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 #include <linux/net_mm.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 53 int tcp_orphan_count_sum(void); 54 55 void tcp_time_wait(struct sock *sk, int state, int timeo); 56 57 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 58 #define MAX_TCP_OPTION_SPACE 40 59 #define TCP_MIN_SND_MSS 48 60 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 61 62 /* 63 * Never offer a window over 32767 without using window scaling. Some 64 * poor stacks do signed 16bit maths! 65 */ 66 #define MAX_TCP_WINDOW 32767U 67 68 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 69 #define TCP_MIN_MSS 88U 70 71 /* The initial MTU to use for probing */ 72 #define TCP_BASE_MSS 1024 73 74 /* probing interval, default to 10 minutes as per RFC4821 */ 75 #define TCP_PROBE_INTERVAL 600 76 77 /* Specify interval when tcp mtu probing will stop */ 78 #define TCP_PROBE_THRESHOLD 8 79 80 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 81 #define TCP_FASTRETRANS_THRESH 3 82 83 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 84 #define TCP_MAX_QUICKACKS 16U 85 86 /* Maximal number of window scale according to RFC1323 */ 87 #define TCP_MAX_WSCALE 14U 88 89 /* urg_data states */ 90 #define TCP_URG_VALID 0x0100 91 #define TCP_URG_NOTYET 0x0200 92 #define TCP_URG_READ 0x0400 93 94 #define TCP_RETR1 3 /* 95 * This is how many retries it does before it 96 * tries to figure out if the gateway is 97 * down. Minimal RFC value is 3; it corresponds 98 * to ~3sec-8min depending on RTO. 99 */ 100 101 #define TCP_RETR2 15 /* 102 * This should take at least 103 * 90 minutes to time out. 104 * RFC1122 says that the limit is 100 sec. 105 * 15 is ~13-30min depending on RTO. 106 */ 107 108 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 109 * when active opening a connection. 110 * RFC1122 says the minimum retry MUST 111 * be at least 180secs. Nevertheless 112 * this value is corresponding to 113 * 63secs of retransmission with the 114 * current initial RTO. 115 */ 116 117 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 118 * when passive opening a connection. 119 * This is corresponding to 31secs of 120 * retransmission with the current 121 * initial RTO. 122 */ 123 124 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 125 * state, about 60 seconds */ 126 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 127 /* BSD style FIN_WAIT2 deadlock breaker. 128 * It used to be 3min, new value is 60sec, 129 * to combine FIN-WAIT-2 timeout with 130 * TIME-WAIT timer. 131 */ 132 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 133 134 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 135 #if HZ >= 100 136 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 137 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 138 #else 139 #define TCP_DELACK_MIN 4U 140 #define TCP_ATO_MIN 4U 141 #endif 142 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 143 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 144 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 145 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 146 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 147 * used as a fallback RTO for the 148 * initial data transmission if no 149 * valid RTT sample has been acquired, 150 * most likely due to retrans in 3WHS. 151 */ 152 153 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 154 * for local resources. 155 */ 156 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 157 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 158 #define TCP_KEEPALIVE_INTVL (75*HZ) 159 160 #define MAX_TCP_KEEPIDLE 32767 161 #define MAX_TCP_KEEPINTVL 32767 162 #define MAX_TCP_KEEPCNT 127 163 #define MAX_TCP_SYNCNT 127 164 165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 167 * after this time. It should be equal 168 * (or greater than) TCP_TIMEWAIT_LEN 169 * to provide reliability equal to one 170 * provided by timewait state. 171 */ 172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 173 * timestamps. It must be less than 174 * minimal timewait lifetime. 175 */ 176 /* 177 * TCP option 178 */ 179 180 #define TCPOPT_NOP 1 /* Padding */ 181 #define TCPOPT_EOL 0 /* End of options */ 182 #define TCPOPT_MSS 2 /* Segment size negotiating */ 183 #define TCPOPT_WINDOW 3 /* Window scaling */ 184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 185 #define TCPOPT_SACK 5 /* SACK Block */ 186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 188 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 189 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 190 #define TCPOPT_EXP 254 /* Experimental */ 191 /* Magic number to be after the option value for sharing TCP 192 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 193 */ 194 #define TCPOPT_FASTOPEN_MAGIC 0xF989 195 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 196 197 /* 198 * TCP option lengths 199 */ 200 201 #define TCPOLEN_MSS 4 202 #define TCPOLEN_WINDOW 3 203 #define TCPOLEN_SACK_PERM 2 204 #define TCPOLEN_TIMESTAMP 10 205 #define TCPOLEN_MD5SIG 18 206 #define TCPOLEN_FASTOPEN_BASE 2 207 #define TCPOLEN_EXP_FASTOPEN_BASE 4 208 #define TCPOLEN_EXP_SMC_BASE 6 209 210 /* But this is what stacks really send out. */ 211 #define TCPOLEN_TSTAMP_ALIGNED 12 212 #define TCPOLEN_WSCALE_ALIGNED 4 213 #define TCPOLEN_SACKPERM_ALIGNED 4 214 #define TCPOLEN_SACK_BASE 2 215 #define TCPOLEN_SACK_BASE_ALIGNED 4 216 #define TCPOLEN_SACK_PERBLOCK 8 217 #define TCPOLEN_MD5SIG_ALIGNED 20 218 #define TCPOLEN_MSS_ALIGNED 4 219 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 220 221 /* Flags in tp->nonagle */ 222 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 223 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 224 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 225 226 /* TCP thin-stream limits */ 227 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 228 229 /* TCP initial congestion window as per rfc6928 */ 230 #define TCP_INIT_CWND 10 231 232 /* Bit Flags for sysctl_tcp_fastopen */ 233 #define TFO_CLIENT_ENABLE 1 234 #define TFO_SERVER_ENABLE 2 235 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 236 237 /* Accept SYN data w/o any cookie option */ 238 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 239 240 /* Force enable TFO on all listeners, i.e., not requiring the 241 * TCP_FASTOPEN socket option. 242 */ 243 #define TFO_SERVER_WO_SOCKOPT1 0x400 244 245 246 /* sysctl variables for tcp */ 247 extern int sysctl_tcp_max_orphans; 248 extern long sysctl_tcp_mem[3]; 249 250 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 251 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 252 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 253 254 extern atomic_long_t tcp_memory_allocated; 255 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 256 257 extern struct percpu_counter tcp_sockets_allocated; 258 extern unsigned long tcp_memory_pressure; 259 260 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 261 static inline bool tcp_under_memory_pressure(const struct sock *sk) 262 { 263 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 264 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 265 return true; 266 267 return READ_ONCE(tcp_memory_pressure); 268 } 269 /* 270 * The next routines deal with comparing 32 bit unsigned ints 271 * and worry about wraparound (automatic with unsigned arithmetic). 272 */ 273 274 static inline bool before(__u32 seq1, __u32 seq2) 275 { 276 return (__s32)(seq1-seq2) < 0; 277 } 278 #define after(seq2, seq1) before(seq1, seq2) 279 280 /* is s2<=s1<=s3 ? */ 281 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 282 { 283 return seq3 - seq2 >= seq1 - seq2; 284 } 285 286 static inline bool tcp_out_of_memory(struct sock *sk) 287 { 288 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 289 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 290 return true; 291 return false; 292 } 293 294 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 295 { 296 sk_wmem_queued_add(sk, -skb->truesize); 297 if (!skb_zcopy_pure(skb)) 298 sk_mem_uncharge(sk, skb->truesize); 299 else 300 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 301 __kfree_skb(skb); 302 } 303 304 void sk_forced_mem_schedule(struct sock *sk, int size); 305 306 bool tcp_check_oom(struct sock *sk, int shift); 307 308 309 extern struct proto tcp_prot; 310 311 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 312 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 315 316 void tcp_tasklet_init(void); 317 318 int tcp_v4_err(struct sk_buff *skb, u32); 319 320 void tcp_shutdown(struct sock *sk, int how); 321 322 int tcp_v4_early_demux(struct sk_buff *skb); 323 int tcp_v4_rcv(struct sk_buff *skb); 324 325 void tcp_remove_empty_skb(struct sock *sk); 326 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 327 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 328 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 329 size_t size, struct ubuf_info *uarg); 330 void tcp_splice_eof(struct socket *sock); 331 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 332 int tcp_wmem_schedule(struct sock *sk, int copy); 333 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 334 int size_goal); 335 void tcp_release_cb(struct sock *sk); 336 void tcp_wfree(struct sk_buff *skb); 337 void tcp_write_timer_handler(struct sock *sk); 338 void tcp_delack_timer_handler(struct sock *sk); 339 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 340 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 341 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 342 void tcp_rcv_space_adjust(struct sock *sk); 343 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 344 void tcp_twsk_destructor(struct sock *sk); 345 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 346 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 347 struct pipe_inode_info *pipe, size_t len, 348 unsigned int flags); 349 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 350 bool force_schedule); 351 352 static