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_PAWS_24DAYS (60 * 60 * 24 * 24) 165 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 166 * after this time. It should be equal 167 * (or greater than) TCP_TIMEWAIT_LEN 168 * to provide reliability equal to one 169 * provided by timewait state. 170 */ 171 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 172 * timestamps. It must be less than 173 * minimal timewait lifetime. 174 */ 175 /* 176 * TCP option 177 */ 178 179 #define TCPOPT_NOP 1 /* Padding */ 180 #define TCPOPT_EOL 0 /* End of options */ 181 #define TCPOPT_MSS 2 /* Segment size negotiating */ 182 #define TCPOPT_WINDOW 3 /* Window scaling */ 183 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 184 #define TCPOPT_SACK 5 /* SACK Block */ 185 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 186 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 187 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 189 #define TCPOPT_EXP 254 /* Experimental */ 190 /* Magic number to be after the option value for sharing TCP 191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 192 */ 193 #define TCPOPT_FASTOPEN_MAGIC 0xF989 194 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 195 196 /* 197 * TCP option lengths 198 */ 199 200 #define TCPOLEN_MSS 4 201 #define TCPOLEN_WINDOW 3 202 #define TCPOLEN_SACK_PERM 2 203 #define TCPOLEN_TIMESTAMP 10 204 #define TCPOLEN_MD5SIG 18 205 #define TCPOLEN_FASTOPEN_BASE 2 206 #define TCPOLEN_EXP_FASTOPEN_BASE 4 207 #define TCPOLEN_EXP_SMC_BASE 6 208 209 /* But this is what stacks really send out. */ 210 #define TCPOLEN_TSTAMP_ALIGNED 12 211 #define TCPOLEN_WSCALE_ALIGNED 4 212 #define TCPOLEN_SACKPERM_ALIGNED 4 213 #define TCPOLEN_SACK_BASE 2 214 #define TCPOLEN_SACK_BASE_ALIGNED 4 215 #define TCPOLEN_SACK_PERBLOCK 8 216 #define TCPOLEN_MD5SIG_ALIGNED 20 217 #define TCPOLEN_MSS_ALIGNED 4 218 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 219 220 /* Flags in tp->nonagle */ 221 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 222 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 223 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 224 225 /* TCP thin-stream limits */ 226 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 227 228 /* TCP initial congestion window as per rfc6928 */ 229 #define TCP_INIT_CWND 10 230 231 /* Bit Flags for sysctl_tcp_fastopen */ 232 #define TFO_CLIENT_ENABLE 1 233 #define TFO_SERVER_ENABLE 2 234 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 235 236 /* Accept SYN data w/o any cookie option */ 237 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 238 239 /* Force enable TFO on all listeners, i.e., not requiring the 240 * TCP_FASTOPEN socket option. 241 */ 242 #define TFO_SERVER_WO_SOCKOPT1 0x400 243 244 245 /* sysctl variables for tcp */ 246 extern int sysctl_tcp_max_orphans; 247 extern long sysctl_tcp_mem[3]; 248 249 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 250 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 251 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 252 253 extern atomic_long_t tcp_memory_allocated; 254 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 255 256 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_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 329 size_t size, struct ubuf_info *uarg); 330 void tcp_splice_eof(struct socket *sock); 331 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 332 int tcp_wmem_schedule(struct sock *sk, int copy); 333 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 334 int size_goal); 335 void tcp_release_cb(struct sock *sk); 336 void tcp_wfree(struct sk_buff *skb); 337 void tcp_write_timer_handler(struct sock *sk); 338 void tcp_delack_timer_handler(struct sock *sk); 339 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 340 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 341 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 342 void tcp_rcv_space_adjust(struct sock *sk); 343 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 344 void tcp_twsk_destructor(struct sock *sk); 345 void tcp_twsk_purge(struct list_head *net_exit_list, int family); 346 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 347 struct pipe_inode_info *pipe, size_t len, 348 unsigned int flags); 349 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 350 bool force_schedule); 351 352 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); 353 static inline void tcp_dec_quickack_mode(struct sock *sk, 354 const unsigned int pkts) 355 { 356 struct inet_connection_sock *icsk = inet_csk(sk); 357 358 if (icsk->icsk_ack.quick) { 359 if (pkts >= icsk->icsk_ack.quick) { 360 icsk->icsk_ack.quick = 0; 361 /* Leaving quickack mode we deflate ATO. */ 362 icsk->icsk_ack.ato = TCP_ATO_MIN; 363 } else 364 icsk->icsk_ack.quick -= pkts; 365 } 366 } 367 368 #define TCP_ECN_OK 1 369 #define TCP_ECN_QUEUE_CWR 2 370 #define TCP_ECN_DEMAND_CWR 4 371 #define TCP_ECN_SEEN 8 372 373 enum tcp_tw_status { 374 TCP_TW_SUCCESS = 0, 375 TCP_TW_RST = 1, 376 TCP_TW_ACK = 2, 377 TCP_TW_SYN = 3 378 }; 379 380 381 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 382 struct sk_buff *skb, 383 const struct tcphdr *th); 384 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 385 struct request_sock *req, bool fastopen, 386 bool *lost_race); 387 int tcp_child_process(struct sock *parent, struct sock *child, 388 struct sk_buff *skb); 389 void tcp_enter_loss(struct sock *sk); 390 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 391 void tcp_clear_retrans(struct tcp_sock *tp); 392 void tcp_update_metrics(struct sock *sk); 393 void tcp_init_metrics(struct sock *sk); 394 void tcp_metrics_init(void); 395 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 396 void __tcp_close(struct sock *sk, long timeout); 397 void tcp_close(struct sock *sk, long timeout); 398 void tcp_init_sock(struct sock *sk); 399 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 400 __poll_t tcp_poll(struct file *file, struct socket *sock, 401 struct poll_table_struct *wait); 402 int do_tcp_getsockopt(struct sock *sk, int level, 403 int optname, sockptr_t optval, sockptr_t optlen); 404 int tcp_getsockopt(struct sock *sk, int level, int optname, 405 char __user *optval, int __user *optlen); 406 bool tcp_bpf_bypass_getsockopt(int level, int optname); 407 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 408 sockptr_t optval, unsigned int optlen); 409 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 410 unsigned int optlen); 411 void tcp_set_keepalive(struct sock *sk, int val); 412 void tcp_syn_ack_timeout(const struct request_sock *req); 413 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 414 int flags, int *addr_len); 415 int tcp_set_rcvlowat(struct sock *sk, int val); 416 int tcp_set_window_clamp(struct sock *sk, int val); 417 void tcp_update_recv_tstamps(struct sk_buff *skb, 418 struct scm_timestamping_internal *tss); 419 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 420 struct scm_timestamping_internal *tss); 421 void tcp_data_ready(struct sock *sk); 422 #ifdef CONFIG_MMU 423 int tcp_mmap(struct file *file, struct socket *sock, 424 struct vm_area_struct *vma); 425 #endif 426 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 427 struct tcp_options_received *opt_rx, 428 int estab, struct tcp_fastopen_cookie *foc); 429 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 430 431 /* 432 * BPF SKB-less helpers 433 */ 434 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 435 struct tcphdr *th, u32 *cookie); 436 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 437 struct tcphdr *th, u32 *cookie); 438 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 439 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 440 const struct tcp_request_sock_ops *af_ops, 441 struct sock *sk, struct tcphdr *th); 442 /* 443 * TCP v4 functions exported for the inet6 API 444 */ 445 446 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 447 void tcp_v4_mtu_reduced(struct sock *sk); 448 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 449 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 450 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 451 struct sock *tcp_create_openreq_child(const struct sock *sk, 452 struct request_sock *req, 453 struct sk_buff *skb); 454 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 455 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 456 struct request_sock *req, 457 struct dst_entry *dst, 458 struct request_sock *req_unhash, 459 bool *own_req); 460 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 461 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 462 int tcp_connect(struct sock *sk); 463 enum tcp_synack_type { 464 TCP_SYNACK_NORMAL, 465 TCP_SYNACK_FASTOPEN, 466 TCP_SYNACK_COOKIE, 467 }; 468 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 469 struct request_sock *req, 470 struct tcp_fastopen_cookie *foc, 471 enum tcp_synack_type synack_type, 472 struct sk_buff *syn_skb); 473 int tcp_disconnect(struct sock *sk, int flags); 474 475 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 476 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 477 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 478 479 /* From syncookies.c */ 480 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 481 struct request_sock *req, 482 struct dst_entry *dst, u32 tsoff); 483 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 484 u32 cookie); 485 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 486 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 487 const struct tcp_request_sock_ops *af_ops, 488 struct sock *sk, struct sk_buff *skb); 489 #ifdef CONFIG_SYN_COOKIES 490 491 /* Syncookies use a monotonic timer which increments every 60 seconds. 