1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP module. 8 * 9 * Version: @(#)tcp.h 1.0.5 05/23/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 */ 14 #ifndef _TCP_H 15 #define _TCP_H 16 17 #define FASTRETRANS_DEBUG 1 18 19 #include <linux/list.h> 20 #include <linux/tcp.h> 21 #include <linux/bug.h> 22 #include <linux/slab.h> 23 #include <linux/cache.h> 24 #include <linux/percpu.h> 25 #include <linux/skbuff.h> 26 #include <linux/kref.h> 27 #include <linux/ktime.h> 28 #include <linux/indirect_call_wrapper.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/inet_ecn.h> 41 #include <net/dst.h> 42 #include <net/mptcp.h> 43 44 #include <linux/seq_file.h> 45 #include <linux/memcontrol.h> 46 #include <linux/bpf-cgroup.h> 47 #include <linux/siphash.h> 48 #include <linux/net_mm.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 53 int tcp_orphan_count_sum(void); 54 55 void tcp_time_wait(struct sock *sk, int state, int timeo); 56 57 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 58 #define MAX_TCP_OPTION_SPACE 40 59 #define TCP_MIN_SND_MSS 48 60 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 61 62 /* 63 * Never offer a window over 32767 without using window scaling. Some 64 * poor stacks do signed 16bit maths! 65 */ 66 #define MAX_TCP_WINDOW 32767U 67 68 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 69 #define TCP_MIN_MSS 88U 70 71 /* The initial MTU to use for probing */ 72 #define TCP_BASE_MSS 1024 73 74 /* probing interval, default to 10 minutes as per RFC4821 */ 75 #define TCP_PROBE_INTERVAL 600 76 77 /* Specify interval when tcp mtu probing will stop */ 78 #define TCP_PROBE_THRESHOLD 8 79 80 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 81 #define TCP_FASTRETRANS_THRESH 3 82 83 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 84 #define TCP_MAX_QUICKACKS 16U 85 86 /* Maximal number of window scale according to RFC1323 */ 87 #define TCP_MAX_WSCALE 14U 88 89 /* urg_data states */ 90 #define TCP_URG_VALID 0x0100 91 #define TCP_URG_NOTYET 0x0200 92 #define TCP_URG_READ 0x0400 93 94 #define TCP_RETR1 3 /* 95 * This is how many retries it does before it 96 * tries to figure out if the gateway is 97 * down. Minimal RFC value is 3; it corresponds 98 * to ~3sec-8min depending on RTO. 99 */ 100 101 #define TCP_RETR2 15 /* 102 * This should take at least 103 * 90 minutes to time out. 104 * RFC1122 says that the limit is 100 sec. 105 * 15 is ~13-30min depending on RTO. 106 */ 107 108 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 109 * when active opening a connection. 110 * RFC1122 says the minimum retry MUST 111 * be at least 180secs. Nevertheless 112 * this value is corresponding to 113 * 63secs of retransmission with the 114 * current initial RTO. 115 */ 116 117 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 118 * when passive opening a connection. 119 * This is corresponding to 31secs of 120 * retransmission with the current 121 * initial RTO. 122 */ 123 124 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 125 * state, about 60 seconds */ 126 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 127 /* BSD style FIN_WAIT2 deadlock breaker. 128 * It used to be 3min, new value is 60sec, 129 * to combine FIN-WAIT-2 timeout with 130 * TIME-WAIT timer. 131 */ 132 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 133 134 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 135 #if HZ >= 100 136 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 137 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 138 #else 139 #define TCP_DELACK_MIN 4U 140 #define TCP_ATO_MIN 4U 141 #endif 142 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 143 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 144 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 145 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 146 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 147 * used as a fallback RTO for the 148 * initial data transmission if no 149 * valid RTT sample has been acquired, 150 * most likely due to retrans in 3WHS. 151 */ 152 153 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 154 * for local resources. 155 */ 156 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 157 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 158 #define TCP_KEEPALIVE_INTVL (75*HZ) 159 160 #define MAX_TCP_KEEPIDLE 32767 161 #define MAX_TCP_KEEPINTVL 32767 162 #define MAX_TCP_KEEPCNT 127 163 #define MAX_TCP_SYNCNT 127 164 165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 167 * after this time. It should be equal 168 * (or greater than) TCP_TIMEWAIT_LEN 169 * to provide reliability equal to one 170 * provided by timewait state. 171 */ 172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 173 * timestamps. It must be less than 174 * minimal timewait lifetime. 175 */ 176 /* 177 * TCP option 178 */ 179 180 #define TCPOPT_NOP 1 /* Padding */ 181 #define TCPOPT_EOL 0 /* End of options */ 182 #define TCPOPT_MSS 2 /* Segment size negotiating */ 183 #define TCPOPT_WINDOW 3 /* Window scaling */ 184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 185 #define TCPOPT_SACK 5 /* SACK Block */ 186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 188 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 189 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 190 #define TCPOPT_EXP 254 /* Experimental */ 191 /* Magic number to be after the option value for sharing TCP 192 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 193 */ 194 #define TCPOPT_FASTOPEN_MAGIC 0xF989 195 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 196 197 /* 198 * TCP option lengths 199 */ 200 201 #define TCPOLEN_MSS 4 202 #define TCPOLEN_WINDOW 3 203 #define TCPOLEN_SACK_PERM 2 204 #define TCPOLEN_TIMESTAMP 10 205 #define TCPOLEN_MD5SIG 18 206 #define TCPOLEN_FASTOPEN_BASE 2 207 #define TCPOLEN_EXP_FASTOPEN_BASE 4 208 #define TCPOLEN_EXP_SMC_BASE 6 209 210 /* But this is what stacks really send out. */ 211 #define TCPOLEN_TSTAMP_ALIGNED 12 212 #define TCPOLEN_WSCALE_ALIGNED 4 213 #define TCPOLEN_SACKPERM_ALIGNED 4 214 #define TCPOLEN_SACK_BASE 2 215 #define TCPOLEN_SACK_BASE_ALIGNED 4 216 #define TCPOLEN_SACK_PERBLOCK 8 217 #define TCPOLEN_MD5SIG_ALIGNED 20 218 #define TCPOLEN_MSS_ALIGNED 4 219 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 220 221 /* Flags in tp->nonagle */ 222 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 223 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 224 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 225 226 /* TCP thin-stream limits */ 227 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 228 229 /* TCP initial congestion window as per rfc6928 */ 230 #define TCP_INIT_CWND 10 231 232 /* Bit Flags for sysctl_tcp_fastopen */ 233 #define TFO_CLIENT_ENABLE 1 234 #define TFO_SERVER_ENABLE 2 235 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 236 237 /* Accept SYN data w/o any cookie option */ 238 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 239 240 /* Force enable TFO on all listeners, i.e., not requiring the 241 * TCP_FASTOPEN socket option. 242 */ 243 #define TFO_SERVER_WO_SOCKOPT1 0x400 244 245 246 /* sysctl variables for tcp */ 247 extern int sysctl_tcp_max_orphans; 248 extern long sysctl_tcp_mem[3]; 249 250 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 251 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 252 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 253 254 extern atomic_long_t tcp_memory_allocated; 255 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 256 257 extern struct percpu_counter tcp_sockets_allocated; 258 extern unsigned long tcp_memory_pressure; 259 260 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 261 static inline bool tcp_under_memory_pressure(const struct sock *sk) 262 { 263 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 264 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 265 return true; 266 267 return READ_ONCE(tcp_memory_pressure); 268 } 269 /* 270 * The next routines deal with comparing 32 bit unsigned ints 271 * and worry about wraparound (automatic with unsigned arithmetic). 272 */ 273 274 static inline bool before(__u32 seq1, __u32 seq2) 275 { 276 return (__s32)(seq1-seq2) < 0; 277 } 278 #define after(seq2, seq1) before(seq1, seq2) 279 280 /* is s2<=s1<=s3 ? */ 281 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 282 { 283 return seq3 - seq2 >= seq1 - seq2; 284 } 285 286 static inline bool tcp_out_of_memory(struct sock *sk) 287 { 288 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 289 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 290 return true; 291 return false; 292 } 293 294 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 295 { 296 sk_wmem_queued_add(sk, -skb->truesize); 297 if (!skb_zcopy_pure(skb)) 298 sk_mem_uncharge(sk, skb->truesize); 299 else 300 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 301 __kfree_skb(skb); 302 } 303 304 void sk_forced_mem_schedule(struct sock *sk, int size); 305 306 bool tcp_check_oom(struct sock *sk, int shift); 307 308 309 extern struct proto tcp_prot; 310 311 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 312 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 315 316 void tcp_tasklet_init(void); 317 318 int tcp_v4_err(struct sk_buff *skb, u32); 319 320 void tcp_shutdown(struct sock *sk, int how); 321 322 int tcp_v4_early_demux(struct sk_buff *skb); 323 int tcp_v4_rcv(struct sk_buff *skb); 324 325 void tcp_remove_empty_skb(struct sock *sk); 326 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 327 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 328 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 329 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 330 size_t size, struct ubuf_info *uarg); 331 void tcp_splice_eof(struct socket *sock); 332 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 333 int tcp_wmem_schedule(struct sock *sk, int copy); 334 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 335 int size_goal); 336 void tcp_release_cb(struct