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 extern struct percpu_counter tcp_sockets_allocated; 257 extern unsigned long tcp_memory_pressure; 258 259 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 260 static inline bool tcp_under_memory_pressure(const struct sock *sk) 261 { 262 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 263 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 264 return true; 265 266 return READ_ONCE(tcp_memory_pressure); 267 } 268 /* 269 * The next routines deal with comparing 32 bit unsigned ints 270 * and worry about wraparound (automatic with unsigned arithmetic). 271 */ 272 273 static inline bool before(__u32 seq1, __u32 seq2) 274 { 275 return (__s32)(seq1-seq2) < 0; 276 } 277 #define after(seq2, seq1) before(seq1, seq2) 278 279 /* is s2<=s1<=s3 ? */ 280 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 281 { 282 return seq3 - seq2 >= seq1 - seq2; 283 } 284 285 static inline bool tcp_out_of_memory(struct sock *sk) 286 { 287 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 288 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 289 return true; 290 return false; 291 } 292 293 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 294 { 295 sk_wmem_queued_add(sk, -skb->truesize); 296 if (!skb_zcopy_pure(skb)) 297 sk_mem_uncharge(sk, skb->truesize); 298 else 299 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 300 __kfree_skb(skb); 301 } 302 303 void sk_forced_mem_schedule(struct sock *sk, int size); 304 305 bool tcp_check_oom(struct sock *sk, int shift); 306 307 308 extern struct proto tcp_prot; 309 310 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 311 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 312 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 313 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 314 315 void tcp_tasklet_init(void); 316 317 int tcp_v4_err(struct sk_buff *skb, u32); 318 319 void tcp_shutdown(struct sock *sk, int how); 320 321 int tcp_v4_early_demux(struct sk_buff *skb); 322 int tcp_v4_rcv(struct sk_buff *skb); 323 324 void tcp_remove_empty_skb(struct sock *sk); 325 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 326 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 327 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 328 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 329 int flags); 330 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 331 size_t size, int flags); 332 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 333 size_t size, int flags); 334 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 335 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 336 int size_goal); 337 void tcp_release_cb(struct sock *sk); 338 void tcp_wfree(struct sk_buff *skb); 339 void tcp_write_timer_handler(struct sock *sk); 340 void tcp_delack_timer_handler(struct sock *sk); 341 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 342 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 343 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 344 void tcp_rcv_space_adjust(struct sock *sk); 345 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 346 void tcp_twsk_destructor(struct sock *sk); 347 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, int size, gfp_t gfp, 351 bool force_schedule); 352 353 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 354 static inline void tcp_dec_quickack_mode(struct sock *sk, 355 const unsigned int pkts) 356 { 357 struct inet_connection_sock *icsk = inet_csk(sk); 358 359 if (icsk->icsk_ack.quick) { 360 if (pkts >= icsk->icsk_ack.quick) { 361 icsk->icsk_ack.quick = 0; 362 /* Leaving quickack mode we deflate ATO. */ 363 icsk->icsk_ack.ato = TCP_ATO_MIN; 364 } else 365 icsk->icsk_ack.quick -= pkts; 366 } 367 } 368 369 #define TCP_ECN_OK 1 370 #define TCP_ECN_QUEUE_CWR 2 371 #define TCP_ECN_DEMAND_CWR 4 372 #define TCP_ECN_SEEN 8 373 374 enum tcp_tw_status { 375 TCP_TW_SUCCESS = 0, 376 TCP_TW_RST = 1, 377 TCP_TW_ACK = 2, 378 TCP_TW_SYN = 3 379 }; 380 381 382 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 383 struct sk_buff *skb, 384 const struct tcphdr *th); 385 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 386 struct request_sock *req, bool fastopen, 387 bool *lost_race); 388 int tcp_child_process(struct sock *parent, struct sock *child, 389 struct sk_buff *skb); 390 void tcp_enter_loss(struct sock *sk); 391 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 392 void tcp_clear_retrans(struct tcp_sock *tp); 393 void tcp_update_metrics(struct sock *sk); 394 void tcp_init_metrics(struct sock *sk); 395 void tcp_metrics_init(void); 396 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 397 void __tcp_close(struct sock *sk, long timeout); 398 void tcp_close(struct sock *sk, long timeout); 399 void tcp_init_sock(struct sock *sk); 400 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 401 __poll_t tcp_poll(struct file *file, struct socket *sock, 402 struct poll_table_struct *wait); 403 int tcp_getsockopt(struct sock *sk, int level, int optname, 404 char __user *optval, int __user *optlen); 405 bool tcp_bpf_bypass_getsockopt(int level, int optname); 406 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 407 unsigned int optlen); 408 void tcp_set_keepalive(struct sock *sk, int val); 409 void tcp_syn_ack_timeout(const struct request_sock *req); 410 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 411 int flags, int *addr_len); 412 int tcp_set_rcvlowat(struct sock *sk, int val); 413 int tcp_set_window_clamp(struct sock *sk, int val); 414 void tcp_update_recv_tstamps(struct sk_buff *skb, 415 struct scm_timestamping_internal *tss); 416 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 417 struct scm_timestamping_internal *tss); 418 void tcp_data_ready(struct sock *sk); 419 #ifdef CONFIG_MMU 420 int tcp_mmap(struct file *file, struct socket *sock, 421 struct vm_area_struct *vma); 422 #endif 423 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 424 struct tcp_options_received *opt_rx, 425 int estab, struct tcp_fastopen_cookie *foc); 426 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 427 428 /* 429 * BPF SKB-less helpers 430 */ 431 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 432 struct tcphdr *th, u32 *cookie); 433 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 434 struct tcphdr *th, u32 *cookie); 435 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 436 const struct tcp_request_sock_ops *af_ops, 437 struct sock *sk, struct tcphdr *th); 438 /* 439 * TCP v4 functions exported for the inet6 API 440 */ 441 442 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 443 void tcp_v4_mtu_reduced(struct sock *sk); 444 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 445 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 446 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 447 struct sock *tcp_create_openreq_child(const struct sock *sk, 448 struct request_sock *req, 449 struct sk_buff *skb); 450 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 451 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 452 struct request_sock *req, 453 struct dst_entry *dst, 454 struct request_sock *req_unhash, 455 bool *own_req); 456 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 457 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 458 int tcp_connect(struct sock *sk); 459 enum tcp_synack_type { 460 TCP_SYNACK_NORMAL, 461 TCP_SYNACK_FASTOPEN, 462 TCP_SYNACK_COOKIE, 463 }; 464 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 465 struct request_sock *req, 466 struct tcp_fastopen_cookie *foc, 467 enum tcp_synack_type synack_type, 468 struct sk_buff *syn_skb); 469 int tcp_disconnect(struct sock *sk, int flags); 470 471 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 472 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 473 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 474 475 /* From syncookies.c */ 476 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 477 struct request_sock *req, 478 struct dst_entry *dst, u32 tsoff); 479 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 480 u32 cookie); 481 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 482 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 483 const struct tcp_request_sock_ops *af_ops, 484 struct sock *sk, struct sk_buff *skb); 485 #ifdef CONFIG_SYN_COOKIES 486 487 /* Syncookies use a monotonic timer which increments every 60 seconds. 488 * This counter is used both as a hash input and partially encoded into 489 * the cookie value. A cookie is only validated further if the delta 490 * between the current counter value and the encoded one is less than this, 491 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 492 * the counter advances immediately after a cookie is generated). 