492 * This counter is used both as a hash input and partially encoded into 493 * the cookie value. A cookie is only validated further if the delta 494 * between the current counter value and the encoded one is less than this, 495 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 496 * the counter advances immediately after a cookie is generated). 497 */ 498 #define MAX_SYNCOOKIE_AGE 2 499 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 500 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 501 502 /* syncookies: remember time of last synqueue overflow 503 * But do not dirty this field too often (once per second is enough) 504 * It is racy as we do not hold a lock, but race is very minor. 505 */ 506 static inline void tcp_synq_overflow(const struct sock *sk) 507 { 508 unsigned int last_overflow; 509 unsigned int now = jiffies; 510 511 if (sk->sk_reuseport) { 512 struct sock_reuseport *reuse; 513 514 reuse = rcu_dereference(sk->sk_reuseport_cb); 515 if (likely(reuse)) { 516 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 517 if (!time_between32(now, last_overflow, 518 last_overflow + HZ)) 519 WRITE_ONCE(reuse->synq_overflow_ts, now); 520 return; 521 } 522 } 523 524 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 525 if (!time_between32(now, last_overflow, last_overflow + HZ)) 526 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 527 } 528 529 /* syncookies: no recent synqueue overflow on this listening socket? */ 530 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 531 { 532 unsigned int last_overflow; 533 unsigned int now = jiffies; 534 535 if (sk->sk_reuseport) { 536 struct sock_reuseport *reuse; 537 538 reuse = rcu_dereference(sk->sk_reuseport_cb); 539 if (likely(reuse)) { 540 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 541 return !time_between32(now, last_overflow - HZ, 542 last_overflow + 543 TCP_SYNCOOKIE_VALID); 544 } 545 } 546 547 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 548 549 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 550 * then we're under synflood. However, we have to use 551 * 'last_overflow - HZ' as lower bound. That's because a concurrent 552 * tcp_synq_overflow() could update .ts_recent_stamp after we read 553 * jiffies but before we store .ts_recent_stamp into last_overflow, 554 * which could lead to rejecting a valid syncookie. 555 */ 556 return !time_between32(now, last_overflow - HZ, 557 last_overflow + TCP_SYNCOOKIE_VALID); 558 } 559 560 static inline u32 tcp_cookie_time(void) 561 { 562 u64 val = get_jiffies_64(); 563 564 do_div(val, TCP_SYNCOOKIE_PERIOD); 565 return val; 566 } 567 568 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 569 u16 *mssp); 570 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 571 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 572 bool cookie_timestamp_decode(const struct net *net, 573 struct tcp_options_received *opt); 574 bool cookie_ecn_ok(const struct tcp_options_received *opt, 575 const struct net *net, const struct dst_entry *dst); 576 577 /* From net/ipv6/syncookies.c */ 578 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 579 u32 cookie); 580 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 581 582 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 583 const struct tcphdr *th, u16 *mssp); 584 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 585 #endif 586 /* tcp_output.c */ 587 588 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 589 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 590 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 591 int nonagle); 592 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 593 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 594 void tcp_retransmit_timer(struct sock *sk); 595 void tcp_xmit_retransmit_queue(struct sock *); 596 void tcp_simple_retransmit(struct sock *); 597 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 598 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 599 enum tcp_queue { 600 TCP_FRAG_IN_WRITE_QUEUE, 601 TCP_FRAG_IN_RTX_QUEUE, 602 }; 603 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 604 struct sk_buff *skb, u32 len, 605 unsigned int mss_now, gfp_t gfp); 606 607 void tcp_send_probe0(struct sock *); 608 void tcp_send_partial(struct sock *); 609 int tcp_write_wakeup(struct sock *, int mib); 610 void tcp_send_fin(struct sock *sk); 611 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 612 int tcp_send_synack(struct sock *); 613 void tcp_push_one(struct sock *, unsigned int mss_now); 614 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 615 void tcp_send_ack(struct sock *sk); 616 void tcp_send_delayed_ack(struct sock *sk); 617 void tcp_send_loss_probe(struct sock *sk); 618 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 619 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 620 const struct sk_buff *next_skb); 621 622 /* tcp_input.c */ 623 void tcp_rearm_rto(struct sock *sk); 624 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 625 void tcp_reset(struct sock *sk, struct sk_buff *skb); 626 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 627 void tcp_fin(struct sock *sk); 628 void tcp_check_space(struct sock *sk); 629 void tcp_sack_compress_send_ack(struct sock *sk); 630 631 /* tcp_timer.c */ 632 void tcp_init_xmit_timers(struct sock *); 633 static inline void tcp_clear_xmit_timers(struct sock *sk) 634 { 635 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 636 __sock_put(sk); 637 638 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 639 __sock_put(sk); 640 641 inet_csk_clear_xmit_timers(sk); 642 } 643 644 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 645 unsigned int tcp_current_mss(struct sock *sk); 646 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 647 648 /* Bound MSS / TSO packet size with the half of the window */ 649 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 650 { 651 int cutoff; 652 653 /* When peer uses tiny windows, there is no use in packetizing 654 * to sub-MSS pieces for the sake of SWS or making sure there 655 * are enough packets in the pipe for fast recovery. 656 * 657 * On the other hand, for extremely large MSS devices, handling 658 * smaller than MSS windows in this way does make sense. 659 */ 660 if (tp->max_window > TCP_MSS_DEFAULT) 661 cutoff = (tp->max_window >> 1); 662 else 663 cutoff = tp->max_window; 664 665 if (cutoff && pktsize > cutoff) 666 return max_t(int, cutoff, 68U - tp->tcp_header_len); 667 else 668 return pktsize; 669 } 670 671 /* tcp.c */ 672 void tcp_get_info(struct sock *, struct tcp_info *); 673 674 /* Read 'sendfile()'-style from a TCP socket */ 675 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 676 sk_read_actor_t recv_actor); 677 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 678 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 679 void tcp_read_done(struct sock *sk, size_t len); 680 681 void tcp_initialize_rcv_mss(struct sock *sk); 682 683 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 684 int tcp_mss_to_mtu(struct sock *sk, int mss); 685 void tcp_mtup_init(struct sock *sk); 686 687 static inline void tcp_bound_rto(const struct sock *sk) 688 { 689 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 690 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 691 } 692 693 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 694 { 695 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 696 } 697 698 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 699 { 700 /* mptcp hooks are only on the slow path */ 701 if (sk_is_mptcp((struct sock *)tp)) 702 return; 703 704 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 705 ntohl(TCP_FLAG_ACK) | 706 snd_wnd); 707 } 708 709 static inline void tcp_fast_path_on(struct tcp_sock *tp) 710 { 711 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 712 } 713 714 static inline void tcp_fast_path_check(struct sock *sk) 715 { 716 struct tcp_sock *tp = tcp_sk(sk); 717 718 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 719 tp->rcv_wnd && 720 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 721 !tp->urg_data) 722 tcp_fast_path_on(tp); 723 } 724 725 /* Compute the actual rto_min value */ 726 static inline u32 tcp_rto_min(struct sock *sk) 727 { 728 const struct dst_entry *dst = __sk_dst_get(sk); 729 u32 rto_min = inet_csk(sk)->icsk_rto_min; 730 731 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 732 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 733 return rto_min; 734 } 735 736 static inline u32 tcp_rto_min_us(struct sock *sk) 737 { 738 return jiffies_to_usecs(tcp_rto_min(sk)); 739 } 740 741 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 742 { 743 return dst_metric_locked(dst, RTAX_CC_ALGO); 744 } 745 746 /* Minimum RTT in usec. ~0 means not available. */ 747 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 748 { 749 return minmax_get(&tp->rtt_min); 750 } 751 752 /* Compute the actual receive window we are currently advertising. 