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 #include <linux/bits.h> 30 31 #include <net/inet_connection_sock.h> 32 #include <net/inet_timewait_sock.h> 33 #include <net/inet_hashtables.h> 34 #include <net/checksum.h> 35 #include <net/request_sock.h> 36 #include <net/sock_reuseport.h> 37 #include <net/sock.h> 38 #include <net/snmp.h> 39 #include <net/ip.h> 40 #include <net/tcp_states.h> 41 #include <net/tcp_ao.h> 42 #include <net/inet_ecn.h> 43 #include <net/dst.h> 44 #include <net/mptcp.h> 45 #include <net/xfrm.h> 46 #include <net/secure_seq.h> 47 48 #include <linux/seq_file.h> 49 #include <linux/memcontrol.h> 50 #include <linux/bpf-cgroup.h> 51 #include <linux/siphash.h> 52 53 extern struct inet_hashinfo tcp_hashinfo; 54 55 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 56 int tcp_orphan_count_sum(void); 57 58 static inline void tcp_orphan_count_inc(void) 59 { 60 this_cpu_inc(tcp_orphan_count); 61 } 62 63 static inline void tcp_orphan_count_dec(void) 64 { 65 this_cpu_dec(tcp_orphan_count); 66 } 67 68 DECLARE_PER_CPU(u32, tcp_tw_isn); 69 70 void tcp_time_wait(struct sock *sk, int state, int timeo); 71 72 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 73 #define MAX_TCP_OPTION_SPACE 40 74 #define TCP_MIN_SND_MSS 48 75 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 76 77 /* 78 * Never offer a window over 32767 without using window scaling. Some 79 * poor stacks do signed 16bit maths! 80 */ 81 #define MAX_TCP_WINDOW 32767U 82 83 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 84 #define TCP_MIN_MSS 88U 85 86 /* The initial MTU to use for probing */ 87 #define TCP_BASE_MSS 1024 88 89 /* probing interval, default to 10 minutes as per RFC4821 */ 90 #define TCP_PROBE_INTERVAL 600 91 92 /* Specify interval when tcp mtu probing will stop */ 93 #define TCP_PROBE_THRESHOLD 8 94 95 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 96 #define TCP_FASTRETRANS_THRESH 3 97 98 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 99 #define TCP_MAX_QUICKACKS 16U 100 101 /* Maximal number of window scale according to RFC1323 */ 102 #define TCP_MAX_WSCALE 14U 103 104 /* Default sending frequency of accurate ECN option per RTT */ 105 #define TCP_ACCECN_OPTION_BEACON 3 106 107 /* urg_data states */ 108 #define TCP_URG_VALID 0x0100 109 #define TCP_URG_NOTYET 0x0200 110 #define TCP_URG_READ 0x0400 111 112 #define TCP_RETR1 3 /* 113 * This is how many retries it does before it 114 * tries to figure out if the gateway is 115 * down. Minimal RFC value is 3; it corresponds 116 * to ~3sec-8min depending on RTO. 117 */ 118 119 #define TCP_RETR2 15 /* 120 * This should take at least 121 * 90 minutes to time out. 122 * RFC1122 says that the limit is 100 sec. 123 * 15 is ~13-30min depending on RTO. 124 */ 125 126 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 127 * when active opening a connection. 128 * RFC1122 says the minimum retry MUST 129 * be at least 180secs. Nevertheless 130 * this value is corresponding to 131 * 63secs of retransmission with the 132 * current initial RTO. 133 */ 134 135 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 136 * when passive opening a connection. 137 * This is corresponding to 31secs of 138 * retransmission with the current 139 * initial RTO. 140 */ 141 142 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 143 * state, about 60 seconds */ 144 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 145 /* BSD style FIN_WAIT2 deadlock breaker. 146 * It used to be 3min, new value is 60sec, 147 * to combine FIN-WAIT-2 timeout with 148 * TIME-WAIT timer. 149 */ 150 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 151 152 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 153 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); 154 155 #if HZ >= 100 156 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 157 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 158 #else 159 #define TCP_DELACK_MIN 4U 160 #define TCP_ATO_MIN 4U 161 #endif 162 #define TCP_RTO_MAX_SEC 120 163 #define TCP_RTO_MAX ((unsigned)(TCP_RTO_MAX_SEC * HZ)) 164 #define TCP_RTO_MIN ((unsigned)(HZ / 5)) 165 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 166 167 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 168 169 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 170 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 171 * used as a fallback RTO for the 172 * initial data transmission if no 173 * valid RTT sample has been acquired, 174 * most likely due to retrans in 3WHS. 175 */ 176 177 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 178 * for local resources. 179 */ 180 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 181 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 182 #define TCP_KEEPALIVE_INTVL (75*HZ) 183 184 #define MAX_TCP_KEEPIDLE 32767 185 #define MAX_TCP_KEEPINTVL 32767 186 #define MAX_TCP_KEEPCNT 127 187 #define MAX_TCP_SYNCNT 127 188 189 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds 190 * to avoid overflows. This assumes a clock smaller than 1 Mhz. 191 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz. 192 */ 193 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC) 194 195 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 196 * after this time. It should be equal 197 * (or greater than) TCP_TIMEWAIT_LEN 198 * to provide reliability equal to one 199 * provided by timewait state. 200 */ 201 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 202 * timestamps. It must be less than 203 * minimal timewait lifetime. 204 */ 205 /* 206 * TCP option 207 */ 208 209 #define TCPOPT_NOP 1 /* Padding */ 210 #define TCPOPT_EOL 0 /* End of options */ 211 #define TCPOPT_MSS 2 /* Segment size negotiating */ 212 #define TCPOPT_WINDOW 3 /* Window scaling */ 213 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 214 #define TCPOPT_SACK 5 /* SACK Block */ 215 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 216 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 217 #define TCPOPT_AO 29 /* Authentication Option (RFC5925) */ 218 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 219 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 220 #define TCPOPT_ACCECN0 172 /* 0xAC: Accurate ECN Order 0 */ 221 #define TCPOPT_ACCECN1 174 /* 0xAE: Accurate ECN Order 1 */ 222 #define TCPOPT_EXP 254 /* Experimental */ 223 /* Magic number to be after the option value for sharing TCP 224 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 225 */ 226 #define TCPOPT_FASTOPEN_MAGIC 0xF989 227 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 228 229 /* 230 * TCP option lengths 231 */ 232 233 #define TCPOLEN_MSS 4 234 #define TCPOLEN_WINDOW 3 235 #define TCPOLEN_SACK_PERM 2 236 #define TCPOLEN_TIMESTAMP 10 237 #define TCPOLEN_MD5SIG 18 238 #define TCPOLEN_FASTOPEN_BASE 2 239 #define TCPOLEN_ACCECN_BASE 2 240 #define TCPOLEN_EXP_FASTOPEN_BASE 4 241 #define TCPOLEN_EXP_SMC_BASE 6 242 243 /* But this is what stacks really send out. */ 244 #define TCPOLEN_TSTAMP_ALIGNED 12 245 #define TCPOLEN_WSCALE_ALIGNED 4 246 #define TCPOLEN_SACKPERM_ALIGNED 4 247 #define TCPOLEN_SACK_BASE 2 248 #define TCPOLEN_SACK_BASE_ALIGNED 4 249 #define TCPOLEN_SACK_PERBLOCK 8 250 #define TCPOLEN_MD5SIG_ALIGNED 20 251 #define TCPOLEN_MSS_ALIGNED 4 252 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 253 #define TCPOLEN_ACCECN_PERFIELD 3 254 255 /* Maximum number of byte counters in AccECN option + size */ 256 #define TCP_ACCECN_NUMFIELDS 3 257 #define TCP_ACCECN_MAXSIZE (TCPOLEN_ACCECN_BASE + \ 258 TCPOLEN_ACCECN_PERFIELD * \ 259 TCP_ACCECN_NUMFIELDS) 260 #define TCP_ACCECN_SAFETY_SHIFT 1 /* SAFETY_FACTOR in accecn draft */ 261 262 /* Flags in tp->nonagle */ 263 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 264 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 265 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 266 267 /* TCP thin-stream limits */ 268 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 269 270 /* TCP initial congestion window as per rfc6928 */ 271 #define TCP_INIT_CWND 10 272 273 /* Bit Flags for sysctl_tcp_fastopen */ 274 #define TFO_CLIENT_ENABLE 1 275 #define TFO_SERVER_ENABLE 2 276 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 277 278 /* Accept SYN data w/o any cookie option */ 279 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 280 281 /* Force enable TFO on all listeners, i.e., not requiring the 282 * TCP_FASTOPEN socket option. 283 */ 284 #define TFO_SERVER_WO_SOCKOPT1 0x400 285 286 287 /* sysctl variables for tcp */ 288 extern int sysctl_tcp_max_orphans; 289 extern long sysctl_tcp_mem[3]; 290 291 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 292 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 293 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 294 295 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 296 297 extern struct percpu_counter tcp_sockets_allocated; 298 extern unsigned long tcp_memory_pressure; 299 300 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 301 static inline bool tcp_under_memory_pressure(const struct sock *sk) 302 { 303 if (mem_cgroup_sk_enabled(sk) && 304 mem_cgroup_sk_under_memory_pressure(sk)) 305 return true; 306 307 if (sk->sk_bypass_prot_mem) 308 return false; 309 310 return READ_ONCE(tcp_memory_pressure); 311 } 312 /* 313 * The next routines deal with comparing 32 bit unsigned ints 314 * and worry about wraparound (automatic with unsigned arithmetic). 315 */ 316 317 static inline bool before(__u32 seq1, __u32 seq2) 318 { 319 return (__s32)(seq1-seq2) < 0; 320 } 321 #define after(seq2, seq1) before(seq1, seq2) 322 323 /* is s2<=s1<=s3 ? */ 324 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 325 { 326 return seq3 - seq2 >= seq1 - seq2; 327 } 328 329 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 330 { 331 sk_wmem_queued_add(sk, -skb->truesize); 332 if (!skb_zcopy_pure(skb)) 333 sk_mem_uncharge(sk, skb->truesize); 334 else 335 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 336 __kfree_skb(skb); 337 } 338 339 void sk_forced_mem_schedule(struct sock *sk, int size); 340 341 bool tcp_check_oom(const struct sock *sk, int shift); 342 343 344 extern struct proto tcp_prot; 345 346 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 347 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 348 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 349 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 350 351 /* 352 * TCP splice context 353 */ 354 struct tcp_splice_state { 355 struct pipe_inode_info *pipe; 356 size_t len; 357 unsigned int flags; 358 }; 359 360 void tcp_tsq_work_init(void); 361 362 int tcp_v4_err(struct sk_buff *skb, u32); 363 364 void tcp_shutdown(struct sock *sk, int how); 365 366 int tcp_v4_rcv(struct sk_buff *skb); 367 368 void tcp_remove_empty_skb(struct sock *sk); 369 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 370 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 371 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 372 size_t size, struct ubuf_info *uarg); 373 void tcp_splice_eof(struct socket *sock); 374 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 375 int tcp_wmem_schedule(struct sock *sk, int copy); 376 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 377 int size_goal); 378 379 void tcp_release_cb(struct sock *sk); 380 381 static inline bool tcp_release_cb_cond(struct sock *sk) 382 { 383 #ifdef CONFIG_INET 384 if (likely(sk->sk_prot->release_cb == tcp_release_cb)) { 385 if (unlikely(smp_load_acquire(&sk->sk_tsq_flags) & TCP_DEFERRED_ALL)) 386 tcp_release_cb(sk); 387 return true; 388 } 389 #endif 390 return false; 391 } 392 393 void tcp_wfree(struct sk_buff *skb); 394 void tcp_write_timer_handler(struct sock *sk); 395 void tcp_delack_timer_handler(struct sock *sk); 396 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 397 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 398 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 399 void tcp_rcvbuf_grow(struct sock *sk, u32 newval); 400 void tcp_rcv_space_adjust(struct sock *sk); 401 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 402 void tcp_twsk_destructor(struct sock *sk); 403 void tcp_twsk_purge(struct