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_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 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 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 331 int flags); 332 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 333 size_t size, int flags); 334 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 335 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 336 int size_goal); 337 void tcp_release_cb(struct sock *sk); 338 void tcp_wfree(struct sk_buff *skb); 339 void tcp_write_timer_handler(struct sock *sk); 340 void tcp_delack_timer_handler(struct sock *sk); 341 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 342 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 344 void tcp_rcv_space_adjust(struct sock *sk); 345 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 346 void tcp_twsk_destructor(struct sock *sk); 347 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 348 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 349 struct pipe_inode_info *pipe, size_t len, 350 unsigned int flags); 351 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 352 bool force_schedule); 353 354 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 355 static inline void tcp_dec_quickack_mode(struct sock *sk, 356 const unsigned int pkts) 357 { 358 struct inet_connection_sock *icsk = inet_csk(sk); 359 360 if (icsk->icsk_ack.quick) { 361 if (pkts >= icsk->icsk_ack.quick) { 362 icsk->icsk_ack.quick = 0; 363 /* Leaving quickack mode we deflate ATO. */ 364 icsk->icsk_ack.ato = TCP_ATO_MIN; 365 } else 366 icsk->icsk_ack.quick -= pkts; 367 } 368 } 369 370 #define TCP_ECN_OK 1 371 #define TCP_ECN_QUEUE_CWR 2 372 #define TCP_ECN_DEMAND_CWR 4 373 #define TCP_ECN_SEEN 8 374 375 enum tcp_tw_status { 376 TCP_TW_SUCCESS = 0, 377 TCP_TW_RST = 1, 378 TCP_TW_ACK = 2, 379 TCP_TW_SYN = 3 380 }; 381 382 383 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 384 struct sk_buff *skb, 385 const struct tcphdr *th); 386 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 387 struct request_sock *req, bool fastopen, 388 bool *lost_race); 389 int tcp_child_process(struct sock *parent, struct sock *child, 390 struct sk_buff *skb); 391 void tcp_enter_loss(struct sock *sk); 392 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 393 void tcp_clear_retrans(struct tcp_sock *tp); 394 void tcp_update_metrics(struct sock *sk); 395 void tcp_init_metrics(struct sock *sk); 396 void tcp_metrics_init(void); 397 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 398 void __tcp_close(struct sock *sk, long timeout); 399 void tcp_close(struct sock *sk, long timeout); 400 void tcp_init_sock(struct sock *sk); 401 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 402 __poll_t tcp_poll(struct file *file, struct socket *sock, 403 struct poll_table_struct *wait); 404 int do_tcp_getsockopt(struct sock *sk, int level, 405 int optname, sockptr_t optval, sockptr_t optlen); 406 int tcp_getsockopt(struct sock *sk, int level, int optname, 407 char __user *optval, int __user *optlen); 408 bool tcp_bpf_bypass_getsockopt(int level, int optname); 409 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 410 sockptr_t optval, unsigned int optlen); 411 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 412 unsigned int optlen); 413 void tcp_set_keepalive(struct sock *sk, int val); 414 void tcp_syn_ack_timeout(const struct request_sock *req); 415 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 416 int flags, int *addr_len); 417 int tcp_set_rcvlowat(struct sock *sk, int val); 418 int tcp_set_window_clamp(struct sock *sk, int val); 419 void tcp_update_recv_tstamps(struct sk_buff *skb, 420 struct scm_timestamping_internal *tss); 421 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 422 struct scm_timestamping_internal *tss); 423 void tcp_data_ready(struct sock *sk); 424 #ifdef CONFIG_MMU 425 int tcp_mmap(struct file *file, struct socket *sock, 426 struct vm_area_struct *vma); 427 #endif 428 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 429 struct tcp_options_received *opt_rx, 430 int estab, struct tcp_fastopen_cookie *foc); 431 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 432 433 /* 434 * BPF SKB-less helpers 435 */ 436 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 437 struct tcphdr *th, u32 *cookie); 438 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 439 struct tcphdr *th, u32 *cookie); 440 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 441 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 442 const struct tcp_request_sock_ops *af_ops, 443 struct sock *sk, struct tcphdr *th); 444 /* 445 * TCP v4 functions exported for the inet6 API 446 */ 447 448 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 449 void tcp_v4_mtu_reduced(struct sock *sk); 450 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 451 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 452 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 453 struct sock *tcp_create_openreq_child(const struct sock *sk, 454 struct request_sock *req, 455 struct sk_buff *skb); 456 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 457 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 458 struct request_sock *req, 459 struct dst_entry *dst, 460 struct request_sock *req_unhash, 461 bool *own_req); 462 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 463 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 464 int tcp_connect(struct sock *sk); 465 enum tcp_synack_type { 466 TCP_SYNACK_NORMAL, 467 TCP_SYNACK_FASTOPEN, 468 TCP_SYNACK_COOKIE, 469 }; 470 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 471 struct request_sock *req, 472 struct tcp_fastopen_cookie *foc, 473 enum tcp_synack_type synack_type, 474 struct sk_buff *syn_skb); 475 int tcp_disconnect(struct sock *sk, int flags); 476 477 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 478 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 479 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 480 481 /* From syncookies.c */ 482 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 483 struct request_sock *req, 484 struct dst_entry *dst, u32 tsoff); 485 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 486 u32 cookie); 487 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 488 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 489 const struct tcp_request_sock_ops *af_ops, 490 struct sock *sk, struct sk_buff *skb); 491 #ifdef CONFIG_SYN_COOKIES 492 493 /* Syncookies use a monotonic timer which increments every 60 seconds. 494 * This counter is used both as a hash input and partially encoded into 495 * the cookie value. A cookie is only validated further if the delta 496 * between the current counter value and the encoded one is less than this, 497 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 498 * the counter advances immediately after a cookie is generated). 499 */ 500 #define MAX_SYNCOOKIE_AGE 2 501 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 502 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 503 504 /* syncookies: remember time of last synqueue overflow 505 * But do not dirty this field too often (once per second is enough) 506 * It is racy as we do not hold a lock, but race is very minor. 507 */ 508 static inline void tcp_synq_overflow(const struct sock *sk) 509 { 510 unsigned int last_overflow; 511 unsigned int now = jiffies; 512 513 if (sk->sk_reuseport) { 514 struct sock_reuseport *reuse; 515 516 reuse = rcu_dereference(sk->sk_reuseport_cb); 517 if (likely(reuse)) { 518 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 519 if (!time_between32(now, last_overflow, 520 last_overflow + HZ)) 521 WRITE_ONCE(reuse->synq_overflow_ts, now); 522 return; 523 } 524 } 525 526 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 527 if (!time_between32(now, last_overflow, last_overflow + HZ)) 528 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 529 } 530 531 /* syncookies: no recent synqueue overflow on this listening socket? */ 532 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 533 { 534 unsigned int last_overflow; 535 unsigned int now = jiffies; 536 537 if (sk->sk_reuseport) { 538 struct sock_reuseport *reuse; 539 540 reuse = rcu_dereference(sk->sk_reuseport_cb); 541 if (likely(reuse)) { 542 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 543 return !time_between32(now, last_overflow - HZ, 544 last_overflow + 545 TCP_SYNCOOKIE_VALID); 546 } 547 } 548 549 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 550 551 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 552 * then we're under synflood. However, we have to use 553 * 'last_overflow - HZ' as lower bound. That's because a concurrent 554 * tcp_synq_overflow() could update .ts_recent_stamp after we read 555 * jiffies but before we store .ts_recent_stamp into last_overflow, 556 * which could lead to rejecting a valid syncookie. 