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_PAWS_24DAYS (60 * 60 * 24 * 24) 165 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 166 * after this time. It should be equal 167 * (or greater than) TCP_TIMEWAIT_LEN 168 * to provide reliability equal to one 169 * provided by timewait state. 170 */ 171 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 172 * timestamps. It must be less than 173 * minimal timewait lifetime. 174 */ 175 /* 176 * TCP option 177 */ 178 179 #define TCPOPT_NOP 1 /* Padding */ 180 #define TCPOPT_EOL 0 /* End of options */ 181 #define TCPOPT_MSS 2 /* Segment size negotiating */ 182 #define TCPOPT_WINDOW 3 /* Window scaling */ 183 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 184 #define TCPOPT_SACK 5 /* SACK Block */ 185 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 186 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 187 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 189 #define TCPOPT_EXP 254 /* Experimental */ 190 /* Magic number to be after the option value for sharing TCP 191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 192 */ 193 #define TCPOPT_FASTOPEN_MAGIC 0xF989 194 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 195 196 /* 197 * TCP option lengths 198 */ 199 200 #define TCPOLEN_MSS 4 201 #define TCPOLEN_WINDOW 3 202 #define TCPOLEN_SACK_PERM 2 203 #define TCPOLEN_TIMESTAMP 10 204 #define TCPOLEN_MD5SIG 18 205 #define TCPOLEN_FASTOPEN_BASE 2 206 #define TCPOLEN_EXP_FASTOPEN_BASE 4 207 #define TCPOLEN_EXP_SMC_BASE 6 208 209 /* But this is what stacks really send out. */ 210 #define TCPOLEN_TSTAMP_ALIGNED 12 211 #define TCPOLEN_WSCALE_ALIGNED 4 212 #define TCPOLEN_SACKPERM_ALIGNED 4 213 #define TCPOLEN_SACK_BASE 2 214 #define TCPOLEN_SACK_BASE_ALIGNED 4 215 #define TCPOLEN_SACK_PERBLOCK 8 216 #define TCPOLEN_MD5SIG_ALIGNED 20 217 #define TCPOLEN_MSS_ALIGNED 4 218 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 219 220 /* Flags in tp->nonagle */ 221 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 222 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 223 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 224 225 /* TCP thin-stream limits */ 226 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 227 228 /* TCP initial congestion window as per rfc6928 */ 229 #define TCP_INIT_CWND 10 230 231 /* Bit Flags for sysctl_tcp_fastopen */ 232 #define TFO_CLIENT_ENABLE 1 233 #define TFO_SERVER_ENABLE 2 234 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 235 236 /* Accept SYN data w/o any cookie option */ 237 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 238 239 /* Force enable TFO on all listeners, i.e., not requiring the 240 * TCP_FASTOPEN socket option. 241 */ 242 #define TFO_SERVER_WO_SOCKOPT1 0x400 243 244 245 /* sysctl variables for tcp */ 246 extern int sysctl_tcp_max_orphans; 247 extern long sysctl_tcp_mem[3]; 248 249 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 250 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 251 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 252 253 extern atomic_long_t tcp_memory_allocated; 254 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 255 256 extern struct percpu_counter tcp_sockets_allocated; 257 extern unsigned long tcp_memory_pressure; 258 259 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 260 static inline bool tcp_under_memory_pressure(const struct sock *sk) 261 { 262 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 263 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 264 return true; 265 266 return READ_ONCE(tcp_memory_pressure); 267 } 268 /* 269 * The next routines deal with comparing 32 bit unsigned ints 270 * and worry about wraparound (automatic with unsigned arithmetic). 271 */ 272 273 static inline bool before(__u32 seq1, __u32 seq2) 274 { 275 return (__s32)(seq1-seq2) < 0; 276 } 277 #define after(seq2, seq1) before(seq1, seq2) 278 279 /* is s2<=s1<=s3 ? */ 280 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 281 { 282 return seq3 - seq2 >= seq1 - seq2; 283 } 284 285 static inline bool tcp_out_of_memory(struct sock *sk) 286 { 287 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 288 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 289 return true; 290 return false; 291 } 292 293 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 294 { 295 sk_wmem_queued_add(sk, -skb->truesize); 296 if (!skb_zcopy_pure(skb)) 297 sk_mem_uncharge(sk, skb->truesize); 298 else 299 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 300 __kfree_skb(skb); 301 } 302 303 void sk_forced_mem_schedule(struct sock *sk, int size); 304 305 bool tcp_check_oom(struct sock *sk, int shift); 306 307 308 extern struct proto tcp_prot; 309 310 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 311 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 312 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 313 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 314 315 void tcp_tasklet_init(void); 316 317 int tcp_v4_err(struct sk_buff *skb, u32); 318 319 void tcp_shutdown(struct sock *sk, int how); 320 321 int tcp_v4_early_demux(struct sk_buff *skb); 322 int tcp_v4_rcv(struct sk_buff *skb); 323 324 void tcp_remove_empty_skb(struct sock *sk); 325 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 326 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 327 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 328 size_t size, struct ubuf_info *uarg); 329 void tcp_splice_eof(struct socket *sock); 330 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 331 int tcp_wmem_schedule(struct sock *sk, int copy); 332 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 333 int size_goal); 334 void tcp_release_cb(struct sock *sk); 335 void tcp_wfree(struct sk_buff *skb); 336 void tcp_write_timer_handler(struct sock *sk); 337 void tcp_delack_timer_handler(struct sock *sk); 338 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 339 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 340 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 341 void tcp_rcv_space_adjust(struct sock *sk); 342 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 343 void tcp_twsk_destructor(struct sock *sk); 344 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 345 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 346 struct pipe_inode_info *pipe, size_t len, 347 unsigned int flags); 348 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 349 bool force_schedule); 350 351 static inline void tcp_dec_quickack_mode(struct sock *sk) 352 { 353 struct inet_connection_sock *icsk = inet_csk(sk); 354 355 if (icsk->icsk_ack.quick) { 356 /* How many ACKs S/ACKing new data have we sent? */ 357 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 358 359 if (pkts >= icsk->icsk_ack.quick) { 360 icsk->icsk_ack.quick = 0; 361 /* Leaving quickack mode we deflate ATO. */ 362 icsk->icsk_ack.ato = TCP_ATO_MIN; 363 } else 364 icsk->icsk_ack.quick -= pkts; 365 } 366 } 367 368 #define TCP_ECN_OK 1 369 #define TCP_ECN_QUEUE_CWR 2 370 #define TCP_ECN_DEMAND_CWR 4 371 #define TCP_ECN_SEEN 8 372 373 enum tcp_tw_status { 374 TCP_TW_SUCCESS = 0, 375 TCP_TW_RST = 1, 376 TCP_TW_ACK = 2, 377 TCP_TW_SYN = 3 378 }; 379 380 381 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 382 struct sk_buff *skb, 383 const struct tcphdr *th); 384 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 385 struct request_sock *req, bool fastopen, 386 bool *lost_race); 387 int tcp_child_process(struct sock *parent, struct sock *child, 388 struct sk_buff *skb); 389 void tcp_enter_loss(struct sock *sk); 390 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 391 void tcp_clear_retrans(struct tcp_sock *tp); 392 void tcp_update_metrics(struct sock *sk); 393 void tcp_init_metrics(struct sock *sk); 394 void tcp_metrics_init(void); 395 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 396 void __tcp_close(struct sock *sk, long timeout); 397 void tcp_close(struct sock *sk, long timeout); 398 void tcp_init_sock(struct sock *sk); 399 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 400 __poll_t tcp_poll(struct file *file, struct socket *sock, 401 struct poll_table_struct *wait); 402 int do_tcp_getsockopt(struct sock *sk, int level, 403 int optname, sockptr_t optval, sockptr_t optlen); 404 int tcp_getsockopt(struct sock *sk, int level, int optname, 405 char __user *optval, int __user *optlen); 406 bool tcp_bpf_bypass_getsockopt(int level, int optname); 407 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 408 sockptr_t optval, unsigned int optlen); 409 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 410 unsigned int optlen); 411 void tcp_set_keepalive(struct sock *sk, int val); 412 void tcp_syn_ack_timeout(const struct request_sock *req); 413 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 414 int flags, int *addr_len); 415 int tcp_set_rcvlowat(struct sock *sk, int val); 416 int tcp_set_window_clamp(struct sock *sk, int val); 417 void tcp_update_recv_tstamps(struct sk_buff *skb, 418 struct scm_timestamping_internal *tss); 419 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 420 struct scm_timestamping_internal *tss); 421 void tcp_data_ready(struct sock *sk); 422 #ifdef CONFIG_MMU 423 int tcp_mmap(struct file *file, struct socket *sock, 424 struct vm_area_struct *vma); 425 #endif 426 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 427 struct tcp_options_received *opt_rx, 428 int estab, struct tcp_fastopen_cookie *foc); 429 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 430 431 /* 432 * BPF SKB-less helpers 433 */ 434 