list_head *net_exit_list); 404 int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, 405 unsigned int offset, size_t len); 406 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 407 struct pipe_inode_info *pipe, size_t len, 408 unsigned int flags); 409 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 410 bool force_schedule); 411 412 static inline void tcp_dec_quickack_mode(struct sock *sk) 413 { 414 struct inet_connection_sock *icsk = inet_csk(sk); 415 416 if (icsk->icsk_ack.quick) { 417 /* How many ACKs S/ACKing new data have we sent? */ 418 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 419 420 if (pkts >= icsk->icsk_ack.quick) { 421 icsk->icsk_ack.quick = 0; 422 /* Leaving quickack mode we deflate ATO. */ 423 icsk->icsk_ack.ato = TCP_ATO_MIN; 424 } else 425 icsk->icsk_ack.quick -= pkts; 426 } 427 } 428 429 #define TCP_ECN_MODE_RFC3168 BIT(0) 430 #define TCP_ECN_QUEUE_CWR BIT(1) 431 #define TCP_ECN_DEMAND_CWR BIT(2) 432 #define TCP_ECN_SEEN BIT(3) 433 #define TCP_ECN_MODE_ACCECN BIT(4) 434 435 #define TCP_ECN_DISABLED 0 436 #define TCP_ECN_MODE_PENDING (TCP_ECN_MODE_RFC3168 | TCP_ECN_MODE_ACCECN) 437 #define TCP_ECN_MODE_ANY (TCP_ECN_MODE_RFC3168 | TCP_ECN_MODE_ACCECN) 438 439 static inline bool tcp_ecn_mode_any(const struct tcp_sock *tp) 440 { 441 return tp->ecn_flags & TCP_ECN_MODE_ANY; 442 } 443 444 static inline bool tcp_ecn_mode_rfc3168(const struct tcp_sock *tp) 445 { 446 return (tp->ecn_flags & TCP_ECN_MODE_ANY) == TCP_ECN_MODE_RFC3168; 447 } 448 449 static inline bool tcp_ecn_mode_accecn(const struct tcp_sock *tp) 450 { 451 return (tp->ecn_flags & TCP_ECN_MODE_ANY) == TCP_ECN_MODE_ACCECN; 452 } 453 454 static inline bool tcp_ecn_disabled(const struct tcp_sock *tp) 455 { 456 return !tcp_ecn_mode_any(tp); 457 } 458 459 static inline bool tcp_ecn_mode_pending(const struct tcp_sock *tp) 460 { 461 return (tp->ecn_flags & TCP_ECN_MODE_PENDING) == TCP_ECN_MODE_PENDING; 462 } 463 464 static inline void tcp_ecn_mode_set(struct tcp_sock *tp, u8 mode) 465 { 466 tp->ecn_flags &= ~TCP_ECN_MODE_ANY; 467 tp->ecn_flags |= mode; 468 } 469 470 enum tcp_tw_status { 471 TCP_TW_SUCCESS = 0, 472 TCP_TW_RST = 1, 473 TCP_TW_ACK = 2, 474 TCP_TW_SYN = 3, 475 TCP_TW_ACK_OOW = 4 476 }; 477 478 479 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 480 struct sk_buff *skb, 481 const struct tcphdr *th, 482 u32 *tw_isn, 483 enum skb_drop_reason *drop_reason); 484 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 485 struct request_sock *req, bool fastopen, 486 bool *lost_race, enum skb_drop_reason *drop_reason); 487 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child, 488 struct sk_buff *skb); 489 void tcp_enter_loss(struct sock *sk); 490 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 491 void tcp_clear_retrans(struct tcp_sock *tp); 492 void tcp_update_pacing_rate(struct sock *sk); 493 void tcp_set_rto(struct sock *sk); 494 void tcp_update_metrics(struct sock *sk); 495 void tcp_init_metrics(struct sock *sk); 496 void tcp_metrics_init(void); 497 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 498 void __tcp_close(struct sock *sk, long timeout); 499 void tcp_close(struct sock *sk, long timeout); 500 void tcp_init_sock(struct sock *sk); 501 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 502 __poll_t tcp_poll(struct file *file, struct socket *sock, 503 struct poll_table_struct *wait); 504 int do_tcp_getsockopt(struct sock *sk, int level, 505 int optname, sockptr_t optval, sockptr_t optlen); 506 int tcp_getsockopt(struct sock *sk, int level, int optname, 507 char __user *optval, int __user *optlen); 508 bool tcp_bpf_bypass_getsockopt(int level, int optname); 509 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 510 sockptr_t optval, unsigned int optlen); 511 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 512 unsigned int optlen); 513 void tcp_reset_keepalive_timer(struct sock *sk, unsigned long timeout); 514 void tcp_set_keepalive(struct sock *sk, int val); 515 void tcp_syn_ack_timeout(const struct request_sock *req); 516 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 517 int flags); 518 int tcp_set_rcvlowat(struct sock *sk, int val); 519 int tcp_set_window_clamp(struct sock *sk, int val); 520 521 static inline void 522 tcp_update_recv_tstamps(struct sk_buff *skb, 523 struct scm_timestamping_internal *tss) 524 { 525 tss->ts[0] = skb->tstamp; 526 tss->ts[2] = skb_hwtstamps(skb)->hwtstamp; 527 } 528 529 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 530 struct scm_timestamping_internal *tss); 531 void tcp_data_ready(struct sock *sk); 532 #ifdef CONFIG_MMU 533 int tcp_mmap(struct file *file, struct socket *sock, 534 struct vm_area_struct *vma); 535 #endif 536 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 537 struct tcp_options_received *opt_rx, 538 int estab, struct tcp_fastopen_cookie *foc); 539 540 /* 541 * BPF SKB-less helpers 542 */ 543 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 544 struct tcphdr *th, u32 *cookie); 545 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 546 struct tcphdr *th, u32 *cookie); 547 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 548 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 549 const struct tcp_request_sock_ops *af_ops, 550 struct sock *sk, struct tcphdr *th); 551 /* 552 * TCP v4 functions exported for the inet6 API 553 */ 554 555 void tcp_v4_mtu_reduced(struct sock *sk); 556 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 557 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 558 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 559 struct sock *tcp_create_openreq_child(const struct sock *sk, 560 struct request_sock *req, 561 struct sk_buff *skb); 562 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 563 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 564 struct request_sock *req, 565 struct dst_entry *dst, 566 struct request_sock *req_unhash, 567 bool *own_req, 568 void (*opt_child_init)(struct sock *newsk, 569 const struct sock *sk)); 570 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 571 int tcp_v4_connect(struct sock *sk, struct sockaddr_unsized *uaddr, int addr_len); 572 int tcp_connect(struct sock *sk); 573 enum tcp_synack_type { 574 TCP_SYNACK_NORMAL, 575 TCP_SYNACK_FASTOPEN, 576 TCP_SYNACK_COOKIE, 577 TCP_SYNACK_RETRANS, 578 }; 579 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 580 struct request_sock *req, 581 struct tcp_fastopen_cookie *foc, 582 enum tcp_synack_type synack_type, 583 struct sk_buff *syn_skb); 584 int tcp_disconnect(struct sock *sk, int flags); 585 586 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 587 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 588 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 589 590 /* From syncookies.c */ 591 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 592 struct request_sock *req, 593 struct dst_entry *dst); 594 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th); 595 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 596 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 597 struct sock *sk, struct sk_buff *skb, 598 struct tcp_options_received *tcp_opt, 599 int mss, u32 tsoff); 600 601 #if IS_ENABLED(CONFIG_BPF) 602 struct bpf_tcp_req_attrs { 603 u32 rcv_tsval; 604 u32 rcv_tsecr; 605 u16 mss; 606 u8 rcv_wscale; 607 u8 snd_wscale; 608 u8 ecn_ok; 609 u8 wscale_ok; 610 u8 sack_ok; 611 u8 tstamp_ok; 612 u8 usec_ts_ok; 613 u8 reserved[3]; 614 }; 615 #endif 616 617 #ifdef CONFIG_SYN_COOKIES 618 619 /* Syncookies use a monotonic timer which increments every 60 seconds. 620 * This counter is used both as a hash input and partially encoded into 621 * the cookie value. A cookie is only validated further if the delta 622 * between the current counter value and the encoded one is less than this, 623 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 624 * the counter advances immediately after a cookie is generated). 625 */ 626 #define MAX_SYNCOOKIE_AGE 2 627 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 628 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 629 630 /* syncookies: remember time of last synqueue overflow 631 * But do not dirty this field too often (once per second is enough) 632 * It is racy as we do not hold a lock, but race is very minor. 633 */ 634 static inline void tcp_synq_overflow(const struct sock *sk) 635 { 636 unsigned int last_overflow; 637 unsigned int now = jiffies; 638 639 if (sk->sk_reuseport) { 640 struct sock_reuseport *reuse; 641 642 reuse = rcu_dereference(sk->sk_reuseport_cb); 643 if (likely(reuse)) { 644 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 645 if (!time_between32(now, last_overflow, 646 last_overflow + HZ)) 647 WRITE_ONCE(reuse->synq_overflow_ts, now); 648 return; 649 } 650 } 651 652 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 653 if (!time_between32(now, last_overflow, last_overflow + HZ)) 654 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 655 } 656 657 /* syncookies: no recent synqueue overflow on this listening socket? */ 658 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 659 { 660 unsigned int last_overflow; 661 unsigned int now = jiffies; 662 663 if (sk->sk_reuseport) { 664 struct sock_reuseport *reuse; 665 666 reuse = rcu_dereference(sk->sk_reuseport_cb); 667 if (likely(reuse)) { 668 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 669 return !time_between32(now, last_overflow - HZ, 670 last_overflow + 671 TCP_SYNCOOKIE_VALID); 672 } 673 } 674 675 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 676 677 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 678 * then we're under synflood. However, we have to use 679 * 'last_overflow - HZ' as lower bound. That's because a concurrent 680 * tcp_synq_overflow() could update .ts_recent_stamp after we read 681 * jiffies but before we store .ts_recent_stamp into last_overflow, 682 * which could lead to rejecting a valid syncookie. 683 */ 684 return !time_between32(now, last_overflow - HZ, 685 last_overflow + TCP_SYNCOOKIE_VALID); 686 } 687 688 static inline u32 tcp_cookie_time(void) 689 { 690 u64 val = get_jiffies_64(); 691 692 do_div(val, TCP_SYNCOOKIE_PERIOD); 693 return val; 694 } 695 696 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */ 697 static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val) 698 { 699 if (usec_ts) 700 return div_u64(val, NSEC_PER_USEC); 701 702 return div_u64(val, NSEC_PER_MSEC); 703 } 704 705 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 706 u16 *mssp); 707 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 708 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 709 bool cookie_timestamp_decode(const struct net *net, 710 struct tcp_options_received *opt); 711 712 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst) 713 { 714 return READ_ONCE(net->ipv4.sysctl_tcp_ecn) || 715 dst_feature(dst, RTAX_FEATURE_ECN); 716 } 717 718 #if IS_ENABLED(CONFIG_BPF) 719 static inline bool cookie_bpf_ok(struct sk_buff *skb) 720 { 721 return skb->sk; 722 } 723 724 struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb); 725 #else 726 static inline bool cookie_bpf_ok(struct sk_buff *skb) 727 { 728 return false; 729 } 730 731 static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk, 732 struct sk_buff *skb) 733 { 734 return NULL; 735 } 736 #endif 737 738 /* From net/ipv6/syncookies.c */ 739 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th); 740 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 741 742 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 743 const struct tcphdr *th, u16 *mssp); 744 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 745 #endif 746 /* tcp_output.c */ 747 748 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 749 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 750 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 751 int nonagle); 752 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 753 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 754 void tcp_retransmit_timer(struct sock *sk); 755 void tcp_xmit_retransmit_queue(struct sock *); 756 void tcp_simple_retransmit(struct sock *); 757 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 