inline void tcp_dec_quickack_mode(struct sock *sk, 353 const unsigned int pkts) 354 { 355 struct inet_connection_sock *icsk = inet_csk(sk); 356 357 if (icsk->icsk_ack.quick) { 358 if (pkts >= icsk->icsk_ack.quick) { 359 icsk->icsk_ack.quick = 0; 360 /* Leaving quickack mode we deflate ATO. */ 361 icsk->icsk_ack.ato = TCP_ATO_MIN; 362 } else 363 icsk->icsk_ack.quick -= pkts; 364 } 365 } 366 367 #define TCP_ECN_OK 1 368 #define TCP_ECN_QUEUE_CWR 2 369 #define TCP_ECN_DEMAND_CWR 4 370 #define TCP_ECN_SEEN 8 371 372 enum tcp_tw_status { 373 TCP_TW_SUCCESS = 0, 374 TCP_TW_RST = 1, 375 TCP_TW_ACK = 2, 376 TCP_TW_SYN = 3 377 }; 378 379 380 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 381 struct sk_buff *skb, 382 const struct tcphdr *th); 383 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 384 struct request_sock *req, bool fastopen, 385 bool *lost_race); 386 int tcp_child_process(struct sock *parent, struct sock *child, 387 struct sk_buff *skb); 388 void tcp_enter_loss(struct sock *sk); 389 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 390 void tcp_clear_retrans(struct tcp_sock *tp); 391 void tcp_update_metrics(struct sock *sk); 392 void tcp_init_metrics(struct sock *sk); 393 void tcp_metrics_init(void); 394 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 395 void __tcp_close(struct sock *sk, long timeout); 396 void tcp_close(struct sock *sk, long timeout); 397 void tcp_init_sock(struct sock *sk); 398 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 399 __poll_t tcp_poll(struct file *file, struct socket *sock, 400 struct poll_table_struct *wait); 401 int do_tcp_getsockopt(struct sock *sk, int level, 402 int optname, sockptr_t optval, sockptr_t optlen); 403 int tcp_getsockopt(struct sock *sk, int level, int optname, 404 char __user *optval, int __user *optlen); 405 bool tcp_bpf_bypass_getsockopt(int level, int optname); 406 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 407 sockptr_t optval, unsigned int optlen); 408 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 409 unsigned int optlen); 410 void tcp_set_keepalive(struct sock *sk, int val); 411 void tcp_syn_ack_timeout(const struct request_sock *req); 412 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 413 int flags, int *addr_len); 414 int tcp_set_rcvlowat(struct sock *sk, int val); 415 int tcp_set_window_clamp(struct sock *sk, int val); 416 void tcp_update_recv_tstamps(struct sk_buff *skb, 417 struct scm_timestamping_internal *tss); 418 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 419 struct scm_timestamping_internal *tss); 420 void tcp_data_ready(struct sock *sk); 421 #ifdef CONFIG_MMU 422 int tcp_mmap(struct file *file, struct socket *sock, 423 struct vm_area_struct *vma); 424 #endif 425 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 426 struct tcp_options_received *opt_rx, 427 int estab, struct tcp_fastopen_cookie *foc); 428 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 429 430 /* 431 * BPF SKB-less helpers 432 */ 433 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 434 struct tcphdr *th, u32 *cookie); 435 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 436 struct tcphdr *th, u32 *cookie); 437 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 438 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 439 const struct tcp_request_sock_ops *af_ops, 440 struct sock *sk, struct tcphdr *th); 441 /* 442 * TCP v4 functions exported for the inet6 API 443 */ 444 445 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 446 void tcp_v4_mtu_reduced(struct sock *sk); 447 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 448 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 449 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 450 struct sock *tcp_create_openreq_child(const struct sock *sk, 451 struct request_sock *req, 452 struct sk_buff *skb); 453 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 454 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 455 struct request_sock *req, 456 struct dst_entry *dst, 457 struct request_sock *req_unhash, 458 bool *own_req); 459 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 460 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 461 int tcp_connect(struct sock *sk); 462 enum tcp_synack_type { 463 TCP_SYNACK_NORMAL, 464 TCP_SYNACK_FASTOPEN, 465 TCP_SYNACK_COOKIE, 466 }; 467 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 468 struct request_sock *req, 469 struct tcp_fastopen_cookie *foc, 470 enum tcp_synack_type synack_type, 471 struct sk_buff *syn_skb); 472 int tcp_disconnect(struct sock *sk, int flags); 473 474 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 475 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 476 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 477 478 /* From syncookies.c */ 479 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 480 struct request_sock *req, 481 struct dst_entry *dst, u32 tsoff); 482 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 483 u32 cookie); 484 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 485 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 486 const struct tcp_request_sock_ops *af_ops, 487 struct sock *sk, struct sk_buff *skb); 488 #ifdef CONFIG_SYN_COOKIES 489 490 /* Syncookies use a monotonic timer which increments every 60 seconds. 491 * This counter is used both as a hash input and partially encoded into 492 * the cookie value. A cookie is only validated further if the delta 493 * between the current counter value and the encoded one is less than this, 494 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 495 * the counter advances immediately after a cookie is generated). 496 */ 497 #define MAX_SYNCOOKIE_AGE 2 498 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 499 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 500 501 /* syncookies: remember time of last synqueue overflow 502 * But do not dirty this field too often (once per second is enough) 503 * It is racy as we do not hold a lock, but race is very minor. 504 */ 505 static inline void tcp_synq_overflow(const struct sock *sk) 506 { 507 unsigned int last_overflow; 508 unsigned int now = jiffies; 509 510 if (sk->sk_reuseport) { 511 struct sock_reuseport *reuse; 512 513 reuse = rcu_dereference(sk->sk_reuseport_cb); 514 if (likely(reuse)) { 515 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 516 if (!time_between32(now, last_overflow, 517 last_overflow + HZ)) 518 WRITE_ONCE(reuse->synq_overflow_ts, now); 519 return; 520 } 521 } 522 523 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 524 if (!time_between32(now, last_overflow, last_overflow + HZ)) 525 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 526 } 527 528 /* syncookies: no recent synqueue overflow on this listening socket? */ 529 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 530 { 531 unsigned int last_overflow; 532 unsigned int now = jiffies; 533 534 if (sk->sk_reuseport) { 535 struct sock_reuseport *reuse; 536 537 reuse = rcu_dereference(sk->sk_reuseport_cb); 538 if (likely(reuse)) { 539 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 540 return !time_between32(now, last_overflow - HZ, 541 last_overflow + 542 TCP_SYNCOOKIE_VALID); 543 } 544 } 545 546 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 547 548 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 549 * then we're under synflood. However, we have to use 550 * 'last_overflow - HZ' as lower bound. That's because a concurrent 551 * tcp_synq_overflow() could update .ts_recent_stamp after we read 552 * jiffies but before we store .ts_recent_stamp into last_overflow, 553 * which could lead to rejecting a valid syncookie. 554 */ 555 return !time_between32(now, last_overflow - HZ, 556 last_overflow + TCP_SYNCOOKIE_VALID); 557 } 558 559 static inline u32 tcp_cookie_time(void) 560 { 561 u64 val = get_jiffies_64(); 562 563 do_div(val, TCP_SYNCOOKIE_PERIOD); 564 return val; 565 } 566 567 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 568 u16 *mssp); 569 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 570 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 571 bool cookie_timestamp_decode(const struct net *net, 572 struct tcp_options_received *opt); 573 bool cookie_ecn_ok(const struct tcp_options_received *opt, 574 const struct net *net, const struct dst_entry *dst); 575 576 /* From net/ipv6/syncookies.c */ 577 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 578 u32 cookie); 579 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 580 581 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 582 const struct tcphdr *th, u16 *mssp); 583 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 584 #endif 585 /* tcp_output.c */ 586 587 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 588 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 589 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 590 int nonagle); 591 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 592 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 593 void tcp_retransmit_timer(struct sock *sk); 594 void tcp_xmit_retransmit_queue(struct sock *); 595 void tcp_simple_retransmit(struct sock *); 596 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 597 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 598 enum tcp_queue { 599 TCP_FRAG_IN_WRITE_QUEUE, 600 TCP_FRAG_IN_RTX_QUEUE, 601 }; 602 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 603 struct sk_buff *skb, u32 len, 604 unsigned int mss_now, gfp_t gfp); 605 606 void tcp_send_probe0(struct sock *); 607 int tcp_write_wakeup(struct sock *, int mib); 608 void tcp_send_fin(struct sock *sk); 609 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 610 int tcp_send_synack(struct sock *); 611 void tcp_push_one(struct sock *, unsigned int mss_now); 612 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 613 void tcp_send_ack(struct sock *sk); 614 void tcp_send_delayed_ack(struct sock *sk); 615 void tcp_send_loss_probe(struct sock *sk); 616 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 617 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 618 const struct sk_buff *next_skb); 619 620 /* tcp_input.c */ 621 void tcp_rearm_rto(struct sock *sk); 622 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 623 void tcp_reset(struct sock *sk, struct sk_buff *skb); 624 void tcp_fin(struct sock *sk); 625 void tcp_check_space(struct sock *sk); 626 void tcp_sack_compress_send_ack(struct sock *sk); 627 628 /* tcp_timer.c */ 629 void tcp_init_xmit_timers(struct sock *); 630 static inline void tcp_clear_xmit_timers(struct sock *sk) 631 { 632 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 633 __sock_put(sk); 634 635 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 636 __sock_put(sk); 637 638 inet_csk_clear_xmit_timers(sk); 639 } 640 641 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 642 unsigned int tcp_current_mss(struct sock *sk); 643 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 644 645 /* Bound MSS / TSO packet size with the half of the window */ 646 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 647 { 648 int cutoff; 649 650 /* When peer uses tiny windows, there is no use in packetizing 651 * to sub-MSS pieces for the sake of SWS or making sure there 652 * are enough packets in the pipe for fast recovery. 