1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP module. 8 * 9 * Version: @(#)tcp.h 1.0.5 05/23/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 */ 14 #ifndef _TCP_H 15 #define _TCP_H 16 17 #define FASTRETRANS_DEBUG 1 18 19 #include <linux/list.h> 20 #include <linux/tcp.h> 21 #include <linux/bug.h> 22 #include <linux/slab.h> 23 #include <linux/cache.h> 24 #include <linux/percpu.h> 25 #include <linux/skbuff.h> 26 #include <linux/kref.h> 27 #include <linux/ktime.h> 28 #include <linux/indirect_call_wrapper.h> 29 30 #include <net/inet_connection_sock.h> 31 #include <net/inet_timewait_sock.h> 32 #include <net/inet_hashtables.h> 33 #include <net/checksum.h> 34 #include <net/request_sock.h> 35 #include <net/sock_reuseport.h> 36 #include <net/sock.h> 37 #include <net/snmp.h> 38 #include <net/ip.h> 39 #include <net/tcp_states.h> 40 #include <net/tcp_ao.h> 41 #include <net/inet_ecn.h> 42 #include <net/dst.h> 43 #include <net/mptcp.h> 44 45 #include <linux/seq_file.h> 46 #include <linux/memcontrol.h> 47 #include <linux/bpf-cgroup.h> 48 #include <linux/siphash.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 DECLARE_PER_CPU(unsigned int, tcp_orphan_count); 53 int tcp_orphan_count_sum(void); 54 55 DECLARE_PER_CPU(u32, tcp_tw_isn); 56 57 void tcp_time_wait(struct sock *sk, int state, int timeo); 58 59 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) 60 #define MAX_TCP_OPTION_SPACE 40 61 #define TCP_MIN_SND_MSS 48 62 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) 63 64 /* 65 * Never offer a window over 32767 without using window scaling. Some 66 * poor stacks do signed 16bit maths! 67 */ 68 #define MAX_TCP_WINDOW 32767U 69 70 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 71 #define TCP_MIN_MSS 88U 72 73 /* The initial MTU to use for probing */ 74 #define TCP_BASE_MSS 1024 75 76 /* probing interval, default to 10 minutes as per RFC4821 */ 77 #define TCP_PROBE_INTERVAL 600 78 79 /* Specify interval when tcp mtu probing will stop */ 80 #define TCP_PROBE_THRESHOLD 8 81 82 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 83 #define TCP_FASTRETRANS_THRESH 3 84 85 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 86 #define TCP_MAX_QUICKACKS 16U 87 88 /* Maximal number of window scale according to RFC1323 */ 89 #define TCP_MAX_WSCALE 14U 90 91 /* urg_data states */ 92 #define TCP_URG_VALID 0x0100 93 #define TCP_URG_NOTYET 0x0200 94 #define TCP_URG_READ 0x0400 95 96 #define TCP_RETR1 3 /* 97 * This is how many retries it does before it 98 * tries to figure out if the gateway is 99 * down. Minimal RFC value is 3; it corresponds 100 * to ~3sec-8min depending on RTO. 101 */ 102 103 #define TCP_RETR2 15 /* 104 * This should take at least 105 * 90 minutes to time out. 106 * RFC1122 says that the limit is 100 sec. 107 * 15 is ~13-30min depending on RTO. 108 */ 109 110 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 111 * when active opening a connection. 112 * RFC1122 says the minimum retry MUST 113 * be at least 180secs. Nevertheless 114 * this value is corresponding to 115 * 63secs of retransmission with the 116 * current initial RTO. 117 */ 118 119 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 120 * when passive opening a connection. 121 * This is corresponding to 31secs of 122 * retransmission with the current 123 * initial RTO. 124 */ 125 126 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 127 * state, about 60 seconds */ 128 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 129 /* BSD style FIN_WAIT2 deadlock breaker. 130 * It used to be 3min, new value is 60sec, 131 * to combine FIN-WAIT-2 timeout with 132 * TIME-WAIT timer. 133 */ 134 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */ 135 136 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 137 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); 138 139 #if HZ >= 100 140 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 141 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 142 #else 143 #define TCP_DELACK_MIN 4U 144 #define TCP_ATO_MIN 4U 145 #endif 146 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 147 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 148 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 149 150 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */ 151 152 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 153 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 154 * used as a fallback RTO for the 155 * initial data transmission if no 156 * valid RTT sample has been acquired, 157 * most likely due to retrans in 3WHS. 158 */ 159 160 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 161 * for local resources. 162 */ 163 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 164 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 165 #define TCP_KEEPALIVE_INTVL (75*HZ) 166 167 #define MAX_TCP_KEEPIDLE 32767 168 #define MAX_TCP_KEEPINTVL 32767 169 #define MAX_TCP_KEEPCNT 127 170 #define MAX_TCP_SYNCNT 127 171 172 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds 173 * to avoid overflows. This assumes a clock smaller than 1 Mhz. 174 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz. 175 */ 176 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC) 177 178 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 179 * after this time. It should be equal 180 * (or greater than) TCP_TIMEWAIT_LEN 181 * to provide reliability equal to one 182 * provided by timewait state. 183 */ 184 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 185 * timestamps. It must be less than 186 * minimal timewait lifetime. 187 */ 188 /* 189 * TCP option 190 */ 191 192 #define TCPOPT_NOP 1 /* Padding */ 193 #define TCPOPT_EOL 0 /* End of options */ 194 #define TCPOPT_MSS 2 /* Segment size negotiating */ 195 #define TCPOPT_WINDOW 3 /* Window scaling */ 196 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 197 #define TCPOPT_SACK 5 /* SACK Block */ 198 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 199 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 200 #define TCPOPT_AO 29 /* Authentication Option (RFC5925) */ 201 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ 202 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 203 #define TCPOPT_EXP 254 /* Experimental */ 204 /* Magic number to be after the option value for sharing TCP 205 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 206 */ 207 #define TCPOPT_FASTOPEN_MAGIC 0xF989 208 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 209 210 /* 211 * TCP option lengths 212 */ 213 214 #define TCPOLEN_MSS 4 215 #define TCPOLEN_WINDOW 3 216 #define TCPOLEN_SACK_PERM 2 217 #define TCPOLEN_TIMESTAMP 10 218 #define TCPOLEN_MD5SIG 18 219 #define TCPOLEN_FASTOPEN_BASE 2 220 #define TCPOLEN_EXP_FASTOPEN_BASE 4 221 #define TCPOLEN_EXP_SMC_BASE 6 222 223 /* But this is what stacks really send out. */ 224 #define TCPOLEN_TSTAMP_ALIGNED 12 225 #define TCPOLEN_WSCALE_ALIGNED 4 226 #define TCPOLEN_SACKPERM_ALIGNED 4 227 #define TCPOLEN_SACK_BASE 2 228 #define TCPOLEN_SACK_BASE_ALIGNED 4 229 #define TCPOLEN_SACK_PERBLOCK 8 230 #define TCPOLEN_MD5SIG_ALIGNED 20 231 #define TCPOLEN_MSS_ALIGNED 4 232 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 233 234 /* Flags in tp->nonagle */ 235 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 236 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 237 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 238 239 /* TCP thin-stream limits */ 240 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 241 242 /* TCP initial congestion window as per rfc6928 */ 243 #define TCP_INIT_CWND 10 244 245 /* Bit Flags for sysctl_tcp_fastopen */ 246 #define TFO_CLIENT_ENABLE 1 247 #define TFO_SERVER_ENABLE 2 248 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 249 250 /* Accept SYN data w/o any cookie option */ 251 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 252 253 /* Force enable TFO on all listeners, i.e., not requiring the 254 * TCP_FASTOPEN socket option. 255 */ 256 #define TFO_SERVER_WO_SOCKOPT1 0x400 257 258 259 /* sysctl variables for tcp */ 260 extern int sysctl_tcp_max_orphans; 261 extern long sysctl_tcp_mem[3]; 262 263 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 264 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 265 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ 266 267 extern atomic_long_t tcp_memory_allocated; 268 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); 269 270 extern struct percpu_counter tcp_sockets_allocated; 271 extern unsigned long tcp_memory_pressure; 272 273 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 274 static inline bool tcp_under_memory_pressure(const struct sock *sk) 275 { 276 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 277 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 278 return true; 279 280 return READ_ONCE(tcp_memory_pressure); 281 } 282 /* 283 * The next routines deal with comparing 32 bit unsigned ints 284 * and worry about wraparound (automatic with unsigned arithmetic). 285 */ 286 287 static inline bool before(__u32 seq1, __u32 seq2) 288 { 289 return (__s32)(seq1-seq2) < 0; 290 } 291 #define after(seq2, seq1) before(seq1, seq2) 292 293 /* is s2<=s1<=s3 ? */ 294 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 295 { 296 return seq3 - seq2 >= seq1 - seq2; 297 } 298 299 static inline bool tcp_out_of_memory(struct sock *sk) 300 { 301 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 302 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 303 return true; 304 return false; 305 } 306 307 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 308 { 309 sk_wmem_queued_add(sk, -skb->truesize); 310 if (!skb_zcopy_pure(skb)) 311 sk_mem_uncharge(sk, skb->truesize); 312 else 313 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 314 __kfree_skb(skb); 315 } 316 317 void sk_forced_mem_schedule(struct sock *sk, int size); 318 319 bool tcp_check_oom(struct sock *sk, int shift); 320 321 322 extern struct proto tcp_prot; 323 324 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 325 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 326 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 327 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 328 329 void tcp_tasklet_init(void); 330 331 int tcp_v4_err(struct sk_buff *skb, u32); 332 333 void tcp_shutdown(struct sock *sk, int how); 334 335 int tcp_v4_early_demux(struct sk_buff *skb); 336 int tcp_v4_rcv(struct sk_buff *skb); 337 338 void tcp_remove_empty_skb(struct sock *sk); 339 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 340 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 341 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 342 size_t size, struct ubuf_info *uarg); 343 void tcp_splice_eof(struct socket *sock); 344 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 345 int tcp_wmem_schedule(struct sock *sk, int copy); 346 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 347 int size_goal); 348 void tcp_release_cb(struct sock *sk); 349 void tcp_wfree(struct sk_buff *skb); 350 void tcp_write_timer_handler(struct sock *sk); 351 void tcp_delack_timer_handler(struct sock *sk); 352 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 353 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 354 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 355 void tcp_rcv_space_adjust(struct sock *sk); 356 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 357 void