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 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 #include <linux/bpf-cgroup.h> 50 #include <linux/siphash.h> 51 52 extern struct inet_hashinfo tcp_hashinfo; 53 54 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 55 int tcp_orphan_count_sum(void); 56 57 static inline void tcp_orphan_count_inc(void) 58 { 59 this_cpu_inc(tcp_orphan_count); 60 } 61 62 static inline void tcp_orphan_count_dec(void) 63 { 64 this_cpu_dec(tcp_orphan_count); 65 } 66 67 DECLARE_PER_CPU(u32, tcp_tw_isn); 68 69 void tcp_time_wait(struct sock *sk, int state, int timeo); 70 71 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 72 #define MAX_TCP_OPTION_SPACE 40 73 #define TCP_MIN_SND_MSS 48 74 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 75 76 /* 77 * Never offer a window over 32767 without using window scaling. Some 78 * poor stacks do signed 16bit maths! 79 */ 80 #define MAX_TCP_WINDOW 32767U 81 82 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 83 #define TCP_MIN_MSS 88U 84 85 /* The initial MTU to use for probing */ 86 #define TCP_BASE_MSS 1024 87 88 /* probing interval, default to 10 minutes as per RFC4821 */ 89 #define TCP_PROBE_INTERVAL 600 90 91 /* Specify interval when tcp mtu probing will stop */ 92 #define TCP_PROBE_THRESHOLD 8 93 94 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 95 #define TCP_FASTRETRANS_THRESH 3 96 97 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 98 #define TCP_MAX_QUICKACKS 16U 99 100 /* Maximal number of window scale according to RFC1323 */ 101 #define TCP_MAX_WSCALE 14U 102 103 /* Default sending frequency of accurate ECN option per RTT */ 104 #define TCP_ACCECN_OPTION_BEACON 3 105 106 /* urg_data states */ 107 #define TCP_URG_VALID 0x0100 108 #define TCP_URG_NOTYET 0x0200 109 #define TCP_URG_READ 0x0400 110 111 #define TCP_RETR1 3 /* 112 * This is how many retries it does before it 113 * tries to figure out if the gateway is 114 * down. Minimal RFC value is 3; it corresponds 115 * to ~3sec-8min depending on RTO. 116 */ 117 118 #define TCP_RETR2 15 /* 119 * This should take at least 120 * 90 minutes to time out. 121 * RFC1122 says that the limit is 100 sec. 122 * 15 is ~13-30min depending on RTO. 123 */ 124 125 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 126 * when active opening a connection. 127 * RFC1122 says the minimum retry MUST 128 * be at least 180secs. Nevertheless 129 * this value is corresponding to 130 * 63secs of retransmission with the 131 * current initial RTO. 132 */ 133 134 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 135 * when passive opening a connection. 136 * This is corresponding to 31secs of 137 * retransmission with the current 138 * initial RTO. 139 */ 140 141 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 142 * state, about 60 seconds */ 143 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 144 /* BSD style FIN_WAIT2 deadlock breaker. 145 * It used to be 3min, new value is 60sec, 146 * to combine FIN-WAIT-2 timeout with 147 * TIME-WAIT timer. 148 */ 149 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 150 151 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 152 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); 153 154 #if HZ >= 100 155 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 156 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 157 #else 158 #define TCP_DELACK_MIN 4U 159 #define TCP_ATO_MIN 4U 160 #endif 161 #define TCP_RTO_MAX_SEC 120 162 #define TCP_RTO_MAX ((unsigned)(TCP_RTO_MAX_SEC * HZ)) 163 #define TCP_RTO_MIN ((unsigned)(HZ / 5)) 164 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 165 166 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 167 168 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 169 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 170 * used as a fallback RTO for the 171 * initial data transmission if no 172 * valid RTT sample has been acquired, 173 * most likely due to retrans in 3WHS. 174 */ 175 176 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 177 * for local resources. 178 */ 179 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 180 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 181 #define TCP_KEEPALIVE_INTVL (75*HZ) 182 183 #define MAX_TCP_KEEPIDLE 32767 184 #define MAX_TCP_KEEPINTVL 32767 185 #define MAX_TCP_KEEPCNT 127 186 #define MAX_TCP_SYNCNT 127 187 188 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds 189 * to avoid overflows. This assumes a clock smaller than 1 Mhz. 190 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz. 191 */ 192 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC) 193 194 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 195 * after this time. It should be equal 196 * (or greater than) TCP_TIMEWAIT_LEN 197 * to provide reliability equal to one 198 * provided by timewait state. 199 */ 200 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 201 * timestamps. It must be less than 202 * minimal timewait lifetime. 203 */ 204 /* 205 * TCP option 206 */ 207 208 #define TCPOPT_NOP 1 /* Padding */ 209 #define TCPOPT_EOL 0 /* End of options */ 210 #define TCPOPT_MSS 2 /* Segment size negotiating */ 211 #define TCPOPT_WINDOW 3 /* Window scaling */ 212 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 213 #define TCPOPT_SACK 5 /* SACK Block */ 214 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 215 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 216 #define TCPOPT_AO 29 /* Authentication Option (RFC5925) */ 217 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 218 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 219 #define TCPOPT_ACCECN0 172 /* 0xAC: Accurate ECN Order 0 */ 220 #define TCPOPT_ACCECN1 174 /* 0xAE: Accurate ECN Order 1 */ 221 #define TCPOPT_EXP 254 /* Experimental */ 222 /* Magic number to be after the option value for sharing TCP 223 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 224 */ 225 #define TCPOPT_FASTOPEN_MAGIC 0xF989 226 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 227 228 /* 229 * TCP option lengths 230 */ 231 232 #define TCPOLEN_MSS 4 233 #define TCPOLEN_WINDOW 3 234 #define TCPOLEN_SACK_PERM 2 235 #define TCPOLEN_TIMESTAMP 10 236 #define TCPOLEN_MD5SIG 18 237 #define TCPOLEN_FASTOPEN_BASE 2 238 #define TCPOLEN_ACCECN_BASE 2 239 #define TCPOLEN_EXP_FASTOPEN_BASE 4 240 #define TCPOLEN_EXP_SMC_BASE 6 241 242 /* But this is what stacks really send out. */ 243 #define TCPOLEN_TSTAMP_ALIGNED 12 244 #define TCPOLEN_WSCALE_ALIGNED 4 245 #define TCPOLEN_SACKPERM_ALIGNED 4 246 #define TCPOLEN_SACK_BASE 2 247 #define TCPOLEN_SACK_BASE_ALIGNED 4 248 #define TCPOLEN_SACK_PERBLOCK 8 249 #define TCPOLEN_MD5SIG_ALIGNED 20 250 #define TCPOLEN_MSS_ALIGNED 4 251 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 252 #define TCPOLEN_ACCECN_PERFIELD 3 253 254 /* Maximum number of byte counters in AccECN option + size */ 255 #define TCP_ACCECN_NUMFIELDS 3 256 #define TCP_ACCECN_MAXSIZE (TCPOLEN_ACCECN_BASE + \ 257 TCPOLEN_ACCECN_PERFIELD * \ 258 TCP_ACCECN_NUMFIELDS) 259 #define TCP_ACCECN_SAFETY_SHIFT 1 /* SAFETY_FACTOR in accecn draft */ 260 261 /* Flags in tp->nonagle */ 262 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 263 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 264 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 265 266 /* TCP thin-stream limits */ 267 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 268 269 /* TCP initial congestion window as per rfc6928 */ 270 #define TCP_INIT_CWND 10 271 272 /* Bit Flags for sysctl_tcp_fastopen */ 273 #define TFO_CLIENT_ENABLE 1 274 #define TFO_SERVER_ENABLE 2 275 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 276 277 /* Accept SYN data w/o any cookie option */ 278 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 279 280 /* Force enable TFO on all listeners, i.e., not requiring the 281 * TCP_FASTOPEN socket option. 282 */ 283 #define TFO_SERVER_WO_SOCKOPT1 0x400 284 285 286 /* sysctl variables for tcp */ 287 extern int sysctl_tcp_max_orphans; 288 extern long sysctl_tcp_mem[3]; 289 290 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 291 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 292 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 293 294 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 295 296 extern struct percpu_counter tcp_sockets_allocated; 297 extern unsigned long tcp_memory_pressure; 298 299 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 300 static inline bool tcp_under_memory_pressure(const struct sock *sk) 301 { 302 if (mem_cgroup_sk_enabled(sk) && 303 mem_cgroup_sk_under_memory_pressure(sk)) 304 return true; 305 306 if (sk->sk_bypass_prot_mem) 307 return false; 308 309 return READ_ONCE(tcp_memory_pressure); 310 } 311 /* 312 * The next routines deal with comparing 32 bit unsigned ints 313 * and worry about wraparound (automatic with unsigned arithmetic). 314 */ 315 316 static inline bool before(__u32 seq1, __u32 seq2) 317 { 318 return (__s32)(seq1-seq2) < 0; 319 } 320 #define after(seq2, seq1) before(seq1, seq2) 321 322 /* is s2<=s1<=s3 ? */ 323 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 324 { 325 return seq3 - seq2 >= seq1 - seq2; 326 } 327 328 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 329 { 330 sk_wmem_queued_add(sk, -skb->truesize); 331 if (!skb_zcopy_pure(skb)) 332 sk_mem_uncharge(sk, skb->truesize); 333 else 334 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 335 __kfree_skb(skb); 336 } 337 338 void sk_forced_mem_schedule(struct sock *sk, int size); 339 340 bool tcp_check_oom(const struct sock *sk, int shift); 341 342 343 extern struct proto tcp_prot; 344 345 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 346 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 347 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 348 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 349 350 /* 351 * TCP splice context 352 */ 353 struct tcp_splice_state { 354 struct pipe_inode_info *pipe; 355 size_t len; 356 unsigned int flags; 357 }; 358 359 void tcp_tsq_work_init(void); 360 361 int tcp_v4_err(struct sk_buff *skb, u32); 362 363 void tcp_shutdown(struct sock *sk, int how); 364 365 int tcp_v4_early_demux(struct sk_buff *skb); 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 void tcp_release_cb(struct sock *sk); 379 void tcp_wfree(struct sk_buff *skb); 380 void tcp_write_timer_handler(struct sock *sk); 381 void tcp_delack_timer_handler(struct sock *sk); 382 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 383 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 384 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 385 void tcp_rcvbuf_grow(struct sock *sk, u32 newval); 386 void tcp_rcv_space_adjust(struct sock *sk); 387 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 388 void tcp_twsk_destructor(struct sock *sk); 389 void tcp_twsk_purge(struct list_head *net_exit_list); 390 int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, 391 unsigned int offset, size_t len); 392 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 393 struct pipe_inode_info *pipe, size_t len, 394 unsigned int flags); 395 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 396 bool force_schedule); 397 398 static inline void tcp_dec_quickack_mode(struct sock *sk) 399 { 400 struct inet_connection_sock *icsk = inet_csk(sk); 401 402 if (icsk->icsk_ack.quick) { 403 /* How many ACKs S/ACKing new data have we sent? */ 404 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 405 406 if (pkts >= icsk->icsk_ack.quick) { 407 icsk->icsk_ack.quick = 0; 408 /* Leaving quickack mode we deflate ATO. */ 409 icsk->icsk_ack.ato = TCP_ATO_MIN; 410 } else 411 icsk->icsk_ack.quick -= pkts; 412 } 413 } 414 415 #define TCP_ECN_MODE_RFC3168 BIT(0) 416 #define TCP_ECN_QUEUE_CWR BIT(1) 417 #define TCP_ECN_DEMAND_CWR BIT(2) 418 #define TCP_ECN_SEEN BIT(3) 419 #define TCP_ECN_MODE_ACCECN BIT(4) 420 421 #define TCP_ECN_DISABLED 0 422 #define TCP_ECN_MODE_PENDING (TCP_ECN_MODE_RFC3168 | TCP_ECN_MODE_ACCECN) 423 #define TCP_ECN_MODE_ANY (TCP_ECN_MODE_RFC3168 | TCP_ECN_MODE_ACCECN) 424 425 static inline bool tcp_ecn_mode_any(const struct tcp_sock *tp) 426 { 427 return tp->ecn_flags & TCP_ECN_MODE_ANY; 428 } 429 430 static inline bool tcp_ecn_mode_rfc3168(const struct tcp_sock *tp) 431 { 432 return (tp->ecn_flags & TCP_ECN_MODE_ANY) == TCP_ECN_MODE_RFC3168; 433 } 434 435 static inline bool tcp_ecn_mode_accecn(const struct tcp_sock *tp) 436 { 437 return (tp->ecn_flags & TCP_ECN_MODE_ANY) == TCP_ECN_MODE_ACCECN; 438 } 439 440 static inline bool tcp_ecn_disabled(const struct tcp_sock *tp) 441 { 442 return !tcp_ecn_mode_any(tp); 443 } 444 445 static inline bool tcp_ecn_mode_pending(const struct tcp_sock *tp) 446 { 447 return (tp->ecn_flags & TCP_ECN_MODE_PENDING) == TCP_ECN_MODE_PENDING; 448 } 449 450 static inline void tcp_ecn_mode_set(struct tcp_sock *tp, u8 mode) 451 { 452 tp->ecn_flags &= ~TCP_ECN_MODE_ANY; 453 tp->ecn_flags |= mode; 454 } 455 456 enum tcp_tw_status { 457 TCP_TW_SUCCESS = 0, 458 TCP_TW_RST = 1, 459 TCP_TW_ACK = 2, 460 TCP_TW_SYN = 3, 461 TCP_TW_ACK_OOW = 4 462 }; 463 464 465 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 466 struct sk_buff *skb, 467 const struct tcphdr *th, 468 u32 *tw_isn, 469 enum skb_drop_reason *drop_reason); 470 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 471 struct request_sock *req, bool fastopen, 472 bool *lost_race, enum skb_drop_reason *drop_reason); 473 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child, 474 struct sk_buff *skb); 475 void tcp_enter_loss(struct sock *sk); 476 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 477 void tcp_clear_retrans(struct tcp_sock *tp); 478 void tcp_update_pacing_rate(struct sock *sk); 479 void tcp_set_rto(struct sock *sk); 480 void tcp_update_metrics(struct sock *sk); 481 void tcp_init_metrics(struct sock *sk); 482 void tcp_metrics_init(void); 483 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 484 void __tcp_close(struct sock *sk, long timeout); 485 void tcp_close(struct sock *sk, long timeout); 486 void tcp_init_sock(struct sock *sk); 487 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 488 __poll_t tcp_poll(struct file *file, struct socket *sock, 489 struct poll_table_struct *wait); 490 int do_tcp_getsockopt(struct sock *sk, int level, 491 int optname, sockptr_t optval, sockptr_t optlen); 492 int tcp_getsockopt(struct sock *sk, int level, int optname, 493 char __user *optval, int __user *optlen); 494 bool tcp_bpf_bypass_getsockopt(int level, int optname); 495 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 496 sockptr_t optval, unsigned int optlen); 497 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 498 unsigned int optlen); 499 void tcp_reset_keepalive_timer(struct sock *sk, unsigned long timeout); 500 void tcp_set_keepalive(struct sock *sk, int val); 501 void tcp_syn_ack_timeout(const struct request_sock *req); 502 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 503 int flags, int *addr_len); 504 int tcp_set_rcvlowat(struct sock *sk, int val); 505 int tcp_set_window_clamp(struct sock *sk, int val); 506 void tcp_update_recv_tstamps(struct sk_buff *skb, 507 struct scm_timestamping_internal *tss); 508 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 509 struct scm_timestamping_internal *tss); 510 void tcp_data_ready(struct sock *sk); 511 #ifdef CONFIG_MMU 512 int tcp_mmap(struct file *file, struct socket *sock, 513 struct vm_area_struct *vma); 514 #endif 515 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 516 struct tcp_options_received *opt_rx, 517 int estab, struct tcp_fastopen_cookie *foc); 518 519 /* 520 * BPF SKB-less helpers 521 */ 522 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 523 struct tcphdr *th, u32 *cookie); 524 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 525 struct tcphdr *th, u32 *cookie); 526 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 527 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 528 const struct tcp_request_sock_ops *af_ops, 529 struct sock *sk, struct tcphdr *th); 530 /* 531 * TCP v4 functions exported for the inet6 API 532 */ 533 534 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 535 void tcp_v4_mtu_reduced(struct sock *sk); 536 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 537 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 538 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 539 struct sock *tcp_create_openreq_child(const struct sock *sk, 540 struct request_sock *req, 541 struct sk_buff *skb); 542 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 543 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 544 struct request_sock *req, 545 struct dst_entry *dst, 546 struct request_sock *req_unhash, 547 bool *own_req); 548 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 549 int tcp_v4_connect(struct sock *sk, struct sockaddr_unsized *uaddr, int addr_len); 550 int tcp_connect(struct sock *sk); 551 enum tcp_synack_type { 552 TCP_SYNACK_NORMAL, 553 TCP_SYNACK_FASTOPEN, 554 TCP_SYNACK_COOKIE, 555 TCP_SYNACK_RETRANS, 556 }; 557 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 558 struct request_sock *req, 559 struct tcp_fastopen_cookie *foc, 560 enum tcp_synack_type synack_type, 561 struct sk_buff *syn_skb); 562 int tcp_disconnect(struct sock *sk, int flags); 563 564 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 565 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 566 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 567 568 /* From syncookies.c */ 569 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 570 struct request_sock *req, 571 struct dst_entry *dst); 572 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th); 573 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 574 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 575 struct sock *sk, struct sk_buff *skb, 576 struct tcp_options_received *tcp_opt, 577 int mss, u32 tsoff); 578 579 #if IS_ENABLED(CONFIG_BPF) 580 struct bpf_tcp_req_attrs { 581 u32 rcv_tsval; 582 u32 rcv_tsecr; 583 u16 mss; 584 u8 rcv_wscale; 585 u8 snd_wscale; 586 u8 ecn_ok; 587 u8 wscale_ok; 588 u8 sack_ok; 589 u8 tstamp_ok; 590 u8 usec_ts_ok; 591 u8 reserved[3]; 592 }; 593 #endif 594 595 #ifdef CONFIG_SYN_COOKIES 596 597 /* Syncookies use a monotonic timer which increments every 60 seconds. 598 * This counter is used both as a hash input and partially encoded into 599 * the cookie value. A cookie is only validated further if the delta 600 * between the current counter value and the encoded one is less than this, 601 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 602 * the counter advances immediately after a cookie is generated). 603 */ 604 #define MAX_SYNCOOKIE_AGE 2 605 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 606 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 607 608 /* syncookies: remember time of last synqueue overflow 609 * But do not dirty this field too often (once per second is enough) 610 * It is racy as we do not hold a lock, but race is very minor. 611 */ 612 static inline void tcp_synq_overflow(const struct sock *sk) 613 { 614 unsigned int last_overflow; 615 unsigned int now = jiffies; 616 617 if (sk->sk_reuseport) { 618 struct sock_reuseport *reuse; 619 620 reuse = rcu_dereference(sk->sk_reuseport_cb); 621 if (likely(reuse)) { 622 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 623 if (!time_between32(now, last_overflow, 624 last_overflow + HZ)) 625 WRITE_ONCE(reuse->synq_overflow_ts, now); 626 return; 627 } 628 } 629 630 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 631 if (!time_between32(now, last_overflow, last_overflow + HZ)) 632 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 633 } 634 635 /* syncookies: no recent synqueue overflow on this listening socket? */ 636 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 637 { 638 unsigned int last_overflow; 639 unsigned int now = jiffies; 640 641 if (sk->sk_reuseport) { 642 struct sock_reuseport *reuse; 643 644 reuse = rcu_dereference(sk->sk_reuseport_cb); 645 if (likely(reuse)) { 646 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 647 return !time_between32(now, last_overflow - HZ, 648 last_overflow + 649 TCP_SYNCOOKIE_VALID); 650 } 651 } 652 653 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 654 655 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 656 * then we're under synflood. However, we have to use 657 * 'last_overflow - HZ' as lower bound. That's because a concurrent 658 * tcp_synq_overflow() could update .ts_recent_stamp after we read 659 * jiffies but before we store .ts_recent_stamp into last_overflow, 660 * which could lead to rejecting a valid syncookie. 661 */ 662 return !time_between32(now, last_overflow - HZ, 663 last_overflow + TCP_SYNCOOKIE_VALID); 664 } 665 666 static inline u32 tcp_cookie_time(void) 667 { 668 u64 val = get_jiffies_64(); 669 670 do_div(val, TCP_SYNCOOKIE_PERIOD); 671 return val; 672 } 673 674 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */ 675 static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val) 676 { 677 if (usec_ts) 678 return div_u64(val, NSEC_PER_USEC); 679 680 return div_u64(val, NSEC_PER_MSEC); 681 } 682 683 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 684 u16 *mssp); 685 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 686 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 687 bool cookie_timestamp_decode(const struct net *net, 688 struct tcp_options_received *opt); 689 690 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst) 691 { 692 return READ_ONCE(net->ipv4.sysctl_tcp_ecn) || 693 dst_feature(dst, RTAX_FEATURE_ECN); 694 } 695 696 #if IS_ENABLED(CONFIG_BPF) 697 static inline bool cookie_bpf_ok(struct sk_buff *skb) 698 { 699 return skb->sk; 700 } 701 702 struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb); 703 #else 704 static inline bool cookie_bpf_ok(struct sk_buff *skb) 705 { 706 return false; 707 } 708 709 static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk, 710 struct sk_buff *skb) 711 { 712 return NULL; 713 } 714 #endif 715 716 /* From net/ipv6/syncookies.c */ 717 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th); 718 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 719 720 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 721 const struct tcphdr *th, u16 *mssp); 722 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 723 #endif 724 /* tcp_output.c */ 725 726 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 727 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 728 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 729 int nonagle); 730 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 731 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 732 void tcp_retransmit_timer(struct sock *sk); 733 void tcp_xmit_retransmit_queue(struct sock *); 734 void tcp_simple_retransmit(struct sock *); 735 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 736 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 737 enum tcp_queue { 738 TCP_FRAG_IN_WRITE_QUEUE, 739 TCP_FRAG_IN_RTX_QUEUE, 740 }; 741 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 742 struct sk_buff *skb, u32 len, 743 unsigned int mss_now, gfp_t gfp); 744 745 void tcp_send_probe0(struct sock *); 746 int tcp_write_wakeup(struct sock *, int mib); 747 void tcp_send_fin(struct sock *sk); 748 void tcp_send_active_reset(struct sock *sk, gfp_t priority, 749 enum sk_rst_reason reason); 750 int tcp_send_synack(struct sock *); 751 void tcp_push_one(struct sock *, unsigned int mss_now); 752 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt, u16 flags); 753 void tcp_send_ack(struct sock *sk); 754 void