753 * Rcv_nxt can be after the window if our peer push more data 754 * than the offered window. 755 */ 756 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 757 { 758 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 759 760 if (win < 0) 761 win = 0; 762 return (u32) win; 763 } 764 765 /* Choose a new window, without checks for shrinking, and without 766 * scaling applied to the result. The caller does these things 767 * if necessary. This is a "raw" window selection. 768 */ 769 u32 __tcp_select_window(struct sock *sk); 770 771 void tcp_send_window_probe(struct sock *sk); 772 773 /* TCP uses 32bit jiffies to save some space. 774 * Note that this is different from tcp_time_stamp, which 775 * historically has been the same until linux-4.13. 776 */ 777 #define tcp_jiffies32 ((u32)jiffies) 778 779 /* 780 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 781 * It is no longer tied to jiffies, but to 1 ms clock. 782 * Note: double check if you want to use tcp_jiffies32 instead of this. 783 */ 784 #define TCP_TS_HZ 1000 785 786 static inline u64 tcp_clock_ns(void) 787 { 788 return ktime_get_ns(); 789 } 790 791 static inline u64 tcp_clock_us(void) 792 { 793 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 794 } 795 796 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 797 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 798 { 799 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 800 } 801 802 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */ 803 static inline u32 tcp_ns_to_ts(u64 ns) 804 { 805 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ); 806 } 807 808 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 809 static inline u32 tcp_time_stamp_raw(void) 810 { 811 return tcp_ns_to_ts(tcp_clock_ns()); 812 } 813 814 void tcp_mstamp_refresh(struct tcp_sock *tp); 815 816 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 817 { 818 return max_t(s64, t1 - t0, 0); 819 } 820 821 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 822 { 823 return tcp_ns_to_ts(skb->skb_mstamp_ns); 824 } 825 826 /* provide the departure time in us unit */ 827 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 828 { 829 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 830 } 831 832 833 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 834 835 #define TCPHDR_FIN 0x01 836 #define TCPHDR_SYN 0x02 837 #define TCPHDR_RST 0x04 838 #define TCPHDR_PSH 0x08 839 #define TCPHDR_ACK 0x10 840 #define TCPHDR_URG 0x20 841 #define TCPHDR_ECE 0x40 842 #define TCPHDR_CWR 0x80 843 844 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 845 846 /* This is what the send packet queuing engine uses to pass 847 * TCP per-packet control information to the transmission code. 848 * We also store the host-order sequence numbers in here too. 849 * This is 44 bytes if IPV6 is enabled. 850 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 851 */ 852 struct tcp_skb_cb { 853 __u32 seq; /* Starting sequence number */ 854 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 855 union { 856 /* Note : tcp_tw_isn is used in input path only 857 * (isn chosen by tcp_timewait_state_process()) 858 * 859 * tcp_gso_segs/size are used in write queue only, 860 * cf tcp_skb_pcount()/tcp_skb_mss() 861 */ 862 __u32 tcp_tw_isn; 863 struct { 864 u16 tcp_gso_segs; 865 u16 tcp_gso_size; 866 }; 867 }; 868 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 869 870 __u8 sacked; /* State flags for SACK. */ 871 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 872 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 873 #define TCPCB_LOST 0x04 /* SKB is lost */ 874 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 875 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ 876 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 877 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 878 TCPCB_REPAIRED) 879 880 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 881 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 882 eor:1, /* Is skb MSG_EOR marked? */ 883 has_rxtstamp:1, /* SKB has a RX timestamp */ 884 unused:5; 885 __u32 ack_seq; /* Sequence number ACK'd */ 886 union { 887 struct { 888 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 889 /* There is space for up to 24 bytes */ 890 __u32 is_app_limited:1, /* cwnd not fully used? */ 891 delivered_ce:20, 892 unused:11; 893 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 894 __u32 delivered; 895 /* start of send pipeline phase */ 896 u64 first_tx_mstamp; 897 /* when we reached the "delivered" count */ 898 u64 delivered_mstamp; 899 } tx; /* only used for outgoing skbs */ 900 union { 901 struct inet_skb_parm h4; 902 #if IS_ENABLED(CONFIG_IPV6) 903 struct inet6_skb_parm h6; 904 #endif 905 } header; /* For incoming skbs */ 906 }; 907 }; 908 909 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 910 911 extern const struct inet_connection_sock_af_ops ipv4_specific; 912 913 #if IS_ENABLED(CONFIG_IPV6) 914 /* This is the variant of inet6_iif() that must be used by TCP, 915 * as TCP moves IP6CB into a different location in skb->cb[] 916 */ 917 static inline int tcp_v6_iif(const struct sk_buff *skb) 918 { 919 return TCP_SKB_CB(skb)->header.h6.iif; 920 } 921 922 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 923 { 924 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 925 926 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 927 } 928 929 /* TCP_SKB_CB reference means this can not be used from early demux */ 930 static inline int tcp_v6_sdif(const struct sk_buff *skb) 931 { 932 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 933 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 934 return TCP_SKB_CB(skb)->header.h6.iif; 935 #endif 936 return 0; 937 } 938 939 extern const struct inet_connection_sock_af_ops ipv6_specific; 940 941 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 942 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 943 void tcp_v6_early_demux(struct sk_buff *skb); 944 945 #endif 946 947 /* TCP_SKB_CB reference means this can not be used from early demux */ 948 static inline int tcp_v4_sdif(struct sk_buff *skb) 949 { 950 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 951 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 952 return TCP_SKB_CB(skb)->header.h4.iif; 953 #endif 954 return 0; 955 } 956 957 /* Due to TSO, an SKB can be composed of multiple actual 958 * packets. To keep these tracked properly, we use this. 959 */ 960 static inline int tcp_skb_pcount(const struct sk_buff *skb) 961 { 962 return TCP_SKB_CB(skb)->tcp_gso_segs; 963 } 964 965 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 966 { 967 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 968 } 969 970 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 971 { 972 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 973 } 974 975 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 976 static inline int tcp_skb_mss(const struct sk_buff *skb) 977 { 978 return TCP_SKB_CB(skb)->tcp_gso_size; 979 } 980 981 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 982 { 983 return likely(!TCP_SKB_CB(skb)->eor); 984 } 985 986 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 987 const struct sk_buff *from) 988 { 989 return likely(tcp_skb_can_collapse_to(to) && 990 mptcp_skb_can_collapse(to, from) && 991 skb_pure_zcopy_same(to, from)); 992 } 993 994 /* Events passed to congestion control interface */ 995 enum tcp_ca_event { 996 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 997 CA_EVENT_CWND_RESTART, /* congestion window restart */ 998 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 999 CA_EVENT_LOSS, /* loss timeout */ 1000 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1001 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1002 }; 1003 1004 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1005 enum tcp_ca_ack_event_flags { 1006 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1007 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1008 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1009 }; 1010 1011 /* 1012 * Interface for adding new TCP congestion control handlers 1013 */ 1014 #define TCP_CA_NAME_MAX 16 1015 #define TCP_CA_MAX 128 1016 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1017 1018 #define TCP_CA_UNSPEC 0 1019 1020 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1021 #define TCP_CONG_NON_RESTRICTED 0x1 1022 /* Requires ECN/ECT set on all packets */ 1023 #define TCP_CONG_NEEDS_ECN 0x2 1024 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1025 1026 union tcp_cc_info; 1027 1028 struct ack_sample { 1029 u32 pkts_acked; 1030 s32 rtt_us; 1031 u32 in_flight; 1032 }; 1033 1034 /* A rate sample measures the number of (original/retransmitted) data 1035 * packets delivered "delivered" over an interval of time "interval_us". 1036 * The tcp_rate.c code fills in the rate sample, and congestion 1037 * control modules that define a cong_control function to run at the end 1038 * of ACK processing can optionally chose to consult this sample when 1039 * setting cwnd and pacing rate. 1040 * A sample is invalid if "delivered" or "interval_us" is negative. 