1567 { 1568 if (tcp_paws_check(rx_opt, 0)) 1569 return false; 1570 1571 /* RST segments are not recommended to carry timestamp, 1572 and, if they do, it is recommended to ignore PAWS because 1573 "their cleanup function should take precedence over timestamps." 1574 Certainly, it is mistake. It is necessary to understand the reasons 1575 of this constraint to relax it: if peer reboots, clock may go 1576 out-of-sync and half-open connections will not be reset. 1577 Actually, the problem would be not existing if all 1578 the implementations followed draft about maintaining clock 1579 via reboots. Linux-2.2 DOES NOT! 1580 1581 However, we can relax time bounds for RST segments to MSL. 1582 */ 1583 if (rst && !time_before32(ktime_get_seconds(), 1584 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1585 return false; 1586 return true; 1587 } 1588 1589 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1590 int mib_idx, u32 *last_oow_ack_time); 1591 1592 static inline void tcp_mib_init(struct net *net) 1593 { 1594 /* See RFC 2012 */ 1595 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1596 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1597 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1598 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1599 } 1600 1601 /* from STCP */ 1602 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1603 { 1604 tp->lost_skb_hint = NULL; 1605 } 1606 1607 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1608 { 1609 tcp_clear_retrans_hints_partial(tp); 1610 tp->retransmit_skb_hint = NULL; 1611 } 1612 1613 union tcp_md5_addr { 1614 struct in_addr a4; 1615 #if IS_ENABLED(CONFIG_IPV6) 1616 struct in6_addr a6; 1617 #endif 1618 }; 1619 1620 /* - key database */ 1621 struct tcp_md5sig_key { 1622 struct hlist_node node; 1623 u8 keylen; 1624 u8 family; /* AF_INET or AF_INET6 */ 1625 u8 prefixlen; 1626 u8 flags; 1627 union tcp_md5_addr addr; 1628 int l3index; /* set if key added with L3 scope */ 1629 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1630 struct rcu_head rcu; 1631 }; 1632 1633 /* - sock block */ 1634 struct tcp_md5sig_info { 1635 struct hlist_head head; 1636 struct rcu_head rcu; 1637 }; 1638 1639 /* - pseudo header */ 1640 struct tcp4_pseudohdr { 1641 __be32 saddr; 1642 __be32 daddr; 1643 __u8 pad; 1644 __u8 protocol; 1645 __be16 len; 1646 }; 1647 1648 struct tcp6_pseudohdr { 1649 struct in6_addr saddr; 1650 struct in6_addr daddr; 1651 __be32 len; 1652 __be32 protocol; /* including padding */ 1653 }; 1654 1655 union tcp_md5sum_block { 1656 struct tcp4_pseudohdr ip4; 1657 #if IS_ENABLED(CONFIG_IPV6) 1658 struct tcp6_pseudohdr ip6; 1659 #endif 1660 }; 1661 1662 /* - pool: digest algorithm, hash description and scratch buffer */ 1663 struct tcp_md5sig_pool { 1664 struct ahash_request *md5_req; 1665 void *scratch; 1666 }; 1667 1668 /* - functions */ 1669 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1670 const struct sock *sk, const struct sk_buff *skb); 1671 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1672 int family, u8 prefixlen, int l3index, u8 flags, 1673 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1674 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1675 int family, u8 prefixlen, int l3index, u8 flags); 1676 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1677 const struct sock *addr_sk); 1678 1679 #ifdef CONFIG_TCP_MD5SIG 1680 #include <linux/jump_label.h> 1681 extern struct static_key_false tcp_md5_needed; 1682 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1683 const union tcp_md5_addr *addr, 1684 int family); 1685 static inline struct tcp_md5sig_key * 1686 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1687 const union tcp_md5_addr *addr, int family) 1688 { 1689 if (!static_branch_unlikely(&tcp_md5_needed)) 1690 return NULL; 1691 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1692 } 1693 1694 enum skb_drop_reason 1695 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1696 const void *saddr, const void *daddr, 1697 int family, int dif, int sdif); 1698 1699 1700 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1701 #else 1702 static inline struct tcp_md5sig_key * 1703 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1704 const union tcp_md5_addr *addr, int family) 1705 { 1706 return NULL; 1707 } 1708 1709 static inline enum skb_drop_reason 1710 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1711 