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 435 struct tcphdr *th, u32 *cookie); 436 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 437 struct tcphdr *th, u32 *cookie); 438 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 439 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 440 const struct tcp_request_sock_ops *af_ops, 441 struct sock *sk, struct tcphdr *th); 442 /* 443 * TCP v4 functions exported for the inet6 API 444 */ 445 446 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 447 void tcp_v4_mtu_reduced(struct sock *sk); 448 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 449 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 450 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 451 struct sock *tcp_create_openreq_child(const struct sock *sk, 452 struct request_sock *req, 453 struct sk_buff *skb); 454 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 455 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 456 struct request_sock *req, 457 struct dst_entry *dst, 458 struct request_sock *req_unhash, 459 bool *own_req); 460 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 461 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 462 int tcp_connect(struct sock *sk); 463 enum tcp_synack_type { 464 TCP_SYNACK_NORMAL, 465 TCP_SYNACK_FASTOPEN, 466 TCP_SYNACK_COOKIE, 467 }; 468 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 469 struct request_sock *req, 470 struct tcp_fastopen_cookie *foc, 471 enum tcp_synack_type synack_type, 472 struct sk_buff *syn_skb); 473 int tcp_disconnect(struct sock *sk, int flags); 474 475 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 476 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 477 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 478 479 /* From syncookies.c */ 480 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 481 struct request_sock *req, 482 struct dst_entry *dst, u32 tsoff); 483 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 484 u32 cookie); 485 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 486 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 487 const struct tcp_request_sock_ops *af_ops, 488 struct sock *sk, struct sk_buff *skb); 489 #ifdef CONFIG_SYN_COOKIES 490 491 /* Syncookies use a monotonic timer which increments every 60 seconds. 492 * This counter is used both as a hash input and partially encoded into 493 * the cookie value. A cookie is only validated further if the delta 494 * between the current counter value and the encoded one is less than this, 495 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 496 * the counter advances immediately after a cookie is generated). 497 */ 498 #define MAX_SYNCOOKIE_AGE 2 499 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 500 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 501 502 /* syncookies: remember time of last synqueue overflow 503 * But do not dirty this field too often (once per second is enough) 504 * It is racy as we do not hold a lock, but race is very minor. 505 */ 506 static inline void tcp_synq_overflow(const struct sock *sk) 507 { 508 unsigned int last_overflow; 509 unsigned int now = jiffies; 510 511 if (sk->sk_reuseport) { 512 struct sock_reuseport *reuse; 513 514 reuse = rcu_dereference(sk->sk_reuseport_cb); 515 if (likely(reuse)) { 516 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 517 if (!time_between32(now, last_overflow, 518 last_overflow + HZ)) 519 WRITE_ONCE(reuse->synq_overflow_ts, now); 520 return; 521 } 522 } 523 524 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 525 if (!time_between32(now, last_overflow, last_overflow + HZ)) 526 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 527 } 528 529 /* syncookies: no recent synqueue overflow on this listening socket? */ 530 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 531 { 532 unsigned int last_overflow; 533 unsigned int now = jiffies; 534 535 if (sk->sk_reuseport) { 536 struct sock_reuseport *reuse; 537 538 reuse = rcu_dereference(sk->sk_reuseport_cb); 539 if (likely(reuse)) { 540 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 541 return !time_between32(now, last_overflow - HZ, 542 last_overflow + 543 TCP_SYNCOOKIE_VALID); 544 } 545 } 546 547 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 548 549 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 550 * then we're under synflood. However, we have to use 551 * 'last_overflow - HZ' as lower bound. That's because a concurrent 552 * tcp_synq_overflow() could update .ts_recent_stamp after we read 553 * jiffies but before we store .ts_recent_stamp into last_overflow, 554 * which could lead to rejecting a valid syncookie. 555 */ 556 return !time_between32(now, last_overflow - HZ, 557 last_overflow + TCP_SYNCOOKIE_VALID); 558 } 559 560 static inline u32 tcp_cookie_time(void) 561 { 562 u64 val = get_jiffies_64(); 563 564 do_div(val, TCP_SYNCOOKIE_PERIOD); 565 return val; 566 } 567 568 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 569 u16 *mssp); 570 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 571 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 572 bool cookie_timestamp_decode(const struct net *net, 573 struct tcp_options_received *opt); 574 bool cookie_ecn_ok(const struct tcp_options_received *opt, 575 const struct net *net, const struct dst_entry *dst); 576 577 /* From net/ipv6/syncookies.c */ 578 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 579 u32 cookie); 580 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 581 582 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 583 const struct tcphdr *th, u16 *mssp); 584 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 585 #endif 586 /* tcp_output.c */ 587 588 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 589 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 590 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 591 int nonagle); 592 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 593 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 594 void tcp_retransmit_timer(struct sock *sk); 595 void tcp_xmit_retransmit_queue(struct sock *); 596 void tcp_simple_retransmit(struct sock *); 597 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 598 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 599 enum tcp_queue { 600 TCP_FRAG_IN_WRITE_QUEUE, 601 TCP_FRAG_IN_RTX_QUEUE, 602 }; 603 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 604 struct sk_buff *skb, u32 len, 605 unsigned int mss_now, gfp_t gfp); 606 607 void tcp_send_probe0(struct sock *); 608 int tcp_write_wakeup(struct sock *, int mib); 609 void tcp_send_fin(struct sock *sk); 610 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 611 int tcp_send_synack(struct sock *); 612 void tcp_push_one(struct sock *, unsigned int mss_now); 613 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 614 void tcp_send_ack(struct sock *sk); 615 void tcp_send_delayed_ack(struct sock *sk); 616 void tcp_send_loss_probe(struct sock *sk); 617 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 618 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 619 const struct sk_buff *next_skb); 620 621 /* tcp_input.c */ 622 void tcp_rearm_rto(struct sock *sk); 623 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 624 void tcp_reset(struct sock *sk, struct sk_buff *skb); 625 void tcp_fin(struct sock *sk); 626 void tcp_check_space(struct sock *sk); 627 void tcp_sack_compress_send_ack(struct sock *sk); 628 629 /* tcp_timer.c */ 630 void tcp_init_xmit_timers(struct sock *); 631 static inline void tcp_clear_xmit_timers(struct sock *sk) 632 { 633 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 634 __sock_put(sk); 635 636 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 637 __sock_put(sk); 638 639 inet_csk_clear_xmit_timers(sk); 640 } 641 642 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 643 unsigned int tcp_current_mss(struct sock *sk); 644 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 645 646 /* Bound MSS / TSO packet size with the half of the window */ 647 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 648 { 649 int cutoff; 650 651 /* When peer uses tiny windows, there is no use in packetizing 652 * to sub-MSS pieces for the sake of SWS or making sure there 653 * are enough packets in the pipe for fast recovery. 654 * 655 * On the other hand, for extremely large MSS devices, handling 656 * smaller than MSS windows in this way does make sense. 