tcp_twsk_destructor(struct sock *sk); 358 void tcp_twsk_purge(struct list_head *net_exit_list); 359 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 360 struct pipe_inode_info *pipe, size_t len, 361 unsigned int flags); 362 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 363 bool force_schedule); 364 365 static inline void tcp_dec_quickack_mode(struct sock *sk) 366 { 367 struct inet_connection_sock *icsk = inet_csk(sk); 368 369 if (icsk->icsk_ack.quick) { 370 /* How many ACKs S/ACKing new data have we sent? */ 371 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 372 373 if (pkts >= icsk->icsk_ack.quick) { 374 icsk->icsk_ack.quick = 0; 375 /* Leaving quickack mode we deflate ATO. */ 376 icsk->icsk_ack.ato = TCP_ATO_MIN; 377 } else 378 icsk->icsk_ack.quick -= pkts; 379 } 380 } 381 382 #define TCP_ECN_OK 1 383 #define TCP_ECN_QUEUE_CWR 2 384 #define TCP_ECN_DEMAND_CWR 4 385 #define TCP_ECN_SEEN 8 386 387 enum tcp_tw_status { 388 TCP_TW_SUCCESS = 0, 389 TCP_TW_RST = 1, 390 TCP_TW_ACK = 2, 391 TCP_TW_SYN = 3 392 }; 393 394 395 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 396 struct sk_buff *skb, 397 const struct tcphdr *th, 398 u32 *tw_isn); 399 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 400 struct request_sock *req, bool fastopen, 401 bool *lost_race); 402 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child, 403 struct sk_buff *skb); 404 void tcp_enter_loss(struct sock *sk); 405 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 406 void tcp_clear_retrans(struct tcp_sock *tp); 407 void tcp_update_metrics(struct sock *sk); 408 void tcp_init_metrics(struct sock *sk); 409 void tcp_metrics_init(void); 410 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 411 void __tcp_close(struct sock *sk, long timeout); 412 void tcp_close(struct sock *sk, long timeout); 413 void tcp_init_sock(struct sock *sk); 414 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 415 __poll_t tcp_poll(struct file *file, struct socket *sock, 416 struct poll_table_struct *wait); 417 int do_tcp_getsockopt(struct sock *sk, int level, 418 int optname, sockptr_t optval, sockptr_t optlen); 419 int tcp_getsockopt(struct sock *sk, int level, int optname, 420 char __user *optval, int __user *optlen); 421 bool tcp_bpf_bypass_getsockopt(int level, int optname); 422 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 423 sockptr_t optval, unsigned int optlen); 424 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 425 unsigned int optlen); 426 void tcp_set_keepalive(struct sock *sk, int val); 427 void tcp_syn_ack_timeout(const struct request_sock *req); 428 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 429 int flags, int *addr_len); 430 int tcp_set_rcvlowat(struct sock *sk, int val); 431 int tcp_set_window_clamp(struct sock *sk, int val); 432 void tcp_update_recv_tstamps(struct sk_buff *skb, 433 struct scm_timestamping_internal *tss); 434 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 435 struct scm_timestamping_internal *tss); 436 void tcp_data_ready(struct sock *sk); 437 #ifdef CONFIG_MMU 438 int tcp_mmap(struct file *file, struct socket *sock, 439 struct vm_area_struct *vma); 440 #endif 441 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 442 struct tcp_options_received *opt_rx, 443 int estab, struct tcp_fastopen_cookie *foc); 444 445 /* 446 * BPF SKB-less helpers 447 */ 448 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 449 struct tcphdr *th, u32 *cookie); 450 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 451 struct tcphdr *th, u32 *cookie); 452 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 453 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 454 const struct tcp_request_sock_ops *af_ops, 455 struct sock *sk, struct tcphdr *th); 456 /* 457 * TCP v4 functions exported for the inet6 API 458 */ 459 460 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 461 void tcp_v4_mtu_reduced(struct sock *sk); 462 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 463 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 464 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 465 struct sock *tcp_create_openreq_child(const struct sock *sk, 466 struct request_sock *req, 467 struct sk_buff *skb); 468 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 469 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 470 struct request_sock *req, 471 struct dst_entry *dst, 472 struct request_sock *req_unhash, 473 bool *own_req); 474 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 475 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 476 int tcp_connect(struct sock *sk); 477 enum tcp_synack_type { 478 TCP_SYNACK_NORMAL, 479 TCP_SYNACK_FASTOPEN, 480 TCP_SYNACK_COOKIE, 481 }; 482 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 483 struct request_sock *req, 484 struct tcp_fastopen_cookie *foc, 485 enum tcp_synack_type synack_type, 486 struct sk_buff *syn_skb); 487 int tcp_disconnect(struct sock *sk, int flags); 488 489 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 490 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 491 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 492 493 /* From syncookies.c */ 494 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 495 struct request_sock *req, 496 struct dst_entry *dst); 497 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th); 498 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 499 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 500 struct sock *sk, struct sk_buff *skb, 501 struct tcp_options_received *tcp_opt, 502 int mss, u32 tsoff); 503 504 #if IS_ENABLED(CONFIG_BPF) 505 struct bpf_tcp_req_attrs { 506 u32 rcv_tsval; 507 u32 rcv_tsecr; 508 u16 mss; 509 u8 rcv_wscale; 510 u8 snd_wscale; 511 u8 ecn_ok; 512 u8 wscale_ok; 513 u8 sack_ok; 514 u8 tstamp_ok; 515 u8 usec_ts_ok; 516 u8 reserved[3]; 517 }; 518 #endif 519 520 #ifdef CONFIG_SYN_COOKIES 521 522 /* Syncookies use a monotonic timer which increments every 60 seconds. 523 * This counter is used both as a hash input and partially encoded into 524 * the cookie value. A cookie is only validated further if the delta 525 * between the current counter value and the encoded one is less than this, 526 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 527 * the counter advances immediately after a cookie is generated). 528 */ 529 #define MAX_SYNCOOKIE_AGE 2 530 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 531 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 532 533 /* syncookies: remember time of last synqueue overflow 534 * But do not dirty this field too often (once per second is enough) 535 * It is racy as we do not hold a lock, but race is very minor. 536 */ 537 static inline void tcp_synq_overflow(const struct sock *sk) 538 { 539 unsigned int last_overflow; 540 unsigned int now = jiffies; 541 542 if (sk->sk_reuseport) { 543 struct sock_reuseport *reuse; 544 545 reuse = rcu_dereference(sk->sk_reuseport_cb); 546 if (likely(reuse)) { 547 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 548 if (!time_between32(now, last_overflow, 549 last_overflow + HZ)) 550 WRITE_ONCE(reuse->synq_overflow_ts, now); 551 return; 552 } 553 } 554 555 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 556 if (!time_between32(now, last_overflow, last_overflow + HZ)) 557 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 558 } 559 560 /* syncookies: no recent synqueue overflow on this listening socket? */ 561 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 562 { 563 unsigned int last_overflow; 564 unsigned int now = jiffies; 565 566 if (sk->sk_reuseport) { 567 struct sock_reuseport *reuse; 568 569 reuse = rcu_dereference(sk->sk_reuseport_cb); 570 if (likely(reuse)) { 571 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 572 return !time_between32(now, last_overflow - HZ, 573 last_overflow + 574 TCP_SYNCOOKIE_VALID); 575 } 576 } 577 578 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 579 580 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 581 * then we're under synflood. However, we have to use 582 * 'last_overflow - HZ' as lower bound. That's because a concurrent 583 * tcp_synq_overflow() could update .ts_recent_stamp after we read 584 * jiffies but before we store .ts_recent_stamp into last_overflow, 585 * which could lead to rejecting a valid syncookie. 586 */ 587 return !time_between32(now, last_overflow - HZ, 588 last_overflow + TCP_SYNCOOKIE_VALID); 589 } 590 591 static inline u32 tcp_cookie_time(void) 592 { 593 u64 val = get_jiffies_64(); 594 595 do_div(val, TCP_SYNCOOKIE_PERIOD); 596 return val; 597 } 598 599 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */ 600 static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val) 601 { 602 if (usec_ts) 603 return div_u64(val, NSEC_PER_USEC); 604 605 return div_u64(val, NSEC_PER_MSEC); 606 } 607 608 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 609 u16 *mssp); 610 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 611 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 612 bool cookie_timestamp_decode(const struct net *net, 613 struct tcp_options_received *opt); 614 615 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst) 616 { 617 return READ_ONCE(net->ipv4.sysctl_tcp_ecn) || 618 dst_feature(dst, RTAX_FEATURE_ECN); 619 } 620 621 #if IS_ENABLED(CONFIG_BPF) 622 static inline bool cookie_bpf_ok(struct sk_buff *skb) 623 { 624 return skb->sk; 625 } 626 627 struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb); 628 #else 629 static inline bool cookie_bpf_ok(struct sk_buff *skb) 630 { 631 return false; 632 } 633 634 static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk, 635 struct sk_buff *skb) 636 { 637 return NULL; 638 } 639 #endif 640 641 /* From net/ipv6/syncookies.c */ 642 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th); 643 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 644 645 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 646 const struct tcphdr *th, u16 *mssp); 647 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 648 #endif 649 /* tcp_output.c */ 650 651 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 652 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 653 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 654 int nonagle); 655 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 656 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 657 void tcp_retransmit_timer(struct sock *sk); 658 void tcp_xmit_retransmit_queue(struct sock *); 659 void tcp_simple_retransmit(struct sock *); 660 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 661 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 662 enum tcp_queue { 663 TCP_FRAG_IN_WRITE_QUEUE, 664 TCP_FRAG_IN_RTX_QUEUE, 665 }; 666 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 667 struct sk_buff *skb, u32 len, 668 unsigned int mss_now, gfp_t gfp); 669 670 void tcp_send_probe0(struct sock *); 671 int tcp_write_wakeup(struct sock *, int mib); 672 void tcp_send_fin(struct sock *sk); 673 void tcp_send_active_reset(struct sock *sk, gfp_t priority, 674 enum sk_rst_reason reason); 675 int tcp_send_synack(struct sock *); 676 void tcp_push_one(struct sock *, unsigned int mss_now); 677 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 678 void tcp_send_ack(struct sock *sk); 679 void tcp_send_delayed_ack(struct sock *sk); 680 void tcp_send_loss_probe(struct sock *sk); 681 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 682 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 683 const struct sk_buff *next_skb); 684 685 /* tcp_input.c */ 686 void tcp_rearm_rto(struct sock *sk); 687 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 688 void tcp_reset(struct sock *sk, struct sk_buff *skb); 689 void tcp_fin(struct sock *sk); 690 void tcp_check_space(struct sock *sk); 691 void tcp_sack_compress_send_ack(struct sock *sk); 692 693 /* tcp_timer.c */ 694 void tcp_init_xmit_timers(struct sock *); 695 static inline void tcp_clear_xmit_timers(struct sock *sk) 696 { 697 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 698 __sock_put(sk); 699 700 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 701 __sock_put(sk); 702 703 inet_csk_clear_xmit_timers(sk); 704 } 705 706 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 707 unsigned int tcp_current_mss(struct sock *sk); 708 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 709 710 /* Bound MSS / TSO packet size with the half of the window */ 711 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 712 { 713 int cutoff; 714 715 /* When peer uses tiny windows, there is no use in packetizing 716 * to sub-MSS pieces for the sake of SWS or making sure there 717 * are enough packets in the pipe for fast recovery. 