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