const void *saddr, const void *daddr, 1712 int family, int dif, int sdif) 1713 { 1714 return SKB_NOT_DROPPED_YET; 1715 } 1716 #define tcp_twsk_md5_key(twsk) NULL 1717 #endif 1718 1719 bool tcp_alloc_md5sig_pool(void); 1720 1721 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1722 static inline void tcp_put_md5sig_pool(void) 1723 { 1724 local_bh_enable(); 1725 } 1726 1727 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1728 unsigned int header_len); 1729 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1730 const struct tcp_md5sig_key *key); 1731 1732 /* From tcp_fastopen.c */ 1733 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1734 struct tcp_fastopen_cookie *cookie); 1735 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1736 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1737 u16 try_exp); 1738 struct tcp_fastopen_request { 1739 /* Fast Open cookie. Size 0 means a cookie request */ 1740 struct tcp_fastopen_cookie cookie; 1741 struct msghdr *data; /* data in MSG_FASTOPEN */ 1742 size_t size; 1743 int copied; /* queued in tcp_connect() */ 1744 struct ubuf_info *uarg; 1745 }; 1746 void tcp_free_fastopen_req(struct tcp_sock *tp); 1747 void tcp_fastopen_destroy_cipher(struct sock *sk); 1748 void tcp_fastopen_ctx_destroy(struct net *net); 1749 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1750 void *primary_key, void *backup_key); 1751 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1752 u64 *key); 1753 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1754 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1755 struct request_sock *req, 1756 struct tcp_fastopen_cookie *foc, 1757 const struct dst_entry *dst); 1758 void tcp_fastopen_init_key_once(struct net *net); 1759 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1760 struct tcp_fastopen_cookie *cookie); 1761 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1762 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1763 #define TCP_FASTOPEN_KEY_MAX 2 1764 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1765 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1766 1767 /* Fastopen key context */ 1768 struct tcp_fastopen_context { 1769 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1770 int num; 1771 struct rcu_head rcu; 1772 }; 1773 1774 void tcp_fastopen_active_disable(struct sock *sk); 1775 bool tcp_fastopen_active_should_disable(struct sock *sk); 1776 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1777 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1778 1779 /* Caller needs to wrap with rcu_read_(un)lock() */ 1780 static inline 1781 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1782 { 1783 struct tcp_fastopen_context *ctx; 1784 1785 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1786 if (!ctx) 1787 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1788 return ctx; 1789 } 1790 1791 static inline 1792 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1793 const struct tcp_fastopen_cookie *orig) 1794 { 1795 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1796 orig->len == foc->len && 1797 !memcmp(orig->val, foc->val, foc->len)) 1798 return true; 1799 return false; 1800 } 1801 1802 static inline 1803 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1804 { 1805 return ctx->num; 1806 } 1807 1808 /* Latencies incurred by various limits for a sender. They are 1809 * chronograph-like stats that are mutually exclusive. 1810 */ 1811 enum tcp_chrono { 1812 TCP_CHRONO_UNSPEC, 1813 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1814 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1815 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1816 __TCP_CHRONO_MAX, 1817 }; 1818 1819 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1820 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1821 1822 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1823 * the same memory storage than skb->destructor/_skb_refdst 1824 */ 1825 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1826 { 1827 skb->destructor = NULL; 1828 skb->_skb_refdst = 0UL; 1829 } 1830 1831 #define tcp_skb_tsorted_save(skb) { \ 1832 unsigned long _save = skb->_skb_refdst; \ 1833 skb->_skb_refdst = 0UL; 1834 1835 #define tcp_skb_tsorted_restore(skb) \ 1836 skb->_skb_refdst = _save; \ 1837 } 1838 1839 void tcp_write_queue_purge(struct sock *sk); 1840 1841 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1842 { 1843 