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 void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 300 { 301 sk_wmem_queued_add(sk, -skb->truesize); 302 if (!skb_zcopy_pure(skb)) 303 sk_mem_uncharge(sk, skb->truesize); 304 else 305 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb))); 306 __kfree_skb(skb); 307 } 308 309 void sk_forced_mem_schedule(struct sock *sk, int size); 310 311 bool tcp_check_oom(const struct sock *sk, int shift); 312 313 314 extern struct proto tcp_prot; 315 316 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 317 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 318 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 319 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 320 321 void tcp_tasklet_init(void); 322 323 int tcp_v4_err(struct sk_buff *skb, u32); 324 325 void tcp_shutdown(struct sock *sk, int how); 326 327 int tcp_v4_early_demux(struct sk_buff *skb); 328 int tcp_v4_rcv(struct sk_buff *skb); 329 330 void tcp_remove_empty_skb(struct sock *sk); 331 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 332 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 333 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied, 334 size_t size, struct ubuf_info *uarg); 335 void tcp_splice_eof(struct socket *sock); 336 int tcp_send_mss(struct sock *sk, int *size_goal, int flags); 337 int tcp_wmem_schedule(struct sock *sk, int copy); 338 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle, 339 int size_goal); 340 void tcp_release_cb(struct sock *sk); 341 void tcp_wfree(struct sk_buff *skb); 342 void tcp_write_timer_handler(struct sock *sk); 343 void tcp_delack_timer_handler(struct sock *sk); 344 int tcp_ioctl(struct sock *sk, int cmd, int *karg); 345 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 346 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); 347 void tcp_rcv_space_adjust(struct sock *sk); 348 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 349 void tcp_twsk_destructor(struct sock *sk); 350 void tcp_twsk_purge(struct list_head *net_exit_list); 351 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 352 struct pipe_inode_info *pipe, size_t len, 353 unsigned int flags); 354 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp, 355 bool force_schedule); 356 357 static inline void tcp_dec_quickack_mode(struct sock *sk) 358 { 359 struct inet_connection_sock *icsk = inet_csk(sk); 360 361 if (icsk->icsk_ack.quick) { 362 /* How many ACKs S/ACKing new data have we sent? */ 363 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0; 364 365 if (pkts >= icsk->icsk_ack.quick) { 366 icsk->icsk_ack.quick = 0; 367 /* Leaving quickack mode we deflate ATO. */ 368 icsk->icsk_ack.ato = TCP_ATO_MIN; 369 } else 370 icsk->icsk_ack.quick -= pkts; 371 } 372 } 373 374 #define TCP_ECN_OK 1 375 #define TCP_ECN_QUEUE_CWR 2 376 #define TCP_ECN_DEMAND_CWR 4 377 #define TCP_ECN_SEEN 8 378 379 enum tcp_tw_status { 380 TCP_TW_SUCCESS = 0, 381 TCP_TW_RST = 1, 382 TCP_TW_ACK = 2, 383 TCP_TW_SYN = 3 384 }; 385 386 387 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 388 struct sk_buff *skb, 389 const struct tcphdr *th, 390 u32 *tw_isn); 391 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 392 struct request_sock *req, bool fastopen, 393 bool *lost_race); 394 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child, 395 struct sk_buff *skb); 396 void tcp_enter_loss(struct sock *sk); 397 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag); 398 void tcp_clear_retrans(struct tcp_sock *tp); 399 void tcp_update_metrics(struct sock *sk); 400 void tcp_init_metrics(struct sock *sk); 401 void tcp_metrics_init(void); 402 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 403 void __tcp_close(struct sock *sk, long timeout); 404 void tcp_close(struct sock *sk, long timeout); 405 void tcp_init_sock(struct sock *sk); 406 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb); 407 __poll_t tcp_poll(struct file *file, struct socket *sock, 408 struct poll_table_struct *wait); 409 int do_tcp_getsockopt(struct sock *sk, int level, 410 int optname, sockptr_t optval, sockptr_t optlen); 411 int tcp_getsockopt(struct sock *sk, int level, int optname, 412 char __user *optval, int __user *optlen); 413 bool tcp_bpf_bypass_getsockopt(int level, int optname); 414 int do_tcp_setsockopt(struct sock *sk, int level, int optname, 415 sockptr_t optval, unsigned int optlen); 416 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 417 unsigned int optlen); 418 void tcp_set_keepalive(struct sock *sk, int val); 419 void tcp_syn_ack_timeout(const struct request_sock *req); 420 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 421 int flags, int *addr_len); 422 int tcp_set_rcvlowat(struct sock *sk, int val); 423 int tcp_set_window_clamp(struct sock *sk, int val); 424 void tcp_update_recv_tstamps(struct sk_buff *skb, 425 struct scm_timestamping_internal *tss); 426 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk, 427 struct scm_timestamping_internal *tss); 428 void tcp_data_ready(struct sock *sk); 429 #ifdef CONFIG_MMU 430 int tcp_mmap(struct file *file, struct socket *sock, 431 struct vm_area_struct *vma); 432 #endif 433 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 434 struct tcp_options_received *opt_rx, 435 int estab, struct tcp_fastopen_cookie *foc); 436 437 /* 438 * BPF SKB-less helpers 439 */ 440 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, 441 struct tcphdr *th, u32 *cookie); 442 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, 443 struct tcphdr *th, u32 *cookie); 444 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss); 445 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, 446 const struct tcp_request_sock_ops *af_ops, 447 struct sock *sk, struct tcphdr *th); 448 /* 449 * TCP v4 functions exported for the inet6 API 450 */ 451 452 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 453 void tcp_v4_mtu_reduced(struct sock *sk); 454 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 455 void tcp_ld_RTO_revert(struct sock *sk, u32 seq); 456 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 457 struct sock *tcp_create_openreq_child(const struct sock *sk, 458 struct request_sock *req, 459 struct sk_buff *skb); 460 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 461 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 462 struct request_sock *req, 463 struct dst_entry *dst, 464 struct request_sock *req_unhash, 465 bool *own_req); 466 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 467 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 468 int tcp_connect(struct sock *sk); 469 enum tcp_synack_type { 470 TCP_SYNACK_NORMAL, 471 TCP_SYNACK_FASTOPEN, 472 TCP_SYNACK_COOKIE, 473 }; 474 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 475 struct request_sock *req, 476 struct tcp_fastopen_cookie *foc, 477 enum tcp_synack_type synack_type, 478 struct sk_buff *syn_skb); 479 int tcp_disconnect(struct sock *sk, int flags); 480 481 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 482 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 483 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 484 485 /* From syncookies.c */ 486 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 487 struct request_sock *req, 488 struct dst_entry *dst); 489 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th); 490 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 491 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops, 492 struct sock *sk, struct sk_buff *skb, 493 struct tcp_options_received *tcp_opt, 494 int mss, u32 tsoff); 495 496 #if IS_ENABLED(CONFIG_BPF) 497 struct bpf_tcp_req_attrs { 498 u32 rcv_tsval; 499 u32 rcv_tsecr; 500 u16 mss; 501 u8 rcv_wscale; 502 u8 snd_wscale; 503 u8 ecn_ok; 504 u8 wscale_ok; 505 u8 sack_ok; 506 u8 tstamp_ok; 507 u8 usec_ts_ok; 508 u8 reserved[3]; 509 }; 510 #endif 511 512 #ifdef CONFIG_SYN_COOKIES 513 514 /* Syncookies use a monotonic timer which increments every 60 seconds. 515 * This counter is used both as a hash input and partially encoded into 516 * the cookie value. A cookie is only validated further if the delta 517 * between the current counter value and the encoded one is less than this, 518 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 519 * the counter advances immediately after a cookie is generated). 