1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP module. 8 * 9 * Version: @(#)tcp.h 1.0.5 05/23/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 */ 14 #ifndef _TCP_H 15 #define _TCP_H 16 17 #define FASTRETRANS_DEBUG 1 18 19 #include <linux/list.h> 20 #include <linux/tcp.h> 21 #include <linux/bug.h> 22 #include <linux/slab.h> 23 #include <linux/cache.h> 24 #include <linux/percpu.h> 25 #include <linux/skbuff.h> 26 #include <linux/kref.h> 27 #include <linux/ktime.h> 28 #include <linux/indirect_call_wrapper.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/inet_ecn.h> 41 #include <net/dst.h> 42 #include <net/mptcp.h> 43 44 #include <linux/seq_file.h> 45 #include <linux/memcontrol.h> 46 #include <linux/bpf-cgroup.h> 47 #include <linux/siphash.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 52 int tcp_orphan_count_sum(void); 53 54 void tcp_time_wait(struct sock *sk, int state, int timeo); 55 56 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 57 #define MAX_TCP_OPTION_SPACE 40 58 #define TCP_MIN_SND_MSS 48 59 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 60 61 /* 62 * Never offer a window over 32767 without using window scaling. Some 63 * poor stacks do signed 16bit maths! 64 */ 65 #define MAX_TCP_WINDOW 32767U 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The initial MTU to use for probing */ 71 #define TCP_BASE_MSS 1024 72 73 /* probing interval, default to 10 minutes as per RFC4821 */ 74 #define TCP_PROBE_INTERVAL 600 75 76 /* Specify interval when tcp mtu probing will stop */ 77 #define TCP_PROBE_THRESHOLD 8 78 79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 80 #define TCP_FASTRETRANS_THRESH 3 81 82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 83 #define TCP_MAX_QUICKACKS 16U 84 85 /* Maximal number of window scale according to RFC1323 */ 86 #define TCP_MAX_WSCALE 14U 87 88 /* urg_data states */ 89 #define TCP_URG_VALID 0x0100 90 #define TCP_URG_NOTYET 0x0200 91 #define TCP_URG_READ 0x0400 92 93 #define TCP_RETR1 3 /* 94 * This is how many retries it does before it 95 * tries to figure out if the gateway is 96 * down. Minimal RFC value is 3; it corresponds 97 * to ~3sec-8min depending on RTO. 98 */ 99 100 #define TCP_RETR2 15 /* 101 * This should take at least 102 * 90 minutes to time out. 103 * RFC1122 says that the limit is 100 sec. 104 * 15 is ~13-30min depending on RTO. 105 */ 106 107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 108 * when active opening a connection. 109 * RFC1122 says the minimum retry MUST 110 * be at least 180secs. Nevertheless 111 * this value is corresponding to 112 * 63secs of retransmission with the 113 * current initial RTO. 114 */ 115 116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 117 * when passive opening a connection. 118 * This is corresponding to 31secs of 119 * retransmission with the current 120 * initial RTO. 121 */ 122 123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 124 * state, about 60 seconds */ 125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 126 /* BSD style FIN_WAIT2 deadlock breaker. 127 * It used to be 3min, new value is 60sec, 128 * to combine FIN-WAIT-2 timeout with 129 * TIME-WAIT timer. 130 */ 131 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 132 133 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 134 #if HZ >= 100 135 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 136 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 137 #else 138 #define TCP_DELACK_MIN 4U 139 #define TCP_ATO_MIN 4U 140 #endif 141 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 142 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 143 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 144 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 145 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 146 * used as a fallback RTO for the 147 * initial data transmission if no 148 * valid RTT sample has been acquired, 149 * most likely due to retrans in 3WHS. 150 */ 151 152 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 153 * for local resources. 154 */ 155 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 156 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 157 #define TCP_KEEPALIVE_INTVL (75*HZ) 158 159 #define MAX_TCP_KEEPIDLE 32767 160 #define MAX_TCP_KEEPINTVL 32767 161 #define MAX_TCP_KEEPCNT 127 162 #define MAX_TCP_SYNCNT 127 163 164 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 165 166 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 167 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 168 * after this time. It should be equal 169 * (or greater than) TCP_TIMEWAIT_LEN 170 * to provide reliability equal to one 171 * provided by timewait state. 172 */ 173 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 174 * timestamps. It must be less than 175 * minimal timewait lifetime. 176 */ 177 /* 178 * TCP option 179 */ 180 181 #define TCPOPT_NOP 1 /* Padding */ 182 #define TCPOPT_EOL 0 /* End of options */ 183 #define TCPOPT_MSS 2 /* Segment size negotiating */ 184 #define TCPOPT_WINDOW 3 /* Window scaling */ 185 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 186 #define TCPOPT_SACK 5 /* SACK Block */ 187 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 188 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 189 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 190 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 191 #define TCPOPT_EXP 254 /* Experimental */ 192 /* Magic number to be after the option value for sharing TCP 193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 194 */ 195 #define TCPOPT_FASTOPEN_MAGIC 0xF989 196 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 197 198 /* 199 * TCP option lengths 200 */ 201 202 #define TCPOLEN_MSS 4 203 #define TCPOLEN_WINDOW 3 204 #define TCPOLEN_SACK_PERM 2 205 #define TCPOLEN_TIMESTAMP 10 206 #define TCPOLEN_MD5SIG 18 207 #define TCPOLEN_FASTOPEN_BASE 2 208 #define TCPOLEN_EXP_FASTOPEN_BASE 4 209 #define TCPOLEN_EXP_SMC_BASE 6 210 211 /* But this is what stacks really send out. */ 212 #define TCPOLEN_TSTAMP_ALIGNED 12 213 #define TCPOLEN_WSCALE_ALIGNED 4 214 #define TCPOLEN_SACKPERM_ALIGNED 4 215 #define TCPOLEN_SACK_BASE 2 216 #define TCPOLEN_SACK_BASE_ALIGNED 4 217 #define TCPOLEN_SACK_PERBLOCK 8 218 #define TCPOLEN_MD5SIG_ALIGNED 20 219 #define TCPOLEN_MSS_ALIGNED 4 220 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 221 222 /* Flags in tp->nonagle */ 223 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 224 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 225 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 226 227 /* TCP thin-stream limits */ 228 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 229 230 /* TCP initial congestion window as per rfc6928 */ 231 #define TCP_INIT_CWND 10 232 233 /* Bit Flags for sysctl_tcp_fastopen */ 234 #define TFO_CLIENT_ENABLE 1 235 #define TFO_SERVER_ENABLE 2 236 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 237 238 /* Accept SYN data w/o any cookie option */ 239 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 240 241 /* Force enable TFO on all listeners, i.e., not requiring the 242 * TCP_FASTOPEN socket option. 243 */ 244 #define TFO_SERVER_WO_SOCKOPT1 0x400 245 246 247 /* sysctl variables for tcp */ 248 extern int sysctl_tcp_max_orphans; 249 extern long sysctl_tcp_mem[3]; 250 251 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 252 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 253 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 254 255 extern atomic_long_t tcp_memory_allocated; 256 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 257 258 extern struct percpu_counter tcp_sockets_allocated; 259 extern unsigned long tcp_memory_pressure; 260 261 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 262 static inline bool tcp_under_memory_pressure(const struct sock *sk) 263 { 264 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 265 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 266 return true; 267 268 return READ_ONCE(tcp_memory_pressure); 269 } 270 /* 271 * The next routines deal with comparing 32 bit unsigned ints 272 * and worry about wraparound (automatic with unsigned arithmetic). 