493 */ 494 #define MAX_SYNCOOKIE_AGE 2 495 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 496 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 497 498 /* syncookies: remember time of last synqueue overflow 499 * But do not dirty this field too often (once per second is enough) 500 * It is racy as we do not hold a lock, but race is very minor. 501 */ 502 static inline void tcp_synq_overflow(const struct sock *sk) 503 { 504 unsigned int last_overflow; 505 unsigned int now = jiffies; 506 507 if (sk->sk_reuseport) { 508 struct sock_reuseport *reuse; 509 510 reuse = rcu_dereference(sk->sk_reuseport_cb); 511 if (likely(reuse)) { 512 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 513 if (!time_between32(now, last_overflow, 514 last_overflow + HZ)) 515 WRITE_ONCE(reuse->synq_overflow_ts, now); 516 return; 517 } 518 } 519 520 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 521 if (!time_between32(now, last_overflow, last_overflow + HZ)) 522 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); 523 } 524 525 /* syncookies: no recent synqueue overflow on this listening socket? */ 526 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 527 { 528 unsigned int last_overflow; 529 unsigned int now = jiffies; 530 531 if (sk->sk_reuseport) { 532 struct sock_reuseport *reuse; 533 534 reuse = rcu_dereference(sk->sk_reuseport_cb); 535 if (likely(reuse)) { 536 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 537 return !time_between32(now, last_overflow - HZ, 538 last_overflow + 539 TCP_SYNCOOKIE_VALID); 540 } 541 } 542 543 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 544 545 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 546 * then we're under synflood. However, we have to use 547 * 'last_overflow - HZ' as lower bound. That's because a concurrent 548 * tcp_synq_overflow() could update .ts_recent_stamp after we read 549 * jiffies but before we store .ts_recent_stamp into last_overflow, 550 * which could lead to rejecting a valid syncookie. 551 */ 552 return !time_between32(now, last_overflow - HZ, 553 last_overflow + TCP_SYNCOOKIE_VALID); 554 } 555 556 static inline u32 tcp_cookie_time(void) 557 { 558 u64 val = get_jiffies_64(); 559 560 do_div(val, TCP_SYNCOOKIE_PERIOD); 561 return val; 562 } 563 564 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 565 u16 *mssp); 566 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 567 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 568 bool cookie_timestamp_decode(const struct net *net, 569 struct tcp_options_received *opt); 570 bool cookie_ecn_ok(const struct tcp_options_received *opt, 571 const struct net *net, const struct dst_entry *dst); 572 573 /* From net/ipv6/syncookies.c */ 574 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 575 u32 cookie); 576 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 577 578 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 579 const struct tcphdr *th, u16 *mssp); 580 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 581 #endif 582 /* tcp_output.c */ 583 584 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 585 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 586 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 587 int nonagle); 588 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 589 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 590 void tcp_retransmit_timer(struct sock *sk); 591 void tcp_xmit_retransmit_queue(struct sock *); 592 void tcp_simple_retransmit(struct sock *); 593 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 594 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 595 enum tcp_queue { 596 TCP_FRAG_IN_WRITE_QUEUE, 597 TCP_FRAG_IN_RTX_QUEUE, 598 }; 599 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 600 struct sk_buff *skb, u32 len, 601 unsigned int mss_now, gfp_t gfp); 602 603 void tcp_send_probe0(struct sock *); 604 void tcp_send_partial(struct sock *); 605 int tcp_write_wakeup(struct sock *, int mib); 606 void tcp_send_fin(struct sock *sk); 607 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 608 int tcp_send_synack(struct sock *); 609 void tcp_push_one(struct sock *, unsigned int mss_now); 610 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 611 void tcp_send_ack(struct sock *sk); 612 void tcp_send_delayed_ack(struct sock *sk); 613 void tcp_send_loss_probe(struct sock *sk); 614 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 615 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 616 const struct sk_buff *next_skb); 617 618 /* tcp_input.c */ 619 void tcp_rearm_rto(struct sock *sk); 620 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 621 void tcp_reset(struct sock *sk, struct sk_buff *skb); 622 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 623 void tcp_fin(struct sock *sk); 624 void tcp_check_space(struct sock *sk); 625 626 /* tcp_timer.c */ 627 void tcp_init_xmit_timers(struct sock *); 628 static inline void tcp_clear_xmit_timers(struct sock *sk) 629 { 630 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 631 __sock_put(sk); 632 633 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 634 __sock_put(sk); 635 636 inet_csk_clear_xmit_timers(sk); 637 } 638 639 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 640 unsigned int tcp_current_mss(struct sock *sk); 641 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 642 643 /* Bound MSS / TSO packet size with the half of the window */ 644 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 645 { 646 int cutoff; 647 648 /* When peer uses tiny windows, there is no use in packetizing 649 * to sub-MSS pieces for the sake of SWS or making sure there 650 * are enough packets in the pipe for fast recovery. 651 * 652 * On the other hand, for extremely large MSS devices, handling 653 * smaller than MSS windows in this way does make sense. 654 */ 655 if (tp->max_window > TCP_MSS_DEFAULT) 656 cutoff = (tp->max_window >> 1); 657 else 658 cutoff = tp->max_window; 659 660 if (cutoff && pktsize > cutoff) 661 return max_t(int, cutoff, 68U - tp->tcp_header_len); 662 else 663 return pktsize; 664 } 665 666 /* tcp.c */ 667 void tcp_get_info(struct sock *, struct tcp_info *); 668 669 /* Read 'sendfile()'-style from a TCP socket */ 670 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 671 sk_read_actor_t recv_actor); 672 673 void tcp_initialize_rcv_mss(struct sock *sk); 674 675 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 676 int tcp_mss_to_mtu(struct sock *sk, int mss); 677 void tcp_mtup_init(struct sock *sk); 678 679 static inline void tcp_bound_rto(const struct sock *sk) 680 { 681 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 682 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 683 } 684 685 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 686 { 687 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 688 } 689 690 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 691 { 692 /* mptcp hooks are only on the slow path */ 693 if (sk_is_mptcp((struct sock *)tp)) 694 return; 695 696 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 697 ntohl(TCP_FLAG_ACK) | 698 snd_wnd); 699 } 700 701 static inline void tcp_fast_path_on(struct tcp_sock *tp) 702 { 703 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 704 } 705 706 static inline void tcp_fast_path_check(struct sock *sk) 707 { 708 struct tcp_sock *tp = tcp_sk(sk); 709 710 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 711 tp->rcv_wnd && 712 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 713 !tp->urg_data) 714 tcp_fast_path_on(tp); 715 } 716 717 /* Compute the actual rto_min value */ 718 static inline u32 tcp_rto_min(struct sock *sk) 719 { 720 const struct dst_entry *dst = __sk_dst_get(sk); 721 u32 rto_min = inet_csk(sk)->icsk_rto_min; 722 723 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 724 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 725 return rto_min; 726 } 727 728 static inline u32 tcp_rto_min_us(struct sock *sk) 729 { 730 return jiffies_to_usecs(tcp_rto_min(sk)); 731 } 732 733 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 734 { 735 return dst_metric_locked(dst, RTAX_CC_ALGO); 736 } 737 738 /* Minimum RTT in usec. ~0 means not available. */ 739 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 740 { 741 return minmax_get(&tp->rtt_min); 742 } 743 744 /* Compute the actual receive window we are currently advertising. 745 * Rcv_nxt can be after the window if our peer push more data 746 * than the offered window. 747 */ 748 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 749 { 750 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 751 752 if (win < 0) 753 win = 0; 754 return (u32) win; 755 } 756 757 /* Choose a new window, without checks for shrinking, and without 758 * scaling applied to the result. The caller does these things 759 * if necessary. This is a "raw" window selection. 