758 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 759 enum tcp_queue { 760 TCP_FRAG_IN_WRITE_QUEUE, 761 TCP_FRAG_IN_RTX_QUEUE, 762 }; 763 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 764 struct sk_buff *skb, u32 len, 765 unsigned int mss_now, gfp_t gfp); 766 767 void tcp_send_probe0(struct sock *); 768 int tcp_write_wakeup(struct sock *, int mib); 769 void tcp_send_fin(struct sock *sk); 770 void tcp_send_active_reset(struct sock *sk, gfp_t priority, 771 enum sk_rst_reason reason); 772 int tcp_send_synack(struct sock *); 773 void tcp_push_one(struct sock *, unsigned int mss_now); 774 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt, u16 flags); 775 void tcp_send_ack(struct sock *sk); 776 void tcp_send_delayed_ack(struct sock *sk); 777 void tcp_send_loss_probe(struct sock *sk); 778 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 779 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 780 const struct sk_buff *next_skb); 781 782 /* tcp_input.c */ 783 void tcp_rearm_rto(struct sock *sk); 784 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 785 void tcp_done_with_error(struct sock *sk, int err); 786 void tcp_reset(struct sock *sk, struct sk_buff *skb); 787 void tcp_fin(struct sock *sk); 788 void __tcp_check_space(struct sock *sk); 789 static inline void tcp_check_space(struct sock *sk) 790 { 791 /* pairs with tcp_poll() */ 792 smp_mb(); 793 794 if (sk->sk_socket && test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) 795 __tcp_check_space(sk); 796 } 797 void tcp_sack_compress_send_ack(struct sock *sk); 798 799 static inline void tcp_cleanup_skb(struct sk_buff *skb) 800 { 801 skb_dst_drop(skb); 802 secpath_reset(skb); 803 } 804 805 static inline void tcp_add_receive_queue(struct sock *sk, struct sk_buff *skb) 806 { 807 DEBUG_NET_WARN_ON_ONCE(skb_dst(skb)); 808 DEBUG_NET_WARN_ON_ONCE(secpath_exists(skb)); 809 __skb_queue_tail(&sk->sk_receive_queue, skb); 810 } 811 812 /* tcp_timer.c */ 813 void tcp_init_xmit_timers(struct sock *); 814 static inline void tcp_clear_xmit_timers(struct sock *sk) 815 { 816 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 817 __sock_put(sk); 818 819 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 820 __sock_put(sk); 821 822 inet_csk_clear_xmit_timers(sk); 823 } 824 825 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 826 unsigned int tcp_current_mss(struct sock *sk); 827 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 828 829 /* Bound MSS / TSO packet size with the half of the window */ 830 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 831 { 832 int cutoff; 833 834 /* When peer uses tiny windows, there is no use in packetizing 835 * to sub-MSS pieces for the sake of SWS or making sure there 836 * are enough packets in the pipe for fast recovery. 837 * 838 * On the other hand, for extremely large MSS devices, handling 839 * smaller than MSS windows in this way does make sense. 840 */ 841 if (tp->max_window > TCP_MSS_DEFAULT) 842 cutoff = (tp->max_window >> 1); 843 else 844 cutoff = tp->max_window; 845 846 if (cutoff && pktsize > cutoff) 847 return max_t(int, cutoff, 68U - tp->tcp_header_len); 848 else 849 return pktsize; 850 } 851 852 /* tcp.c */ 853 void tcp_get_info(struct sock *, struct tcp_info *); 854 void tcp_rate_check_app_limited(struct sock *sk); 855 856 /* Read 'sendfile()'-style from a TCP socket */ 857 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 858 sk_read_actor_t recv_actor); 859 int tcp_read_sock_noack(struct sock *sk, read_descriptor_t *desc, 860 sk_read_actor_t recv_actor, bool noack, 861 u32 *copied_seq); 862 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 863 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 864 void tcp_read_done(struct sock *sk, size_t len); 865 866 void tcp_initialize_rcv_mss(struct sock *sk); 867 868 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 869 int tcp_mss_to_mtu(struct sock *sk, int mss); 870 void tcp_mtup_init(struct sock *sk); 871 872 static inline unsigned int tcp_rto_max(const struct sock *sk) 873 { 874 return READ_ONCE(inet_csk(sk)->icsk_rto_max); 875 } 876 877 static inline void tcp_bound_rto(struct sock *sk) 878 { 879 inet_csk(sk)->icsk_rto = min(inet_csk(sk)->icsk_rto, tcp_rto_max(sk)); 880 } 881 882 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 883 { 884 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 885 } 886 887 static inline unsigned long tcp_reqsk_timeout(struct request_sock *req) 888 { 889 u64 timeout = (u64)req->timeout << req->num_timeout; 890 891 return (unsigned long)min_t(u64, timeout, 892 tcp_rto_max(req->rsk_listener)); 893 } 894 895 u32 tcp_delack_max(const struct sock *sk); 896 897 /* Compute the actual rto_min value */ 898 static inline u32 tcp_rto_min(const struct sock *sk) 899 { 900 const struct dst_entry *dst = __sk_dst_get(sk); 901 u32 rto_min = READ_ONCE(inet_csk(sk)->icsk_rto_min); 902 903 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 904 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 905 return rto_min; 906 } 907 908 static inline u32 tcp_rto_min_us(const struct sock *sk) 909 { 910 return jiffies_to_usecs(tcp_rto_min(sk)); 911 } 912 913 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 914 { 915 return dst_metric_locked(dst, RTAX_CC_ALGO); 916 } 917 918 /* Minimum RTT in usec. ~0 means not available. */ 919 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 920 { 921 return minmax_get(&tp->rtt_min); 922 } 923 924 /* Compute the actual receive window we are currently advertising. 925 * Rcv_nxt can be after the window if our peer push more data 926 * than the offered window. 927 */ 928 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 929 { 930 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 931 932 if (win < 0) 933 win = 0; 934 return (u32) win; 935 } 936 937 /* Choose a new window, without checks for shrinking, and without 938 * scaling applied to the result. The caller does these things 939 * if necessary. This is a "raw" window selection. 940 */ 941 u32 __tcp_select_window(struct sock *sk); 942 943 void tcp_send_window_probe(struct sock *sk); 944 945 /* TCP uses 32bit jiffies to save some space. 946 * Note that this is different from tcp_time_stamp, which 947 * historically has been the same until linux-4.13. 948 */ 949 #define tcp_jiffies32 ((u32)jiffies) 950 951 /* 952 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 953 * It is no longer tied to jiffies, but to 1 ms clock. 954 * Note: double check if you want to use tcp_jiffies32 instead of this. 955 */ 956 #define TCP_TS_HZ 1000 957 958 static inline u64 tcp_clock_ns(void) 959 { 960 return ktime_get_ns(); 961 } 962 963 static inline u64 tcp_clock_us(void) 964 { 965 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 966 } 967 968 static inline u64 tcp_clock_ms(void) 969 { 970 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 971 } 972 973 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 974 * or usec resolution. Each socket carries a flag to select one or other 975 * resolution, as the route attribute could change anytime. 976 * Each flow must stick to initial resolution. 977 */ 978 static inline u32 tcp_clock_ts(bool usec_ts) 979 { 980 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 981 } 982 983 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 984 { 985 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 986 } 987 988 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 989 { 990 if (tp->tcp_usec_ts) 991 return tp->tcp_mstamp; 992 return tcp_time_stamp_ms(tp); 993 } 994 995 void tcp_mstamp_refresh(struct tcp_sock *tp); 996 997 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 998 { 999 return max_t(s64, t1 - t0, 0); 1000 } 1001 1002 /* provide the departure time in us unit */ 1003 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 1004 { 1005 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 1006 } 1007 1008 /* Provide skb TSval in usec or ms unit */ 1009 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 1010 { 1011 if (usec_ts) 1012 return tcp_skb_timestamp_us(skb); 1013 1014 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 1015 } 1016 1017 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 1018 { 1019 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 1020 } 1021 1022 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 1023 { 1024 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 1025 } 1026 1027 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 1028 1029 #define TCPHDR_FIN BIT(0) 1030 #define TCPHDR_SYN BIT(1) 1031 #define TCPHDR_RST BIT(2) 1032 #define TCPHDR_PSH BIT(3) 1033 #define TCPHDR_ACK BIT(4) 1034 #define TCPHDR_URG BIT(5) 1035 #define TCPHDR_ECE BIT(6) 1036 #define TCPHDR_CWR BIT(7) 1037 #define TCPHDR_AE BIT(8) 1038 #define TCPHDR_FLAGS_MASK (TCPHDR_FIN | TCPHDR_SYN | TCPHDR_RST | \ 1039 TCPHDR_PSH | TCPHDR_ACK | TCPHDR_URG | \ 1040 TCPHDR_ECE | TCPHDR_CWR | TCPHDR_AE) 1041 #define tcp_flags_ntohs(th) (ntohs(*(__be16 *)&tcp_flag_word(th)) & \ 1042 TCPHDR_FLAGS_MASK) 1043 1044 #define TCPHDR_ACE (TCPHDR_ECE | TCPHDR_CWR | TCPHDR_AE) 1045 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 1046 #define TCPHDR_SYNACK_ACCECN (TCPHDR_SYN | TCPHDR_ACK | TCPHDR_CWR) 1047 1048 #define TCP_ACCECN_CEP_ACE_MASK 0x7 1049 #define TCP_ACCECN_ACE_MAX_DELTA 6 1050 1051 /* To avoid/detect middlebox interference, not all counters start at 0. 1052 * See draft-ietf-tcpm-accurate-ecn for the latest values. 1053 */ 1054 #define TCP_ACCECN_CEP_INIT_OFFSET 5 1055 #define TCP_ACCECN_E1B_INIT_OFFSET 1 1056 #define TCP_ACCECN_E0B_INIT_OFFSET 1 1057 #define TCP_ACCECN_CEB_INIT_OFFSET 0 1058 1059 /* State flags for sacked in struct tcp_skb_cb */ 1060 enum tcp_skb_cb_sacked_flags { 1061 TCPCB_SACKED_ACKED = (1 << 0), /* SKB ACK'd by a SACK block */ 1062 TCPCB_SACKED_RETRANS = (1 << 1), /* SKB retransmitted */ 1063 TCPCB_LOST = (1 << 2), /* SKB is lost */ 1064 TCPCB_TAGBITS = (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS | 1065 TCPCB_LOST), /* All tag bits */ 1066 TCPCB_REPAIRED = (1 << 4), /* SKB repaired (no skb_mstamp_ns) */ 1067 TCPCB_EVER_RETRANS = (1 << 7), /* Ever retransmitted frame */ 1068 TCPCB_RETRANS = (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS | 1069 TCPCB_REPAIRED), 1070 }; 1071 1072 /* This is what the send packet queuing engine uses to pass 1073 * TCP per-packet control information to the transmission code. 1074 * We also store the host-order sequence numbers in here too. 1075 * This is 44 bytes if IPV6 is enabled. 