657 */ 658 if (tp->max_window > TCP_MSS_DEFAULT) 659 cutoff = (tp->max_window >> 1); 660 else 661 cutoff = tp->max_window; 662 663 if (cutoff && pktsize > cutoff) 664 return max_t(int, cutoff, 68U - tp->tcp_header_len); 665 else 666 return pktsize; 667 } 668 669 /* tcp.c */ 670 void tcp_get_info(struct sock *, struct tcp_info *); 671 672 /* Read 'sendfile()'-style from a TCP socket */ 673 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 674 sk_read_actor_t recv_actor); 675 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 676 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 677 void tcp_read_done(struct sock *sk, size_t len); 678 679 void tcp_initialize_rcv_mss(struct sock *sk); 680 681 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 682 int tcp_mss_to_mtu(struct sock *sk, int mss); 683 void tcp_mtup_init(struct sock *sk); 684 685 static inline void tcp_bound_rto(const struct sock *sk) 686 { 687 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 688 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 689 } 690 691 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 692 { 693 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 694 } 695 696 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 697 { 698 /* mptcp hooks are only on the slow path */ 699 if (sk_is_mptcp((struct sock *)tp)) 700 return; 701 702 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 703 ntohl(TCP_FLAG_ACK) | 704 snd_wnd); 705 } 706 707 static inline void tcp_fast_path_on(struct tcp_sock *tp) 708 { 709 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 710 } 711 712 static inline void tcp_fast_path_check(struct sock *sk) 713 { 714 struct tcp_sock *tp = tcp_sk(sk); 715 716 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 717 tp->rcv_wnd && 718 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 719 !tp->urg_data) 720 tcp_fast_path_on(tp); 721 } 722 723 /* Compute the actual rto_min value */ 724 static inline u32 tcp_rto_min(struct sock *sk) 725 { 726 const struct dst_entry *dst = __sk_dst_get(sk); 727 u32 rto_min = inet_csk(sk)->icsk_rto_min; 728 729 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 730 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 731 return rto_min; 732 } 733 734 static inline u32 tcp_rto_min_us(struct sock *sk) 735 { 736 return jiffies_to_usecs(tcp_rto_min(sk)); 737 } 738 739 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 740 { 741 return dst_metric_locked(dst, RTAX_CC_ALGO); 742 } 743 744 /* Minimum RTT in usec. ~0 means not available. */ 745 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 746 { 747 return minmax_get(&tp->rtt_min); 748 } 749 750 /* Compute the actual receive window we are currently advertising. 751 * Rcv_nxt can be after the window if our peer push more data 752 * than the offered window. 753 */ 754 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 755 { 756 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 757 758 if (win < 0) 759 win = 0; 760 return (u32) win; 761 } 762 763 /* Choose a new window, without checks for shrinking, and without 764 * scaling applied to the result. The caller does these things 765 * if necessary. This is a "raw" window selection. 766 */ 767 u32 __tcp_select_window(struct sock *sk); 768 769 void tcp_send_window_probe(struct sock *sk); 770 771 /* TCP uses 32bit jiffies to save some space. 772 * Note that this is different from tcp_time_stamp, which 773 * historically has been the same until linux-4.13. 774 */ 775 #define tcp_jiffies32 ((u32)jiffies) 776 777 /* 778 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 779 * It is no longer tied to jiffies, but to 1 ms clock. 780 * Note: double check if you want to use tcp_jiffies32 instead of this. 781 */ 782 #define TCP_TS_HZ 1000 783 784 static inline u64 tcp_clock_ns(void) 785 { 786 return ktime_get_ns(); 787 } 788 789 static inline u64 tcp_clock_us(void) 790 { 791 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 792 } 793 794 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 795 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 796 { 797 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 798 } 799 800 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 801 static inline u32 tcp_ns_to_ts(u64 ns) 802 { 803 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 804 } 805 806 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 807 static inline u32 tcp_time_stamp_raw(void) 808 { 809 return tcp_ns_to_ts(tcp_clock_ns()); 810 } 811 812 void tcp_mstamp_refresh(struct tcp_sock *tp); 813 814 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 815 { 816 return max_t(s64, t1 - t0, 0); 817 } 818 819 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 820 { 821 return tcp_ns_to_ts(skb->skb_mstamp_ns); 822 } 823 824 /* provide the departure time in us unit */ 825 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 826 { 827 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 828 } 829 830 831 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 832 833 #define TCPHDR_FIN 0x01 834 #define TCPHDR_SYN 0x02 835 #define TCPHDR_RST 0x04 836 #define TCPHDR_PSH 0x08 837 #define TCPHDR_ACK 0x10 838 #define TCPHDR_URG 0x20 839 #define TCPHDR_ECE 0x40 840 #define TCPHDR_CWR 0x80 841 842 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 843 844 /* This is what the send packet queuing engine uses to pass 845 * TCP per-packet control information to the transmission code. 846 * We also store the host-order sequence numbers in here too. 847 * This is 44 bytes if IPV6 is enabled. 848 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 849 */ 850 struct tcp_skb_cb { 851 __u32 seq; /* Starting sequence number */ 852 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 853 union { 854 /* Note : tcp_tw_isn is used in input path only 855 * (isn chosen by tcp_timewait_state_process()) 856 * 857 * tcp_gso_segs/size are used in write queue only, 858 * cf tcp_skb_pcount()/tcp_skb_mss() 859 */ 860 __u32 tcp_tw_isn; 861 struct { 862 u16 tcp_gso_segs; 863 u16 tcp_gso_size; 864 }; 865 }; 866 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 867 868 __u8 sacked; /* State flags for SACK. */ 869 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 870 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 871 #define TCPCB_LOST 0x04 /* SKB is lost */ 872 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 873 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 874 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 875 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 876 TCPCB_REPAIRED) 877 878 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 879 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 880 eor:1, /* Is skb MSG_EOR marked? */ 881 has_rxtstamp:1, /* SKB has a RX timestamp */ 882 unused:5; 883 __u32 ack_seq; /* Sequence number ACK'd */ 884 union { 885 struct { 886 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 887 /* There is space for up to 24 bytes */ 888 __u32 is_app_limited:1, /* cwnd not fully used? */ 889 delivered_ce:20, 890 unused:11; 891 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 892 __u32 delivered; 893 /* start of send pipeline phase */ 894 u64 first_tx_mstamp; 895 /* when we reached the "delivered" count */ 896 u64 delivered_mstamp; 897 } tx; /* only used for outgoing skbs */ 898 union { 899 struct inet_skb_parm h4; 900 #if IS_ENABLED(CONFIG_IPV6) 901 struct inet6_skb_parm h6; 902 #endif 903 } header; /* For incoming skbs */ 904 }; 905 }; 906 907 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 908 909 extern const struct inet_connection_sock_af_ops ipv4_specific; 910 911 #if IS_ENABLED(CONFIG_IPV6) 912 /* This is the variant of inet6_iif() that must be used by TCP, 913 * as TCP moves IP6CB into a different location in skb->cb[] 914 */ 915 static inline int tcp_v6_iif(const struct sk_buff *skb) 916 { 917 return TCP_SKB_CB(skb)->header.h6.iif; 918 } 919 920 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 921 { 922 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 923 924 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 925 } 926 927 /* TCP_SKB_CB reference means this can not be used from early demux */ 928 static inline int tcp_v6_sdif(const struct sk_buff *skb) 929 { 930 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 931 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 932 return TCP_SKB_CB(skb)->header.h6.iif; 933 #endif 934 return 0; 935 } 936 937 extern const struct inet_connection_sock_af_ops ipv6_specific; 938 939 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 940 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 941 void tcp_v6_early_demux(struct sk_buff *skb); 942 943 #endif 944 945 /* TCP_SKB_CB reference means this can not be used from early demux */ 946 static inline int tcp_v4_sdif(struct sk_buff *skb) 947 { 948 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 949 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 950 return TCP_SKB_CB(skb)->header.h4.iif; 951 #endif 952 return 0; 953 } 954 955 /* Due to TSO, an SKB can be composed of multiple actual 956 * packets. To keep these tracked properly, we use this. 957 */ 958 static inline int tcp_skb_pcount(const struct sk_buff *skb) 959 { 960 return TCP_SKB_CB(skb)->tcp_gso_segs; 961 } 962 963 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 964 { 965 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 966 } 967 968 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 969 { 970 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 971 } 972 973 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 974 static inline int tcp_skb_mss(const struct sk_buff *skb) 975 { 976 return TCP_SKB_CB(skb)->tcp_gso_size; 977 } 978 979 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 980 { 981 return likely(!TCP_SKB_CB(skb)->eor); 982 } 983 984 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 985 const struct sk_buff *from) 986 { 987 return likely(tcp_skb_can_collapse_to(to) && 988 mptcp_skb_can_collapse(to, from) && 989 skb_pure_zcopy_same(to, from)); 990 } 991 992 /* Events passed to congestion control interface */ 993 enum tcp_ca_event { 994 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 995 CA_EVENT_CWND_RESTART, /* congestion window restart */ 996 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 997 CA_EVENT_LOSS, /* loss timeout */ 998 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 999 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1000 }; 1001 1002 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1003 enum tcp_ca_ack_event_flags { 1004 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1005 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1006 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1007 }; 1008 1009 /* 1010 * Interface for adding new TCP congestion control handlers 1011 */ 1012 #define TCP_CA_NAME_MAX 16 1013 #define TCP_CA_MAX 128 1014 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1015 1016 #define TCP_CA_UNSPEC 0 1017 1018 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1019 #define TCP_CONG_NON_RESTRICTED 0x1 1020 /* Requires ECN/ECT set on all packets */ 1021 #define TCP_CONG_NEEDS_ECN 0x2 1022 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1023 1024 union tcp_cc_info; 1025 1026 struct ack_sample { 1027 u32 pkts_acked; 1028 s32 rtt_us; 1029 u32 in_flight; 1030 }; 1031 1032 /* A rate sample measures the number of (original/retransmitted) data 1033 * packets delivered "delivered" over an interval of time "interval_us". 