718 * 719 * On the other hand, for extremely large MSS devices, handling 720 * smaller than MSS windows in this way does make sense. 721 */ 722 if (tp->max_window > TCP_MSS_DEFAULT) 723 cutoff = (tp->max_window >> 1); 724 else 725 cutoff = tp->max_window; 726 727 if (cutoff && pktsize > cutoff) 728 return max_t(int, cutoff, 68U - tp->tcp_header_len); 729 else 730 return pktsize; 731 } 732 733 /* tcp.c */ 734 void tcp_get_info(struct sock *, struct tcp_info *); 735 736 /* Read 'sendfile()'-style from a TCP socket */ 737 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 738 sk_read_actor_t recv_actor); 739 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 740 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 741 void tcp_read_done(struct sock *sk, size_t len); 742 743 void tcp_initialize_rcv_mss(struct sock *sk); 744 745 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 746 int tcp_mss_to_mtu(struct sock *sk, int mss); 747 void tcp_mtup_init(struct sock *sk); 748 749 static inline void tcp_bound_rto(struct sock *sk) 750 { 751 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 752 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 753 } 754 755 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 756 { 757 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 758 } 759 760 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 761 { 762 /* mptcp hooks are only on the slow path */ 763 if (sk_is_mptcp((struct sock *)tp)) 764 return; 765 766 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 767 ntohl(TCP_FLAG_ACK) | 768 snd_wnd); 769 } 770 771 static inline void tcp_fast_path_on(struct tcp_sock *tp) 772 { 773 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 774 } 775 776 static inline void tcp_fast_path_check(struct sock *sk) 777 { 778 struct tcp_sock *tp = tcp_sk(sk); 779 780 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 781 tp->rcv_wnd && 782 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 783 !tp->urg_data) 784 tcp_fast_path_on(tp); 785 } 786 787 u32 tcp_delack_max(const struct sock *sk); 788 789 /* Compute the actual rto_min value */ 790 static inline u32 tcp_rto_min(const struct sock *sk) 791 { 792 const struct dst_entry *dst = __sk_dst_get(sk); 793 u32 rto_min = inet_csk(sk)->icsk_rto_min; 794 795 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 796 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 797 return rto_min; 798 } 799 800 static inline u32 tcp_rto_min_us(const struct sock *sk) 801 { 802 return jiffies_to_usecs(tcp_rto_min(sk)); 803 } 804 805 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 806 { 807 return dst_metric_locked(dst, RTAX_CC_ALGO); 808 } 809 810 /* Minimum RTT in usec. ~0 means not available. */ 811 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 812 { 813 return minmax_get(&tp->rtt_min); 814 } 815 816 /* Compute the actual receive window we are currently advertising. 817 * Rcv_nxt can be after the window if our peer push more data 818 * than the offered window. 819 */ 820 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 821 { 822 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 823 824 if (win < 0) 825 win = 0; 826 return (u32) win; 827 } 828 829 /* Choose a new window, without checks for shrinking, and without 830 * scaling applied to the result. The caller does these things 831 * if necessary. This is a "raw" window selection. 832 */ 833 u32 __tcp_select_window(struct sock *sk); 834 835 void tcp_send_window_probe(struct sock *sk); 836 837 /* TCP uses 32bit jiffies to save some space. 838 * Note that this is different from tcp_time_stamp, which 839 * historically has been the same until linux-4.13. 840 */ 841 #define tcp_jiffies32 ((u32)jiffies) 842 843 /* 844 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 845 * It is no longer tied to jiffies, but to 1 ms clock. 846 * Note: double check if you want to use tcp_jiffies32 instead of this. 847 */ 848 #define TCP_TS_HZ 1000 849 850 static inline u64 tcp_clock_ns(void) 851 { 852 return ktime_get_ns(); 853 } 854 855 static inline u64 tcp_clock_us(void) 856 { 857 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 858 } 859 860 static inline u64 tcp_clock_ms(void) 861 { 862 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 863 } 864 865 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 866 * or usec resolution. Each socket carries a flag to select one or other 867 * resolution, as the route attribute could change anytime. 868 * Each flow must stick to initial resolution. 869 */ 870 static inline u32 tcp_clock_ts(bool usec_ts) 871 { 872 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 873 } 874 875 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 876 { 877 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 878 } 879 880 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 881 { 882 if (tp->tcp_usec_ts) 883 return tp->tcp_mstamp; 884 return tcp_time_stamp_ms(tp); 885 } 886 887 void tcp_mstamp_refresh(struct tcp_sock *tp); 888 889 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 890 { 891 return max_t(s64, t1 - t0, 0); 892 } 893 894 /* provide the departure time in us unit */ 895 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 896 { 897 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 898 } 899 900 /* Provide skb TSval in usec or ms unit */ 901 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 902 { 903 if (usec_ts) 904 return tcp_skb_timestamp_us(skb); 905 906 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 907 } 908 909 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 910 { 911 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 912 } 913 914 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 915 { 916 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 917 } 918 919 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 920 921 #define TCPHDR_FIN 0x01 922 #define TCPHDR_SYN 0x02 923 #define TCPHDR_RST 0x04 924 #define TCPHDR_PSH 0x08 925 #define TCPHDR_ACK 0x10 926 #define TCPHDR_URG 0x20 927 #define TCPHDR_ECE 0x40 928 #define TCPHDR_CWR 0x80 929 930 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 931 932 /* State flags for sacked in struct tcp_skb_cb */ 933 enum tcp_skb_cb_sacked_flags { 934 TCPCB_SACKED_ACKED = (1 << 0), /* SKB ACK'd by a SACK block */ 935 TCPCB_SACKED_RETRANS = (1 << 1), /* SKB retransmitted */ 936 TCPCB_LOST = (1 << 2), /* SKB is lost */ 937 TCPCB_TAGBITS = (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS | 938 TCPCB_LOST), /* All tag bits */ 939 TCPCB_REPAIRED = (1 << 4), /* SKB repaired (no skb_mstamp_ns) */ 940 TCPCB_EVER_RETRANS = (1 << 7), /* Ever retransmitted frame */ 941 TCPCB_RETRANS = (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS | 942 TCPCB_REPAIRED), 943 }; 944 945 /* This is what the send packet queuing engine uses to pass 946 * TCP per-packet control information to the transmission code. 947 * We also store the host-order sequence numbers in here too. 948 * This is 44 bytes if IPV6 is enabled. 949 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 950 */ 951 struct tcp_skb_cb { 952 __u32 seq; /* Starting sequence number */ 953 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 954 union { 955 /* Note : 956 * tcp_gso_segs/size are used in write queue only, 957 * cf tcp_skb_pcount()/tcp_skb_mss() 958 */ 959 struct { 960 u16 tcp_gso_segs; 961 u16 tcp_gso_size; 962 }; 963 }; 964 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 965 966 __u8 sacked; /* State flags for SACK. */ 967 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 968 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 969 eor:1, /* Is skb MSG_EOR marked? */ 970 has_rxtstamp:1, /* SKB has a RX timestamp */ 971 unused:5; 972 __u32 ack_seq; /* Sequence number ACK'd */ 973 union { 974 struct { 975 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 976 /* There is space for up to 24 bytes */ 977 __u32 is_app_limited:1, /* cwnd not fully used? */ 978 delivered_ce:20, 979 unused:11; 980 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 981 __u32 delivered; 982 /* start of send pipeline phase */ 983 u64 first_tx_mstamp; 984 /* when we reached the "delivered" count */ 985 u64 delivered_mstamp; 986 } tx; /* only used for outgoing skbs */ 987 union { 988 struct inet_skb_parm h4; 989 #if IS_ENABLED(CONFIG_IPV6) 990 struct inet6_skb_parm h6; 991 #endif 992 } header; /* For incoming skbs */ 993 }; 994 }; 995 996 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 997 998 extern const struct inet_connection_sock_af_ops ipv4_specific; 999 1000 #if IS_ENABLED(CONFIG_IPV6) 1001 /* This is the variant of inet6_iif() that must be used by TCP, 1002 * as TCP moves IP6CB into a different location in skb->cb[] 1003 */ 1004 static inline int tcp_v6_iif(const struct sk_buff *skb) 1005 { 1006 return TCP_SKB_CB(skb)->header.h6.iif; 1007 } 1008 1009 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 1010 { 1011 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 1012 1013 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 1014 } 1015 1016 /* TCP_SKB_CB reference means this can not be used from early demux */ 1017 static inline int tcp_v6_sdif(const struct sk_buff *skb) 1018 { 1019 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1020 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 1021 return TCP_SKB_CB(skb)->header.h6.iif; 1022 #endif 1023 return 0; 1024 } 1025 1026 extern const struct inet_connection_sock_af_ops ipv6_specific; 1027 1028 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 1029 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 1030 void tcp_v6_early_demux(struct sk_buff *skb); 1031 1032 #endif 1033 1034 /* TCP_SKB_CB reference means this can not be used from early demux */ 1035 static inline int tcp_v4_sdif(struct sk_buff *skb) 1036 { 1037 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1038 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 1039 return TCP_SKB_CB(skb)->header.h4.iif; 1040 #endif 1041 return 0; 1042 } 1043 1044 /* Due to TSO, an SKB can be composed of multiple actual 1045 * packets. To keep these tracked properly, we use this. 1046 */ 1047 static inline int