653 * 654 * On the other hand, for extremely large MSS devices, handling 655 * smaller than MSS windows in this way does make sense. 656 */ 657 if (tp->max_window > TCP_MSS_DEFAULT) 658 cutoff = (tp->max_window >> 1); 659 else 660 cutoff = tp->max_window; 661 662 if (cutoff && pktsize > cutoff) 663 return max_t(int, cutoff, 68U - tp->tcp_header_len); 664 else 665 return pktsize; 666 } 667 668 /* tcp.c */ 669 void tcp_get_info(struct sock *, struct tcp_info *); 670 671 /* Read 'sendfile()'-style from a TCP socket */ 672 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 673 sk_read_actor_t recv_actor); 674 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 675 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 676 void tcp_read_done(struct sock *sk, size_t len); 677 678 void tcp_initialize_rcv_mss(struct sock *sk); 679 680 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 681 int tcp_mss_to_mtu(struct sock *sk, int mss); 682 void tcp_mtup_init(struct sock *sk); 683 684 static inline void tcp_bound_rto(const struct sock *sk) 685 { 686 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 687 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 688 } 689 690 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 691 { 692 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 693 } 694 695 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 696 { 697 /* mptcp hooks are only on the slow path */ 698 if (sk_is_mptcp((struct sock *)tp)) 699 return; 700 701 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 702 ntohl(TCP_FLAG_ACK) | 703 snd_wnd); 704 } 705 706 static inline void tcp_fast_path_on(struct tcp_sock *tp) 707 { 708 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 709 } 710 711 static inline void tcp_fast_path_check(struct sock *sk) 712 { 713 struct tcp_sock *tp = tcp_sk(sk); 714 715 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 716 tp->rcv_wnd && 717 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 718 !tp->urg_data) 719 tcp_fast_path_on(tp); 720 } 721 722 /* Compute the actual rto_min value */ 723 static inline u32 tcp_rto_min(struct sock *sk) 724 { 725 const struct dst_entry *dst = __sk_dst_get(sk); 726 u32 rto_min = inet_csk(sk)->icsk_rto_min; 727 728 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 729 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 730 return rto_min; 731 } 732 733 static inline u32 tcp_rto_min_us(struct sock *sk) 734 { 735 return jiffies_to_usecs(tcp_rto_min(sk)); 736 } 737 738 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 739 { 740 return dst_metric_locked(dst, RTAX_CC_ALGO); 741 } 742 743 /* Minimum RTT in usec. ~0 means not available. */ 744 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 745 { 746 return minmax_get(&tp->rtt_min); 747 } 748 749 /* Compute the actual receive window we are currently advertising. 750 * Rcv_nxt can be after the window if our peer push more data 751 * than the offered window. 752 */ 753 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 754 { 755 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 756 757 if (win < 0) 758 win = 0; 759 return (u32) win; 760 } 761 762 /* Choose a new window, without checks for shrinking, and without 763 * scaling applied to the result. The caller does these things 764 * if necessary. This is a "raw" window selection. 765 */ 766 u32 __tcp_select_window(struct sock *sk); 767 768 void tcp_send_window_probe(struct sock *sk); 769 770 /* TCP uses 32bit jiffies to save some space. 771 * Note that this is different from tcp_time_stamp, which 772 * historically has been the same until linux-4.13. 773 */ 774 #define tcp_jiffies32 ((u32)jiffies) 775 776 /* 777 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 778 * It is no longer tied to jiffies, but to 1 ms clock. 779 * Note: double check if you want to use tcp_jiffies32 instead of this. 780 */ 781 #define TCP_TS_HZ 1000 782 783 static inline u64 tcp_clock_ns(void) 784 { 785 return ktime_get_ns(); 786 } 787 788 static inline u64 tcp_clock_us(void) 789 { 790 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 791 } 792 793 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 794 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 795 { 796 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 797 } 798 799 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 800 static inline u32 tcp_ns_to_ts(u64 ns) 801 { 802 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 803 } 804 805 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 806 static inline u32 tcp_time_stamp_raw(void) 807 { 808 return tcp_ns_to_ts(tcp_clock_ns()); 809 } 810 811 void tcp_mstamp_refresh(struct tcp_sock *tp); 812 813 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 814 { 815 return max_t(s64, t1 - t0, 0); 816 } 817 818 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 819 { 820 return tcp_ns_to_ts(skb->skb_mstamp_ns); 821 } 822 823 /* provide the departure time in us unit */ 824 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 825 { 826 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 827 } 828 829 830 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 831 832 #define TCPHDR_FIN 0x01 833 #define TCPHDR_SYN 0x02 834 #define TCPHDR_RST 0x04 835 #define TCPHDR_PSH 0x08 836 #define TCPHDR_ACK 0x10 837 #define TCPHDR_URG 0x20 838 #define TCPHDR_ECE 0x40 839 #define TCPHDR_CWR 0x80 840 841 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 842 843 /* This is what the send packet queuing engine uses to pass 844 * TCP per-packet control information to the transmission code. 845 * We also store the host-order sequence numbers in here too. 846 * This is 44 bytes if IPV6 is enabled. 847 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 848 */ 849 struct tcp_skb_cb { 850 __u32 seq; /* Starting sequence number */ 851 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 852 union { 853 /* Note : tcp_tw_isn is used in input path only 854 * (isn chosen by tcp_timewait_state_process()) 855 * 856 * tcp_gso_segs/size are used in write queue only, 857 * cf tcp_skb_pcount()/tcp_skb_mss() 858 */ 859 __u32 tcp_tw_isn; 860 struct { 861 u16 tcp_gso_segs; 862 u16 tcp_gso_size; 863 }; 864 }; 865 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 866 867 __u8 sacked; /* State flags for SACK. */ 868 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 869 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 870 #define TCPCB_LOST 0x04 /* SKB is lost */ 871 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 872 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 873 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 874 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 875 TCPCB_REPAIRED) 876 877 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 878 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 879 eor:1, /* Is skb MSG_EOR marked? */ 880 has_rxtstamp:1, /* SKB has a RX timestamp */ 881 unused:5; 882 __u32 ack_seq; /* Sequence number ACK'd */ 883 union { 884 struct { 885 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 886 /* There is space for up to 24 bytes */ 887 __u32 is_app_limited:1, /* cwnd not fully used? */ 888 delivered_ce:20, 889 unused:11; 890 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 891 __u32 delivered; 892 /* start of send pipeline phase */ 893 u64 first_tx_mstamp; 894 /* when we reached the "delivered" count */ 895 u64 delivered_mstamp; 896 } tx; /* only used for outgoing skbs */ 897 union { 898 struct inet_skb_parm h4; 899 #if IS_ENABLED(CONFIG_IPV6) 900 struct inet6_skb_parm h6; 901 #endif 902 } header; /* For incoming skbs */ 903 }; 904 }; 905 906 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 907 908 extern const struct inet_connection_sock_af_ops ipv4_specific; 909 910 #if IS_ENABLED(CONFIG_IPV6) 911 /* This is the variant of inet6_iif() that must be used by TCP, 912 * as TCP moves IP6CB into a different location in skb->cb[] 913 */ 914 static inline int tcp_v6_iif(const struct sk_buff *skb) 915 { 916 return TCP_SKB_CB(skb)->header.h6.iif; 917 } 918 919 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 920 { 921 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 922 923 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 924 } 925 926 /* TCP_SKB_CB reference means this can not be used from early demux */ 927 static inline int tcp_v6_sdif(const struct sk_buff *skb) 928 { 929 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 930 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 931 return TCP_SKB_CB(skb)->header.h6.iif; 932 #endif 933 return 0; 934 } 935 936 extern const struct inet_connection_sock_af_ops ipv6_specific; 937 938 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 939 