557 */ 558 return !time_between32(now, last_overflow - HZ, 559 last_overflow + TCP_SYNCOOKIE_VALID); 560 } 561 562 static inline u32 tcp_cookie_time(void) 563 { 564 u64 val = get_jiffies_64(); 565 566 do_div(val, TCP_SYNCOOKIE_PERIOD); 567 return val; 568 } 569 570 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 571 u16 *mssp); 572 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 573 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 574 bool cookie_timestamp_decode(const struct net *net, 575 struct tcp_options_received *opt); 576 bool cookie_ecn_ok(const struct tcp_options_received *opt, 577 const struct net *net, const struct dst_entry *dst); 578 579 /* From net/ipv6/syncookies.c */ 580 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 581 u32 cookie); 582 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 583 584 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 585 const struct tcphdr *th, u16 *mssp); 586 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 587 #endif 588 /* tcp_output.c */ 589 590 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 591 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 592 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 593 int nonagle); 594 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 595 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 596 void tcp_retransmit_timer(struct sock *sk); 597 void tcp_xmit_retransmit_queue(struct sock *); 598 void tcp_simple_retransmit(struct sock *); 599 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 600 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 601 enum tcp_queue { 602 TCP_FRAG_IN_WRITE_QUEUE, 603 TCP_FRAG_IN_RTX_QUEUE, 604 }; 605 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 606 struct sk_buff *skb, u32 len, 607 unsigned int mss_now, gfp_t gfp); 608 609 void tcp_send_probe0(struct sock *); 610 void tcp_send_partial(struct sock *); 611 int tcp_write_wakeup(struct sock *, int mib); 612 void tcp_send_fin(struct sock *sk); 613 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 614 int tcp_send_synack(struct sock *); 615 void tcp_push_one(struct sock *, unsigned int mss_now); 616 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 617 void tcp_send_ack(struct sock *sk); 618 void tcp_send_delayed_ack(struct sock *sk); 619 void tcp_send_loss_probe(struct sock *sk); 620 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 621 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 622 const struct sk_buff *next_skb); 623 624 /* tcp_input.c */ 625 void tcp_rearm_rto(struct sock *sk); 626 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 627 void tcp_reset(struct sock *sk, struct sk_buff *skb); 628 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 629 void tcp_fin(struct sock *sk); 630 void tcp_check_space(struct sock *sk); 631 void tcp_sack_compress_send_ack(struct sock *sk); 632 633 /* tcp_timer.c */ 634 void tcp_init_xmit_timers(struct sock *); 635 static inline void tcp_clear_xmit_timers(struct sock *sk) 636 { 637 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 638 __sock_put(sk); 639 640 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 641 __sock_put(sk); 642 643 inet_csk_clear_xmit_timers(sk); 644 } 645 646 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 647 unsigned int tcp_current_mss(struct sock *sk); 648 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 649 650 /* Bound MSS / TSO packet size with the half of the window */ 651 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 652 { 653 int cutoff; 654 655 /* When peer uses tiny windows, there is no use in packetizing 656 * to sub-MSS pieces for the sake of SWS or making sure there 657 * are enough packets in the pipe for fast recovery. 658 * 659 * On the other hand, for extremely large MSS devices, handling 660 * smaller than MSS windows in this way does make sense. 661 */ 662 if (tp->max_window > TCP_MSS_DEFAULT) 663 cutoff = (tp->max_window >> 1); 664 else 665 cutoff = tp->max_window; 666 667 if (cutoff && pktsize > cutoff) 668 return max_t(int, cutoff, 68U - tp->tcp_header_len); 669 else 670 return pktsize; 671 } 672 673 /* tcp.c */ 674 void tcp_get_info(struct sock *, struct tcp_info *); 675 676 /* Read 'sendfile()'-style from a TCP socket */ 677 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 678 sk_read_actor_t recv_actor); 679 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 680 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 681 void tcp_read_done(struct sock *sk, size_t len); 682 683 void tcp_initialize_rcv_mss(struct sock *sk); 684 685 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 686 int tcp_mss_to_mtu(struct sock *sk, int mss); 687 void tcp_mtup_init(struct sock *sk); 688 689 static inline void tcp_bound_rto(const struct sock *sk) 690 { 691 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 692 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 693 } 694 695 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 696 { 697 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 698 } 699 700 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 701 { 702 /* mptcp hooks are only on the slow path */ 703 if (sk_is_mptcp((struct sock *)tp)) 704 return; 705 706 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 707 ntohl(TCP_FLAG_ACK) | 708 snd_wnd); 709 } 710 711 static inline void tcp_fast_path_on(struct tcp_sock *tp) 712 { 713 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 714 } 715 716 static inline void tcp_fast_path_check(struct sock *sk) 717 { 718 struct tcp_sock *tp = tcp_sk(sk); 719 720 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 721 tp->rcv_wnd && 722 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 723 !tp->urg_data) 724 tcp_fast_path_on(tp); 725 } 726 727 /* Compute the actual rto_min value */ 728 static inline u32 tcp_rto_min(struct sock *sk) 729 { 730 const struct dst_entry *dst = __sk_dst_get(sk); 731 u32 rto_min = inet_csk(sk)->icsk_rto_min; 732 733 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 734 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 735 return rto_min; 736 } 737 738 static inline u32 tcp_rto_min_us(struct sock *sk) 739 { 740 return jiffies_to_usecs(tcp_rto_min(sk)); 741 } 742 743 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 744 { 745 return dst_metric_locked(dst, RTAX_CC_ALGO); 746 } 747 748 /* Minimum RTT in usec. ~0 means not available. */ 749 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 750 { 751 return minmax_get(&tp->rtt_min); 752 } 753 754 /* Compute the actual receive window we are currently advertising. 755 * Rcv_nxt can be after the window if our peer push more data 756 * than the offered window. 757 */ 758 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 759 { 760 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 761 762 if (win < 0) 763 win = 0; 764 return (u32) win; 765 } 766 767 /* Choose a new window, without checks for shrinking, and without 768 * scaling applied to the result. The caller does these things 769 * if necessary. This is a "raw" window selection. 770 */ 771 u32 __tcp_select_window(struct sock *sk); 772 773 void tcp_send_window_probe(struct sock *sk); 774 775 /* TCP uses 32bit jiffies to save some space. 776 * Note that this is different from tcp_time_stamp, which 777 * historically has been the same until linux-4.13. 778 */ 779 #define tcp_jiffies32 ((u32)jiffies) 780 781 /* 782 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 783 * It is no longer tied to jiffies, but to 1 ms clock. 784 * Note: double check if you want to use tcp_jiffies32 instead of this. 785 */ 786 #define TCP_TS_HZ 1000 787 788 static inline u64 tcp_clock_ns(void) 789 { 790 return ktime_get_ns(); 791 } 792 793 static inline u64 tcp_clock_us(void) 794 { 795 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 796 } 797 798 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 799 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 800 { 801 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 802 } 803 804 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 805 static inline u32 tcp_ns_to_ts(u64 ns) 806 { 807 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 808 } 809 810 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 811 static inline u32 tcp_time_stamp_raw(void) 812 { 813 return tcp_ns_to_ts(tcp_clock_ns()); 814 } 815 816 void tcp_mstamp_refresh(struct tcp_sock *tp); 817 818 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 819 { 820 return max_t(s64, t1 - t0, 0); 821 } 822 823 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 824 { 825 return tcp_ns_to_ts(skb->skb_mstamp_ns); 826 } 827 828 /* provide the departure time in us unit */ 829 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 830 { 831 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 832 } 833 834 835 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 836 837 #define TCPHDR_FIN 0x01 838 #define TCPHDR_SYN 0x02 839 #define TCPHDR_RST 0x04 840 #define TCPHDR_PSH 0x08 841 #define TCPHDR_ACK 0x10 842 #define TCPHDR_URG 0x20 843 #define TCPHDR_ECE 0x40 844 #define TCPHDR_CWR 0x80 845 846 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 847 848 /* This is what the send packet queuing engine uses to pass 849 * TCP per-packet control information to the transmission code. 850 * We also store the host-order sequence numbers in here too. 851 * This is 44 bytes if IPV6 is enabled. 