tcp_skb_pcount(const struct sk_buff *skb) 1048 { 1049 return TCP_SKB_CB(skb)->tcp_gso_segs; 1050 } 1051 1052 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 1053 { 1054 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1055 } 1056 1057 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1058 { 1059 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1060 } 1061 1062 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1063 static inline int tcp_skb_mss(const struct sk_buff *skb) 1064 { 1065 return TCP_SKB_CB(skb)->tcp_gso_size; 1066 } 1067 1068 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1069 { 1070 return likely(!TCP_SKB_CB(skb)->eor); 1071 } 1072 1073 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1074 const struct sk_buff *from) 1075 { 1076 return likely(tcp_skb_can_collapse_to(to) && 1077 mptcp_skb_can_collapse(to, from) && 1078 skb_pure_zcopy_same(to, from)); 1079 } 1080 1081 /* Events passed to congestion control interface */ 1082 enum tcp_ca_event { 1083 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1084 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1085 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1086 CA_EVENT_LOSS, /* loss timeout */ 1087 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1088 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1089 }; 1090 1091 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1092 enum tcp_ca_ack_event_flags { 1093 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1094 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1095 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1096 }; 1097 1098 /* 1099 * Interface for adding new TCP congestion control handlers 1100 */ 1101 #define TCP_CA_NAME_MAX 16 1102 #define TCP_CA_MAX 128 1103 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1104 1105 #define TCP_CA_UNSPEC 0 1106 1107 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1108 #define TCP_CONG_NON_RESTRICTED 0x1 1109 /* Requires ECN/ECT set on all packets */ 1110 #define TCP_CONG_NEEDS_ECN 0x2 1111 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1112 1113 union tcp_cc_info; 1114 1115 struct ack_sample { 1116 u32 pkts_acked; 1117 s32 rtt_us; 1118 u32 in_flight; 1119 }; 1120 1121 /* A rate sample measures the number of (original/retransmitted) data 1122 * packets delivered "delivered" over an interval of time "interval_us". 1123 * The tcp_rate.c code fills in the rate sample, and congestion 1124 * control modules that define a cong_control function to run at the end 1125 * of ACK processing can optionally chose to consult this sample when 1126 * setting cwnd and pacing rate. 1127 * A sample is invalid if "delivered" or "interval_us" is negative. 1128 */ 1129 struct rate_sample { 1130 u64 prior_mstamp; /* starting timestamp for interval */ 1131 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1132 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1133 s32 delivered; /* number of packets delivered over interval */ 1134 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1135 long interval_us; /* time for tp->delivered to incr "delivered" */ 1136 u32 snd_interval_us; /* snd interval for delivered packets */ 1137 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1138 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1139 int losses; /* number of packets marked lost upon ACK */ 1140 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1141 u32 prior_in_flight; /* in flight before this ACK */ 1142 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1143 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1144 bool is_retrans; /* is sample from retransmission? */ 1145 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1146 }; 1147 1148 struct tcp_congestion_ops { 1149 /* fast path fields are put first to fill one cache line */ 1150 1151 /* return slow start threshold (required) */ 1152 u32 (*ssthresh)(struct sock *sk); 1153 1154 /* do new cwnd calculation (required) */ 1155 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1156 1157 /* call before changing ca_state (optional) */ 1158 void (*set_state)(struct sock *sk, u8 new_state); 1159 1160 /* call when cwnd event occurs (optional) */ 1161 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1162 1163 /* call when ack arrives (optional) */ 1164 void (*in_ack_event)(struct sock *sk, u32 flags); 1165 1166 /* hook for packet ack accounting (optional) */ 1167 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1168 1169 /* override sysctl_tcp_min_tso_segs */ 1170 u32 (*min_tso_segs)(struct sock *sk); 1171 1172 /* call when packets are delivered to update cwnd and pacing rate, 1173 * after all the ca_state processing. (optional) 1174 */ 1175 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 1176 1177 1178 /* new value of cwnd after loss (required) */ 1179 u32 (*undo_cwnd)(struct sock *sk); 1180 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1181 u32 (*sndbuf_expand)(struct sock *sk); 1182 1183 /* control/slow paths put last */ 1184 /* get info for inet_diag (optional) */ 1185 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1186 union tcp_cc_info *info); 1187 1188 char name[TCP_CA_NAME_MAX]; 1189 struct module *owner; 1190 struct list_head list; 1191 u32 key; 1192 u32 flags; 1193 1194 /* initialize private data (optional) */ 1195 void (*init)(struct sock *sk); 1196 /* cleanup private data (optional) */ 1197 void (*release)(struct sock *sk); 1198 } ____cacheline_aligned_in_smp; 1199 1200 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1201 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1202 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1203 struct tcp_congestion_ops *old_type); 1204 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1205 1206 void tcp_assign_congestion_control(struct sock *sk); 1207 void tcp_init_congestion_control(struct sock *sk); 1208 void tcp_cleanup_congestion_control(struct sock *sk); 1209 int tcp_set_default_congestion_control(struct net *net, const char *name); 1210 void tcp_get_default_congestion_control(struct net *net, char *name); 1211 void tcp_get_available_congestion_control(char *buf, size_t len); 1212 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1213 int tcp_set_allowed_congestion_control(char *allowed); 1214 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1215 bool cap_net_admin); 1216 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1217 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1218 1219 u32 tcp_reno_ssthresh(struct sock *sk); 1220 u32 tcp_reno_undo_cwnd(struct sock *sk); 1221 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1222 extern struct tcp_congestion_ops tcp_reno; 1223 1224 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1225 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1226 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); 1227 #ifdef CONFIG_INET 1228 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1229 #else 1230 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1231 { 1232 return NULL; 1233 } 1234 #endif 1235 1236 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1237 { 1238 const struct inet_connection_sock *icsk = inet_csk(sk); 1239 1240 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1241 } 1242 1243 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1244 { 1245 const struct inet_connection_sock *icsk = inet_csk(sk); 1246 1247 if (icsk->icsk_ca_ops->cwnd_event) 1248 icsk->icsk_ca_ops->cwnd_event(sk, event); 1249 } 1250 1251 /* From tcp_cong.c */ 1252 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1253 1254 /* From tcp_rate.c */ 1255 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1256 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1257 struct rate_sample *rs); 1258 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1259 bool is_sack_reneg, struct rate_sample *rs); 1260 void tcp_rate_check_app_limited(struct sock *sk); 1261 1262 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1263 { 1264 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1265 } 1266 1267 /* These functions determine how the current flow behaves in respect of SACK 1268 * handling. SACK is negotiated with the peer, and therefore it can vary 1269 * between different flows. 1270 * 1271 * tcp_is_sack - SACK enabled 1272 * tcp_is_reno - No SACK 1273 */ 1274 static inline int tcp_is_sack(const struct tcp_sock *tp) 1275 { 1276 return likely(tp->rx_opt.sack_ok); 1277 } 1278 1279 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1280 { 1281 return !tcp_is_sack(tp); 1282 } 1283 1284 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1285 { 1286 return tp->sacked_out + tp->lost_out; 1287 } 1288 1289 /* This determines how many packets are "in the network" to the best 1290 * of our knowledge. In many cases it is conservative, but where 1291 * detailed information is available from the receiver (via SACK 1292 * blocks etc.) we can make more aggressive calculations. 1293 * 1294 * Use this for decisions involving congestion control, use just 1295 * tp->packets_out to determine if the send queue is empty or not. 1296 * 1297 * Read this equation as: 1298 * 1299 * "Packets sent once on transmission queue" MINUS 1300 * "Packets left network, but not honestly ACKed yet" PLUS 1301 * "Packets fast retransmitted" 1302 */ 1303 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1304 { 1305 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1306 } 1307 1308 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1309 1310 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1311 { 1312 return tp->snd_cwnd; 1313 } 1314 1315 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1316 { 1317 WARN_ON_ONCE((int)val <= 0); 1318 tp->snd_cwnd = val; 1319 } 1320 1321 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1322 { 1323 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1324 } 1325 1326 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1327 { 1328 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1329 } 1330 1331 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1332 { 1333 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1334 (1 << inet_csk(sk)->icsk_ca_state); 1335 } 1336 1337 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1338 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1339 * ssthresh. 1340 */ 1341 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1342 { 1343 const struct tcp_sock *tp = tcp_sk(sk); 1344 1345 if (tcp_in_cwnd_reduction(sk)) 1346 return tp->snd_ssthresh; 1347 else 1348 return max(tp->snd_ssthresh, 1349 ((tcp_snd_cwnd(tp) >> 1) + 1350 (tcp_snd_cwnd(tp) >> 2))); 1351 } 1352 1353 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1354 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1355 1356 void tcp_enter_cwr(struct sock *sk); 1357 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1358 1359 /* The maximum number of MSS of available cwnd for which TSO defers 1360 * sending if not using sysctl_tcp_tso_win_divisor. 1361 */ 1362 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1363 { 1364 return 3; 1365 } 1366 1367 /* Returns end sequence number of the receiver's advertised window */ 1368 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1369 { 1370 return tp->snd_una + tp->snd_wnd; 1371 } 1372 1373 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1374 * flexible approach. The RFC suggests cwnd should not be raised unless 1375 * it was fully used previously. And that's exactly what we do in 1376 * congestion avoidance mode. But in slow start we allow cwnd to grow 1377 * as long as the application has used half the cwnd. 1378 * Example : 1379 * cwnd is 10 (IW10), but application sends 9 frames. 