1034 * The tcp_rate.c code fills in the rate sample, and congestion 1035 * control modules that define a cong_control function to run at the end 1036 * of ACK processing can optionally chose to consult this sample when 1037 * setting cwnd and pacing rate. 1038 * A sample is invalid if "delivered" or "interval_us" is negative. 1039 */ 1040 struct rate_sample { 1041 u64 prior_mstamp; /* starting timestamp for interval */ 1042 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1043 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1044 s32 delivered; /* number of packets delivered over interval */ 1045 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1046 long interval_us; /* time for tp->delivered to incr "delivered" */ 1047 u32 snd_interval_us; /* snd interval for delivered packets */ 1048 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1049 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1050 int losses; /* number of packets marked lost upon ACK */ 1051 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1052 u32 prior_in_flight; /* in flight before this ACK */ 1053 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1054 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1055 bool is_retrans; /* is sample from retransmission? */ 1056 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1057 }; 1058 1059 struct tcp_congestion_ops { 1060 /* fast path fields are put first to fill one cache line */ 1061 1062 /* return slow start threshold (required) */ 1063 u32 (*ssthresh)(struct sock *sk); 1064 1065 /* do new cwnd calculation (required) */ 1066 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1067 1068 /* call before changing ca_state (optional) */ 1069 void (*set_state)(struct sock *sk, u8 new_state); 1070 1071 /* call when cwnd event occurs (optional) */ 1072 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1073 1074 /* call when ack arrives (optional) */ 1075 void (*in_ack_event)(struct sock *sk, u32 flags); 1076 1077 /* hook for packet ack accounting (optional) */ 1078 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1079 1080 /* override sysctl_tcp_min_tso_segs */ 1081 u32 (*min_tso_segs)(struct sock *sk); 1082 1083 /* call when packets are delivered to update cwnd and pacing rate, 1084 * after all the ca_state processing. (optional) 1085 */ 1086 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1087 1088 1089 /* new value of cwnd after loss (required) */ 1090 u32 (*undo_cwnd)(struct sock *sk); 1091 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1092 u32 (*sndbuf_expand)(struct sock *sk); 1093 1094 /* control/slow paths put last */ 1095 /* get info for inet_diag (optional) */ 1096 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1097 union tcp_cc_info *info); 1098 1099 char name[TCP_CA_NAME_MAX]; 1100 struct module *owner; 1101 struct list_head list; 1102 u32 key; 1103 u32 flags; 1104 1105 /* initialize private data (optional) */ 1106 void (*init)(struct sock *sk); 1107 /* cleanup private data (optional) */ 1108 void (*release)(struct sock *sk); 1109 } ____cacheline_aligned_in_smp; 1110 1111 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1112 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1113 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1114 struct tcp_congestion_ops *old_type); 1115 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1116 1117 void tcp_assign_congestion_control(struct sock *sk); 1118 void tcp_init_congestion_control(struct sock *sk); 1119 void tcp_cleanup_congestion_control(struct sock *sk); 1120 int tcp_set_default_congestion_control(struct net *net, const char *name); 1121 void tcp_get_default_congestion_control(struct net *net, char *name); 1122 void tcp_get_available_congestion_control(char *buf, size_t len); 1123 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1124 int tcp_set_allowed_congestion_control(char *allowed); 1125 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1126 bool cap_net_admin); 1127 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1128 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1129 1130 u32 tcp_reno_ssthresh(struct sock *sk); 1131 u32 tcp_reno_undo_cwnd(struct sock *sk); 1132 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1133 extern struct tcp_congestion_ops tcp_reno; 1134 1135 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1136 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1137 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1138 #ifdef CONFIG_INET 1139 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1140 #else 1141 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1142 { 1143 return NULL; 1144 } 1145 #endif 1146 1147 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1148 { 1149 const struct inet_connection_sock *icsk = inet_csk(sk); 1150 1151 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1152 } 1153 1154 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1155 { 1156 const struct inet_connection_sock *icsk = inet_csk(sk); 1157 1158 if (icsk->icsk_ca_ops->cwnd_event) 1159 icsk->icsk_ca_ops->cwnd_event(sk, event); 1160 } 1161 1162 /* From tcp_cong.c */ 1163 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1164 1165 /* From tcp_rate.c */ 1166 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1167 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1168 struct rate_sample *rs); 1169 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1170 bool is_sack_reneg, struct rate_sample *rs); 1171 void tcp_rate_check_app_limited(struct sock *sk); 1172 1173 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1174 { 1175 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1176 } 1177 1178 /* These functions determine how the current flow behaves in respect of SACK 1179 * handling. SACK is negotiated with the peer, and therefore it can vary 1180 * between different flows. 1181 * 1182 * tcp_is_sack - SACK enabled 1183 * tcp_is_reno - No SACK 1184 */ 1185 static inline int tcp_is_sack(const struct tcp_sock *tp) 1186 { 1187 return likely(tp->rx_opt.sack_ok); 1188 } 1189 1190 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1191 { 1192 return !tcp_is_sack(tp); 1193 } 1194 1195 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1196 { 1197 return tp->sacked_out + tp->lost_out; 1198 } 1199 1200 /* This determines how many packets are "in the network" to the best 1201 * of our knowledge. In many cases it is conservative, but where 1202 * detailed information is available from the receiver (via SACK 1203 * blocks etc.) we can make more aggressive calculations. 1204 * 1205 * Use this for decisions involving congestion control, use just 1206 * tp->packets_out to determine if the send queue is empty or not. 1207 * 1208 * Read this equation as: 1209 * 1210 * "Packets sent once on transmission queue" MINUS 1211 * "Packets left network, but not honestly ACKed yet" PLUS 1212 * "Packets fast retransmitted" 1213 */ 1214 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1215 { 1216 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1217 } 1218 1219 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1220 1221 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1222 { 1223 return tp->snd_cwnd; 1224 } 1225 1226 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1227 { 1228 WARN_ON_ONCE((int)val <= 0); 1229 tp->snd_cwnd = val; 1230 } 1231 1232 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1233 { 1234 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1235 } 1236 1237 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1238 { 1239 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1240 } 1241 1242 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1243 { 1244 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1245 (1 << inet_csk(sk)->icsk_ca_state); 1246 } 1247 1248 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1249 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1250 * ssthresh. 1251 */ 1252 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1253 { 1254 const struct tcp_sock *tp = tcp_sk(sk); 1255 1256 if (tcp_in_cwnd_reduction(sk)) 1257 return tp->snd_ssthresh; 1258 else 1259 return max(tp->snd_ssthresh, 1260 ((tcp_snd_cwnd(tp) >> 1) + 1261 (tcp_snd_cwnd(tp) >> 2))); 1262 } 1263 1264 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1265 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1266 1267 void tcp_enter_cwr(struct sock *sk); 1268 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1269 1270 /* The maximum number of MSS of available cwnd for which TSO defers 1271 * sending if not using sysctl_tcp_tso_win_divisor. 