sock *sk); 337 void tcp_wfree(struct sk_buff *skb); 338 void tcp_write_timer_handler(struct sock *sk); 339 void tcp_delack_timer_handler(struct sock *sk); 340 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 341 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 342 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_space_adjust(struct sock *sk); 344 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 345 void tcp_twsk_destructor(struct sock *sk); 346 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 347 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 348 struct pipe_inode_info *pipe, size_t len, 349 unsigned int flags); 350 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 351 bool force_schedule); 352 353 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 354 static inline void tcp_dec_quickack_mode(struct sock *sk, 355 const unsigned int pkts) 356 { 357 struct inet_connection_sock *icsk = inet_csk(sk); 358 359 if (icsk->icsk_ack.quick) { 360 if (pkts >= icsk->icsk_ack.quick) { 361 icsk->icsk_ack.quick = 0; 362 /* Leaving quickack mode we deflate ATO. */ 363 icsk->icsk_ack.ato = TCP_ATO_MIN; 364 } else 365 icsk->icsk_ack.quick -= pkts; 366 } 367 } 368 369 #define TCP_ECN_OK 1 370 #define TCP_ECN_QUEUE_CWR 2 371 #define TCP_ECN_DEMAND_CWR 4 372 #define TCP_ECN_SEEN 8 373 374 enum tcp_tw_status { 375 TCP_TW_SUCCESS = 0, 376 TCP_TW_RST = 1, 377 TCP_TW_ACK = 2, 378 TCP_TW_SYN = 3 379 }; 380 381 382 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 383 struct sk_buff *skb, 384 const struct tcphdr *th); 385 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 386 struct request_sock *req, bool fastopen, 387 bool *lost_race); 388 int tcp_child_process(struct sock *parent, struct sock *child, 389 struct sk_buff *skb); 390 void tcp_enter_loss(struct sock *sk); 391 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 392 void tcp_clear_retrans(struct tcp_sock *tp); 393 void tcp_update_metrics(struct sock *sk); 394 void tcp_init_metrics(struct sock *sk); 395 void tcp_metrics_init(void); 396 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 397 void __tcp_close(struct sock *sk, long timeout); 398 void tcp_close(struct sock *sk, long timeout); 399 void tcp_init_sock(struct sock *sk); 400 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 401 __poll_t tcp_poll(struct file *file, struct socket *sock, 402 struct poll_table_struct *wait); 403 int do_tcp_getsockopt(struct sock *sk, int level, 404 int optname, sockptr_t optval, sockptr_t optlen); 405 int tcp_getsockopt(struct sock *sk, int level, int optname, 406 char __user *optval, int __user *optlen); 407 bool tcp_bpf_bypass_getsockopt(int level, int optname); 408 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 409 sockptr_t optval, unsigned int optlen); 410 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 411 unsigned int optlen); 412 void tcp_set_keepalive(struct sock *sk, int val); 413 void tcp_syn_ack_timeout(const struct request_sock *req); 414 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 415 int flags, int *addr_len); 416 int tcp_set_rcvlowat(struct sock *sk, int val); 417 int tcp_set_window_clamp(struct sock *sk, int val); 418 void tcp_update_recv_tstamps(struct sk_buff *skb, 419 struct scm_timestamping_internal *tss); 420 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 421 struct scm_timestamping_internal *tss); 422 void tcp_data_ready(struct sock *sk); 423 #ifdef CONFIG_MMU 424 int tcp_mmap(struct file *file, struct socket *sock, 425 struct vm_area_struct *vma); 426 #endif 427 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 428 struct tcp_options_received *opt_rx, 429 int estab, struct tcp_fastopen_cookie *foc); 430 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 431 432 /* 433 * BPF SKB-less helpers 434 */ 435 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 436 struct tcphdr *th, u32 *cookie); 437 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 438 struct tcphdr *th, u32 *cookie); 439 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 440 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 441 const struct tcp_request_sock_ops *af_ops, 442 struct sock *sk, struct tcphdr *th); 443 /* 444 * TCP v4 functions exported for the inet6 API 445 */ 446 447 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 448 void tcp_v4_mtu_reduced(struct sock *sk); 449 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 450 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 451 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 452 struct sock *tcp_create_openreq_child(const struct sock *sk, 453 struct request_sock *req, 454 struct sk_buff *skb); 455 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 456 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 457 struct request_sock *req, 458 struct dst_entry *dst, 459 struct request_sock *req_unhash, 460 bool *own_req); 461 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 462 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 463 int tcp_connect(struct sock *sk); 464 enum tcp_synack_type { 465 TCP_SYNACK_NORMAL, 466 TCP_SYNACK_FASTOPEN, 467 TCP_SYNACK_COOKIE, 468 }; 469 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 470 struct request_sock *req, 471 struct tcp_fastopen_cookie *foc, 472 enum tcp_synack_type synack_type, 473 struct sk_buff *syn_skb); 474 int tcp_disconnect(struct sock *sk, int flags); 475 476 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 477 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 478 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 479 480 /* From syncookies.c */ 481 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 482 struct request_sock *req, 483 struct dst_entry *dst, u32 tsoff); 484 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 485 u32 cookie); 486 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 487 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 488 const struct tcp_request_sock_ops *af_ops, 489 struct sock *sk, struct sk_buff *skb); 490 #ifdef CONFIG_SYN_COOKIES 491 492 /* Syncookies use a monotonic timer which increments every 60 seconds. 493 * This counter is used both as a hash input and partially encoded into 494 * the cookie value. A cookie is only validated further if the delta 495 * between the current counter value and the encoded one is less than this, 496 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 497 * the counter advances immediately after a cookie is generated). 498 */ 499 #define MAX_SYNCOOKIE_AGE 2 500 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 501 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 502 503 /* syncookies: remember time of last synqueue overflow 504 * But do not dirty this field too often (once per second is enough) 505 * It is racy as we do not hold a lock, but race is very minor. 506 */ 507 static inline void tcp_synq_overflow(const struct sock *sk) 508 { 509 unsigned int last_overflow; 510 unsigned int now = jiffies; 511 512 if (sk->sk_reuseport) { 513 struct sock_reuseport *reuse; 514 515 reuse = rcu_dereference(sk->sk_reuseport_cb); 516 if (likely(reuse)) { 517 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 518 if (!time_between32(now, last_overflow, 519 last_overflow + HZ)) 520 WRITE_ONCE(reuse->synq_overflow_ts, now); 521 return; 522 } 523 } 524 525 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 526 if (!time_between32(now, last_overflow, last_overflow + HZ)) 527 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 528 } 529 530 /* syncookies: no recent synqueue overflow on this listening socket? */ 531 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 532 { 533 unsigned int last_overflow; 534 unsigned int now = jiffies; 535 536 if (sk->sk_reuseport) { 537 struct sock_reuseport *reuse; 538 539 reuse = rcu_dereference(sk->sk_reuseport_cb); 540 if (likely(reuse)) { 541 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 542 return !time_between32(now, last_overflow - HZ, 543 last_overflow + 544 TCP_SYNCOOKIE_VALID); 545 } 546 } 547 548 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 549 550 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 551 * then we're under synflood. However, we have to use 552 * 'last_overflow - HZ' as lower bound. That's because a concurrent 553 * tcp_synq_overflow() could update .ts_recent_stamp after we read 554 * jiffies but before we store .ts_recent_stamp into last_overflow, 555 * which could lead to rejecting a valid syncookie. 556 */ 557 return !time_between32(now, last_overflow - HZ, 558 last_overflow + TCP_SYNCOOKIE_VALID); 559 } 560 561 static inline u32 tcp_cookie_time(void) 562 { 563 u64 val = get_jiffies_64(); 564 565 do_div(val, TCP_SYNCOOKIE_PERIOD); 566 return val; 567 } 568 569 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 570 u16 *mssp); 571 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 572 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 573 bool cookie_timestamp_decode(const struct net *net, 574 struct tcp_options_received *opt); 575 bool cookie_ecn_ok(const struct tcp_options_received *opt, 576 const struct net *net, const struct dst_entry *dst); 577 578 /* From net/ipv6/syncookies.c */ 579 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 580 u32 cookie); 581 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 582 583 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 584 const struct tcphdr *th, u16 *mssp); 585 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 586 #endif 587 /* tcp_output.c */ 588 589 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 590 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 591 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 592 int nonagle); 593 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 594 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 595 void tcp_retransmit_timer(struct sock *sk); 596 void tcp_xmit_retransmit_queue(struct sock *); 597 void tcp_simple_retransmit(struct sock *); 598 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 