760 */ 761 u32 __tcp_select_window(struct sock *sk); 762 763 void tcp_send_window_probe(struct sock *sk); 764 765 /* TCP uses 32bit jiffies to save some space. 766 * Note that this is different from tcp_time_stamp, which 767 * historically has been the same until linux-4.13. 768 */ 769 #define tcp_jiffies32 ((u32)jiffies) 770 771 /* 772 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 773 * It is no longer tied to jiffies, but to 1 ms clock. 774 * Note: double check if you want to use tcp_jiffies32 instead of this. 775 */ 776 #define TCP_TS_HZ 1000 777 778 static inline u64 tcp_clock_ns(void) 779 { 780 return ktime_get_ns(); 781 } 782 783 static inline u64 tcp_clock_us(void) 784 { 785 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 786 } 787 788 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 789 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 790 { 791 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 792 } 793 794 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 795 static inline u32 tcp_ns_to_ts(u64 ns) 796 { 797 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 798 } 799 800 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 801 static inline u32 tcp_time_stamp_raw(void) 802 { 803 return tcp_ns_to_ts(tcp_clock_ns()); 804 } 805 806 void tcp_mstamp_refresh(struct tcp_sock *tp); 807 808 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 809 { 810 return max_t(s64, t1 - t0, 0); 811 } 812 813 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 814 { 815 return tcp_ns_to_ts(skb->skb_mstamp_ns); 816 } 817 818 /* provide the departure time in us unit */ 819 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 820 { 821 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 822 } 823 824 825 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 826 827 #define TCPHDR_FIN 0x01 828 #define TCPHDR_SYN 0x02 829 #define TCPHDR_RST 0x04 830 #define TCPHDR_PSH 0x08 831 #define TCPHDR_ACK 0x10 832 #define TCPHDR_URG 0x20 833 #define TCPHDR_ECE 0x40 834 #define TCPHDR_CWR 0x80 835 836 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 837 838 /* This is what the send packet queuing engine uses to pass 839 * TCP per-packet control information to the transmission code. 840 * We also store the host-order sequence numbers in here too. 841 * This is 44 bytes if IPV6 is enabled. 842 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 843 */ 844 struct tcp_skb_cb { 845 __u32 seq; /* Starting sequence number */ 846 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 847 union { 848 /* Note : tcp_tw_isn is used in input path only 849 * (isn chosen by tcp_timewait_state_process()) 850 * 851 * tcp_gso_segs/size are used in write queue only, 852 * cf tcp_skb_pcount()/tcp_skb_mss() 853 */ 854 __u32 tcp_tw_isn; 855 struct { 856 u16 tcp_gso_segs; 857 u16 tcp_gso_size; 858 }; 859 }; 860 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 861 862 __u8 sacked; /* State flags for SACK. */ 863 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 864 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 865 #define TCPCB_LOST 0x04 /* SKB is lost */ 866 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 867 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 868 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 869 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 870 TCPCB_REPAIRED) 871 872 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 873 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 874 eor:1, /* Is skb MSG_EOR marked? */ 875 has_rxtstamp:1, /* SKB has a RX timestamp */ 876 unused:5; 877 __u32 ack_seq; /* Sequence number ACK'd */ 878 union { 879 struct { 880 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 881 /* There is space for up to 24 bytes */ 882 __u32 is_app_limited:1, /* cwnd not fully used? */ 883 delivered_ce:20, 884 unused:11; 885 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 886 __u32 delivered; 887 /* start of send pipeline phase */ 888 u64 first_tx_mstamp; 889 /* when we reached the "delivered" count */ 890 u64 delivered_mstamp; 891 } tx; /* only used for outgoing skbs */ 892 union { 893 struct inet_skb_parm h4; 894 #if IS_ENABLED(CONFIG_IPV6) 895 struct inet6_skb_parm h6; 896 #endif 897 } header; /* For incoming skbs */ 898 }; 899 }; 900 901 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 902 903 extern const struct inet_connection_sock_af_ops ipv4_specific; 904 905 #if IS_ENABLED(CONFIG_IPV6) 906 /* This is the variant of inet6_iif() that must be used by TCP, 907 * as TCP moves IP6CB into a different location in skb->cb[] 908 */ 909 static inline int tcp_v6_iif(const struct sk_buff *skb) 910 { 911 return TCP_SKB_CB(skb)->header.h6.iif; 912 } 913 914 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 915 { 916 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 917 918 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 919 } 920 921 /* TCP_SKB_CB reference means this can not be used from early demux */ 922 static inline int tcp_v6_sdif(const struct sk_buff *skb) 923 { 924 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 925 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 926 return TCP_SKB_CB(skb)->header.h6.iif; 927 #endif 928 return 0; 929 } 930 931 extern const struct inet_connection_sock_af_ops ipv6_specific; 932 933 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 934 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 935 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb)); 936 937 #endif 938 939 /* TCP_SKB_CB reference means this can not be used from early demux */ 940 static inline int tcp_v4_sdif(struct sk_buff *skb) 941 { 942 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 943 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 944 return TCP_SKB_CB(skb)->header.h4.iif; 945 #endif 946 return 0; 947 } 948 949 /* Due to TSO, an SKB can be composed of multiple actual 950 * packets. To keep these tracked properly, we use this. 951 */ 952 static inline int tcp_skb_pcount(const struct sk_buff *skb) 953 { 954 return TCP_SKB_CB(skb)->tcp_gso_segs; 955 } 956 957 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 958 { 959 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 960 } 961 962 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 963 { 964 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 965 } 966 967 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 968 static inline int tcp_skb_mss(const struct sk_buff *skb) 969 { 970 return TCP_SKB_CB(skb)->tcp_gso_size; 971 } 972 973 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 974 { 975 return likely(!TCP_SKB_CB(skb)->eor); 976 } 977 978 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 979 const struct sk_buff *from) 980 { 981 return likely(tcp_skb_can_collapse_to(to) && 982 mptcp_skb_can_collapse(to, from) && 983 skb_pure_zcopy_same(to, from)); 984 } 985 986 /* Events passed to congestion control interface */ 987 enum tcp_ca_event { 988 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 989 CA_EVENT_CWND_RESTART, /* congestion window restart */ 990 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 991 CA_EVENT_LOSS, /* loss timeout */ 992 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 993 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 994 }; 995 996 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 997 enum tcp_ca_ack_event_flags { 998 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 999 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1000 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1001 }; 1002 1003 /* 1004 * Interface for adding new TCP congestion control handlers 1005 */ 1006 #define TCP_CA_NAME_MAX 16 1007 #define TCP_CA_MAX 128 1008 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1009 1010 #define TCP_CA_UNSPEC 0 1011 1012 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1013 #define TCP_CONG_NON_RESTRICTED 0x1 1014 /* Requires ECN/ECT set on all packets */ 1015 #define TCP_CONG_NEEDS_ECN 0x2 1016 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1017 1018 union tcp_cc_info; 1019 1020 struct ack_sample { 1021 u32 pkts_acked; 1022 s32 rtt_us; 1023 u32 in_flight; 1024 }; 1025 1026 /* A rate sample measures the number of (original/retransmitted) data 1027 * packets delivered "delivered" over an interval of time "interval_us". 1028 * The tcp_rate.c code fills in the rate sample, and congestion 1029 * control modules that define a cong_control function to run at the end 1030 * of ACK processing can optionally chose to consult this sample when 1031 * setting cwnd and pacing rate. 1032 * A sample is invalid if "delivered" or "interval_us" is negative. 