tcp_send_delayed_ack(struct sock *sk); 755 void tcp_send_loss_probe(struct sock *sk); 756 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 757 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 758 const struct sk_buff *next_skb); 759 760 /* tcp_input.c */ 761 void tcp_rearm_rto(struct sock *sk); 762 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 763 void tcp_done_with_error(struct sock *sk, int err); 764 void tcp_reset(struct sock *sk, struct sk_buff *skb); 765 void tcp_fin(struct sock *sk); 766 void __tcp_check_space(struct sock *sk); 767 static inline void tcp_check_space(struct sock *sk) 768 { 769 /* pairs with tcp_poll() */ 770 smp_mb(); 771 772 if (sk->sk_socket && test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) 773 __tcp_check_space(sk); 774 } 775 void tcp_sack_compress_send_ack(struct sock *sk); 776 777 static inline void tcp_cleanup_skb(struct sk_buff *skb) 778 { 779 skb_dst_drop(skb); 780 secpath_reset(skb); 781 } 782 783 static inline void tcp_add_receive_queue(struct sock *sk, struct sk_buff *skb) 784 { 785 DEBUG_NET_WARN_ON_ONCE(skb_dst(skb)); 786 DEBUG_NET_WARN_ON_ONCE(secpath_exists(skb)); 787 __skb_queue_tail(&sk->sk_receive_queue, skb); 788 } 789 790 /* tcp_timer.c */ 791 void tcp_init_xmit_timers(struct sock *); 792 static inline void tcp_clear_xmit_timers(struct sock *sk) 793 { 794 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 795 __sock_put(sk); 796 797 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 798 __sock_put(sk); 799 800 inet_csk_clear_xmit_timers(sk); 801 } 802 803 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 804 unsigned int tcp_current_mss(struct sock *sk); 805 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 806 807 /* Bound MSS / TSO packet size with the half of the window */ 808 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 809 { 810 int cutoff; 811 812 /* When peer uses tiny windows, there is no use in packetizing 813 * to sub-MSS pieces for the sake of SWS or making sure there 814 * are enough packets in the pipe for fast recovery. 815 * 816 * On the other hand, for extremely large MSS devices, handling 817 * smaller than MSS windows in this way does make sense. 818 */ 819 if (tp->max_window > TCP_MSS_DEFAULT) 820 cutoff = (tp->max_window >> 1); 821 else 822 cutoff = tp->max_window; 823 824 if (cutoff && pktsize > cutoff) 825 return max_t(int, cutoff, 68U - tp->tcp_header_len); 826 else 827 return pktsize; 828 } 829 830 /* tcp.c */ 831 void tcp_get_info(struct sock *, struct tcp_info *); 832 void tcp_rate_check_app_limited(struct sock *sk); 833 834 /* Read 'sendfile()'-style from a TCP socket */ 835 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 836 sk_read_actor_t recv_actor); 837 int tcp_read_sock_noack(struct sock *sk, read_descriptor_t *desc, 838 sk_read_actor_t recv_actor, bool noack, 839 u32 *copied_seq); 840 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 841 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 842 void tcp_read_done(struct sock *sk, size_t len); 843 844 void tcp_initialize_rcv_mss(struct sock *sk); 845 846 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 847 int tcp_mss_to_mtu(struct sock *sk, int mss); 848 void tcp_mtup_init(struct sock *sk); 849 850 static inline unsigned int tcp_rto_max(const struct sock *sk) 851 { 852 return READ_ONCE(inet_csk(sk)->icsk_rto_max); 853 } 854 855 static inline void tcp_bound_rto(struct sock *sk) 856 { 857 inet_csk(sk)->icsk_rto = min(inet_csk(sk)->icsk_rto, tcp_rto_max(sk)); 858 } 859 860 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 861 { 862 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 863 } 864 865 static inline unsigned long tcp_reqsk_timeout(struct request_sock *req) 866 { 867 u64 timeout = (u64)req->timeout << req->num_timeout; 868 869 return (unsigned long)min_t(u64, timeout, 870 tcp_rto_max(req->rsk_listener)); 871 } 872 873 u32 tcp_delack_max(const struct sock *sk); 874 875 /* Compute the actual rto_min value */ 876 static inline u32 tcp_rto_min(const struct sock *sk) 877 { 878 const struct dst_entry *dst = __sk_dst_get(sk); 879 u32 rto_min = READ_ONCE(inet_csk(sk)->icsk_rto_min); 880 881 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 882 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 883 return rto_min; 884 } 885 886 static inline u32 tcp_rto_min_us(const struct sock *sk) 887 { 888 return jiffies_to_usecs(tcp_rto_min(sk)); 889 } 890 891 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 892 { 893 return dst_metric_locked(dst, RTAX_CC_ALGO); 894 } 895 896 /* Minimum RTT in usec. ~0 means not available. */ 897 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 898 { 899 return minmax_get(&tp->rtt_min); 900 } 901 902 /* Compute the actual receive window we are currently advertising. 903 * Rcv_nxt can be after the window if our peer push more data 904 * than the offered window. 905 */ 906 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 907 { 908 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 909 910 if (win < 0) 911 win = 0; 912 return (u32) win; 913 } 914 915 /* Choose a new window, without checks for shrinking, and without 916 * scaling applied to the result. The caller does these things 917 * if necessary. This is a "raw" window selection. 918 */ 919 u32 __tcp_select_window(struct sock *sk); 920 921 void tcp_send_window_probe(struct sock *sk); 922 923 /* TCP uses 32bit jiffies to save some space. 924 * Note that this is different from tcp_time_stamp, which 925 * historically has been the same until linux-4.13. 926 */ 927 #define tcp_jiffies32 ((u32)jiffies) 928 929 /* 930 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 931 * It is no longer tied to jiffies, but to 1 ms clock. 932 * Note: double check if you want to use tcp_jiffies32 instead of this. 933 */ 934 #define TCP_TS_HZ 1000 935 936 static inline u64 tcp_clock_ns(void) 937 { 938 return ktime_get_ns(); 939 } 940 941 static inline u64 tcp_clock_us(void) 942 { 943 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 944 } 945 946 static inline u64 tcp_clock_ms(void) 947 { 948 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 949 } 950 951 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 952 * or usec resolution. Each socket carries a flag to select one or other 953 * resolution, as the route attribute could change anytime. 954 * Each flow must stick to initial resolution. 955 */ 956 static inline u32 tcp_clock_ts(bool usec_ts) 957 { 958 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 959 } 960 961 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 962 { 963 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 964 } 965 966 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 967 { 968 if (tp->tcp_usec_ts) 969 return tp->tcp_mstamp; 970 return tcp_time_stamp_ms(tp); 971 } 972 973 void tcp_mstamp_refresh(struct tcp_sock *tp); 974 975 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 976 { 977 return max_t(s64, t1 - t0, 0); 978 } 979 980 /* provide the departure time in us unit */ 981 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 982 { 983 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 984 } 985 986 /* Provide skb TSval in usec or ms unit */ 987 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 988 { 989 if (usec_ts) 990 return tcp_skb_timestamp_us(skb); 991 992 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 993 } 994 995 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 996 { 997 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 998 } 999 1000 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 1001 { 1002 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 1003 } 1004 1005 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 1006 1007 #define TCPHDR_FIN BIT(0) 1008 #define TCPHDR_SYN BIT(1) 1009 #define TCPHDR_RST BIT(2) 1010 #define TCPHDR_PSH BIT(3) 1011 #define TCPHDR_ACK BIT(4) 1012 #define TCPHDR_URG BIT(5) 1013 #define TCPHDR_ECE BIT(6) 1014 #define TCPHDR_CWR BIT(7) 1015 #define TCPHDR_AE BIT(8) 1016 #define TCPHDR_FLAGS_MASK (TCPHDR_FIN | TCPHDR_SYN | TCPHDR_RST | \ 1017 TCPHDR_PSH | TCPHDR_ACK | TCPHDR_URG | \ 1018 TCPHDR_ECE | TCPHDR_CWR | TCPHDR_AE) 1019 #define tcp_flags_ntohs(th) (ntohs(*(__be16 *)&tcp_flag_word(th)) & \ 1020 TCPHDR_FLAGS_MASK) 1021 1022 #define TCPHDR_ACE (TCPHDR_ECE | TCPHDR_CWR | TCPHDR_AE) 1023 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 1024 #define TCPHDR_SYNACK_ACCECN (TCPHDR_SYN | TCPHDR_ACK | TCPHDR_CWR) 1025 1026 #define TCP_ACCECN_CEP_ACE_MASK 0x7 1027 #define TCP_ACCECN_ACE_MAX_DELTA 6 1028 1029 /* To avoid/detect middlebox interference, not all counters start at 0. 1030 * See draft-ietf-tcpm-accurate-ecn for the latest values. 1031 */ 1032 #define TCP_ACCECN_CEP_INIT_OFFSET 5 1033 #define TCP_ACCECN_E1B_INIT_OFFSET 1 1034 #define TCP_ACCECN_E0B_INIT_OFFSET 1 1035 #define TCP_ACCECN_CEB_INIT_OFFSET 0 1036 1037 /* State flags for sacked in struct tcp_skb_cb */ 1038 enum tcp_skb_cb_sacked_flags { 1039 TCPCB_SACKED_ACKED = (1 << 0), /* SKB ACK'd by a SACK block */ 1040 TCPCB_SACKED_RETRANS = (1 << 1), /* SKB retransmitted */ 1041 TCPCB_LOST = (1 << 2), /* SKB is lost */ 1042 TCPCB_TAGBITS = (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS | 1043 TCPCB_LOST), /* All tag bits */ 1044 TCPCB_REPAIRED = (1 << 4), /* SKB repaired (no skb_mstamp_ns) */ 1045 TCPCB_EVER_RETRANS = (1 << 7), /* Ever retransmitted frame */ 1046 TCPCB_RETRANS = (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS | 1047 TCPCB_REPAIRED), 1048 }; 1049 1050 /* This is what the send packet queuing engine uses to pass 1051 * TCP per-packet control information to the transmission code. 1052 * We also store the host-order sequence numbers in here too. 1053 * This is 44 bytes if IPV6 is enabled. 1054 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 1055 */ 1056 struct tcp_skb_cb { 1057 __u32 seq; /* Starting sequence number */ 1058 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 1059 union { 1060 /* Note : 1061 * tcp_gso_segs/size are used in write queue only, 1062 * cf tcp_skb_pcount()/tcp_skb_mss() 1063 */ 1064 struct { 1065 u16 tcp_gso_segs; 1066 u16 tcp_gso_size; 1067 }; 1068 }; 1069 __u16 tcp_flags; /* TCP header flags (tcp[12-13])*/ 1070 1071 __u8 sacked; /* State flags for SACK. */ 1072 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 1073 #define TSTAMP_ACK_SK 0x1 1074 #define TSTAMP_ACK_BPF 0x2 1075 __u8 txstamp_ack:2, /* Record TX timestamp for ack? */ 1076 eor:1, /* Is skb MSG_EOR marked? */ 1077 has_rxtstamp:1, /* SKB has a RX timestamp */ 1078 unused:4; 1079 __u32 ack_seq; /* Sequence number ACK'd */ 1080 union { 1081 struct { 1082 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 1083 /* There is space for up to 24 bytes */ 1084 __u32 is_app_limited:1, /* cwnd not fully used? */ 1085 delivered_ce:20, 1086 unused:11; 1087 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 1088 __u32 delivered; 1089 /* start of send pipeline phase */ 1090 u64 first_tx_mstamp; 1091 /* when we reached the "delivered" count */ 1092 u64 delivered_mstamp; 1093 } tx; /* only used for outgoing skbs */ 1094 union { 1095 struct inet_skb_parm h4; 1096 #if IS_ENABLED(CONFIG_IPV6) 1097 struct inet6_skb_parm h6; 1098 #endif 1099 } header; /* For incoming skbs */ 1100 }; 1101 }; 1102 1103 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 1104 1105 extern const struct inet_connection_sock_af_ops ipv4_specific; 1106 1107 #if IS_ENABLED(CONFIG_IPV6) 1108 /* This is the variant of inet6_iif() that must be used by TCP, 1109 * as TCP moves IP6CB into a different location in skb->cb[] 1110 */ 1111 static inline int tcp_v6_iif(const struct sk_buff *skb) 1112 { 1113 return TCP_SKB_CB(skb)->header.h6.iif; 1114 } 1115 1116 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 1117 { 1118 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 