273 */ 274 275 static inline bool before(__u32 seq1, __u32 seq2) 276 { 277 return (__s32)(seq1-seq2) < 0; 278 } 279 #define after(seq2, seq1) before(seq1, seq2) 280 281 /* is s2<=s1<=s3 ? */ 282 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 283 { 284 return seq3 - seq2 >= seq1 - seq2; 285 } 286 287 static inline bool tcp_out_of_memory(struct sock *sk) 288 { 289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 291 return true; 292 return false; 293 } 294 295 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 296 { 297 sk_wmem_queued_add(sk, -skb->truesize); 298 if (!skb_zcopy_pure(skb)) 299 sk_mem_uncharge(sk, skb->truesize); 300 else 301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 302 __kfree_skb(skb); 303 } 304 305 void sk_forced_mem_schedule(struct sock *sk, int size); 306 307 bool tcp_check_oom(struct sock *sk, int shift); 308 309 310 extern struct proto tcp_prot; 311 312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 313 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 316 317 void tcp_tasklet_init(void); 318 319 int tcp_v4_err(struct sk_buff *skb, u32); 320 321 void tcp_shutdown(struct sock *sk, int how); 322 323 int tcp_v4_early_demux(struct sk_buff *skb); 324 int tcp_v4_rcv(struct sk_buff *skb); 325 326 void tcp_remove_empty_skb(struct sock *sk); 327 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 328 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 329 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 330 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 331 int flags); 332 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 333 size_t size, int flags); 334 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 335 size_t size, int flags); 336 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 337 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 338 int size_goal); 339 void tcp_release_cb(struct sock *sk); 340 void tcp_wfree(struct sk_buff *skb); 341 void tcp_write_timer_handler(struct sock *sk); 342 void tcp_delack_timer_handler(struct sock *sk); 343 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 344 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 345 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 346 void tcp_rcv_space_adjust(struct sock *sk); 347 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 348 void tcp_twsk_destructor(struct sock *sk); 349 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 350 struct pipe_inode_info *pipe, size_t len, 351 unsigned int flags); 352 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 353 bool force_schedule); 354 355 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 356 static inline void tcp_dec_quickack_mode(struct sock *sk, 357 const unsigned int pkts) 358 { 359 struct inet_connection_sock *icsk = inet_csk(sk); 360 361 if (icsk->icsk_ack.quick) { 362 if (pkts >= icsk->icsk_ack.quick) { 363 icsk->icsk_ack.quick = 0; 364 /* Leaving quickack mode we deflate ATO. */ 365 icsk->icsk_ack.ato = TCP_ATO_MIN; 366 } else 367 icsk->icsk_ack.quick -= pkts; 368 } 369 } 370 371 #define TCP_ECN_OK 1 372 #define TCP_ECN_QUEUE_CWR 2 373 #define TCP_ECN_DEMAND_CWR 4 374 #define TCP_ECN_SEEN 8 375 376 enum tcp_tw_status { 377 TCP_TW_SUCCESS = 0, 378 TCP_TW_RST = 1, 379 TCP_TW_ACK = 2, 380 TCP_TW_SYN = 3 381 }; 382 383 384 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 385 struct sk_buff *skb, 386 const struct tcphdr *th); 387 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 388 struct request_sock *req, bool fastopen, 389 bool *lost_race); 390 int tcp_child_process(struct sock *parent, struct sock *child, 391 struct sk_buff *skb); 392 void tcp_enter_loss(struct sock *sk); 393 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 394 void tcp_clear_retrans(struct tcp_sock *tp); 395 void tcp_update_metrics(struct sock *sk); 396 void tcp_init_metrics(struct sock *sk); 397 void tcp_metrics_init(void); 398 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 399 void __tcp_close(struct sock *sk, long timeout); 400 void tcp_close(struct sock *sk, long timeout); 401 void tcp_init_sock(struct sock *sk); 402 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 403 __poll_t tcp_poll(struct file *file, struct socket *sock, 404 struct poll_table_struct *wait); 405 int do_tcp_getsockopt(struct sock *sk, int level, 406 int optname, sockptr_t optval, sockptr_t optlen); 407 int tcp_getsockopt(struct sock *sk, int level, int optname, 408 char __user *optval, int __user *optlen); 409 bool tcp_bpf_bypass_getsockopt(int level, int optname); 410 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 411 sockptr_t optval, unsigned int optlen); 412 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 413 unsigned int optlen); 414 void tcp_set_keepalive(struct sock *sk, int val); 415 void tcp_syn_ack_timeout(const struct request_sock *req); 416 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 417 int flags, int *addr_len); 418 int tcp_set_rcvlowat(struct sock *sk, int val); 419 int tcp_set_window_clamp(struct sock *sk, int val); 420 void tcp_update_recv_tstamps(struct sk_buff *skb, 421 struct scm_timestamping_internal *tss); 422 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 423 struct scm_timestamping_internal *tss); 424 void tcp_data_ready(struct sock *sk); 425 #ifdef CONFIG_MMU 426 int tcp_mmap(struct file *file, struct socket *sock, 427 struct vm_area_struct *vma); 428 #endif 429 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 430 struct tcp_options_received *opt_rx, 431 int estab, struct tcp_fastopen_cookie *foc); 432 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 433 434 /* 435 * BPF SKB-less helpers 436 */ 437 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 438 struct tcphdr *th, u32 *cookie); 439 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 440 struct tcphdr *th, u32 *cookie); 441 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 442 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 443 const struct tcp_request_sock_ops *af_ops, 444 struct sock *sk, struct tcphdr *th); 445 /* 446 * TCP v4 functions exported for the inet6 API 447 */ 448 449 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 450 void tcp_v4_mtu_reduced(struct sock *sk); 451 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 452 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 453 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 454 struct sock *tcp_create_openreq_child(const struct sock *sk, 455 struct request_sock *req, 456 struct sk_buff *skb); 457 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 458 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 459 struct request_sock *req, 460 struct dst_entry *dst, 461 struct request_sock *req_unhash, 462 bool *own_req); 463 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 464 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 465 int tcp_connect(struct sock *sk); 466 enum tcp_synack_type { 467 TCP_SYNACK_NORMAL, 468 TCP_SYNACK_FASTOPEN, 469 TCP_SYNACK_COOKIE, 470 }; 471 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 472 struct request_sock *req, 473 struct tcp_fastopen_cookie *foc, 474 enum tcp_synack_type synack_type, 475 struct sk_buff *syn_skb); 476 int tcp_disconnect(struct sock *sk, int flags); 477 478 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 479 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 480 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 481 482 /* From syncookies.c */ 483 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 484 struct request_sock *req, 485 struct dst_entry *dst, u32 tsoff); 486 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 487 u32 cookie); 488 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 489 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 490 const struct tcp_request_sock_ops *af_ops, 491 struct sock *sk, struct sk_buff *skb); 492 #ifdef CONFIG_SYN_COOKIES 493 494 /* Syncookies use a monotonic timer which increments every 60 seconds. 495 * This counter is used both as a hash input and partially encoded into 496 * the cookie value. A cookie is only validated further if the delta 497 * between the current counter value and the encoded one is less than this, 498 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 499 * the counter advances immediately after a cookie is generated). 500 */ 501 #define MAX_SYNCOOKIE_AGE 2 502 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 503 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 504 505 /* syncookies: remember time of last synqueue overflow 506 * But do not dirty this field too often (once per second is enough) 507 * It is racy as we do not hold a lock, but race is very minor. 508 */ 509 static inline void tcp_synq_overflow(const struct sock *sk) 510 { 511 unsigned int last_overflow; 512 unsigned int now = jiffies; 513 514 if (sk->sk_reuseport) { 515 struct sock_reuseport *reuse; 516 517 reuse = rcu_dereference(sk->sk_reuseport_cb); 518 if (likely(reuse)) { 519 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 520 if (!time_between32(now, last_overflow, 521 last_overflow + HZ)) 522 WRITE_ONCE(reuse->synq_overflow_ts, now); 523 return; 524 } 525 } 526 527 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 528 if (!time_between32(now, last_overflow, last_overflow + HZ)) 529 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 530 } 531 532 /* syncookies: no recent synqueue overflow on this listening socket? */ 533 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 534 { 535 unsigned int last_overflow; 536 unsigned int now = jiffies; 537 538 if (sk->sk_reuseport) { 539 struct sock_reuseport *reuse; 540 541 reuse = rcu_dereference(sk->sk_reuseport_cb); 542 if (likely(reuse)) { 543 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 544 return !time_between32(now, last_overflow - HZ, 545 last_overflow + 546 TCP_SYNCOOKIE_VALID); 547 } 548 } 549 550 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 551 552 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 553 * then we're under synflood. However, we have to use 554 * 'last_overflow - HZ' as lower bound. That's because a concurrent 555 * tcp_synq_overflow() could update .ts_recent_stamp after we read 556 * jiffies but before we store .ts_recent_stamp into last_overflow, 557 * which could lead to rejecting a valid syncookie. 