return skb_rb_first(&sk->tcp_rtx_queue); 1844 } 1845 1846 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1847 { 1848 return skb_rb_last(&sk->tcp_rtx_queue); 1849 } 1850 1851 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1852 { 1853 return skb_peek_tail(&sk->sk_write_queue); 1854 } 1855 1856 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1857 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1858 1859 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1860 { 1861 return skb_peek(&sk->sk_write_queue); 1862 } 1863 1864 static inline bool tcp_skb_is_last(const struct sock *sk, 1865 const struct sk_buff *skb) 1866 { 1867 return skb_queue_is_last(&sk->sk_write_queue, skb); 1868 } 1869 1870 /** 1871 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1872 * @sk: socket 1873 * 1874 * Since the write queue can have a temporary empty skb in it, 1875 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1876 */ 1877 static inline bool tcp_write_queue_empty(const struct sock *sk) 1878 { 1879 const struct tcp_sock *tp = tcp_sk(sk); 1880 1881 return tp->write_seq == tp->snd_nxt; 1882 } 1883 1884 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1885 { 1886 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1887 } 1888 1889 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1890 { 1891 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1892 } 1893 1894 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1895 { 1896 __skb_queue_tail(&sk->sk_write_queue, skb); 1897 1898 /* Queue it, remembering where we must start sending. */ 1899 if (sk->sk_write_queue.next == skb) 1900 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1901 } 1902 1903 /* Insert new before skb on the write queue of sk. */ 1904 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1905 struct sk_buff *skb, 1906 struct sock *sk) 1907 { 1908 __skb_queue_before(&sk->sk_write_queue, skb, new); 1909 } 1910 1911 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1912 { 1913 tcp_skb_tsorted_anchor_cleanup(skb); 1914 __skb_unlink(skb, &sk->sk_write_queue); 1915 } 1916 1917 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1918 1919 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1920 { 1921 tcp_skb_tsorted_anchor_cleanup(skb); 1922 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1923 } 1924 1925 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1926 { 1927 list_del(&skb->tcp_tsorted_anchor); 1928 tcp_rtx_queue_unlink(skb, sk); 1929 tcp_wmem_free_skb(sk, skb); 1930 } 1931 1932 static inline void tcp_push_pending_frames(struct sock *sk) 1933 { 1934 if (tcp_send_head(sk)) { 1935 struct tcp_sock *tp = tcp_sk(sk); 1936 1937 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1938 } 1939 } 1940 1941 /* Start sequence of the skb just after the highest skb with SACKed 1942 * bit, valid only if sacked_out > 0 or when the caller has ensured 1943 * validity by itself. 1944 */ 1945 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1946 { 1947 if (!tp->sacked_out) 1948 return tp->snd_una; 1949 1950 if (tp->highest_sack == NULL) 1951 return tp->snd_nxt; 1952 1953 return TCP_SKB_CB(tp->highest_sack)->seq; 1954 } 1955 1956 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1957 { 1958 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1959 } 1960 1961 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1962 { 1963 return tcp_sk(sk)->highest_sack; 1964 } 1965 1966 static inline void tcp_highest_sack_reset(struct sock *sk) 1967 { 1968 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1969 } 1970 1971 /* Called when old skb is about to be deleted and replaced by new skb */ 1972 static inline void tcp_highest_sack_replace(struct sock *sk, 1973 struct sk_buff *old, 1974 struct sk_buff *new) 1975 { 1976 if (old == tcp_highest_sack(sk)) 1977 tcp_sk(sk)->highest_sack = new; 1978 } 1979 1980 /* This helper checks if socket has IP_TRANSPARENT set */ 1981 static inline bool inet_sk_transparent(const struct sock *sk) 1982 { 1983 switch (sk->sk_state) { 1984 case TCP_TIME_WAIT: 1985 return inet_twsk(sk)->tw_transparent; 1986 case TCP_NEW_SYN_RECV: 1987 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1988 } 1989 return inet_sk(sk)->transparent; 1990 } 1991 1992 /* Determines whether this is a thin stream (which may suffer from 1993 * increased latency). Used to trigger latency-reducing mechanisms. 