1076 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 1077 */ 1078 struct tcp_skb_cb { 1079 __u32 seq; /* Starting sequence number */ 1080 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 1081 union { 1082 /* Note : 1083 * tcp_gso_segs/size are used in write queue only, 1084 * cf tcp_skb_pcount()/tcp_skb_mss() 1085 */ 1086 struct { 1087 u16 tcp_gso_segs; 1088 u16 tcp_gso_size; 1089 }; 1090 }; 1091 __u16 tcp_flags; /* TCP header flags (tcp[12-13])*/ 1092 1093 __u8 sacked; /* State flags for SACK. */ 1094 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 1095 #define TSTAMP_ACK_SK 0x1 1096 #define TSTAMP_ACK_BPF 0x2 1097 __u8 txstamp_ack:2, /* Record TX timestamp for ack? */ 1098 eor:1, /* Is skb MSG_EOR marked? */ 1099 has_rxtstamp:1, /* SKB has a RX timestamp */ 1100 unused:4; 1101 __u32 ack_seq; /* Sequence number ACK'd */ 1102 union { 1103 struct { 1104 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 1105 /* There is space for up to 24 bytes */ 1106 __u32 is_app_limited:1, /* cwnd not fully used? */ 1107 delivered_ce:20, 1108 unused:11; 1109 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 1110 __u32 delivered; 1111 /* start of send pipeline phase */ 1112 u64 first_tx_mstamp; 1113 /* when we reached the "delivered" count */ 1114 u64 delivered_mstamp; 1115 } tx; /* only used for outgoing skbs */ 1116 union { 1117 struct inet_skb_parm h4; 1118 #if IS_ENABLED(CONFIG_IPV6) 1119 struct inet6_skb_parm h6; 1120 #endif 1121 } header; /* For incoming skbs */ 1122 }; 1123 }; 1124 1125 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 1126 1127 extern const struct inet_connection_sock_af_ops ipv4_specific; 1128 1129 #if IS_ENABLED(CONFIG_IPV6) 1130 /* This is the variant of inet6_iif() that must be used by TCP, 1131 * as TCP moves IP6CB into a different location in skb->cb[] 1132 */ 1133 static inline int tcp_v6_iif(const struct sk_buff *skb) 1134 { 1135 return TCP_SKB_CB(skb)->header.h6.iif; 1136 } 1137 1138 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 1139 { 1140 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 1141 1142 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 1143 } 1144 1145 /* TCP_SKB_CB reference means this can not be used from early demux */ 1146 static inline int tcp_v6_sdif(const struct sk_buff *skb) 1147 { 1148 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1149 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 1150 return TCP_SKB_CB(skb)->header.h6.iif; 1151 #endif 1152 return 0; 1153 } 1154 1155 extern const struct inet_connection_sock_af_ops ipv6_specific; 1156 1157 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 1158 1159 #endif 1160 1161 /* TCP_SKB_CB reference means this can not be used from early demux */ 1162 static inline int tcp_v4_sdif(struct sk_buff *skb) 1163 { 1164 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1165 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 1166 return TCP_SKB_CB(skb)->header.h4.iif; 1167 #endif 1168 return 0; 1169 } 1170 1171 /* Due to TSO, an SKB can be composed of multiple actual 1172 * packets. To keep these tracked properly, we use this. 1173 */ 1174 static inline int tcp_skb_pcount(const struct sk_buff *skb) 1175 { 1176 return TCP_SKB_CB(skb)->tcp_gso_segs; 1177 } 1178 1179 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 1180 { 1181 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1182 } 1183 1184 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1185 { 1186 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1187 } 1188 1189 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1190 static inline int tcp_skb_mss(const struct sk_buff *skb) 1191 { 1192 return TCP_SKB_CB(skb)->tcp_gso_size; 1193 } 1194 1195 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1196 { 1197 return likely(!TCP_SKB_CB(skb)->eor); 1198 } 1199 1200 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1201 const struct sk_buff *from) 1202 { 1203 /* skb_cmp_decrypted() not needed, use tcp_write_collapse_fence() */ 1204 return likely(tcp_skb_can_collapse_to(to) && 1205 mptcp_skb_can_collapse(to, from) && 1206 skb_pure_zcopy_same(to, from) && 1207 skb_frags_readable(to) == skb_frags_readable(from)); 1208 } 1209 1210 static inline bool tcp_skb_can_collapse_rx(const struct sk_buff *to, 1211 const struct sk_buff *from) 1212 { 1213 return likely(mptcp_skb_can_collapse(to, from) && 1214 !skb_cmp_decrypted(to, from)); 1215 } 1216 1217 /* Events passed to congestion control interface */ 1218 enum tcp_ca_event { 1219 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1220 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1221 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1222 CA_EVENT_LOSS, /* loss timeout */ 1223 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1224 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1225 }; 1226 1227 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1228 enum tcp_ca_ack_event_flags { 1229 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1230 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1231 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1232 }; 1233 1234 /* 1235 * Interface for adding new TCP congestion control handlers 1236 */ 1237 #define TCP_CA_NAME_MAX 16 1238 #define TCP_CA_MAX 128 1239 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1240 1241 #define TCP_CA_UNSPEC 0 1242 1243 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1244 #define TCP_CONG_NON_RESTRICTED BIT(0) 1245 /* Requires ECN/ECT set on all packets */ 1246 #define TCP_CONG_NEEDS_ECN BIT(1) 1247 /* Require successfully negotiated AccECN capability */ 1248 #define TCP_CONG_NEEDS_ACCECN BIT(2) 1249 /* Use ECT(1) instead of ECT(0) while the CA is uninitialized */ 1250 #define TCP_CONG_ECT_1_NEGOTIATION BIT(3) 1251 /* Cannot fallback to RFC3168 during AccECN negotiation */ 1252 #define TCP_CONG_NO_FALLBACK_RFC3168 BIT(4) 1253 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN | \ 1254 TCP_CONG_NEEDS_ACCECN | TCP_CONG_ECT_1_NEGOTIATION | \ 1255 TCP_CONG_NO_FALLBACK_RFC3168) 1256 1257 union tcp_cc_info; 1258 1259 struct ack_sample { 1260 u32 pkts_acked; 1261 s32 rtt_us; 1262 u32 in_flight; 1263 }; 1264 1265 /* A rate sample measures the number of (original/retransmitted) data 1266 * packets delivered "delivered" over an interval of time "interval_us". 1267 * The tcp_rate.c code fills in the rate sample, and congestion 1268 * control modules that define a cong_control function to run at the end 1269 * of ACK processing can optionally chose to consult this sample when 1270 * setting cwnd and pacing rate. 1271 * A sample is invalid if "delivered" or "interval_us" is negative. 1272 */ 1273 struct rate_sample { 1274 u64 prior_mstamp; /* starting timestamp for interval */ 1275 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1276 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1277 s32 delivered; /* number of packets delivered over interval */ 1278 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1279 long interval_us; /* time for tp->delivered to incr "delivered" */ 1280 u32 snd_interval_us; /* snd interval for delivered packets */ 1281 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1282 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1283 int losses; /* number of packets marked lost upon ACK */ 1284 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1285 u32 prior_in_flight; /* in flight before this ACK */ 1286 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1287 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1288 bool is_retrans; /* is sample from retransmission? */ 1289 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1290 }; 1291 1292 struct tcp_congestion_ops { 1293 /* fast path fields are put first to fill one cache line */ 1294 1295 /* A congestion control (CC) must provide one of either: 1296 * 1297 * (a) a cong_avoid function, if the CC wants to use the core TCP 1298 * stack's default functionality to implement a "classic" 1299 * (Reno/CUBIC-style) response to packet loss, RFC3168 ECN, 1300 * idle periods, pacing rate computations, etc. 1301 * 1302 * (b) a cong_control function, if the CC wants custom behavior and 1303 * complete control of all congestion control behaviors. 1304 */ 1305 /* (a) "classic" response: calculate new cwnd. 1306 */ 1307 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1308 /* (b) "custom" response: call when packets are delivered to update 1309 * cwnd and pacing rate, after all the ca_state processing. 1310 */ 1311 void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs); 1312 1313 /* return slow start threshold (required) */ 1314 u32 (*ssthresh)(struct sock *sk); 1315 1316 /* call before changing ca_state (optional) */ 1317 void (*set_state)(struct sock *sk, u8 new_state); 1318 1319 /* call when cwnd event occurs (optional) */ 1320 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1321 1322 /* call when ack arrives (optional) */ 1323 void (*in_ack_event)(struct sock *sk, u32 flags); 1324 1325 /* hook for packet ack accounting (optional) */ 1326 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1327 1328 /* override sysctl_tcp_min_tso_segs (optional) */ 1329 u32 (*min_tso_segs)(struct sock *sk); 1330 1331 /* new value of cwnd after loss (required) */ 1332 u32 (*undo_cwnd)(struct sock *sk); 1333 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1334 u32 (*sndbuf_expand)(struct sock *sk); 1335 1336 /* control/slow paths put last */ 1337 /* get info for inet_diag (optional) */ 1338 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1339 union tcp_cc_info *info); 1340 1341 char name[TCP_CA_NAME_MAX]; 1342 struct module *owner; 1343 struct list_head list; 1344 u32 key; 1345 u32 flags; 1346 1347 /* initialize private data (optional) */ 1348 void (*init)(struct sock *sk); 1349 /* cleanup private data (optional) */ 1350 void (*release)(struct sock *sk); 1351 } ____cacheline_aligned_in_smp; 1352 1353 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1354 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1355 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1356 struct tcp_congestion_ops *old_type); 1357 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1358 1359 void tcp_assign_congestion_control(struct sock *sk); 1360 void tcp_init_congestion_control(struct sock *sk); 1361 void tcp_cleanup_congestion_control(struct sock *sk); 1362 int tcp_set_default_congestion_control(struct net *net, const char *name); 1363 void tcp_get_default_congestion_control(struct net *net, char *name); 1364 void tcp_get_available_congestion_control(char *buf, size_t len); 1365 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1366 int tcp_set_allowed_congestion_control(char *allowed); 1367 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1368 bool cap_net_admin); 1369 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1370 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1371 1372 u32 tcp_reno_ssthresh(struct sock *sk); 1373 u32 tcp_reno_undo_cwnd(struct sock *sk); 1374 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1375 extern struct tcp_congestion_ops tcp_reno; 1376 1377 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1378 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1379 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1380 #ifdef CONFIG_INET 1381 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1382 #else 1383 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1384 { 1385 return NULL; 1386 } 1387 #endif 1388 1389 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1390 { 1391 const struct inet_connection_sock *icsk = inet_csk(sk); 1392 1393 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1394 } 1395 1396 static inline bool tcp_ca_needs_accecn(const struct sock *sk) 1397 { 1398 const struct inet_connection_sock *icsk = inet_csk(sk); 1399 1400 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ACCECN; 1401 } 1402 1403 static inline bool tcp_ca_ect_1_negotiation(const struct sock *sk) 1404 { 1405 const struct inet_connection_sock *icsk = inet_csk(sk); 1406 1407 return icsk->icsk_ca_ops->flags & TCP_CONG_ECT_1_NEGOTIATION; 1408 } 1409 1410 static inline bool tcp_ca_no_fallback_rfc3168(const struct sock *sk) 1411 { 1412 const struct inet_connection_sock *icsk = inet_csk(sk); 1413 1414 return icsk->icsk_ca_ops->flags & TCP_CONG_NO_FALLBACK_RFC3168; 1415 } 1416 1417 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1418 { 1419 const struct inet_connection_sock *icsk = inet_csk(sk); 1420 1421 if (icsk->icsk_ca_ops->cwnd_event) 1422 icsk->icsk_ca_ops->cwnd_event(sk, event); 1423 } 1424 1425 /* From tcp_cong.c */ 1426 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1427 1428 1429 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1430 { 1431 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1432 } 1433 1434 /* These functions determine how the current flow behaves in respect of SACK 1435 * handling. SACK is negotiated with the peer, and therefore it can vary 1436 * between different flows. 1437 * 1438 * tcp_is_sack - SACK enabled 1439 * tcp_is_reno - No SACK 1440 */ 1441 static inline int tcp_is_sack(const struct tcp_sock *tp) 1442 { 1443 return likely(tp->rx_opt.sack_ok); 1444 } 1445 1446 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1447 { 1448 return !tcp_is_sack(tp); 1449 } 1450 1451 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1452 { 1453 return tp->sacked_out + tp->lost_out; 1454 } 1455 1456 /* This determines how many packets are "in the network" to the best 1457 * of our knowledge. In many cases it is conservative, but where 1458 * detailed information is available from the receiver (via SACK 1459 * blocks etc.) we can make more aggressive calculations. 