1033 */ 1034 struct rate_sample { 1035 u64 prior_mstamp; /* starting timestamp for interval */ 1036 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1037 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1038 s32 delivered; /* number of packets delivered over interval */ 1039 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1040 long interval_us; /* time for tp->delivered to incr "delivered" */ 1041 u32 snd_interval_us; /* snd interval for delivered packets */ 1042 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1043 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1044 int losses; /* number of packets marked lost upon ACK */ 1045 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1046 u32 prior_in_flight; /* in flight before this ACK */ 1047 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1048 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1049 bool is_retrans; /* is sample from retransmission? */ 1050 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1051 }; 1052 1053 struct tcp_congestion_ops { 1054 /* fast path fields are put first to fill one cache line */ 1055 1056 /* return slow start threshold (required) */ 1057 u32 (*ssthresh)(struct sock *sk); 1058 1059 /* do new cwnd calculation (required) */ 1060 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1061 1062 /* call before changing ca_state (optional) */ 1063 void (*set_state)(struct sock *sk, u8 new_state); 1064 1065 /* call when cwnd event occurs (optional) */ 1066 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1067 1068 /* call when ack arrives (optional) */ 1069 void (*in_ack_event)(struct sock *sk, u32 flags); 1070 1071 /* hook for packet ack accounting (optional) */ 1072 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1073 1074 /* override sysctl_tcp_min_tso_segs */ 1075 u32 (*min_tso_segs)(struct sock *sk); 1076 1077 /* call when packets are delivered to update cwnd and pacing rate, 1078 * after all the ca_state processing. (optional) 1079 */ 1080 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1081 1082 1083 /* new value of cwnd after loss (required) */ 1084 u32 (*undo_cwnd)(struct sock *sk); 1085 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1086 u32 (*sndbuf_expand)(struct sock *sk); 1087 1088 /* control/slow paths put last */ 1089 /* get info for inet_diag (optional) */ 1090 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1091 union tcp_cc_info *info); 1092 1093 char name[TCP_CA_NAME_MAX]; 1094 struct module *owner; 1095 struct list_head list; 1096 u32 key; 1097 u32 flags; 1098 1099 /* initialize private data (optional) */ 1100 void (*init)(struct sock *sk); 1101 /* cleanup private data (optional) */ 1102 void (*release)(struct sock *sk); 1103 } ____cacheline_aligned_in_smp; 1104 1105 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1106 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1107 1108 void tcp_assign_congestion_control(struct sock *sk); 1109 void tcp_init_congestion_control(struct sock *sk); 1110 void tcp_cleanup_congestion_control(struct sock *sk); 1111 int tcp_set_default_congestion_control(struct net *net, const char *name); 1112 void tcp_get_default_congestion_control(struct net *net, char *name); 1113 void tcp_get_available_congestion_control(char *buf, size_t len); 1114 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1115 int tcp_set_allowed_congestion_control(char *allowed); 1116 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1117 bool cap_net_admin); 1118 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1119 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1120 1121 u32 tcp_reno_ssthresh(struct sock *sk); 1122 u32 tcp_reno_undo_cwnd(struct sock *sk); 1123 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1124 extern struct tcp_congestion_ops tcp_reno; 1125 1126 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1127 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1128 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1129 #ifdef CONFIG_INET 1130 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1131 #else 1132 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1133 { 1134 return NULL; 1135 } 1136 #endif 1137 1138 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1139 { 1140 const struct inet_connection_sock *icsk = inet_csk(sk); 1141 1142 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1143 } 1144 1145 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1146 { 1147 struct inet_connection_sock *icsk = inet_csk(sk); 1148 1149 if (icsk->icsk_ca_ops->set_state) 1150 icsk->icsk_ca_ops->set_state(sk, ca_state); 1151 icsk->icsk_ca_state = ca_state; 1152 } 1153 1154 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1155 { 1156 const struct inet_connection_sock *icsk = inet_csk(sk); 1157 1158 if (icsk->icsk_ca_ops->cwnd_event) 1159 icsk->icsk_ca_ops->cwnd_event(sk, event); 1160 } 1161 1162 /* From tcp_rate.c */ 1163 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1164 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1165 struct rate_sample *rs); 1166 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1167 bool is_sack_reneg, struct rate_sample *rs); 1168 void tcp_rate_check_app_limited(struct sock *sk); 1169 1170 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1171 { 1172 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1173 } 1174 1175 /* These functions determine how the current flow behaves in respect of SACK 1176 * handling. SACK is negotiated with the peer, and therefore it can vary 1177 * between different flows. 1178 * 1179 * tcp_is_sack - SACK enabled 1180 * tcp_is_reno - No SACK 1181 */ 1182 static inline int tcp_is_sack(const struct tcp_sock *tp) 1183 { 1184 return likely(tp->rx_opt.sack_ok); 1185 } 1186 1187 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1188 { 1189 return !tcp_is_sack(tp); 1190 } 1191 1192 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1193 { 1194 return tp->sacked_out + tp->lost_out; 1195 } 1196 1197 /* This determines how many packets are "in the network" to the best 1198 * of our knowledge. In many cases it is conservative, but where 1199 * detailed information is available from the receiver (via SACK 1200 * blocks etc.) we can make more aggressive calculations. 1201 * 1202 * Use this for decisions involving congestion control, use just 1203 * tp->packets_out to determine if the send queue is empty or not. 1204 * 1205 * Read this equation as: 1206 * 1207 * "Packets sent once on transmission queue" MINUS 1208 * "Packets left network, but not honestly ACKed yet" PLUS 1209 * "Packets fast retransmitted" 1210 */ 1211 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1212 { 1213 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1214 } 1215 1216 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1217 1218 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1219 { 1220 return tp->snd_cwnd < tp->snd_ssthresh; 1221 } 1222 1223 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1224 { 1225 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1226 } 1227 1228 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1229 { 1230 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1231 (1 << inet_csk(sk)->icsk_ca_state); 1232 } 1233 1234 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1235 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1236 * ssthresh. 1237 */ 1238 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1239 { 1240 const struct tcp_sock *tp = tcp_sk(sk); 1241 1242 if (tcp_in_cwnd_reduction(sk)) 1243 return tp->snd_ssthresh; 1244 else 1245 return max(tp->snd_ssthresh, 1246 ((tp->snd_cwnd >> 1) + 1247 (tp->snd_cwnd >> 2))); 1248 } 1249 1250 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1251 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1252 1253 void tcp_enter_cwr(struct sock *sk); 1254 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1255 1256 /* The maximum number of MSS of available cwnd for which TSO defers 1257 * sending if not using sysctl_tcp_tso_win_divisor. 1258 */ 1259 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1260 { 1261 return 3; 1262 } 1263 1264 /* Returns end sequence number of the receiver's advertised window */ 1265 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1266 { 1267 return tp->snd_una + tp->snd_wnd; 1268 } 1269 1270 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1271 * flexible approach. The RFC suggests cwnd should not be raised unless 1272 * it was fully used previously. And that's exactly what we do in 1273 * congestion avoidance mode. But in slow start we allow cwnd to grow 1274 * as long as the application has used half the cwnd. 1275 * Example : 1276 * cwnd is 10 (IW10), but application sends 9 frames. 1277 * We allow cwnd to reach 18 when all frames are ACKed. 