1119 1120 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 1121 } 1122 1123 /* TCP_SKB_CB reference means this can not be used from early demux */ 1124 static inline int tcp_v6_sdif(const struct sk_buff *skb) 1125 { 1126 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1127 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 1128 return TCP_SKB_CB(skb)->header.h6.iif; 1129 #endif 1130 return 0; 1131 } 1132 1133 extern const struct inet_connection_sock_af_ops ipv6_specific; 1134 1135 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 1136 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 1137 void tcp_v6_early_demux(struct sk_buff *skb); 1138 1139 #endif 1140 1141 /* TCP_SKB_CB reference means this can not be used from early demux */ 1142 static inline int tcp_v4_sdif(struct sk_buff *skb) 1143 { 1144 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1145 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 1146 return TCP_SKB_CB(skb)->header.h4.iif; 1147 #endif 1148 return 0; 1149 } 1150 1151 /* Due to TSO, an SKB can be composed of multiple actual 1152 * packets. To keep these tracked properly, we use this. 1153 */ 1154 static inline int tcp_skb_pcount(const struct sk_buff *skb) 1155 { 1156 return TCP_SKB_CB(skb)->tcp_gso_segs; 1157 } 1158 1159 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 1160 { 1161 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1162 } 1163 1164 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1165 { 1166 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1167 } 1168 1169 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1170 static inline int tcp_skb_mss(const struct sk_buff *skb) 1171 { 1172 return TCP_SKB_CB(skb)->tcp_gso_size; 1173 } 1174 1175 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1176 { 1177 return likely(!TCP_SKB_CB(skb)->eor); 1178 } 1179 1180 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1181 const struct sk_buff *from) 1182 { 1183 /* skb_cmp_decrypted() not needed, use tcp_write_collapse_fence() */ 1184 return likely(tcp_skb_can_collapse_to(to) && 1185 mptcp_skb_can_collapse(to, from) && 1186 skb_pure_zcopy_same(to, from) && 1187 skb_frags_readable(to) == skb_frags_readable(from)); 1188 } 1189 1190 static inline bool tcp_skb_can_collapse_rx(const struct sk_buff *to, 1191 const struct sk_buff *from) 1192 { 1193 return likely(mptcp_skb_can_collapse(to, from) && 1194 !skb_cmp_decrypted(to, from)); 1195 } 1196 1197 /* Events passed to congestion control interface */ 1198 enum tcp_ca_event { 1199 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1200 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1201 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1202 CA_EVENT_LOSS, /* loss timeout */ 1203 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1204 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1205 }; 1206 1207 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1208 enum tcp_ca_ack_event_flags { 1209 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1210 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1211 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1212 }; 1213 1214 /* 1215 * Interface for adding new TCP congestion control handlers 1216 */ 1217 #define TCP_CA_NAME_MAX 16 1218 #define TCP_CA_MAX 128 1219 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1220 1221 #define TCP_CA_UNSPEC 0 1222 1223 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1224 #define TCP_CONG_NON_RESTRICTED BIT(0) 1225 /* Requires ECN/ECT set on all packets */ 1226 #define TCP_CONG_NEEDS_ECN BIT(1) 1227 /* Require successfully negotiated AccECN capability */ 1228 #define TCP_CONG_NEEDS_ACCECN BIT(2) 1229 /* Use ECT(1) instead of ECT(0) while the CA is uninitialized */ 1230 #define TCP_CONG_ECT_1_NEGOTIATION BIT(3) 1231 /* Cannot fallback to RFC3168 during AccECN negotiation */ 1232 #define TCP_CONG_NO_FALLBACK_RFC3168 BIT(4) 1233 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN | \ 1234 TCP_CONG_NEEDS_ACCECN | TCP_CONG_ECT_1_NEGOTIATION | \ 1235 TCP_CONG_NO_FALLBACK_RFC3168) 1236 1237 union tcp_cc_info; 1238 1239 struct ack_sample { 1240 u32 pkts_acked; 1241 s32 rtt_us; 1242 u32 in_flight; 1243 }; 1244 1245 /* A rate sample measures the number of (original/retransmitted) data 1246 * packets delivered "delivered" over an interval of time "interval_us". 1247 * The tcp_rate.c code fills in the rate sample, and congestion 1248 * control modules that define a cong_control function to run at the end 1249 * of ACK processing can optionally chose to consult this sample when 1250 * setting cwnd and pacing rate. 1251 * A sample is invalid if "delivered" or "interval_us" is negative. 1252 */ 1253 struct rate_sample { 1254 u64 prior_mstamp; /* starting timestamp for interval */ 1255 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1256 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1257 s32 delivered; /* number of packets delivered over interval */ 1258 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1259 long interval_us; /* time for tp->delivered to incr "delivered" */ 1260 u32 snd_interval_us; /* snd interval for delivered packets */ 1261 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1262 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1263 int losses; /* number of packets marked lost upon ACK */ 1264 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1265 u32 prior_in_flight; /* in flight before this ACK */ 1266 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1267 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1268 bool is_retrans; /* is sample from retransmission? */ 1269 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1270 }; 1271 1272 struct tcp_congestion_ops { 1273 /* fast path fields are put first to fill one cache line */ 1274 1275 /* A congestion control (CC) must provide one of either: 1276 * 1277 * (a) a cong_avoid function, if the CC wants to use the core TCP 1278 * stack's default functionality to implement a "classic" 1279 * (Reno/CUBIC-style) response to packet loss, RFC3168 ECN, 1280 * idle periods, pacing rate computations, etc. 1281 * 1282 * (b) a cong_control function, if the CC wants custom behavior and 1283 * complete control of all congestion control behaviors. 1284 */ 1285 /* (a) "classic" response: calculate new cwnd. 1286 */ 1287 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1288 /* (b) "custom" response: call when packets are delivered to update 1289 * cwnd and pacing rate, after all the ca_state processing. 1290 */ 1291 void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs); 1292 1293 /* return slow start threshold (required) */ 1294 u32 (*ssthresh)(struct sock *sk); 1295 1296 /* call before changing ca_state (optional) */ 1297 void (*set_state)(struct sock *sk, u8 new_state); 1298 1299 /* call when cwnd event occurs (optional) */ 1300 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1301 1302 /* call when ack arrives (optional) */ 1303 void (*in_ack_event)(struct sock *sk, u32 flags); 1304 1305 /* hook for packet ack accounting (optional) */ 1306 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1307 1308 /* override sysctl_tcp_min_tso_segs (optional) */ 1309 u32 (*min_tso_segs)(struct sock *sk); 1310 1311 /* new value of cwnd after loss (required) */ 1312 u32 (*undo_cwnd)(struct sock *sk); 1313 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1314 u32 (*sndbuf_expand)(struct sock *sk); 1315 1316 /* control/slow paths put last */ 1317 /* get info for inet_diag (optional) */ 1318 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1319 union tcp_cc_info *info); 1320 1321 char name[TCP_CA_NAME_MAX]; 1322 struct module *owner; 1323 struct list_head list; 1324 u32 key; 1325 u32 flags; 1326 1327 /* initialize private data (optional) */ 1328 void (*init)(struct sock *sk); 1329 /* cleanup private data (optional) */ 1330 void (*release)(struct sock *sk); 1331 } ____cacheline_aligned_in_smp; 1332 1333 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1334 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1335 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1336 struct tcp_congestion_ops *old_type); 1337 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1338 1339 void tcp_assign_congestion_control(struct sock *sk); 1340 void tcp_init_congestion_control(struct sock *sk); 1341 void tcp_cleanup_congestion_control(struct sock *sk); 1342 int tcp_set_default_congestion_control(struct net *net, const char *name); 1343 void tcp_get_default_congestion_control(struct net *net, char *name); 1344 void tcp_get_available_congestion_control(char *buf, size_t len); 1345 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1346 int tcp_set_allowed_congestion_control(char *allowed); 1347 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1348 bool cap_net_admin); 1349 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1350 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1351 1352 u32 tcp_reno_ssthresh(struct sock *sk); 1353 u32 tcp_reno_undo_cwnd(struct sock *sk); 1354 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1355 extern struct tcp_congestion_ops tcp_reno; 1356 1357 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1358 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1359 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1360 #ifdef CONFIG_INET 1361 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1362 #else 1363 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1364 { 1365 return NULL; 1366 } 1367 #endif 1368 1369 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1370 { 1371 const struct inet_connection_sock *icsk = inet_csk(sk); 1372 1373 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1374 } 1375 1376 static inline bool tcp_ca_needs_accecn(const struct sock *sk) 1377 { 1378 const struct inet_connection_sock *icsk = inet_csk(sk); 1379 1380 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ACCECN; 1381 } 1382 1383 static inline bool tcp_ca_ect_1_negotiation(const struct sock *sk) 1384 { 1385 const struct inet_connection_sock *icsk = inet_csk(sk); 1386 1387 return icsk->icsk_ca_ops->flags & TCP_CONG_ECT_1_NEGOTIATION; 1388 } 1389 1390 static inline bool tcp_ca_no_fallback_rfc3168(const struct sock *sk) 1391 { 1392 const struct inet_connection_sock *icsk = inet_csk(sk); 1393 1394 return icsk->icsk_ca_ops->flags & TCP_CONG_NO_FALLBACK_RFC3168; 1395 } 1396 1397 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1398 { 1399 const struct inet_connection_sock *icsk = inet_csk(sk); 1400 1401 if (icsk->icsk_ca_ops->cwnd_event) 1402 icsk->icsk_ca_ops->cwnd_event(sk, event); 1403 } 1404 1405 /* From tcp_cong.c */ 1406 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1407 1408 1409 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1410 { 1411 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1412 } 1413 1414 /* These functions determine how the current flow behaves in respect of SACK 1415 * handling. SACK is negotiated with the peer, and therefore it can vary 1416 * between different flows. 1417 * 1418 * tcp_is_sack - SACK enabled 1419 * tcp_is_reno - No SACK 1420 */ 1421 static inline int tcp_is_sack(const struct tcp_sock *tp) 1422 { 1423 return likely(tp->rx_opt.sack_ok); 1424 } 1425 1426 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1427 { 1428 return !tcp_is_sack(tp); 1429 } 1430 1431 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1432 { 1433 return tp->sacked_out + tp->lost_out; 1434 } 1435 1436 /* This determines how many packets are "in the network" to the best 1437 * of our knowledge. In many cases it is conservative, but where 1438 * detailed information is available from the receiver (via SACK 1439 * blocks etc.) we can make more aggressive calculations. 