1460 * 1461 * Use this for decisions involving congestion control, use just 1462 * tp->packets_out to determine if the send queue is empty or not. 1463 * 1464 * Read this equation as: 1465 * 1466 * "Packets sent once on transmission queue" MINUS 1467 * "Packets left network, but not honestly ACKed yet" PLUS 1468 * "Packets fast retransmitted" 1469 */ 1470 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1471 { 1472 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1473 } 1474 1475 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1476 1477 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1478 { 1479 return tp->snd_cwnd; 1480 } 1481 1482 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1483 { 1484 WARN_ON_ONCE((int)val <= 0); 1485 tp->snd_cwnd = val; 1486 } 1487 1488 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1489 { 1490 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1491 } 1492 1493 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1494 { 1495 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1496 } 1497 1498 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1499 { 1500 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1501 (1 << inet_csk(sk)->icsk_ca_state); 1502 } 1503 1504 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1505 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1506 * ssthresh. 1507 */ 1508 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1509 { 1510 const struct tcp_sock *tp = tcp_sk(sk); 1511 1512 if (tcp_in_cwnd_reduction(sk)) 1513 return tp->snd_ssthresh; 1514 else 1515 return max(tp->snd_ssthresh, 1516 ((tcp_snd_cwnd(tp) >> 1) + 1517 (tcp_snd_cwnd(tp) >> 2))); 1518 } 1519 1520 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1521 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1522 1523 void tcp_enter_cwr(struct sock *sk); 1524 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1525 1526 /* The maximum number of MSS of available cwnd for which TSO defers 1527 * sending if not using sysctl_tcp_tso_win_divisor. 1528 */ 1529 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1530 { 1531 return 3; 1532 } 1533 1534 /* Returns end sequence number of the receiver's advertised window */ 1535 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1536 { 1537 return tp->snd_una + tp->snd_wnd; 1538 } 1539 1540 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1541 * flexible approach. The RFC suggests cwnd should not be raised unless 1542 * it was fully used previously. And that's exactly what we do in 1543 * congestion avoidance mode. But in slow start we allow cwnd to grow 1544 * as long as the application has used half the cwnd. 1545 * Example : 1546 * cwnd is 10 (IW10), but application sends 9 frames. 1547 * We allow cwnd to reach 18 when all frames are ACKed. 1548 * This check is safe because it's as aggressive as slow start which already 1549 * risks 100% overshoot. The advantage is that we discourage application to 1550 * either send more filler packets or data to artificially blow up the cwnd 1551 * usage, and allow application-limited process to probe bw more aggressively. 1552 */ 1553 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1554 { 1555 const struct tcp_sock *tp = tcp_sk(sk); 1556 1557 if (tp->is_cwnd_limited) 1558 return true; 1559 1560 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1561 if (tcp_in_slow_start(tp)) 1562 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1563 1564 return false; 1565 } 1566 1567 /* BBR congestion control needs pacing. 1568 * Same remark for SO_MAX_PACING_RATE. 1569 * sch_fq packet scheduler is efficiently handling pacing, 1570 * but is not always installed/used. 1571 * Return true if TCP stack should pace packets itself. 1572 */ 1573 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1574 { 1575 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1576 } 1577 1578 /* Estimates in how many jiffies next packet for this flow can be sent. 1579 * Scheduling a retransmit timer too early would be silly. 1580 */ 1581 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1582 { 1583 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1584 1585 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1586 } 1587 1588 static inline void tcp_reset_xmit_timer(struct sock *sk, 1589 const int what, 1590 unsigned long when, 1591 bool pace_delay) 1592 { 1593 if (pace_delay) 1594 when += tcp_pacing_delay(sk); 1595 inet_csk_reset_xmit_timer(sk, what, when, 1596 tcp_rto_max(sk)); 1597 } 1598 1599 /* Something is really bad, we could not queue an additional packet, 1600 * because qdisc is full or receiver sent a 0 window, or we are paced. 1601 * We do not want to add fuel to the fire, or abort too early, 1602 * so make sure the timer we arm now is at least 200ms in the future, 1603 * regardless of current icsk_rto value (as it could be ~2ms) 1604 */ 1605 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1606 { 1607 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1608 } 1609 1610 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1611 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1612 unsigned long max_when) 1613 { 1614 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1615 inet_csk(sk)->icsk_backoff); 1616 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1617 1618 return (unsigned long)min_t(u64, when, max_when); 1619 } 1620 1621 static inline void tcp_check_probe_timer(struct sock *sk) 1622 { 1623 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1624 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1625 tcp_probe0_base(sk), true); 1626 } 1627 1628 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1629 { 1630 tp->snd_wl1 = seq; 1631 } 1632 1633 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1634 { 1635 tp->snd_wl1 = seq; 1636 } 1637 1638 /* 1639 * Calculate(/check) TCP checksum 1640 */ 1641 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1642 __be32 daddr, __wsum base) 1643 { 1644 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1645 } 1646 1647 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1648 { 1649 return !skb_csum_unnecessary(skb) && 1650 __skb_checksum_complete(skb); 1651 } 1652 1653 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1654 enum skb_drop_reason *reason); 1655 1656 static inline int tcp_filter(struct sock *sk, struct sk_buff *skb, 1657 enum skb_drop_reason *reason) 1658 { 1659 const struct tcphdr *th = (const struct tcphdr *)skb->data; 1660 1661 return sk_filter_trim_cap(sk, skb, __tcp_hdrlen(th), reason); 1662 } 1663 1664 void tcp_set_state(struct sock *sk, int state); 1665 void tcp_done(struct sock *sk); 1666 int tcp_abort(struct sock *sk, int err); 1667 1668 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1669 { 1670 rx_opt->dsack = 0; 1671 rx_opt->num_sacks = 0; 1672 } 1673 1674 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1675 1676 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1677 { 1678 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1679 struct tcp_sock *tp = tcp_sk(sk); 1680 s32 delta; 1681 1682 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1683 tp->packets_out || ca_ops->cong_control) 1684 return; 1685 delta = tcp_jiffies32 - tp->lsndtime; 1686 if (delta > inet_csk(sk)->icsk_rto) 1687 tcp_cwnd_restart(sk, delta); 1688 } 1689 1690 /* Determine a window scaling and initial window to offer. */ 1691 void tcp_select_initial_window(const struct sock *sk, int __space, 1692 __u32 mss, __u32 *rcv_wnd, 1693 __u32 *window_clamp, int wscale_ok, 1694 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1695 1696 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1697 { 1698 s64 scaled_space = (s64)space * scaling_ratio; 1699 1700 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1701 } 1702 1703 static inline int tcp_win_from_space(const struct sock *sk, int space) 1704 { 1705 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1706 } 1707 1708 /* inverse of __tcp_win_from_space() */ 1709 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1710 { 1711 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1712 1713 do_div(val, scaling_ratio); 1714 return val; 1715 } 1716 1717 static inline int tcp_space_from_win(const struct sock *sk, int win) 1718 { 1719 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1720 } 1721 1722 /* Assume a 50% default for skb->len/skb->truesize ratio. 1723 * This may be adjusted later in tcp_measure_rcv_mss(). 1724 */ 1725 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1)) 1726 1727 static inline void tcp_scaling_ratio_init(struct sock *sk) 1728 { 1729 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1730 } 1731 1732 /* Note: caller must be prepared to deal with negative returns */ 1733 static inline int tcp_space(const struct sock *sk) 1734 { 1735 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1736 READ_ONCE(sk->sk_backlog.len) - 1737 atomic_read(&sk->sk_rmem_alloc)); 1738 } 1739 1740 static inline int tcp_full_space(const struct sock *sk) 1741 { 1742 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1743 } 1744 1745 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh) 1746 { 1747 int unused_mem = sk_unused_reserved_mem(sk); 1748 struct tcp_sock *tp = tcp_sk(sk); 1749 1750 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh); 1751 if (unused_mem) 1752 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1753 tcp_win_from_space(sk, unused_mem)); 1754 } 1755 1756 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1757 { 1758 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss); 1759 } 1760 1761 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1762 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1763 1764 1765 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1766 * If 87.5 % (7/8) of the space has been consumed, we want to override 1767 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1768 * len/truesize ratio. 1769 */ 1770 static inline bool tcp_rmem_pressure(const struct sock *sk) 1771 { 1772 int rcvbuf, threshold; 1773 1774 if (tcp_under_memory_pressure(sk)) 1775 return true; 1776 1777 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1778 threshold = rcvbuf - (rcvbuf >> 3); 1779 1780 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1781 } 1782 1783 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1784 { 1785 const struct tcp_sock *tp = tcp_sk(sk); 1786 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1787 1788 if (avail <= 0) 1789 return false; 1790 1791 return (avail >= target) || tcp_rmem_pressure(sk) || 1792 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1793 } 1794 1795 extern void tcp_openreq_init_rwin(struct request_sock *req, 1796 const struct sock *sk_listener, 1797 const struct dst_entry *dst); 1798 1799 void tcp_enter_memory_pressure(struct sock *sk); 1800 void tcp_leave_memory_pressure(struct sock *sk); 1801 1802 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1803 { 1804 struct net *net = sock_net((struct sock *)tp); 1805 int val; 1806 1807 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1808 * and do_tcp_setsockopt(). 1809 */ 1810 val = READ_ONCE(tp->keepalive_intvl); 1811 1812 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1813 } 1814 1815 static inline int keepalive_time_when(const struct tcp_sock *tp) 1816 { 1817 struct net *net = sock_net((struct sock *)tp); 1818 int val; 1819 1820 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1821 val = READ_ONCE(tp->keepalive_time); 1822 1823 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1824 } 1825 1826 static inline int keepalive_probes(const struct tcp_sock *tp) 1827 { 1828 struct net *net = sock_net((struct sock *)tp); 1829 int val; 1830 1831 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1832 * and do_tcp_setsockopt(). 