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