599 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 600 enum tcp_queue { 601 TCP_FRAG_IN_WRITE_QUEUE, 602 TCP_FRAG_IN_RTX_QUEUE, 603 }; 604 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 605 struct sk_buff *skb, u32 len, 606 unsigned int mss_now, gfp_t gfp); 607 608 void tcp_send_probe0(struct sock *); 609 void tcp_send_partial(struct sock *); 610 int tcp_write_wakeup(struct sock *, int mib); 611 void tcp_send_fin(struct sock *sk); 612 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 613 int tcp_send_synack(struct sock *); 614 void tcp_push_one(struct sock *, unsigned int mss_now); 615 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 616 void tcp_send_ack(struct sock *sk); 617 void tcp_send_delayed_ack(struct sock *sk); 618 void tcp_send_loss_probe(struct sock *sk); 619 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 620 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 621 const struct sk_buff *next_skb); 622 623 /* tcp_input.c */ 624 void tcp_rearm_rto(struct sock *sk); 625 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 626 void tcp_reset(struct sock *sk, struct sk_buff *skb); 627 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 628 void tcp_fin(struct sock *sk); 629 void tcp_check_space(struct sock *sk); 630 void tcp_sack_compress_send_ack(struct sock *sk); 631 632 /* tcp_timer.c */ 633 void tcp_init_xmit_timers(struct sock *); 634 static inline void tcp_clear_xmit_timers(struct sock *sk) 635 { 636 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 637 __sock_put(sk); 638 639 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 640 __sock_put(sk); 641 642 inet_csk_clear_xmit_timers(sk); 643 } 644 645 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 646 unsigned int tcp_current_mss(struct sock *sk); 647 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 648 649 /* Bound MSS / TSO packet size with the half of the window */ 650 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 651 { 652 int cutoff; 653 654 /* When peer uses tiny windows, there is no use in packetizing 655 * to sub-MSS pieces for the sake of SWS or making sure there 656 * are enough packets in the pipe for fast recovery. 657 * 658 * On the other hand, for extremely large MSS devices, handling 659 * smaller than MSS windows in this way does make sense. 660 */ 661 if (tp->max_window > TCP_MSS_DEFAULT) 662 cutoff = (tp->max_window >> 1); 663 else 664 cutoff = tp->max_window; 665 666 if (cutoff && pktsize > cutoff) 667 return max_t(int, cutoff, 68U - tp->tcp_header_len); 668 else 669 return pktsize; 670 } 671 672 /* tcp.c */ 673 void tcp_get_info(struct sock *, struct tcp_info *); 674 675 /* Read 'sendfile()'-style from a TCP socket */ 676 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 677 sk_read_actor_t recv_actor); 678 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 679 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 680 void tcp_read_done(struct sock *sk, size_t len); 681 682 void tcp_initialize_rcv_mss(struct sock *sk); 683 684 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 685 int tcp_mss_to_mtu(struct sock *sk, int mss); 686 void tcp_mtup_init(struct sock *sk); 687 688 static inline void tcp_bound_rto(const struct sock *sk) 689 { 690 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 691 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 692 } 693 694 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 695 { 696 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 697 } 698 699 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 700 { 701 /* mptcp hooks are only on the slow path */ 702 if (sk_is_mptcp((struct sock *)tp)) 703 return; 704 705 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 706 ntohl(TCP_FLAG_ACK) | 707 snd_wnd); 708 } 709 710 static inline void tcp_fast_path_on(struct tcp_sock *tp) 711 { 712 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 713 } 714 715 static inline void tcp_fast_path_check(struct sock *sk) 716 { 717 struct tcp_sock *tp = tcp_sk(sk); 718 719 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 720 tp->rcv_wnd && 721 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 722 !tp->urg_data) 723 tcp_fast_path_on(tp); 724 } 725 726 /* Compute the actual rto_min value */ 727 static inline u32 tcp_rto_min(struct sock *sk) 728 { 729 const struct dst_entry *dst = __sk_dst_get(sk); 730 u32 rto_min = inet_csk(sk)->icsk_rto_min; 731 732 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 733 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 734 return rto_min; 735 } 736 737 static inline u32 tcp_rto_min_us(struct sock *sk) 738 { 739 return jiffies_to_usecs(tcp_rto_min(sk)); 740 } 741 742 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 743 { 744 return dst_metric_locked(dst, RTAX_CC_ALGO); 745 } 746 747 /* Minimum RTT in usec. ~0 means not available. */ 748 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 749 { 750 return minmax_get(&tp->rtt_min); 751 } 752 753 /* Compute the actual receive window we are currently advertising. 754 * Rcv_nxt can be after the window if our peer push more data 755 * than the offered window. 756 */ 757 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 758 { 759 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 760 761 if (win < 0) 762 win = 0; 763 return (u32) win; 764 } 765 766 /* Choose a new window, without checks for shrinking, and without 767 * scaling applied to the result. The caller does these things 768 * if necessary. This is a "raw" window selection. 769 */ 770 u32 __tcp_select_window(struct sock *sk); 771 772 void tcp_send_window_probe(struct sock *sk); 773 774 /* TCP uses 32bit jiffies to save some space. 775 * Note that this is different from tcp_time_stamp, which 776 * historically has been the same until linux-4.13. 777 */ 778 #define tcp_jiffies32 ((u32)jiffies) 779 780 /* 781 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 782 * It is no longer tied to jiffies, but to 1 ms clock. 783 * Note: double check if you want to use tcp_jiffies32 instead of this. 784 */ 785 #define TCP_TS_HZ 1000 786 787 static inline u64 tcp_clock_ns(void) 788 { 789 return ktime_get_ns(); 790 } 791 792 static inline u64 tcp_clock_us(void) 793 { 794 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 795 } 796 797 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 798 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 799 { 800 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 801 } 802 803 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 804 static inline u32 tcp_ns_to_ts(u64 ns) 805 { 806 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 807 } 808 809 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 810 static inline u32 tcp_time_stamp_raw(void) 811 { 812 return tcp_ns_to_ts(tcp_clock_ns()); 813 } 814 815 void tcp_mstamp_refresh(struct tcp_sock *tp); 816 817 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 818 { 819 return max_t(s64, t1 - t0, 0); 820 } 821 822 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 823 { 824 return tcp_ns_to_ts(skb->skb_mstamp_ns); 825 } 826 827 /* provide the departure time in us unit */ 828 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 829 { 830 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 831 } 832 833 834 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 835 836 #define TCPHDR_FIN 0x01 837 #define TCPHDR_SYN 0x02 838 #define TCPHDR_RST 0x04 839 #define TCPHDR_PSH 0x08 840 #define TCPHDR_ACK 0x10 841 #define TCPHDR_URG 0x20 842 #define TCPHDR_ECE 0x40 843 #define TCPHDR_CWR 0x80 844 845 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 846 847 /* This is what the send packet queuing engine uses to pass 848 * TCP per-packet control information to the transmission code. 849 * We also store the host-order sequence numbers in here too. 850 * This is 44 bytes if IPV6 is enabled. 851 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 852 */ 853 struct tcp_skb_cb { 854 __u32 seq; /* Starting sequence number */ 855 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 856 union { 857 /* Note : tcp_tw_isn is used in input path only 858 * (isn chosen by tcp_timewait_state_process()) 859 * 860 * tcp_gso_segs/size are used in write queue only, 861 * cf tcp_skb_pcount()/tcp_skb_mss() 862 */ 863 __u32 tcp_tw_isn; 864 struct { 865 u16 tcp_gso_segs; 866 u16 tcp_gso_size; 867 }; 868 }; 869 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 870 871 __u8 sacked; /* State flags for SACK. */ 872 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 873 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 874 #define TCPCB_LOST 0x04 /* SKB is lost */ 875 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 876 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 877 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 878 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 879 TCPCB_REPAIRED) 880 881 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 882 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 883 eor:1, /* Is skb MSG_EOR marked? */ 884 has_rxtstamp:1, /* SKB has a RX timestamp */ 885 unused:5; 886 __u32 ack_seq; /* Sequence number ACK'd */ 887 union { 888 struct { 889 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 890 /* There is space for up to 24 bytes */ 891 __u32 is_app_limited:1, /* cwnd not fully used? */ 892 delivered_ce:20, 893 unused:11; 894 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 895 __u32 delivered; 896 /* start of send pipeline phase */ 897 u64 first_tx_mstamp; 898 /* when we reached the "delivered" count */ 899 u64 delivered_mstamp; 900 } tx; /* only used for outgoing skbs */ 901 union { 902 struct inet_skb_parm h4; 903 #if IS_ENABLED(CONFIG_IPV6) 904 struct inet6_skb_parm h6; 905 #endif 906 } header; /* For incoming skbs */ 907 }; 908 }; 909 910 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 911 912 extern const struct inet_connection_sock_af_ops ipv4_specific; 913 914 #if IS_ENABLED(CONFIG_IPV6) 915 /* This is the variant