1278 * This check is safe because it's as aggressive as slow start which already 1279 * risks 100% overshoot. The advantage is that we discourage application to 1280 * either send more filler packets or data to artificially blow up the cwnd 1281 * usage, and allow application-limited process to probe bw more aggressively. 1282 */ 1283 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1284 { 1285 const struct tcp_sock *tp = tcp_sk(sk); 1286 1287 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1288 if (tcp_in_slow_start(tp)) 1289 return tp->snd_cwnd < 2 * tp->max_packets_out; 1290 1291 return tp->is_cwnd_limited; 1292 } 1293 1294 /* BBR congestion control needs pacing. 1295 * Same remark for SO_MAX_PACING_RATE. 1296 * sch_fq packet scheduler is efficiently handling pacing, 1297 * but is not always installed/used. 1298 * Return true if TCP stack should pace packets itself. 1299 */ 1300 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1301 { 1302 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1303 } 1304 1305 /* Estimates in how many jiffies next packet for this flow can be sent. 1306 * Scheduling a retransmit timer too early would be silly. 1307 */ 1308 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1309 { 1310 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1311 1312 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1313 } 1314 1315 static inline void tcp_reset_xmit_timer(struct sock *sk, 1316 const int what, 1317 unsigned long when, 1318 const unsigned long max_when) 1319 { 1320 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1321 max_when); 1322 } 1323 1324 /* Something is really bad, we could not queue an additional packet, 1325 * because qdisc is full or receiver sent a 0 window, or we are paced. 1326 * We do not want to add fuel to the fire, or abort too early, 1327 * so make sure the timer we arm now is at least 200ms in the future, 1328 * regardless of current icsk_rto value (as it could be ~2ms) 1329 */ 1330 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1331 { 1332 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1333 } 1334 1335 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1336 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1337 unsigned long max_when) 1338 { 1339 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1340 inet_csk(sk)->icsk_backoff); 1341 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1342 1343 return (unsigned long)min_t(u64, when, max_when); 1344 } 1345 1346 static inline void tcp_check_probe_timer(struct sock *sk) 1347 { 1348 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1349 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1350 tcp_probe0_base(sk), TCP_RTO_MAX); 1351 } 1352 1353 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1354 { 1355 tp->snd_wl1 = seq; 1356 } 1357 1358 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1359 { 1360 tp->snd_wl1 = seq; 1361 } 1362 1363 /* 1364 * Calculate(/check) TCP checksum 1365 */ 1366 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1367 __be32 daddr, __wsum base) 1368 { 1369 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1370 } 1371 1372 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1373 { 1374 return !skb_csum_unnecessary(skb) && 1375 __skb_checksum_complete(skb); 1376 } 1377 1378 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1379 enum skb_drop_reason *reason); 1380 1381 #ifdef CONFIG_INET 1382 void __sk_defer_free_flush(struct sock *sk); 1383 1384 static inline void sk_defer_free_flush(struct sock *sk) 1385 { 1386 if (llist_empty(&sk->defer_list)) 1387 return; 1388 __sk_defer_free_flush(sk); 1389 } 1390 #else 1391 static inline void sk_defer_free_flush(struct sock *sk) {} 1392 #endif 1393 1394 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1395 void tcp_set_state(struct sock *sk, int state); 1396 void tcp_done(struct sock *sk); 1397 int tcp_abort(struct sock *sk, int err); 1398 1399 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1400 { 1401 rx_opt->dsack = 0; 1402 rx_opt->num_sacks = 0; 1403 } 1404 1405 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1406 1407 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1408 { 1409 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1410 struct tcp_sock *tp = tcp_sk(sk); 1411 s32 delta; 1412 1413 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || 1414 ca_ops->cong_control) 1415 return; 1416 delta = tcp_jiffies32 - tp->lsndtime; 1417 if (delta > inet_csk(sk)->icsk_rto) 1418 tcp_cwnd_restart(sk, delta); 1419 } 1420 1421 /* Determine a window scaling and initial window to offer. */ 1422 void tcp_select_initial_window(const struct sock *sk, int __space, 1423 __u32 mss, __u32 *rcv_wnd, 1424 __u32 *window_clamp, int wscale_ok, 1425 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1426 1427 static inline int tcp_win_from_space(const struct sock *sk, int space) 1428 { 1429 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; 1430 1431 return tcp_adv_win_scale <= 0 ? 1432 (space>>(-tcp_adv_win_scale)) : 1433 space - (space>>tcp_adv_win_scale); 1434 } 1435 1436 /* Note: caller must be prepared to deal with negative returns */ 1437 static inline int tcp_space(const struct sock *sk) 1438 { 1439 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1440 READ_ONCE(sk->sk_backlog.len) - 1441 atomic_read(&sk->sk_rmem_alloc)); 1442 } 1443 1444 static inline int tcp_full_space(const struct sock *sk) 1445 { 1446 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1447 } 1448 1449 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1450 { 1451 int unused_mem = sk_unused_reserved_mem(sk); 1452 struct tcp_sock *tp = tcp_sk(sk); 1453 1454 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1455 if (unused_mem) 1456 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1457 tcp_win_from_space(sk, unused_mem)); 1458 } 1459 1460 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1461 1462 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1463 * If 87.5 % (7/8) of the space has been consumed, we want to override 1464 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1465 * len/truesize ratio. 1466 */ 1467 static inline bool tcp_rmem_pressure(const struct sock *sk) 1468 { 1469 int rcvbuf, threshold; 1470 1471 if (tcp_under_memory_pressure(sk)) 1472 return true; 1473 1474 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1475 threshold = rcvbuf - (rcvbuf >> 3); 1476 1477 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1478 } 1479 1480 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1481 { 1482 const struct tcp_sock *tp = tcp_sk(sk); 1483 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1484 1485 if (avail <= 0) 1486 return false; 1487 1488 return (avail >= target) || tcp_rmem_pressure(sk) || 1489 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1490 } 1491 1492 extern void tcp_openreq_init_rwin(struct request_sock *req, 1493 const struct sock *sk_listener, 1494 const struct dst_entry *dst); 1495 1496 void tcp_enter_memory_pressure(struct sock *sk); 1497 void tcp_leave_memory_pressure(struct sock *sk); 1498 1499 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1500 { 1501 struct net *net = sock_net((struct sock *)tp); 1502 1503 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1504 } 1505 1506 static inline int keepalive_time_when(const struct tcp_sock *tp) 1507 { 1508 struct net *net = sock_net((struct sock *)tp); 1509 1510 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1511 } 1512 1513 static inline int keepalive_probes(const struct tcp_sock *tp) 1514 { 1515 struct net *net = sock_net((struct sock *)tp); 1516 1517 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1518 } 1519 1520 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1521 { 1522 const struct inet_connection_sock *icsk = &tp->inet_conn; 1523 1524 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1525 tcp_jiffies32 - tp->rcv_tstamp); 1526 } 1527 1528 static inline int tcp_fin_time(const struct sock *sk) 1529 { 1530 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1531 const int rto = inet_csk(sk)->icsk_rto; 1532 1533 if (fin_timeout < (rto << 2) - (rto >> 1)) 1534 fin_timeout = (rto << 2) - (rto >> 1); 1535 1536 return fin_timeout; 1537 } 1538 1539 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1540 int paws_win) 1541 { 1542 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1543 return true; 1544 if (unlikely(!time_before32(ktime_get_seconds(), 1545 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1546 return true; 1547 /* 1548 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1549 * then following tcp messages have valid values. Ignore 0 value, 1550 * or else 'negative' tsval might forbid us to accept their packets. 1551 */ 1552 if (!rx_opt->ts_recent) 1553 return true; 1554 return false; 1555 } 1556 1557 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1558 int rst) 1559 { 1560 if (tcp_paws_check(rx_opt, 0)) 1561 return false; 1562 1563 /* RST segments are not recommended to carry timestamp, 1564 and, if they do, it is recommended to ignore PAWS because 1565 "their cleanup function should take precedence over timestamps." 