1041 */ 1042 struct rate_sample { 1043 u64 prior_mstamp; /* starting timestamp for interval */ 1044 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1045 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1046 s32 delivered; /* number of packets delivered over interval */ 1047 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1048 long interval_us; /* time for tp->delivered to incr "delivered" */ 1049 u32 snd_interval_us; /* snd interval for delivered packets */ 1050 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1051 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1052 int losses; /* number of packets marked lost upon ACK */ 1053 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1054 u32 prior_in_flight; /* in flight before this ACK */ 1055 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1056 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1057 bool is_retrans; /* is sample from retransmission? */ 1058 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1059 }; 1060 1061 struct tcp_congestion_ops { 1062 /* fast path fields are put first to fill one cache line */ 1063 1064 /* return slow start threshold (required) */ 1065 u32 (*ssthresh)(struct sock *sk); 1066 1067 /* do new cwnd calculation (required) */ 1068 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1069 1070 /* call before changing ca_state (optional) */ 1071 void (*set_state)(struct sock *sk, u8 new_state); 1072 1073 /* call when cwnd event occurs (optional) */ 1074 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1075 1076 /* call when ack arrives (optional) */ 1077 void (*in_ack_event)(struct sock *sk, u32 flags); 1078 1079 /* hook for packet ack accounting (optional) */ 1080 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1081 1082 /* override sysctl_tcp_min_tso_segs */ 1083 u32 (*min_tso_segs)(struct sock *sk); 1084 1085 /* call when packets are delivered to update cwnd and pacing rate, 1086 * after all the ca_state processing. (optional) 1087 */ 1088 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1089 1090 1091 /* new value of cwnd after loss (required) */ 1092 u32 (*undo_cwnd)(struct sock *sk); 1093 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1094 u32 (*sndbuf_expand)(struct sock *sk); 1095 1096 /* control/slow paths put last */ 1097 /* get info for inet_diag (optional) */ 1098 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1099 union tcp_cc_info *info); 1100 1101 char name[TCP_CA_NAME_MAX]; 1102 struct module *owner; 1103 struct list_head list; 1104 u32 key; 1105 u32 flags; 1106 1107 /* initialize private data (optional) */ 1108 void (*init)(struct sock *sk); 1109 /* cleanup private data (optional) */ 1110 void (*release)(struct sock *sk); 1111 } ____cacheline_aligned_in_smp; 1112 1113 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1114 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1115 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1116 struct tcp_congestion_ops *old_type); 1117 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1118 1119 void tcp_assign_congestion_control(struct sock *sk); 1120 void tcp_init_congestion_control(struct sock *sk); 1121 void tcp_cleanup_congestion_control(struct sock *sk); 1122 int tcp_set_default_congestion_control(struct net *net, const char *name); 1123 void tcp_get_default_congestion_control(struct net *net, char *name); 1124 void tcp_get_available_congestion_control(char *buf, size_t len); 1125 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1126 int tcp_set_allowed_congestion_control(char *allowed); 1127 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1128 bool cap_net_admin); 1129 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1130 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1131 1132 u32 tcp_reno_ssthresh(struct sock *sk); 1133 u32 tcp_reno_undo_cwnd(struct sock *sk); 1134 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1135 extern struct tcp_congestion_ops tcp_reno; 1136 1137 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1138 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1139 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1140 #ifdef CONFIG_INET 1141 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1142 #else 1143 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1144 { 1145 return NULL; 1146 } 1147 #endif 1148 1149 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1150 { 1151 const struct inet_connection_sock *icsk = inet_csk(sk); 1152 1153 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1154 } 1155 1156 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1157 { 1158 const struct inet_connection_sock *icsk = inet_csk(sk); 1159 1160 if (icsk->icsk_ca_ops->cwnd_event) 1161 icsk->icsk_ca_ops->cwnd_event(sk, event); 1162 } 1163 1164 /* From tcp_cong.c */ 1165 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1166 1167 /* From tcp_rate.c */ 1168 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1169 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1170 struct rate_sample *rs); 1171 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1172 bool is_sack_reneg, struct rate_sample *rs); 1173 void tcp_rate_check_app_limited(struct sock *sk); 1174 1175 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1176 { 1177 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1178 } 1179 1180 /* These functions determine how the current flow behaves in respect of SACK 1181 * handling. SACK is negotiated with the peer, and therefore it can vary 1182 * between different flows. 1183 * 1184 * tcp_is_sack - SACK enabled 1185 * tcp_is_reno - No SACK 1186 */ 1187 static inline int tcp_is_sack(const struct tcp_sock *tp) 1188 { 1189 return likely(tp->rx_opt.sack_ok); 1190 } 1191 1192 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1193 { 1194 return !tcp_is_sack(tp); 1195 } 1196 1197 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1198 { 1199 return tp->sacked_out + tp->lost_out; 1200 } 1201 1202 /* This determines how many packets are "in the network" to the best 1203 * of our knowledge. In many cases it is conservative, but where 1204 * detailed information is available from the receiver (via SACK 1205 * blocks etc.) we can make more aggressive calculations. 1206 * 1207 * Use this for decisions involving congestion control, use just 1208 * tp->packets_out to determine if the send queue is empty or not. 1209 * 1210 * Read this equation as: 1211 * 1212 * "Packets sent once on transmission queue" MINUS 1213 * "Packets left network, but not honestly ACKed yet" PLUS 1214 * "Packets fast retransmitted" 1215 */ 1216 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1217 { 1218 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1219 } 1220 1221 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1222 1223 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1224 { 1225 return tp->snd_cwnd; 1226 } 1227 1228 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1229 { 1230 WARN_ON_ONCE((int)val <= 0); 1231 tp->snd_cwnd = val; 1232 } 1233 1234 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1235 { 1236 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1237 } 1238 1239 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1240 { 1241 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1242 } 1243 1244 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1245 { 1246 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1247 (1 << inet_csk(sk)->icsk_ca_state); 1248 } 1249 1250 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1251 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1252 * ssthresh. 1253 */ 1254 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1255 { 1256 const struct tcp_sock *tp = tcp_sk(sk); 1257 1258 if (tcp_in_cwnd_reduction(sk)) 1259 return tp->snd_ssthresh; 1260 else 1261 return max(tp->snd_ssthresh, 1262 ((tcp_snd_cwnd(tp) >> 1) + 1263 (tcp_snd_cwnd(tp) >> 2))); 1264 } 1265 1266 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1267 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1268 1269 void tcp_enter_cwr(struct sock *sk); 1270 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1271 1272 /* The maximum number of MSS of available cwnd for which TSO defers 1273 * sending if not using sysctl_tcp_tso_win_divisor. 1274 */ 1275 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1276 { 1277 return 3; 1278 } 1279 1280 /* Returns end sequence number of the receiver's advertised window */ 1281 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1282 { 1283 return tp->snd_una + tp->snd_wnd; 1284 } 1285 1286 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1287 * flexible approach. The RFC suggests cwnd should not be raised unless 1288 * it was fully used previously. And that's exactly what we do in 1289 * congestion avoidance mode. But in slow start we allow cwnd to grow 1290 * as long as the application has used half the cwnd. 1291 * Example : 1292 * cwnd is 10 (IW10), but application sends 9 frames. 1293 * We allow cwnd to reach 18 when all frames are ACKed. 