520 */ 521 #define MAX_SYNCOOKIE_AGE 2 522 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 523 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 524 525 /* syncookies: remember time of last synqueue overflow 526 * But do not dirty this field too often (once per second is enough) 527 * It is racy as we do not hold a lock, but race is very minor. 528 */ 529 static inline void tcp_synq_overflow(const struct sock *sk) 530 { 531 unsigned int last_overflow; 532 unsigned int now = jiffies; 533 534 if (sk->sk_reuseport) { 535 struct sock_reuseport *reuse; 536 537 reuse = rcu_dereference(sk->sk_reuseport_cb); 538 if (likely(reuse)) { 539 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 540 if (!time_between32(now, last_overflow, 541 last_overflow + HZ)) 542 WRITE_ONCE(reuse->synq_overflow_ts, now); 543 return; 544 } 545 } 546 547 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 548 if (!time_between32(now, last_overflow, last_overflow + HZ)) 549 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now); 550 } 551 552 /* syncookies: no recent synqueue overflow on this listening socket? */ 553 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 554 { 555 unsigned int last_overflow; 556 unsigned int now = jiffies; 557 558 if (sk->sk_reuseport) { 559 struct sock_reuseport *reuse; 560 561 reuse = rcu_dereference(sk->sk_reuseport_cb); 562 if (likely(reuse)) { 563 last_overflow = READ_ONCE(reuse->synq_overflow_ts); 564 return !time_between32(now, last_overflow - HZ, 565 last_overflow + 566 TCP_SYNCOOKIE_VALID); 567 } 568 } 569 570 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); 571 572 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, 573 * then we're under synflood. However, we have to use 574 * 'last_overflow - HZ' as lower bound. That's because a concurrent 575 * tcp_synq_overflow() could update .ts_recent_stamp after we read 576 * jiffies but before we store .ts_recent_stamp into last_overflow, 577 * which could lead to rejecting a valid syncookie. 578 */ 579 return !time_between32(now, last_overflow - HZ, 580 last_overflow + TCP_SYNCOOKIE_VALID); 581 } 582 583 static inline u32 tcp_cookie_time(void) 584 { 585 u64 val = get_jiffies_64(); 586 587 do_div(val, TCP_SYNCOOKIE_PERIOD); 588 return val; 589 } 590 591 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */ 592 static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val) 593 { 594 if (usec_ts) 595 return div_u64(val, NSEC_PER_USEC); 596 597 return div_u64(val, NSEC_PER_MSEC); 598 } 599 600 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 601 u16 *mssp); 602 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 603 u64 cookie_init_timestamp(struct request_sock *req, u64 now); 604 bool cookie_timestamp_decode(const struct net *net, 605 struct tcp_options_received *opt); 606 607 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst) 608 { 609 return READ_ONCE(net->ipv4.sysctl_tcp_ecn) || 610 dst_feature(dst, RTAX_FEATURE_ECN); 611 } 612 613 #if IS_ENABLED(CONFIG_BPF) 614 static inline bool cookie_bpf_ok(struct sk_buff *skb) 615 { 616 return skb->sk; 617 } 618 619 struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb); 620 #else 621 static inline bool cookie_bpf_ok(struct sk_buff *skb) 622 { 623 return false; 624 } 625 626 static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk, 627 struct sk_buff *skb) 628 { 629 return NULL; 630 } 631 #endif 632 633 /* From net/ipv6/syncookies.c */ 634 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th); 635 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 636 637 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 638 const struct tcphdr *th, u16 *mssp); 639 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 640 #endif 641 /* tcp_output.c */ 642 643 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb); 644 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb); 645 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 646 int nonagle); 647 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 648 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 649 void tcp_retransmit_timer(struct sock *sk); 650 void tcp_xmit_retransmit_queue(struct sock *); 651 void tcp_simple_retransmit(struct sock *); 652 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 