1272 */ 1273 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1274 { 1275 return 3; 1276 } 1277 1278 /* Returns end sequence number of the receiver's advertised window */ 1279 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1280 { 1281 return tp->snd_una + tp->snd_wnd; 1282 } 1283 1284 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1285 * flexible approach. The RFC suggests cwnd should not be raised unless 1286 * it was fully used previously. And that's exactly what we do in 1287 * congestion avoidance mode. But in slow start we allow cwnd to grow 1288 * as long as the application has used half the cwnd. 1289 * Example : 1290 * cwnd is 10 (IW10), but application sends 9 frames. 1291 * We allow cwnd to reach 18 when all frames are ACKed. 1292 * This check is safe because it's as aggressive as slow start which already 1293 * risks 100% overshoot. The advantage is that we discourage application to 1294 * either send more filler packets or data to artificially blow up the cwnd 1295 * usage, and allow application-limited process to probe bw more aggressively. 1296 */ 1297 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1298 { 1299 const struct tcp_sock *tp = tcp_sk(sk); 1300 1301 if (tp->is_cwnd_limited) 1302 return true; 1303 1304 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1305 if (tcp_in_slow_start(tp)) 1306 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1307 1308 return false; 1309 } 1310 1311 /* BBR congestion control needs pacing. 1312 * Same remark for SO_MAX_PACING_RATE. 1313 * sch_fq packet scheduler is efficiently handling pacing, 1314 * but is not always installed/used. 1315 * Return true if TCP stack should pace packets itself. 1316 */ 1317 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1318 { 1319 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1320 } 1321 1322 /* Estimates in how many jiffies next packet for this flow can be sent. 1323 * Scheduling a retransmit timer too early would be silly. 1324 */ 1325 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1326 { 1327 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1328 1329 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1330 } 1331 1332 static inline void tcp_reset_xmit_timer(struct sock *sk, 1333 const int what, 1334 unsigned long when, 1335 const unsigned long max_when) 1336 { 1337 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1338 max_when); 1339 } 1340 1341 /* Something is really bad, we could not queue an additional packet, 1342 * because qdisc is full or receiver sent a 0 window, or we are paced. 1343 * We do not want to add fuel to the fire, or abort too early, 1344 * so make sure the timer we arm now is at least 200ms in the future, 1345 * regardless of current icsk_rto value (as it could be ~2ms) 1346 */ 1347 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1348 { 1349 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1350 } 1351 1352 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1353 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1354 unsigned long max_when) 1355 { 1356 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1357 inet_csk(sk)->icsk_backoff); 1358 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1359 1360 return (unsigned long)min_t(u64, when, max_when); 1361 } 1362 1363 static inline void tcp_check_probe_timer(struct sock *sk) 1364 { 1365 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1366 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1367 tcp_probe0_base(sk), TCP_RTO_MAX); 1368 } 1369 1370 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1371 { 1372 tp->snd_wl1 = seq; 1373 } 1374 1375 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1376 { 1377 tp->snd_wl1 = seq; 1378 } 1379 1380 /* 1381 * Calculate(/check) TCP checksum 1382 */ 1383 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1384 __be32 daddr, __wsum base) 1385 { 1386 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1387 } 1388 1389 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1390 { 1391 return !skb_csum_unnecessary(skb) && 1392 __skb_checksum_complete(skb); 1393 } 1394 1395 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1396 enum skb_drop_reason *reason); 1397 1398 1399 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1400 void tcp_set_state(struct sock *sk, int state); 1401 void tcp_done(struct sock *sk); 1402 int tcp_abort(struct sock *sk, int err); 1403 1404 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1405 { 1406 rx_opt->dsack = 0; 1407 rx_opt->num_sacks = 0; 1408 } 1409 1410 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1411 1412 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1413 { 1414 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1415 struct tcp_sock *tp = tcp_sk(sk); 1416 s32 delta; 1417 1418 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1419 tp->packets_out || ca_ops->cong_control) 1420 return; 1421 delta = tcp_jiffies32 - tp->lsndtime; 1422 if (delta > inet_csk(sk)->icsk_rto) 1423 tcp_cwnd_restart(sk, delta); 1424 } 1425 1426 /* Determine a window scaling and initial window to offer. */ 1427 void tcp_select_initial_window(const struct sock *sk, int __space, 1428 __u32 mss, __u32 *rcv_wnd, 1429 __u32 *window_clamp, int wscale_ok, 1430 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1431 1432 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1433 { 1434 s64 scaled_space = (s64)space * scaling_ratio; 1435 1436 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1437 } 1438 1439 static inline int tcp_win_from_space(const struct sock *sk, int space) 1440 { 1441 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1442 } 1443 1444 /* inverse of __tcp_win_from_space() */ 1445 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1446 { 1447 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1448 1449 do_div(val, scaling_ratio); 1450 return val; 1451 } 1452 1453 static inline int tcp_space_from_win(const struct sock *sk, int win) 1454 { 1455 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1456 } 1457 1458 static inline void tcp_scaling_ratio_init(struct sock *sk) 1459 { 1460 /* Assume a conservative default of 1200 bytes of payload per 4K page. 1461 * This may be adjusted later in tcp_measure_rcv_mss(). 1462 */ 1463 tcp_sk(sk)->scaling_ratio = (1200 << TCP_RMEM_TO_WIN_SCALE) / 1464 SKB_TRUESIZE(4096); 1465 } 1466 1467 /* Note: caller must be prepared to deal with negative returns */ 1468 static inline int tcp_space(const struct sock *sk) 1469 { 1470 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1471 READ_ONCE(sk->sk_backlog.len) - 1472 atomic_read(&sk->sk_rmem_alloc)); 1473 } 1474 1475 static inline int tcp_full_space(const struct sock *sk) 1476 { 1477 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1478 } 1479 1480 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1481 { 1482 int unused_mem = sk_unused_reserved_mem(sk); 1483 struct tcp_sock *tp = tcp_sk(sk); 1484 1485 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1486 if (unused_mem) 1487 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1488 tcp_win_from_space(sk, unused_mem)); 1489 } 1490 1491 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1492 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1493 1494 1495 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1496 * If 87.5 % (7/8) of the space has been consumed, we want to override 1497 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1498 * len/truesize ratio. 1499 */ 1500 static inline bool tcp_rmem_pressure(const struct sock *sk) 1501 { 1502 int rcvbuf, threshold; 1503 1504 if (tcp_under_memory_pressure(sk)) 1505 return true; 1506 1507 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1508 threshold = rcvbuf - (rcvbuf >> 3); 1509 1510 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1511 } 1512 1513 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1514 { 1515 const struct tcp_sock *tp = tcp_sk(sk); 1516 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1517 1518 if (avail <= 0) 1519 return false; 1520 1521 return (avail >= target) || tcp_rmem_pressure(sk) || 1522 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1523 } 1524 1525 extern void tcp_openreq_init_rwin(struct request_sock *req, 1526 const struct sock *sk_listener, 1527 const struct dst_entry *dst); 1528 1529 void tcp_enter_memory_pressure(struct sock *sk); 1530 void tcp_leave_memory_pressure(struct sock *sk); 1531 1532 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1533 { 1534 struct net *net = sock_net((struct sock *)tp); 1535 int val; 1536 1537 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1538 * and do_tcp_setsockopt(). 1539 */ 1540 val = READ_ONCE(tp->keepalive_intvl); 1541 1542 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1543 } 1544 1545 static inline int keepalive_time_when(const struct tcp_sock *tp) 1546 { 1547 struct net *net = sock_net((struct sock *)tp); 1548 int val; 1549 1550 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1551 val = READ_ONCE(tp->keepalive_time); 1552 1553 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1554 } 1555 1556 static inline int keepalive_probes(const struct tcp_sock *tp) 1557 { 1558 struct net *net = sock_net((struct sock *)tp); 1559 int val; 1560 1561 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1562 * and do_tcp_setsockopt(). 