852 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 853 */ 854 struct tcp_skb_cb { 855 __u32 seq; /* Starting sequence number */ 856 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 857 union { 858 /* Note : tcp_tw_isn is used in input path only 859 * (isn chosen by tcp_timewait_state_process()) 860 * 861 * tcp_gso_segs/size are used in write queue only, 862 * cf tcp_skb_pcount()/tcp_skb_mss() 863 */ 864 __u32 tcp_tw_isn; 865 struct { 866 u16 tcp_gso_segs; 867 u16 tcp_gso_size; 868 }; 869 }; 870 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 871 872 __u8 sacked; /* State flags for SACK. */ 873 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 874 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 875 #define TCPCB_LOST 0x04 /* SKB is lost */ 876 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 877 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 878 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 879 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 880 TCPCB_REPAIRED) 881 882 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 883 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 884 eor:1, /* Is skb MSG_EOR marked? */ 885 has_rxtstamp:1, /* SKB has a RX timestamp */ 886 unused:5; 887 __u32 ack_seq; /* Sequence number ACK'd */ 888 union { 889 struct { 890 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 891 /* There is space for up to 24 bytes */ 892 __u32 is_app_limited:1, /* cwnd not fully used? */ 893 delivered_ce:20, 894 unused:11; 895 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 896 __u32 delivered; 897 /* start of send pipeline phase */ 898 u64 first_tx_mstamp; 899 /* when we reached the "delivered" count */ 900 u64 delivered_mstamp; 901 } tx; /* only used for outgoing skbs */ 902 union { 903 struct inet_skb_parm h4; 904 #if IS_ENABLED(CONFIG_IPV6) 905 struct inet6_skb_parm h6; 906 #endif 907 } header; /* For incoming skbs */ 908 }; 909 }; 910 911 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 912 913 extern const struct inet_connection_sock_af_ops ipv4_specific; 914 915 #if IS_ENABLED(CONFIG_IPV6) 916 /* This is the variant of inet6_iif() that must be used by TCP, 917 * as TCP moves IP6CB into a different location in skb->cb[] 918 */ 919 static inline int tcp_v6_iif(const struct sk_buff *skb) 920 { 921 return TCP_SKB_CB(skb)->header.h6.iif; 922 } 923 924 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 925 { 926 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 927 928 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 929 } 930 931 /* TCP_SKB_CB reference means this can not be used from early demux */ 932 static inline int tcp_v6_sdif(const struct sk_buff *skb) 933 { 934 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 935 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 936 return TCP_SKB_CB(skb)->header.h6.iif; 937 #endif 938 return 0; 939 } 940 941 extern const struct inet_connection_sock_af_ops ipv6_specific; 942 943 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 944 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 945 void tcp_v6_early_demux(struct sk_buff *skb); 946 947 #endif 948 949 /* TCP_SKB_CB reference means this can not be used from early demux */ 950 static inline int tcp_v4_sdif(struct sk_buff *skb) 951 { 952 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 953 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 954 return TCP_SKB_CB(skb)->header.h4.iif; 955 #endif 956 return 0; 957 } 958 959 /* Due to TSO, an SKB can be composed of multiple actual 960 * packets. To keep these tracked properly, we use this. 961 */ 962 static inline int tcp_skb_pcount(const struct sk_buff *skb) 963 { 964 return TCP_SKB_CB(skb)->tcp_gso_segs; 965 } 966 967 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 968 { 969 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 970 } 971 972 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 973 { 974 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 975 } 976 977 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 978 static inline int tcp_skb_mss(const struct sk_buff *skb) 979 { 980 return TCP_SKB_CB(skb)->tcp_gso_size; 981 } 982 983 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 984 { 985 return likely(!TCP_SKB_CB(skb)->eor); 986 } 987 988 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 989 const struct sk_buff *from) 990 { 991 return likely(tcp_skb_can_collapse_to(to) && 992 mptcp_skb_can_collapse(to, from) && 993 skb_pure_zcopy_same(to, from)); 994 } 995 996 /* Events passed to congestion control interface */ 997 enum tcp_ca_event { 998 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 999 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1000 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1001 CA_EVENT_LOSS, /* loss timeout */ 1002 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1003 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1004 }; 1005 1006 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1007 enum tcp_ca_ack_event_flags { 1008 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1009 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1010 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1011 }; 1012 1013 /* 1014 * Interface for adding new TCP congestion control handlers 1015 */ 1016 #define TCP_CA_NAME_MAX 16 1017 #define TCP_CA_MAX 128 1018 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1019 1020 #define TCP_CA_UNSPEC 0 1021 1022 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1023 #define TCP_CONG_NON_RESTRICTED 0x1 1024 /* Requires ECN/ECT set on all packets */ 1025 #define TCP_CONG_NEEDS_ECN 0x2 1026 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1027 1028 union tcp_cc_info; 1029 1030 struct ack_sample { 1031 u32 pkts_acked; 1032 s32 rtt_us; 1033 u32 in_flight; 1034 }; 1035 1036 /* A rate sample measures the number of (original/retransmitted) data 1037 * packets delivered "delivered" over an interval of time "interval_us". 1038 * The tcp_rate.c code fills in the rate sample, and congestion 1039 * control modules that define a cong_control function to run at the end 1040 * of ACK processing can optionally chose to consult this sample when 1041 * setting cwnd and pacing rate. 1042 * A sample is invalid if "delivered" or "interval_us" is negative. 1043 */ 1044 struct rate_sample { 1045 u64 prior_mstamp; /* starting timestamp for interval */ 1046 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1047 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1048 s32 delivered; /* number of packets delivered over interval */ 1049 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1050 long interval_us; /* time for tp->delivered to incr "delivered" */ 1051 u32 snd_interval_us; /* snd interval for delivered packets */ 1052 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1053 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1054 int losses; /* number of packets marked lost upon ACK */ 1055 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1056 u32 prior_in_flight; /* in flight before this ACK */ 1057 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1058 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1059 bool is_retrans; /* is sample from retransmission? */ 1060 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1061 }; 1062 1063 struct tcp_congestion_ops { 1064 /* fast path fields are put first to fill one cache line */ 1065 1066 /* return slow start threshold (required) */ 1067 u32 (*ssthresh)(struct sock *sk); 1068 1069 /* do new cwnd calculation (required) */ 1070 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1071 1072 /* call before changing ca_state (optional) */ 1073 void (*set_state)(struct sock *sk, u8 new_state); 1074 1075 /* call when cwnd event occurs (optional) */ 1076 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1077 1078 /* call when ack arrives (optional) */ 1079 void (*in_ack_event)(struct sock *sk, u32 flags); 1080 1081 /* hook for packet ack accounting (optional) */ 1082 