1833 */ 1834 val = READ_ONCE(tp->keepalive_probes); 1835 1836 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1837 } 1838 1839 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1840 { 1841 const struct inet_connection_sock *icsk = &tp->inet_conn; 1842 1843 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1844 tcp_jiffies32 - tp->rcv_tstamp); 1845 } 1846 1847 static inline int tcp_fin_time(const struct sock *sk) 1848 { 1849 int fin_timeout = tcp_sk(sk)->linger2 ? : 1850 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1851 const int rto = inet_csk(sk)->icsk_rto; 1852 1853 if (fin_timeout < (rto << 2) - (rto >> 1)) 1854 fin_timeout = (rto << 2) - (rto >> 1); 1855 1856 return fin_timeout; 1857 } 1858 1859 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1860 int paws_win) 1861 { 1862 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1863 return true; 1864 if (unlikely(!time_before32(ktime_get_seconds(), 1865 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1866 return true; 1867 /* 1868 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1869 * then following tcp messages have valid values. Ignore 0 value, 1870 * or else 'negative' tsval might forbid us to accept their packets. 1871 */ 1872 if (!rx_opt->ts_recent) 1873 return true; 1874 return false; 1875 } 1876 1877 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1878 int rst) 1879 { 1880 if (tcp_paws_check(rx_opt, 0)) 1881 return false; 1882 1883 /* RST segments are not recommended to carry timestamp, 1884 and, if they do, it is recommended to ignore PAWS because 1885 "their cleanup function should take precedence over timestamps." 1886 Certainly, it is mistake. It is necessary to understand the reasons 1887 of this constraint to relax it: if peer reboots, clock may go 1888 out-of-sync and half-open connections will not be reset. 1889 Actually, the problem would be not existing if all 1890 the implementations followed draft about maintaining clock 1891 via reboots. Linux-2.2 DOES NOT! 1892 1893 However, we can relax time bounds for RST segments to MSL. 1894 */ 1895 if (rst && !time_before32(ktime_get_seconds(), 1896 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1897 return false; 1898 return true; 1899 } 1900 1901 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 1902 { 1903 u32 ace; 1904 1905 /* mptcp hooks are only on the slow path */ 1906 if (sk_is_mptcp((struct sock *)tp)) 1907 return; 1908 1909 ace = tcp_ecn_mode_accecn(tp) ? 1910 ((tp->delivered_ce + TCP_ACCECN_CEP_INIT_OFFSET) & 1911 TCP_ACCECN_CEP_ACE_MASK) : 0; 1912 1913 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 1914 (ace << 22) | 1915 ntohl(TCP_FLAG_ACK) | 1916 snd_wnd); 1917 } 1918 1919 static inline void tcp_fast_path_on(struct tcp_sock *tp) 1920 { 1921 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 1922 } 1923 1924 static inline void tcp_fast_path_check(struct sock *sk) 1925 { 1926 struct tcp_sock *tp = tcp_sk(sk); 1927 1928 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 1929 tp->rcv_wnd && 1930 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 1931 !tp->urg_data) 1932 tcp_fast_path_on(tp); 1933 } 1934 1935 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1936 int mib_idx, u32 *last_oow_ack_time); 1937 1938 static inline void tcp_mib_init(struct net *net) 1939 { 1940 /* See RFC 2012 */ 1941 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1942 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1943 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1944 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1945 } 1946 1947 /* from STCP */ 1948 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1949 { 1950 tp->retransmit_skb_hint = NULL; 1951 } 1952 1953 #define tcp_md5_addr tcp_ao_addr 1954 1955 /* - key database */ 1956 struct tcp_md5sig_key { 1957 struct hlist_node node; 1958 u8 keylen; 1959 u8 family; /* AF_INET or AF_INET6 */ 1960 u8 prefixlen; 1961 u8 flags; 1962 union tcp_md5_addr addr; 1963 int l3index; /* set if key added with L3 scope */ 1964 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1965 struct rcu_head rcu; 1966 }; 1967 1968 /* - sock block */ 1969 struct tcp_md5sig_info { 1970 struct hlist_head head; 1971 struct rcu_head rcu; 1972 }; 1973 1974 /* - pseudo header */ 1975 struct tcp4_pseudohdr { 1976 __be32 saddr; 1977 __be32 daddr; 1978 __u8 pad; 1979 __u8 protocol; 1980 __be16 len; 1981 }; 1982 1983 struct tcp6_pseudohdr { 1984 struct in6_addr saddr; 1985 struct in6_addr daddr; 1986 __be32 len; 1987 __be32 protocol; /* including padding */ 1988 }; 1989 1990 /* 1991 * struct tcp_sigpool - per-CPU pool of ahash_requests 1992 * @scratch: per-CPU temporary area, that can be used between 1993 * tcp_sigpool_start() and tcp_sigpool_end() to perform 1994 * crypto request 1995 * @req: pre-allocated ahash request 1996 */ 1997 struct tcp_sigpool { 1998 void *scratch; 1999 struct ahash_request *req; 2000 }; 2001 2002 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size); 2003 void tcp_sigpool_get(unsigned int id); 2004 void tcp_sigpool_release(unsigned int id); 2005 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp, 2006 const struct sk_buff *skb, 2007 unsigned int header_len); 2008 2009 /** 2010 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash 2011 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash() 2012 * @c: returned tcp_sigpool for usage (uninitialized on failure) 2013 * 2014 * Returns: 0 on success, error otherwise. 2015 */ 2016 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c); 2017 /** 2018 * tcp_sigpool_end - enable bh and stop using tcp_sigpool 2019 * @c: tcp_sigpool context that was returned by tcp_sigpool_start() 2020 */ 2021 void tcp_sigpool_end(struct tcp_sigpool *c); 2022 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len); 2023 /* - functions */ 2024 void tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 2025 const struct sock *sk, const struct sk_buff *skb); 2026 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 2027 int family, u8 prefixlen, int l3index, u8 flags, 2028 const u8 *newkey, u8 newkeylen); 2029 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 2030 int family, u8 prefixlen, int l3index, 2031 struct tcp_md5sig_key *key); 2032 2033 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 2034 int family, u8 prefixlen, int l3index, u8 flags); 2035 void tcp_clear_md5_list(struct sock *sk); 2036 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 2037 const struct sock *addr_sk); 2038 2039 #ifdef CONFIG_TCP_MD5SIG 2040 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 2041 const union tcp_md5_addr *addr, 2042 int family, bool any_l3index); 2043 static inline struct tcp_md5sig_key * 2044 tcp_md5_do_lookup(const struct sock *sk, int l3index, 2045 const union tcp_md5_addr *addr, int family) 2046 { 2047 if (!static_branch_unlikely(&tcp_md5_needed.key)) 2048 return NULL; 2049 return __tcp_md5_do_lookup(sk, l3index, addr, family, false); 2050 } 2051 2052 static inline struct tcp_md5sig_key * 2053 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 2054 const union tcp_md5_addr *addr, int family) 2055 { 2056 if (!static_branch_unlikely(&tcp_md5_needed.key)) 2057 return NULL; 2058 return __tcp_md5_do_lookup(sk, 0, addr, family, true); 2059 } 2060 2061 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 2062 void tcp_md5_destruct_sock(struct sock *sk); 2063 #else 2064 static inline struct tcp_md5sig_key * 2065 tcp_md5_do_lookup(const struct sock *sk, int l3index, 2066 const union tcp_md5_addr *addr, int family) 2067 { 2068 return NULL; 2069 } 2070 2071 static inline struct tcp_md5sig_key * 2072 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 2073 const union tcp_md5_addr *addr, int family) 2074 { 2075 return NULL; 2076 } 2077 2078 #define tcp_twsk_md5_key(twsk) NULL 2079 static inline void tcp_md5_destruct_sock(struct sock *sk) 2080 { 2081 } 2082 #endif 2083 2084 struct md5_ctx; 2085 void tcp_md5_hash_skb_data(struct md5_ctx *ctx, const struct sk_buff *skb, 2086 unsigned int header_len); 2087 void tcp_md5_hash_key(struct md5_ctx *ctx, const struct tcp_md5sig_key *key); 2088 2089 /* From tcp_fastopen.c */ 2090 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 2091 struct tcp_fastopen_cookie *cookie); 2092 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 2093 struct tcp_fastopen_cookie *cookie, bool syn_lost, 2094 u16 try_exp); 2095 struct tcp_fastopen_request { 2096 /* Fast Open cookie. Size 0 means a cookie request */ 2097 struct tcp_fastopen_cookie cookie; 2098 struct msghdr *data; /* data in MSG_FASTOPEN */ 2099 size_t size; 2100 int copied; /* queued in tcp_connect() */ 2101 struct ubuf_info *uarg; 2102 }; 2103 void tcp_free_fastopen_req(struct tcp_sock *tp); 2104 void tcp_fastopen_destroy_cipher(struct sock *sk); 2105 void tcp_fastopen_ctx_destroy(struct net *net); 2106 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 2107 void *primary_key, void *backup_key); 2108 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 2109 u64 *key); 2110 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 2111 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 2112 struct request_sock *req, 2113 struct tcp_fastopen_cookie *foc, 2114 const struct dst_entry *dst); 2115 void tcp_fastopen_init_key_once(struct net *net); 2116 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 2117 struct tcp_fastopen_cookie *cookie); 2118 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 2119 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 2120 #define TCP_FASTOPEN_KEY_MAX 2 2121 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 2122 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 2123 2124 /* Fastopen key context */ 2125 struct tcp_fastopen_context { 2126 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 2127 int num; 2128 struct rcu_head rcu; 2129 }; 2130 2131 void tcp_fastopen_active_disable(struct sock *sk); 2132 bool tcp_fastopen_active_should_disable(struct sock *sk); 2133 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 2134 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 2135 2136 /* Caller needs to wrap with rcu_read_(un)lock() */ 2137 static inline 2138 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 2139 { 2140 struct tcp_fastopen_context *ctx; 2141 2142 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 2143 if (!ctx) 2144 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 2145 return ctx; 2146 } 2147 2148 static inline 2149 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 2150 const struct tcp_fastopen_cookie *orig) 2151 { 2152 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 2153 orig->len == foc->len && 2154 !memcmp(orig->val, foc->val, foc->len)) 2155 return true; 2156 return false; 2157 } 2158 2159 static inline 2160 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 2161 { 2162 return ctx->num; 2163 } 2164 2165 /* Latencies incurred by various limits for a sender. They are 2166 * chronograph-like stats that are mutually exclusive. 