1563 */ 1564 val = READ_ONCE(tp->keepalive_probes); 1565 1566 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1567 } 1568 1569 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1570 { 1571 const struct inet_connection_sock *icsk = &tp->inet_conn; 1572 1573 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1574 tcp_jiffies32 - tp->rcv_tstamp); 1575 } 1576 1577 static inline int tcp_fin_time(const struct sock *sk) 1578 { 1579 int fin_timeout = tcp_sk(sk)->linger2 ? : 1580 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1581 const int rto = inet_csk(sk)->icsk_rto; 1582 1583 if (fin_timeout < (rto << 2) - (rto >> 1)) 1584 fin_timeout = (rto << 2) - (rto >> 1); 1585 1586 return fin_timeout; 1587 } 1588 1589 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1590 int paws_win) 1591 { 1592 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1593 return true; 1594 if (unlikely(!time_before32(ktime_get_seconds(), 1595 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1596 return true; 1597 /* 1598 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1599 * then following tcp messages have valid values. Ignore 0 value, 1600 * or else 'negative' tsval might forbid us to accept their packets. 1601 */ 1602 if (!rx_opt->ts_recent) 1603 return true; 1604 return false; 1605 } 1606 1607 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1608 int rst) 1609 { 1610 if (tcp_paws_check(rx_opt, 0)) 1611 return false; 1612 1613 /* RST segments are not recommended to carry timestamp, 1614 and, if they do, it is recommended to ignore PAWS because 1615 "their cleanup function should take precedence over timestamps." 1616 Certainly, it is mistake. It is necessary to understand the reasons 1617 of this constraint to relax it: if peer reboots, clock may go 1618 out-of-sync and half-open connections will not be reset. 1619 Actually, the problem would be not existing if all 1620 the implementations followed draft about maintaining clock 1621 via reboots. Linux-2.2 DOES NOT! 1622 1623 However, we can relax time bounds for RST segments to MSL. 1624 */ 1625 if (rst && !time_before32(ktime_get_seconds(), 1626 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1627 return false; 1628 return true; 1629 } 1630 1631 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1632 int mib_idx, u32 *last_oow_ack_time); 1633 1634 static inline void tcp_mib_init(struct net *net) 1635 { 1636 /* See RFC 2012 */ 1637 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1638 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1639 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1640 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1641 } 1642 1643 /* from STCP */ 1644 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1645 { 1646 tp->lost_skb_hint = NULL; 1647 } 1648 1649 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1650 { 1651 tcp_clear_retrans_hints_partial(tp); 1652 tp->retransmit_skb_hint = NULL; 1653 } 1654 1655 union tcp_md5_addr { 1656 struct in_addr a4; 1657 #if IS_ENABLED(CONFIG_IPV6) 1658 struct in6_addr a6; 1659 #endif 1660 }; 1661 1662 /* - key database */ 1663 struct tcp_md5sig_key { 1664 struct hlist_node node; 1665 u8 keylen; 1666 u8 family; /* AF_INET or AF_INET6 */ 1667 u8 prefixlen; 1668 u8 flags; 1669 union tcp_md5_addr addr; 1670 int l3index; /* set if key added with L3 scope */ 1671 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1672 struct rcu_head rcu; 1673 }; 1674 1675 /* - sock block */ 1676 struct tcp_md5sig_info { 1677 struct hlist_head head; 1678 struct rcu_head rcu; 1679 }; 1680 1681 /* - pseudo header */ 1682 struct tcp4_pseudohdr { 1683 __be32 saddr; 1684 __be32 daddr; 1685 __u8 pad; 1686 __u8 protocol; 1687 __be16 len; 1688 }; 1689 1690 struct tcp6_pseudohdr { 1691 struct in6_addr saddr; 1692 struct in6_addr daddr; 1693 __be32 len; 1694 __be32 protocol; /* including padding */ 1695 }; 1696 1697 union tcp_md5sum_block { 1698 struct tcp4_pseudohdr ip4; 1699 #if IS_ENABLED(CONFIG_IPV6) 1700 struct tcp6_pseudohdr ip6; 1701 #endif 1702 }; 1703 1704 /* - pool: digest algorithm, hash description and scratch buffer */ 1705 struct tcp_md5sig_pool { 1706 struct ahash_request *md5_req; 1707 void *scratch; 1708 }; 1709 1710 /* - functions */ 1711 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1712 const struct sock *sk, const struct sk_buff *skb); 1713 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1714 int family, u8 prefixlen, int l3index, u8 flags, 1715 const u8 *newkey, u8 newkeylen); 1716 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1717 int family, u8 prefixlen, int l3index, 1718 struct tcp_md5sig_key *key); 1719 1720 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1721 int family, u8 prefixlen, int l3index, u8 flags); 1722 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1723 const struct sock *addr_sk); 1724 1725 #ifdef CONFIG_TCP_MD5SIG 1726 #include <linux/jump_label.h> 1727 extern struct static_key_false_deferred tcp_md5_needed; 1728 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1729 const union tcp_md5_addr *addr, 1730 int family); 1731 static inline struct tcp_md5sig_key * 1732 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1733 const union tcp_md5_addr *addr, int family) 1734 { 1735 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1736 return NULL; 1737 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1738 } 1739 1740 enum skb_drop_reason 1741 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1742 const void *saddr, const void *daddr, 1743 int family, int dif, int sdif); 1744 1745 1746 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1747 #else 1748 static inline struct tcp_md5sig_key * 1749 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1750 const union tcp_md5_addr *addr, int family) 1751 { 1752 return NULL; 1753 } 1754 1755 static inline enum skb_drop_reason 1756 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1757 const void *saddr, const void *daddr, 1758 int family, int dif, int sdif) 1759 { 1760 return SKB_NOT_DROPPED_YET; 1761 } 1762 #define tcp_twsk_md5_key(twsk) NULL 1763 #endif 1764 1765 bool tcp_alloc_md5sig_pool(void); 1766 1767 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1768 static inline void tcp_put_md5sig_pool(void) 1769 { 1770 local_bh_enable(); 1771 } 1772 1773 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1774 unsigned int header_len); 1775 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1776 const struct tcp_md5sig_key *key); 1777 1778 /* From tcp_fastopen.c */ 1779 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1780 struct tcp_fastopen_cookie *cookie); 1781 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1782 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1783 u16 try_exp); 1784 struct tcp_fastopen_request { 1785 /* Fast Open cookie. Size 0 means a cookie request */ 1786 struct tcp_fastopen_cookie cookie; 1787 struct msghdr *data; /* data in MSG_FASTOPEN */ 1788 size_t size; 1789 int copied; /* queued in tcp_connect() */ 1790 struct ubuf_info *uarg; 1791 }; 1792 void tcp_free_fastopen_req(struct tcp_sock *tp); 1793 void tcp_fastopen_destroy_cipher(struct sock *sk); 1794 void tcp_fastopen_ctx_destroy(struct net *net); 1795 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1796 void *primary_key, void *backup_key); 1797 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1798 u64 *key); 1799 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1800 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1801 struct request_sock *req, 1802 struct tcp_fastopen_cookie *foc, 1803 const struct dst_entry *dst); 1804 void tcp_fastopen_init_key_once(struct net *net); 1805 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1806 struct tcp_fastopen_cookie *cookie); 1807 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1808 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1809 #define TCP_FASTOPEN_KEY_MAX 2 1810 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1811 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1812 1813 /* Fastopen key context */ 1814 struct tcp_fastopen_context { 1815 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1816 int num; 1817 struct rcu_head rcu; 1818 }; 1819 1820 void tcp_fastopen_active_disable(struct sock *sk); 1821 bool tcp_fastopen_active_should_disable(struct sock *sk); 1822 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1823 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1824 1825 /* Caller needs to wrap with rcu_read_(un)lock() */ 1826 static inline 1827 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1828 { 1829 struct tcp_fastopen_context *ctx; 1830 1831 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1832 if (!ctx) 1833 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1834 return ctx; 1835 } 1836 1837 static inline 1838 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1839 const struct tcp_fastopen_cookie *orig) 1840 { 1841 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1842 orig->len == foc->len && 1843 !memcmp(orig->val, foc->val, foc->len)) 1844 return true; 1845 return false; 1846 } 1847 1848 static inline 1849 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1850 { 1851 return ctx->num; 1852 } 1853 1854 /* Latencies incurred by various limits for a sender. They are 1855 * chronograph-like stats that are mutually exclusive. 