558 */ 559 return !time_between32(now, last_overflow - HZ, 560 last_overflow + TCP_SYNCOOKIE_VALID); 561 } 562 563 static inline u32 tcp_cookie_time(void) 564 { 565 u64 val = get_jiffies_64(); 566 567 do_div(val, TCP_SYNCOOKIE_PERIOD); 568 return val; 569 } 570 571 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 572 u16 *mssp); 573 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 574 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 575 bool cookie_timestamp_decode(const struct net *net, 576 struct tcp_options_received *opt); 577 bool cookie_ecn_ok(const struct tcp_options_received *opt, 578 const struct net *net, const struct dst_entry *dst); 579 580 /* From net/ipv6/syncookies.c */ 581 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 582 u32 cookie); 583 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 584 585 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 586 const struct tcphdr *th, u16 *mssp); 587 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 588 #endif 589 /* tcp_output.c */ 590 591 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 592 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 593 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 594 int nonagle); 595 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 596 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 597 void tcp_retransmit_timer(struct sock *sk); 598 void tcp_xmit_retransmit_queue(struct sock *); 599 void tcp_simple_retransmit(struct sock *); 600 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 601 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 602 enum tcp_queue { 603 TCP_FRAG_IN_WRITE_QUEUE, 604 TCP_FRAG_IN_RTX_QUEUE, 605 }; 606 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 607 struct sk_buff *skb, u32 len, 608 unsigned int mss_now, gfp_t gfp); 609 610 void tcp_send_probe0(struct sock *); 611 void tcp_send_partial(struct sock *); 612 int tcp_write_wakeup(struct sock *, int mib); 613 void tcp_send_fin(struct sock *sk); 614 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 615 int tcp_send_synack(struct sock *); 616 void tcp_push_one(struct sock *, unsigned int mss_now); 617 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 618 void tcp_send_ack(struct sock *sk); 619 void tcp_send_delayed_ack(struct sock *sk); 620 void tcp_send_loss_probe(struct sock *sk); 621 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 622 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 623 const struct sk_buff *next_skb); 624 625 /* tcp_input.c */ 626 void tcp_rearm_rto(struct sock *sk); 627 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 628 void tcp_reset(struct sock *sk, struct sk_buff *skb); 629 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 630 void tcp_fin(struct sock *sk); 631 void tcp_check_space(struct sock *sk); 632 633 /* tcp_timer.c */ 634 void tcp_init_xmit_timers(struct sock *); 635 static inline void tcp_clear_xmit_timers(struct sock *sk) 636 { 637 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 638 __sock_put(sk); 639 640 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 641 __sock_put(sk); 642 643 inet_csk_clear_xmit_timers(sk); 644 } 645 646 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 647 unsigned int tcp_current_mss(struct sock *sk); 648 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 649 650 /* Bound MSS / TSO packet size with the half of the window */ 651 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 652 { 653 int cutoff; 654 655 /* When peer uses tiny windows, there is no use in packetizing 656 * to sub-MSS pieces for the sake of SWS or making sure there 657 * are enough packets in the pipe for fast recovery. 658 * 659 * On the other hand, for extremely large MSS devices, handling 660 * smaller than MSS windows in this way does make sense. 661 */ 662 if (tp->max_window > TCP_MSS_DEFAULT) 663 cutoff = (tp->max_window >> 1); 664 else 665 cutoff = tp->max_window; 666 667 if (cutoff && pktsize > cutoff) 668 return max_t(int, cutoff, 68U - tp->tcp_header_len); 669 else 670 return pktsize; 671 } 672 673 /* tcp.c */ 674 void tcp_get_info(struct sock *, struct tcp_info *); 675 676 /* Read 'sendfile()'-style from a TCP socket */ 677 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 678 sk_read_actor_t recv_actor); 679 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 680 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 681 void tcp_read_done(struct sock *sk, size_t len); 682 683 void tcp_initialize_rcv_mss(struct sock *sk); 684 685 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 686 int tcp_mss_to_mtu(struct sock *sk, int mss); 687 void tcp_mtup_init(struct sock *sk); 688 689 static inline void tcp_bound_rto(const struct sock *sk) 690 { 691 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 692 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 693 } 694 695 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 696 { 697 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 698 } 699 700 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 701 { 702 /* mptcp hooks are only on the slow path */ 703 if (sk_is_mptcp((struct sock *)tp)) 704 return; 705 706 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 707 ntohl(TCP_FLAG_ACK) | 708 snd_wnd); 709 } 710 711 static inline void tcp_fast_path_on(struct tcp_sock *tp) 712 { 713 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 714 } 715 716 static inline void tcp_fast_path_check(struct sock *sk) 717 { 718 struct tcp_sock *tp = tcp_sk(sk); 719 720 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 721 tp->rcv_wnd && 722 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 723 !tp->urg_data) 724 tcp_fast_path_on(tp); 725 } 726 727 /* Compute the actual rto_min value */ 728 static inline u32 tcp_rto_min(struct sock *sk) 729 { 730 const struct dst_entry *dst = __sk_dst_get(sk); 731 u32 rto_min = inet_csk(sk)->icsk_rto_min; 732 733 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 734 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 735 return rto_min; 736 } 737 738 static inline u32 tcp_rto_min_us(struct sock *sk) 739 { 740 return jiffies_to_usecs(tcp_rto_min(sk)); 741 } 742 743 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 744 { 745 return dst_metric_locked(dst, RTAX_CC_ALGO); 746 } 747 748 /* Minimum RTT in usec. ~0 means not available. */ 749 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 750 { 751 return minmax_get(&tp->rtt_min); 752 } 753 754 /* Compute the actual receive window we are currently advertising. 755 * Rcv_nxt can be after the window if our peer push more data 756 * than the offered window. 757 */ 758 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 759 { 760 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 761 762 if (win < 0) 763 win = 0; 764 return (u32) win; 765 } 766 767 /* Choose a new window, without checks for shrinking, and without 768 * scaling applied to the result. The caller does these things 769 * if necessary. This is a "raw" window selection. 770 */ 771 u32 __tcp_select_window(struct sock *sk); 772 773 void tcp_send_window_probe(struct sock *sk); 774 775 /* TCP uses 32bit jiffies to save some space. 776 * Note that this is different from tcp_time_stamp, which 777 * historically has been the same until linux-4.13. 778 */ 779 #define tcp_jiffies32 ((u32)jiffies) 780 781 /* 782 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 783 * It is no longer tied to jiffies, but to 1 ms clock. 784 * Note: double check if you want to use tcp_jiffies32 instead of this. 785 */ 786 #define TCP_TS_HZ 1000 787 788 static inline u64 tcp_clock_ns(void) 789 { 790 return ktime_get_ns(); 791 } 792 793 static inline u64 tcp_clock_us(void) 794 { 795 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 796 } 797 798 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 799 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 800 { 801 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 802 } 803 804 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 805 static inline u32 tcp_ns_to_ts(u64 ns) 806 { 807 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 808 } 809 810 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 811 static inline u32 tcp_time_stamp_raw(void) 812 { 813 return tcp_ns_to_ts(tcp_clock_ns()); 814 } 815 816 void tcp_mstamp_refresh(struct tcp_sock *tp); 817 818 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 819 { 820 return max_t(s64, t1 - t0, 0); 821 } 822 823 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 824 { 825 return tcp_ns_to_ts(skb->skb_mstamp_ns); 826 } 827 828 /* provide the departure time in us unit */ 829 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 830 { 831 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 832 } 833 834 835 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 836 837 #define TCPHDR_FIN 0x01 838 #define TCPHDR_SYN 0x02 839 #define TCPHDR_RST 0x04 840 #define TCPHDR_PSH 0x08 841 #define TCPHDR_ACK 0x10 842 #define TCPHDR_URG 0x20 843 #define TCPHDR_ECE 0x40 844 #define TCPHDR_CWR 0x80 845 846 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 847 848 /* This is what the send packet queuing engine uses to pass 849 * TCP per-packet control information to the transmission code. 850 * We also store the host-order sequence numbers in here too. 851 * This is 44 bytes if IPV6 is enabled. 