1380 * We allow cwnd to reach 18 when all frames are ACKed. 1381 * This check is safe because it's as aggressive as slow start which already 1382 * risks 100% overshoot. The advantage is that we discourage application to 1383 * either send more filler packets or data to artificially blow up the cwnd 1384 * usage, and allow application-limited process to probe bw more aggressively. 1385 */ 1386 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1387 { 1388 const struct tcp_sock *tp = tcp_sk(sk); 1389 1390 if (tp->is_cwnd_limited) 1391 return true; 1392 1393 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1394 if (tcp_in_slow_start(tp)) 1395 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1396 1397 return false; 1398 } 1399 1400 /* BBR congestion control needs pacing. 1401 * Same remark for SO_MAX_PACING_RATE. 1402 * sch_fq packet scheduler is efficiently handling pacing, 1403 * but is not always installed/used. 1404 * Return true if TCP stack should pace packets itself. 1405 */ 1406 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1407 { 1408 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1409 } 1410 1411 /* Estimates in how many jiffies next packet for this flow can be sent. 1412 * Scheduling a retransmit timer too early would be silly. 1413 */ 1414 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1415 { 1416 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1417 1418 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1419 } 1420 1421 static inline void tcp_reset_xmit_timer(struct sock *sk, 1422 const int what, 1423 unsigned long when, 1424 const unsigned long max_when) 1425 { 1426 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1427 max_when); 1428 } 1429 1430 /* Something is really bad, we could not queue an additional packet, 1431 * because qdisc is full or receiver sent a 0 window, or we are paced. 1432 * We do not want to add fuel to the fire, or abort too early, 1433 * so make sure the timer we arm now is at least 200ms in the future, 1434 * regardless of current icsk_rto value (as it could be ~2ms) 1435 */ 1436 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1437 { 1438 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1439 } 1440 1441 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1442 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1443 unsigned long max_when) 1444 { 1445 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1446 inet_csk(sk)->icsk_backoff); 1447 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1448 1449 return (unsigned long)min_t(u64, when, max_when); 1450 } 1451 1452 static inline void tcp_check_probe_timer(struct sock *sk) 1453 { 1454 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1455 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1456 tcp_probe0_base(sk), TCP_RTO_MAX); 1457 } 1458 1459 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1460 { 1461 tp->snd_wl1 = seq; 1462 } 1463 1464 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1465 { 1466 tp->snd_wl1 = seq; 1467 } 1468 1469 /* 1470 * Calculate(/check) TCP checksum 1471 */ 1472 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1473 __be32 daddr, __wsum base) 1474 { 1475 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1476 } 1477 1478 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1479 { 1480 return !skb_csum_unnecessary(skb) && 1481 __skb_checksum_complete(skb); 1482 } 1483 1484 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1485 enum skb_drop_reason *reason); 1486 1487 1488 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1489 void tcp_set_state(struct sock *sk, int state); 1490 void tcp_done(struct sock *sk); 1491 int tcp_abort(struct sock *sk, int err); 1492 1493 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1494 { 1495 rx_opt->dsack = 0; 1496 rx_opt->num_sacks = 0; 1497 } 1498 1499 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1500 1501 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1502 { 1503 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1504 struct tcp_sock *tp = tcp_sk(sk); 1505 s32 delta; 1506 1507 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1508 tp->packets_out || ca_ops->cong_control) 1509 return; 1510 delta = tcp_jiffies32 - tp->lsndtime; 1511 if (delta > inet_csk(sk)->icsk_rto) 1512 tcp_cwnd_restart(sk, delta); 1513 } 1514 1515 /* Determine a window scaling and initial window to offer. */ 1516 void tcp_select_initial_window(const struct sock *sk, int __space, 1517 __u32 mss, __u32 *rcv_wnd, 1518 __u32 *window_clamp, int wscale_ok, 1519 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1520 1521 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1522 { 1523 s64 scaled_space = (s64)space * scaling_ratio; 1524 1525 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1526 } 1527 1528 static inline int tcp_win_from_space(const struct sock *sk, int space) 1529 { 1530 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1531 } 1532 1533 /* inverse of __tcp_win_from_space() */ 1534 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1535 { 1536 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1537 1538 do_div(val, scaling_ratio); 1539 return val; 1540 } 1541 1542 static inline int tcp_space_from_win(const struct sock *sk, int win) 1543 { 1544 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1545 } 1546 1547 /* Assume a 50% default for skb->len/skb->truesize ratio. 1548 * This may be adjusted later in tcp_measure_rcv_mss(). 1549 */ 1550 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1)) 1551 1552 static inline void tcp_scaling_ratio_init(struct sock *sk) 1553 { 1554 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1555 } 1556 1557 /* Note: caller must be prepared to deal with negative returns */ 1558 static inline int tcp_space(const struct sock *sk) 1559 { 1560 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1561 READ_ONCE(sk->sk_backlog.len) - 1562 atomic_read(&sk->sk_rmem_alloc)); 1563 } 1564 1565 static inline int tcp_full_space(const struct sock *sk) 1566 { 1567 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1568 } 1569 1570 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh) 1571 { 1572 int unused_mem = sk_unused_reserved_mem(sk); 1573 struct tcp_sock *tp = tcp_sk(sk); 1574 1575 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh); 1576 if (unused_mem) 1577 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1578 tcp_win_from_space(sk, unused_mem)); 1579 } 1580 1581 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1582 { 1583 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss); 1584 } 1585 1586 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1587 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1588 1589 1590 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1591 * If 87.5 % (7/8) of the space has been consumed, we want to override 1592 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1593 * len/truesize ratio. 1594 */ 1595 static inline bool tcp_rmem_pressure(const struct sock *sk) 1596 { 1597 int rcvbuf, threshold; 1598 1599 if (tcp_under_memory_pressure(sk)) 1600 return true; 1601 1602 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1603 threshold = rcvbuf - (rcvbuf >> 3); 1604 1605 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1606 } 1607 1608 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1609 { 1610 const struct tcp_sock *tp = tcp_sk(sk); 1611 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1612 1613 if (avail <= 0) 1614 return false; 1615 1616 return (avail >= target) || tcp_rmem_pressure(sk) || 1617 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1618 } 1619 1620 extern void tcp_openreq_init_rwin(struct request_sock *req, 1621 const struct sock *sk_listener, 1622 const struct dst_entry *dst); 1623 1624 void tcp_enter_memory_pressure(struct sock *sk); 1625 void tcp_leave_memory_pressure(struct sock *sk); 1626 1627 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1628 { 1629 struct net *net = sock_net((struct sock *)tp); 1630 int val; 1631 1632 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1633 * and do_tcp_setsockopt(). 1634 */ 1635 val = READ_ONCE(tp->keepalive_intvl); 1636 1637 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1638 } 1639 1640 static inline int keepalive_time_when(const struct tcp_sock *tp) 1641 { 1642 struct net *net = sock_net((struct sock *)tp); 1643 int val; 1644 1645 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1646 val = READ_ONCE(tp->keepalive_time); 1647 1648 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1649 } 1650 1651 static inline int keepalive_probes(const struct tcp_sock *tp) 1652 { 1653 struct net *net = sock_net((struct sock *)tp); 1654 int val; 1655 1656 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1657 * and do_tcp_setsockopt(). 1658 */ 1659 val = READ_ONCE(tp->keepalive_probes); 1660 1661 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1662 } 1663 1664 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1665 { 1666 const struct inet_connection_sock *icsk = &tp->inet_conn; 1667 1668 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1669 tcp_jiffies32 - tp->rcv_tstamp); 1670 } 1671 1672 static inline int tcp_fin_time(const struct sock *sk) 1673 { 1674 int fin_timeout = tcp_sk(sk)->linger2 ? : 1675 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1676 const int rto = inet_csk(sk)->icsk_rto; 1677 1678 if (fin_timeout < (rto << 2) - (rto >> 1)) 1679 fin_timeout = (rto << 2) - (rto >> 1); 1680 1681 return fin_timeout; 1682 } 1683 1684 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1685 int paws_win) 1686 { 1687 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1688 return true; 1689 if (unlikely(!time_before32(ktime_get_seconds(), 1690 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1691 return true; 1692 /* 1693 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1694 * then following tcp messages have valid values. Ignore 0 value, 1695 * or else 'negative' tsval might forbid us to accept their packets. 1696 */ 1697 if (!rx_opt->ts_recent) 1698 return true; 1699 return false; 1700 } 1701 1702 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1703 int rst) 1704 { 1705 if (tcp_paws_check(rx_opt, 0)) 1706 return false; 1707 1708 /* RST segments are not recommended to carry timestamp, 1709 and, if they do, it is recommended to ignore PAWS because 1710 "their cleanup function should take precedence over timestamps." 1711 Certainly, it is mistake. It is necessary to understand the reasons 1712 of this constraint to relax it: if peer reboots, clock may go 1713 out-of-sync and half-open connections will not be reset. 