1994 */ 1995 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1996 { 1997 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1998 } 1999 2000 /* /proc */ 2001 enum tcp_seq_states { 2002 TCP_SEQ_STATE_LISTENING, 2003 TCP_SEQ_STATE_ESTABLISHED, 2004 }; 2005 2006 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2007 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2008 void tcp_seq_stop(struct seq_file *seq, void *v); 2009 2010 struct tcp_seq_afinfo { 2011 sa_family_t family; 2012 }; 2013 2014 struct tcp_iter_state { 2015 struct seq_net_private p; 2016 enum tcp_seq_states state; 2017 struct sock *syn_wait_sk; 2018 int bucket, offset, sbucket, num; 2019 loff_t last_pos; 2020 }; 2021 2022 extern struct request_sock_ops tcp_request_sock_ops; 2023 extern struct request_sock_ops tcp6_request_sock_ops; 2024 2025 void tcp_v4_destroy_sock(struct sock *sk); 2026 2027 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2028 netdev_features_t features); 2029 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2030 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2031 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2032 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2033 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2034 int tcp_gro_complete(struct sk_buff *skb); 2035 2036 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2037 2038 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2039 { 2040 struct net *net = sock_net((struct sock *)tp); 2041 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2042 } 2043 2044 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2045 2046 #ifdef CONFIG_PROC_FS 2047 int tcp4_proc_init(void); 2048 void tcp4_proc_exit(void); 2049 #endif 2050 2051 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2052 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2053 const struct tcp_request_sock_ops *af_ops, 2054 struct sock *sk, struct sk_buff *skb); 2055 2056 /* TCP af-specific functions */ 2057 struct tcp_sock_af_ops { 2058 #ifdef CONFIG_TCP_MD5SIG 2059 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2060 const struct sock *addr_sk); 2061 int (*calc_md5_hash)(char *location, 2062 const struct tcp_md5sig_key *md5, 2063 const struct sock *sk, 2064 const struct sk_buff *skb); 2065 int (*md5_parse)(struct sock *sk, 2066 int optname, 2067 sockptr_t optval, 2068 int optlen); 2069 #endif 2070 }; 2071 2072 struct tcp_request_sock_ops { 2073 u16 mss_clamp; 2074 #ifdef CONFIG_TCP_MD5SIG 2075 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2076 const struct sock *addr_sk); 2077 int (*calc_md5_hash) (char *location, 2078 const struct tcp_md5sig_key *md5, 2079 const struct sock *sk, 2080 const struct sk_buff *skb); 2081 #endif 2082 #ifdef CONFIG_SYN_COOKIES 2083 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2084 __u16 *mss); 2085 #endif 2086 struct dst_entry *(*route_req)(const struct sock *sk, 2087 struct sk_buff *skb, 2088 struct flowi *fl, 2089 struct request_sock *req); 2090 u32 (*init_seq)(const struct sk_buff *skb); 2091 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2092 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2093 struct flowi *fl, struct request_sock *req, 2094 struct tcp_fastopen_cookie *foc, 2095 enum tcp_synack_type synack_type, 2096 struct sk_buff *syn_skb); 2097 }; 2098 2099 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2100 #if IS_ENABLED(CONFIG_IPV6) 2101 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2102 #endif 2103 2104 #ifdef CONFIG_SYN_COOKIES 2105 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2106 const struct sock *sk, struct sk_buff *skb, 2107 __u16 *mss) 2108 { 2109 tcp_synq_overflow(sk); 2110 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2111 return ops->cookie_init_seq(skb, mss); 2112 } 2113 #else 2114 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2115 const struct sock *sk, struct sk_buff *skb, 2116 __u16 *mss) 2117 { 2118 return 0; 2119 } 2120 #endif 2121 2122 int tcpv4_offload_init(void); 2123 2124 void tcp_v4_init(void); 2125 void tcp_init(void); 2126 2127 /* tcp_recovery.c */ 2128 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2129 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2130 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2131 u32 reo_wnd); 2132 extern bool tcp_rack_mark_lost(struct sock *sk); 2133 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2134 u64 xmit_time); 2135 extern void tcp_rack_reo_timeout(struct