852 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 853 */ 854 struct tcp_skb_cb { 855 __u32 seq; /* Starting sequence number */ 856 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 857 union { 858 /* Note : tcp_tw_isn is used in input path only 859 * (isn chosen by tcp_timewait_state_process()) 860 * 861 * tcp_gso_segs/size are used in write queue only, 862 * cf tcp_skb_pcount()/tcp_skb_mss() 863 */ 864 __u32 tcp_tw_isn; 865 struct { 866 u16 tcp_gso_segs; 867 u16 tcp_gso_size; 868 }; 869 }; 870 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 871 872 __u8 sacked; /* State flags for SACK. */ 873 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 874 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 875 #define TCPCB_LOST 0x04 /* SKB is lost */ 876 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 877 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 878 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 879 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 880 TCPCB_REPAIRED) 881 882 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 883 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 884 eor:1, /* Is skb MSG_EOR marked? */ 885 has_rxtstamp:1, /* SKB has a RX timestamp */ 886 unused:5; 887 __u32 ack_seq; /* Sequence number ACK'd */ 888 union { 889 struct { 890 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 891 /* There is space for up to 24 bytes */ 892 __u32 is_app_limited:1, /* cwnd not fully used? */ 893 delivered_ce:20, 894 unused:11; 895 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 896 __u32 delivered; 897 /* start of send pipeline phase */ 898 u64 first_tx_mstamp; 899 /* when we reached the "delivered" count */ 900 u64 delivered_mstamp; 901 } tx; /* only used for outgoing skbs */ 902 union { 903 struct inet_skb_parm h4; 904 #if IS_ENABLED(CONFIG_IPV6) 905 struct inet6_skb_parm h6; 906 #endif 907 } header; /* For incoming skbs */ 908 }; 909 }; 910 911 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 912 913 extern const struct inet_connection_sock_af_ops ipv4_specific; 914 915 #if IS_ENABLED(CONFIG_IPV6) 916 /* This is the variant of inet6_iif() that must be used by TCP, 917 * as TCP moves IP6CB into a different location in skb->cb[] 918 */ 919 static inline int tcp_v6_iif(const struct sk_buff *skb) 920 { 921 return TCP_SKB_CB(skb)->header.h6.iif; 922 } 923 924 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 925 { 926 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 927 928 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 929 } 930 931 /* TCP_SKB_CB reference means this can not be used from early demux */ 932 static inline int tcp_v6_sdif(const struct sk_buff *skb) 933 { 934 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 935 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 936 return TCP_SKB_CB(skb)->header.h6.iif; 937 #endif 938 return 0; 939 } 940 941 extern const struct inet_connection_sock_af_ops ipv6_specific; 942 943 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 944 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 945 void tcp_v6_early_demux(struct sk_buff *skb); 946 947 #endif 948 949 /* TCP_SKB_CB reference means this can not be used from early demux */ 950 static inline int tcp_v4_sdif(struct sk_buff *skb) 951 { 952 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 953 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 954 return TCP_SKB_CB(skb)->header.h4.iif; 955 #endif 956 return 0; 957 } 958 959 /* Due to TSO, an SKB can be composed of multiple actual 960 * packets. To keep these tracked properly, we use this. 961 */ 962 static inline int tcp_skb_pcount(const struct sk_buff *skb) 963 { 964 return TCP_SKB_CB(skb)->tcp_gso_segs; 965 } 966 967 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 968 { 969 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 970 } 971 972 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 973 { 974 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 975 } 976 977 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 978 static inline int tcp_skb_mss(const struct sk_buff *skb) 979 { 980 return TCP_SKB_CB(skb)->tcp_gso_size; 981 } 982 983 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 984 { 985 return likely(!TCP_SKB_CB(skb)->eor); 986 } 987 988 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 989 const struct sk_buff *from) 990 { 991 return likely(tcp_skb_can_collapse_to(to) && 992 mptcp_skb_can_collapse(to, from) && 993 skb_pure_zcopy_same(to, from)); 994 } 995 996 /* Events passed to congestion control interface */ 997 enum tcp_ca_event { 998 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 999 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1000 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1001 CA_EVENT_LOSS, /* loss timeout */ 1002 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1003 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1004 }; 1005 1006 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1007 enum tcp_ca_ack_event_flags { 1008 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1009 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1010 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1011 }; 1012 1013 /* 1014 * Interface for adding new TCP congestion control handlers 1015 */ 1016 #define TCP_CA_NAME_MAX 16 1017 #define TCP_CA_MAX 128 1018 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1019 1020 #define TCP_CA_UNSPEC 0 1021 1022 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1023 #define TCP_CONG_NON_RESTRICTED 0x1 1024 /* Requires ECN/ECT set on all packets */ 1025 #define TCP_CONG_NEEDS_ECN 0x2 1026 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1027 1028 union tcp_cc_info; 1029 1030 struct ack_sample { 1031 u32 pkts_acked; 1032 s32 rtt_us; 1033 u32 in_flight; 1034 }; 1035 1036 /* A rate sample measures the number of (original/retransmitted) data 1037 * packets delivered "delivered" over an interval of time "interval_us". 1038 * The tcp_rate.c code fills in the rate sample, and congestion 1039 * control modules that define a cong_control function to run at the end 1040 * of ACK processing can optionally chose to consult this sample when 1041 * setting cwnd and pacing rate. 1042 * A sample is invalid if "delivered" or "interval_us" is negative. 1043 */ 1044 struct rate_sample { 1045 u64 prior_mstamp; /* starting timestamp for interval */ 1046 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1047 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1048 s32 delivered; /* number of packets delivered over interval */ 1049 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1050 long interval_us; /* time for tp->delivered to incr "delivered" */ 1051 u32 snd_interval_us; /* snd interval for delivered packets */ 1052 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1053 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1054 int losses; /* number of packets marked lost upon ACK */ 1055 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1056 u32 prior_in_flight; /* in flight before this ACK */ 1057 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1058 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1059 bool is_retrans; /* is sample from retransmission? */ 1060 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1061 }; 1062 1063 struct tcp_congestion_ops { 1064 /* fast path fields are put first to fill one cache line */ 1065 1066 /* return slow start threshold (required) */ 1067 u32 (*ssthresh)(struct sock *sk); 1068 1069 /* do new cwnd calculation (required) */ 1070 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1071 1072 /* call before changing ca_state (optional) */ 1073 void (*set_state)(struct sock *sk, u8 new_state); 1074 1075 /* call when cwnd event occurs (optional) */ 1076 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1077 1078 /* call when ack arrives (optional) */ 1079 void (*in_ack_event)(struct sock *sk, u32 flags); 1080 1081 /* hook for packet ack accounting (optional) */ 1082 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1083 