1294 * This check is safe because it's as aggressive as slow start which already 1295 * risks 100% overshoot. The advantage is that we discourage application to 1296 * either send more filler packets or data to artificially blow up the cwnd 1297 * usage, and allow application-limited process to probe bw more aggressively. 1298 */ 1299 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1300 { 1301 const struct tcp_sock *tp = tcp_sk(sk); 1302 1303 if (tp->is_cwnd_limited) 1304 return true; 1305 1306 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1307 if (tcp_in_slow_start(tp)) 1308 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1309 1310 return false; 1311 } 1312 1313 /* BBR congestion control needs pacing. 1314 * Same remark for SO_MAX_PACING_RATE. 1315 * sch_fq packet scheduler is efficiently handling pacing, 1316 * but is not always installed/used. 1317 * Return true if TCP stack should pace packets itself. 1318 */ 1319 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1320 { 1321 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1322 } 1323 1324 /* Estimates in how many jiffies next packet for this flow can be sent. 1325 * Scheduling a retransmit timer too early would be silly. 1326 */ 1327 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1328 { 1329 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1330 1331 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1332 } 1333 1334 static inline void tcp_reset_xmit_timer(struct sock *sk, 1335 const int what, 1336 unsigned long when, 1337 const unsigned long max_when) 1338 { 1339 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1340 max_when); 1341 } 1342 1343 /* Something is really bad, we could not queue an additional packet, 1344 * because qdisc is full or receiver sent a 0 window, or we are paced. 1345 * We do not want to add fuel to the fire, or abort too early, 1346 * so make sure the timer we arm now is at least 200ms in the future, 1347 * regardless of current icsk_rto value (as it could be ~2ms) 1348 */ 1349 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1350 { 1351 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1352 } 1353 1354 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1355 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1356 unsigned long max_when) 1357 { 1358 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1359 inet_csk(sk)->icsk_backoff); 1360 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1361 1362 return (unsigned long)min_t(u64, when, max_when); 1363 } 1364 1365 static inline void tcp_check_probe_timer(struct sock *sk) 1366 { 1367 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1368 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1369 tcp_probe0_base(sk), TCP_RTO_MAX); 1370 } 1371 1372 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1373 { 1374 tp->snd_wl1 = seq; 1375 } 1376 1377 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1378 { 1379 tp->snd_wl1 = seq; 1380 } 1381 1382 /* 1383 * Calculate(/check) TCP checksum 1384 */ 1385 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1386 __be32 daddr, __wsum base) 1387 { 1388 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1389 } 1390 1391 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1392 { 1393 return !skb_csum_unnecessary(skb) && 1394 __skb_checksum_complete(skb); 1395 } 1396 1397 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1398 enum skb_drop_reason *reason); 1399 1400 1401 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1402 void tcp_set_state(struct sock *sk, int state); 1403 void tcp_done(struct sock *sk); 1404 int tcp_abort(struct sock *sk, int err); 1405 1406 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1407 { 1408 rx_opt->dsack = 0; 1409 rx_opt->num_sacks = 0; 1410 } 1411 1412 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1413 1414 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1415 { 1416 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1417 struct tcp_sock *tp = tcp_sk(sk); 1418 s32 delta; 1419 1420 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1421 tp->packets_out || ca_ops->cong_control) 1422 return; 1423 delta = tcp_jiffies32 - tp->lsndtime; 1424 if (delta > inet_csk(sk)->icsk_rto) 1425 tcp_cwnd_restart(sk, delta); 1426 } 1427 1428 /* Determine a window scaling and initial window to offer. */ 1429 void tcp_select_initial_window(const struct sock *sk, int __space, 1430 __u32 mss, __u32 *rcv_wnd, 1431 __u32 *window_clamp, int wscale_ok, 1432 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1433 1434 static inline int tcp_win_from_space(const struct sock *sk, int space) 1435 { 1436 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale); 1437 1438 return tcp_adv_win_scale <= 0 ? 1439 (space>>(-tcp_adv_win_scale)) : 1440 space - (space>>tcp_adv_win_scale); 1441 } 1442 1443 /* Note: caller must be prepared to deal with negative returns */ 1444 static inline int tcp_space(const struct sock *sk) 1445 { 1446 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1447 READ_ONCE(sk->sk_backlog.len) - 1448 atomic_read(&sk->sk_rmem_alloc)); 1449 } 1450 1451 static inline int tcp_full_space(const struct sock *sk) 1452 { 1453 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1454 } 1455 1456 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1457 { 1458 int unused_mem = sk_unused_reserved_mem(sk); 1459 struct tcp_sock *tp = tcp_sk(sk); 1460 1461 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 1462 if (unused_mem) 1463 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1464 tcp_win_from_space(sk, unused_mem)); 1465 } 1466 1467 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1468 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1469 1470 1471 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1472 * If 87.5 % (7/8) of the space has been consumed, we want to override 1473 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1474 * len/truesize ratio. 1475 */ 1476 static inline bool tcp_rmem_pressure(const struct sock *sk) 1477 { 1478 int rcvbuf, threshold; 1479 1480 if (tcp_under_memory_pressure(sk)) 1481 return true; 1482 1483 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1484 threshold = rcvbuf - (rcvbuf >> 3); 1485 1486 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1487 } 1488 1489 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1490 { 1491 const struct tcp_sock *tp = tcp_sk(sk); 1492 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1493 1494 if (avail <= 0) 1495 return false; 1496 1497 return (avail >= target) || tcp_rmem_pressure(sk) || 1498 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1499 } 1500 1501 extern void tcp_openreq_init_rwin(struct request_sock *req, 1502 const struct sock *sk_listener, 1503 const struct dst_entry *dst); 1504 1505 void tcp_enter_memory_pressure(struct sock *sk); 1506 void tcp_leave_memory_pressure(struct sock *sk); 1507 1508 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1509 { 1510 struct net *net = sock_net((struct sock *)tp); 1511 int val; 1512 1513 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1514 * and do_tcp_setsockopt(). 1515 */ 1516 val = READ_ONCE(tp->keepalive_intvl); 1517 1518 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1519 } 1520 1521 static inline int keepalive_time_when(const struct tcp_sock *tp) 1522 { 1523 struct net *net = sock_net((struct sock *)tp); 1524 int val; 1525 1526 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1527 val = READ_ONCE(tp->keepalive_time); 1528 1529 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1530 } 1531 1532 static inline int keepalive_probes(const struct tcp_sock *tp) 1533 { 1534 struct net *net = sock_net((struct sock *)tp); 1535 int val; 1536 1537 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1538 * and do_tcp_setsockopt(). 1539 */ 1540 val = READ_ONCE(tp->keepalive_probes); 1541 1542 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1543 } 1544 1545 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1546 { 1547 const struct inet_connection_sock *icsk = &tp->inet_conn; 1548 1549 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1550 tcp_jiffies32 - tp->rcv_tstamp); 1551 } 1552 1553 static inline int tcp_fin_time(const struct sock *sk) 1554 { 1555 int fin_timeout = tcp_sk(sk)->linger2 ? : 1556 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1557 const int rto = inet_csk(sk)->icsk_rto; 1558 1559 if (fin_timeout < (rto << 2) - (rto >> 1)) 1560 fin_timeout = (rto << 2) - (rto >> 1); 1561 1562 return fin_timeout; 1563 } 1564 1565 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1566 int paws_win) 1567 { 1568 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1569 return true; 1570 if (unlikely(!time_before32(ktime_get_seconds(), 1571 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) 1572 return true; 1573 /* 1574 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1575 * then following tcp messages have valid values. Ignore 0 value, 1576 * or else 'negative' tsval might forbid us to accept their packets. 1577 */ 1578 if (!rx_opt->ts_recent) 1579 return true; 1580 return false; 1581 } 1582 1583 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1584 int rst) 1585 { 1586 if (tcp_paws_check(rx_opt, 0)) 1587 return false; 1588 1589 /* RST segments are not recommended to carry timestamp, 1590 and, if they do, it is recommended to ignore PAWS because 1591 "their cleanup function should take precedence over timestamps." 