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