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1083 1084 /* override sysctl_tcp_min_tso_segs */ 1085 u32 (*min_tso_segs)(struct sock *sk); 1086 1087 /* call when packets are delivered to update cwnd and pacing rate, 1088 * after all the ca_state processing. (optional) 1089 */ 1090 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1091 1092 1093 /* new value of cwnd after loss (required) */ 1094 u32 (*undo_cwnd)(struct sock *sk); 1095 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1096 u32 (*sndbuf_expand)(struct sock *sk); 1097 1098 /* control/slow paths put last */ 1099 /* get info for inet_diag (optional) */ 1100 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1101 union tcp_cc_info *info); 1102 1103 char name[TCP_CA_NAME_MAX]; 1104 struct module *owner; 1105 struct list_head list; 1106 u32 key; 1107 u32 flags; 1108 1109 /* initialize private data (optional) */ 1110 void (*init)(struct sock *sk); 1111 /* cleanup private data (optional) */ 1112 void (*release)(struct sock *sk); 1113 } ____cacheline_aligned_in_smp; 1114 1115 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1116 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1117 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1118 struct tcp_congestion_ops *old_type); 1119 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1120 1121 void tcp_assign_congestion_control(struct sock *sk); 1122 void tcp_init_congestion_control(struct sock *sk); 1123 void tcp_cleanup_congestion_control(struct sock *sk); 1124 int tcp_set_default_congestion_control(struct net *net, const char *name); 1125 void tcp_get_default_congestion_control(struct net *net, char *name); 1126 void tcp_get_available_congestion_control(char *buf, size_t len); 1127 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1128 int tcp_set_allowed_congestion_control(char *allowed); 1129 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1130 bool cap_net_admin); 1131 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1132 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1133 1134 u32 tcp_reno_ssthresh(struct sock *sk); 1135 u32 tcp_reno_undo_cwnd(struct sock *sk); 1136 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1137 extern struct tcp_congestion_ops tcp_reno; 1138 1139 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1140 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1141 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1142 #ifdef CONFIG_INET 1143 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1144 #else 1145 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1146 { 1147 return NULL; 1148 } 1149 #endif 1150 1151 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1152 { 1153 const struct inet_connection_sock *icsk = inet_csk(sk); 1154 1155 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1156 } 1157 1158 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1159 { 1160 const struct inet_connection_sock *icsk = inet_csk(sk); 1161 1162 if (icsk->icsk_ca_ops->cwnd_event) 1163 icsk->icsk_ca_ops->cwnd_event(sk, event); 1164 } 1165 1166 /* From tcp_cong.c */ 1167 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1168 1169 /* From tcp_rate.c */ 1170 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1171 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1172 struct rate_sample *rs); 1173 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1174 bool is_sack_reneg, struct rate_sample *rs); 1175 void tcp_rate_check_app_limited(struct sock *sk); 1176 1177 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1178 { 1179 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1180 } 1181 1182 /* These functions determine how the current flow behaves in respect of SACK 1183 * handling. SACK is negotiated with the peer, and therefore it can vary 1184 * between different flows. 1185 * 1186 * tcp_is_sack - SACK enabled 1187 * tcp_is_reno - No SACK 1188 */ 1189 static inline int tcp_is_sack(const struct tcp_sock *tp) 1190 { 1191 return likely(tp->rx_opt.sack_ok); 1192 } 1193 1194 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1195 { 1196 return !tcp_is_sack(tp); 1197 } 1198 1199 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1200 { 1201 return tp->sacked_out + tp->lost_out; 1202 } 1203 1204 /* This determines how many packets are "in the network" to the best 1205 * of our knowledge. In many cases it is conservative, but where 1206 * detailed information is available from the receiver (via SACK 1207 * blocks etc.) we can make more aggressive calculations. 1208 * 1209 * Use this for decisions involving congestion control, use just 1210 * tp->packets_out to determine if the send queue is empty or not. 1211 * 1212 * Read this equation as: 1213 * 1214 * "Packets sent once on transmission queue" MINUS 1215 * "Packets left network, but not honestly ACKed yet" PLUS 1216 * "Packets fast retransmitted" 1217 */ 1218 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1219 { 1220 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1221 } 1222 1223 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1224 1225 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1226 { 1227 return tp->snd_cwnd; 1228 } 1229 1230 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1231 { 1232 WARN_ON_ONCE((int)val <= 0); 1233 tp->snd_cwnd = val; 1234 } 1235 1236 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1237 { 1238 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1239 } 1240 1241 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1242 { 1243 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1244 } 1245 1246 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1247 { 1248 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1249 (1 << inet_csk(sk)->icsk_ca_state); 1250 } 1251 1252 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1253 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1254 * ssthresh. 1255 */ 1256 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1257 { 1258 const struct tcp_sock *tp = tcp_sk(sk); 1259 1260 if (tcp_in_cwnd_reduction(sk)) 1261 return tp->snd_ssthresh; 1262 else 1263 return max(tp->snd_ssthresh, 1264 ((tcp_snd_cwnd(tp) >> 1) + 1265 (tcp_snd_cwnd(tp) >> 2))); 1266 } 1267 1268 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1269 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1270 1271 void tcp_enter_cwr(struct sock *sk); 1272 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1273 1274 /* The maximum number of MSS of available cwnd for which TSO defers 1275 * sending if not using sysctl_tcp_tso_win_divisor. 1276 */ 1277 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1278 { 1279 return 3; 1280 } 1281 1282 /* Returns end sequence number of the receiver's advertised window */ 1283 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1284 { 1285 return tp->snd_una + tp->snd_wnd; 1286 } 1287 1288 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1289 * flexible approach. The RFC suggests cwnd should not be raised unless 1290 * it was fully used previously. And that's exactly what we do in 1291 * congestion avoidance mode. But in slow start we allow cwnd to grow 1292 * as long as the application has used half the cwnd. 1293 * Example : 1294 * cwnd is 10 (IW10), but application sends 9 frames. 1295 * We allow cwnd to reach 18 when all frames are ACKed. 1296 * This check is safe because it's as aggressive as slow start which already 1297 * risks 100% overshoot. The advantage is that we discourage application to 1298 * either send more filler packets or data to artificially blow up the cwnd 1299 * usage, and allow application-limited process to probe bw more aggressively. 1300 */ 1301 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1302 { 1303 const struct tcp_sock *tp = tcp_sk(sk); 1304 1305 if (tp->is_cwnd_limited) 1306 return true; 1307 1308 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1309 if (tcp_in_slow_start(tp)) 1310 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1311 1312 return false; 1313 } 1314 1315 /* BBR congestion control needs pacing. 1316 * Same remark for SO_MAX_PACING_RATE. 1317 * sch_fq packet scheduler is efficiently handling pacing, 1318 * but is not always installed/used. 1319 * Return true if TCP stack should pace packets itself. 1320 */ 1321 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1322 { 1323 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1324 } 1325 1326 /* Estimates in how many jiffies next packet for this flow can be sent. 1327 * Scheduling a retransmit timer too early would be silly. 