1592 Certainly, it is mistake. It is necessary to understand the reasons 1593 of this constraint to relax it: if peer reboots, clock may go 1594 out-of-sync and half-open connections will not be reset. 1595 Actually, the problem would be not existing if all 1596 the implementations followed draft about maintaining clock 1597 via reboots. Linux-2.2 DOES NOT! 1598 1599 However, we can relax time bounds for RST segments to MSL. 1600 */ 1601 if (rst && !time_before32(ktime_get_seconds(), 1602 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1603 return false; 1604 return true; 1605 } 1606 1607 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1608 int mib_idx, u32 *last_oow_ack_time); 1609 1610 static inline void tcp_mib_init(struct net *net) 1611 { 1612 /* See RFC 2012 */ 1613 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1614 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1615 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1616 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1617 } 1618 1619 /* from STCP */ 1620 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1621 { 1622 tp->lost_skb_hint = NULL; 1623 } 1624 1625 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1626 { 1627 tcp_clear_retrans_hints_partial(tp); 1628 tp->retransmit_skb_hint = NULL; 1629 } 1630 1631 union tcp_md5_addr { 1632 struct in_addr a4; 1633 #if IS_ENABLED(CONFIG_IPV6) 1634 struct in6_addr a6; 1635 #endif 1636 }; 1637 1638 /* - key database */ 1639 struct tcp_md5sig_key { 1640 struct hlist_node node; 1641 u8 keylen; 1642 u8 family; /* AF_INET or AF_INET6 */ 1643 u8 prefixlen; 1644 u8 flags; 1645 union tcp_md5_addr addr; 1646 int l3index; /* set if key added with L3 scope */ 1647 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1648 struct rcu_head rcu; 1649 }; 1650 1651 /* - sock block */ 1652 struct tcp_md5sig_info { 1653 struct hlist_head head; 1654 struct rcu_head rcu; 1655 }; 1656 1657 /* - pseudo header */ 1658 struct tcp4_pseudohdr { 1659 __be32 saddr; 1660 __be32 daddr; 1661 __u8 pad; 1662 __u8 protocol; 1663 __be16 len; 1664 }; 1665 1666 struct tcp6_pseudohdr { 1667 struct in6_addr saddr; 1668 struct in6_addr daddr; 1669 __be32 len; 1670 __be32 protocol; /* including padding */ 1671 }; 1672 1673 union tcp_md5sum_block { 1674 struct tcp4_pseudohdr ip4; 1675 #if IS_ENABLED(CONFIG_IPV6) 1676 struct tcp6_pseudohdr ip6; 1677 #endif 1678 }; 1679 1680 /* - pool: digest algorithm, hash description and scratch buffer */ 1681 struct tcp_md5sig_pool { 1682 struct ahash_request *md5_req; 1683 void *scratch; 1684 }; 1685 1686 /* - functions */ 1687 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1688 const struct sock *sk, const struct sk_buff *skb); 1689 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1690 int family, u8 prefixlen, int l3index, u8 flags, 1691 const u8 *newkey, u8 newkeylen); 1692 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1693 int family, u8 prefixlen, int l3index, 1694 struct tcp_md5sig_key *key); 1695 1696 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1697 int family, u8 prefixlen, int l3index, u8 flags); 1698 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1699 const struct sock *addr_sk); 1700 1701 #ifdef CONFIG_TCP_MD5SIG 1702 #include <linux/jump_label.h> 1703 extern struct static_key_false_deferred tcp_md5_needed; 1704 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1705 const union tcp_md5_addr *addr, 1706 int family); 1707 static inline struct tcp_md5sig_key * 1708 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1709 const union tcp_md5_addr *addr, int family) 1710 { 1711 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1712 return NULL; 1713 return __tcp_md5_do_lookup(sk, l3index, addr, family); 1714 } 1715 1716 enum skb_drop_reason 1717 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1718 const void *saddr, const void *daddr, 1719 int family, int dif, int sdif); 1720 1721 1722 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1723 #else 1724 static inline struct tcp_md5sig_key * 1725 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1726 const union tcp_md5_addr *addr, int family) 1727 { 1728 return NULL; 1729 } 1730 1731 static inline enum skb_drop_reason 1732 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1733 const void *saddr, const void *daddr, 1734 int family, int dif, int sdif) 1735 { 1736 return SKB_NOT_DROPPED_YET; 1737 } 1738 #define tcp_twsk_md5_key(twsk) NULL 1739 #endif 1740 1741 bool tcp_alloc_md5sig_pool(void); 1742 1743 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1744 static inline void tcp_put_md5sig_pool(void) 1745 { 1746 local_bh_enable(); 1747 } 1748 1749 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1750 unsigned int header_len); 1751 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1752 const struct tcp_md5sig_key *key); 1753 1754 /* From tcp_fastopen.c */ 1755 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1756 struct tcp_fastopen_cookie *cookie); 1757 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1758 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1759 u16 try_exp); 1760 struct tcp_fastopen_request { 1761 /* Fast Open cookie. Size 0 means a cookie request */ 1762 struct tcp_fastopen_cookie cookie; 1763 struct msghdr *data; /* data in MSG_FASTOPEN */ 1764 size_t size; 1765 int copied; /* queued in tcp_connect() */ 1766 struct ubuf_info *uarg; 1767 }; 1768 void tcp_free_fastopen_req(struct tcp_sock *tp); 1769 void tcp_fastopen_destroy_cipher(struct sock *sk); 1770 void tcp_fastopen_ctx_destroy(struct net *net); 1771 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1772 void *primary_key, void *backup_key); 1773 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1774 u64 *key); 1775 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1776 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1777 struct request_sock *req, 1778 struct tcp_fastopen_cookie *foc, 1779 const struct dst_entry *dst); 1780 void tcp_fastopen_init_key_once(struct net *net); 1781 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1782 struct tcp_fastopen_cookie *cookie); 1783 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1784 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1785 #define TCP_FASTOPEN_KEY_MAX 2 1786 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1787 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1788 1789 /* Fastopen key context */ 1790 struct tcp_fastopen_context { 1791 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1792 int num; 1793 struct rcu_head rcu; 1794 }; 1795 1796 void tcp_fastopen_active_disable(struct sock *sk); 1797 bool tcp_fastopen_active_should_disable(struct sock *sk); 1798 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1799 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1800 1801 /* Caller needs to wrap with rcu_read_(un)lock() */ 1802 static inline 1803 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1804 { 1805 struct tcp_fastopen_context *ctx; 1806 1807 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1808 if (!ctx) 1809 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1810 return ctx; 1811 } 1812 1813 static inline 1814 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1815 const struct tcp_fastopen_cookie *orig) 1816 { 1817 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1818 orig->len == foc->len && 1819 !memcmp(orig->val, foc->val, foc->len)) 1820 return true; 1821 return false; 1822 } 1823 1824 static inline 1825 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1826 { 1827 return ctx->num; 1828 } 1829 1830 /* Latencies incurred by various limits for a sender. They are 1831 * chronograph-like stats that are mutually exclusive. 1832 */ 1833 enum tcp_chrono { 1834 TCP_CHRONO_UNSPEC, 1835 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1836 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1837 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1838 __TCP_CHRONO_MAX, 1839 }; 1840 1841 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1842 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1843 1844 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1845 * the same memory storage than skb->destructor/_skb_refdst 1846 */ 1847 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1848 { 1849 skb->destructor = NULL; 1850 skb->_skb_refdst = 0UL; 1851 } 1852 1853 #define tcp_skb_tsorted_save(skb) { \ 1854 unsigned long _save = skb->_skb_refdst; \ 1855 skb->_skb_refdst = 0UL; 1856 1857 #define tcp_skb_tsorted_restore(skb) \ 1858 skb->_skb_refdst = _save; \ 1859 } 1860 1861 void tcp_write_queue_purge(struct sock *sk); 1862 1863 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1864 { 1865 return skb_rb_first(&sk->tcp_rtx_queue); 1866 } 1867 1868 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 1869 { 1870 return skb_rb_last(&sk->tcp_rtx_queue); 1871 } 1872 1873 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1874 { 1875 return skb_peek_tail(&sk->sk_write_queue); 1876 } 1877 1878 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1879 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1880 1881 