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