1566 Certainly, it is mistake. It is necessary to understand the reasons 1567 of this constraint to relax it: if peer reboots, clock may go 1568 out-of-sync and half-open connections will not be reset. 1569 Actually, the problem would be not existing if all 1570 the implementations followed draft about maintaining clock 1571 via reboots. Linux-2.2 DOES NOT! 1572 1573 However, we can relax time bounds for RST segments to MSL. 1574 */ 1575 if (rst && !time_before32(ktime_get_seconds(), 1576 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1577 return false; 1578 return true; 1579 } 1580 1581 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1582 int mib_idx, u32 *last_oow_ack_time); 1583 1584 static inline void tcp_mib_init(struct net *net) 1585 { 1586 /* See RFC 2012 */ 1587 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1588 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1589 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1590 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1591 } 1592 1593 /* from STCP */ 1594 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1595 { 1596 tp->lost_skb_hint = NULL; 1597 } 1598 1599 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1600 { 1601 tcp_clear_retrans_hints_partial(tp); 1602 tp->retransmit_skb_hint = NULL; 1603 } 1604 1605 union tcp_md5_addr { 1606 struct in_addr a4; 1607 #if IS_ENABLED(CONFIG_IPV6) 1608 struct in6_addr a6; 1609 #endif 1610 }; 1611 1612 /* - key database */ 1613 struct tcp_md5sig_key { 1614 struct hlist_node node; 1615 u8 keylen; 1616 u8 family; /* AF_INET or AF_INET6 */ 1617 u8 prefixlen; 1618 u8 flags; 1619 union tcp_md5_addr addr; 1620 int l3index; /* set if key added with L3 scope */ 1621 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1622 struct rcu_head rcu; 1623 }; 1624 1625 /* - sock block */ 1626 struct tcp_md5sig_info { 1627 struct hlist_head head; 1628 struct rcu_head rcu; 1629 }; 1630 1631 /* - pseudo header */ 1632 struct tcp4_pseudohdr { 1633 __be32 saddr; 1634 __be32 daddr; 1635 __u8 pad; 1636 __u8 protocol; 1637 __be16 len; 1638 }; 1639 1640 struct tcp6_pseudohdr { 1641 struct in6_addr saddr; 1642 struct in6_addr daddr; 1643 __be32 len; 1644 __be32 protocol; /* including padding */ 1645 }; 1646 1647 union tcp_md5sum_block { 1648 struct tcp4_pseudohdr ip4; 1649 #if IS_ENABLED(CONFIG_IPV6) 1650 struct tcp6_pseudohdr ip6; 1651 #endif 1652 }; 1653 1654 /* - pool: digest algorithm, hash description and scratch buffer */ 1655 struct tcp_md5sig_pool { 1656 struct ahash_request *md5_req; 1657 void *scratch; 1658 }; 1659 1660 /* - functions */ 1661 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1662 const struct sock *sk, const struct sk_buff *skb); 1663 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1664 int family, u8 prefixlen, int l3index, u8 flags, 1665 const u8 *newkey, u8 newkeylen, gfp_t gfp); 1666 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1667 int family, u8 prefixlen, int l3index, u8 flags); 1668 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1669 const struct sock *addr_sk); 1670 1671 #ifdef CONFIG_TCP_MD5SIG 1672 #include <linux/jump_label.h> 1673 extern struct static_key_false tcp_md5_needed; 1674 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1675 const union tcp_md5_addr *addr, 1676 int family); 1677 static inline struct tcp_md5sig_key * 1678 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1679 const union tcp_md5_addr *addr, int family) 1680 { 1681 if (!static_branch_unlikely(&tcp_md5_needed)) 1682 return NULL; 1683 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1684 } 1685 1686 enum skb_drop_reason 1687 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1688 const void *saddr, const void *daddr, 1689 int family, int dif, int sdif); 1690 1691 1692 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1693 #else 1694 static inline struct tcp_md5sig_key * 1695 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1696 const union tcp_md5_addr *addr, int family) 1697 { 1698 return NULL; 1699 } 1700 1701 static inline enum skb_drop_reason 1702 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1703 const void *saddr, const void *daddr, 1704 int family, int dif, int sdif) 1705 { 1706 return SKB_NOT_DROPPED_YET; 1707 } 1708 #define tcp_twsk_md5_key(twsk) NULL 1709 #endif 1710 1711 bool tcp_alloc_md5sig_pool(void); 1712 1713 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1714 static inline void tcp_put_md5sig_pool(void) 1715 { 1716 local_bh_enable(); 1717 } 1718 1719 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1720 unsigned int header_len); 1721 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1722 const struct tcp_md5sig_key *key); 1723 1724 /* From tcp_fastopen.c */ 1725 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1726 struct tcp_fastopen_cookie *cookie); 1727 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1728 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1729 u16 try_exp); 1730 struct tcp_fastopen_request { 1731 /* Fast Open cookie. Size 0 means a cookie request */ 1732 struct tcp_fastopen_cookie cookie; 1733 struct msghdr *data; /* data in MSG_FASTOPEN */ 1734 size_t size; 1735 int copied; /* queued in tcp_connect() */ 1736 struct ubuf_info *uarg; 1737 }; 1738 void tcp_free_fastopen_req(struct tcp_sock *tp); 1739 void tcp_fastopen_destroy_cipher(struct sock *sk); 1740 void tcp_fastopen_ctx_destroy(struct net *net); 1741 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1742 void *primary_key, void *backup_key); 1743 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1744 u64 *key); 1745 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1746 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1747 struct request_sock *req, 1748 struct tcp_fastopen_cookie *foc, 1749 const struct dst_entry *dst); 1750 void tcp_fastopen_init_key_once(struct net *net); 1751 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1752 struct tcp_fastopen_cookie *cookie); 1753 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1754 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1755 #define TCP_FASTOPEN_KEY_MAX 2 1756 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1757 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1758 1759 /* Fastopen key context */ 1760 struct tcp_fastopen_context { 1761 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1762 int num; 1763 struct rcu_head rcu; 1764 }; 1765 1766 void tcp_fastopen_active_disable(struct sock *sk); 1767 bool tcp_fastopen_active_should_disable(struct sock *sk); 1768 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1769 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1770 1771 /* Caller needs to wrap with rcu_read_(un)lock() */ 1772 static inline 1773 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1774 { 1775 struct tcp_fastopen_context *ctx; 1776 1777 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1778 if (!ctx) 1779 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1780 return ctx; 1781 } 1782 1783 static inline 1784 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1785 const struct tcp_fastopen_cookie *orig) 1786 { 1787 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1788 orig->len == foc->len && 1789 !memcmp(orig->val, foc->val, foc->len)) 1790 return true; 1791 return false; 1792 } 1793 1794 static inline 1795 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1796 { 1797 return ctx->num; 1798 } 1799 1800 /* Latencies incurred by various limits for a sender. They are 1801 * chronograph-like stats that are mutually exclusive. 