sock *sk); 2136 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2137 2138 /* At how many usecs into the future should the RTO fire? */ 2139 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2140 { 2141 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2142 u32 rto = inet_csk(sk)->icsk_rto; 2143 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2144 2145 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2146 } 2147 2148 /* 2149 * Save and compile IPv4 options, return a pointer to it 2150 */ 2151 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2152 struct sk_buff *skb) 2153 { 2154 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2155 struct ip_options_rcu *dopt = NULL; 2156 2157 if (opt->optlen) { 2158 int opt_size = sizeof(*dopt) + opt->optlen; 2159 2160 dopt = kmalloc(opt_size, GFP_ATOMIC); 2161 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2162 kfree(dopt); 2163 dopt = NULL; 2164 } 2165 } 2166 return dopt; 2167 } 2168 2169 /* locally generated TCP pure ACKs have skb->truesize == 2 2170 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2171 * This is much faster than dissecting the packet to find out. 2172 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2173 */ 2174 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2175 { 2176 return skb->truesize == 2; 2177 } 2178 2179 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2180 { 2181 skb->truesize = 2; 2182 } 2183 2184 static inline int tcp_inq(struct sock *sk) 2185 { 2186 struct tcp_sock *tp = tcp_sk(sk); 2187 int answ; 2188 2189 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2190 answ = 0; 2191 } else if (sock_flag(sk, SOCK_URGINLINE) || 2192 !tp->urg_data || 2193 before(tp->urg_seq, tp->copied_seq) || 2194 !before(tp->urg_seq, tp->rcv_nxt)) { 2195 2196 answ = tp->rcv_nxt - tp->copied_seq; 2197 2198 /* Subtract 1, if FIN was received */ 2199 if (answ && sock_flag(sk, SOCK_DONE)) 2200 answ--; 2201 } else { 2202 answ = tp->urg_seq - tp->copied_seq; 2203 } 2204 2205 return answ; 2206 } 2207 2208 int tcp_peek_len(struct socket *sock); 2209 2210 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2211 { 2212 u16 segs_in; 2213 2214 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2215 2216 /* We update these fields while other threads might 2217 * read them from tcp_get_info() 2218 */ 2219 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2220 if (skb->len > tcp_hdrlen(skb)) 2221 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2222 } 2223 2224 /* 2225 * TCP listen path runs lockless. 2226 * We forced "struct sock" to be const qualified to make sure 2227 * we don't modify one of its field by mistake. 2228 * Here, we increment sk_drops which is an atomic_t, so we can safely 2229 * make sock writable again. 2230 */ 2231 static inline void tcp_listendrop(const struct sock *sk) 2232 { 2233 atomic_inc(&((struct sock *)sk)->sk_drops); 2234 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2235 } 2236 2237 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2238 2239 /* 2240 * Interface for adding Upper Level Protocols over TCP 2241 */ 2242 2243 #define TCP_ULP_NAME_MAX 16 2244 #define TCP_ULP_MAX 128 2245 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2246 2247 struct tcp_ulp_ops { 2248 struct list_head list; 2249 2250 /* initialize ulp */ 2251 int (*init)(struct sock *sk); 2252 /* update ulp */ 2253 void (*update)(struct sock *sk, struct proto *p, 2254 void (*write_space)(struct sock *sk)); 2255 /* cleanup ulp */ 2256 void (*release)(struct sock *sk); 2257 /* diagnostic */ 2258 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2259 size_t (*get_info_size)(const struct sock *sk); 2260 /* clone ulp */ 2261 void (*clone)(const struct request_sock *req, struct sock *newsk, 2262 const gfp_t priority); 2263 2264 char name[TCP_ULP_NAME_MAX]; 2265 struct module *owner; 2266 }; 2267 int tcp_register_ulp(struct tcp_ulp_ops *type); 2268 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2269 int tcp_set_ulp(struct sock *sk, const char *name); 2270 void tcp_get_available_ulp(char *buf, size_t len); 2271 void tcp_cleanup_ulp(struct sock *sk); 2272 void tcp_update_ulp(struct sock *sk, struct proto *p, 2273 void (*write_space)(struct sock *sk)); 2274 2275 #define MODULE_ALIAS_TCP_ULP(name) \ 2276 __MODULE_INFO(alias, alias_userspace, name); \ 2277 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2278 2279 #ifdef