of inet6_iif() that must be used by TCP, 916 * as TCP moves IP6CB into a different location in skb->cb[] 917 */ 918 static inline int tcp_v6_iif(const struct sk_buff *skb) 919 { 920 return TCP_SKB_CB(skb)->header.h6.iif; 921 } 922 923 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 924 { 925 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 926 927 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 928 } 929 930 /* TCP_SKB_CB reference means this can not be used from early demux */ 931 static inline int tcp_v6_sdif(const struct sk_buff *skb) 932 { 933 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 934 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 935 return TCP_SKB_CB(skb)->header.h6.iif; 936 #endif 937 return 0; 938 } 939 940 extern const struct inet_connection_sock_af_ops ipv6_specific; 941 942 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 943 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 944 void tcp_v6_early_demux(struct sk_buff *skb); 945 946 #endif 947 948 /* TCP_SKB_CB reference means this can not be used from early demux */ 949 static inline int tcp_v4_sdif(struct sk_buff *skb) 950 { 951 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 952 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 953 return TCP_SKB_CB(skb)->header.h4.iif; 954 #endif 955 return 0; 956 } 957 958 /* Due to TSO, an SKB can be composed of multiple actual 959 * packets. To keep these tracked properly, we use this. 960 */ 961 static inline int tcp_skb_pcount(const struct sk_buff *skb) 962 { 963 return TCP_SKB_CB(skb)->tcp_gso_segs; 964 } 965 966 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 967 { 968 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 969 } 970 971 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 972 { 973 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 974 } 975 976 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 977 static inline int tcp_skb_mss(const struct sk_buff *skb) 978 { 979 return TCP_SKB_CB(skb)->tcp_gso_size; 980 } 981 982 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 983 { 984 return likely(!TCP_SKB_CB(skb)->eor); 985 } 986 987 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 988 const struct sk_buff *from) 989 { 990 return likely(tcp_skb_can_collapse_to(to) && 991 mptcp_skb_can_collapse(to, from) && 992 skb_pure_zcopy_same(to, from)); 993 } 994 995 /* Events passed to congestion control interface */ 996 enum tcp_ca_event { 997 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 998 CA_EVENT_CWND_RESTART, /* congestion window restart */ 999 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1000 CA_EVENT_LOSS, /* loss timeout */ 1001 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1002 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1003 }; 1004 1005 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1006 enum tcp_ca_ack_event_flags { 1007 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1008 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1009 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1010 }; 1011 1012 /* 1013 * Interface for adding new TCP congestion control handlers 1014 */ 1015 #define TCP_CA_NAME_MAX 16 1016 #define TCP_CA_MAX 128 1017 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1018 1019 #define TCP_CA_UNSPEC 0 1020 1021 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1022 #define TCP_CONG_NON_RESTRICTED 0x1 1023 /* Requires ECN/ECT set on all packets */ 1024 #define TCP_CONG_NEEDS_ECN 0x2 1025 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1026 1027 union tcp_cc_info; 1028 1029 struct ack_sample { 1030 u32 pkts_acked; 1031 s32 rtt_us; 1032 u32 in_flight; 1033 }; 1034 1035 /* A rate sample measures the number of (original/retransmitted) data 1036 * packets delivered "delivered" over an interval of time "interval_us". 1037 * The tcp_rate.c code fills in the rate sample, and congestion 1038 * control modules that define a cong_control function to run at the end 1039 * of ACK processing can optionally chose to consult this sample when 1040 * setting cwnd and pacing rate. 1041 * A sample is invalid if "delivered" or "interval_us" is negative. 1042 */ 1043 struct rate_sample { 1044 u64 prior_mstamp; /* starting timestamp for interval */ 1045 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1046 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1047 s32 delivered; /* number of packets delivered over interval */ 1048 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1049 long interval_us; /* time for tp->delivered to incr "delivered" */ 1050 u32 snd_interval_us; /* snd interval for delivered packets */ 1051 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1052 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1053 int losses; /* number of packets marked lost upon ACK */ 1054 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1055 u32 prior_in_flight; /* in flight before this ACK */ 1056 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1057 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1058 bool is_retrans; /* is sample from retransmission? */ 1059 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1060 }; 1061 1062 struct tcp_congestion_ops { 1063 /* fast path fields are put first to fill one cache line */ 1064 1065 /* return slow start threshold (required) */ 1066 u32 (*ssthresh)(struct sock *sk); 1067 1068 /* do new cwnd calculation (required) */ 1069 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1070 1071 /* call before changing ca_state (optional) */ 1072 void (*set_state)(struct sock *sk, u8 new_state); 1073 1074 /* call when cwnd event occurs (optional) */ 1075 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1076 1077 /* call when ack arrives (optional) */ 1078 void (*in_ack_event)(struct sock *sk, u32 flags); 1079 1080 /* hook for packet ack accounting (optional) */ 1081 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1082 1083 /* override sysctl_tcp_min_tso_segs */ 1084 u32 (*min_tso_segs)(struct sock *sk); 1085 1086 /* call when packets are delivered to update cwnd and pacing rate, 1087 * after all the ca_state processing. (optional) 1088 */ 1089 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1090 1091 1092 /* new value of cwnd after loss (required) */ 1093 u32 (*undo_cwnd)(struct sock *sk); 1094 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1095 u32 (*sndbuf_expand)(struct sock *sk); 1096 1097 /* control/slow paths put last */ 1098 /* get info for inet_diag (optional) */ 1099 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1100 union tcp_cc_info *info); 1101 1102 char name[TCP_CA_NAME_MAX]; 1103 struct module *owner; 1104 struct list_head list; 1105 u32 key; 1106 u32 flags; 1107 1108 /* initialize private data (optional) */ 1109 void (*init)(struct sock *sk); 1110 /* cleanup private data (optional) */ 1111 void (*release)(struct sock *sk); 1112 } ____cacheline_aligned_in_smp; 1113 1114 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1115 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1116 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1117 struct tcp_congestion_ops *old_type); 1118 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1119 1120 void tcp_assign_congestion_control(struct sock *sk); 1121 void tcp_init_congestion_control(struct sock *sk); 1122 void tcp_cleanup_congestion_control(struct sock *sk); 1123 int tcp_set_default_congestion_control(struct net *net, const char *name); 1124 void tcp_get_default_congestion_control(struct net *net, char *name); 1125 void tcp_get_available_congestion_control(char *buf, size_t len); 1126 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1127 int tcp_set_allowed_congestion_control(char *allowed); 1128 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1129 bool cap_net_admin); 1130 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1131 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1132 1133 u32 tcp_reno_ssthresh(struct sock *sk); 1134 u32 tcp_reno_undo_cwnd(struct sock *sk); 1135 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1136 extern struct tcp_congestion_ops tcp_reno; 1137 1138 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1139 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1140 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1141 #ifdef CONFIG_INET 1142 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1143 #else 1144 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1145 { 1146 return NULL; 1147 } 1148 #endif 1149 1150 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1151 { 1152 const struct inet_connection_sock *icsk = inet_csk(sk); 1153 1154 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1155 } 1156 1157 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1158 { 1159 const struct inet_connection_sock *icsk = inet_csk(sk); 1160 1161 if (icsk->icsk_ca_ops->cwnd_event) 1162 icsk->icsk_ca_ops->cwnd_event(sk, event); 1163 } 1164 1165 /* From tcp_cong.c */ 1166 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1167 1168 /* From tcp_rate.c */ 1169 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1170 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1171 struct rate_sample *rs); 1172 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1173 bool is_sack_reneg, struct rate_sample *rs); 1174 void tcp_rate_check_app_limited(struct sock *sk); 1175 1176 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1177 { 1178 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1179 } 1180 1181 /* These functions determine how the current flow behaves in respect of SACK 1182 * handling. SACK is negotiated with the peer, and therefore it can vary 1183 * between different flows. 