653 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 654 enum tcp_queue { 655 TCP_FRAG_IN_WRITE_QUEUE, 656 TCP_FRAG_IN_RTX_QUEUE, 657 }; 658 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 659 struct sk_buff *skb, u32 len, 660 unsigned int mss_now, gfp_t gfp); 661 662 void tcp_send_probe0(struct sock *); 663 int tcp_write_wakeup(struct sock *, int mib); 664 void tcp_send_fin(struct sock *sk); 665 void tcp_send_active_reset(struct sock *sk, gfp_t priority, 666 enum sk_rst_reason reason); 667 int tcp_send_synack(struct sock *); 668 void tcp_push_one(struct sock *, unsigned int mss_now); 669 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); 670 void tcp_send_ack(struct sock *sk); 671 void tcp_send_delayed_ack(struct sock *sk); 672 void tcp_send_loss_probe(struct sock *sk); 673 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); 674 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 675 const struct sk_buff *next_skb); 676 677 /* tcp_input.c */ 678 void tcp_rearm_rto(struct sock *sk); 679 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 680 void tcp_done_with_error(struct sock *sk, int err); 681 void tcp_reset(struct sock *sk, struct sk_buff *skb); 682 void tcp_fin(struct sock *sk); 683 void tcp_check_space(struct sock *sk); 684 void tcp_sack_compress_send_ack(struct sock *sk); 685 686 /* tcp_timer.c */ 687 void tcp_init_xmit_timers(struct sock *); 688 static inline void tcp_clear_xmit_timers(struct sock *sk) 689 { 690 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) 691 __sock_put(sk); 692 693 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) 694 __sock_put(sk); 695 696 inet_csk_clear_xmit_timers(sk); 697 } 698 699 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 700 unsigned int tcp_current_mss(struct sock *sk); 701 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); 702 703 /* Bound MSS / TSO packet size with the half of the window */ 704 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 705 { 706 int cutoff; 707 708 /* When peer uses tiny windows, there is no use in packetizing 709 * to sub-MSS pieces for the sake of SWS or making sure there 710 * are enough packets in the pipe for fast recovery. 711 * 712 * On the other hand, for extremely large MSS devices, handling 713 * smaller than MSS windows in this way does make sense. 714 */ 715 if (tp->max_window > TCP_MSS_DEFAULT) 716 cutoff = (tp->max_window >> 1); 717 else 718 cutoff = tp->max_window; 719 720 if (cutoff && pktsize > cutoff) 721 return max_t(int, cutoff, 68U - tp->tcp_header_len); 722 else 723 return pktsize; 724 } 725 726 /* tcp.c */ 727 void tcp_get_info(struct sock *, struct tcp_info *); 728 729 /* Read 'sendfile()'-style from a TCP socket */ 730 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 731 sk_read_actor_t recv_actor); 732 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor); 733 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off); 734 void tcp_read_done(struct sock *sk, size_t len); 735 736 void tcp_initialize_rcv_mss(struct sock *sk); 737 738 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 739 int tcp_mss_to_mtu(struct sock *sk, int mss); 740 void tcp_mtup_init(struct sock *sk); 741 742 static inline void tcp_bound_rto(struct sock *sk) 743 { 744 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 745 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 746 } 747 748 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 749 { 750 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 751 } 752 753 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 754 { 755 /* mptcp hooks are only on the slow path */ 756 if (sk_is_mptcp((struct sock *)tp)) 757 return; 758 759 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 760 ntohl(TCP_FLAG_ACK) | 761 snd_wnd); 762 } 763 764 static inline void tcp_fast_path_on(struct tcp_sock *tp) 765 { 766 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 767 } 768 769 static inline void tcp_fast_path_check(struct sock *sk) 770 { 771 struct tcp_sock *tp = tcp_sk(sk); 772 773 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 774 tp->rcv_wnd && 775 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 776 !tp->urg_data) 777 tcp_fast_path_on(tp); 778 } 779 780 u32 tcp_delack_max(const struct sock *sk); 781 782 /* Compute the actual rto_min value */ 783 static inline u32 tcp_rto_min(const struct sock *sk) 784 { 785 const struct dst_entry *dst = __sk_dst_get(sk); 786 u32 rto_min = inet_csk(sk)->icsk_rto_min; 787 788 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 789 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 790 return rto_min; 791 } 792 793 static inline u32 tcp_rto_min_us(const struct sock *sk) 794 { 795 return jiffies_to_usecs(tcp_rto_min(sk)); 796 } 797 798 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 799 { 800 return dst_metric_locked(dst, RTAX_CC_ALGO); 801 } 802 803 /* Minimum RTT in usec. ~0 means not available. */ 804 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 805 { 806 return minmax_get(&tp->rtt_min); 807 } 808 809 /* Compute the actual receive window we are currently advertising. 810 * Rcv_nxt can be after the window if our peer push more data 811 * than the offered window. 812 */ 813 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 814 { 815 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 816 817 if (win < 0) 818 win = 0; 819 return (u32) win; 820 } 821 822 /* Choose a new window, without checks for shrinking, and without 823 * scaling applied to the result. The caller does these things 824 * if necessary. This is a "raw" window selection. 825 */ 826 u32 __tcp_select_window(struct sock *sk); 827 828 void tcp_send_window_probe(struct sock *sk); 829 830 /* TCP uses 32bit jiffies to save some space. 831 * Note that this is different from tcp_time_stamp, which 832 * historically has been the same until linux-4.13. 833 */ 834 #define tcp_jiffies32 ((u32)jiffies) 835 836 /* 837 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 838 * It is no longer tied to jiffies, but to 1 ms clock. 839 * Note: double check if you want to use tcp_jiffies32 instead of this. 840 */ 841 #define TCP_TS_HZ 1000 842 843 static inline u64 tcp_clock_ns(void) 844 { 845 return ktime_get_ns(); 846 } 847 848 static inline u64 tcp_clock_us(void) 849 { 850 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 851 } 852 853 static inline u64 tcp_clock_ms(void) 854 { 855 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC); 856 } 857 858 /* TCP Timestamp included in TS option (RFC 1323) can either use ms 859 * or usec resolution. Each socket carries a flag to select one or other 860 * resolution, as the route attribute could change anytime. 861 * Each flow must stick to initial resolution. 862 */ 863 static inline u32 tcp_clock_ts(bool usec_ts) 864 { 865 return usec_ts ? tcp_clock_us() : tcp_clock_ms(); 866 } 867 868 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp) 869 { 870 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC); 871 } 872 873 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp) 874 { 875 if (tp->tcp_usec_ts) 876 return tp->tcp_mstamp; 877 return tcp_time_stamp_ms(tp); 878 } 879 880 void tcp_mstamp_refresh(struct tcp_sock *tp); 881 882 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 883 { 884 return max_t(s64, t1 - t0, 0); 885 } 886 887 /* provide the departure time in us unit */ 888 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) 889 { 890 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); 891 } 892 893 /* Provide skb TSval in usec or ms unit */ 894 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb) 895 { 896 if (usec_ts) 897 return tcp_skb_timestamp_us(skb); 898 899 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC); 900 } 901 902 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw) 903 { 904 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset; 905 } 906 907 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) 908 { 909 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off; 910 } 911 912 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 913 914 #define TCPHDR_FIN 0x01 915 #define TCPHDR_SYN 0x02 916 #define TCPHDR_RST 0x04 917 #define TCPHDR_PSH 0x08 918 #define TCPHDR_ACK 0x10 919 #define TCPHDR_URG 0x20 920 #define TCPHDR_ECE 0x40 921 #define TCPHDR_CWR 0x80 922 923 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 