1856 */ 1857 enum tcp_chrono { 1858 TCP_CHRONO_UNSPEC, 1859 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1860 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1861 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1862 __TCP_CHRONO_MAX, 1863 }; 1864 1865 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1866 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1867 1868 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1869 * the same memory storage than skb->destructor/_skb_refdst 1870 */ 1871 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1872 { 1873 skb->destructor = NULL; 1874 skb->_skb_refdst = 0UL; 1875 } 1876 1877 #define tcp_skb_tsorted_save(skb) { \ 1878 unsigned long _save = skb->_skb_refdst; \ 1879 skb->_skb_refdst = 0UL; 1880 1881 #define tcp_skb_tsorted_restore(skb) \ 1882 skb->_skb_refdst = _save; \ 1883 } 1884 1885 void tcp_write_queue_purge(struct sock *sk); 1886 1887 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1888 { 1889 return skb_rb_first(&sk->tcp_rtx_queue); 1890 } 1891 1892 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1893 { 1894 return skb_rb_last(&sk->tcp_rtx_queue); 1895 } 1896 1897 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1898 { 1899 return skb_peek_tail(&sk->sk_write_queue); 1900 } 1901 1902 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1903 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1904 1905 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1906 { 1907 return skb_peek(&sk->sk_write_queue); 1908 } 1909 1910 static inline bool tcp_skb_is_last(const struct sock *sk, 1911 const struct sk_buff *skb) 1912 { 1913 return skb_queue_is_last(&sk->sk_write_queue, skb); 1914 } 1915 1916 /** 1917 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1918 * @sk: socket 1919 * 1920 * Since the write queue can have a temporary empty skb in it, 1921 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1922 */ 1923 static inline bool tcp_write_queue_empty(const struct sock *sk) 1924 { 1925 const struct tcp_sock *tp = tcp_sk(sk); 1926 1927 return tp->write_seq == tp->snd_nxt; 1928 } 1929 1930 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1931 { 1932 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1933 } 1934 1935 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1936 { 1937 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1938 } 1939 1940 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1941 { 1942 __skb_queue_tail(&sk->sk_write_queue, skb); 1943 1944 /* Queue it, remembering where we must start sending. */ 1945 if (sk->sk_write_queue.next == skb) 1946 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1947 } 1948 1949 /* Insert new before skb on the write queue of sk. */ 1950 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1951 struct sk_buff *skb, 1952 struct sock *sk) 1953 { 1954 __skb_queue_before(&sk->sk_write_queue, skb, new); 1955 } 1956 1957 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1958 { 1959 tcp_skb_tsorted_anchor_cleanup(skb); 1960 __skb_unlink(skb, &sk->sk_write_queue); 1961 } 1962 1963 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1964 1965 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1966 { 1967 tcp_skb_tsorted_anchor_cleanup(skb); 1968 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1969 } 1970 1971 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1972 { 1973 list_del(&skb->tcp_tsorted_anchor); 1974 tcp_rtx_queue_unlink(skb, sk); 1975 tcp_wmem_free_skb(sk, skb); 1976 } 1977 1978 static inline void tcp_push_pending_frames(struct sock *sk) 1979 { 1980 if (tcp_send_head(sk)) { 1981 struct tcp_sock *tp = tcp_sk(sk); 1982 1983 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1984 } 1985 } 1986 1987 /* Start sequence of the skb just after the highest skb with SACKed 1988 * bit, valid only if sacked_out > 0 or when the caller has ensured 1989 * validity by itself. 1990 */ 1991 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1992 { 1993 if (!tp->sacked_out) 1994 return tp->snd_una; 1995 1996 if (tp->highest_sack == NULL) 1997 return tp->snd_nxt; 1998 1999 return TCP_SKB_CB(tp->highest_sack)->seq; 2000 } 2001 2002 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2003 { 2004 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2005 } 2006 2007 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2008 { 2009 return tcp_sk(sk)->highest_sack; 2010 } 2011 2012 static inline void tcp_highest_sack_reset(struct sock *sk) 2013 { 2014 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2015 } 2016 2017 /* Called when old skb is about to be deleted and replaced by new skb */ 2018 static inline void tcp_highest_sack_replace(struct sock *sk, 2019 struct sk_buff *old, 2020 struct sk_buff *new) 2021 { 2022 if (old == tcp_highest_sack(sk)) 2023 tcp_sk(sk)->highest_sack = new; 2024 } 2025 2026 /* This helper checks if socket has IP_TRANSPARENT set */ 2027 static inline bool inet_sk_transparent(const struct sock *sk) 2028 { 2029 switch (sk->sk_state) { 2030 case TCP_TIME_WAIT: 2031 return inet_twsk(sk)->tw_transparent; 2032 case TCP_NEW_SYN_RECV: 2033 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2034 } 2035 return inet_test_bit(TRANSPARENT, sk); 2036 } 2037 2038 /* Determines whether this is a thin stream (which may suffer from 2039 * increased latency). Used to trigger latency-reducing mechanisms. 2040 */ 2041 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2042 { 2043 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2044 } 2045 2046 /* /proc */ 2047 enum tcp_seq_states { 2048 TCP_SEQ_STATE_LISTENING, 2049 TCP_SEQ_STATE_ESTABLISHED, 2050 }; 2051 2052 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2053 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2054 void tcp_seq_stop(struct seq_file *seq, void *v); 2055 2056 struct tcp_seq_afinfo { 2057 sa_family_t family; 2058 }; 2059 2060 struct tcp_iter_state { 2061 struct seq_net_private p; 2062 enum tcp_seq_states state; 2063 struct sock *syn_wait_sk; 2064 int bucket, offset, sbucket, num; 2065 loff_t last_pos; 2066 }; 2067 2068 extern struct request_sock_ops tcp_request_sock_ops; 2069 extern struct request_sock_ops tcp6_request_sock_ops; 2070 2071 void tcp_v4_destroy_sock(struct sock *sk); 2072 2073 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2074 netdev_features_t features); 2075 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2076 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2077 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2078 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2079 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2080 void tcp_gro_complete(struct sk_buff *skb); 2081 2082 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2083 2084 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2085 { 2086 struct net *net = sock_net((struct sock *)tp); 2087 u32 val; 2088 2089 val = READ_ONCE(tp->notsent_lowat); 2090 2091 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2092 } 2093 2094 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2095 2096 #ifdef CONFIG_PROC_FS 2097 int tcp4_proc_init(void); 2098 void tcp4_proc_exit(void); 2099 #endif 2100 2101 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2102 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2103 const struct tcp_request_sock_ops *af_ops, 2104 struct sock *sk, struct sk_buff *skb); 2105 2106 /* TCP af-specific functions */ 2107 struct tcp_sock_af_ops { 2108 #ifdef CONFIG_TCP_MD5SIG 2109 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2110 const struct sock *addr_sk); 2111 int (*calc_md5_hash)(char *location, 2112 const struct tcp_md5sig_key *md5, 2113 const struct sock *sk, 2114 const struct sk_buff *skb); 2115 int (*md5_parse)(struct sock *sk, 2116 int optname, 2117 sockptr_t optval, 2118 int optlen); 2119 #endif 2120 }; 2121 2122 struct tcp_request_sock_ops { 2123 u16 mss_clamp; 2124 #ifdef CONFIG_TCP_MD5SIG 2125 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2126 const struct sock *addr_sk); 2127 int (*calc_md5_hash) (char *location, 2128 const struct tcp_md5sig_key *md5, 2129 const struct sock *sk, 2130 const struct sk_buff *skb); 2131 #endif 2132 #ifdef CONFIG_SYN_COOKIES 2133 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2134 __u16 *mss); 2135 #endif 2136 struct dst_entry *(*route_req)(const struct sock *sk, 2137 struct sk_buff *skb, 2138 struct flowi *fl, 2139 struct request_sock *req); 2140 u32 (*init_seq)(const struct sk_buff *skb); 2141 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2142 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2143 struct flowi *fl, struct request_sock *req, 2144 struct tcp_fastopen_cookie *foc, 2145 enum tcp_synack_type synack_type, 2146 struct sk_buff *syn_skb); 2147 }; 2148 2149 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2150 #if IS_ENABLED(CONFIG_IPV6) 2151 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2152 #endif 2153 2154 #ifdef CONFIG_SYN_COOKIES 2155 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2156 const struct sock *sk, struct sk_buff *skb, 2157 __u16 *mss) 2158 { 2159 tcp_synq_overflow(sk); 2160 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2161 return ops->cookie_init_seq(skb, mss); 2162 } 2163 #else 2164 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2165 const struct sock *sk, struct sk_buff *skb, 2166 __u16 *mss) 2167 { 2168 return 0; 2169 } 2170 #endif 2171 2172 int tcpv4_offload_init(void); 2173 2174 void tcp_v4_init(void); 2175 void tcp_init(void); 2176 2177 /* tcp_recovery.c */ 2178 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2179 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2180 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2181 u32 reo_wnd); 2182 extern bool tcp_rack_mark_lost(struct sock *sk); 2183 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2184 u64 xmit_time); 2185 extern void tcp_rack_reo_timeout(struct sock *sk); 2186 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2187 2188 /* tcp_plb.c */ 2189 2190 /* 2191 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2192 * expects cong_ratio which represents fraction of traffic that experienced 2193 * congestion over a single RTT. In order to avoid floating point operations, 2194 