1184 * 1185 * tcp_is_sack - SACK enabled 1186 * tcp_is_reno - No SACK 1187 */ 1188 static inline int tcp_is_sack(const struct tcp_sock *tp) 1189 { 1190 return likely(tp->rx_opt.sack_ok); 1191 } 1192 1193 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1194 { 1195 return !tcp_is_sack(tp); 1196 } 1197 1198 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1199 { 1200 return tp->sacked_out + tp->lost_out; 1201 } 1202 1203 /* This determines how many packets are "in the network" to the best 1204 * of our knowledge. In many cases it is conservative, but where 1205 * detailed information is available from the receiver (via SACK 1206 * blocks etc.) we can make more aggressive calculations. 1207 * 1208 * Use this for decisions involving congestion control, use just 1209 * tp->packets_out to determine if the send queue is empty or not. 1210 * 1211 * Read this equation as: 1212 * 1213 * "Packets sent once on transmission queue" MINUS 1214 * "Packets left network, but not honestly ACKed yet" PLUS 1215 * "Packets fast retransmitted" 1216 */ 1217 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1218 { 1219 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1220 } 1221 1222 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1223 1224 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1225 { 1226 return tp->snd_cwnd; 1227 } 1228 1229 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1230 { 1231 WARN_ON_ONCE((int)val <= 0); 1232 tp->snd_cwnd = val; 1233 } 1234 1235 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1236 { 1237 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1238 } 1239 1240 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1241 { 1242 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1243 } 1244 1245 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1246 { 1247 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1248 (1 << inet_csk(sk)->icsk_ca_state); 1249 } 1250 1251 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1252 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1253 * ssthresh. 1254 */ 1255 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1256 { 1257 const struct tcp_sock *tp = tcp_sk(sk); 1258 1259 if (tcp_in_cwnd_reduction(sk)) 1260 return tp->snd_ssthresh; 1261 else 1262 return max(tp->snd_ssthresh, 1263 ((tcp_snd_cwnd(tp) >> 1) + 1264 (tcp_snd_cwnd(tp) >> 2))); 1265 } 1266 1267 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1268 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1269 1270 void tcp_enter_cwr(struct sock *sk); 1271 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1272 1273 /* The maximum number of MSS of available cwnd for which TSO defers 1274 * sending if not using sysctl_tcp_tso_win_divisor. 1275 */ 1276 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1277 { 1278 return 3; 1279 } 1280 1281 /* Returns end sequence number of the receiver's advertised window */ 1282 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1283 { 1284 return tp->snd_una + tp->snd_wnd; 1285 } 1286 1287 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1288 * flexible approach. The RFC suggests cwnd should not be raised unless 1289 * it was fully used previously. And that's exactly what we do in 1290 * congestion avoidance mode. But in slow start we allow cwnd to grow 1291 * as long as the application has used half the cwnd. 1292 * Example : 1293 * cwnd is 10 (IW10), but application sends 9 frames. 1294 * We allow cwnd to reach 18 when all frames are ACKed. 1295 * This check is safe because it's as aggressive as slow start which already 1296 * risks 100% overshoot. The advantage is that we discourage application to 1297 * either send more filler packets or data to artificially blow up the cwnd 1298 * usage, and allow application-limited process to probe bw more aggressively. 1299 */ 1300 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1301 { 1302 const struct tcp_sock *tp = tcp_sk(sk); 1303 1304 if (tp->is_cwnd_limited) 1305 return true; 1306 1307 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1308 if (tcp_in_slow_start(tp)) 1309 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1310 1311 return false; 1312 } 1313 1314 /* BBR congestion control needs pacing. 1315 * Same remark for SO_MAX_PACING_RATE. 1316 * sch_fq packet scheduler is efficiently handling pacing, 1317 * but is not always installed/used. 1318 * Return true if TCP stack should pace packets itself. 1319 */ 1320 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1321 { 1322 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1323 } 1324 1325 /* Estimates in how many jiffies next packet for this flow can be sent. 1326 * Scheduling a retransmit timer too early would be silly. 1327 */ 1328 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1329 { 1330 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1331 1332 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1333 } 1334 1335 static inline void tcp_reset_xmit_timer(struct sock *sk, 1336 const int what, 1337 unsigned long when, 1338 const unsigned long max_when) 1339 { 1340 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1341 max_when); 1342 } 1343 1344 /* Something is really bad, we could not queue an additional packet, 1345 * because qdisc is full or receiver sent a 0 window, or we are paced. 1346 * We do not want to add fuel to the fire, or abort too early, 1347 * so make sure the timer we arm now is at least 200ms in the future, 1348 * regardless of current icsk_rto value (as it could be ~2ms) 1349 */ 1350 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1351 { 1352 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1353 } 1354 1355 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1356 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1357 unsigned long max_when) 1358 { 1359 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1360 inet_csk(sk)->icsk_backoff); 1361 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1362 1363 return (unsigned long)min_t(u64, when, max_when); 1364 } 1365 1366 static inline void tcp_check_probe_timer(struct sock *sk) 1367 { 1368 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1369 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1370 tcp_probe0_base(sk), TCP_RTO_MAX); 1371 } 1372 1373 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1374 { 1375 tp->snd_wl1 = seq; 1376 } 1377 1378 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1379 { 1380 tp->snd_wl1 = seq; 1381 } 1382 1383 /* 1384 * Calculate(/check) TCP checksum 1385 */ 1386 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1387 __be32 daddr, __wsum base) 1388 { 1389 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1390 } 1391 1392 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1393 { 1394 return !skb_csum_unnecessary(skb) && 1395 __skb_checksum_complete(skb); 1396 } 1397 1398 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1399 enum skb_drop_reason *reason); 1400 1401 1402 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1403 void tcp_set_state(struct sock *sk, int state); 1404 void tcp_done(struct sock *sk); 1405 int tcp_abort(struct sock *sk, int err); 1406 1407 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1408 { 1409 rx_opt->dsack = 0; 1410 rx_opt->num_sacks = 0; 1411 } 1412 1413 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1414 1415 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1416 { 1417 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1418 struct tcp_sock *tp = tcp_sk(sk); 1419 s32 delta; 1420 1421 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1422 tp->packets_out || ca_ops->cong_control) 1423 return; 1424 delta = tcp_jiffies32 - tp->lsndtime; 1425 if (delta > inet_csk(sk)->icsk_rto) 1426 tcp_cwnd_restart(sk, delta); 1427 } 1428 1429 /* Determine a window scaling and initial window to offer. */ 1430 void tcp_select_initial_window(const struct sock *sk, int __space, 1431 __u32 mss, __u32 *rcv_wnd, 1432 __u32 *window_clamp, int wscale_ok, 1433 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1434 1435 static inline int tcp_win_from_space(const struct sock *sk, int space) 1436 { 1437 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale); 1438 1439 return tcp_adv_win_scale <= 0 ? 1440 (space>>(-tcp_adv_win_scale)) : 1441 space - (space>>tcp_adv_win_scale); 1442 } 1443 1444 /* Note: caller must be prepared to deal with negative returns */ 1445 static inline int tcp_space(const struct sock *sk) 1446 { 1447 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1448 READ_ONCE(sk->sk_backlog.len) - 1449 atomic_read(&sk->sk_rmem_alloc)); 1450 } 1451 1452 static inline int tcp_full_space(const struct sock *sk) 1453 { 1454 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1455 } 1456 1457 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1458 { 1459 int unused_mem = sk_unused_reserved_mem(sk); 1460 struct tcp_sock *tp = tcp_sk(sk); 1461 1462 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1463 if (unused_mem) 1464 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1465 tcp_win_from_space(sk, unused_mem)); 1466 } 1467 1468 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1469 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1470 1471 1472 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1473 * If 87.5 % (7/8) of the space has been consumed, we want to override 1474 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1475 * len/truesize ratio. 1476 */ 1477 static inline bool tcp_rmem_pressure(const struct sock *sk) 1478 { 1479 int rcvbuf, threshold; 1480 1481 if (tcp_under_memory_pressure(sk)) 1482 return true; 1483 1484 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1485 threshold = rcvbuf - (rcvbuf >> 3); 1486 1487 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1488 } 1489 1490 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1491 { 1492 const struct tcp_sock *tp = tcp_sk(sk); 1493 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1494 1495 if (avail <= 0) 1496 return false; 1497 1498 return (avail >= target) || tcp_rmem_pressure(sk) || 1499 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1500 } 1501 1502 extern void tcp_openreq_init_rwin(struct request_sock *req, 1503 const struct sock *sk_listener, 1504 const struct dst_entry *dst); 1505 1506 void tcp_enter_memory_pressure(struct sock *sk); 1507 void tcp_leave_memory_pressure(struct sock *sk); 1508 1509 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1510 { 1511 struct net *net = sock_net((struct sock *)tp); 1512 int val; 1513 1514 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1515 * and do_tcp_setsockopt(). 