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1882 { 1883 return skb_peek(&sk->sk_write_queue); 1884 } 1885 1886 static inline bool tcp_skb_is_last(const struct sock *sk, 1887 const struct sk_buff *skb) 1888 { 1889 return skb_queue_is_last(&sk->sk_write_queue, skb); 1890 } 1891 1892 /** 1893 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 1894 * @sk: socket 1895 * 1896 * Since the write queue can have a temporary empty skb in it, 1897 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 1898 */ 1899 static inline bool tcp_write_queue_empty(const struct sock *sk) 1900 { 1901 const struct tcp_sock *tp = tcp_sk(sk); 1902 1903 return tp->write_seq == tp->snd_nxt; 1904 } 1905 1906 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1907 { 1908 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1909 } 1910 1911 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1912 { 1913 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1914 } 1915 1916 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1917 { 1918 __skb_queue_tail(&sk->sk_write_queue, skb); 1919 1920 /* Queue it, remembering where we must start sending. */ 1921 if (sk->sk_write_queue.next == skb) 1922 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1923 } 1924 1925 /* Insert new before skb on the write queue of sk. */ 1926 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1927 struct sk_buff *skb, 1928 struct sock *sk) 1929 { 1930 __skb_queue_before(&sk->sk_write_queue, skb, new); 1931 } 1932 1933 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1934 { 1935 tcp_skb_tsorted_anchor_cleanup(skb); 1936 __skb_unlink(skb, &sk->sk_write_queue); 1937 } 1938 1939 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1940 1941 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1942 { 1943 tcp_skb_tsorted_anchor_cleanup(skb); 1944 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1945 } 1946 1947 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1948 { 1949 list_del(&skb->tcp_tsorted_anchor); 1950 tcp_rtx_queue_unlink(skb, sk); 1951 tcp_wmem_free_skb(sk, skb); 1952 } 1953 1954 static inline void tcp_push_pending_frames(struct sock *sk) 1955 { 1956 if (tcp_send_head(sk)) { 1957 struct tcp_sock *tp = tcp_sk(sk); 1958 1959 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1960 } 1961 } 1962 1963 /* Start sequence of the skb just after the highest skb with SACKed 1964 * bit, valid only if sacked_out > 0 or when the caller has ensured 1965 * validity by itself. 1966 */ 1967 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1968 { 1969 if (!tp->sacked_out) 1970 return tp->snd_una; 1971 1972 if (tp->highest_sack == NULL) 1973 return tp->snd_nxt; 1974 1975 return TCP_SKB_CB(tp->highest_sack)->seq; 1976 } 1977 1978 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1979 { 1980 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 1981 } 1982 1983 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1984 { 1985 return tcp_sk(sk)->highest_sack; 1986 } 1987 1988 static inline void tcp_highest_sack_reset(struct sock *sk) 1989 { 1990 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 1991 } 1992 1993 /* Called when old skb is about to be deleted and replaced by new skb */ 1994 static inline void tcp_highest_sack_replace(struct sock *sk, 1995 struct sk_buff *old, 1996 struct sk_buff *new) 1997 { 1998 if (old == tcp_highest_sack(sk)) 1999 tcp_sk(sk)->highest_sack = new; 2000 } 2001 2002 /* This helper checks if socket has IP_TRANSPARENT set */ 2003 static inline bool inet_sk_transparent(const struct sock *sk) 2004 { 2005 switch (sk->sk_state) { 2006 case TCP_TIME_WAIT: 2007 return inet_twsk(sk)->tw_transparent; 2008 case TCP_NEW_SYN_RECV: 2009 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2010 } 2011 return inet_sk(sk)->transparent; 2012 } 2013 2014 /* Determines whether this is a thin stream (which may suffer from 2015 * increased latency). Used to trigger latency-reducing mechanisms. 2016 */ 2017 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2018 { 2019 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2020 } 2021 2022 /* /proc */ 2023 enum tcp_seq_states { 2024 TCP_SEQ_STATE_LISTENING, 2025 TCP_SEQ_STATE_ESTABLISHED, 2026 }; 2027 2028 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2029 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2030 void tcp_seq_stop(struct seq_file *seq, void *v); 2031 2032 struct tcp_seq_afinfo { 2033 sa_family_t family; 2034 }; 2035 2036 struct tcp_iter_state { 2037 struct seq_net_private p; 2038 enum tcp_seq_states state; 2039 struct sock *syn_wait_sk; 2040 int bucket, offset, sbucket, num; 2041 loff_t last_pos; 2042 }; 2043 2044 extern struct request_sock_ops tcp_request_sock_ops; 2045 extern struct request_sock_ops tcp6_request_sock_ops; 2046 2047 void tcp_v4_destroy_sock(struct sock *sk); 2048 2049 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2050 netdev_features_t features); 2051 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2052 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2053 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2054 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2055 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2056 void tcp_gro_complete(struct sk_buff *skb); 2057 2058 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2059 2060 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2061 { 2062 struct net *net = sock_net((struct sock *)tp); 2063 u32 val; 2064 2065 val = READ_ONCE(tp->notsent_lowat); 2066 2067 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2068 } 2069 2070 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2071 2072 #ifdef CONFIG_PROC_FS 2073 int tcp4_proc_init(void); 2074 void tcp4_proc_exit(void); 2075 #endif 2076 2077 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2078 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2079 const struct tcp_request_sock_ops *af_ops, 2080 struct sock *sk, struct sk_buff *skb); 2081 2082 /* TCP af-specific functions */ 2083 struct tcp_sock_af_ops { 2084 #ifdef CONFIG_TCP_MD5SIG 2085 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2086 const struct sock *addr_sk); 2087 int (*calc_md5_hash)(char *location, 2088 const struct tcp_md5sig_key *md5, 2089 const struct sock *sk, 2090 const struct sk_buff *skb); 2091 int (*md5_parse)(struct sock *sk, 2092 int optname, 2093 sockptr_t optval, 2094 int optlen); 2095 #endif 2096 }; 2097 2098 struct tcp_request_sock_ops { 2099 u16 mss_clamp; 2100 #ifdef CONFIG_TCP_MD5SIG 2101 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2102 const struct sock *addr_sk); 2103 int (*calc_md5_hash) (char *location, 2104 const struct tcp_md5sig_key *md5, 2105 const struct sock *sk, 2106 const struct sk_buff *skb); 2107 #endif 2108 #ifdef CONFIG_SYN_COOKIES 2109 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2110 __u16 *mss); 2111 #endif 2112 struct dst_entry *(*route_req)(const struct sock *sk, 2113 struct sk_buff *skb, 2114 struct flowi *fl, 2115 struct request_sock *req); 2116 u32 (*init_seq)(const struct sk_buff *skb); 2117 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2118 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2119 struct flowi *fl, struct request_sock *req, 2120 struct tcp_fastopen_cookie *foc, 2121 enum tcp_synack_type synack_type, 2122 struct sk_buff *syn_skb); 2123 }; 2124 2125 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2126 #if IS_ENABLED(CONFIG_IPV6) 2127 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2128 #endif 2129 2130 #ifdef CONFIG_SYN_COOKIES 2131 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2132 const struct sock *sk, struct sk_buff *skb, 2133 __u16 *mss) 2134 { 2135 tcp_synq_overflow(sk); 2136 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2137 return ops->cookie_init_seq(skb, mss); 2138 } 2139 #else 2140 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2141 const struct sock *sk, struct sk_buff *skb, 2142 __u16 *mss) 2143 { 2144 return 0; 2145 } 2146 #endif 2147 2148 int tcpv4_offload_init(void); 2149 2150 void tcp_v4_init(void); 2151 void tcp_init(void); 2152 2153 /* tcp_recovery.c */ 2154 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2155 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2156 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2157 u32 reo_wnd); 2158 extern bool tcp_rack_mark_lost(struct sock *sk); 2159 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2160 u64 xmit_time); 2161 extern void tcp_rack_reo_timeout(struct sock *sk); 2162 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2163 2164 /* tcp_plb.c */ 2165 2166 /* 2167 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2168 * expects cong_ratio which represents fraction of traffic that experienced 2169 * congestion over a single RTT. In order to avoid floating point operations, 2170 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2171 */ 2172 #define TCP_PLB_SCALE 8 2173 2174 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2175 struct tcp_plb_state { 2176 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2177 unused:3; 2178 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2179 }; 2180 2181 static inline void tcp_plb_init(const struct sock *sk, 2182 struct tcp_plb_state *plb) 2183 { 2184 