1084 /* override sysctl_tcp_min_tso_segs */ 1085 u32 (*min_tso_segs)(struct sock *sk); 1086 1087 /* call when packets are delivered to update cwnd and pacing rate, 1088 * after all the ca_state processing. (optional) 1089 */ 1090 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1091 1092 1093 /* new value of cwnd after loss (required) */ 1094 u32 (*undo_cwnd)(struct sock *sk); 1095 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1096 u32 (*sndbuf_expand)(struct sock *sk); 1097 1098 /* control/slow paths put last */ 1099 /* get info for inet_diag (optional) */ 1100 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1101 union tcp_cc_info *info); 1102 1103 char name[TCP_CA_NAME_MAX]; 1104 struct module *owner; 1105 struct list_head list; 1106 u32 key; 1107 u32 flags; 1108 1109 /* initialize private data (optional) */ 1110 void (*init)(struct sock *sk); 1111 /* cleanup private data (optional) */ 1112 void (*release)(struct sock *sk); 1113 } ____cacheline_aligned_in_smp; 1114 1115 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1116 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1117 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 in slow start, ensure cwnd grows to twice what was ACKed. */ 1303 if (tcp_in_slow_start(tp)) 1304 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1305 1306 return tp->is_cwnd_limited; 1307 } 1308 1309 /* BBR congestion control needs pacing. 1310 * Same remark for SO_MAX_PACING_RATE. 1311 * sch_fq packet scheduler is efficiently handling pacing, 1312 * but is not always installed/used. 1313 * Return true if TCP stack should pace packets itself. 1314 */ 1315 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1316 { 1317 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1318 } 1319 1320 /* Estimates in how many jiffies next packet for this flow can be sent. 1321 * Scheduling a retransmit timer too early would be silly. 1322 */ 1323 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1324 { 1325 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1326 1327 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1328 } 1329 1330 static inline void tcp_reset_xmit_timer(struct sock *sk, 1331 const int what, 1332 unsigned long when, 1333 const unsigned long max_when) 1334 { 1335 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1336 max_when); 1337 } 1338 1339 /* Something is really bad, we could not queue an additional packet, 1340 * because qdisc is full or receiver sent a 0 window, or we are paced. 1341 * We do not want to add fuel to the fire, or abort too early, 1342 * so make sure the timer we arm now is at least 200ms in the future, 1343 * regardless of current icsk_rto value (as it could be ~2ms) 1344 */ 1345 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1346 { 1347 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1348 } 1349 1350 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1351 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1352 unsigned long max_when) 1353 { 1354 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1355 inet_csk(sk)->icsk_backoff); 1356 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1357 1358 return (unsigned long)min_t(u64, when, max_when); 1359 } 1360 1361 static inline void tcp_check_probe_timer(struct sock *sk) 1362 { 1363 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1364 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1365 tcp_probe0_base(sk), TCP_RTO_MAX); 1366 } 1367 1368 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1369 { 1370 tp->snd_wl1 = seq; 1371 } 1372 1373 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1374 { 1375 tp->snd_wl1 = seq; 1376 } 1377 1378 /* 1379 * Calculate(/check) TCP checksum 1380 */ 1381 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1382 __be32 daddr, __wsum base) 1383 { 1384 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1385 } 1386 1387 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1388 { 1389 return !skb_csum_unnecessary(skb) && 1390 __skb_checksum_complete(skb); 1391 } 1392 1393 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1394 enum skb_drop_reason *reason); 1395 1396 1397 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1398 void tcp_set_state(struct sock *sk, int state); 1399 void tcp_done(struct sock *sk); 1400 int tcp_abort(struct sock *sk, int err); 1401 1402 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1403 { 1404 rx_opt->dsack = 0; 1405 rx_opt->num_sacks = 0; 1406 } 1407 1408 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1409 1410 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1411 { 1412 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1413 struct tcp_sock *tp = tcp_sk(sk); 1414 s32 delta; 1415 1416 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1417 tp->packets_out || ca_ops->cong_control) 1418 return; 1419 delta = tcp_jiffies32 - tp->lsndtime; 1420 if (delta > inet_csk(sk)->icsk_rto) 1421 tcp_cwnd_restart(sk, delta); 1422 } 1423 1424 /* Determine a window scaling and initial window to offer. */ 1425 void tcp_select_initial_window(const struct sock *sk, int __space, 1426 __u32 mss, __u32 *rcv_wnd, 1427 __u32 *window_clamp, int wscale_ok, 1428 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1429 1430 static inline int tcp_win_from_space(const struct sock *sk, int space) 1431 { 1432 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale); 1433 1434 return tcp_adv_win_scale <= 0 ? 1435 (space>>(-tcp_adv_win_scale)) : 1436 space - (space>>tcp_adv_win_scale); 1437 } 1438 1439 /* Note: caller must be prepared to deal with negative returns */ 1440 static inline int tcp_space(const struct sock *sk) 1441 { 1442 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1443 READ_ONCE(sk->sk_backlog.len) - 1444 atomic_read(&sk->sk_rmem_alloc)); 1445 } 1446 1447 static inline int tcp_full_space(const struct sock *sk) 1448 { 1449 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1450 } 1451 1452 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1453 { 1454 int unused_mem = sk_unused_reserved_mem(sk); 1455 struct tcp_sock *tp = tcp_sk(sk); 1456 1457 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1458 if (unused_mem) 1459 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1460 tcp_win_from_space(sk, unused_mem)); 1461 } 1462 1463 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1464 1465 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1466 * If 87.5 % (7/8) of the space has been consumed, we want to override 1467 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1468 * len/truesize ratio. 1469 */ 1470 static inline bool tcp_rmem_pressure(const struct sock *sk) 1471 { 1472 int rcvbuf, threshold; 1473 1474 if (tcp_under_memory_pressure(sk)) 1475 return true; 1476 1477 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1478 threshold = rcvbuf - (rcvbuf >> 3); 1479 1480 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1481 } 1482 1483 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1484 { 1485 const struct tcp_sock *tp = tcp_sk(sk); 1486 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1487 1488 if (avail <= 0) 1489 return false; 1490 1491 return (avail >= target) || tcp_rmem_pressure(sk) || 1492 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1493 } 1494 1495 extern void tcp_openreq_init_rwin(struct request_sock *req, 1496 const struct sock *sk_listener, 1497 const struct dst_entry *dst); 1498 1499 void tcp_enter_memory_pressure(struct sock *sk); 1500 void tcp_leave_memory_pressure(struct sock *sk); 1501 1502 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1503 { 1504 struct net *net = sock_net((struct sock *)tp); 1505 1506 return tp->keepalive_intvl ? : 1507 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1508 } 1509 1510 static inline int keepalive_time_when(const struct tcp_sock *tp) 1511 { 1512 struct net *net = sock_net((struct sock *)tp); 1513 1514 return tp->keepalive_time ? : 1515 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1516 } 1517 1518 static inline int keepalive_probes(const struct tcp_sock *tp) 1519 { 1520 struct net *net = sock_net((struct sock *)tp); 1521 1522 return tp->keepalive_probes ? : 1523 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1524 } 1525 1526 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1527 { 1528 const struct inet_connection_sock *icsk = &tp->inet_conn; 1529 1530 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1531 tcp_jiffies32 - tp->rcv_tstamp); 1532 } 1533 1534 static inline int tcp_fin_time(const struct sock *sk) 1535 { 1536 int fin_timeout = tcp_sk(sk)->linger2 ? : 1537 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1538 const int rto = inet_csk(sk)->icsk_rto; 1539 1540 if (fin_timeout < (rto << 2) - (rto >> 1)) 1541 fin_timeout = (rto << 2) - (rto >> 1); 1542 1543 return fin_timeout; 1544 } 1545 1546 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1547 int paws_win) 1548 { 1549 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1550 return true; 1551 if (unlikely(!time_before32(ktime_get_seconds(), 1552 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1553 return true; 1554 /* 1555 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1556 * then following tcp messages have valid values. Ignore 0 value, 1557 * or else 'negative' tsval might forbid us to accept their packets. 