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 940 void tcp_v6_early_demux(struct sk_buff *skb); 941 942 #endif 943 944 /* TCP_SKB_CB reference means this can not be used from early demux */ 945 static inline int tcp_v4_sdif(struct sk_buff *skb) 946 { 947 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 948 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 949 return TCP_SKB_CB(skb)->header.h4.iif; 950 #endif 951 return 0; 952 } 953 954 /* Due to TSO, an SKB can be composed of multiple actual 955 * packets. To keep these tracked properly, we use this. 956 */ 957 static inline int tcp_skb_pcount(const struct sk_buff *skb) 958 { 959 return TCP_SKB_CB(skb)->tcp_gso_segs; 960 } 961 962 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 963 { 964 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 965 } 966 967 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 968 { 969 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 970 } 971 972 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 973 static inline int tcp_skb_mss(const struct sk_buff *skb) 974 { 975 return TCP_SKB_CB(skb)->tcp_gso_size; 976 } 977 978 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 979 { 980 return likely(!TCP_SKB_CB(skb)->eor); 981 } 982 983 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 984 const struct sk_buff *from) 985 { 986 return likely(tcp_skb_can_collapse_to(to) && 987 mptcp_skb_can_collapse(to, from) && 988 skb_pure_zcopy_same(to, from)); 989 } 990 991 /* Events passed to congestion control interface */ 992 enum tcp_ca_event { 993 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 994 CA_EVENT_CWND_RESTART, /* congestion window restart */ 995 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 996 CA_EVENT_LOSS, /* loss timeout */ 997 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 998 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 999 }; 1000 1001 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1002 enum tcp_ca_ack_event_flags { 1003 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1004 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1005 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1006 }; 1007 1008 /* 1009 * Interface for adding new TCP congestion control handlers 1010 */ 1011 #define TCP_CA_NAME_MAX 16 1012 #define TCP_CA_MAX 128 1013 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1014 1015 #define TCP_CA_UNSPEC 0 1016 1017 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1018 #define TCP_CONG_NON_RESTRICTED 0x1 1019 /* Requires ECN/ECT set on all packets */ 1020 #define TCP_CONG_NEEDS_ECN 0x2 1021 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1022 1023 union tcp_cc_info; 1024 1025 struct ack_sample { 1026 u32 pkts_acked; 1027 s32 rtt_us; 1028 u32 in_flight; 1029 }; 1030 1031 /* A rate sample measures the number of (original/retransmitted) data 1032 * packets delivered "delivered" over an interval of time "interval_us". 1033 * The tcp_rate.c code fills in the rate sample, and congestion 1034 * control modules that define a cong_control function to run at the end 1035 * of ACK processing can optionally chose to consult this sample when 1036 * setting cwnd and pacing rate. 1037 * A sample is invalid if "delivered" or "interval_us" is negative. 1038 */ 1039 struct rate_sample { 1040 u64 prior_mstamp; /* starting timestamp for interval */ 1041 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1042 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1043 s32 delivered; /* number of packets delivered over interval */ 1044 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1045 long interval_us; /* time for tp->delivered to incr "delivered" */ 1046 u32 snd_interval_us; /* snd interval for delivered packets */ 1047 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1048 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1049 int losses; /* number of packets marked lost upon ACK */ 1050 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1051 u32 prior_in_flight; /* in flight before this ACK */ 1052 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1053 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1054 bool is_retrans; /* is sample from retransmission? */ 1055 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1056 }; 1057 1058 struct tcp_congestion_ops { 1059 /* fast path fields are put first to fill one cache line */ 1060 1061 /* return slow start threshold (required) */ 1062 u32 (*ssthresh)(struct sock *sk); 1063 1064 /* do new cwnd calculation (required) */ 1065 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1066 1067 /* call before changing ca_state (optional) */ 1068 void (*set_state)(struct sock *sk, u8 new_state); 1069 1070 /* call when cwnd event occurs (optional) */ 1071 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1072 1073 /* call when ack arrives (optional) */ 1074 void (*in_ack_event)(struct sock *sk, u32 flags); 1075 1076 /* hook for packet ack accounting (optional) */ 1077 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1078 1079 /* override sysctl_tcp_min_tso_segs */ 1080 u32 (*min_tso_segs)(struct sock *sk); 1081 1082 /* call when packets are delivered to update cwnd and pacing rate, 1083 * after all the ca_state processing. (optional) 1084 */ 1085 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1086 1087 1088 /* new value of cwnd after loss (required) */ 1089 u32 (*undo_cwnd)(struct sock *sk); 1090 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1091 u32 (*sndbuf_expand)(struct sock *sk); 1092 1093 /* control/slow paths put last */ 1094 /* get info for inet_diag (optional) */ 1095 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1096 union tcp_cc_info *info); 1097 1098 char name[TCP_CA_NAME_MAX]; 1099 struct module *owner; 1100 struct list_head list; 1101 u32 key; 1102 u32 flags; 1103 1104 /* initialize private data (optional) */ 1105 void (*init)(struct sock *sk); 1106 /* cleanup private data (optional) */ 1107 void (*release)(struct sock *sk); 1108 } ____cacheline_aligned_in_smp; 1109 1110 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1111 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1112 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1113 struct tcp_congestion_ops *old_type); 1114 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1115 1116 void tcp_assign_congestion_control(struct sock *sk); 1117 void tcp_init_congestion_control(struct sock *sk); 1118 void tcp_cleanup_congestion_control(struct sock *sk); 1119 int tcp_set_default_congestion_control(struct net *net, const char *name); 1120 void tcp_get_default_congestion_control(struct net *net, char *name); 1121 void tcp_get_available_congestion_control(char *buf, size_t len); 1122 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1123 int tcp_set_allowed_congestion_control(char *allowed); 1124 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1125 bool cap_net_admin); 1126 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1127 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1128 1129 u32 tcp_reno_ssthresh(struct sock *sk); 1130 u32 tcp_reno_undo_cwnd(struct sock *sk); 1131 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1132 extern struct tcp_congestion_ops tcp_reno; 1133 1134 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1135 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1136 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1137 #ifdef CONFIG_INET 1138 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1139 #else 1140 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1141 { 1142 return NULL; 1143 } 1144 #endif 1145 1146 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1147 { 1148 const struct inet_connection_sock *icsk = inet_csk(sk); 1149 1150 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1151 } 1152 1153 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1154 { 1155 const struct inet_connection_sock *icsk = inet_csk(sk); 1156 1157 if (icsk->icsk_ca_ops->cwnd_event) 1158 icsk->icsk_ca_ops->cwnd_event(sk, event); 1159 } 1160 1161 /* From tcp_cong.c */ 1162 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1163 1164 /* From tcp_rate.c */ 1165 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1166 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1167 struct rate_sample *rs); 1168 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1169 bool is_sack_reneg, struct rate_sample *rs); 1170 void tcp_rate_check_app_limited(struct sock *sk); 1171 1172 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1173 { 1174 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1175 } 1176 1177 /* These functions determine how the current flow behaves in respect of SACK 1178 * handling. SACK is negotiated with the peer, and therefore it can vary 1179 * between different flows. 1180 * 1181 * tcp_is_sack - SACK enabled 1182 * tcp_is_reno - No SACK 1183 */ 1184 static inline int tcp_is_sack(const struct tcp_sock *tp) 1185 { 1186 return likely(tp->rx_opt.sack_ok); 1187 } 1188 1189 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1190 { 1191 return !tcp_is_sack(tp); 1192 } 1193 1194 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1195 { 1196 return tp->sacked_out + tp->lost_out; 1197 } 1198 1199 /* This determines how many packets are "in the network" to the best 1200 * of our knowledge. In many cases it is conservative, but where 1201 * detailed information is available from the receiver (via SACK 1202 * blocks etc.) we can make more aggressive calculations. 