1516 */ 1517 val = READ_ONCE(tp->keepalive_intvl); 1518 1519 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1520 } 1521 1522 static inline int keepalive_time_when(const struct tcp_sock *tp) 1523 { 1524 struct net *net = sock_net((struct sock *)tp); 1525 int val; 1526 1527 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1528 val = READ_ONCE(tp->keepalive_time); 1529 1530 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1531 } 1532 1533 static inline int keepalive_probes(const struct tcp_sock *tp) 1534 { 1535 struct net *net = sock_net((struct sock *)tp); 1536 int val; 1537 1538 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1539 * and do_tcp_setsockopt(). 1540 */ 1541 val = READ_ONCE(tp->keepalive_probes); 1542 1543 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1544 } 1545 1546 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1547 { 1548 const struct inet_connection_sock *icsk = &tp->inet_conn; 1549 1550 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1551 tcp_jiffies32 - tp->rcv_tstamp); 1552 } 1553 1554 static inline int tcp_fin_time(const struct sock *sk) 1555 { 1556 int fin_timeout = tcp_sk(sk)->linger2 ? : 1557 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1558 const int rto = inet_csk(sk)->icsk_rto; 1559 1560 if (fin_timeout < (rto << 2) - (rto >> 1)) 1561 fin_timeout = (rto << 2) - (rto >> 1); 1562 1563 return fin_timeout; 1564 } 1565 1566 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1567 int paws_win) 1568 { 1569 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1570 return true; 1571 if (unlikely(!time_before32(ktime_get_seconds(), 1572 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1573 return true; 1574 /* 1575 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1576 * then following tcp messages have valid values. Ignore 0 value, 1577 * or else 'negative' tsval might forbid us to accept their packets. 1578 */ 1579 if (!rx_opt->ts_recent) 1580 return true; 1581 return false; 1582 } 1583 1584 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1585 int rst) 1586 { 1587 if (tcp_paws_check(rx_opt, 0)) 1588 return false; 1589 1590 /* RST segments are not recommended to carry timestamp, 1591 and, if they do, it is recommended to ignore PAWS because 1592 "their cleanup function should take precedence over timestamps." 1593 Certainly, it is mistake. It is necessary to understand the reasons 1594 of this constraint to relax it: if peer reboots, clock may go 1595 out-of-sync and half-open connections will not be reset. 1596 Actually, the problem would be not existing if all 1597 the implementations followed draft about maintaining clock 1598 via reboots. Linux-2.2 DOES NOT! 1599 1600 However, we can relax time bounds for RST segments to MSL. 1601 */ 1602 if (rst && !time_before32(ktime_get_seconds(), 1603 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1604 return false; 1605 return true; 1606 } 1607 1608 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1609 int mib_idx, u32 *last_oow_ack_time); 1610 1611 static inline void tcp_mib_init(struct net *net) 1612 { 1613 /* See RFC 2012 */ 1614 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1615 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1616 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1617 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1618 } 1619 1620 /* from STCP */ 1621 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1622 { 1623 tp->lost_skb_hint = NULL; 1624 } 1625 1626 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1627 { 1628 tcp_clear_retrans_hints_partial(tp); 1629 tp->retransmit_skb_hint = NULL; 1630 } 1631 1632 union tcp_md5_addr { 1633 struct in_addr a4; 1634 #if IS_ENABLED(CONFIG_IPV6) 1635 struct in6_addr a6; 1636 #endif 1637 }; 1638 1639 /* - key database */ 1640 struct tcp_md5sig_key { 1641 struct hlist_node node; 1642 u8 keylen; 1643 u8 family; /* AF_INET or AF_INET6 */ 1644 u8 prefixlen; 1645 u8 flags; 1646 union tcp_md5_addr addr; 1647 int l3index; /* set if key added with L3 scope */ 1648 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1649 struct rcu_head rcu; 1650 }; 1651 1652 /* - sock block */ 1653 struct tcp_md5sig_info { 1654 struct hlist_head head; 1655 struct rcu_head rcu; 1656 }; 1657 1658 /* - pseudo header */ 1659 struct tcp4_pseudohdr { 1660 __be32 saddr; 1661 __be32 daddr; 1662 __u8 pad; 1663 __u8 protocol; 1664 __be16 len; 1665 }; 1666 1667 struct tcp6_pseudohdr { 1668 struct in6_addr saddr; 1669 struct in6_addr daddr; 1670 __be32 len; 1671 __be32 protocol; /* including padding */ 1672 }; 1673 1674 union tcp_md5sum_block { 1675 struct tcp4_pseudohdr ip4; 1676 #if IS_ENABLED(CONFIG_IPV6) 1677 struct tcp6_pseudohdr ip6; 1678 #endif 1679 }; 1680 1681 /* - pool: digest algorithm, hash description and scratch buffer */ 1682 struct tcp_md5sig_pool { 1683 struct ahash_request *md5_req; 1684 void *scratch; 1685 }; 1686 1687 /* - functions */ 1688 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1689 const struct sock *sk, const struct sk_buff *skb); 1690 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1691 int family, u8 prefixlen, int l3index, u8 flags, 1692 const u8 *newkey, u8 newkeylen); 1693 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1694 int family, u8 prefixlen, int l3index, 1695 struct tcp_md5sig_key *key); 1696 1697 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1698 int family, u8 prefixlen, int l3index, u8 flags); 1699 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1700 const struct sock *addr_sk); 1701 1702 #ifdef CONFIG_TCP_MD5SIG 1703 #include <linux/jump_label.h> 1704 extern struct static_key_false_deferred tcp_md5_needed; 1705 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1706 const union tcp_md5_addr *addr, 1707 int family); 1708 static inline struct tcp_md5sig_key * 1709 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1710 const union tcp_md5_addr *addr, int family) 1711 { 1712 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1713 return NULL; 1714 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1715 } 1716 1717 enum skb_drop_reason 1718 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1719 const void *saddr, const void *daddr, 1720 int family, int dif, int sdif); 1721 1722 1723 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1724 #else 1725 static inline struct tcp_md5sig_key * 1726 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1727 const union tcp_md5_addr *addr, int family) 1728 { 1729 return NULL; 1730 } 1731 1732 static inline enum skb_drop_reason 1733 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1734 const void *saddr, const void *daddr, 1735 int family, int dif, int sdif) 1736 { 1737 return SKB_NOT_DROPPED_YET; 1738 } 1739 #define tcp_twsk_md5_key(twsk) NULL 1740 #endif 1741 1742 bool tcp_alloc_md5sig_pool(void); 1743 1744 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1745 static inline void tcp_put_md5sig_pool(void) 1746 { 1747 local_bh_enable(); 1748 } 1749 1750 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1751 unsigned int header_len); 1752 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1753 const struct tcp_md5sig_key *key); 1754 1755 /* From tcp_fastopen.c */ 1756 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1757 struct tcp_fastopen_cookie *cookie); 1758 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1759 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1760 u16 try_exp); 1761 struct tcp_fastopen_request { 1762 /* Fast Open cookie. Size 0 means a cookie request */ 1763 struct tcp_fastopen_cookie cookie; 1764 struct msghdr *data; /* data in MSG_FASTOPEN */ 1765 size_t size; 1766 int copied; /* queued in tcp_connect() */ 1767 struct ubuf_info *uarg; 1768 }; 1769 void tcp_free_fastopen_req(struct tcp_sock *tp); 1770 void tcp_fastopen_destroy_cipher(struct sock *sk); 1771 void tcp_fastopen_ctx_destroy(struct net *net); 1772 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1773 void *primary_key, void *backup_key); 1774 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1775 u64 *key); 1776 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1777 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1778 struct request_sock *req, 1779 struct tcp_fastopen_cookie *foc, 1780 const struct dst_entry *dst); 1781 void tcp_fastopen_init_key_once(struct net *net); 1782 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1783 struct tcp_fastopen_cookie *cookie); 1784 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1785 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1786 #define TCP_FASTOPEN_KEY_MAX 2 1787 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1788 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1789 1790 /* Fastopen key context */ 1791 struct tcp_fastopen_context { 1792 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1793 int num; 1794 struct rcu_head rcu; 1795 }; 1796 1797 void tcp_fastopen_active_disable(struct sock *sk); 1798 bool tcp_fastopen_active_should_disable(struct sock *sk); 1799 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1800 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1801 1802 /* Caller needs to wrap with rcu_read_(un)lock() */ 1803 static inline 1804 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1805 { 1806 struct tcp_fastopen_context *ctx; 1807 1808 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1809 if (!ctx) 1810 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1811 return ctx; 1812 } 1813 1814 static inline 1815 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1816 const struct tcp_fastopen_cookie *orig) 1817 { 1818 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1819 orig->len == foc->len && 1820 !memcmp(orig->val, foc->val, foc->len)) 1821 return true; 1822 return false; 1823 } 1824 1825 static inline 1826 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1827 { 1828 return ctx->num; 1829 } 1830 1831 /* Latencies incurred by various limits for a sender. They are 1832 * chronograph-like stats that are mutually exclusive. 