1714 Actually, the problem would be not existing if all 1715 the implementations followed draft about maintaining clock 1716 via reboots. Linux-2.2 DOES NOT! 1717 1718 However, we can relax time bounds for RST segments to MSL. 1719 */ 1720 if (rst && !time_before32(ktime_get_seconds(), 1721 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1722 return false; 1723 return true; 1724 } 1725 1726 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1727 int mib_idx, u32 *last_oow_ack_time); 1728 1729 static inline void tcp_mib_init(struct net *net) 1730 { 1731 /* See RFC 2012 */ 1732 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1733 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1734 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1735 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1736 } 1737 1738 /* from STCP */ 1739 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1740 { 1741 tp->lost_skb_hint = NULL; 1742 } 1743 1744 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1745 { 1746 tcp_clear_retrans_hints_partial(tp); 1747 tp->retransmit_skb_hint = NULL; 1748 } 1749 1750 #define tcp_md5_addr tcp_ao_addr 1751 1752 /* - key database */ 1753 struct tcp_md5sig_key { 1754 struct hlist_node node; 1755 u8 keylen; 1756 u8 family; /* AF_INET or AF_INET6 */ 1757 u8 prefixlen; 1758 u8 flags; 1759 union tcp_md5_addr addr; 1760 int l3index; /* set if key added with L3 scope */ 1761 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1762 struct rcu_head rcu; 1763 }; 1764 1765 /* - sock block */ 1766 struct tcp_md5sig_info { 1767 struct hlist_head head; 1768 struct rcu_head rcu; 1769 }; 1770 1771 /* - pseudo header */ 1772 struct tcp4_pseudohdr { 1773 __be32 saddr; 1774 __be32 daddr; 1775 __u8 pad; 1776 __u8 protocol; 1777 __be16 len; 1778 }; 1779 1780 struct tcp6_pseudohdr { 1781 struct in6_addr saddr; 1782 struct in6_addr daddr; 1783 __be32 len; 1784 __be32 protocol; /* including padding */ 1785 }; 1786 1787 union tcp_md5sum_block { 1788 struct tcp4_pseudohdr ip4; 1789 #if IS_ENABLED(CONFIG_IPV6) 1790 struct tcp6_pseudohdr ip6; 1791 #endif 1792 }; 1793 1794 /* 1795 * struct tcp_sigpool - per-CPU pool of ahash_requests 1796 * @scratch: per-CPU temporary area, that can be used between 1797 * tcp_sigpool_start() and tcp_sigpool_end() to perform 1798 * crypto request 1799 * @req: pre-allocated ahash request 1800 */ 1801 struct tcp_sigpool { 1802 void *scratch; 1803 struct ahash_request *req; 1804 }; 1805 1806 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size); 1807 void tcp_sigpool_get(unsigned int id); 1808 void tcp_sigpool_release(unsigned int id); 1809 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp, 1810 const struct sk_buff *skb, 1811 unsigned int header_len); 1812 1813 /** 1814 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash 1815 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash() 1816 * @c: returned tcp_sigpool for usage (uninitialized on failure) 1817 * 1818 * Returns 0 on success, error otherwise. 1819 */ 1820 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c); 1821 /** 1822 * tcp_sigpool_end - enable bh and stop using tcp_sigpool 1823 * @c: tcp_sigpool context that was returned by tcp_sigpool_start() 1824 */ 1825 void tcp_sigpool_end(struct tcp_sigpool *c); 1826 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len); 1827 /* - functions */ 1828 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1829 const struct sock *sk, const struct sk_buff *skb); 1830 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1831 int family, u8 prefixlen, int l3index, u8 flags, 1832 const u8 *newkey, u8 newkeylen); 1833 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1834 int family, u8 prefixlen, int l3index, 1835 struct tcp_md5sig_key *key); 1836 1837 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1838 int family, u8 prefixlen, int l3index, u8 flags); 1839 void tcp_clear_md5_list(struct sock *sk); 1840 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1841 const struct sock *addr_sk); 1842 1843 #ifdef CONFIG_TCP_MD5SIG 1844 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1845 const union tcp_md5_addr *addr, 1846 int family, bool any_l3index); 1847 static inline struct tcp_md5sig_key * 1848 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1849 const union tcp_md5_addr *addr, int family) 1850 { 1851 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1852 return NULL; 1853 return __tcp_md5_do_lookup(sk, l3index, addr, family, false); 1854 } 1855 1856 static inline struct tcp_md5sig_key * 1857 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1858 const union tcp_md5_addr *addr, int family) 1859 { 1860 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1861 return NULL; 1862 return __tcp_md5_do_lookup(sk, 0, addr, family, true); 1863 } 1864 1865 enum skb_drop_reason 1866 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1867 const void *saddr, const void *daddr, 1868 int family, int l3index, const __u8 *hash_location); 1869 1870 1871 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1872 #else 1873 static inline struct tcp_md5sig_key * 1874 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1875 const union tcp_md5_addr *addr, int family) 1876 { 1877 return NULL; 1878 } 1879 1880 static inline struct tcp_md5sig_key * 1881 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1882 const union tcp_md5_addr *addr, int family) 1883 { 1884 return NULL; 1885 } 1886 1887 static inline enum skb_drop_reason 1888 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1889 const void *saddr, const void *daddr, 1890 int family, int l3index, const __u8 *hash_location) 1891 { 1892 return SKB_NOT_DROPPED_YET; 1893 } 1894 #define tcp_twsk_md5_key(twsk) NULL 1895 #endif 1896 1897 int tcp_md5_alloc_sigpool(void); 1898 void tcp_md5_release_sigpool(void); 1899 void tcp_md5_add_sigpool(void); 1900 extern int tcp_md5_sigpool_id; 1901 1902 int tcp_md5_hash_key(struct tcp_sigpool *hp, 1903 const struct tcp_md5sig_key *key); 1904 1905 /* From tcp_fastopen.c */ 1906 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1907 struct tcp_fastopen_cookie *cookie); 1908 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1909 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1910 u16 try_exp); 1911 struct tcp_fastopen_request { 1912 /* Fast Open cookie. Size 0 means a cookie request */ 1913 struct tcp_fastopen_cookie cookie; 1914 struct msghdr *data; /* data in MSG_FASTOPEN */ 1915 size_t size; 1916 int copied; /* queued in tcp_connect() */ 1917 struct ubuf_info *uarg; 1918 }; 1919 void tcp_free_fastopen_req(struct tcp_sock *tp); 1920 void tcp_fastopen_destroy_cipher(struct sock *sk); 1921 void tcp_fastopen_ctx_destroy(struct net *net); 1922 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1923 void *primary_key, void *backup_key); 1924 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1925 u64 *key); 1926 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1927 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1928 struct request_sock *req, 1929 struct tcp_fastopen_cookie *foc, 1930 const struct dst_entry *dst); 1931 void tcp_fastopen_init_key_once(struct net *net); 1932 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1933 struct tcp_fastopen_cookie *cookie); 1934 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1935 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1936 #define TCP_FASTOPEN_KEY_MAX 2 1937 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1938 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1939 1940 /* Fastopen key context */ 1941 struct tcp_fastopen_context { 1942 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1943 int num; 1944 struct rcu_head rcu; 1945 }; 1946 1947 void tcp_fastopen_active_disable(struct sock *sk); 1948 bool tcp_fastopen_active_should_disable(struct sock *sk); 1949 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1950 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1951 1952 /* Caller needs to wrap with rcu_read_(un)lock() */ 1953 static inline 1954 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1955 { 1956 struct tcp_fastopen_context *ctx; 1957 1958 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1959 if (!ctx) 1960 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1961 return ctx; 1962 } 1963 1964 static inline 1965 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1966 const struct tcp_fastopen_cookie *orig) 1967 { 1968 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1969 orig->len == foc->len && 1970 !memcmp(orig->val, foc->val, foc->len)) 1971 return true; 1972 return false; 1973 } 1974 1975 static inline 1976 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1977 { 1978 return ctx->num; 1979 } 1980 1981 /* Latencies incurred by various limits for a sender. They are 1982 * chronograph-like stats that are mutually exclusive. 1983 */ 1984 enum tcp_chrono { 1985 TCP_CHRONO_UNSPEC, 1986 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1987 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1988 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1989 __TCP_CHRONO_MAX, 1990 }; 1991 1992 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1993 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1994 1995 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1996 * the same memory storage than skb->destructor/_skb_refdst 1997 */ 1998 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1999 { 2000 skb->destructor = NULL; 2001 skb->_skb_refdst = 0UL; 2002 } 2003 2004 #define tcp_skb_tsorted_save(skb) { \ 2005 unsigned long _save = skb->_skb_refdst; \ 2006 skb->_skb_refdst = 0UL; 2007 2008 #define tcp_skb_tsorted_restore(skb) \ 2009 skb->_skb_refdst = _save; \ 2010 } 2011 2012 void tcp_write_queue_purge(struct sock *sk); 2013 2014 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 2015 { 2016 return skb_rb_first(&sk->tcp_rtx_queue); 2017 } 2018 2019 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 2020 { 2021 return skb_rb_last(&sk->tcp_rtx_queue); 2022 } 2023 2024 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 2025 { 2026 return skb_peek_tail(&sk->sk_write_queue); 2027 } 2028 2029 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 2030 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 2031 2032 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 2033 { 2034 return skb_peek(&sk->sk_write_queue); 2035 } 2036 2037 static inline bool tcp_skb_is_last(const struct sock *sk, 2038 const struct sk_buff *skb) 2039 { 2040 return skb_queue_is_last(&sk->sk_write_queue, skb); 2041 } 2042 2043 /** 2044 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 2045 * @sk: socket 2046 * 2047 * Since the write queue can have a temporary empty skb in it, 2048 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 2049 */ 2050 static inline bool tcp_write_queue_empty(const struct sock *sk) 2051 { 2052 const struct tcp_sock *tp = tcp_sk(sk); 2053 2054 return tp->write_seq == tp->snd_nxt; 2055 } 2056 2057 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 2058 { 2059 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 2060 } 2061 2062 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 2063 { 2064 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 2065 } 2066 2067 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 2068 { 2069 __skb_queue_tail(&sk->sk_write_queue, skb); 2070 2071 /* Queue it, remembering where we must start sending. */ 2072 if (sk->sk_write_queue.next == skb) 2073 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 2074 } 2075 2076 /* Insert new before skb on the write queue of sk. */ 2077 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 2078 struct sk_buff *skb, 2079 struct sock *sk) 2080 { 2081 __skb_queue_before(&sk->sk_write_queue, skb, new); 2082 } 2083 2084 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 2085 { 2086 tcp_skb_tsorted_anchor_cleanup(skb); 2087 __skb_unlink(skb, &sk->sk_write_queue); 2088 } 2089 2090 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 2091 2092 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2093 { 2094 tcp_skb_tsorted_anchor_cleanup(skb); 2095 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2096 } 2097 2098 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2099 { 2100 list_del(&skb->tcp_tsorted_anchor); 2101 tcp_rtx_queue_unlink(skb, sk); 2102 tcp_wmem_free_skb(sk, skb); 2103 } 2104 2105 static inline void tcp_push_pending_frames(struct sock *sk) 2106 { 2107 if (tcp_send_head(sk)) { 2108 struct tcp_sock *tp = tcp_sk(sk); 2109 2110 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2111 } 2112 } 2113 2114 /* Start sequence of the skb just after the highest skb with SACKed 2115 * bit, valid only if sacked_out > 0 or when the caller has ensured 2116 * validity by itself. 