1802 */ 1803 enum tcp_chrono { 1804 TCP_CHRONO_UNSPEC, 1805 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1806 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1807 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1808 __TCP_CHRONO_MAX, 1809 }; 1810 1811 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1812 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1813 1814 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1815 * the same memory storage than skb->destructor/_skb_refdst 1816 */ 1817 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1818 { 1819 skb->destructor = NULL; 1820 skb->_skb_refdst = 0UL; 1821 } 1822 1823 #define tcp_skb_tsorted_save(skb) { \ 1824 unsigned long _save = skb->_skb_refdst; \ 1825 skb->_skb_refdst = 0UL; 1826 1827 #define tcp_skb_tsorted_restore(skb) \ 1828 skb->_skb_refdst = _save; \ 1829 } 1830 1831 void tcp_write_queue_purge(struct sock *sk); 1832 1833 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1834 { 1835 return skb_rb_first(&sk->tcp_rtx_queue); 1836 } 1837 1838 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1839 { 1840 return skb_rb_last(&sk->tcp_rtx_queue); 1841 } 1842 1843 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1844 { 1845 return skb_peek_tail(&sk->sk_write_queue); 1846 } 1847 1848 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1849 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1850 1851 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1852 { 1853 return skb_peek(&sk->sk_write_queue); 1854 } 1855 1856 static inline bool tcp_skb_is_last(const struct sock *sk, 1857 const struct sk_buff *skb) 1858 { 1859 return skb_queue_is_last(&sk->sk_write_queue, skb); 1860 } 1861 1862 /** 1863 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1864 * @sk: socket 1865 * 1866 * Since the write queue can have a temporary empty skb in it, 1867 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1868 */ 1869 static inline bool tcp_write_queue_empty(const struct sock *sk) 1870 { 1871 const struct tcp_sock *tp = tcp_sk(sk); 1872 1873 return tp->write_seq == tp->snd_nxt; 1874 } 1875 1876 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1877 { 1878 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1879 } 1880 1881 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1882 { 1883 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1884 } 1885 1886 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1887 { 1888 __skb_queue_tail(&sk->sk_write_queue, skb); 1889 1890 /* Queue it, remembering where we must start sending. */ 1891 if (sk->sk_write_queue.next == skb) 1892 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1893 } 1894 1895 /* Insert new before skb on the write queue of sk. */ 1896 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1897 struct sk_buff *skb, 1898 struct sock *sk) 1899 { 1900 __skb_queue_before(&sk->sk_write_queue, skb, new); 1901 } 1902 1903 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1904 { 1905 tcp_skb_tsorted_anchor_cleanup(skb); 1906 __skb_unlink(skb, &sk->sk_write_queue); 1907 } 1908 1909 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1910 1911 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1912 { 1913 tcp_skb_tsorted_anchor_cleanup(skb); 1914 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1915 } 1916 1917 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1918 { 1919 list_del(&skb->tcp_tsorted_anchor); 1920 tcp_rtx_queue_unlink(skb, sk); 1921 tcp_wmem_free_skb(sk, skb); 1922 } 1923 1924 static inline void tcp_push_pending_frames(struct sock *sk) 1925 { 1926 if (tcp_send_head(sk)) { 1927 struct tcp_sock *tp = tcp_sk(sk); 1928 1929 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1930 } 1931 } 1932 1933 /* Start sequence of the skb just after the highest skb with SACKed 1934 * bit, valid only if sacked_out > 0 or when the caller has ensured 1935 * validity by itself. 1936 */ 1937 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1938 { 1939 if (!tp->sacked_out) 1940 return tp->snd_una; 1941 1942 if (tp->highest_sack == NULL) 1943 return tp->snd_nxt; 1944 1945 return TCP_SKB_CB(tp->highest_sack)->seq; 1946 } 1947 1948 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1949 { 1950 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1951 } 1952 1953 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1954 { 1955 return tcp_sk(sk)->highest_sack; 1956 } 1957 1958 static inline void tcp_highest_sack_reset(struct sock *sk) 1959 { 1960 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1961 } 1962 1963 /* Called when old skb is about to be deleted and replaced by new skb */ 1964 static inline void tcp_highest_sack_replace(struct sock *sk, 1965 struct sk_buff *old, 1966 struct sk_buff *new) 1967 { 1968 if (old == tcp_highest_sack(sk)) 1969 tcp_sk(sk)->highest_sack = new; 1970 } 1971 1972 /* This helper checks if socket has IP_TRANSPARENT set */ 1973 static inline bool inet_sk_transparent(const struct sock *sk) 1974 { 1975 switch (sk->sk_state) { 1976 case TCP_TIME_WAIT: 1977 return inet_twsk(sk)->tw_transparent; 1978 case TCP_NEW_SYN_RECV: 1979 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1980 } 1981 return inet_sk(sk)->transparent; 1982 } 1983 1984 /* Determines whether this is a thin stream (which may suffer from 1985 * increased latency). Used to trigger latency-reducing mechanisms. 1986 */ 1987 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1988 { 1989 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1990 } 1991 1992 /* /proc */ 1993 enum tcp_seq_states { 1994 TCP_SEQ_STATE_LISTENING, 1995 TCP_SEQ_STATE_ESTABLISHED, 1996 }; 1997 1998 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 1999 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2000 void tcp_seq_stop(struct seq_file *seq, void *v); 2001 2002 struct tcp_seq_afinfo { 2003 sa_family_t family; 2004 }; 2005 2006 struct tcp_iter_state { 2007 struct seq_net_private p; 2008 enum tcp_seq_states state; 2009 struct sock *syn_wait_sk; 2010 int bucket, offset, sbucket, num; 2011 loff_t last_pos; 2012 }; 2013 2014 extern struct request_sock_ops tcp_request_sock_ops; 2015 extern struct request_sock_ops tcp6_request_sock_ops; 2016 2017 void tcp_v4_destroy_sock(struct sock *sk); 2018 2019 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2020 netdev_features_t features); 2021 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2022 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2023 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2024 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2025 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2026 int tcp_gro_complete(struct sk_buff *skb); 2027 2028 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2029 2030 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2031 { 2032 struct net *net = sock_net((struct sock *)tp); 2033 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 2034 } 2035 2036 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2037 2038 #ifdef CONFIG_PROC_FS 2039 int tcp4_proc_init(void); 2040 void tcp4_proc_exit(void); 2041 #endif 2042 2043 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2044 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2045 const struct tcp_request_sock_ops *af_ops, 2046 struct sock *sk, struct sk_buff *skb); 2047 2048 /* TCP af-specific functions */ 2049 struct tcp_sock_af_ops { 2050 #ifdef CONFIG_TCP_MD5SIG 2051 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2052 const struct sock *addr_sk); 2053 int (*calc_md5_hash)(char *location, 2054 const struct tcp_md5sig_key *md5, 2055 const struct sock *sk, 2056 const struct sk_buff *skb); 2057 int (*md5_parse)(struct sock *sk, 2058 int optname, 2059 sockptr_t optval, 2060 int optlen); 2061 #endif 2062 }; 2063 2064 struct tcp_request_sock_ops { 2065 u16 mss_clamp; 2066 #ifdef CONFIG_TCP_MD5SIG 2067 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2068 const struct sock *addr_sk); 2069 int (*calc_md5_hash) (char *location, 2070 const struct tcp_md5sig_key *md5, 2071 const struct sock *sk, 2072 const struct sk_buff *skb); 2073 #endif 2074 #ifdef CONFIG_SYN_COOKIES 2075 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2076 __u16 *mss); 2077 #endif 2078 struct dst_entry *(*route_req)(const struct sock *sk, 2079 struct sk_buff *skb, 2080 struct flowi *fl, 2081 struct request_sock *req); 2082 u32 (*init_seq)(const struct sk_buff *skb); 2083 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2084 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2085 struct flowi *fl, struct request_sock *req, 2086 struct tcp_fastopen_cookie *foc, 2087 enum tcp_synack_type synack_type, 2088 struct sk_buff *syn_skb); 2089 }; 2090 2091 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2092 #if IS_ENABLED(CONFIG_IPV6) 2093 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2094 #endif 2095 2096 #ifdef CONFIG_SYN_COOKIES 2097 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2098 const struct sock *sk, struct sk_buff *skb, 2099 __u16 *mss) 2100 { 2101 tcp_synq_overflow(sk); 2102 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2103 return ops->cookie_init_seq(skb, mss); 2104 } 2105 #else 2106 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2107 const struct sock *sk, struct sk_buff *skb, 2108 __u16 *mss) 2109 { 2110 return 0; 2111 } 2112 #endif 2113 2114 int tcpv4_offload_init(void); 2115 2116 void tcp_v4_init(void); 2117 void tcp_init(void); 2118 2119 /* tcp_recovery.c */ 2120 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2121 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2122 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2123 u32 reo_wnd); 2124 extern bool tcp_rack_mark_lost(struct sock *sk); 2125 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2126 u64 xmit_time); 2127 extern void tcp_rack_reo_timeout(struct sock *sk); 2128 