1203 * 1204 * Use this for decisions involving congestion control, use just 1205 * tp->packets_out to determine if the send queue is empty or not. 1206 * 1207 * Read this equation as: 1208 * 1209 * "Packets sent once on transmission queue" MINUS 1210 * "Packets left network, but not honestly ACKed yet" PLUS 1211 * "Packets fast retransmitted" 1212 */ 1213 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1214 { 1215 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1216 } 1217 1218 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1219 1220 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1221 { 1222 return tp->snd_cwnd; 1223 } 1224 1225 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1226 { 1227 WARN_ON_ONCE((int)val <= 0); 1228 tp->snd_cwnd = val; 1229 } 1230 1231 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1232 { 1233 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1234 } 1235 1236 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1237 { 1238 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1239 } 1240 1241 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1242 { 1243 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1244 (1 << inet_csk(sk)->icsk_ca_state); 1245 } 1246 1247 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1248 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1249 * ssthresh. 1250 */ 1251 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1252 { 1253 const struct tcp_sock *tp = tcp_sk(sk); 1254 1255 if (tcp_in_cwnd_reduction(sk)) 1256 return tp->snd_ssthresh; 1257 else 1258 return max(tp->snd_ssthresh, 1259 ((tcp_snd_cwnd(tp) >> 1) + 1260 (tcp_snd_cwnd(tp) >> 2))); 1261 } 1262 1263 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1264 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1265 1266 void tcp_enter_cwr(struct sock *sk); 1267 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1268 1269 /* The maximum number of MSS of available cwnd for which TSO defers 1270 * sending if not using sysctl_tcp_tso_win_divisor. 1271 */ 1272 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1273 { 1274 return 3; 1275 } 1276 1277 /* Returns end sequence number of the receiver's advertised window */ 1278 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1279 { 1280 return tp->snd_una + tp->snd_wnd; 1281 } 1282 1283 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1284 * flexible approach. The RFC suggests cwnd should not be raised unless 1285 * it was fully used previously. And that's exactly what we do in 1286 * congestion avoidance mode. But in slow start we allow cwnd to grow 1287 * as long as the application has used half the cwnd. 1288 * Example : 1289 * cwnd is 10 (IW10), but application sends 9 frames. 1290 * We allow cwnd to reach 18 when all frames are ACKed. 1291 * This check is safe because it's as aggressive as slow start which already 1292 * risks 100% overshoot. The advantage is that we discourage application to 1293 * either send more filler packets or data to artificially blow up the cwnd 1294 * usage, and allow application-limited process to probe bw more aggressively. 1295 */ 1296 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1297 { 1298 const struct tcp_sock *tp = tcp_sk(sk); 1299 1300 if (tp->is_cwnd_limited) 1301 return true; 1302 1303 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1304 if (tcp_in_slow_start(tp)) 1305 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1306 1307 return false; 1308 } 1309 1310 /* BBR congestion control needs pacing. 1311 * Same remark for SO_MAX_PACING_RATE. 1312 * sch_fq packet scheduler is efficiently handling pacing, 1313 * but is not always installed/used. 1314 * Return true if TCP stack should pace packets itself. 1315 */ 1316 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1317 { 1318 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1319 } 1320 1321 /* Estimates in how many jiffies next packet for this flow can be sent. 1322 * Scheduling a retransmit timer too early would be silly. 1323 */ 1324 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1325 { 1326 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1327 1328 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1329 } 1330 1331 static inline void tcp_reset_xmit_timer(struct sock *sk, 1332 const int what, 1333 unsigned long when, 1334 const unsigned long max_when) 1335 { 1336 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1337 max_when); 1338 } 1339 1340 /* Something is really bad, we could not queue an additional packet, 1341 * because qdisc is full or receiver sent a 0 window, or we are paced. 1342 * We do not want to add fuel to the fire, or abort too early, 1343 * so make sure the timer we arm now is at least 200ms in the future, 1344 * regardless of current icsk_rto value (as it could be ~2ms) 1345 */ 1346 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1347 { 1348 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1349 } 1350 1351 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1352 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1353 unsigned long max_when) 1354 { 1355 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1356 inet_csk(sk)->icsk_backoff); 1357 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1358 1359 return (unsigned long)min_t(u64, when, max_when); 1360 } 1361 1362 static inline void tcp_check_probe_timer(struct sock *sk) 1363 { 1364 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1365 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1366 tcp_probe0_base(sk), TCP_RTO_MAX); 1367 } 1368 1369 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1370 { 1371 tp->snd_wl1 = seq; 1372 } 1373 1374 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1375 { 1376 tp->snd_wl1 = seq; 1377 } 1378 1379 /* 1380 * Calculate(/check) TCP checksum 1381 */ 1382 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1383 __be32 daddr, __wsum base) 1384 { 1385 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1386 } 1387 1388 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1389 { 1390 return !skb_csum_unnecessary(skb) && 1391 __skb_checksum_complete(skb); 1392 } 1393 1394 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1395 enum skb_drop_reason *reason); 1396 1397 1398 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1399 void tcp_set_state(struct sock *sk, int state); 1400 void tcp_done(struct sock *sk); 1401 int tcp_abort(struct sock *sk, int err); 1402 1403 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1404 { 1405 rx_opt->dsack = 0; 1406 rx_opt->num_sacks = 0; 1407 } 1408 1409 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1410 1411 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1412 { 1413 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1414 struct tcp_sock *tp = tcp_sk(sk); 1415 s32 delta; 1416 1417 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1418 tp->packets_out || ca_ops->cong_control) 1419 return; 1420 delta = tcp_jiffies32 - tp->lsndtime; 1421 if (delta > inet_csk(sk)->icsk_rto) 1422 tcp_cwnd_restart(sk, delta); 1423 } 1424 1425 /* Determine a window scaling and initial window to offer. */ 1426 void tcp_select_initial_window(const struct sock *sk, int __space, 1427 __u32 mss, __u32 *rcv_wnd, 1428 __u32 *window_clamp, int wscale_ok, 1429 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1430 1431 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1432 { 1433 s64 scaled_space = (s64)space * scaling_ratio; 1434 1435 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1436 } 1437 1438 static inline int tcp_win_from_space(const struct sock *sk, int space) 1439 { 1440 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1441 } 1442 1443 /* inverse of __tcp_win_from_space() */ 1444 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1445 { 1446 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1447 1448 do_div(val, scaling_ratio); 1449 return val; 1450 } 1451 1452 static inline int tcp_space_from_win(const struct sock *sk, int win) 1453 { 1454 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1455 } 1456 1457 static inline void tcp_scaling_ratio_init(struct sock *sk) 1458 { 1459 /* Assume a conservative default of 1200 bytes of payload per 4K page. 1460 * This may be adjusted later in tcp_measure_rcv_mss(). 1461 */ 1462 tcp_sk(sk)->scaling_ratio = (1200 << TCP_RMEM_TO_WIN_SCALE) / 1463 SKB_TRUESIZE(4096); 1464 } 1465 1466 /* Note: caller must be prepared to deal with negative returns */ 1467 static inline int tcp_space(const struct sock *sk) 1468 { 1469 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1470 READ_ONCE(sk->sk_backlog.len) - 1471 atomic_read(&sk->sk_rmem_alloc)); 1472 } 1473 1474 static inline int tcp_full_space(const struct sock *sk) 1475 { 1476 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1477 } 1478 1479 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1480 { 1481 int unused_mem = sk_unused_reserved_mem(sk); 1482 struct tcp_sock *tp = tcp_sk(sk); 1483 1484 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1485 if (unused_mem) 1486 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1487 tcp_win_from_space(sk, unused_mem)); 1488 } 1489 1490 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1491 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1492 1493 1494 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1495 * If 87.5 % (7/8) of the space has been consumed, we want to override 1496 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1497 * len/truesize ratio. 1498 */ 1499 static inline bool tcp_rmem_pressure(const struct sock *sk) 1500 { 1501 int rcvbuf, threshold; 1502 1503 if (tcp_under_memory_pressure(sk)) 1504 return true; 1505 1506 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1507 threshold = rcvbuf - (rcvbuf >> 3); 1508 1509 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1510 } 1511 1512 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1513 { 1514 const struct tcp_sock *tp = tcp_sk(sk); 1515 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1516 1517 if (avail <= 0) 1518 return false; 1519 1520 return (avail >= target) || tcp_rmem_pressure(sk) || 1521 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1522 } 1523 1524 extern void tcp_openreq_init_rwin(struct request_sock *req, 1525 const struct sock *sk_listener, 1526 const struct dst_entry *dst); 1527 1528 void tcp_enter_memory_pressure(struct sock *sk); 1529 void tcp_leave_memory_pressure(struct sock *sk); 1530 1531 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1532 { 1533 struct net *net = sock_net((struct sock *)tp); 1534 int val; 1535 1536 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1537 * and do_tcp_setsockopt(). 