2117 */ 2118 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2119 { 2120 if (!tp->sacked_out) 2121 return tp->snd_una; 2122 2123 if (tp->highest_sack == NULL) 2124 return tp->snd_nxt; 2125 2126 return TCP_SKB_CB(tp->highest_sack)->seq; 2127 } 2128 2129 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2130 { 2131 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2132 } 2133 2134 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2135 { 2136 return tcp_sk(sk)->highest_sack; 2137 } 2138 2139 static inline void tcp_highest_sack_reset(struct sock *sk) 2140 { 2141 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2142 } 2143 2144 /* Called when old skb is about to be deleted and replaced by new skb */ 2145 static inline void tcp_highest_sack_replace(struct sock *sk, 2146 struct sk_buff *old, 2147 struct sk_buff *new) 2148 { 2149 if (old == tcp_highest_sack(sk)) 2150 tcp_sk(sk)->highest_sack = new; 2151 } 2152 2153 /* This helper checks if socket has IP_TRANSPARENT set */ 2154 static inline bool inet_sk_transparent(const struct sock *sk) 2155 { 2156 switch (sk->sk_state) { 2157 case TCP_TIME_WAIT: 2158 return inet_twsk(sk)->tw_transparent; 2159 case TCP_NEW_SYN_RECV: 2160 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2161 } 2162 return inet_test_bit(TRANSPARENT, sk); 2163 } 2164 2165 /* Determines whether this is a thin stream (which may suffer from 2166 * increased latency). Used to trigger latency-reducing mechanisms. 2167 */ 2168 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2169 { 2170 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2171 } 2172 2173 /* /proc */ 2174 enum tcp_seq_states { 2175 TCP_SEQ_STATE_LISTENING, 2176 TCP_SEQ_STATE_ESTABLISHED, 2177 }; 2178 2179 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2180 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2181 void tcp_seq_stop(struct seq_file *seq, void *v); 2182 2183 struct tcp_seq_afinfo { 2184 sa_family_t family; 2185 }; 2186 2187 struct tcp_iter_state { 2188 struct seq_net_private p; 2189 enum tcp_seq_states state; 2190 struct sock *syn_wait_sk; 2191 int bucket, offset, sbucket, num; 2192 loff_t last_pos; 2193 }; 2194 2195 extern struct request_sock_ops tcp_request_sock_ops; 2196 extern struct request_sock_ops tcp6_request_sock_ops; 2197 2198 void tcp_v4_destroy_sock(struct sock *sk); 2199 2200 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2201 netdev_features_t features); 2202 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); 2203 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2204 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2205 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2206 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2207 #ifdef CONFIG_INET 2208 void tcp_gro_complete(struct sk_buff *skb); 2209 #else 2210 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2211 #endif 2212 2213 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2214 2215 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2216 { 2217 struct net *net = sock_net((struct sock *)tp); 2218 u32 val; 2219 2220 val = READ_ONCE(tp->notsent_lowat); 2221 2222 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2223 } 2224 2225 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2226 2227 #ifdef CONFIG_PROC_FS 2228 int tcp4_proc_init(void); 2229 void tcp4_proc_exit(void); 2230 #endif 2231 2232 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2233 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2234 const struct tcp_request_sock_ops *af_ops, 2235 struct sock *sk, struct sk_buff *skb); 2236 2237 /* TCP af-specific functions */ 2238 struct tcp_sock_af_ops { 2239 #ifdef CONFIG_TCP_MD5SIG 2240 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2241 const struct sock *addr_sk); 2242 int (*calc_md5_hash)(char *location, 2243 const struct tcp_md5sig_key *md5, 2244 const struct sock *sk, 2245 const struct sk_buff *skb); 2246 int (*md5_parse)(struct sock *sk, 2247 int optname, 2248 sockptr_t optval, 2249 int optlen); 2250 #endif 2251 #ifdef CONFIG_TCP_AO 2252 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen); 2253 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2254 struct sock *addr_sk, 2255 int sndid, int rcvid); 2256 int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key, 2257 const struct sock *sk, 2258 __be32 sisn, __be32 disn, bool send); 2259 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao, 2260 const struct sock *sk, const struct sk_buff *skb, 2261 const u8 *tkey, int hash_offset, u32 sne); 2262 #endif 2263 }; 2264 2265 struct tcp_request_sock_ops { 2266 u16 mss_clamp; 2267 #ifdef CONFIG_TCP_MD5SIG 2268 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2269 const struct sock *addr_sk); 2270 int (*calc_md5_hash) (char *location, 2271 const struct tcp_md5sig_key *md5, 2272 const struct sock *sk, 2273 const struct sk_buff *skb); 2274 #endif 2275 #ifdef CONFIG_TCP_AO 2276 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2277 struct request_sock *req, 2278 int sndid, int rcvid); 2279 int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk); 2280 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt, 2281 struct request_sock *req, const struct sk_buff *skb, 2282 int hash_offset, u32 sne); 2283 #endif 2284 #ifdef CONFIG_SYN_COOKIES 2285 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2286 __u16 *mss); 2287 #endif 2288 struct dst_entry *(*route_req)(const struct sock *sk, 2289 struct sk_buff *skb, 2290 struct flowi *fl, 2291 struct request_sock *req, 2292 u32 tw_isn); 2293 u32 (*init_seq)(const struct sk_buff *skb); 2294 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2295 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2296 struct flowi *fl, struct request_sock *req, 2297 struct tcp_fastopen_cookie *foc, 2298 enum tcp_synack_type synack_type, 2299 struct sk_buff *syn_skb); 2300 }; 2301 2302 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2303 #if IS_ENABLED(CONFIG_IPV6) 2304 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2305 #endif 2306 2307 #ifdef CONFIG_SYN_COOKIES 2308 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2309 const struct sock *sk, struct sk_buff *skb, 2310 __u16 *mss) 2311 { 2312 tcp_synq_overflow(sk); 2313 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2314 return ops->cookie_init_seq(skb, mss); 2315 } 2316 #else 2317 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2318 const struct sock *sk, struct sk_buff *skb, 2319 __u16 *mss) 2320 { 2321 return 0; 2322 } 2323 #endif 2324 2325 struct tcp_key { 2326 union { 2327 struct { 2328 struct tcp_ao_key *ao_key; 2329 char *traffic_key; 2330 u32 sne; 2331 u8 rcv_next; 2332 }; 2333 struct tcp_md5sig_key *md5_key; 2334 }; 2335 enum { 2336 TCP_KEY_NONE = 0, 2337 TCP_KEY_MD5, 2338 TCP_KEY_AO, 2339 } type; 2340 }; 2341 2342 static inline void tcp_get_current_key(const struct sock *sk, 2343 struct tcp_key *out) 2344 { 2345 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG) 2346 const struct tcp_sock *tp = tcp_sk(sk); 2347 #endif 2348 2349 #ifdef CONFIG_TCP_AO 2350 if (static_branch_unlikely(&tcp_ao_needed.key)) { 2351 struct tcp_ao_info *ao; 2352 2353 ao = rcu_dereference_protected(tp->ao_info, 2354 lockdep_sock_is_held(sk)); 2355 if (ao) { 2356 out->ao_key = READ_ONCE(ao->current_key); 2357 out->type = TCP_KEY_AO; 2358 return; 2359 } 2360 } 2361 #endif 2362 #ifdef CONFIG_TCP_MD5SIG 2363 if (static_branch_unlikely(&tcp_md5_needed.key) && 2364 rcu_access_pointer(tp->md5sig_info)) { 2365 out->md5_key = tp->af_specific->md5_lookup(sk, sk); 2366 if (out->md5_key) { 2367 out->type = TCP_KEY_MD5; 2368 return; 2369 } 2370 } 2371 #endif 2372 out->type = TCP_KEY_NONE; 2373 } 2374 2375 static inline bool tcp_key_is_md5(const struct tcp_key *key) 2376 { 2377 #ifdef CONFIG_TCP_MD5SIG 2378 if (static_branch_unlikely(&tcp_md5_needed.key) && 2379 key->type == TCP_KEY_MD5) 2380 return true; 2381 #endif 2382 return false; 2383 } 2384 2385 static inline bool tcp_key_is_ao(const struct tcp_key *key) 2386 { 2387 #ifdef CONFIG_TCP_AO 2388 if (static_branch_unlikely(&tcp_ao_needed.key) && 2389 key->type == TCP_KEY_AO) 2390 return true; 2391 #endif 2392 return false; 2393 } 2394 2395 int tcpv4_offload_init(void); 2396 2397 void tcp_v4_init(void); 2398 void tcp_init(void); 2399 2400 /* tcp_recovery.c */ 2401 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2402 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2403 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2404 u32 reo_wnd); 2405 extern bool tcp_rack_mark_lost(struct sock *sk); 2406 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2407 u64 xmit_time); 2408 extern void tcp_rack_reo_timeout(struct sock *sk); 2409 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2410 2411 /* tcp_plb.c */ 2412 2413 /* 2414 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2415 * expects cong_ratio which represents fraction of traffic that experienced 2416 * congestion over a single RTT. In order to avoid floating point operations, 2417 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2418 */ 2419 #define TCP_PLB_SCALE 8 2420 2421 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2422 struct tcp_plb_state { 2423 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2424 unused:3; 2425 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2426 }; 2427 2428 static inline void tcp_plb_init(const struct sock *sk, 2429 struct tcp_plb_state *plb) 2430 { 2431 plb->consec_cong_rounds = 0; 2432 plb->pause_until = 0; 2433 } 2434 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2435 const int cong_ratio); 2436 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2437 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2438 2439 /* At how many usecs into the future should the RTO fire? */ 2440 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2441 { 2442 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2443 u32 rto = inet_csk(sk)->icsk_rto; 2444 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2445 2446 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2447 } 2448 2449 /* 2450 * Save and compile IPv4 options, return a pointer to it 2451 */ 2452 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2453 struct sk_buff *skb) 2454 { 2455 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2456 struct ip_options_rcu *dopt = NULL; 2457 2458 if (opt->optlen) { 2459 int opt_size = sizeof(*dopt) + opt->optlen; 2460 2461 dopt = kmalloc(opt_size, GFP_ATOMIC); 2462 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2463 kfree(dopt); 2464 dopt = NULL; 2465 } 2466 } 2467 return dopt; 2468 } 2469 2470 /* locally generated TCP pure ACKs have skb->truesize == 2 2471 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2472 * This is much faster than dissecting the packet to find out. 2473 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2474 */ 2475 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2476 { 2477 return skb->truesize == 2; 2478 } 2479 2480 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2481 { 2482 skb->truesize = 2; 2483 } 2484 2485 static inline int tcp_inq(struct sock *sk) 2486 { 2487 struct tcp_sock *tp = tcp_sk(sk); 2488 int answ; 2489 2490 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2491 answ = 0; 2492 } else if (sock_flag(sk, SOCK_URGINLINE) || 2493 !tp->urg_data || 2494 before(tp->urg_seq, tp->copied_seq) || 2495 !before(tp->urg_seq, tp->rcv_nxt)) { 2496 2497 answ = tp->rcv_nxt - tp->copied_seq; 2498 2499 /* Subtract 1, if FIN was received */ 2500 if (answ && sock_flag(sk, SOCK_DONE)) 2501 answ--; 2502 } else { 2503 answ = tp->urg_seq - tp->copied_seq; 2504 } 2505 2506 return answ; 2507 } 2508 2509 int tcp_peek_len(struct socket *sock); 2510 2511 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2512 { 2513 u16 segs_in; 2514 2515 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2516 2517 /* We update these fields while other threads might 2518 * read them from tcp_get_info() 2519 */ 2520 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2521 if (skb->len > tcp_hdrlen(skb)) 2522 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2523 } 2524 2525 /* 2526 * TCP listen path runs lockless. 