1558 */ 1559 if (!rx_opt->ts_recent) 1560 return true; 1561 return false; 1562 } 1563 1564 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1565 int rst) 1566 { 1567 if (tcp_paws_check(rx_opt, 0)) 1568 return false; 1569 1570 /* RST segments are not recommended to carry timestamp, 1571 and, if they do, it is recommended to ignore PAWS because 1572 "their cleanup function should take precedence over timestamps." 1573 Certainly, it is mistake. It is necessary to understand the reasons 1574 of this constraint to relax it: if peer reboots, clock may go 1575 out-of-sync and half-open connections will not be reset. 1576 Actually, the problem would be not existing if all 1577 the implementations followed draft about maintaining clock 1578 via reboots. Linux-2.2 DOES NOT! 1579 1580 However, we can relax time bounds for RST segments to MSL. 1581 */ 1582 if (rst && !time_before32(ktime_get_seconds(), 1583 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1584 return false; 1585 return true; 1586 } 1587 1588 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1589 int mib_idx, u32 *last_oow_ack_time); 1590 1591 static inline void tcp_mib_init(struct net *net) 1592 { 1593 /* See RFC 2012 */ 1594 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1595 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1596 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1597 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1598 } 1599 1600 /* from STCP */ 1601 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1602 { 1603 tp->lost_skb_hint = NULL; 1604 } 1605 1606 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1607 { 1608 tcp_clear_retrans_hints_partial(tp); 1609 tp->retransmit_skb_hint = NULL; 1610 } 1611 1612 union tcp_md5_addr { 1613 struct in_addr a4; 1614 #if IS_ENABLED(CONFIG_IPV6) 1615 struct in6_addr a6; 1616 #endif 1617 }; 1618 1619 /* - key database */ 1620 struct tcp_md5sig_key { 1621 struct hlist_node node; 1622 u8 keylen; 1623 u8 family; /* AF_INET or AF_INET6 */ 1624 u8 prefixlen; 1625 u8 flags; 1626 union tcp_md5_addr addr; 1627 int l3index; /* set if key added with L3 scope */ 1628 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1629 struct rcu_head rcu; 1630 }; 1631 1632 /* - sock block */ 1633 struct tcp_md5sig_info { 1634 struct hlist_head head; 1635 struct rcu_head rcu; 1636 }; 1637 1638 /* - pseudo header */ 1639 struct tcp4_pseudohdr { 1640 __be32 saddr; 1641 __be32 daddr; 1642 __u8 pad; 1643 __u8 protocol; 1644 __be16 len; 1645 }; 1646 1647 struct tcp6_pseudohdr { 1648 struct in6_addr saddr; 1649 struct in6_addr daddr; 1650 __be32 len; 1651 __be32 protocol; /* including padding */ 1652 }; 1653 1654 union tcp_md5sum_block { 1655 struct tcp4_pseudohdr ip4; 1656 #if IS_ENABLED(CONFIG_IPV6) 1657 struct tcp6_pseudohdr ip6; 1658 #endif 1659 }; 1660 1661 /* - pool: digest algorithm, hash description and scratch buffer */ 1662 struct tcp_md5sig_pool { 1663 struct ahash_request *md5_req; 1664 void *scratch; 1665 }; 1666 1667 /* - functions */ 1668 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1669 const struct sock *sk, const struct sk_buff *skb); 1670 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1671 int family, u8 prefixlen, int l3index, u8 flags, 1672 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1673 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1674 int family, u8 prefixlen, int l3index, u8 flags); 1675 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1676 const struct sock *addr_sk); 1677 1678 #ifdef CONFIG_TCP_MD5SIG 1679 #include <linux/jump_label.h> 1680 extern struct static_key_false tcp_md5_needed; 1681 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1682 const union tcp_md5_addr *addr, 1683 int family); 1684 static inline struct tcp_md5sig_key * 1685 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1686 const union tcp_md5_addr *addr, int family) 1687 { 1688 if (!static_branch_unlikely(&tcp_md5_needed)) 1689 return NULL; 1690 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1691 } 1692 1693 enum skb_drop_reason 1694 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1695 const void *saddr, const void *daddr, 1696 int family, int dif, int sdif); 1697 1698 1699 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1700 #else 1701 static inline struct tcp_md5sig_key * 1702 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1703 const union tcp_md5_addr *addr, int family) 1704 { 1705 return NULL; 1706 } 1707 1708 static inline enum skb_drop_reason 1709 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1710 const void *saddr, const void *daddr, 1711 int family, int dif, int sdif) 1712 { 1713 return SKB_NOT_DROPPED_YET; 1714 } 1715 #define tcp_twsk_md5_key(twsk) NULL 1716 #endif 1717 1718 bool tcp_alloc_md5sig_pool(void); 1719 1720 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1721 static inline void tcp_put_md5sig_pool(void) 1722 { 1723 local_bh_enable(); 1724 } 1725 1726 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1727 unsigned int header_len); 1728 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1729 const struct tcp_md5sig_key *key); 1730 1731 /* From tcp_fastopen.c */ 1732 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1733 struct tcp_fastopen_cookie *cookie); 1734 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1735 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1736 u16 try_exp); 1737 struct tcp_fastopen_request { 1738 /* Fast Open cookie. Size 0 means a cookie request */ 1739 struct tcp_fastopen_cookie cookie; 1740 struct msghdr *data; /* data in MSG_FASTOPEN */ 1741 size_t size; 1742 int copied; /* queued in tcp_connect() */ 1743 struct ubuf_info *uarg; 1744 }; 1745 void tcp_free_fastopen_req(struct tcp_sock *tp); 1746 void tcp_fastopen_destroy_cipher(struct sock *sk); 1747 void tcp_fastopen_ctx_destroy(struct net *net); 1748 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1749 void *primary_key, void *backup_key); 1750 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1751 u64 *key); 1752 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1753 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1754 struct request_sock *req, 1755 struct tcp_fastopen_cookie *foc, 1756 const struct dst_entry *dst); 1757 void tcp_fastopen_init_key_once(struct net *net); 1758 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1759 struct tcp_fastopen_cookie *cookie); 1760 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1761 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1762 #define TCP_FASTOPEN_KEY_MAX 2 1763 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1764 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1765 1766 /* Fastopen key context */ 1767 struct tcp_fastopen_context { 1768 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1769 int num; 1770 struct rcu_head rcu; 1771 }; 1772 1773 void tcp_fastopen_active_disable(struct sock *sk); 1774 bool tcp_fastopen_active_should_disable(struct sock *sk); 1775 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1776 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1777 1778 /* Caller needs to wrap with rcu_read_(un)lock() */ 1779 static inline 1780 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1781 { 1782 struct tcp_fastopen_context *ctx; 1783 1784 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1785 if (!ctx) 1786 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1787 return ctx; 1788 } 1789 1790 static inline 1791 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1792 const struct tcp_fastopen_cookie *orig) 1793 { 1794 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1795 orig->len == foc->len && 1796 !memcmp(orig->val, foc->val, foc->len)) 1797 return true; 1798 return false; 1799 } 1800 1801 static inline 1802 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1803 { 1804 return ctx->num; 1805 } 1806 1807 /* Latencies incurred by various limits for a sender. They are 1808 * chronograph-like stats that are mutually exclusive. 