1538 */ 1539 val = READ_ONCE(tp->keepalive_intvl); 1540 1541 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1542 } 1543 1544 static inline int keepalive_time_when(const struct tcp_sock *tp) 1545 { 1546 struct net *net = sock_net((struct sock *)tp); 1547 int val; 1548 1549 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1550 val = READ_ONCE(tp->keepalive_time); 1551 1552 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1553 } 1554 1555 static inline int keepalive_probes(const struct tcp_sock *tp) 1556 { 1557 struct net *net = sock_net((struct sock *)tp); 1558 int val; 1559 1560 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1561 * and do_tcp_setsockopt(). 1562 */ 1563 val = READ_ONCE(tp->keepalive_probes); 1564 1565 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1566 } 1567 1568 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1569 { 1570 const struct inet_connection_sock *icsk = &tp->inet_conn; 1571 1572 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1573 tcp_jiffies32 - tp->rcv_tstamp); 1574 } 1575 1576 static inline int tcp_fin_time(const struct sock *sk) 1577 { 1578 int fin_timeout = tcp_sk(sk)->linger2 ? : 1579 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1580 const int rto = inet_csk(sk)->icsk_rto; 1581 1582 if (fin_timeout < (rto << 2) - (rto >> 1)) 1583 fin_timeout = (rto << 2) - (rto >> 1); 1584 1585 return fin_timeout; 1586 } 1587 1588 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1589 int paws_win) 1590 { 1591 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1592 return true; 1593 if (unlikely(!time_before32(ktime_get_seconds(), 1594 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1595 return true; 1596 /* 1597 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1598 * then following tcp messages have valid values. Ignore 0 value, 1599 * or else 'negative' tsval might forbid us to accept their packets. 1600 */ 1601 if (!rx_opt->ts_recent) 1602 return true; 1603 return false; 1604 } 1605 1606 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1607 int rst) 1608 { 1609 if (tcp_paws_check(rx_opt, 0)) 1610 return false; 1611 1612 /* RST segments are not recommended to carry timestamp, 1613 and, if they do, it is recommended to ignore PAWS because 1614 "their cleanup function should take precedence over timestamps." 1615 Certainly, it is mistake. It is necessary to understand the reasons 1616 of this constraint to relax it: if peer reboots, clock may go 1617 out-of-sync and half-open connections will not be reset. 1618 Actually, the problem would be not existing if all 1619 the implementations followed draft about maintaining clock 1620 via reboots. Linux-2.2 DOES NOT! 1621 1622 However, we can relax time bounds for RST segments to MSL. 1623 */ 1624 if (rst && !time_before32(ktime_get_seconds(), 1625 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1626 return false; 1627 return true; 1628 } 1629 1630 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1631 int mib_idx, u32 *last_oow_ack_time); 1632 1633 static inline void tcp_mib_init(struct net *net) 1634 { 1635 /* See RFC 2012 */ 1636 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1637 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1638 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1639 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1640 } 1641 1642 /* from STCP */ 1643 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1644 { 1645 tp->lost_skb_hint = NULL; 1646 } 1647 1648 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1649 { 1650 tcp_clear_retrans_hints_partial(tp); 1651 tp->retransmit_skb_hint = NULL; 1652 } 1653 1654 union tcp_md5_addr { 1655 struct in_addr a4; 1656 #if IS_ENABLED(CONFIG_IPV6) 1657 struct in6_addr a6; 1658 #endif 1659 }; 1660 1661 /* - key database */ 1662 struct tcp_md5sig_key { 1663 struct hlist_node node; 1664 u8 keylen; 1665 u8 family; /* AF_INET or AF_INET6 */ 1666 u8 prefixlen; 1667 u8 flags; 1668 union tcp_md5_addr addr; 1669 int l3index; /* set if key added with L3 scope */ 1670 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1671 struct rcu_head rcu; 1672 }; 1673 1674 /* - sock block */ 1675 struct tcp_md5sig_info { 1676 struct hlist_head head; 1677 struct rcu_head rcu; 1678 }; 1679 1680 /* - pseudo header */ 1681 struct tcp4_pseudohdr { 1682 __be32 saddr; 1683 __be32 daddr; 1684 __u8 pad; 1685 __u8 protocol; 1686 __be16 len; 1687 }; 1688 1689 struct tcp6_pseudohdr { 1690 struct in6_addr saddr; 1691 struct in6_addr daddr; 1692 __be32 len; 1693 __be32 protocol; /* including padding */ 1694 }; 1695 1696 union tcp_md5sum_block { 1697 struct tcp4_pseudohdr ip4; 1698 #if IS_ENABLED(CONFIG_IPV6) 1699 struct tcp6_pseudohdr ip6; 1700 #endif 1701 }; 1702 1703 /* - pool: digest algorithm, hash description and scratch buffer */ 1704 struct tcp_md5sig_pool { 1705 struct ahash_request *md5_req; 1706 void *scratch; 1707 }; 1708 1709 /* - functions */ 1710 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1711 const struct sock *sk, const struct sk_buff *skb); 1712 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1713 int family, u8 prefixlen, int l3index, u8 flags, 1714 const u8 *newkey, u8 newkeylen); 1715 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1716 int family, u8 prefixlen, int l3index, 1717 struct tcp_md5sig_key *key); 1718 1719 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1720 int family, u8 prefixlen, int l3index, u8 flags); 1721 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1722 const struct sock *addr_sk); 1723 1724 #ifdef CONFIG_TCP_MD5SIG 1725 #include <linux/jump_label.h> 1726 extern struct static_key_false_deferred tcp_md5_needed; 1727 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1728 const union tcp_md5_addr *addr, 1729 int family); 1730 static inline struct tcp_md5sig_key * 1731 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1732 const union tcp_md5_addr *addr, int family) 1733 { 1734 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1735 return NULL; 1736 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1737 } 1738 1739 enum skb_drop_reason 1740 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1741 const void *saddr, const void *daddr, 1742 int family, int dif, int sdif); 1743 1744 1745 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1746 #else 1747 static inline struct tcp_md5sig_key * 1748 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1749 const union tcp_md5_addr *addr, int family) 1750 { 1751 return NULL; 1752 } 1753 1754 static inline enum skb_drop_reason 1755 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1756 const void *saddr, const void *daddr, 1757 int family, int dif, int sdif) 1758 { 1759 return SKB_NOT_DROPPED_YET; 1760 } 1761 #define tcp_twsk_md5_key(twsk) NULL 1762 #endif 1763 1764 bool tcp_alloc_md5sig_pool(void); 1765 1766 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1767 static inline void tcp_put_md5sig_pool(void) 1768 { 1769 local_bh_enable(); 1770 } 1771 1772 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1773 unsigned int header_len); 1774 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1775 const struct tcp_md5sig_key *key); 1776 1777 /* From tcp_fastopen.c */ 1778 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1779 struct tcp_fastopen_cookie *cookie); 1780 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1781 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1782 u16 try_exp); 1783 struct tcp_fastopen_request { 1784 /* Fast Open cookie. Size 0 means a cookie request */ 1785 struct tcp_fastopen_cookie cookie; 1786 struct msghdr *data; /* data in MSG_FASTOPEN */ 1787 size_t size; 1788 int copied; /* queued in tcp_connect() */ 1789 struct ubuf_info *uarg; 1790 }; 1791 void tcp_free_fastopen_req(struct tcp_sock *tp); 1792 void tcp_fastopen_destroy_cipher(struct sock *sk); 1793 void tcp_fastopen_ctx_destroy(struct net *net); 1794 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1795 void *primary_key, void *backup_key); 1796 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1797 u64 *key); 1798 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1799 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1800 struct request_sock *req, 1801 struct tcp_fastopen_cookie *foc, 1802 const struct dst_entry *dst); 1803 void tcp_fastopen_init_key_once(struct net *net); 1804 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1805 struct tcp_fastopen_cookie *cookie); 1806 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1807 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1808 #define TCP_FASTOPEN_KEY_MAX 2 1809 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1810 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1811 1812 /* Fastopen key context */ 1813 struct tcp_fastopen_context { 1814 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1815 int num; 1816 struct rcu_head rcu; 1817 }; 1818 1819 void tcp_fastopen_active_disable(struct sock *sk); 1820 bool tcp_fastopen_active_should_disable(struct sock *sk); 1821 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1822 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1823 1824 /* Caller needs to wrap with rcu_read_(un)lock() */ 1825 static inline 1826 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1827 { 1828 struct tcp_fastopen_context *ctx; 1829 1830 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1831 if (!ctx) 1832 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1833 return ctx; 1834 } 1835 1836 static inline 1837 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1838 const struct tcp_fastopen_cookie *orig) 1839 { 1840 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1841 orig->len == foc->len && 1842 !memcmp(orig->val, foc->val, foc->len)) 1843 return true; 1844 return false; 1845 } 1846 1847 static inline 1848 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1849 { 1850 return ctx->num; 1851 } 1852 1853 /* Latencies incurred by various limits for a sender. They are 1854 * chronograph-like stats that are mutually exclusive. 