1328 */ 1329 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1330 { 1331 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1332 1333 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1334 } 1335 1336 static inline void tcp_reset_xmit_timer(struct sock *sk, 1337 const int what, 1338 unsigned long when, 1339 const unsigned long max_when) 1340 { 1341 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1342 max_when); 1343 } 1344 1345 /* Something is really bad, we could not queue an additional packet, 1346 * because qdisc is full or receiver sent a 0 window, or we are paced. 1347 * We do not want to add fuel to the fire, or abort too early, 1348 * so make sure the timer we arm now is at least 200ms in the future, 1349 * regardless of current icsk_rto value (as it could be ~2ms) 1350 */ 1351 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1352 { 1353 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1354 } 1355 1356 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1357 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1358 unsigned long max_when) 1359 { 1360 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1361 inet_csk(sk)->icsk_backoff); 1362 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1363 1364 return (unsigned long)min_t(u64, when, max_when); 1365 } 1366 1367 static inline void tcp_check_probe_timer(struct sock *sk) 1368 { 1369 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1370 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1371 tcp_probe0_base(sk), TCP_RTO_MAX); 1372 } 1373 1374 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1375 { 1376 tp->snd_wl1 = seq; 1377 } 1378 1379 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1380 { 1381 tp->snd_wl1 = seq; 1382 } 1383 1384 /* 1385 * Calculate(/check) TCP checksum 1386 */ 1387 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1388 __be32 daddr, __wsum base) 1389 { 1390 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1391 } 1392 1393 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1394 { 1395 return !skb_csum_unnecessary(skb) && 1396 __skb_checksum_complete(skb); 1397 } 1398 1399 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1400 enum skb_drop_reason *reason); 1401 1402 1403 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1404 void tcp_set_state(struct sock *sk, int state); 1405 void tcp_done(struct sock *sk); 1406 int tcp_abort(struct sock *sk, int err); 1407 1408 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1409 { 1410 rx_opt->dsack = 0; 1411 rx_opt->num_sacks = 0; 1412 } 1413 1414 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1415 1416 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1417 { 1418 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1419 struct tcp_sock *tp = tcp_sk(sk); 1420 s32 delta; 1421 1422 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1423 tp->packets_out || ca_ops->cong_control) 1424 return; 1425 delta = tcp_jiffies32 - tp->lsndtime; 1426 if (delta > inet_csk(sk)->icsk_rto) 1427 tcp_cwnd_restart(sk, delta); 1428 } 1429 1430 /* Determine a window scaling and initial window to offer. */ 1431 void tcp_select_initial_window(const struct sock *sk, int __space, 1432 __u32 mss, __u32 *rcv_wnd, 1433 __u32 *window_clamp, int wscale_ok, 1434 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1435 1436 static inline int tcp_win_from_space(const struct sock *sk, int space) 1437 { 1438 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale); 1439 1440 return tcp_adv_win_scale <= 0 ? 1441 (space>>(-tcp_adv_win_scale)) : 1442 space - (space>>tcp_adv_win_scale); 1443 } 1444 1445 /* Note: caller must be prepared to deal with negative returns */ 1446 static inline int tcp_space(const struct sock *sk) 1447 { 1448 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1449 READ_ONCE(sk->sk_backlog.len) - 1450 atomic_read(&sk->sk_rmem_alloc)); 1451 } 1452 1453 static inline int tcp_full_space(const struct sock *sk) 1454 { 1455 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1456 } 1457 1458 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1459 { 1460 int unused_mem = sk_unused_reserved_mem(sk); 1461 struct tcp_sock *tp = tcp_sk(sk); 1462 1463 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1464 if (unused_mem) 1465 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1466 tcp_win_from_space(sk, unused_mem)); 1467 } 1468 1469 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1470 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1471 1472 1473 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1474 * If 87.5 % (7/8) of the space has been consumed, we want to override 1475 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1476 * len/truesize ratio. 1477 */ 1478 static inline bool tcp_rmem_pressure(const struct sock *sk) 1479 { 1480 int rcvbuf, threshold; 1481 1482 if (tcp_under_memory_pressure(sk)) 1483 return true; 1484 1485 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1486 threshold = rcvbuf - (rcvbuf >> 3); 1487 1488 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1489 } 1490 1491 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1492 { 1493 const struct tcp_sock *tp = tcp_sk(sk); 1494 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1495 1496 if (avail <= 0) 1497 return false; 1498 1499 return (avail >= target) || tcp_rmem_pressure(sk) || 1500 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1501 } 1502 1503 extern void tcp_openreq_init_rwin(struct request_sock *req, 1504 const struct sock *sk_listener, 1505 const struct dst_entry *dst); 1506 1507 void tcp_enter_memory_pressure(struct sock *sk); 1508 void tcp_leave_memory_pressure(struct sock *sk); 1509 1510 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1511 { 1512 struct net *net = sock_net((struct sock *)tp); 1513 1514 return tp->keepalive_intvl ? : 1515 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1516 } 1517 1518 static inline int keepalive_time_when(const struct tcp_sock *tp) 1519 { 1520 struct net *net = sock_net((struct sock *)tp); 1521 1522 return tp->keepalive_time ? : 1523 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1524 } 1525 1526 static inline int keepalive_probes(const struct tcp_sock *tp) 1527 { 1528 struct net *net = sock_net((struct sock *)tp); 1529 1530 return tp->keepalive_probes ? : 1531 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1532 } 1533 1534 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1535 { 1536 const struct inet_connection_sock *icsk = &tp->inet_conn; 1537 1538 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1539 tcp_jiffies32 - tp->rcv_tstamp); 1540 } 1541 1542 static inline int tcp_fin_time(const struct sock *sk) 1543 { 1544 int fin_timeout = tcp_sk(sk)->linger2 ? : 1545 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1546 const int rto = inet_csk(sk)->icsk_rto; 1547 1548 if (fin_timeout < (rto << 2) - (rto >> 1)) 1549 fin_timeout = (rto << 2) - (rto >> 1); 1550 1551 return fin_timeout; 1552 } 1553 1554 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1555 int paws_win) 1556 { 1557 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1558 return true; 1559 if (unlikely(!time_before32(ktime_get_seconds(), 1560 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1561 return true; 1562 /* 1563 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1564 * then following tcp messages have valid values. Ignore 0 value, 1565 * or else 'negative' tsval might forbid us to accept their packets. 1566 */ 1567 if (!rx_opt->ts_recent) 1568 return true; 1569 return false; 1570 } 1571 1572 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1573 int rst) 1574 { 1575 if (tcp_paws_check(rx_opt, 0)) 1576 return false; 1577 1578 /* RST segments are not recommended to carry timestamp, 1579 and, if they do, it is recommended to ignore PAWS because 1580 "their cleanup function should take precedence over timestamps." 1581 Certainly, it is mistake. It is necessary to understand the reasons 1582 of this constraint to relax it: if peer reboots, clock may go 1583 out-of-sync and half-open connections will not be reset. 