1809 */ 1810 enum tcp_chrono { 1811 TCP_CHRONO_UNSPEC, 1812 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1813 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1814 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1815 __TCP_CHRONO_MAX, 1816 }; 1817 1818 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1819 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1820 1821 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1822 * the same memory storage than skb->destructor/_skb_refdst 1823 */ 1824 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1825 { 1826 skb->destructor = NULL; 1827 skb->_skb_refdst = 0UL; 1828 } 1829 1830 #define tcp_skb_tsorted_save(skb) { \ 1831 unsigned long _save = skb->_skb_refdst; \ 1832 skb->_skb_refdst = 0UL; 1833 1834 #define tcp_skb_tsorted_restore(skb) \ 1835 skb->_skb_refdst = _save; \ 1836 } 1837 1838 void tcp_write_queue_purge(struct sock *sk); 1839 1840 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1841 { 1842 return skb_rb_first(&sk->tcp_rtx_queue); 1843 } 1844 1845 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1846 { 1847 return skb_rb_last(&sk->tcp_rtx_queue); 1848 } 1849 1850 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1851 { 1852 return skb_peek_tail(&sk->sk_write_queue); 1853 } 1854 1855 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1856 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1857 1858 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1859 { 1860 return skb_peek(&sk->sk_write_queue); 1861 } 1862 1863 static inline bool tcp_skb_is_last(const struct sock *sk, 1864 const struct sk_buff *skb) 1865 { 1866 return skb_queue_is_last(&sk->sk_write_queue, skb); 1867 } 1868 1869 /** 1870 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1871 * @sk: socket 1872 * 1873 * Since the write queue can have a temporary empty skb in it, 1874 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1875 */ 1876 static inline bool tcp_write_queue_empty(const struct sock *sk) 1877 { 1878 const struct tcp_sock *tp = tcp_sk(sk); 1879 1880 return tp->write_seq == tp->snd_nxt; 1881 } 1882 1883 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1884 { 1885 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1886 } 1887 1888 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1889 { 1890 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1891 } 1892 1893 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1894 { 1895 __skb_queue_tail(&sk->sk_write_queue, skb); 1896 1897 /* Queue it, remembering where we must start sending. */ 1898 if (sk->sk_write_queue.next == skb) 1899 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1900 } 1901 1902 /* Insert new before skb on the write queue of sk. */ 1903 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1904 struct sk_buff *skb, 1905 struct sock *sk) 1906 { 1907 __skb_queue_before(&sk->sk_write_queue, skb, new); 1908 } 1909 1910 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1911 { 1912 tcp_skb_tsorted_anchor_cleanup(skb); 1913 __skb_unlink(skb, &sk->sk_write_queue); 1914 } 1915 1916 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1917 1918 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1919 { 1920 tcp_skb_tsorted_anchor_cleanup(skb); 1921 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1922 } 1923 1924 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1925 { 1926 list_del(&skb->tcp_tsorted_anchor); 1927 tcp_rtx_queue_unlink(skb, sk); 1928 tcp_wmem_free_skb(sk, skb); 1929 } 1930 1931 static inline void tcp_push_pending_frames(struct sock *sk) 1932 { 1933 if (tcp_send_head(sk)) { 1934 struct tcp_sock *tp = tcp_sk(sk); 1935 1936 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1937 } 1938 } 1939 1940 /* Start sequence of the skb just after the highest skb with SACKed 1941 * bit, valid only if sacked_out > 0 or when the caller has ensured 1942 * validity by itself. 1943 */ 1944 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1945 { 1946 if (!tp->sacked_out) 1947 return tp->snd_una; 1948 1949 if (tp->highest_sack == NULL) 1950 return tp->snd_nxt; 1951 1952 return TCP_SKB_CB(tp->highest_sack)->seq; 1953 } 1954 1955 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1956 { 1957 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1958 } 1959 1960 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1961 { 1962 return tcp_sk(sk)->highest_sack; 1963 } 1964 1965 static inline void tcp_highest_sack_reset(struct sock *sk) 1966 { 1967 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1968 } 1969 1970 /* Called when old skb is about to be deleted and replaced by new skb */ 1971 static inline void tcp_highest_sack_replace(struct sock *sk, 1972 struct sk_buff *old, 1973 struct sk_buff *new) 1974 { 1975 if (old == tcp_highest_sack(sk)) 1976 tcp_sk(sk)->highest_sack = new; 1977 } 1978 1979 /* This helper checks if socket has IP_TRANSPARENT set */ 1980 static inline bool inet_sk_transparent(const struct sock *sk) 1981 { 1982 switch (sk->sk_state) { 1983 case TCP_TIME_WAIT: 1984 return inet_twsk(sk)->tw_transparent; 1985 case TCP_NEW_SYN_RECV: 1986 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1987 } 1988 return inet_sk(sk)->transparent; 1989 } 1990 1991 /* Determines whether this is a thin stream (which may suffer from 1992 * increased latency). Used to trigger latency-reducing mechanisms. 1993 */ 1994 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1995 { 1996 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1997 } 1998 1999 /* /proc */ 2000 enum tcp_seq_states { 2001 TCP_SEQ_STATE_LISTENING, 2002 TCP_SEQ_STATE_ESTABLISHED, 2003 }; 2004 2005 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2006 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2007 void tcp_seq_stop(struct seq_file *seq, void *v); 2008 2009 struct tcp_seq_afinfo { 2010 sa_family_t family; 2011 }; 2012 2013 struct tcp_iter_state { 2014 struct seq_net_private p; 2015 enum tcp_seq_states state; 2016 struct sock *syn_wait_sk; 2017 int bucket, offset, sbucket, num; 2018 loff_t last_pos; 2019 }; 2020 2021 extern struct request_sock_ops tcp_request_sock_ops; 2022 extern struct request_sock_ops tcp6_request_sock_ops; 2023 2024 void tcp_v4_destroy_sock(struct sock *sk); 2025 2026 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2027 netdev_features_t features); 2028 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2029 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2030 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2031 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2032 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2033 int tcp_gro_complete(struct sk_buff *skb); 2034 2035 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2036 2037 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2038 { 2039 struct net *net = sock_net((struct sock *)tp); 2040 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2041 } 2042 2043 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2044 2045 #ifdef CONFIG_PROC_FS 2046 int tcp4_proc_init(void); 2047 void tcp4_proc_exit(void); 2048 #endif 2049 2050 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2051 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2052 const struct tcp_request_sock_ops *af_ops, 2053 struct sock *sk, struct sk_buff *skb); 2054 2055 /* TCP af-specific functions */ 2056 struct tcp_sock_af_ops { 2057 #ifdef CONFIG_TCP_MD5SIG 2058 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2059 const struct sock *addr_sk); 2060 int (*calc_md5_hash)(char *location, 2061 const struct tcp_md5sig_key *md5, 2062 const struct sock *sk, 2063 const struct sk_buff *skb); 2064 int (*md5_parse)(struct sock *sk, 2065 int optname, 2066 sockptr_t optval, 2067 int optlen); 2068 #endif 2069 }; 2070 2071 struct tcp_request_sock_ops { 2072 u16 mss_clamp; 2073 #ifdef CONFIG_TCP_MD5SIG 2074 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2075 const struct sock *addr_sk); 2076 int (*calc_md5_hash) (char *location, 2077 const struct tcp_md5sig_key *md5, 2078 const struct sock *sk, 2079 const struct sk_buff *skb); 2080 #endif 2081 #ifdef CONFIG_SYN_COOKIES 2082 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2083 __u16 *mss); 2084 #endif 2085 struct dst_entry *(*route_req)(const struct sock *sk, 2086 struct sk_buff *skb, 2087 struct flowi *fl, 2088 struct request_sock *req); 2089 u32 (*init_seq)(const struct sk_buff *skb); 2090 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2091 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2092 struct flowi *fl, struct request_sock *req, 2093 struct tcp_fastopen_cookie *foc, 2094 enum tcp_synack_type synack_type, 2095 struct sk_buff *syn_skb); 2096 }; 2097 2098 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2099 #if IS_ENABLED(CONFIG_IPV6) 2100 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2101 #endif 2102 2103 #ifdef CONFIG_SYN_COOKIES 2104 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2105 const struct sock *sk, struct sk_buff *skb, 2106 __u16 *mss) 2107 { 2108 tcp_synq_overflow(sk); 2109 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2110 return ops->cookie_init_seq(skb, mss); 2111 } 2112 #else 2113 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2114 const struct sock *sk, struct sk_buff *skb, 2115 __u16 *mss) 2116 { 2117 return 0; 2118 } 2119 #endif 2120 2121 int tcpv4_offload_init(void); 2122 2123 void tcp_v4_init(void); 2124 void tcp_init(void); 2125 2126 /* tcp_recovery.c */ 2127 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2128 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2129 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2130 u32 reo_wnd); 2131 extern