1833 */ 1834 enum tcp_chrono { 1835 TCP_CHRONO_UNSPEC, 1836 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1837 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1838 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1839 __TCP_CHRONO_MAX, 1840 }; 1841 1842 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1843 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1844 1845 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1846 * the same memory storage than skb->destructor/_skb_refdst 1847 */ 1848 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1849 { 1850 skb->destructor = NULL; 1851 skb->_skb_refdst = 0UL; 1852 } 1853 1854 #define tcp_skb_tsorted_save(skb) { \ 1855 unsigned long _save = skb->_skb_refdst; \ 1856 skb->_skb_refdst = 0UL; 1857 1858 #define tcp_skb_tsorted_restore(skb) \ 1859 skb->_skb_refdst = _save; \ 1860 } 1861 1862 void tcp_write_queue_purge(struct sock *sk); 1863 1864 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1865 { 1866 return skb_rb_first(&sk->tcp_rtx_queue); 1867 } 1868 1869 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1870 { 1871 return skb_rb_last(&sk->tcp_rtx_queue); 1872 } 1873 1874 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1875 { 1876 return skb_peek_tail(&sk->sk_write_queue); 1877 } 1878 1879 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1880 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1881 1882 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1883 { 1884 return skb_peek(&sk->sk_write_queue); 1885 } 1886 1887 static inline bool tcp_skb_is_last(const struct sock *sk, 1888 const struct sk_buff *skb) 1889 { 1890 return skb_queue_is_last(&sk->sk_write_queue, skb); 1891 } 1892 1893 /** 1894 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1895 * @sk: socket 1896 * 1897 * Since the write queue can have a temporary empty skb in it, 1898 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1899 */ 1900 static inline bool tcp_write_queue_empty(const struct sock *sk) 1901 { 1902 const struct tcp_sock *tp = tcp_sk(sk); 1903 1904 return tp->write_seq == tp->snd_nxt; 1905 } 1906 1907 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1908 { 1909 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1910 } 1911 1912 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1913 { 1914 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1915 } 1916 1917 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1918 { 1919 __skb_queue_tail(&sk->sk_write_queue, skb); 1920 1921 /* Queue it, remembering where we must start sending. */ 1922 if (sk->sk_write_queue.next == skb) 1923 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1924 } 1925 1926 /* Insert new before skb on the write queue of sk. */ 1927 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1928 struct sk_buff *skb, 1929 struct sock *sk) 1930 { 1931 __skb_queue_before(&sk->sk_write_queue, skb, new); 1932 } 1933 1934 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1935 { 1936 tcp_skb_tsorted_anchor_cleanup(skb); 1937 __skb_unlink(skb, &sk->sk_write_queue); 1938 } 1939 1940 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1941 1942 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1943 { 1944 tcp_skb_tsorted_anchor_cleanup(skb); 1945 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1946 } 1947 1948 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1949 { 1950 list_del(&skb->tcp_tsorted_anchor); 1951 tcp_rtx_queue_unlink(skb, sk); 1952 tcp_wmem_free_skb(sk, skb); 1953 } 1954 1955 static inline void tcp_push_pending_frames(struct sock *sk) 1956 { 1957 if (tcp_send_head(sk)) { 1958 struct tcp_sock *tp = tcp_sk(sk); 1959 1960 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1961 } 1962 } 1963 1964 /* Start sequence of the skb just after the highest skb with SACKed 1965 * bit, valid only if sacked_out > 0 or when the caller has ensured 1966 * validity by itself. 1967 */ 1968 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1969 { 1970 if (!tp->sacked_out) 1971 return tp->snd_una; 1972 1973 if (tp->highest_sack == NULL) 1974 return tp->snd_nxt; 1975 1976 return TCP_SKB_CB(tp->highest_sack)->seq; 1977 } 1978 1979 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1980 { 1981 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1982 } 1983 1984 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1985 { 1986 return tcp_sk(sk)->highest_sack; 1987 } 1988 1989 static inline void tcp_highest_sack_reset(struct sock *sk) 1990 { 1991 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1992 } 1993 1994 /* Called when old skb is about to be deleted and replaced by new skb */ 1995 static inline void tcp_highest_sack_replace(struct sock *sk, 1996 struct sk_buff *old, 1997 struct sk_buff *new) 1998 { 1999 if (old == tcp_highest_sack(sk)) 2000 tcp_sk(sk)->highest_sack = new; 2001 } 2002 2003 /* This helper checks if socket has IP_TRANSPARENT set */ 2004 static inline bool inet_sk_transparent(const struct sock *sk) 2005 { 2006 switch (sk->sk_state) { 2007 case TCP_TIME_WAIT: 2008 return inet_twsk(sk)->tw_transparent; 2009 case TCP_NEW_SYN_RECV: 2010 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2011 } 2012 return inet_sk(sk)->transparent; 2013 } 2014 2015 /* Determines whether this is a thin stream (which may suffer from 2016 * increased latency). Used to trigger latency-reducing mechanisms. 2017 */ 2018 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2019 { 2020 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2021 } 2022 2023 /* /proc */ 2024 enum tcp_seq_states { 2025 TCP_SEQ_STATE_LISTENING, 2026 TCP_SEQ_STATE_ESTABLISHED, 2027 }; 2028 2029 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2030 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2031 void tcp_seq_stop(struct seq_file *seq, void *v); 2032 2033 struct tcp_seq_afinfo { 2034 sa_family_t family; 2035 }; 2036 2037 struct tcp_iter_state { 2038 struct seq_net_private p; 2039 enum tcp_seq_states state; 2040 struct sock *syn_wait_sk; 2041 int bucket, offset, sbucket, num; 2042 loff_t last_pos; 2043 }; 2044 2045 extern struct request_sock_ops tcp_request_sock_ops; 2046 extern struct request_sock_ops tcp6_request_sock_ops; 2047 2048 void tcp_v4_destroy_sock(struct sock *sk); 2049 2050 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2051 netdev_features_t features); 2052 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2053 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2054 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2055 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2056 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2057 void tcp_gro_complete(struct sk_buff *skb); 2058 2059 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2060 2061 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2062 { 2063 struct net *net = sock_net((struct sock *)tp); 2064 u32 val; 2065 2066 val = READ_ONCE(tp->notsent_lowat); 2067 2068 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2069 } 2070 2071 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2072 2073 #ifdef CONFIG_PROC_FS 2074 int tcp4_proc_init(void); 2075 void tcp4_proc_exit(void); 2076 #endif 2077 2078 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2079 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2080 const struct tcp_request_sock_ops *af_ops, 2081 struct sock *sk, struct sk_buff *skb); 2082 2083 /* TCP af-specific functions */ 2084 struct tcp_sock_af_ops { 2085 #ifdef CONFIG_TCP_MD5SIG 2086 struct tcp_md5sig_key *(*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 int (*md5_parse)(struct sock *sk, 2093 int optname, 2094 sockptr_t optval, 2095 int optlen); 2096 #endif 2097 }; 2098 2099 struct tcp_request_sock_ops { 2100 u16 mss_clamp; 2101 #ifdef CONFIG_TCP_MD5SIG 2102 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2103 const struct sock *addr_sk); 2104 int (*calc_md5_hash) (char *location, 2105 const struct tcp_md5sig_key *md5, 2106 const struct sock *sk, 2107 const struct sk_buff *skb); 2108 #endif 2109 #ifdef CONFIG_SYN_COOKIES 2110 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2111 __u16 *mss); 2112 #endif 2113 struct dst_entry *(*route_req)(const struct sock *sk, 2114 struct sk_buff *skb, 2115 struct flowi *fl, 2116 struct request_sock *req); 2117 u32 (*init_seq)(const struct sk_buff *skb); 2118 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2119 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2120 struct flowi *fl, struct request_sock *req, 2121 struct tcp_fastopen_cookie *foc, 2122 enum tcp_synack_type synack_type, 2123 struct sk_buff *syn_skb); 2124 }; 2125 2126 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2127 #if IS_ENABLED(CONFIG_IPV6) 2128 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2129 #endif 2130 2131 #ifdef CONFIG_SYN_COOKIES 2132 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2133 const struct sock *sk, struct sk_buff *skb, 2134 __u16 *mss) 2135 { 2136 tcp_synq_overflow(sk); 2137 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2138 return ops->cookie_init_seq(skb, mss); 2139 } 2140 #else 2141 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2142 const struct sock *sk, struct sk_buff *skb, 2143 __u16 *mss) 2144 { 2145 return 0; 2146 } 2147 #endif 2148 2149 int tcpv4_offload_init(void); 2150 2151 void tcp_v4_init(void); 2152 void tcp_init(void); 2153 2154 /* tcp_recovery.c */ 2155 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2156 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2157 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2158 u32 reo_wnd); 2159 extern bool tcp_rack_mark_lost(struct sock *sk); 2160 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2161 u64 xmit_time); 2162 extern void