1584 Actually, the problem would be not existing if all 1585 the implementations followed draft about maintaining clock 1586 via reboots. Linux-2.2 DOES NOT! 1587 1588 However, we can relax time bounds for RST segments to MSL. 1589 */ 1590 if (rst && !time_before32(ktime_get_seconds(), 1591 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1592 return false; 1593 return true; 1594 } 1595 1596 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1597 int mib_idx, u32 *last_oow_ack_time); 1598 1599 static inline void tcp_mib_init(struct net *net) 1600 { 1601 /* See RFC 2012 */ 1602 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1603 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1604 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1605 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1606 } 1607 1608 /* from STCP */ 1609 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1610 { 1611 tp->lost_skb_hint = NULL; 1612 } 1613 1614 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1615 { 1616 tcp_clear_retrans_hints_partial(tp); 1617 tp->retransmit_skb_hint = NULL; 1618 } 1619 1620 union tcp_md5_addr { 1621 struct in_addr a4; 1622 #if IS_ENABLED(CONFIG_IPV6) 1623 struct in6_addr a6; 1624 #endif 1625 }; 1626 1627 /* - key database */ 1628 struct tcp_md5sig_key { 1629 struct hlist_node node; 1630 u8 keylen; 1631 u8 family; /* AF_INET or AF_INET6 */ 1632 u8 prefixlen; 1633 u8 flags; 1634 union tcp_md5_addr addr; 1635 int l3index; /* set if key added with L3 scope */ 1636 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1637 struct rcu_head rcu; 1638 }; 1639 1640 /* - sock block */ 1641 struct tcp_md5sig_info { 1642 struct hlist_head head; 1643 struct rcu_head rcu; 1644 }; 1645 1646 /* - pseudo header */ 1647 struct tcp4_pseudohdr { 1648 __be32 saddr; 1649 __be32 daddr; 1650 __u8 pad; 1651 __u8 protocol; 1652 __be16 len; 1653 }; 1654 1655 struct tcp6_pseudohdr { 1656 struct in6_addr saddr; 1657 struct in6_addr daddr; 1658 __be32 len; 1659 __be32 protocol; /* including padding */ 1660 }; 1661 1662 union tcp_md5sum_block { 1663 struct tcp4_pseudohdr ip4; 1664 #if IS_ENABLED(CONFIG_IPV6) 1665 struct tcp6_pseudohdr ip6; 1666 #endif 1667 }; 1668 1669 /* - pool: digest algorithm, hash description and scratch buffer */ 1670 struct tcp_md5sig_pool { 1671 struct ahash_request *md5_req; 1672 void *scratch; 1673 }; 1674 1675 /* - functions */ 1676 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1677 const struct sock *sk, const struct sk_buff *skb); 1678 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1679 int family, u8 prefixlen, int l3index, u8 flags, 1680 const u8 *newkey, u8 newkeylen); 1681 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1682 int family, u8 prefixlen, int l3index, 1683 struct tcp_md5sig_key *key); 1684 1685 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1686 int family, u8 prefixlen, int l3index, u8 flags); 1687 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1688 const struct sock *addr_sk); 1689 1690 #ifdef CONFIG_TCP_MD5SIG 1691 #include <linux/jump_label.h> 1692 extern struct static_key_false_deferred tcp_md5_needed; 1693 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1694 const union tcp_md5_addr *addr, 1695 int family); 1696 static inline struct tcp_md5sig_key * 1697 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1698 const union tcp_md5_addr *addr, int family) 1699 { 1700 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1701 return NULL; 1702 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1703 } 1704 1705 enum skb_drop_reason 1706 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1707 const void *saddr, const void *daddr, 1708 int family, int dif, int sdif); 1709 1710 1711 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1712 #else 1713 static inline struct tcp_md5sig_key * 1714 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1715 const union tcp_md5_addr *addr, int family) 1716 { 1717 return NULL; 1718 } 1719 1720 static inline enum skb_drop_reason 1721 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1722 const void *saddr, const void *daddr, 1723 int family, int dif, int sdif) 1724 { 1725 return SKB_NOT_DROPPED_YET; 1726 } 1727 #define tcp_twsk_md5_key(twsk) NULL 1728 #endif 1729 1730 bool tcp_alloc_md5sig_pool(void); 1731 1732 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1733 static inline void tcp_put_md5sig_pool(void) 1734 { 1735 local_bh_enable(); 1736 } 1737 1738 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1739 unsigned int header_len); 1740 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1741 const struct tcp_md5sig_key *key); 1742 1743 /* From tcp_fastopen.c */ 1744 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1745 struct tcp_fastopen_cookie *cookie); 1746 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1747 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1748 u16 try_exp); 1749 struct tcp_fastopen_request { 1750 /* Fast Open cookie. Size 0 means a cookie request */ 1751 struct tcp_fastopen_cookie cookie; 1752 struct msghdr *data; /* data in MSG_FASTOPEN */ 1753 size_t size; 1754 int copied; /* queued in tcp_connect() */ 1755 struct ubuf_info *uarg; 1756 }; 1757 void tcp_free_fastopen_req(struct tcp_sock *tp); 1758 void tcp_fastopen_destroy_cipher(struct sock *sk); 1759 void tcp_fastopen_ctx_destroy(struct net *net); 1760 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1761 void *primary_key, void *backup_key); 1762 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1763 u64 *key); 1764 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1765 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1766 struct request_sock *req, 1767 struct tcp_fastopen_cookie *foc, 1768 const struct dst_entry *dst); 1769 void tcp_fastopen_init_key_once(struct net *net); 1770 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1771 struct tcp_fastopen_cookie *cookie); 1772 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1773 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1774 #define TCP_FASTOPEN_KEY_MAX 2 1775 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1776 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1777 1778 /* Fastopen key context */ 1779 struct tcp_fastopen_context { 1780 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1781 int num; 1782 struct rcu_head rcu; 1783 }; 1784 1785 void tcp_fastopen_active_disable(struct sock *sk); 1786 bool tcp_fastopen_active_should_disable(struct sock *sk); 1787 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1788 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1789 1790 /* Caller needs to wrap with rcu_read_(un)lock() */ 1791 static inline 1792 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1793 { 1794 struct tcp_fastopen_context *ctx; 1795 1796 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1797 if (!ctx) 1798 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1799 return ctx; 1800 } 1801 1802 static inline 1803 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1804 const struct tcp_fastopen_cookie *orig) 1805 { 1806 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1807 orig->len == foc->len && 1808 !memcmp(orig->val, foc->val, foc->len)) 1809 return true; 1810 return false; 1811 } 1812 1813 static inline 1814 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1815 { 1816 return ctx->num; 1817 } 1818 1819 /* Latencies incurred by various limits for a sender. They are 1820 * chronograph-like stats that are mutually exclusive. 1821 */ 1822 enum tcp_chrono { 1823 TCP_CHRONO_UNSPEC, 1824 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1825 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1826 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1827 __TCP_CHRONO_MAX, 1828 }; 1829 1830 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1831 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1832 1833 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1834 * the same memory storage than skb->destructor/_skb_refdst 1835 */ 1836 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1837 { 1838 skb->destructor = NULL; 1839 skb->_skb_refdst = 0UL; 1840 } 1841 1842 #define tcp_skb_tsorted_save(skb) { \ 1843 unsigned long _save = skb->_skb_refdst; \ 1844 skb->_skb_refdst = 0UL; 1845 1846 #define tcp_skb_tsorted_restore(skb) \ 1847 skb->_skb_refdst = _save; \ 1848 } 1849 1850 void tcp_write_queue_purge(struct sock *sk); 1851 1852 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1853 { 1854 return skb_rb_first(&sk->tcp_rtx_queue); 1855 } 1856 1857 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1858 { 1859 return skb_rb_last(&sk->tcp_rtx_queue); 1860 } 1861 1862 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1863 { 1864 return skb_peek_tail(&sk->sk_write_queue); 1865 } 1866 1867 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1868 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1869 1870 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1871 { 1872 return skb_peek(&sk->sk_write_queue); 1873 } 1874 1875 static inline bool tcp_skb_is_last(const struct sock *sk, 1876 const struct sk_buff *skb) 1877 { 1878 return skb_queue_is_last(&sk->sk_write_queue, skb); 1879 } 1880 1881 /** 1882 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1883 * @sk: socket 1884 * 1885 * Since the write queue can have a temporary empty skb in it, 1886 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1887 */ 1888 static inline bool tcp_write_queue_empty(const struct sock *sk) 1889 { 1890 const struct tcp_sock *tp = tcp_sk(sk); 1891 1892 return tp->write_seq == tp->snd_nxt; 1893 } 1894 1895 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1896 { 1897 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1898 } 1899 1900 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1901 { 1902 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1903 } 1904 1905 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1906 { 1907 __skb_queue_tail(&sk->sk_write_queue, skb); 1908 1909 /* Queue it, remembering where we must start sending. */ 1910 if (sk->sk_write_queue.next == skb) 1911 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1912 } 1913 1914 /* Insert new before skb on the write queue of sk. */ 1915 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1916 struct sk_buff *skb, 1917 struct sock *sk) 1918 { 1919 __skb_queue_before(&sk->sk_write_queue, skb, new); 1920 } 1921 1922 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1923 { 1924 tcp_skb_tsorted_anchor_cleanup(skb); 1925 __skb_unlink(skb, &sk->sk_write_queue); 1926 } 1927 1928 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1929 1930 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1931 { 1932 tcp_skb_tsorted_anchor_cleanup(skb); 1933 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1934 } 1935 1936 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1937 { 1938 list_del(&skb->tcp_tsorted_anchor); 1939 tcp_rtx_queue_unlink(skb, sk); 1940 tcp_wmem_free_skb(sk, skb); 1941 } 1942 1943 static inline void tcp_push_pending_frames(struct sock *sk) 1944 { 1945 if (tcp_send_head(sk)) { 1946 struct tcp_sock *tp = tcp_sk(sk); 1947 1948 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1949 } 1950 } 1951 1952 /* Start sequence of the skb just after the highest skb with SACKed 1953 * bit, valid only if sacked_out > 0 or when the caller has ensured 1954 * validity by itself. 1955 */ 1956 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1957 { 1958 if (!tp->sacked_out) 1959 return tp->snd_una; 1960 1961 if (tp->highest_sack == NULL) 1962 return tp->snd_nxt; 1963 1964 return TCP_SKB_CB(tp->highest_sack)->seq; 1965 } 1966 1967 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1968 { 1969 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1970 } 1971 1972 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1973 { 1974 return tcp_sk(sk)->highest_sack; 1975 } 1976 1977 static inline void tcp_highest_sack_reset(struct sock *sk) 1978 { 1979 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1980 } 1981 1982 /* Called when old skb is about to be deleted and replaced by new skb */ 1983 static inline void tcp_highest_sack_replace(struct sock *sk, 1984 struct sk_buff *old, 1985 struct sk_buff *new) 1986 { 1987 if (old == tcp_highest_sack(sk)) 1988 tcp_sk(sk)->highest_sack = new; 1989 } 1990 1991 /* This helper checks if socket has IP_TRANSPARENT set */ 1992 static inline bool inet_sk_transparent(const struct sock *sk) 1993 { 1994 switch (sk->sk_state) { 1995 case TCP_TIME_WAIT: 1996 return inet_twsk(sk)->tw_transparent; 1997 case TCP_NEW_SYN_RECV: 1998 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1999 } 2000 return inet_sk(sk)->transparent; 2001 } 2002 2003 /* Determines whether this is a thin stream (which may suffer from 2004 * increased latency). Used to trigger latency-reducing mechanisms. 2005 */ 2006 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2007 { 2008 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2009 } 2010 2011 /* /proc */ 2012 enum tcp_seq_states { 2013 TCP_SEQ_STATE_LISTENING, 2014 TCP_SEQ_STATE_ESTABLISHED, 2015 }; 2016 2017 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2018 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2019 void tcp_seq_stop(struct seq_file *seq, void *v); 2020 2021 struct tcp_seq_afinfo { 2022 sa_family_t family; 2023 }; 2024 2025 struct tcp_iter_state { 2026 struct seq_net_private p; 2027 enum tcp_seq_states state; 2028 struct sock *syn_wait_sk; 2029 int bucket, offset, sbucket, num; 2030 loff_t last_pos; 2031 }; 2032 2033 extern struct request_sock_ops tcp_request_sock_ops; 2034 extern struct request_sock_ops tcp6_request_sock_ops; 2035 2036 void tcp_v4_destroy_sock(struct sock *sk); 2037 2038 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2039 netdev_features_t features); 2040 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2041 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2042 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2043 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2044 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2045 void tcp_gro_complete(struct sk_buff *skb); 2046 2047 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2048 2049 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2050 { 2051 struct net *net = sock_net((struct sock *)tp); 2052 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2053 } 2054 2055 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2056 2057 #ifdef CONFIG_PROC_FS 2058 int tcp4_proc_init(void); 2059 void tcp4_proc_exit(void); 2060 #endif 2061 2062 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2063 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2064 const struct tcp_request_sock_ops *af_ops, 2065 struct sock *sk, struct sk_buff *skb); 2066 2067 /* TCP af-specific functions */ 2068 struct tcp_sock_af_ops { 2069 #ifdef CONFIG_TCP_MD5SIG 2070 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2071 const struct sock *addr_sk); 2072 int (*calc_md5_hash)(char *location, 2073 const struct tcp_md5sig_key *md5, 2074 const struct sock *sk, 2075 const struct sk_buff *skb); 2076 int (*md5_parse)(struct sock *sk, 2077 int optname, 2078 sockptr_t optval, 2079 int optlen); 2080 #endif 2081 }; 2082 2083 struct tcp_request_sock_ops { 2084 u16 mss_clamp; 2085 #ifdef CONFIG_TCP_MD5SIG 2086 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2087 const struct sock *addr_sk); 2088 int (*calc_md5_hash) (char *location, 2089 const struct tcp_md5sig_key *md5, 2090 const struct sock *sk, 2091 const struct sk_buff *skb); 2092 #endif 2093 #ifdef CONFIG_SYN_COOKIES 2094 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2095 __u16 *mss); 2096 #endif 2097 struct dst_entry *(*route_req)(const struct sock *sk, 2098 struct sk_buff *skb, 2099 struct flowi *fl, 2100 struct request_sock *req); 2101 u32 (*init_seq)(const struct sk_buff *skb); 2102 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2103 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2104 struct flowi *fl, struct request_sock *req, 2105 struct tcp_fastopen_cookie *foc, 2106 enum tcp_synack_type synack_type, 2107 struct sk_buff *syn_skb); 2108 }; 2109 2110 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2111 #if IS_ENABLED(CONFIG_IPV6) 2112 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2113 #endif 2114 2115 #ifdef CONFIG_SYN_COOKIES 2116 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2117 const struct sock *sk, struct sk_buff *skb, 2118 __u16 *mss) 2119 { 2120 tcp_synq_overflow(sk); 2121 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2122 return ops->cookie_init_seq(skb, mss); 2123 } 2124 #else 2125 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2126 const struct sock *sk, struct sk_buff *skb, 2127 __u16 *mss) 2128 { 2129 return 0; 2130 } 2131 #endif 2132 2133 int tcpv4_offload_init(void); 2134 2135 void tcp_v4_init(void); 2136 void tcp_init(void); 2137 2138 /* tcp_recovery.c */ 2139 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2140 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2141 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2142 u32 reo_wnd); 2143 extern bool tcp_rack_mark_lost(struct sock *sk); 2144 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2145 u64 xmit_time); 2146 extern void tcp_rack_reo_timeout(struct sock *sk); 2147 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2148 2149 /* tcp_plb.c */ 2150 2151 /* 2152 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2153 * expects cong_ratio which represents fraction of traffic that experienced 2154 * congestion over a single RTT. In order to avoid floating point operations, 2155 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2156 */ 2157 #define TCP_PLB_SCALE 8 2158 2159 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2160 struct tcp_plb_state { 2161 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2162 unused:3; 2163 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2164 }; 2165 2166 static inline void tcp_plb_init(const struct sock *sk, 2167 struct tcp_plb_state *plb) 2168 { 2169 plb->consec_cong_rounds = 0; 2170 plb->pause_until = 0; 2171 } 2172 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2173 const int cong_ratio); 2174 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2175 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2176 2177 /* At how many usecs into the future should the RTO fire? */ 2178 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2179 { 2180 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2181 