bool tcp_rack_mark_lost(struct sock *sk); 2132 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2133 u64 xmit_time); 2134 extern void tcp_rack_reo_timeout(struct sock *sk); 2135 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2136 2137 /* At how many usecs into the future should the RTO fire? */ 2138 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2139 { 2140 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2141 u32 rto = inet_csk(sk)->icsk_rto; 2142 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2143 2144 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2145 } 2146 2147 /* 2148 * Save and compile IPv4 options, return a pointer to it 2149 */ 2150 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2151 struct sk_buff *skb) 2152 { 2153 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2154 struct ip_options_rcu *dopt = NULL; 2155 2156 if (opt->optlen) { 2157 int opt_size = sizeof(*dopt) + opt->optlen; 2158 2159 dopt = kmalloc(opt_size, GFP_ATOMIC); 2160 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2161 kfree(dopt); 2162 dopt = NULL; 2163 } 2164 } 2165 return dopt; 2166 } 2167 2168 /* locally generated TCP pure ACKs have skb->truesize == 2 2169 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2170 * This is much faster than dissecting the packet to find out. 2171 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2172 */ 2173 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2174 { 2175 return skb->truesize == 2; 2176 } 2177 2178 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2179 { 2180 skb->truesize = 2; 2181 } 2182 2183 static inline int tcp_inq(struct sock *sk) 2184 { 2185 struct tcp_sock *tp = tcp_sk(sk); 2186 int answ; 2187 2188 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2189 answ = 0; 2190 } else if (sock_flag(sk, SOCK_URGINLINE) || 2191 !tp->urg_data || 2192 before(tp->urg_seq, tp->copied_seq) || 2193 !before(tp->urg_seq, tp->rcv_nxt)) { 2194 2195 answ = tp->rcv_nxt - tp->copied_seq; 2196 2197 /* Subtract 1, if FIN was received */ 2198 if (answ && sock_flag(sk, SOCK_DONE)) 2199 answ--; 2200 } else { 2201 answ = tp->urg_seq - tp->copied_seq; 2202 } 2203 2204 return answ; 2205 } 2206 2207 int tcp_peek_len(struct socket *sock); 2208 2209 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2210 { 2211 u16 segs_in; 2212 2213 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2214 2215 /* We update these fields while other threads might 2216 * read them from tcp_get_info() 2217 */ 2218 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2219 if (skb->len > tcp_hdrlen(skb)) 2220 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2221 } 2222 2223 /* 2224 * TCP listen path runs lockless. 2225 * We forced "struct sock" to be const qualified to make sure 2226 * we don't modify one of its field by mistake. 2227 * Here, we increment sk_drops which is an atomic_t, so we can safely 2228 * make sock writable again. 2229 */ 2230 static inline void tcp_listendrop(const struct sock *sk) 2231 { 2232 atomic_inc(&((struct sock *)sk)->sk_drops); 2233 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2234 } 2235 2236 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2237 2238 /* 2239 * Interface for adding Upper Level Protocols over TCP 2240 */ 2241 2242 #define TCP_ULP_NAME_MAX 16 2243 #define TCP_ULP_MAX 128 2244 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2245 2246 struct tcp_ulp_ops { 2247 struct list_head list; 2248 2249 /* initialize ulp */ 2250 int (*init)(struct sock *sk); 2251 /* update ulp */ 2252 void (*update)(struct sock *sk, struct proto *p, 2253 void (*write_space)(struct sock *sk)); 2254 /* cleanup ulp */ 2255 void (*release)(struct sock *sk); 2256 /* diagnostic */ 2257 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2258 size_t (*get_info_size)(const struct sock *sk); 2259 /* clone ulp */ 2260 void (*clone)(const struct request_sock *req, struct sock *newsk, 2261 const gfp_t priority); 2262 2263 char name[TCP_ULP_NAME_MAX]; 2264 struct module *owner; 2265 }; 2266 int tcp_register_ulp(struct tcp_ulp_ops *type); 2267 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2268 int tcp_set_ulp(struct sock *sk, const char *name); 2269 void tcp_get_available_ulp(char *buf, size_t len); 2270 void tcp_cleanup_ulp(struct sock *sk); 2271 void tcp_update_ulp(struct sock *sk, struct proto *p, 2272 void (*write_space)(struct sock *sk)); 2273 2274 #define MODULE_ALIAS_TCP_ULP(name) \ 2275 __MODULE_INFO(alias, alias_userspace, name); \ 2276 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2277 2278 #ifdef CONFIG_NET_SOCK_MSG 2279 struct sk_msg; 2280 struct sk_psock; 2281 2282 #ifdef CONFIG_BPF_SYSCALL 2283 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2284 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2285 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2286 #endif /* CONFIG_BPF_SYSCALL */ 2287 2288 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2289 int flags); 2290 #endif /* CONFIG_NET_SOCK_MSG */ 2291 2292 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2293 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2294 { 2295 } 2296 #endif 2297 2298 #ifdef CONFIG_CGROUP_BPF 2299 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2300 struct sk_buff *skb, 2301 unsigned int end_offset) 2302 { 2303 skops->skb = skb; 2304 skops->skb_data_end = skb->data + end_offset; 2305 } 2306 #else 2307 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2308 struct sk_buff *skb, 2309 unsigned int end_offset) 2310 { 2311 } 2312 #endif 2313 2314 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2315 * is < 0, then the BPF op failed (for example if the loaded BPF 2316 * program does not support the chosen operation or there is no BPF 2317 * program loaded). 2318 */ 2319 #ifdef CONFIG_BPF 2320 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2321 { 2322 struct bpf_sock_ops_kern sock_ops; 2323 int ret; 2324 2325 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2326 if (sk_fullsock(sk)) { 2327 sock_ops.is_fullsock = 1; 2328 sock_owned_by_me(sk); 2329 } 2330 2331 sock_ops.sk = sk; 2332 sock_ops.op = op; 2333 if (nargs > 0) 2334 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2335 2336 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2337 if (ret == 0) 2338 ret = sock_ops.reply; 2339 else 2340 ret = -1; 2341 return ret; 2342 } 2343 2344 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2345 { 2346 u32 args[2] = {arg1, arg2}; 2347 2348 return tcp_call_bpf(sk, op, 2, args); 2349 } 2350 2351 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2352 u32 arg3) 2353 { 2354 u32 args[3] = {arg1, arg2, arg3}; 2355 2356 return tcp_call_bpf(sk, op, 3, args); 2357 } 2358 2359 #else 2360 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2361 { 2362 return -EPERM; 2363 } 2364 2365 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2366 { 2367 return -EPERM; 2368 } 2369 2370 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2371 u32 arg3) 2372 { 2373 return -EPERM; 2374 } 2375 2376 #endif 2377 2378 static inline u32 tcp_timeout_init(struct sock *sk) 2379 { 2380 int timeout; 2381 2382 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2383 2384 if (timeout <= 0) 2385 timeout = TCP_TIMEOUT_INIT; 2386 return min_t(int, timeout, TCP_RTO_MAX); 2387 } 2388 2389 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2390 { 2391 int rwnd; 2392 2393 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2394 2395 if (rwnd < 0) 2396 rwnd = 0; 2397 return rwnd; 2398 } 2399 2400 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2401 { 2402 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2403 } 2404 2405 static inline void tcp_bpf_rtt(struct sock *sk) 2406 { 2407 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2408 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2409 } 2410 2411 #if IS_ENABLED(CONFIG_SMC) 2412 extern struct static_key_false tcp_have_smc; 2413 #endif 2414 2415 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2416 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2417 void (*cad)(struct sock *sk, u32 ack_seq)); 2418 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2419 void clean_acked_data_flush(void); 2420 #endif 2421 2422 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2423 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2424 const struct tcp_sock *tp) 2425 { 2426 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2427 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2428 } 2429 2430 /* Compute Earliest Departure Time for some control packets 2431 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2432 */ 2433 static inline u64 tcp_transmit_time(const struct sock *sk) 2434 { 2435 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2436 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2437 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2438 2439 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2440 } 2441 return 0; 2442 } 2443 2444 #endif /* _TCP_H */ 2445