1440 * 1441 * Use this for decisions involving congestion control, use just 1442 * tp->packets_out to determine if the send queue is empty or not. 1443 * 1444 * Read this equation as: 1445 * 1446 * "Packets sent once on transmission queue" MINUS 1447 * "Packets left network, but not honestly ACKed yet" PLUS 1448 * "Packets fast retransmitted" 1449 */ 1450 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1451 { 1452 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1453 } 1454 1455 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1456 1457 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1458 { 1459 return tp->snd_cwnd; 1460 } 1461 1462 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1463 { 1464 WARN_ON_ONCE((int)val <= 0); 1465 tp->snd_cwnd = val; 1466 } 1467 1468 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1469 { 1470 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1471 } 1472 1473 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1474 { 1475 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1476 } 1477 1478 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1479 { 1480 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1481 (1 << inet_csk(sk)->icsk_ca_state); 1482 } 1483 1484 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1485 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1486 * ssthresh. 1487 */ 1488 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1489 { 1490 const struct tcp_sock *tp = tcp_sk(sk); 1491 1492 if (tcp_in_cwnd_reduction(sk)) 1493 return tp->snd_ssthresh; 1494 else 1495 return max(tp->snd_ssthresh, 1496 ((tcp_snd_cwnd(tp) >> 1) + 1497 (tcp_snd_cwnd(tp) >> 2))); 1498 } 1499 1500 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1501 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1502 1503 void tcp_enter_cwr(struct sock *sk); 1504 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1505 1506 /* The maximum number of MSS of available cwnd for which TSO defers 1507 * sending if not using sysctl_tcp_tso_win_divisor. 1508 */ 1509 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1510 { 1511 return 3; 1512 } 1513 1514 /* Returns end sequence number of the receiver's advertised window */ 1515 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1516 { 1517 return tp->snd_una + tp->snd_wnd; 1518 } 1519 1520 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1521 * flexible approach. The RFC suggests cwnd should not be raised unless 1522 * it was fully used previously. And that's exactly what we do in 1523 * congestion avoidance mode. But in slow start we allow cwnd to grow 1524 * as long as the application has used half the cwnd. 1525 * Example : 1526 * cwnd is 10 (IW10), but application sends 9 frames. 1527 * We allow cwnd to reach 18 when all frames are ACKed. 1528 * This check is safe because it's as aggressive as slow start which already 1529 * risks 100% overshoot. The advantage is that we discourage application to 1530 * either send more filler packets or data to artificially blow up the cwnd 1531 * usage, and allow application-limited process to probe bw more aggressively. 1532 */ 1533 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1534 { 1535 const struct tcp_sock *tp = tcp_sk(sk); 1536 1537 if (tp->is_cwnd_limited) 1538 return true; 1539 1540 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1541 if (tcp_in_slow_start(tp)) 1542 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1543 1544 return false; 1545 } 1546 1547 /* BBR congestion control needs pacing. 1548 * Same remark for SO_MAX_PACING_RATE. 1549 * sch_fq packet scheduler is efficiently handling pacing, 1550 * but is not always installed/used. 1551 * Return true if TCP stack should pace packets itself. 1552 */ 1553 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1554 { 1555 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1556 } 1557 1558 /* Estimates in how many jiffies next packet for this flow can be sent. 1559 * Scheduling a retransmit timer too early would be silly. 1560 */ 1561 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1562 { 1563 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1564 1565 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1566 } 1567 1568 static inline void tcp_reset_xmit_timer(struct sock *sk, 1569 const int what, 1570 unsigned long when, 1571 bool pace_delay) 1572 { 1573 if (pace_delay) 1574 when += tcp_pacing_delay(sk); 1575 inet_csk_reset_xmit_timer(sk, what, when, 1576 tcp_rto_max(sk)); 1577 } 1578 1579 /* Something is really bad, we could not queue an additional packet, 1580 * because qdisc is full or receiver sent a 0 window, or we are paced. 1581 * We do not want to add fuel to the fire, or abort too early, 1582 * so make sure the timer we arm now is at least 200ms in the future, 1583 * regardless of current icsk_rto value (as it could be ~2ms) 1584 */ 1585 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1586 { 1587 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1588 } 1589 1590 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1591 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1592 unsigned long max_when) 1593 { 1594 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1595 inet_csk(sk)->icsk_backoff); 1596 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1597 1598 return (unsigned long)min_t(u64, when, max_when); 1599 } 1600 1601 static inline void tcp_check_probe_timer(struct sock *sk) 1602 { 1603 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1604 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1605 tcp_probe0_base(sk), true); 1606 } 1607 1608 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1609 { 1610 tp->snd_wl1 = seq; 1611 } 1612 1613 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1614 { 1615 tp->snd_wl1 = seq; 1616 } 1617 1618 /* 1619 * Calculate(/check) TCP checksum 1620 */ 1621 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1622 __be32 daddr, __wsum base) 1623 { 1624 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1625 } 1626 1627 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1628 { 1629 return !skb_csum_unnecessary(skb) && 1630 __skb_checksum_complete(skb); 1631 } 1632 1633 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1634 enum skb_drop_reason *reason); 1635 1636 1637 int tcp_filter(struct sock *sk, struct sk_buff *skb, enum skb_drop_reason *reason); 1638 void tcp_set_state(struct sock *sk, int state); 1639 void tcp_done(struct sock *sk); 1640 int tcp_abort(struct sock *sk, int err); 1641 1642 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1643 { 1644 rx_opt->dsack = 0; 1645 rx_opt->num_sacks = 0; 1646 } 1647 1648 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1649 1650 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1651 { 1652 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1653 struct tcp_sock *tp = tcp_sk(sk); 1654 s32 delta; 1655 1656 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1657 tp->packets_out || ca_ops->cong_control) 1658 return; 1659 delta = tcp_jiffies32 - tp->lsndtime; 1660 if (delta > inet_csk(sk)->icsk_rto) 1661 tcp_cwnd_restart(sk, delta); 1662 } 1663 1664 /* Determine a window scaling and initial window to offer. */ 1665 void tcp_select_initial_window(const struct sock *sk, int __space, 1666 __u32 mss, __u32 *rcv_wnd, 1667 __u32 *window_clamp, int wscale_ok, 1668 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1669 1670 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1671 { 1672 s64 scaled_space = (s64)space * scaling_ratio; 1673 1674 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1675 } 1676 1677 static inline int tcp_win_from_space(const struct sock *sk, int space) 1678 { 1679 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1680 } 1681 1682 /* inverse of __tcp_win_from_space() */ 1683 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1684 { 1685 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1686 1687 do_div(val, scaling_ratio); 1688 return val; 1689 } 1690 1691 static inline int tcp_space_from_win(const struct sock *sk, int win) 1692 { 1693 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1694 } 1695 1696 /* Assume a 50% default for skb->len/skb->truesize ratio. 1697 * This may be adjusted later in tcp_measure_rcv_mss(). 1698 */ 1699 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1)) 1700 1701 static inline void tcp_scaling_ratio_init(struct sock *sk) 1702 { 1703 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1704 } 1705 1706 /* Note: caller must be prepared to deal with negative returns */ 1707 static inline int tcp_space(const struct sock *sk) 1708 { 1709 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1710 READ_ONCE(sk->sk_backlog.len) - 1711 atomic_read(&sk->sk_rmem_alloc)); 1712 } 1713 1714 static inline int tcp_full_space(const struct sock *sk) 1715 { 1716 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1717 } 1718 1719 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh) 1720 { 1721 int unused_mem = sk_unused_reserved_mem(sk); 1722 struct tcp_sock *tp = tcp_sk(sk); 1723 1724 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh); 1725 if (unused_mem) 1726 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1727 tcp_win_from_space(sk, unused_mem)); 1728 } 1729 1730 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1731 { 1732 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss); 1733 } 1734 1735 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1736 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1737 1738 1739 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1740 * If 87.5 % (7/8) of the space has been consumed, we want to override 1741 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1742 * len/truesize ratio. 1743 */ 1744 static inline bool tcp_rmem_pressure(const struct sock *sk) 1745 { 1746 int rcvbuf, threshold; 1747 1748 if (tcp_under_memory_pressure(sk)) 1749 return true; 1750 1751 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1752 threshold = rcvbuf - (rcvbuf >> 3); 1753 1754 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1755 } 1756 1757 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1758 { 1759 const struct tcp_sock *tp = tcp_sk(sk); 1760 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1761 1762 if (avail <= 0) 1763 return false; 1764 1765 return (avail >= target) || tcp_rmem_pressure(sk) || 1766 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1767 } 1768 1769 extern void tcp_openreq_init_rwin(struct request_sock *req, 1770 const struct sock *sk_listener, 1771 const struct dst_entry *dst); 1772 1773 void tcp_enter_memory_pressure(struct sock *sk); 1774 void tcp_leave_memory_pressure(struct sock *sk); 1775 1776 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1777 { 1778 struct net *net = sock_net((struct sock *)tp); 1779 int val; 1780 1781 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1782 * and do_tcp_setsockopt(). 1783 */ 1784 val = READ_ONCE(tp->keepalive_intvl); 1785 1786 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1787 } 1788 1789 static inline int keepalive_time_when(const struct tcp_sock *tp) 1790 { 1791 struct net *net = sock_net((struct sock *)tp); 1792 int val; 1793 1794 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1795 val = READ_ONCE(tp->keepalive_time); 1796 1797 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1798 } 1799 1800 static inline int keepalive_probes(const struct tcp_sock *tp) 1801 { 1802 struct net *net = sock_net((struct sock *)tp); 1803 int val; 1804 1805 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1806 * and do_tcp_setsockopt(). 