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2195 */ 2196 #define TCP_PLB_SCALE 8 2197 2198 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2199 struct tcp_plb_state { 2200 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2201 unused:3; 2202 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2203 }; 2204 2205 static inline void tcp_plb_init(const struct sock *sk, 2206 struct tcp_plb_state *plb) 2207 { 2208 plb->consec_cong_rounds = 0; 2209 plb->pause_until = 0; 2210 } 2211 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2212 const int cong_ratio); 2213 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2214 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2215 2216 /* At how many usecs into the future should the RTO fire? */ 2217 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2218 { 2219 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2220 u32 rto = inet_csk(sk)->icsk_rto; 2221 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2222 2223 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2224 } 2225 2226 /* 2227 * Save and compile IPv4 options, return a pointer to it 2228 */ 2229 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2230 struct sk_buff *skb) 2231 { 2232 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2233 struct ip_options_rcu *dopt = NULL; 2234 2235 if (opt->optlen) { 2236 int opt_size = sizeof(*dopt) + opt->optlen; 2237 2238 dopt = kmalloc(opt_size, GFP_ATOMIC); 2239 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2240 kfree(dopt); 2241 dopt = NULL; 2242 } 2243 } 2244 return dopt; 2245 } 2246 2247 /* locally generated TCP pure ACKs have skb->truesize == 2 2248 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2249 * This is much faster than dissecting the packet to find out. 2250 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2251 */ 2252 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2253 { 2254 return skb->truesize == 2; 2255 } 2256 2257 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2258 { 2259 skb->truesize = 2; 2260 } 2261 2262 static inline int tcp_inq(struct sock *sk) 2263 { 2264 struct tcp_sock *tp = tcp_sk(sk); 2265 int answ; 2266 2267 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2268 answ = 0; 2269 } else if (sock_flag(sk, SOCK_URGINLINE) || 2270 !tp->urg_data || 2271 before(tp->urg_seq, tp->copied_seq) || 2272 !before(tp->urg_seq, tp->rcv_nxt)) { 2273 2274 answ = tp->rcv_nxt - tp->copied_seq; 2275 2276 /* Subtract 1, if FIN was received */ 2277 if (answ && sock_flag(sk, SOCK_DONE)) 2278 answ--; 2279 } else { 2280 answ = tp->urg_seq - tp->copied_seq; 2281 } 2282 2283 return answ; 2284 } 2285 2286 int tcp_peek_len(struct socket *sock); 2287 2288 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2289 { 2290 u16 segs_in; 2291 2292 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2293 2294 /* We update these fields while other threads might 2295 * read them from tcp_get_info() 2296 */ 2297 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2298 if (skb->len > tcp_hdrlen(skb)) 2299 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2300 } 2301 2302 /* 2303 * TCP listen path runs lockless. 2304 * We forced "struct sock" to be const qualified to make sure 2305 * we don't modify one of its field by mistake. 2306 * Here, we increment sk_drops which is an atomic_t, so we can safely 2307 * make sock writable again. 2308 */ 2309 static inline void tcp_listendrop(const struct sock *sk) 2310 { 2311 atomic_inc(&((struct sock *)sk)->sk_drops); 2312 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2313 } 2314 2315 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2316 2317 /* 2318 * Interface for adding Upper Level Protocols over TCP 2319 */ 2320 2321 #define TCP_ULP_NAME_MAX 16 2322 #define TCP_ULP_MAX 128 2323 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2324 2325 struct tcp_ulp_ops { 2326 struct list_head list; 2327 2328 /* initialize ulp */ 2329 int (*init)(struct sock *sk); 2330 /* update ulp */ 2331 void (*update)(struct sock *sk, struct proto *p, 2332 void (*write_space)(struct sock *sk)); 2333 /* cleanup ulp */ 2334 void (*release)(struct sock *sk); 2335 /* diagnostic */ 2336 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2337 size_t (*get_info_size)(const struct sock *sk); 2338 /* clone ulp */ 2339 void (*clone)(const struct request_sock *req, struct sock *newsk, 2340 const gfp_t priority); 2341 2342 char name[TCP_ULP_NAME_MAX]; 2343 struct module *owner; 2344 }; 2345 int tcp_register_ulp(struct tcp_ulp_ops *type); 2346 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2347 int tcp_set_ulp(struct sock *sk, const char *name); 2348 void tcp_get_available_ulp(char *buf, size_t len); 2349 void tcp_cleanup_ulp(struct sock *sk); 2350 void tcp_update_ulp(struct sock *sk, struct proto *p, 2351 void (*write_space)(struct sock *sk)); 2352 2353 #define MODULE_ALIAS_TCP_ULP(name) \ 2354 __MODULE_INFO(alias, alias_userspace, name); \ 2355 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2356 2357 #ifdef CONFIG_NET_SOCK_MSG 2358 struct sk_msg; 2359 struct sk_psock; 2360 2361 #ifdef CONFIG_BPF_SYSCALL 2362 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2363 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2364 #endif /* CONFIG_BPF_SYSCALL */ 2365 2366 #ifdef CONFIG_INET 2367 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2368 #else 2369 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2370 { 2371 } 2372 #endif 2373 2374 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2375 struct sk_msg *msg, u32 bytes, int flags); 2376 #endif /* CONFIG_NET_SOCK_MSG */ 2377 2378 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2379 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2380 { 2381 } 2382 #endif 2383 2384 #ifdef CONFIG_CGROUP_BPF 2385 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2386 struct sk_buff *skb, 2387 unsigned int end_offset) 2388 { 2389 skops->skb = skb; 2390 skops->skb_data_end = skb->data + end_offset; 2391 } 2392 #else 2393 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2394 struct sk_buff *skb, 2395 unsigned int end_offset) 2396 { 2397 } 2398 #endif 2399 2400 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2401 * is < 0, then the BPF op failed (for example if the loaded BPF 2402 * program does not support the chosen operation or there is no BPF 2403 * program loaded). 2404 */ 2405 #ifdef CONFIG_BPF 2406 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2407 { 2408 struct bpf_sock_ops_kern sock_ops; 2409 int ret; 2410 2411 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2412 if (sk_fullsock(sk)) { 2413 sock_ops.is_fullsock = 1; 2414 sock_owned_by_me(sk); 2415 } 2416 2417 sock_ops.sk = sk; 2418 sock_ops.op = op; 2419 if (nargs > 0) 2420 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2421 2422 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2423 if (ret == 0) 2424 ret = sock_ops.reply; 2425 else 2426 ret = -1; 2427 return ret; 2428 } 2429 2430 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2431 { 2432 u32 args[2] = {arg1, arg2}; 2433 2434 return tcp_call_bpf(sk, op, 2, args); 2435 } 2436 2437 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2438 u32 arg3) 2439 { 2440 u32 args[3] = {arg1, arg2, arg3}; 2441 2442 return tcp_call_bpf(sk, op, 3, args); 2443 } 2444 2445 #else 2446 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2447 { 2448 return -EPERM; 2449 } 2450 2451 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2452 { 2453 return -EPERM; 2454 } 2455 2456 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2457 u32 arg3) 2458 { 2459 return -EPERM; 2460 } 2461 2462 #endif 2463 2464 static inline u32 tcp_timeout_init(struct sock *sk) 2465 { 2466 int timeout; 2467 2468 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2469 2470 if (timeout <= 0) 2471 timeout = TCP_TIMEOUT_INIT; 2472 return min_t(int, timeout, TCP_RTO_MAX); 2473 } 2474 2475 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2476 { 2477 int rwnd; 2478 2479 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2480 2481 if (rwnd < 0) 2482 rwnd = 0; 2483 return rwnd; 2484 } 2485 2486 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2487 { 2488 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2489 } 2490 2491 static inline void tcp_bpf_rtt(struct sock *sk) 2492 { 2493 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2494 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2495 } 2496 2497 #if IS_ENABLED(CONFIG_SMC) 2498 extern struct static_key_false tcp_have_smc; 2499 #endif 2500 2501 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2502 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2503 void (*cad)(struct sock *sk, u32 ack_seq)); 2504 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2505 void clean_acked_data_flush(void); 2506 #endif 2507 2508 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2509 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2510 const struct tcp_sock *tp) 2511 { 2512 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2513 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2514 } 2515 2516 /* Compute Earliest Departure Time for some control packets 2517 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2518 */ 2519 static inline u64 tcp_transmit_time(const struct sock *sk) 2520 { 2521 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2522 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2523 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2524 2525 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2526 } 2527 return 0; 2528 } 2529 2530 #endif /* _TCP_H */ 2531