2167 */ 2168 enum tcp_chrono { 2169 TCP_CHRONO_UNSPEC, 2170 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 2171 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 2172 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 2173 __TCP_CHRONO_MAX, 2174 }; 2175 2176 static inline void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new) 2177 { 2178 const u32 now = tcp_jiffies32; 2179 enum tcp_chrono old = tp->chrono_type; 2180 2181 if (old > TCP_CHRONO_UNSPEC) 2182 tp->chrono_stat[old - 1] += now - tp->chrono_start; 2183 tp->chrono_start = now; 2184 tp->chrono_type = new; 2185 } 2186 2187 static inline void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type) 2188 { 2189 struct tcp_sock *tp = tcp_sk(sk); 2190 2191 /* If there are multiple conditions worthy of tracking in a 2192 * chronograph then the highest priority enum takes precedence 2193 * over the other conditions. So that if something "more interesting" 2194 * starts happening, stop the previous chrono and start a new one. 2195 */ 2196 if (type > tp->chrono_type) 2197 tcp_chrono_set(tp, type); 2198 } 2199 2200 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 2201 2202 /* This helper is needed, because skb->tcp_tsorted_anchor uses 2203 * the same memory storage than skb->destructor/_skb_refdst 2204 */ 2205 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 2206 { 2207 skb->destructor = NULL; 2208 skb->_skb_refdst = 0UL; 2209 } 2210 2211 #define tcp_skb_tsorted_save(skb) { \ 2212 unsigned long _save = skb->_skb_refdst; \ 2213 skb->_skb_refdst = 0UL; 2214 2215 #define tcp_skb_tsorted_restore(skb) \ 2216 skb->_skb_refdst = _save; \ 2217 } 2218 2219 void tcp_write_queue_purge(struct sock *sk); 2220 2221 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 2222 { 2223 return skb_rb_first(&sk->tcp_rtx_queue); 2224 } 2225 2226 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 2227 { 2228 return skb_rb_last(&sk->tcp_rtx_queue); 2229 } 2230 2231 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 2232 { 2233 return skb_peek_tail(&sk->sk_write_queue); 2234 } 2235 2236 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 2237 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 2238 2239 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 2240 { 2241 return skb_peek(&sk->sk_write_queue); 2242 } 2243 2244 static inline bool tcp_skb_is_last(const struct sock *sk, 2245 const struct sk_buff *skb) 2246 { 2247 return skb_queue_is_last(&sk->sk_write_queue, skb); 2248 } 2249 2250 /** 2251 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 2252 * @sk: socket 2253 * 2254 * Since the write queue can have a temporary empty skb in it, 2255 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 2256 */ 2257 static inline bool tcp_write_queue_empty(const struct sock *sk) 2258 { 2259 const struct tcp_sock *tp = tcp_sk(sk); 2260 2261 return tp->write_seq == tp->snd_nxt; 2262 } 2263 2264 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 2265 { 2266 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 2267 } 2268 2269 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 2270 { 2271 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 2272 } 2273 2274 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 2275 { 2276 __skb_queue_tail(&sk->sk_write_queue, skb); 2277 2278 /* Queue it, remembering where we must start sending. */ 2279 if (sk->sk_write_queue.next == skb) 2280 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 2281 } 2282 2283 /* Insert new before skb on the write queue of sk. */ 2284 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 2285 struct sk_buff *skb, 2286 struct sock *sk) 2287 { 2288 __skb_queue_before(&sk->sk_write_queue, skb, new); 2289 } 2290 2291 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 2292 { 2293 tcp_skb_tsorted_anchor_cleanup(skb); 2294 __skb_unlink(skb, &sk->sk_write_queue); 2295 } 2296 2297 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 2298 2299 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2300 { 2301 tcp_skb_tsorted_anchor_cleanup(skb); 2302 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2303 } 2304 2305 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2306 { 2307 list_del(&skb->tcp_tsorted_anchor); 2308 tcp_rtx_queue_unlink(skb, sk); 2309 tcp_wmem_free_skb(sk, skb); 2310 } 2311 2312 static inline void tcp_write_collapse_fence(struct sock *sk) 2313 { 2314 struct sk_buff *skb = tcp_write_queue_tail(sk); 2315 2316 if (skb) 2317 TCP_SKB_CB(skb)->eor = 1; 2318 } 2319 2320 static inline void tcp_push_pending_frames(struct sock *sk) 2321 { 2322 if (tcp_send_head(sk)) { 2323 struct tcp_sock *tp = tcp_sk(sk); 2324 2325 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2326 } 2327 } 2328 2329 /* Start sequence of the skb just after the highest skb with SACKed 2330 * bit, valid only if sacked_out > 0 or when the caller has ensured 2331 * validity by itself. 2332 */ 2333 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2334 { 2335 if (!tp->sacked_out) 2336 return tp->snd_una; 2337 2338 if (tp->highest_sack == NULL) 2339 return tp->snd_nxt; 2340 2341 return TCP_SKB_CB(tp->highest_sack)->seq; 2342 } 2343 2344 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2345 { 2346 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2347 } 2348 2349 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2350 { 2351 return tcp_sk(sk)->highest_sack; 2352 } 2353 2354 static inline void tcp_highest_sack_reset(struct sock *sk) 2355 { 2356 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2357 } 2358 2359 /* Called when old skb is about to be deleted and replaced by new skb */ 2360 static inline void tcp_highest_sack_replace(struct sock *sk, 2361 struct sk_buff *old, 2362 struct sk_buff *new) 2363 { 2364 if (old == tcp_highest_sack(sk)) 2365 tcp_sk(sk)->highest_sack = new; 2366 } 2367 2368 /* This helper checks if socket has IP_TRANSPARENT set */ 2369 static inline bool inet_sk_transparent(const struct sock *sk) 2370 { 2371 switch (sk->sk_state) { 2372 case TCP_TIME_WAIT: 2373 return inet_twsk(sk)->tw_transparent; 2374 case TCP_NEW_SYN_RECV: 2375 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2376 } 2377 return inet_test_bit(TRANSPARENT, sk); 2378 } 2379 2380 /* Determines whether this is a thin stream (which may suffer from 2381 * increased latency). Used to trigger latency-reducing mechanisms. 2382 */ 2383 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2384 { 2385 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2386 } 2387 2388 /* /proc */ 2389 enum tcp_seq_states { 2390 TCP_SEQ_STATE_LISTENING, 2391 TCP_SEQ_STATE_ESTABLISHED, 2392 }; 2393 2394 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2395 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2396 void tcp_seq_stop(struct seq_file *seq, void *v); 2397 2398 struct tcp_seq_afinfo { 2399 sa_family_t family; 2400 }; 2401 2402 struct tcp_iter_state { 2403 struct seq_net_private p; 2404 enum tcp_seq_states state; 2405 struct sock *syn_wait_sk; 2406 int bucket, offset, sbucket, num; 2407 loff_t last_pos; 2408 }; 2409 2410 extern struct request_sock_ops tcp_request_sock_ops; 2411 extern struct request_sock_ops tcp6_request_sock_ops; 2412 2413 void tcp_v4_destroy_sock(struct sock *sk); 2414 2415 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2416 netdev_features_t features); 2417 struct sk_buff *tcp_gro_lookup(struct list_head *head, struct tcphdr *th); 2418 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb, 2419 struct tcphdr *th); 2420 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2421 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2422 #ifdef CONFIG_INET 2423 void tcp_gro_complete(struct sk_buff *skb); 2424 #else 2425 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2426 #endif 2427 2428 static inline void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, 2429 __be32 daddr) 2430 { 2431 struct tcphdr *th = tcp_hdr(skb); 2432 2433 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); 2434 skb->csum_start = skb_transport_header(skb) - skb->head; 2435 skb->csum_offset = offsetof(struct tcphdr, check); 2436 } 2437 2438 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2439 { 2440 struct net *net = sock_net((struct sock *)tp); 2441 u32 val; 2442 2443 val = READ_ONCE(tp->notsent_lowat); 2444 2445 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2446 } 2447 2448 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2449 2450 #ifdef CONFIG_PROC_FS 2451 int tcp4_proc_init(void); 2452 void tcp4_proc_exit(void); 2453 #endif 2454 2455 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2456 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2457 const struct tcp_request_sock_ops *af_ops, 2458 struct sock *sk, struct sk_buff *skb); 2459 2460 /* TCP af-specific functions */ 2461 struct tcp_sock_af_ops { 2462 #ifdef CONFIG_TCP_MD5SIG 2463 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2464 const struct sock *addr_sk); 2465 void (*calc_md5_hash)(char *location, 2466 const struct tcp_md5sig_key *md5, 2467 const struct sock *sk, 2468 const struct sk_buff *skb); 2469 int (*md5_parse)(struct sock *sk, 2470 int optname, 2471 sockptr_t optval, 2472 int optlen); 2473 #endif 2474 #ifdef CONFIG_TCP_AO 2475 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen); 2476 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2477 struct sock *addr_sk, 2478 int sndid, int rcvid); 2479 int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key, 2480 const struct sock *sk, 2481 __be32 sisn, __be32 disn, bool send); 2482 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao, 2483 const struct sock *sk, const struct sk_buff *skb, 2484 const u8 *tkey, int hash_offset, u32 sne); 2485 #endif 2486 }; 2487 2488 struct tcp_request_sock_ops { 2489 u16 mss_clamp; 2490 #ifdef CONFIG_TCP_MD5SIG 2491 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2492 const struct sock *addr_sk); 2493 void (*calc_md5_hash) (char *location, 2494 const struct tcp_md5sig_key *md5, 2495 const struct sock *sk, 2496 const struct sk_buff *skb); 2497 #endif 2498 #ifdef CONFIG_TCP_AO 2499 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2500 struct request_sock *req, 2501 int sndid, int rcvid); 2502 int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk); 2503 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt, 2504 struct request_sock *req, const struct sk_buff *skb, 2505 int hash_offset, u32 sne); 2506 #endif 2507 #ifdef CONFIG_SYN_COOKIES 2508 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2509 __u16 *mss); 2510 #endif 2511 struct dst_entry *(*route_req)(const struct sock *sk, 2512 struct sk_buff *skb, 2513 struct flowi *fl, 2514 struct request_sock *req, 2515 u32 tw_isn); 2516 union tcp_seq_and_ts_off (*init_seq_and_ts_off)( 2517 const struct net *net, 2518 const struct sk_buff *skb); 2519 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2520 struct flowi *fl, struct request_sock *req, 2521 struct tcp_fastopen_cookie *foc, 2522 enum tcp_synack_type synack_type, 2523 struct sk_buff *syn_skb); 2524 }; 2525 2526 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2527 #if IS_ENABLED(CONFIG_IPV6) 2528 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2529 #endif 2530 2531 #ifdef CONFIG_SYN_COOKIES 2532 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2533 const struct sock *sk, struct sk_buff *skb, 2534 __u16 *mss) 2535 { 2536 tcp_synq_overflow(sk); 2537 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2538 return ops->cookie_init_seq(skb, mss); 2539 } 2540 #else 2541 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2542 const struct sock *sk, struct sk_buff *skb, 2543 __u16 *mss) 2544 { 2545 return 0; 2546 } 2547 #endif 2548 2549 struct tcp_key { 2550 union { 2551 struct { 2552 struct tcp_ao_key *ao_key; 2553 char *traffic_key; 2554 u32 sne; 2555 u8 rcv_next; 2556 }; 2557 struct tcp_md5sig_key *md5_key; 2558 }; 2559 enum { 2560 TCP_KEY_NONE = 0, 2561 TCP_KEY_MD5, 2562 TCP_KEY_AO, 2563 } type; 2564 }; 2565 2566 static inline void tcp_get_current_key(const struct sock *sk, 2567 struct tcp_key *out) 2568 { 2569 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG) 2570 const struct tcp_sock *tp = tcp_sk(sk); 2571 #endif 2572 2573 #ifdef CONFIG_TCP_AO 2574 if (static_branch_unlikely(&tcp_ao_needed.key)) { 2575 struct tcp_ao_info *ao; 2576 2577 ao = rcu_dereference_protected(tp->ao_info, 2578 lockdep_sock_is_held(sk)); 2579 if (ao) { 2580 out->ao_key = READ_ONCE(ao->current_key); 2581 out->type = TCP_KEY_AO; 2582 return; 2583 } 2584 } 2585 #endif 2586 #ifdef CONFIG_TCP_MD5SIG 2587 if (static_branch_unlikely(&tcp_md5_needed.key) && 2588 rcu_access_pointer(tp->md5sig_info)) { 2589 out->md5_key = tp->af_specific->md5_lookup(sk, sk); 2590 if (out->md5_key) { 2591 out->type = TCP_KEY_MD5; 2592 return; 2593 } 2594 } 2595 #endif 2596 out->type = TCP_KEY_NONE; 2597 } 2598 2599 static inline bool tcp_key_is_md5(const struct tcp_key *key) 2600 { 2601 if (static_branch_tcp_md5()) 2602 return key->type == TCP_KEY_MD5; 2603 return false; 2604 } 2605 2606 static inline bool tcp_key_is_ao(const struct tcp_key *key) 2607 { 2608 if (static_branch_tcp_ao()) 2609 return key->type == TCP_KEY_AO; 2610 return false; 2611 } 2612 2613 int tcpv4_offload_init(void); 2614 2615 void tcp_v4_init(void); 2616 void tcp_init(void); 2617 2618 /* tcp_recovery.c */ 2619 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2620 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2621 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2622 u32 reo_wnd); 2623 extern bool tcp_rack_mark_lost(struct sock *sk); 2624 extern void tcp_rack_reo_timeout(struct sock *sk); 2625 2626 /* tcp_plb.c */ 2627 2628 /* 2629 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2630 * expects cong_ratio which represents fraction of traffic that