1855 */ 1856 enum tcp_chrono { 1857 TCP_CHRONO_UNSPEC, 1858 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1859 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1860 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1861 __TCP_CHRONO_MAX, 1862 }; 1863 1864 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1865 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1866 1867 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1868 * the same memory storage than skb->destructor/_skb_refdst 1869 */ 1870 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1871 { 1872 skb->destructor = NULL; 1873 skb->_skb_refdst = 0UL; 1874 } 1875 1876 #define tcp_skb_tsorted_save(skb) { \ 1877 unsigned long _save = skb->_skb_refdst; \ 1878 skb->_skb_refdst = 0UL; 1879 1880 #define tcp_skb_tsorted_restore(skb) \ 1881 skb->_skb_refdst = _save; \ 1882 } 1883 1884 void tcp_write_queue_purge(struct sock *sk); 1885 1886 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1887 { 1888 return skb_rb_first(&sk->tcp_rtx_queue); 1889 } 1890 1891 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1892 { 1893 return skb_rb_last(&sk->tcp_rtx_queue); 1894 } 1895 1896 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1897 { 1898 return skb_peek_tail(&sk->sk_write_queue); 1899 } 1900 1901 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1902 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1903 1904 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1905 { 1906 return skb_peek(&sk->sk_write_queue); 1907 } 1908 1909 static inline bool tcp_skb_is_last(const struct sock *sk, 1910 const struct sk_buff *skb) 1911 { 1912 return skb_queue_is_last(&sk->sk_write_queue, skb); 1913 } 1914 1915 /** 1916 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1917 * @sk: socket 1918 * 1919 * Since the write queue can have a temporary empty skb in it, 1920 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1921 */ 1922 static inline bool tcp_write_queue_empty(const struct sock *sk) 1923 { 1924 const struct tcp_sock *tp = tcp_sk(sk); 1925 1926 return tp->write_seq == tp->snd_nxt; 1927 } 1928 1929 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1930 { 1931 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1932 } 1933 1934 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1935 { 1936 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1937 } 1938 1939 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1940 { 1941 __skb_queue_tail(&sk->sk_write_queue, skb); 1942 1943 /* Queue it, remembering where we must start sending. */ 1944 if (sk->sk_write_queue.next == skb) 1945 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1946 } 1947 1948 /* Insert new before skb on the write queue of sk. */ 1949 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1950 struct sk_buff *skb, 1951 struct sock *sk) 1952 { 1953 __skb_queue_before(&sk->sk_write_queue, skb, new); 1954 } 1955 1956 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1957 { 1958 tcp_skb_tsorted_anchor_cleanup(skb); 1959 __skb_unlink(skb, &sk->sk_write_queue); 1960 } 1961 1962 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1963 1964 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1965 { 1966 tcp_skb_tsorted_anchor_cleanup(skb); 1967 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1968 } 1969 1970 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1971 { 1972 list_del(&skb->tcp_tsorted_anchor); 1973 tcp_rtx_queue_unlink(skb, sk); 1974 tcp_wmem_free_skb(sk, skb); 1975 } 1976 1977 static inline void tcp_push_pending_frames(struct sock *sk) 1978 { 1979 if (tcp_send_head(sk)) { 1980 struct tcp_sock *tp = tcp_sk(sk); 1981 1982 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1983 } 1984 } 1985 1986 /* Start sequence of the skb just after the highest skb with SACKed 1987 * bit, valid only if sacked_out > 0 or when the caller has ensured 1988 * validity by itself. 1989 */ 1990 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1991 { 1992 if (!tp->sacked_out) 1993 return tp->snd_una; 1994 1995 if (tp->highest_sack == NULL) 1996 return tp->snd_nxt; 1997 1998 return TCP_SKB_CB(tp->highest_sack)->seq; 1999 } 2000 2001 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2002 { 2003 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2004 } 2005 2006 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2007 { 2008 return tcp_sk(sk)->highest_sack; 2009 } 2010 2011 static inline void tcp_highest_sack_reset(struct sock *sk) 2012 { 2013 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2014 } 2015 2016 /* Called when old skb is about to be deleted and replaced by new skb */ 2017 static inline void tcp_highest_sack_replace(struct sock *sk, 2018 struct sk_buff *old, 2019 struct sk_buff *new) 2020 { 2021 if (old == tcp_highest_sack(sk)) 2022 tcp_sk(sk)->highest_sack = new; 2023 } 2024 2025 /* This helper checks if socket has IP_TRANSPARENT set */ 2026 static inline bool inet_sk_transparent(const struct sock *sk) 2027 { 2028 switch (sk->sk_state) { 2029 case TCP_TIME_WAIT: 2030 return inet_twsk(sk)->tw_transparent; 2031 case TCP_NEW_SYN_RECV: 2032 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2033 } 2034 return inet_sk(sk)->transparent; 2035 } 2036 2037 /* Determines whether this is a thin stream (which may suffer from 2038 * increased latency). Used to trigger latency-reducing mechanisms. 2039 */ 2040 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2041 { 2042 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2043 } 2044 2045 /* /proc */ 2046 enum tcp_seq_states { 2047 TCP_SEQ_STATE_LISTENING, 2048 TCP_SEQ_STATE_ESTABLISHED, 2049 }; 2050 2051 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2052 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2053 void tcp_seq_stop(struct seq_file *seq, void *v); 2054 2055 struct tcp_seq_afinfo { 2056 sa_family_t family; 2057 }; 2058 2059 struct tcp_iter_state { 2060 struct seq_net_private p; 2061 enum tcp_seq_states state; 2062 struct sock *syn_wait_sk; 2063 int bucket, offset, sbucket, num; 2064 loff_t last_pos; 2065 }; 2066 2067 extern struct request_sock_ops tcp_request_sock_ops; 2068 extern struct request_sock_ops tcp6_request_sock_ops; 2069 2070 void tcp_v4_destroy_sock(struct sock *sk); 2071 2072 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2073 netdev_features_t features); 2074 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2075 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2076 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2077 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2078 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2079 void tcp_gro_complete(struct sk_buff *skb); 2080 2081 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2082 2083 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2084 { 2085 struct net *net = sock_net((struct sock *)tp); 2086 u32 val; 2087 2088 val = READ_ONCE(tp->notsent_lowat); 2089 2090 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2091 } 2092 2093 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2094 2095 #ifdef CONFIG_PROC_FS 2096 int tcp4_proc_init(void); 2097 void tcp4_proc_exit(void); 2098 #endif 2099 2100 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2101 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2102 const struct tcp_request_sock_ops *af_ops, 2103 struct sock *sk, struct sk_buff *skb); 2104 2105 /* TCP af-specific functions */ 2106 struct tcp_sock_af_ops { 2107 #ifdef CONFIG_TCP_MD5SIG 2108 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2109 const struct sock *addr_sk); 2110 int (*calc_md5_hash)(char *location, 2111 const struct tcp_md5sig_key *md5, 2112 const struct sock *sk, 2113 const struct sk_buff *skb); 2114 int (*md5_parse)(struct sock *sk, 2115 int optname, 2116 sockptr_t optval, 2117 int optlen); 2118 #endif 2119 }; 2120 2121 struct tcp_request_sock_ops { 2122 u16 mss_clamp; 2123 #ifdef CONFIG_TCP_MD5SIG 2124 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2125 const struct sock *addr_sk); 2126 int (*calc_md5_hash) (char *location, 2127 const struct tcp_md5sig_key *md5, 2128 const struct sock *sk, 2129 const struct sk_buff *skb); 2130 #endif 2131 #ifdef CONFIG_SYN_COOKIES 2132 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2133 __u16 *mss); 2134 #endif 2135 struct dst_entry *(*route_req)(const struct sock *sk, 2136 struct sk_buff *skb, 2137 struct flowi *fl, 2138 struct request_sock *req); 2139 u32 (*init_seq)(const struct sk_buff *skb); 2140 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2141 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2142 struct flowi *fl, struct request_sock *req, 2143 struct tcp_fastopen_cookie *foc, 2144 enum tcp_synack_type synack_type, 2145 struct sk_buff *syn_skb); 2146 }; 2147 2148 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2149 #if IS_ENABLED(CONFIG_IPV6) 2150 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2151 #endif 2152 2153 #ifdef CONFIG_SYN_COOKIES 2154 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2155 const struct sock *sk, struct sk_buff *skb, 2156 __u16 *mss) 2157 { 2158 tcp_synq_overflow(sk); 2159 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2160 return ops->cookie_init_seq(skb, mss); 2161 } 2162 #else 2163 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2164 const struct sock *sk, struct sk_buff *skb, 2165 __u16 *mss) 2166 { 2167 return 0; 2168 } 2169 #endif 2170 2171 int tcpv4_offload_init(void); 2172 2173 void tcp_v4_init(void); 2174 void tcp_init(void); 2175 2176 /* tcp_recovery.c */ 2177 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2178 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2179 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2180 u32 reo_wnd); 2181 extern bool tcp_rack_mark_lost(struct sock *sk); 2182 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2183 