CONFIG_NET_SOCK_MSG 2280 struct sk_msg; 2281 struct sk_psock; 2282 2283 #ifdef CONFIG_BPF_SYSCALL 2284 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2285 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2286 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2287 #endif /* CONFIG_BPF_SYSCALL */ 2288 2289 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2290 int flags); 2291 #endif /* CONFIG_NET_SOCK_MSG */ 2292 2293 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2294 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2295 { 2296 } 2297 #endif 2298 2299 #ifdef CONFIG_CGROUP_BPF 2300 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2301 struct sk_buff *skb, 2302 unsigned int end_offset) 2303 { 2304 skops->skb = skb; 2305 skops->skb_data_end = skb->data + end_offset; 2306 } 2307 #else 2308 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2309 struct sk_buff *skb, 2310 unsigned int end_offset) 2311 { 2312 } 2313 #endif 2314 2315 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2316 * is < 0, then the BPF op failed (for example if the loaded BPF 2317 * program does not support the chosen operation or there is no BPF 2318 * program loaded). 2319 */ 2320 #ifdef CONFIG_BPF 2321 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2322 { 2323 struct bpf_sock_ops_kern sock_ops; 2324 int ret; 2325 2326 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2327 if (sk_fullsock(sk)) { 2328 sock_ops.is_fullsock = 1; 2329 sock_owned_by_me(sk); 2330 } 2331 2332 sock_ops.sk = sk; 2333 sock_ops.op = op; 2334 if (nargs > 0) 2335 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2336 2337 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2338 if (ret == 0) 2339 ret = sock_ops.reply; 2340 else 2341 ret = -1; 2342 return ret; 2343 } 2344 2345 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2346 { 2347 u32 args[2] = {arg1, arg2}; 2348 2349 return tcp_call_bpf(sk, op, 2, args); 2350 } 2351 2352 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2353 u32 arg3) 2354 { 2355 u32 args[3] = {arg1, arg2, arg3}; 2356 2357 return tcp_call_bpf(sk, op, 3, args); 2358 } 2359 2360 #else 2361 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2362 { 2363 return -EPERM; 2364 } 2365 2366 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2367 { 2368 return -EPERM; 2369 } 2370 2371 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2372 u32 arg3) 2373 { 2374 return -EPERM; 2375 } 2376 2377 #endif 2378 2379 static inline u32 tcp_timeout_init(struct sock *sk) 2380 { 2381 int timeout; 2382 2383 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2384 2385 if (timeout <= 0) 2386 timeout = TCP_TIMEOUT_INIT; 2387 return min_t(int, timeout, TCP_RTO_MAX); 2388 } 2389 2390 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2391 { 2392 int rwnd; 2393 2394 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2395 2396 if (rwnd < 0) 2397 rwnd = 0; 2398 return rwnd; 2399 } 2400 2401 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2402 { 2403 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2404 } 2405 2406 static inline void tcp_bpf_rtt(struct sock *sk) 2407 { 2408 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2409 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2410 } 2411 2412 #if IS_ENABLED(CONFIG_SMC) 2413 extern struct static_key_false tcp_have_smc; 2414 #endif 2415 2416 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2417 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2418 void (*cad)(struct sock *sk, u32 ack_seq)); 2419 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2420 void clean_acked_data_flush(void); 2421 #endif 2422 2423 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2424 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2425 const struct tcp_sock *tp) 2426 { 2427 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2428 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2429 } 2430 2431 /* Compute Earliest Departure Time for some control packets 2432 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2433 */ 2434 static inline u64 tcp_transmit_time(const struct sock *sk) 2435 { 2436 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2437 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2438 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2439 2440 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2441 } 2442 return 0; 2443 } 2444 2445 #endif /* _TCP_H */ 2446