1807 */ 1808 val = READ_ONCE(tp->keepalive_probes); 1809 1810 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1811 } 1812 1813 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1814 { 1815 const struct inet_connection_sock *icsk = &tp->inet_conn; 1816 1817 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1818 tcp_jiffies32 - tp->rcv_tstamp); 1819 } 1820 1821 static inline int tcp_fin_time(const struct sock *sk) 1822 { 1823 int fin_timeout = tcp_sk(sk)->linger2 ? : 1824 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1825 const int rto = inet_csk(sk)->icsk_rto; 1826 1827 if (fin_timeout < (rto << 2) - (rto >> 1)) 1828 fin_timeout = (rto << 2) - (rto >> 1); 1829 1830 return fin_timeout; 1831 } 1832 1833 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1834 int paws_win) 1835 { 1836 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1837 return true; 1838 if (unlikely(!time_before32(ktime_get_seconds(), 1839 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1840 return true; 1841 /* 1842 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1843 * then following tcp messages have valid values. Ignore 0 value, 1844 * or else 'negative' tsval might forbid us to accept their packets. 1845 */ 1846 if (!rx_opt->ts_recent) 1847 return true; 1848 return false; 1849 } 1850 1851 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1852 int rst) 1853 { 1854 if (tcp_paws_check(rx_opt, 0)) 1855 return false; 1856 1857 /* RST segments are not recommended to carry timestamp, 1858 and, if they do, it is recommended to ignore PAWS because 1859 "their cleanup function should take precedence over timestamps." 1860 Certainly, it is mistake. It is necessary to understand the reasons 1861 of this constraint to relax it: if peer reboots, clock may go 1862 out-of-sync and half-open connections will not be reset. 1863 Actually, the problem would be not existing if all 1864 the implementations followed draft about maintaining clock 1865 via reboots. Linux-2.2 DOES NOT! 1866 1867 However, we can relax time bounds for RST segments to MSL. 1868 */ 1869 if (rst && !time_before32(ktime_get_seconds(), 1870 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1871 return false; 1872 return true; 1873 } 1874 1875 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 1876 { 1877 u32 ace; 1878 1879 /* mptcp hooks are only on the slow path */ 1880 if (sk_is_mptcp((struct sock *)tp)) 1881 return; 1882 1883 ace = tcp_ecn_mode_accecn(tp) ? 1884 ((tp->delivered_ce + TCP_ACCECN_CEP_INIT_OFFSET) & 1885 TCP_ACCECN_CEP_ACE_MASK) : 0; 1886 1887 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 1888 (ace << 22) | 1889 ntohl(TCP_FLAG_ACK) | 1890 snd_wnd); 1891 } 1892 1893 static inline void tcp_fast_path_on(struct tcp_sock *tp) 1894 { 1895 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 1896 } 1897 1898 static inline void tcp_fast_path_check(struct sock *sk) 1899 { 1900 struct tcp_sock *tp = tcp_sk(sk); 1901 1902 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 1903 tp->rcv_wnd && 1904 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 1905 !tp->urg_data) 1906 tcp_fast_path_on(tp); 1907 } 1908 1909 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1910 int mib_idx, u32 *last_oow_ack_time); 1911 1912 static inline void tcp_mib_init(struct net *net) 1913 { 1914 /* See RFC 2012 */ 1915 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1916 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1917 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1918 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1919 } 1920 1921 /* from STCP */ 1922 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1923 { 1924 tp->retransmit_skb_hint = NULL; 1925 } 1926 1927 #define tcp_md5_addr tcp_ao_addr 1928 1929 /* - key database */ 1930 struct tcp_md5sig_key { 1931 struct hlist_node node; 1932 u8 keylen; 1933 u8 family; /* AF_INET or AF_INET6 */ 1934 u8 prefixlen; 1935 u8 flags; 1936 union tcp_md5_addr addr; 1937 int l3index; /* set if key added with L3 scope */ 1938 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1939 struct rcu_head rcu; 1940 }; 1941 1942 /* - sock block */ 1943 struct tcp_md5sig_info { 1944 struct hlist_head head; 1945 struct rcu_head rcu; 1946 }; 1947 1948 /* - pseudo header */ 1949 struct tcp4_pseudohdr { 1950 __be32 saddr; 1951 __be32 daddr; 1952 __u8 pad; 1953 __u8 protocol; 1954 __be16 len; 1955 }; 1956 1957 struct tcp6_pseudohdr { 1958 struct in6_addr saddr; 1959 struct in6_addr daddr; 1960 __be32 len; 1961 __be32 protocol; /* including padding */ 1962 }; 1963 1964 /* 1965 * struct tcp_sigpool - per-CPU pool of ahash_requests 1966 * @scratch: per-CPU temporary area, that can be used between 1967 * tcp_sigpool_start() and tcp_sigpool_end() to perform 1968 * crypto request 1969 * @req: pre-allocated ahash request 1970 */ 1971 struct tcp_sigpool { 1972 void *scratch; 1973 struct ahash_request *req; 1974 }; 1975 1976 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size); 1977 void tcp_sigpool_get(unsigned int id); 1978 void tcp_sigpool_release(unsigned int id); 1979 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp, 1980 const struct sk_buff *skb, 1981 unsigned int header_len); 1982 1983 /** 1984 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash 1985 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash() 1986 * @c: returned tcp_sigpool for usage (uninitialized on failure) 1987 * 1988 * Returns: 0 on success, error otherwise. 1989 */ 1990 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c); 1991 /** 1992 * tcp_sigpool_end - enable bh and stop using tcp_sigpool 1993 * @c: tcp_sigpool context that was returned by tcp_sigpool_start() 1994 */ 1995 void tcp_sigpool_end(struct tcp_sigpool *c); 1996 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len); 1997 /* - functions */ 1998 void tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1999 const struct sock *sk, const struct sk_buff *skb); 2000 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 2001 int family, u8 prefixlen, int l3index, u8 flags, 2002 const u8 *newkey, u8 newkeylen); 2003 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 2004 int family, u8 prefixlen, int l3index, 2005 struct tcp_md5sig_key *key); 2006 2007 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 2008 int family, u8 prefixlen, int l3index, u8 flags); 2009 void tcp_clear_md5_list(struct sock *sk); 2010 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 2011 const struct sock *addr_sk); 2012 2013 #ifdef CONFIG_TCP_MD5SIG 2014 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 2015 const union tcp_md5_addr *addr, 2016 int family, bool any_l3index); 2017 static inline struct tcp_md5sig_key * 2018 tcp_md5_do_lookup(const struct sock *sk, int l3index, 2019 const union tcp_md5_addr *addr, int family) 2020 { 2021 if (!static_branch_unlikely(&tcp_md5_needed.key)) 2022 return NULL; 2023 return __tcp_md5_do_lookup(sk, l3index, addr, family, false); 2024 } 2025 2026 static inline struct tcp_md5sig_key * 2027 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 2028 const union tcp_md5_addr *addr, int family) 2029 { 2030 if (!static_branch_unlikely(&tcp_md5_needed.key)) 2031 return NULL; 2032 return __tcp_md5_do_lookup(sk, 0, addr, family, true); 2033 } 2034 2035 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 2036 void tcp_md5_destruct_sock(struct sock *sk); 2037 #else 2038 static inline struct tcp_md5sig_key * 2039 tcp_md5_do_lookup(const struct sock *sk, int l3index, 2040 const union tcp_md5_addr *addr, int family) 2041 { 2042 return NULL; 2043 } 2044 2045 static inline struct tcp_md5sig_key * 2046 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 2047 const union tcp_md5_addr *addr, int family) 2048 { 2049 return NULL; 2050 } 2051 2052 #define tcp_twsk_md5_key(twsk) NULL 2053 static inline void tcp_md5_destruct_sock(struct sock *sk) 2054 { 2055 } 2056 #endif 2057 2058 struct md5_ctx; 2059 void tcp_md5_hash_skb_data(struct md5_ctx *ctx, const struct sk_buff *skb, 2060 unsigned int header_len); 2061 void tcp_md5_hash_key(struct md5_ctx *ctx, const struct tcp_md5sig_key *key); 2062 2063 /* From tcp_fastopen.c */ 2064 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 2065 struct tcp_fastopen_cookie *cookie); 2066 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 2067 struct tcp_fastopen_cookie *cookie, bool syn_lost, 2068 u16 try_exp); 2069 struct tcp_fastopen_request { 2070 /* Fast Open cookie. Size 0 means a cookie request */ 2071 struct tcp_fastopen_cookie cookie; 2072 struct msghdr *data; /* data in MSG_FASTOPEN */ 2073 size_t size; 2074 int copied; /* queued in tcp_connect() */ 2075 struct ubuf_info *uarg; 2076 }; 2077 void tcp_free_fastopen_req(struct tcp_sock *tp); 2078 void tcp_fastopen_destroy_cipher(struct sock *sk); 2079 void tcp_fastopen_ctx_destroy(struct net *net); 2080 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 2081 void *primary_key, void *backup_key); 2082 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 2083 u64 *key); 2084 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 2085 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 2086 struct request_sock *req, 2087 struct tcp_fastopen_cookie *foc, 2088 const struct dst_entry *dst); 2089 void tcp_fastopen_init_key_once(struct net *net); 2090 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 2091 struct tcp_fastopen_cookie *cookie); 2092 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 2093 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 2094 #define TCP_FASTOPEN_KEY_MAX 2 2095 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 2096 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 2097 2098 /* Fastopen key context */ 2099 struct tcp_fastopen_context { 2100 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 2101 int num; 2102 struct rcu_head rcu; 2103 }; 2104 2105 void tcp_fastopen_active_disable(struct sock *sk); 2106 bool tcp_fastopen_active_should_disable(struct sock *sk); 2107 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 2108 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 2109 2110 /* Caller needs to wrap with rcu_read_(un)lock() */ 2111 static inline 2112 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 2113 { 2114 struct tcp_fastopen_context *ctx; 2115 2116 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 2117 if (!ctx) 2118 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 2119 return ctx; 2120 } 2121 2122 static inline 2123 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 2124 const struct tcp_fastopen_cookie *orig) 2125 { 2126 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 2127 orig->len == foc->len && 2128 !memcmp(orig->val, foc->val, foc->len)) 2129 return true; 2130 return false; 2131 } 2132 2133 static inline 2134 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 2135 { 2136 return ctx->num; 2137 } 2138 2139 /* Latencies incurred by various limits for a sender. They are 2140 * chronograph-like stats that are mutually exclusive. 