tcp_rack_reo_timeout(struct sock *sk); 2163 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2164 2165 /* tcp_plb.c */ 2166 2167 /* 2168 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2169 * expects cong_ratio which represents fraction of traffic that experienced 2170 * congestion over a single RTT. In order to avoid floating point operations, 2171 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2172 */ 2173 #define TCP_PLB_SCALE 8 2174 2175 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2176 struct tcp_plb_state { 2177 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2178 unused:3; 2179 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2180 }; 2181 2182 static inline void tcp_plb_init(const struct sock *sk, 2183 struct tcp_plb_state *plb) 2184 { 2185 plb->consec_cong_rounds = 0; 2186 plb->pause_until = 0; 2187 } 2188 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2189 const int cong_ratio); 2190 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2191 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2192 2193 /* At how many usecs into the future should the RTO fire? */ 2194 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2195 { 2196 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2197 u32 rto = inet_csk(sk)->icsk_rto; 2198 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2199 2200 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2201 } 2202 2203 /* 2204 * Save and compile IPv4 options, return a pointer to it 2205 */ 2206 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2207 struct sk_buff *skb) 2208 { 2209 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2210 struct ip_options_rcu *dopt = NULL; 2211 2212 if (opt->optlen) { 2213 int opt_size = sizeof(*dopt) + opt->optlen; 2214 2215 dopt = kmalloc(opt_size, GFP_ATOMIC); 2216 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2217 kfree(dopt); 2218 dopt = NULL; 2219 } 2220 } 2221 return dopt; 2222 } 2223 2224 /* locally generated TCP pure ACKs have skb->truesize == 2 2225 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2226 * This is much faster than dissecting the packet to find out. 2227 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2228 */ 2229 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2230 { 2231 return skb->truesize == 2; 2232 } 2233 2234 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2235 { 2236 skb->truesize = 2; 2237 } 2238 2239 static inline int tcp_inq(struct sock *sk) 2240 { 2241 struct tcp_sock *tp = tcp_sk(sk); 2242 int answ; 2243 2244 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2245 answ = 0; 2246 } else if (sock_flag(sk, SOCK_URGINLINE) || 2247 !tp->urg_data || 2248 before(tp->urg_seq, tp->copied_seq) || 2249 !before(tp->urg_seq, tp->rcv_nxt)) { 2250 2251 answ = tp->rcv_nxt - tp->copied_seq; 2252 2253 /* Subtract 1, if FIN was received */ 2254 if (answ && sock_flag(sk, SOCK_DONE)) 2255 answ--; 2256 } else { 2257 answ = tp->urg_seq - tp->copied_seq; 2258 } 2259 2260 return answ; 2261 } 2262 2263 int tcp_peek_len(struct socket *sock); 2264 2265 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2266 { 2267 u16 segs_in; 2268 2269 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2270 2271 /* We update these fields while other threads might 2272 * read them from tcp_get_info() 2273 */ 2274 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2275 if (skb->len > tcp_hdrlen(skb)) 2276 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2277 } 2278 2279 /* 2280 * TCP listen path runs lockless. 2281 * We forced "struct sock" to be const qualified to make sure 2282 * we don't modify one of its field by mistake. 2283 * Here, we increment sk_drops which is an atomic_t, so we can safely 2284 * make sock writable again. 2285 */ 2286 static inline void tcp_listendrop(const struct sock *sk) 2287 { 2288 atomic_inc(&((struct sock *)sk)->sk_drops); 2289 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2290 } 2291 2292 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2293 2294 /* 2295 * Interface for adding Upper Level Protocols over TCP 2296 */ 2297 2298 #define TCP_ULP_NAME_MAX 16 2299 #define TCP_ULP_MAX 128 2300 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2301 2302 struct tcp_ulp_ops { 2303 struct list_head list; 2304 2305 /* initialize ulp */ 2306 int (*init)(struct sock *sk); 2307 /* update ulp */ 2308 void (*update)(struct sock *sk, struct proto *p, 2309 void (*write_space)(struct sock *sk)); 2310 /* cleanup ulp */ 2311 void (*release)(struct sock *sk); 2312 /* diagnostic */ 2313 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2314 size_t (*get_info_size)(const struct sock *sk); 2315 /* clone ulp */ 2316 void (*clone)(const struct request_sock *req, struct sock *newsk, 2317 const gfp_t priority); 2318 2319 char name[TCP_ULP_NAME_MAX]; 2320 struct module *owner; 2321 }; 2322 int tcp_register_ulp(struct tcp_ulp_ops *type); 2323 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2324 int tcp_set_ulp(struct sock *sk, const char *name); 2325 void tcp_get_available_ulp(char *buf, size_t len); 2326 void tcp_cleanup_ulp(struct sock *sk); 2327 void tcp_update_ulp(struct sock *sk, struct proto *p, 2328 void (*write_space)(struct sock *sk)); 2329 2330 #define MODULE_ALIAS_TCP_ULP(name) \ 2331 __MODULE_INFO(alias, alias_userspace, name); \ 2332 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2333 2334 #ifdef CONFIG_NET_SOCK_MSG 2335 struct sk_msg; 2336 struct sk_psock; 2337 2338 #ifdef CONFIG_BPF_SYSCALL 2339 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2340 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2341 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2342 #endif /* CONFIG_BPF_SYSCALL */ 2343 2344 #ifdef CONFIG_INET 2345 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2346 #else 2347 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2348 { 2349 } 2350 #endif 2351 2352 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2353 struct sk_msg *msg, u32 bytes, int flags); 2354 #endif /* CONFIG_NET_SOCK_MSG */ 2355 2356 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2357 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2358 { 2359 } 2360 #endif 2361 2362 #ifdef CONFIG_CGROUP_BPF 2363 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2364 struct sk_buff *skb, 2365 unsigned int end_offset) 2366 { 2367 skops->skb = skb; 2368 skops->skb_data_end = skb->data + end_offset; 2369 } 2370 #else 2371 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2372 struct sk_buff *skb, 2373 unsigned int end_offset) 2374 { 2375 } 2376 #endif 2377 2378 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2379 * is < 0, then the BPF op failed (for example if the loaded BPF 2380 * program does not support the chosen operation or there is no BPF 2381 * program loaded). 2382 */ 2383 #ifdef CONFIG_BPF 2384 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2385 { 2386 struct bpf_sock_ops_kern sock_ops; 2387 int ret; 2388 2389 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2390 if (sk_fullsock(sk)) { 2391 sock_ops.is_fullsock = 1; 2392 sock_owned_by_me(sk); 2393 } 2394 2395 sock_ops.sk = sk; 2396 sock_ops.op = op; 2397 if (nargs > 0) 2398 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2399 2400 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2401 if (ret == 0) 2402 ret = sock_ops.reply; 2403 else 2404 ret = -1; 2405 return ret; 2406 } 2407 2408 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2409 { 2410 u32 args[2] = {arg1, arg2}; 2411 2412 return tcp_call_bpf(sk, op, 2, args); 2413 } 2414 2415 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2416 u32 arg3) 2417 { 2418 u32 args[3] = {arg1, arg2, arg3}; 2419 2420 return tcp_call_bpf(sk, op, 3, args); 2421 } 2422 2423 #else 2424 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2425 { 2426 return -EPERM; 2427 } 2428 2429 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2430 { 2431 return -EPERM; 2432 } 2433 2434 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2435 u32 arg3) 2436 { 2437 return -EPERM; 2438 } 2439 2440 #endif 2441 2442 static inline u32 tcp_timeout_init(struct sock *sk) 2443 { 2444 int timeout; 2445 2446 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2447 2448 if (timeout <= 0) 2449 timeout = TCP_TIMEOUT_INIT; 2450 return min_t(int, timeout, TCP_RTO_MAX); 2451 } 2452 2453 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2454 { 2455 int rwnd; 2456 2457 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2458 2459 if (rwnd < 0) 2460 rwnd = 0; 2461 return rwnd; 2462 } 2463 2464 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2465 { 2466 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2467 } 2468 2469 static inline void tcp_bpf_rtt(struct sock *sk) 2470 { 2471 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2472 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2473 } 2474 2475 #if IS_ENABLED(CONFIG_SMC) 2476 extern struct static_key_false tcp_have_smc; 2477 #endif 2478 2479 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2480 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2481 void (*cad)(struct sock *sk, u32 ack_seq)); 2482 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2483 void clean_acked_data_flush(void); 2484 #endif 2485 2486 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2487 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2488 const struct tcp_sock *tp) 2489 { 2490 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2491 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2492 } 2493 2494 /* Compute Earliest Departure Time for some control packets 2495 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2496 */ 2497 static inline u64 tcp_transmit_time(const struct sock *sk) 2498 { 2499 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2500 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2501 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2502 2503 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2504 } 2505 return 0; 2506 } 2507 2508 #endif /* _TCP_H */ 2509