u32 rto = inet_csk(sk)->icsk_rto; 2182 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2183 2184 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2185 } 2186 2187 /* 2188 * Save and compile IPv4 options, return a pointer to it 2189 */ 2190 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2191 struct sk_buff *skb) 2192 { 2193 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2194 struct ip_options_rcu *dopt = NULL; 2195 2196 if (opt->optlen) { 2197 int opt_size = sizeof(*dopt) + opt->optlen; 2198 2199 dopt = kmalloc(opt_size, GFP_ATOMIC); 2200 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2201 kfree(dopt); 2202 dopt = NULL; 2203 } 2204 } 2205 return dopt; 2206 } 2207 2208 /* locally generated TCP pure ACKs have skb->truesize == 2 2209 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2210 * This is much faster than dissecting the packet to find out. 2211 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2212 */ 2213 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2214 { 2215 return skb->truesize == 2; 2216 } 2217 2218 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2219 { 2220 skb->truesize = 2; 2221 } 2222 2223 static inline int tcp_inq(struct sock *sk) 2224 { 2225 struct tcp_sock *tp = tcp_sk(sk); 2226 int answ; 2227 2228 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2229 answ = 0; 2230 } else if (sock_flag(sk, SOCK_URGINLINE) || 2231 !tp->urg_data || 2232 before(tp->urg_seq, tp->copied_seq) || 2233 !before(tp->urg_seq, tp->rcv_nxt)) { 2234 2235 answ = tp->rcv_nxt - tp->copied_seq; 2236 2237 /* Subtract 1, if FIN was received */ 2238 if (answ && sock_flag(sk, SOCK_DONE)) 2239 answ--; 2240 } else { 2241 answ = tp->urg_seq - tp->copied_seq; 2242 } 2243 2244 return answ; 2245 } 2246 2247 int tcp_peek_len(struct socket *sock); 2248 2249 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2250 { 2251 u16 segs_in; 2252 2253 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2254 2255 /* We update these fields while other threads might 2256 * read them from tcp_get_info() 2257 */ 2258 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2259 if (skb->len > tcp_hdrlen(skb)) 2260 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2261 } 2262 2263 /* 2264 * TCP listen path runs lockless. 2265 * We forced "struct sock" to be const qualified to make sure 2266 * we don't modify one of its field by mistake. 2267 * Here, we increment sk_drops which is an atomic_t, so we can safely 2268 * make sock writable again. 2269 */ 2270 static inline void tcp_listendrop(const struct sock *sk) 2271 { 2272 atomic_inc(&((struct sock *)sk)->sk_drops); 2273 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2274 } 2275 2276 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2277 2278 /* 2279 * Interface for adding Upper Level Protocols over TCP 2280 */ 2281 2282 #define TCP_ULP_NAME_MAX 16 2283 #define TCP_ULP_MAX 128 2284 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2285 2286 struct tcp_ulp_ops { 2287 struct list_head list; 2288 2289 /* initialize ulp */ 2290 int (*init)(struct sock *sk); 2291 /* update ulp */ 2292 void (*update)(struct sock *sk, struct proto *p, 2293 void (*write_space)(struct sock *sk)); 2294 /* cleanup ulp */ 2295 void (*release)(struct sock *sk); 2296 /* diagnostic */ 2297 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2298 size_t (*get_info_size)(const struct sock *sk); 2299 /* clone ulp */ 2300 void (*clone)(const struct request_sock *req, struct sock *newsk, 2301 const gfp_t priority); 2302 2303 char name[TCP_ULP_NAME_MAX]; 2304 struct module *owner; 2305 }; 2306 int tcp_register_ulp(struct tcp_ulp_ops *type); 2307 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2308 int tcp_set_ulp(struct sock *sk, const char *name); 2309 void tcp_get_available_ulp(char *buf, size_t len); 2310 void tcp_cleanup_ulp(struct sock *sk); 2311 void tcp_update_ulp(struct sock *sk, struct proto *p, 2312 void (*write_space)(struct sock *sk)); 2313 2314 #define MODULE_ALIAS_TCP_ULP(name) \ 2315 __MODULE_INFO(alias, alias_userspace, name); \ 2316 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2317 2318 #ifdef CONFIG_NET_SOCK_MSG 2319 struct sk_msg; 2320 struct sk_psock; 2321 2322 #ifdef CONFIG_BPF_SYSCALL 2323 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2324 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2325 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2326 #endif /* CONFIG_BPF_SYSCALL */ 2327 2328 #ifdef CONFIG_INET 2329 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2330 #else 2331 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2332 { 2333 } 2334 #endif 2335 2336 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2337 struct sk_msg *msg, u32 bytes, int flags); 2338 #endif /* CONFIG_NET_SOCK_MSG */ 2339 2340 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2341 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2342 { 2343 } 2344 #endif 2345 2346 #ifdef CONFIG_CGROUP_BPF 2347 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2348 struct sk_buff *skb, 2349 unsigned int end_offset) 2350 { 2351 skops->skb = skb; 2352 skops->skb_data_end = skb->data + end_offset; 2353 } 2354 #else 2355 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2356 struct sk_buff *skb, 2357 unsigned int end_offset) 2358 { 2359 } 2360 #endif 2361 2362 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2363 * is < 0, then the BPF op failed (for example if the loaded BPF 2364 * program does not support the chosen operation or there is no BPF 2365 * program loaded). 2366 */ 2367 #ifdef CONFIG_BPF 2368 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2369 { 2370 struct bpf_sock_ops_kern sock_ops; 2371 int ret; 2372 2373 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2374 if (sk_fullsock(sk)) { 2375 sock_ops.is_fullsock = 1; 2376 sock_owned_by_me(sk); 2377 } 2378 2379 sock_ops.sk = sk; 2380 sock_ops.op = op; 2381 if (nargs > 0) 2382 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2383 2384 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2385 if (ret == 0) 2386 ret = sock_ops.reply; 2387 else 2388 ret = -1; 2389 return ret; 2390 } 2391 2392 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2393 { 2394 u32 args[2] = {arg1, arg2}; 2395 2396 return tcp_call_bpf(sk, op, 2, args); 2397 } 2398 2399 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2400 u32 arg3) 2401 { 2402 u32 args[3] = {arg1, arg2, arg3}; 2403 2404 return tcp_call_bpf(sk, op, 3, args); 2405 } 2406 2407 #else 2408 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2409 { 2410 return -EPERM; 2411 } 2412 2413 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2414 { 2415 return -EPERM; 2416 } 2417 2418 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2419 u32 arg3) 2420 { 2421 return -EPERM; 2422 } 2423 2424 #endif 2425 2426 static inline u32 tcp_timeout_init(struct sock *sk) 2427 { 2428 int timeout; 2429 2430 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2431 2432 if (timeout <= 0) 2433 timeout = TCP_TIMEOUT_INIT; 2434 return min_t(int, timeout, TCP_RTO_MAX); 2435 } 2436 2437 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2438 { 2439 int rwnd; 2440 2441 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2442 2443 if (rwnd < 0) 2444 rwnd = 0; 2445 return rwnd; 2446 } 2447 2448 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2449 { 2450 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2451 } 2452 2453 static inline void tcp_bpf_rtt(struct sock *sk) 2454 { 2455 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2456 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2457 } 2458 2459 #if IS_ENABLED(CONFIG_SMC) 2460 extern struct static_key_false tcp_have_smc; 2461 #endif 2462 2463 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2464 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2465 void (*cad)(struct sock *sk, u32 ack_seq)); 2466 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2467 void clean_acked_data_flush(void); 2468 #endif 2469 2470 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2471 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2472 const struct tcp_sock *tp) 2473 { 2474 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2475 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2476 } 2477 2478 /* Compute Earliest Departure Time for some control packets 2479 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2480 */ 2481 static inline u64 tcp_transmit_time(const struct sock *sk) 2482 { 2483 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2484 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2485 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2486 2487 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2488 } 2489 return 0; 2490 } 2491 2492 #endif /* _TCP_H */ 2493