2141 */ 2142 enum tcp_chrono { 2143 TCP_CHRONO_UNSPEC, 2144 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 2145 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 2146 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 2147 __TCP_CHRONO_MAX, 2148 }; 2149 2150 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 2151 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 2152 2153 /* This helper is needed, because skb->tcp_tsorted_anchor uses 2154 * the same memory storage than skb->destructor/_skb_refdst 2155 */ 2156 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 2157 { 2158 skb->destructor = NULL; 2159 skb->_skb_refdst = 0UL; 2160 } 2161 2162 #define tcp_skb_tsorted_save(skb) { \ 2163 unsigned long _save = skb->_skb_refdst; \ 2164 skb->_skb_refdst = 0UL; 2165 2166 #define tcp_skb_tsorted_restore(skb) \ 2167 skb->_skb_refdst = _save; \ 2168 } 2169 2170 void tcp_write_queue_purge(struct sock *sk); 2171 2172 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 2173 { 2174 return skb_rb_first(&sk->tcp_rtx_queue); 2175 } 2176 2177 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 2178 { 2179 return skb_rb_last(&sk->tcp_rtx_queue); 2180 } 2181 2182 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 2183 { 2184 return skb_peek_tail(&sk->sk_write_queue); 2185 } 2186 2187 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 2188 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 2189 2190 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 2191 { 2192 return skb_peek(&sk->sk_write_queue); 2193 } 2194 2195 static inline bool tcp_skb_is_last(const struct sock *sk, 2196 const struct sk_buff *skb) 2197 { 2198 return skb_queue_is_last(&sk->sk_write_queue, skb); 2199 } 2200 2201 /** 2202 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 2203 * @sk: socket 2204 * 2205 * Since the write queue can have a temporary empty skb in it, 2206 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 2207 */ 2208 static inline bool tcp_write_queue_empty(const struct sock *sk) 2209 { 2210 const struct tcp_sock *tp = tcp_sk(sk); 2211 2212 return tp->write_seq == tp->snd_nxt; 2213 } 2214 2215 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 2216 { 2217 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 2218 } 2219 2220 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 2221 { 2222 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 2223 } 2224 2225 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 2226 { 2227 __skb_queue_tail(&sk->sk_write_queue, skb); 2228 2229 /* Queue it, remembering where we must start sending. */ 2230 if (sk->sk_write_queue.next == skb) 2231 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 2232 } 2233 2234 /* Insert new before skb on the write queue of sk. */ 2235 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 2236 struct sk_buff *skb, 2237 struct sock *sk) 2238 { 2239 __skb_queue_before(&sk->sk_write_queue, skb, new); 2240 } 2241 2242 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 2243 { 2244 tcp_skb_tsorted_anchor_cleanup(skb); 2245 __skb_unlink(skb, &sk->sk_write_queue); 2246 } 2247 2248 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 2249 2250 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2251 { 2252 tcp_skb_tsorted_anchor_cleanup(skb); 2253 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2254 } 2255 2256 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2257 { 2258 list_del(&skb->tcp_tsorted_anchor); 2259 tcp_rtx_queue_unlink(skb, sk); 2260 tcp_wmem_free_skb(sk, skb); 2261 } 2262 2263 static inline void tcp_write_collapse_fence(struct sock *sk) 2264 { 2265 struct sk_buff *skb = tcp_write_queue_tail(sk); 2266 2267 if (skb) 2268 TCP_SKB_CB(skb)->eor = 1; 2269 } 2270 2271 static inline void tcp_push_pending_frames(struct sock *sk) 2272 { 2273 if (tcp_send_head(sk)) { 2274 struct tcp_sock *tp = tcp_sk(sk); 2275 2276 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2277 } 2278 } 2279 2280 /* Start sequence of the skb just after the highest skb with SACKed 2281 * bit, valid only if sacked_out > 0 or when the caller has ensured 2282 * validity by itself. 2283 */ 2284 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2285 { 2286 if (!tp->sacked_out) 2287 return tp->snd_una; 2288 2289 if (tp->highest_sack == NULL) 2290 return tp->snd_nxt; 2291 2292 return TCP_SKB_CB(tp->highest_sack)->seq; 2293 } 2294 2295 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2296 { 2297 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2298 } 2299 2300 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2301 { 2302 return tcp_sk(sk)->highest_sack; 2303 } 2304 2305 static inline void tcp_highest_sack_reset(struct sock *sk) 2306 { 2307 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2308 } 2309 2310 /* Called when old skb is about to be deleted and replaced by new skb */ 2311 static inline void tcp_highest_sack_replace(struct sock *sk, 2312 struct sk_buff *old, 2313 struct sk_buff *new) 2314 { 2315 if (old == tcp_highest_sack(sk)) 2316 tcp_sk(sk)->highest_sack = new; 2317 } 2318 2319 /* This helper checks if socket has IP_TRANSPARENT set */ 2320 static inline bool inet_sk_transparent(const struct sock *sk) 2321 { 2322 switch (sk->sk_state) { 2323 case TCP_TIME_WAIT: 2324 return inet_twsk(sk)->tw_transparent; 2325 case TCP_NEW_SYN_RECV: 2326 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2327 } 2328 return inet_test_bit(TRANSPARENT, sk); 2329 } 2330 2331 /* Determines whether this is a thin stream (which may suffer from 2332 * increased latency). Used to trigger latency-reducing mechanisms. 2333 */ 2334 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2335 { 2336 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2337 } 2338 2339 /* /proc */ 2340 enum tcp_seq_states { 2341 TCP_SEQ_STATE_LISTENING, 2342 TCP_SEQ_STATE_ESTABLISHED, 2343 }; 2344 2345 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2346 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2347 void tcp_seq_stop(struct seq_file *seq, void *v); 2348 2349 struct tcp_seq_afinfo { 2350 sa_family_t family; 2351 }; 2352 2353 struct tcp_iter_state { 2354 struct seq_net_private p; 2355 enum tcp_seq_states state; 2356 struct sock *syn_wait_sk; 2357 int bucket, offset, sbucket, num; 2358 loff_t last_pos; 2359 }; 2360 2361 extern struct request_sock_ops tcp_request_sock_ops; 2362 extern struct request_sock_ops tcp6_request_sock_ops; 2363 2364 void tcp_v4_destroy_sock(struct sock *sk); 2365 2366 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2367 netdev_features_t features); 2368 struct sk_buff *tcp_gro_lookup(struct list_head *head, struct tcphdr *th); 2369 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb, 2370 struct tcphdr *th); 2371 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2372 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2373 #ifdef CONFIG_INET 2374 void tcp_gro_complete(struct sk_buff *skb); 2375 #else 2376 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2377 #endif 2378 2379 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2380 2381 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2382 { 2383 struct net *net = sock_net((struct sock *)tp); 2384 u32 val; 2385 2386 val = READ_ONCE(tp->notsent_lowat); 2387 2388 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2389 } 2390 2391 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2392 2393 #ifdef CONFIG_PROC_FS 2394 int tcp4_proc_init(void); 2395 void tcp4_proc_exit(void); 2396 #endif 2397 2398 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2399 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2400 const struct tcp_request_sock_ops *af_ops, 2401 struct sock *sk, struct sk_buff *skb); 2402 2403 /* TCP af-specific functions */ 2404 struct tcp_sock_af_ops { 2405 #ifdef CONFIG_TCP_MD5SIG 2406 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2407 const struct sock *addr_sk); 2408 void (*calc_md5_hash)(char *location, 2409 const struct tcp_md5sig_key *md5, 2410 const struct sock *sk, 2411 const struct sk_buff *skb); 2412 int (*md5_parse)(struct sock *sk, 2413 int optname, 2414 sockptr_t optval, 2415 int optlen); 2416 #endif 2417 #ifdef CONFIG_TCP_AO 2418 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen); 2419 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2420 struct sock *addr_sk, 2421 int sndid, int rcvid); 2422 int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key, 2423 const struct sock *sk, 2424 __be32 sisn, __be32 disn, bool send); 2425 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao, 2426 const struct sock *sk, const struct sk_buff *skb, 2427 const u8 *tkey, int hash_offset, u32 sne); 2428 #endif 2429 }; 2430 2431 struct tcp_request_sock_ops { 2432 u16 mss_clamp; 2433 #ifdef CONFIG_TCP_MD5SIG 2434 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2435 const struct sock *addr_sk); 2436 void (*calc_md5_hash) (char *location, 2437 const struct tcp_md5sig_key *md5, 2438 const struct sock *sk, 2439 const struct sk_buff *skb); 2440 #endif 2441 #ifdef CONFIG_TCP_AO 2442 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2443 struct request_sock *req, 2444 int sndid, int rcvid); 2445 int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk); 2446 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt, 2447 struct request_sock *req, const struct sk_buff *skb, 2448 int hash_offset, u32 sne); 2449 #endif 2450 #ifdef CONFIG_SYN_COOKIES 2451 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2452 __u16 *mss); 2453 #endif 2454 struct dst_entry *(*route_req)(const struct sock *sk, 2455 struct sk_buff *skb, 2456 struct flowi *fl, 2457 struct request_sock *req, 2458 u32 tw_isn); 2459 u32 (*init_seq)(const struct sk_buff *skb); 2460 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2461 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2462 struct flowi *fl, struct request_sock *req, 2463 struct tcp_fastopen_cookie *foc, 2464 enum tcp_synack_type synack_type, 2465 struct sk_buff *syn_skb); 2466 }; 2467 2468 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2469 #if IS_ENABLED(CONFIG_IPV6) 2470 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2471 #endif 2472 2473 #ifdef CONFIG_SYN_COOKIES 2474 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2475 const struct sock *sk, struct sk_buff *skb, 2476 __u16 *mss) 2477 { 2478 tcp_synq_overflow(sk); 2479 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2480 return ops->cookie_init_seq(skb, mss); 2481 } 2482 #else 2483 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2484 const struct sock *sk, struct sk_buff *skb, 2485 __u16 *mss) 2486 { 2487 return 0; 2488 } 2489 #endif 2490 2491 struct tcp_key { 2492 union { 2493 struct { 2494 struct tcp_ao_key *ao_key; 2495 char *traffic_key; 2496 u32 sne; 2497 u8 rcv_next; 2498 }; 2499 struct tcp_md5sig_key *md5_key; 2500 }; 2501 enum { 2502 TCP_KEY_NONE = 0, 2503 TCP_KEY_MD5, 2504 TCP_KEY_AO, 2505 } type; 2506 }; 2507 2508 static inline void tcp_get_current_key(const struct sock *sk, 2509 struct tcp_key *out) 2510 { 2511 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG) 2512 const struct tcp_sock *tp = tcp_sk(sk); 2513 #endif 2514 2515 #ifdef CONFIG_TCP_AO 2516 if (static_branch_unlikely(&tcp_ao_needed.key)) { 2517 struct tcp_ao_info *ao; 2518 2519 ao = rcu_dereference_protected(tp->ao_info, 2520 lockdep_sock_is_held(sk)); 2521 if (ao) { 2522 out->ao_key = READ_ONCE(ao->current_key); 2523 out->type = TCP_KEY_AO; 2524 return; 2525 } 2526 } 2527 #endif 2528 #ifdef CONFIG_TCP_MD5SIG 2529 if (static_branch_unlikely(&tcp_md5_needed.key) && 2530 rcu_access_pointer(tp->md5sig_info)) { 2531 out->md5_key = tp->af_specific->md5_lookup(sk, sk); 2532 if (out->md5_key) { 2533 out->type = TCP_KEY_MD5; 2534 return; 2535 } 2536 } 2537 #endif 2538 out->type = TCP_KEY_NONE; 2539 } 2540 2541 static inline bool tcp_key_is_md5(const struct tcp_key *key) 2542 { 2543 if (static_branch_tcp_md5()) 2544 return key->type == TCP_KEY_MD5; 2545 return false; 2546 } 2547 2548 static inline bool tcp_key_is_ao(const struct tcp_key *key) 2549 { 2550 if (static_branch_tcp_ao()) 2551 return key->type == TCP_KEY_AO; 2552 return false; 2553 } 2554 2555 int tcpv4_offload_init(void); 2556 2557 void tcp_v4_init(void); 2558 void tcp_init(void); 2559 2560 /* tcp_recovery.c */ 2561 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2562 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2563 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2564 u32 reo_wnd); 2565 extern bool tcp_rack_mark_lost(struct sock *sk); 2566 extern void tcp_rack_reo_timeout(struct sock *sk); 2567 2568 /* tcp_plb.c */ 2569 2570 /* 2571 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2572 * expects cong_ratio which represents fraction of traffic that experienced 2573 * congestion over a single RTT. In order to avoid floating point operations, 2574 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2575 */ 2576 #define TCP_PLB_SCALE 8 2577 2578 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2579 struct