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2129 2130 /* At how many usecs into the future should the RTO fire? */ 2131 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2132 { 2133 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2134 u32 rto = inet_csk(sk)->icsk_rto; 2135 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2136 2137 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2138 } 2139 2140 /* 2141 * Save and compile IPv4 options, return a pointer to it 2142 */ 2143 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2144 struct sk_buff *skb) 2145 { 2146 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2147 struct ip_options_rcu *dopt = NULL; 2148 2149 if (opt->optlen) { 2150 int opt_size = sizeof(*dopt) + opt->optlen; 2151 2152 dopt = kmalloc(opt_size, GFP_ATOMIC); 2153 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2154 kfree(dopt); 2155 dopt = NULL; 2156 } 2157 } 2158 return dopt; 2159 } 2160 2161 /* locally generated TCP pure ACKs have skb->truesize == 2 2162 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2163 * This is much faster than dissecting the packet to find out. 2164 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2165 */ 2166 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2167 { 2168 return skb->truesize == 2; 2169 } 2170 2171 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2172 { 2173 skb->truesize = 2; 2174 } 2175 2176 static inline int tcp_inq(struct sock *sk) 2177 { 2178 struct tcp_sock *tp = tcp_sk(sk); 2179 int answ; 2180 2181 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2182 answ = 0; 2183 } else if (sock_flag(sk, SOCK_URGINLINE) || 2184 !tp->urg_data || 2185 before(tp->urg_seq, tp->copied_seq) || 2186 !before(tp->urg_seq, tp->rcv_nxt)) { 2187 2188 answ = tp->rcv_nxt - tp->copied_seq; 2189 2190 /* Subtract 1, if FIN was received */ 2191 if (answ && sock_flag(sk, SOCK_DONE)) 2192 answ--; 2193 } else { 2194 answ = tp->urg_seq - tp->copied_seq; 2195 } 2196 2197 return answ; 2198 } 2199 2200 int tcp_peek_len(struct socket *sock); 2201 2202 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2203 { 2204 u16 segs_in; 2205 2206 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2207 2208 /* We update these fields while other threads might 2209 * read them from tcp_get_info() 2210 */ 2211 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2212 if (skb->len > tcp_hdrlen(skb)) 2213 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2214 } 2215 2216 /* 2217 * TCP listen path runs lockless. 2218 * We forced "struct sock" to be const qualified to make sure 2219 * we don't modify one of its field by mistake. 2220 * Here, we increment sk_drops which is an atomic_t, so we can safely 2221 * make sock writable again. 2222 */ 2223 static inline void tcp_listendrop(const struct sock *sk) 2224 { 2225 atomic_inc(&((struct sock *)sk)->sk_drops); 2226 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2227 } 2228 2229 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2230 2231 /* 2232 * Interface for adding Upper Level Protocols over TCP 2233 */ 2234 2235 #define TCP_ULP_NAME_MAX 16 2236 #define TCP_ULP_MAX 128 2237 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2238 2239 struct tcp_ulp_ops { 2240 struct list_head list; 2241 2242 /* initialize ulp */ 2243 int (*init)(struct sock *sk); 2244 /* update ulp */ 2245 void (*update)(struct sock *sk, struct proto *p, 2246 void (*write_space)(struct sock *sk)); 2247 /* cleanup ulp */ 2248 void (*release)(struct sock *sk); 2249 /* diagnostic */ 2250 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2251 size_t (*get_info_size)(const struct sock *sk); 2252 /* clone ulp */ 2253 void (*clone)(const struct request_sock *req, struct sock *newsk, 2254 const gfp_t priority); 2255 2256 char name[TCP_ULP_NAME_MAX]; 2257 struct module *owner; 2258 }; 2259 int tcp_register_ulp(struct tcp_ulp_ops *type); 2260 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2261 int tcp_set_ulp(struct sock *sk, const char *name); 2262 void tcp_get_available_ulp(char *buf, size_t len); 2263 void tcp_cleanup_ulp(struct sock *sk); 2264 void tcp_update_ulp(struct sock *sk, struct proto *p, 2265 void (*write_space)(struct sock *sk)); 2266 2267 #define MODULE_ALIAS_TCP_ULP(name) \ 2268 __MODULE_INFO(alias, alias_userspace, name); \ 2269 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2270 2271 #ifdef CONFIG_NET_SOCK_MSG 2272 struct sk_msg; 2273 struct sk_psock; 2274 2275 #ifdef CONFIG_BPF_SYSCALL 2276 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2277 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2278 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2279 #endif /* CONFIG_BPF_SYSCALL */ 2280 2281 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, 2282 int flags); 2283 #endif /* CONFIG_NET_SOCK_MSG */ 2284 2285 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2286 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2287 { 2288 } 2289 #endif 2290 2291 #ifdef CONFIG_CGROUP_BPF 2292 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2293 struct sk_buff *skb, 2294 unsigned int end_offset) 2295 { 2296 skops->skb = skb; 2297 skops->skb_data_end = skb->data + end_offset; 2298 } 2299 #else 2300 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2301 struct sk_buff *skb, 2302 unsigned int end_offset) 2303 { 2304 } 2305 #endif 2306 2307 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2308 * is < 0, then the BPF op failed (for example if the loaded BPF 2309 * program does not support the chosen operation or there is no BPF 2310 * program loaded). 2311 */ 2312 #ifdef CONFIG_BPF 2313 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2314 { 2315 struct bpf_sock_ops_kern sock_ops; 2316 int ret; 2317 2318 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2319 if (sk_fullsock(sk)) { 2320 sock_ops.is_fullsock = 1; 2321 sock_owned_by_me(sk); 2322 } 2323 2324 sock_ops.sk = sk; 2325 sock_ops.op = op; 2326 if (nargs > 0) 2327 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2328 2329 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2330 if (ret == 0) 2331 ret = sock_ops.reply; 2332 else 2333 ret = -1; 2334 return ret; 2335 } 2336 2337 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2338 { 2339 u32 args[2] = {arg1, arg2}; 2340 2341 return tcp_call_bpf(sk, op, 2, args); 2342 } 2343 2344 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2345 u32 arg3) 2346 { 2347 u32 args[3] = {arg1, arg2, arg3}; 2348 2349 return tcp_call_bpf(sk, op, 3, args); 2350 } 2351 2352 #else 2353 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2354 { 2355 return -EPERM; 2356 } 2357 2358 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2359 { 2360 return -EPERM; 2361 } 2362 2363 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2364 u32 arg3) 2365 { 2366 return -EPERM; 2367 } 2368 2369 #endif 2370 2371 static inline u32 tcp_timeout_init(struct sock *sk) 2372 { 2373 int timeout; 2374 2375 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2376 2377 if (timeout <= 0) 2378 timeout = TCP_TIMEOUT_INIT; 2379 return min_t(int, timeout, TCP_RTO_MAX); 2380 } 2381 2382 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2383 { 2384 int rwnd; 2385 2386 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2387 2388 if (rwnd < 0) 2389 rwnd = 0; 2390 return rwnd; 2391 } 2392 2393 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2394 { 2395 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2396 } 2397 2398 static inline void tcp_bpf_rtt(struct sock *sk) 2399 { 2400 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2401 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2402 } 2403 2404 #if IS_ENABLED(CONFIG_SMC) 2405 extern struct static_key_false tcp_have_smc; 2406 #endif 2407 2408 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2409 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2410 void (*cad)(struct sock *sk, u32 ack_seq)); 2411 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2412 void clean_acked_data_flush(void); 2413 #endif 2414 2415 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2416 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2417 const struct tcp_sock *tp) 2418 { 2419 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2420 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2421 } 2422 2423 /* Compute Earliest Departure Time for some control packets 2424 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2425 */ 2426 static inline u64 tcp_transmit_time(const struct sock *sk) 2427 { 2428 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2429 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2430 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2431 2432 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2433 } 2434 return 0; 2435 } 2436 2437 #endif /* _TCP_H */ 2438