u64 xmit_time); 2184 extern void tcp_rack_reo_timeout(struct sock *sk); 2185 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2186 2187 /* tcp_plb.c */ 2188 2189 /* 2190 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2191 * expects cong_ratio which represents fraction of traffic that experienced 2192 * congestion over a single RTT. In order to avoid floating point operations, 2193 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2194 */ 2195 #define TCP_PLB_SCALE 8 2196 2197 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2198 struct tcp_plb_state { 2199 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2200 unused:3; 2201 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2202 }; 2203 2204 static inline void tcp_plb_init(const struct sock *sk, 2205 struct tcp_plb_state *plb) 2206 { 2207 plb->consec_cong_rounds = 0; 2208 plb->pause_until = 0; 2209 } 2210 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2211 const int cong_ratio); 2212 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2213 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2214 2215 /* At how many usecs into the future should the RTO fire? */ 2216 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2217 { 2218 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2219 u32 rto = inet_csk(sk)->icsk_rto; 2220 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2221 2222 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2223 } 2224 2225 /* 2226 * Save and compile IPv4 options, return a pointer to it 2227 */ 2228 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2229 struct sk_buff *skb) 2230 { 2231 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2232 struct ip_options_rcu *dopt = NULL; 2233 2234 if (opt->optlen) { 2235 int opt_size = sizeof(*dopt) + opt->optlen; 2236 2237 dopt = kmalloc(opt_size, GFP_ATOMIC); 2238 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2239 kfree(dopt); 2240 dopt = NULL; 2241 } 2242 } 2243 return dopt; 2244 } 2245 2246 /* locally generated TCP pure ACKs have skb->truesize == 2 2247 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2248 * This is much faster than dissecting the packet to find out. 2249 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2250 */ 2251 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2252 { 2253 return skb->truesize == 2; 2254 } 2255 2256 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2257 { 2258 skb->truesize = 2; 2259 } 2260 2261 static inline int tcp_inq(struct sock *sk) 2262 { 2263 struct tcp_sock *tp = tcp_sk(sk); 2264 int answ; 2265 2266 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2267 answ = 0; 2268 } else if (sock_flag(sk, SOCK_URGINLINE) || 2269 !tp->urg_data || 2270 before(tp->urg_seq, tp->copied_seq) || 2271 !before(tp->urg_seq, tp->rcv_nxt)) { 2272 2273 answ = tp->rcv_nxt - tp->copied_seq; 2274 2275 /* Subtract 1, if FIN was received */ 2276 if (answ && sock_flag(sk, SOCK_DONE)) 2277 answ--; 2278 } else { 2279 answ = tp->urg_seq - tp->copied_seq; 2280 } 2281 2282 return answ; 2283 } 2284 2285 int tcp_peek_len(struct socket *sock); 2286 2287 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2288 { 2289 u16 segs_in; 2290 2291 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2292 2293 /* We update these fields while other threads might 2294 * read them from tcp_get_info() 2295 */ 2296 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2297 if (skb->len > tcp_hdrlen(skb)) 2298 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2299 } 2300 2301 /* 2302 * TCP listen path runs lockless. 2303 * We forced "struct sock" to be const qualified to make sure 2304 * we don't modify one of its field by mistake. 2305 * Here, we increment sk_drops which is an atomic_t, so we can safely 2306 * make sock writable again. 2307 */ 2308 static inline void tcp_listendrop(const struct sock *sk) 2309 { 2310 atomic_inc(&((struct sock *)sk)->sk_drops); 2311 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2312 } 2313 2314 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2315 2316 /* 2317 * Interface for adding Upper Level Protocols over TCP 2318 */ 2319 2320 #define TCP_ULP_NAME_MAX 16 2321 #define TCP_ULP_MAX 128 2322 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2323 2324 struct tcp_ulp_ops { 2325 struct list_head list; 2326 2327 /* initialize ulp */ 2328 int (*init)(struct sock *sk); 2329 /* update ulp */ 2330 void (*update)(struct sock *sk, struct proto *p, 2331 void (*write_space)(struct sock *sk)); 2332 /* cleanup ulp */ 2333 void (*release)(struct sock *sk); 2334 /* diagnostic */ 2335 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2336 size_t (*get_info_size)(const struct sock *sk); 2337 /* clone ulp */ 2338 void (*clone)(const struct request_sock *req, struct sock *newsk, 2339 const gfp_t priority); 2340 2341 char name[TCP_ULP_NAME_MAX]; 2342 struct module *owner; 2343 }; 2344 int tcp_register_ulp(struct tcp_ulp_ops *type); 2345 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2346 int tcp_set_ulp(struct sock *sk, const char *name); 2347 void tcp_get_available_ulp(char *buf, size_t len); 2348 void tcp_cleanup_ulp(struct sock *sk); 2349 void tcp_update_ulp(struct sock *sk, struct proto *p, 2350 void (*write_space)(struct sock *sk)); 2351 2352 #define MODULE_ALIAS_TCP_ULP(name) \ 2353 __MODULE_INFO(alias, alias_userspace, name); \ 2354 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2355 2356 #ifdef CONFIG_NET_SOCK_MSG 2357 struct sk_msg; 2358 struct sk_psock; 2359 2360 #ifdef CONFIG_BPF_SYSCALL 2361 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2362 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2363 #endif /* CONFIG_BPF_SYSCALL */ 2364 2365 #ifdef CONFIG_INET 2366 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2367 #else 2368 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2369 { 2370 } 2371 #endif 2372 2373 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2374 struct sk_msg *msg, u32 bytes, int flags); 2375 #endif /* CONFIG_NET_SOCK_MSG */ 2376 2377 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2378 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2379 { 2380 } 2381 #endif 2382 2383 #ifdef CONFIG_CGROUP_BPF 2384 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2385 struct sk_buff *skb, 2386 unsigned int end_offset) 2387 { 2388 skops->skb = skb; 2389 skops->skb_data_end = skb->data + end_offset; 2390 } 2391 #else 2392 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2393 struct sk_buff *skb, 2394 unsigned int end_offset) 2395 { 2396 } 2397 #endif 2398 2399 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2400 * is < 0, then the BPF op failed (for example if the loaded BPF 2401 * program does not support the chosen operation or there is no BPF 2402 * program loaded). 2403 */ 2404 #ifdef CONFIG_BPF 2405 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2406 { 2407 struct bpf_sock_ops_kern sock_ops; 2408 int ret; 2409 2410 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2411 if (sk_fullsock(sk)) { 2412 sock_ops.is_fullsock = 1; 2413 sock_owned_by_me(sk); 2414 } 2415 2416 sock_ops.sk = sk; 2417 sock_ops.op = op; 2418 if (nargs > 0) 2419 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2420 2421 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2422 if (ret == 0) 2423 ret = sock_ops.reply; 2424 else 2425 ret = -1; 2426 return ret; 2427 } 2428 2429 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2430 { 2431 u32 args[2] = {arg1, arg2}; 2432 2433 return tcp_call_bpf(sk, op, 2, args); 2434 } 2435 2436 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2437 u32 arg3) 2438 { 2439 u32 args[3] = {arg1, arg2, arg3}; 2440 2441 return tcp_call_bpf(sk, op, 3, args); 2442 } 2443 2444 #else 2445 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2446 { 2447 return -EPERM; 2448 } 2449 2450 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2451 { 2452 return -EPERM; 2453 } 2454 2455 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2456 u32 arg3) 2457 { 2458 return -EPERM; 2459 } 2460 2461 #endif 2462 2463 static inline u32 tcp_timeout_init(struct sock *sk) 2464 { 2465 int timeout; 2466 2467 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2468 2469 if (timeout <= 0) 2470 timeout = TCP_TIMEOUT_INIT; 2471 return min_t(int, timeout, TCP_RTO_MAX); 2472 } 2473 2474 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2475 { 2476 int rwnd; 2477 2478 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2479 2480 if (rwnd < 0) 2481 rwnd = 0; 2482 return rwnd; 2483 } 2484 2485 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2486 { 2487 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2488 } 2489 2490 static inline void tcp_bpf_rtt(struct sock *sk) 2491 { 2492 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2493 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2494 } 2495 2496 #if IS_ENABLED(CONFIG_SMC) 2497 extern struct static_key_false tcp_have_smc; 2498 #endif 2499 2500 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2501 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2502 void (*cad)(struct sock *sk, u32 ack_seq)); 2503 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2504 void clean_acked_data_flush(void); 2505 #endif 2506 2507 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2508 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2509 const struct tcp_sock *tp) 2510 { 2511 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2512 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2513 } 2514 2515 /* Compute Earliest Departure Time for some control packets 2516 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2517 */ 2518 static inline u64 tcp_transmit_time(const struct sock *sk) 2519 { 2520 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2521 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2522 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2523 2524 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2525 } 2526 return 0; 2527 } 2528 2529 #endif /* _TCP_H */ 2530