tcp_plb_state { 2580 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2581 unused:3; 2582 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2583 }; 2584 2585 static inline void tcp_plb_init(const struct sock *sk, 2586 struct tcp_plb_state *plb) 2587 { 2588 plb->consec_cong_rounds = 0; 2589 plb->pause_until = 0; 2590 } 2591 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2592 const int cong_ratio); 2593 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2594 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2595 2596 static inline void tcp_warn_once(const struct sock *sk, bool cond, const char *str) 2597 { 2598 WARN_ONCE(cond, 2599 "%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", 2600 str, 2601 tcp_snd_cwnd(tcp_sk(sk)), 2602 tcp_sk(sk)->packets_out, tcp_sk(sk)->sacked_out, 2603 tcp_sk(sk)->lost_out, tcp_sk(sk)->retrans_out, 2604 tcp_sk(sk)->tlp_high_seq, sk->sk_state, 2605 inet_csk(sk)->icsk_ca_state, 2606 tcp_sk(sk)->advmss, tcp_sk(sk)->mss_cache, 2607 inet_csk(sk)->icsk_pmtu_cookie); 2608 } 2609 2610 /* At how many usecs into the future should the RTO fire? */ 2611 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2612 { 2613 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2614 u32 rto = inet_csk(sk)->icsk_rto; 2615 2616 if (likely(skb)) { 2617 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2618 2619 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2620 } else { 2621 tcp_warn_once(sk, 1, "rtx queue empty: "); 2622 return jiffies_to_usecs(rto); 2623 } 2624 2625 } 2626 2627 /* 2628 * Save and compile IPv4 options, return a pointer to it 2629 */ 2630 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2631 struct sk_buff *skb) 2632 { 2633 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2634 struct ip_options_rcu *dopt = NULL; 2635 2636 if (opt->optlen) { 2637 int opt_size = sizeof(*dopt) + opt->optlen; 2638 2639 dopt = kmalloc(opt_size, GFP_ATOMIC); 2640 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2641 kfree(dopt); 2642 dopt = NULL; 2643 } 2644 } 2645 return dopt; 2646 } 2647 2648 /* locally generated TCP pure ACKs have skb->truesize == 2 2649 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2650 * This is much faster than dissecting the packet to find out. 2651 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2652 */ 2653 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2654 { 2655 return skb->truesize == 2; 2656 } 2657 2658 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2659 { 2660 skb->truesize = 2; 2661 } 2662 2663 static inline int tcp_inq(struct sock *sk) 2664 { 2665 struct tcp_sock *tp = tcp_sk(sk); 2666 int answ; 2667 2668 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2669 answ = 0; 2670 } else if (sock_flag(sk, SOCK_URGINLINE) || 2671 !tp->urg_data || 2672 before(tp->urg_seq, tp->copied_seq) || 2673 !before(tp->urg_seq, tp->rcv_nxt)) { 2674 2675 answ = tp->rcv_nxt - tp->copied_seq; 2676 2677 /* Subtract 1, if FIN was received */ 2678 if (answ && sock_flag(sk, SOCK_DONE)) 2679 answ--; 2680 } else { 2681 answ = tp->urg_seq - tp->copied_seq; 2682 } 2683 2684 return answ; 2685 } 2686 2687 int tcp_peek_len(struct socket *sock); 2688 2689 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2690 { 2691 u16 segs_in; 2692 2693 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2694 2695 /* We update these fields while other threads might 2696 * read them from tcp_get_info() 2697 */ 2698 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2699 if (skb->len > tcp_hdrlen(skb)) 2700 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2701 } 2702 2703 /* 2704 * TCP listen path runs lockless. 2705 * We forced "struct sock" to be const qualified to make sure 2706 * we don't modify one of its field by mistake. 2707 * Here, we increment sk_drops which is an atomic_t, so we can safely 2708 * make sock writable again. 2709 */ 2710 static inline void tcp_listendrop(const struct sock *sk) 2711 { 2712 sk_drops_inc((struct sock *)sk); 2713 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2714 } 2715 2716 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2717 2718 /* 2719 * Interface for adding Upper Level Protocols over TCP 2720 */ 2721 2722 #define TCP_ULP_NAME_MAX 16 2723 #define TCP_ULP_MAX 128 2724 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2725 2726 struct tcp_ulp_ops { 2727 struct list_head list; 2728 2729 /* initialize ulp */ 2730 int (*init)(struct sock *sk); 2731 /* update ulp */ 2732 void (*update)(struct sock *sk, struct proto *p, 2733 void (*write_space)(struct sock *sk)); 2734 /* cleanup ulp */ 2735 void (*release)(struct sock *sk); 2736 /* diagnostic */ 2737 int (*get_info)(struct sock *sk, struct sk_buff *skb, bool net_admin); 2738 size_t (*get_info_size)(const struct sock *sk, bool net_admin); 2739 /* clone ulp */ 2740 void (*clone)(const struct request_sock *req, struct sock *newsk, 2741 const gfp_t priority); 2742 2743 char name[TCP_ULP_NAME_MAX]; 2744 struct module *owner; 2745 }; 2746 int tcp_register_ulp(struct tcp_ulp_ops *type); 2747 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2748 int tcp_set_ulp(struct sock *sk, const char *name); 2749 void tcp_get_available_ulp(char *buf, size_t len); 2750 void tcp_cleanup_ulp(struct sock *sk); 2751 void tcp_update_ulp(struct sock *sk, struct proto *p, 2752 void (*write_space)(struct sock *sk)); 2753 2754 #define MODULE_ALIAS_TCP_ULP(name) \ 2755 MODULE_INFO(alias, name); \ 2756 MODULE_INFO(alias, "tcp-ulp-" name) 2757 2758 #ifdef CONFIG_NET_SOCK_MSG 2759 struct sk_msg; 2760 struct sk_psock; 2761 2762 #ifdef CONFIG_BPF_SYSCALL 2763 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2764 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2765 #ifdef CONFIG_BPF_STREAM_PARSER 2766 struct strparser; 2767 int tcp_bpf_strp_read_sock(struct strparser *strp, read_descriptor_t *desc, 2768 sk_read_actor_t recv_actor); 2769 #endif /* CONFIG_BPF_STREAM_PARSER */ 2770 #endif /* CONFIG_BPF_SYSCALL */ 2771 2772 #ifdef CONFIG_INET 2773 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2774 #else 2775 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2776 { 2777 } 2778 #endif 2779 2780 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2781 struct sk_msg *msg, u32 bytes, int flags); 2782 #endif /* CONFIG_NET_SOCK_MSG */ 2783 2784 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2785 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2786 { 2787 } 2788 #endif 2789 2790 #ifdef CONFIG_CGROUP_BPF 2791 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2792 struct sk_buff *skb, 2793 unsigned int end_offset) 2794 { 2795 skops->skb = skb; 2796 skops->skb_data_end = skb->data + end_offset; 2797 } 2798 #else 2799 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2800 struct sk_buff *skb, 2801 unsigned int end_offset) 2802 { 2803 } 2804 #endif 2805 2806 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2807 * is < 0, then the BPF op failed (for example if the loaded BPF 2808 * program does not support the chosen operation or there is no BPF 2809 * program loaded). 2810 */ 2811 #ifdef CONFIG_BPF 2812 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2813 { 2814 struct bpf_sock_ops_kern sock_ops; 2815 int ret; 2816 2817 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2818 if (sk_fullsock(sk)) { 2819 sock_ops.is_fullsock = 1; 2820 sock_ops.is_locked_tcp_sock = 1; 2821 sock_owned_by_me(sk); 2822 } 2823 2824 sock_ops.sk = sk; 2825 sock_ops.op = op; 2826 if (nargs > 0) 2827 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2828 2829 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2830 if (ret == 0) 2831 ret = sock_ops.reply; 2832 else 2833 ret = -1; 2834 return ret; 2835 } 2836 2837 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2838 { 2839 u32 args[2] = {arg1, arg2}; 2840 2841 return tcp_call_bpf(sk, op, 2, args); 2842 } 2843 2844 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2845 u32 arg3) 2846 { 2847 u32 args[3] = {arg1, arg2, arg3}; 2848 2849 return tcp_call_bpf(sk, op, 3, args); 2850 } 2851 2852 #else 2853 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2854 { 2855 return -EPERM; 2856 } 2857 2858 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2859 { 2860 return -EPERM; 2861 } 2862 2863 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2864 u32 arg3) 2865 { 2866 return -EPERM; 2867 } 2868 2869 #endif 2870 2871 static inline u32 tcp_timeout_init(struct sock *sk) 2872 { 2873 int timeout; 2874 2875 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2876 2877 if (timeout <= 0) 2878 timeout = TCP_TIMEOUT_INIT; 2879 return min_t(int, timeout, TCP_RTO_MAX); 2880 } 2881 2882 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2883 { 2884 int rwnd; 2885 2886 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2887 2888 if (rwnd < 0) 2889 rwnd = 0; 2890 return rwnd; 2891 } 2892 2893 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2894 { 2895 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2896 } 2897 2898 static inline void tcp_bpf_rtt(struct sock *sk, long mrtt, u32 srtt) 2899 { 2900 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2901 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_RTT_CB, mrtt, srtt); 2902 } 2903 2904 #if IS_ENABLED(CONFIG_SMC) 2905 extern struct static_key_false tcp_have_smc; 2906 #endif 2907 2908 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2909 void clean_acked_data_enable(struct tcp_sock *tp, 2910 void (*cad)(struct sock *sk, u32 ack_seq)); 2911 void clean_acked_data_disable(struct tcp_sock *tp); 2912 void clean_acked_data_flush(void); 2913 #endif 2914 2915 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2916 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2917 const struct tcp_sock *tp) 2918 { 2919 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2920 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2921 } 2922 2923 /* Compute Earliest Departure Time for some control packets 2924 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2925 */ 2926 static inline u64 tcp_transmit_time(const struct sock *sk) 2927 { 2928 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2929 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2930 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2931 2932 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2933 } 2934 return 0; 2935 } 2936 2937 static inline int tcp_parse_auth_options(const struct tcphdr *th, 2938 const u8 **md5_hash, const struct tcp_ao_hdr **aoh) 2939 { 2940 const u8 *md5_tmp, *ao_tmp; 2941 int ret; 2942 2943 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp); 2944 if (ret) 2945 return ret; 2946 2947 if (md5_hash) 2948 *md5_hash = md5_tmp; 2949 2950 if (aoh) { 2951 if (!ao_tmp) 2952 *aoh = NULL; 2953 else 2954 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2); 2955 } 2956 2957 return 0; 2958 } 2959 2960 static inline bool tcp_ao_required(struct sock *sk, const void *saddr, 2961 int family, int l3index, bool stat_inc) 2962 { 2963 #ifdef CONFIG_TCP_AO 2964 struct tcp_ao_info *ao_info; 2965 struct tcp_ao_key *ao_key; 2966 2967 if (!static_branch_unlikely(&tcp_ao_needed.key)) 2968 return false; 2969 2970 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info, 2971 lockdep_sock_is_held(sk)); 2972 if (!ao_info) 2973 return false; 2974 2975 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1); 2976 if (ao_info->ao_required || ao_key) { 2977 if (stat_inc) { 2978 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED); 2979 atomic64_inc(&ao_info->counters.ao_required); 2980 } 2981 return true; 2982 } 2983 #endif 2984 return false; 2985 } 2986 2987 enum skb_drop_reason tcp_inbound_hash(struct sock *sk, 2988 const struct request_sock *req, const struct sk_buff *skb, 2989 const void *saddr, const void *daddr, 2990 int family, int dif, int sdif); 2991 2992 #endif /* _TCP_H */ 2993