experienced 2631 * congestion over a single RTT. In order to avoid floating point operations, 2632 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2633 */ 2634 #define TCP_PLB_SCALE 8 2635 2636 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2637 struct tcp_plb_state { 2638 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2639 unused:3; 2640 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2641 }; 2642 2643 static inline void tcp_plb_init(const struct sock *sk, 2644 struct tcp_plb_state *plb) 2645 { 2646 plb->consec_cong_rounds = 0; 2647 plb->pause_until = 0; 2648 } 2649 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2650 const int cong_ratio); 2651 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2652 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2653 2654 static inline void tcp_warn_once(const struct sock *sk, bool cond, const char *str) 2655 { 2656 WARN_ONCE(cond, 2657 "%scwn:%u out:%u sacked:%u lost:%u retrans:%u tlp_high_seq:%u sk_state:%u ca_state:%u advmss:%u mss_cache:%u pmtu:%u\n", 2658 str, 2659 tcp_snd_cwnd(tcp_sk(sk)), 2660 tcp_sk(sk)->packets_out, tcp_sk(sk)->sacked_out, 2661 tcp_sk(sk)->lost_out, tcp_sk(sk)->retrans_out, 2662 tcp_sk(sk)->tlp_high_seq, sk->sk_state, 2663 inet_csk(sk)->icsk_ca_state, 2664 tcp_sk(sk)->advmss, tcp_sk(sk)->mss_cache, 2665 inet_csk(sk)->icsk_pmtu_cookie); 2666 } 2667 2668 /* At how many usecs into the future should the RTO fire? */ 2669 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2670 { 2671 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2672 u32 rto = inet_csk(sk)->icsk_rto; 2673 2674 if (likely(skb)) { 2675 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2676 2677 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2678 } else { 2679 tcp_warn_once(sk, 1, "rtx queue empty: "); 2680 return jiffies_to_usecs(rto); 2681 } 2682 2683 } 2684 2685 /* 2686 * Save and compile IPv4 options, return a pointer to it 2687 */ 2688 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2689 struct sk_buff *skb) 2690 { 2691 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2692 struct ip_options_rcu *dopt = NULL; 2693 2694 if (opt->optlen) { 2695 int opt_size = sizeof(*dopt) + opt->optlen; 2696 2697 dopt = kmalloc(opt_size, GFP_ATOMIC); 2698 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2699 kfree(dopt); 2700 dopt = NULL; 2701 } 2702 } 2703 return dopt; 2704 } 2705 2706 /* locally generated TCP pure ACKs have skb->truesize == 2 2707 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2708 * This is much faster than dissecting the packet to find out. 2709 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2710 */ 2711 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2712 { 2713 return skb->truesize == 2; 2714 } 2715 2716 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2717 { 2718 skb->truesize = 2; 2719 } 2720 2721 static inline int tcp_inq(struct sock *sk) 2722 { 2723 struct tcp_sock *tp = tcp_sk(sk); 2724 int answ; 2725 2726 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2727 answ = 0; 2728 } else if (sock_flag(sk, SOCK_URGINLINE) || 2729 !tp->urg_data || 2730 before(tp->urg_seq, tp->copied_seq) || 2731 !before(tp->urg_seq, tp->rcv_nxt)) { 2732 2733 answ = tp->rcv_nxt - tp->copied_seq; 2734 2735 /* Subtract 1, if FIN was received */ 2736 if (answ && sock_flag(sk, SOCK_DONE)) 2737 answ--; 2738 } else { 2739 answ = tp->urg_seq - tp->copied_seq; 2740 } 2741 2742 return answ; 2743 } 2744 2745 int tcp_peek_len(struct socket *sock); 2746 2747 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2748 { 2749 u16 segs_in; 2750 2751 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2752 2753 /* We update these fields while other threads might 2754 * read them from tcp_get_info() 2755 */ 2756 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2757 if (skb->len > tcp_hdrlen(skb)) 2758 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2759 } 2760 2761 /* 2762 * TCP listen path runs lockless. 2763 * We forced "struct sock" to be const qualified to make sure 2764 * we don't modify one of its field by mistake. 2765 * Here, we increment sk_drops which is an atomic_t, so we can safely 2766 * make sock writable again. 2767 */ 2768 static inline void tcp_listendrop(const struct sock *sk) 2769 { 2770 sk_drops_inc((struct sock *)sk); 2771 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2772 } 2773 2774 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2775 2776 /* 2777 * Interface for adding Upper Level Protocols over TCP 2778 */ 2779 2780 #define TCP_ULP_NAME_MAX 16 2781 #define TCP_ULP_MAX 128 2782 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2783 2784 struct tcp_ulp_ops { 2785 struct list_head list; 2786 2787 /* initialize ulp */ 2788 int (*init)(struct sock *sk); 2789 /* update ulp */ 2790 void (*update)(struct sock *sk, struct proto *p, 2791 void (*write_space)(struct sock *sk)); 2792 /* cleanup ulp */ 2793 void (*release)(struct sock *sk); 2794 /* diagnostic */ 2795 int (*get_info)(struct sock *sk, struct sk_buff *skb, bool net_admin); 2796 size_t (*get_info_size)(const struct sock *sk, bool net_admin); 2797 /* clone ulp */ 2798 void (*clone)(const struct request_sock *req, struct sock *newsk, 2799 const gfp_t priority); 2800 2801 char name[TCP_ULP_NAME_MAX]; 2802 struct module *owner; 2803 }; 2804 int tcp_register_ulp(struct tcp_ulp_ops *type); 2805 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2806 int tcp_set_ulp(struct sock *sk, const char *name); 2807 void tcp_get_available_ulp(char *buf, size_t len); 2808 void tcp_cleanup_ulp(struct sock *sk); 2809 void tcp_update_ulp(struct sock *sk, struct proto *p, 2810 void (*write_space)(struct sock *sk)); 2811 2812 #define MODULE_ALIAS_TCP_ULP(name) \ 2813 MODULE_INFO(alias, name); \ 2814 MODULE_INFO(alias, "tcp-ulp-" name) 2815 2816 #ifdef CONFIG_NET_SOCK_MSG 2817 struct sk_msg; 2818 struct sk_psock; 2819 2820 #ifdef CONFIG_BPF_SYSCALL 2821 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2822 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2823 #ifdef CONFIG_BPF_STREAM_PARSER 2824 struct strparser; 2825 int tcp_bpf_strp_read_sock(struct strparser *strp, read_descriptor_t *desc, 2826 sk_read_actor_t recv_actor); 2827 #endif /* CONFIG_BPF_STREAM_PARSER */ 2828 #endif /* CONFIG_BPF_SYSCALL */ 2829 2830 #ifdef CONFIG_INET 2831 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2832 #else 2833 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2834 { 2835 } 2836 #endif 2837 2838 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2839 struct sk_msg *msg, u32 bytes, int flags); 2840 #endif /* CONFIG_NET_SOCK_MSG */ 2841 2842 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2843 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2844 { 2845 } 2846 #endif 2847 2848 #ifdef CONFIG_CGROUP_BPF 2849 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2850 struct sk_buff *skb, 2851 unsigned int end_offset) 2852 { 2853 skops->skb = skb; 2854 skops->skb_data_end = skb->data + end_offset; 2855 } 2856 #else 2857 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2858 struct sk_buff *skb, 2859 unsigned int end_offset) 2860 { 2861 } 2862 #endif 2863 2864 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2865 * is < 0, then the BPF op failed (for example if the loaded BPF 2866 * program does not support the chosen operation or there is no BPF 2867 * program loaded). 2868 */ 2869 #ifdef CONFIG_BPF 2870 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2871 { 2872 struct bpf_sock_ops_kern sock_ops; 2873 int ret; 2874 2875 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2876 if (sk_fullsock(sk)) { 2877 sock_ops.is_fullsock = 1; 2878 sock_ops.is_locked_tcp_sock = 1; 2879 sock_owned_by_me(sk); 2880 } 2881 2882 sock_ops.sk = sk; 2883 sock_ops.op = op; 2884 if (nargs > 0) 2885 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2886 2887 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2888 if (ret == 0) 2889 ret = sock_ops.reply; 2890 else 2891 ret = -1; 2892 return ret; 2893 } 2894 2895 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2896 { 2897 u32 args[2] = {arg1, arg2}; 2898 2899 return tcp_call_bpf(sk, op, 2, args); 2900 } 2901 2902 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2903 u32 arg3) 2904 { 2905 u32 args[3] = {arg1, arg2, arg3}; 2906 2907 return tcp_call_bpf(sk, op, 3, args); 2908 } 2909 2910 #else 2911 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2912 { 2913 return -EPERM; 2914 } 2915 2916 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2917 { 2918 return -EPERM; 2919 } 2920 2921 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2922 u32 arg3) 2923 { 2924 return -EPERM; 2925 } 2926 2927 #endif 2928 2929 static inline u32 tcp_timeout_init(struct sock *sk) 2930 { 2931 int timeout; 2932 2933 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2934 2935 if (timeout <= 0) 2936 timeout = TCP_TIMEOUT_INIT; 2937 return min_t(int, timeout, TCP_RTO_MAX); 2938 } 2939 2940 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2941 { 2942 int rwnd; 2943 2944 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2945 2946 if (rwnd < 0) 2947 rwnd = 0; 2948 return rwnd; 2949 } 2950 2951 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2952 { 2953 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2954 } 2955 2956 static inline void tcp_bpf_rtt(struct sock *sk, long mrtt, u32 srtt) 2957 { 2958 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2959 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_RTT_CB, mrtt, srtt); 2960 } 2961 2962 #if IS_ENABLED(CONFIG_SMC) 2963 extern struct static_key_false tcp_have_smc; 2964 #endif 2965 2966 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2967 void clean_acked_data_enable(struct tcp_sock *tp, 2968 void (*cad)(struct sock *sk, u32 ack_seq)); 2969 void clean_acked_data_disable(struct tcp_sock *tp); 2970 void clean_acked_data_flush(void); 2971 #endif 2972 2973 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2974 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2975 const struct tcp_sock *tp) 2976 { 2977 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2978 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2979 } 2980 2981 /* Compute Earliest Departure Time for some control packets 2982 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2983 */ 2984 static inline u64 tcp_transmit_time(const struct sock *sk) 2985 { 2986 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2987 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2988 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2989 2990 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2991 } 2992 return 0; 2993 } 2994 2995 static inline int tcp_parse_auth_options(const struct tcphdr *th, 2996 const u8 **md5_hash, const struct tcp_ao_hdr **aoh) 2997 { 2998 const u8 *md5_tmp, *ao_tmp; 2999 int ret; 3000 3001 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp); 3002 if (ret) 3003 return ret; 3004 3005 if (md5_hash) 3006 *md5_hash = md5_tmp; 3007 3008 if (aoh) { 3009 if (!ao_tmp) 3010 *aoh = NULL; 3011 else 3012 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2); 3013 } 3014 3015 return 0; 3016 } 3017 3018 static inline bool tcp_ao_required(struct sock *sk, const void *saddr, 3019 int family, int l3index, bool stat_inc) 3020 { 3021 #ifdef CONFIG_TCP_AO 3022 struct tcp_ao_info *ao_info; 3023 struct tcp_ao_key *ao_key; 3024 3025 if (!static_branch_unlikely(&tcp_ao_needed.key)) 3026 return false; 3027 3028 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info, 3029 lockdep_sock_is_held(sk)); 3030 if (!ao_info) 3031 return false; 3032 3033 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1); 3034 if (ao_info->ao_required || ao_key) { 3035 if (stat_inc) { 3036 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED); 3037 atomic64_inc(&ao_info->counters.ao_required); 3038 } 3039 return true; 3040 } 3041 #endif 3042 return false; 3043 } 3044 3045 enum skb_drop_reason tcp_inbound_hash(struct sock *sk, 3046 const struct request_sock *req, const struct sk_buff *skb, 3047 const void *saddr, const void *daddr, 3048 int family, int dif, int sdif); 3049 3050 #endif /* _TCP_H */ 3051