2527 * We forced "struct sock" to be const qualified to make sure 2528 * we don't modify one of its field by mistake. 2529 * Here, we increment sk_drops which is an atomic_t, so we can safely 2530 * make sock writable again. 2531 */ 2532 static inline void tcp_listendrop(const struct sock *sk) 2533 { 2534 atomic_inc(&((struct sock *)sk)->sk_drops); 2535 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2536 } 2537 2538 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2539 2540 /* 2541 * Interface for adding Upper Level Protocols over TCP 2542 */ 2543 2544 #define TCP_ULP_NAME_MAX 16 2545 #define TCP_ULP_MAX 128 2546 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2547 2548 struct tcp_ulp_ops { 2549 struct list_head list; 2550 2551 /* initialize ulp */ 2552 int (*init)(struct sock *sk); 2553 /* update ulp */ 2554 void (*update)(struct sock *sk, struct proto *p, 2555 void (*write_space)(struct sock *sk)); 2556 /* cleanup ulp */ 2557 void (*release)(struct sock *sk); 2558 /* diagnostic */ 2559 int (*get_info)(struct sock *sk, struct sk_buff *skb); 2560 size_t (*get_info_size)(const struct sock *sk); 2561 /* clone ulp */ 2562 void (*clone)(const struct request_sock *req, struct sock *newsk, 2563 const gfp_t priority); 2564 2565 char name[TCP_ULP_NAME_MAX]; 2566 struct module *owner; 2567 }; 2568 int tcp_register_ulp(struct tcp_ulp_ops *type); 2569 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2570 int tcp_set_ulp(struct sock *sk, const char *name); 2571 void tcp_get_available_ulp(char *buf, size_t len); 2572 void tcp_cleanup_ulp(struct sock *sk); 2573 void tcp_update_ulp(struct sock *sk, struct proto *p, 2574 void (*write_space)(struct sock *sk)); 2575 2576 #define MODULE_ALIAS_TCP_ULP(name) \ 2577 __MODULE_INFO(alias, alias_userspace, name); \ 2578 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2579 2580 #ifdef CONFIG_NET_SOCK_MSG 2581 struct sk_msg; 2582 struct sk_psock; 2583 2584 #ifdef CONFIG_BPF_SYSCALL 2585 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2586 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2587 #endif /* CONFIG_BPF_SYSCALL */ 2588 2589 #ifdef CONFIG_INET 2590 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2591 #else 2592 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2593 { 2594 } 2595 #endif 2596 2597 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2598 struct sk_msg *msg, u32 bytes, int flags); 2599 #endif /* CONFIG_NET_SOCK_MSG */ 2600 2601 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2602 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2603 { 2604 } 2605 #endif 2606 2607 #ifdef CONFIG_CGROUP_BPF 2608 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2609 struct sk_buff *skb, 2610 unsigned int end_offset) 2611 { 2612 skops->skb = skb; 2613 skops->skb_data_end = skb->data + end_offset; 2614 } 2615 #else 2616 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2617 struct sk_buff *skb, 2618 unsigned int end_offset) 2619 { 2620 } 2621 #endif 2622 2623 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2624 * is < 0, then the BPF op failed (for example if the loaded BPF 2625 * program does not support the chosen operation or there is no BPF 2626 * program loaded). 2627 */ 2628 #ifdef CONFIG_BPF 2629 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2630 { 2631 struct bpf_sock_ops_kern sock_ops; 2632 int ret; 2633 2634 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2635 if (sk_fullsock(sk)) { 2636 sock_ops.is_fullsock = 1; 2637 sock_owned_by_me(sk); 2638 } 2639 2640 sock_ops.sk = sk; 2641 sock_ops.op = op; 2642 if (nargs > 0) 2643 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2644 2645 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2646 if (ret == 0) 2647 ret = sock_ops.reply; 2648 else 2649 ret = -1; 2650 return ret; 2651 } 2652 2653 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2654 { 2655 u32 args[2] = {arg1, arg2}; 2656 2657 return tcp_call_bpf(sk, op, 2, args); 2658 } 2659 2660 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2661 u32 arg3) 2662 { 2663 u32 args[3] = {arg1, arg2, arg3}; 2664 2665 return tcp_call_bpf(sk, op, 3, args); 2666 } 2667 2668 #else 2669 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2670 { 2671 return -EPERM; 2672 } 2673 2674 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2675 { 2676 return -EPERM; 2677 } 2678 2679 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2680 u32 arg3) 2681 { 2682 return -EPERM; 2683 } 2684 2685 #endif 2686 2687 static inline u32 tcp_timeout_init(struct sock *sk) 2688 { 2689 int timeout; 2690 2691 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2692 2693 if (timeout <= 0) 2694 timeout = TCP_TIMEOUT_INIT; 2695 return min_t(int, timeout, TCP_RTO_MAX); 2696 } 2697 2698 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2699 { 2700 int rwnd; 2701 2702 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2703 2704 if (rwnd < 0) 2705 rwnd = 0; 2706 return rwnd; 2707 } 2708 2709 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2710 { 2711 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2712 } 2713 2714 static inline void tcp_bpf_rtt(struct sock *sk) 2715 { 2716 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2717 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); 2718 } 2719 2720 #if IS_ENABLED(CONFIG_SMC) 2721 extern struct static_key_false tcp_have_smc; 2722 #endif 2723 2724 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2725 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2726 void (*cad)(struct sock *sk, u32 ack_seq)); 2727 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2728 void clean_acked_data_flush(void); 2729 #endif 2730 2731 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2732 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2733 const struct tcp_sock *tp) 2734 { 2735 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2736 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2737 } 2738 2739 /* Compute Earliest Departure Time for some control packets 2740 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2741 */ 2742 static inline u64 tcp_transmit_time(const struct sock *sk) 2743 { 2744 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2745 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2746 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2747 2748 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2749 } 2750 return 0; 2751 } 2752 2753 static inline int tcp_parse_auth_options(const struct tcphdr *th, 2754 const u8 **md5_hash, const struct tcp_ao_hdr **aoh) 2755 { 2756 const u8 *md5_tmp, *ao_tmp; 2757 int ret; 2758 2759 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp); 2760 if (ret) 2761 return ret; 2762 2763 if (md5_hash) 2764 *md5_hash = md5_tmp; 2765 2766 if (aoh) { 2767 if (!ao_tmp) 2768 *aoh = NULL; 2769 else 2770 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2); 2771 } 2772 2773 return 0; 2774 } 2775 2776 static inline bool tcp_ao_required(struct sock *sk, const void *saddr, 2777 int family, int l3index, bool stat_inc) 2778 { 2779 #ifdef CONFIG_TCP_AO 2780 struct tcp_ao_info *ao_info; 2781 struct tcp_ao_key *ao_key; 2782 2783 if (!static_branch_unlikely(&tcp_ao_needed.key)) 2784 return false; 2785 2786 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info, 2787 lockdep_sock_is_held(sk)); 2788 if (!ao_info) 2789 return false; 2790 2791 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1); 2792 if (ao_info->ao_required || ao_key) { 2793 if (stat_inc) { 2794 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED); 2795 atomic64_inc(&ao_info->counters.ao_required); 2796 } 2797 return true; 2798 } 2799 #endif 2800 return false; 2801 } 2802 2803 /* Called with rcu_read_lock() */ 2804 static inline enum skb_drop_reason 2805 tcp_inbound_hash(struct sock *sk, const struct request_sock *req, 2806 const struct sk_buff *skb, 2807 const void *saddr, const void *daddr, 2808 int family, int dif, int sdif) 2809 { 2810 const struct tcphdr *th = tcp_hdr(skb); 2811 const struct tcp_ao_hdr *aoh; 2812 const __u8 *md5_location; 2813 int l3index; 2814 2815 /* Invalid option or two times meet any of auth options */ 2816 if (tcp_parse_auth_options(th, &md5_location, &aoh)) { 2817 tcp_hash_fail("TCP segment has incorrect auth options set", 2818 family, skb, ""); 2819 return SKB_DROP_REASON_TCP_AUTH_HDR; 2820 } 2821 2822 if (req) { 2823 if (tcp_rsk_used_ao(req) != !!aoh) { 2824 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD); 2825 tcp_hash_fail("TCP connection can't start/end using TCP-AO", 2826 family, skb, "%s", 2827 !aoh ? "missing AO" : "AO signed"); 2828 return SKB_DROP_REASON_TCP_AOFAILURE; 2829 } 2830 } 2831 2832 /* sdif set, means packet ingressed via a device 2833 * in an L3 domain and dif is set to the l3mdev 2834 */ 2835 l3index = sdif ? dif : 0; 2836 2837 /* Fast path: unsigned segments */ 2838 if (likely(!md5_location && !aoh)) { 2839 /* Drop if there's TCP-MD5 or TCP-AO key with any rcvid/sndid 2840 * for the remote peer. On TCP-AO established connection 2841 * the last key is impossible to remove, so there's 2842 * always at least one current_key. 2843 */ 2844 if (tcp_ao_required(sk, saddr, family, l3index, true)) { 2845 tcp_hash_fail("AO hash is required, but not found", 2846 family, skb, "L3 index %d", l3index); 2847 return SKB_DROP_REASON_TCP_AONOTFOUND; 2848 } 2849 if (unlikely(tcp_md5_do_lookup(sk, l3index, saddr, family))) { 2850 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 2851 tcp_hash_fail("MD5 Hash not found", 2852 family, skb, "L3 index %d", l3index); 2853 return SKB_DROP_REASON_TCP_MD5NOTFOUND; 2854 } 2855 return SKB_NOT_DROPPED_YET; 2856 } 2857 2858 if (aoh) 2859 return tcp_inbound_ao_hash(sk, skb, family, req, l3index, aoh); 2860 2861 return tcp_inbound_md5_hash(sk, skb, saddr, daddr, family, 2862 l3index, md5_location); 2863 } 2864 2865 #endif /* _TCP_H */ 2866