plb->consec_cong_rounds = 0; 2185 plb->pause_until = 0; 2186 } 2187 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2188 const int cong_ratio); 2189 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2190 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2191 2192 /* At how many usecs into the future should the RTO fire? */ 2193 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2194 { 2195 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2196 u32 rto = inet_csk(sk)->icsk_rto; 2197 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2198 2199 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2200 } 2201 2202 /* 2203 * Save and compile IPv4 options, return a pointer to it 2204 */ 2205 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2206 struct sk_buff *skb) 2207 { 2208 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2209 struct ip_options_rcu *dopt = NULL; 2210 2211 if (opt->optlen) { 2212 int opt_size = sizeof(*dopt) + opt->optlen; 2213 2214 dopt = kmalloc(opt_size, GFP_ATOMIC); 2215 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2216 kfree(dopt); 2217 dopt = NULL; 2218 } 2219 } 2220 return dopt; 2221 } 2222 2223 /* locally generated TCP pure ACKs have skb->truesize == 2 2224 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2225 * This is much faster than dissecting the packet to find out. 2226 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2227 */ 2228 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2229 { 2230 return skb->truesize == 2; 2231 } 2232 2233 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2234 { 2235 skb->truesize = 2; 2236 } 2237 2238 static inline int tcp_inq(struct sock *sk) 2239 { 2240 struct tcp_sock *tp = tcp_sk(sk); 2241 int answ; 2242 2243 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2244 answ = 0; 2245 } else if (sock_flag(sk, SOCK_URGINLINE) || 2246 !tp->urg_data || 2247 before(tp->urg_seq, tp->copied_seq) || 2248 !before(tp->urg_seq, tp->rcv_nxt)) { 2249 2250 answ = tp->rcv_nxt - tp->copied_seq; 2251 2252 /* Subtract 1, if FIN was received */ 2253 if (answ && sock_flag(sk, SOCK_DONE)) 2254 answ--; 2255 } else { 2256 answ = tp->urg_seq - tp->copied_seq; 2257 } 2258 2259 return answ; 2260 } 2261 2262 int tcp_peek_len(struct socket *sock); 2263 2264 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2265 { 2266 u16 segs_in; 2267 2268 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2269 2270 /* We update these fields while other threads might 2271 * read them from tcp_get_info() 2272 */ 2273 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2274 if (skb->len > tcp_hdrlen(skb)) 2275 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2276 } 2277 2278 /* 2279 * TCP listen path runs lockless. 2280 * We forced "struct sock" to be const qualified to make sure 2281 * we don't modify one of its field by mistake. 2282 * Here, we increment sk_drops which is an atomic_t, so we can safely 2283 * make sock writable again. 2284 */ 2285 static inline void tcp_listendrop(const struct sock *sk) 2286 { 2287 atomic_inc(&((struct sock *)sk)->sk_drops); 2288 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2289 } 2290 2291 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2292 2293 /* 2294 * Interface for adding Upper Level Protocols over TCP 2295 */ 2296 2297 #define TCP_ULP_NAME_MAX 16 2298 #define TCP_ULP_MAX 128 2299 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2300 2301 struct tcp_ulp_ops { 2302 struct list_head list; 2303 2304 /* initialize ulp */ 2305 int (*init)(struct sock *sk); 2306 /* update ulp */ 2307 void (*update)(struct sock *sk, struct proto *p, 2308 void (*write_space)(struct sock *sk)); 2309 /* cleanup ulp */ 2310 void (*release)(struct sock *sk); 2311 /* diagnostic */ 2312 int (*get_info)(const struct sock *sk, struct sk_buff *skb); 2313 size_t (*get_info_size)(const struct sock *sk); 2314 /* clone ulp */ 2315 void (*clone)(const struct request_sock *req, struct sock *newsk, 2316 const gfp_t priority); 2317 2318 char name[TCP_ULP_NAME_MAX]; 2319 struct module *owner; 2320 }; 2321 int tcp_register_ulp(struct tcp_ulp_ops *type); 2322 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2323 int tcp_set_ulp(struct sock *sk, const char *name); 2324 void tcp_get_available_ulp(char *buf, size_t len); 2325 void tcp_cleanup_ulp(struct sock *sk); 2326 void tcp_update_ulp(struct sock *sk, struct proto *p, 2327 void (*write_space)(struct sock *sk)); 2328 2329 #define MODULE_ALIAS_TCP_ULP(name) \ 2330 __MODULE_INFO(alias, alias_userspace, name); \ 2331 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2332 2333 #ifdef CONFIG_NET_SOCK_MSG 2334 struct sk_msg; 2335 struct sk_psock; 2336 2337 #ifdef CONFIG_BPF_SYSCALL 2338 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock); 2339 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2340 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2341 #endif /* CONFIG_BPF_SYSCALL */ 2342 2343 #ifdef CONFIG_INET 2344 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2345 #else 2346 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2347 { 2348 } 2349 #endif 2350 2351 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2352 struct sk_msg *msg, u32 bytes, int flags); 2353 #endif /* CONFIG_NET_SOCK_MSG */ 2354 2355 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2356 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2357 { 2358 } 2359 #endif 2360 2361 #ifdef CONFIG_CGROUP_BPF 2362 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2363 struct sk_buff *skb, 2364 unsigned int end_offset) 2365 { 2366 skops->skb = skb; 2367 skops->skb_data_end = skb->data + end_offset; 2368 } 2369 #else 2370 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2371 struct sk_buff *skb, 2372 unsigned int end_offset) 2373 { 2374 } 2375 #endif 2376 2377 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2378 * is < 0, then the BPF op failed (for example if the loaded BPF 2379 * program does not support the chosen operation or there is no BPF 2380 * program loaded). 2381 */ 2382 #ifdef CONFIG_BPF 2383 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2384 { 2385 struct bpf_sock_ops_kern sock_ops; 2386 int ret; 2387 2388 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2389 if (sk_fullsock(sk)) { 2390 sock_ops.is_fullsock = 1; 2391 sock_owned_by_me(sk); 2392 } 2393 2394 sock_ops.sk = sk; 2395 sock_ops.op = op; 2396 if (nargs > 0) 2397 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2398 2399 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2400 if (ret == 0) 2401 ret = sock_ops.reply; 2402 else 2403 ret = -1; 2404 return ret; 2405 } 2406 2407 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2408 { 2409 u32 args[2] = {arg1, arg2}; 2410 2411 return tcp_call_bpf(sk, op, 2, args); 2412 } 2413 2414 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2415 u32 arg3) 2416 { 2417 u32 args[3] = {arg1, arg2, arg3}; 2418 2419 return tcp_call_bpf(sk, op, 3, args); 2420 } 2421 2422 #else 2423 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2424 { 2425 return -EPERM; 2426 } 2427 2428 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2429 { 2430 return -EPERM; 2431 } 2432 2433 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2434 u32 arg3) 2435 { 2436 return -EPERM; 2437 } 2438 2439 #endif 2440 2441 static inline u32 tcp_timeout_init(struct sock *sk) 2442 { 2443 int timeout; 2444 2445 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2446 2447 if (timeout <= 0) 2448 timeout = TCP_TIMEOUT_INIT; 2449 return min_t(int, timeout, TCP_RTO_MAX); 2450 } 2451 2452 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2453 { 2454 int rwnd; 2455 2456 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2457 2458 if (rwnd < 0) 2459 rwnd = 0; 2460 return rwnd; 2461 } 2462 2463 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2464 { 2465 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2466 } 2467 2468 static inline void tcp_bpf_rtt(struct sock *sk) 2469 { 2470 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2471 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2472 } 2473 2474 #if IS_ENABLED(CONFIG_SMC) 2475 extern struct static_key_false tcp_have_smc; 2476 #endif 2477 2478 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2479 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2480 void (*cad)(struct sock *sk, u32 ack_seq)); 2481 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2482 void clean_acked_data_flush(void); 2483 #endif 2484 2485 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2486 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2487 const struct tcp_sock *tp) 2488 { 2489 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2490 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2491 } 2492 2493 /* Compute Earliest Departure Time for some control packets 2494 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2495 */ 2496 static inline u64 tcp_transmit_time(const struct sock *sk) 2497 { 2498 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2499 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2500 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2501 2502 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2503 } 2504 return 0; 2505 } 2506 2507 #endif /* _TCP_H */ 2508