924 925 /* State flags for sacked in struct tcp_skb_cb */ 926 enum tcp_skb_cb_sacked_flags { 927 TCPCB_SACKED_ACKED = (1 << 0), /* SKB ACK'd by a SACK block */ 928 TCPCB_SACKED_RETRANS = (1 << 1), /* SKB retransmitted */ 929 TCPCB_LOST = (1 << 2), /* SKB is lost */ 930 TCPCB_TAGBITS = (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS | 931 TCPCB_LOST), /* All tag bits */ 932 TCPCB_REPAIRED = (1 << 4), /* SKB repaired (no skb_mstamp_ns) */ 933 TCPCB_EVER_RETRANS = (1 << 7), /* Ever retransmitted frame */ 934 TCPCB_RETRANS = (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS | 935 TCPCB_REPAIRED), 936 }; 937 938 /* This is what the send packet queuing engine uses to pass 939 * TCP per-packet control information to the transmission code. 940 * We also store the host-order sequence numbers in here too. 941 * This is 44 bytes if IPV6 is enabled. 942 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 943 */ 944 struct tcp_skb_cb { 945 __u32 seq; /* Starting sequence number */ 946 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 947 union { 948 /* Note : 949 * tcp_gso_segs/size are used in write queue only, 950 * cf tcp_skb_pcount()/tcp_skb_mss() 951 */ 952 struct { 953 u16 tcp_gso_segs; 954 u16 tcp_gso_size; 955 }; 956 }; 957 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 958 959 __u8 sacked; /* State flags for SACK. */ 960 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 961 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 962 eor:1, /* Is skb MSG_EOR marked? */ 963 has_rxtstamp:1, /* SKB has a RX timestamp */ 964 unused:5; 965 __u32 ack_seq; /* Sequence number ACK'd */ 966 union { 967 struct { 968 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1) 969 /* There is space for up to 24 bytes */ 970 __u32 is_app_limited:1, /* cwnd not fully used? */ 971 delivered_ce:20, 972 unused:11; 973 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 974 __u32 delivered; 975 /* start of send pipeline phase */ 976 u64 first_tx_mstamp; 977 /* when we reached the "delivered" count */ 978 u64 delivered_mstamp; 979 } tx; /* only used for outgoing skbs */ 980 union { 981 struct inet_skb_parm h4; 982 #if IS_ENABLED(CONFIG_IPV6) 983 struct inet6_skb_parm h6; 984 #endif 985 } header; /* For incoming skbs */ 986 }; 987 }; 988 989 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 990 991 extern const struct inet_connection_sock_af_ops ipv4_specific; 992 993 #if IS_ENABLED(CONFIG_IPV6) 994 /* This is the variant of inet6_iif() that must be used by TCP, 995 * as TCP moves IP6CB into a different location in skb->cb[] 996 */ 997 static inline int tcp_v6_iif(const struct sk_buff *skb) 998 { 999 return TCP_SKB_CB(skb)->header.h6.iif; 1000 } 1001 1002 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) 1003 { 1004 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 1005 1006 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 1007 } 1008 1009 /* TCP_SKB_CB reference means this can not be used from early demux */ 1010 static inline int tcp_v6_sdif(const struct sk_buff *skb) 1011 { 1012 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1013 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 1014 return TCP_SKB_CB(skb)->header.h6.iif; 1015 #endif 1016 return 0; 1017 } 1018 1019 extern const struct inet_connection_sock_af_ops ipv6_specific; 1020 1021 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)); 1022 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb)); 1023 void tcp_v6_early_demux(struct sk_buff *skb); 1024 1025 #endif 1026 1027 /* TCP_SKB_CB reference means this can not be used from early demux */ 1028 static inline int tcp_v4_sdif(struct sk_buff *skb) 1029 { 1030 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 1031 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 1032 return TCP_SKB_CB(skb)->header.h4.iif; 1033 #endif 1034 return 0; 1035 } 1036 1037 /* Due to TSO, an SKB can be composed of multiple actual 1038 * packets. To keep these tracked properly, we use this. 1039 */ 1040 static inline int tcp_skb_pcount(const struct sk_buff *skb) 1041 { 1042 return TCP_SKB_CB(skb)->tcp_gso_segs; 1043 } 1044 1045 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 1046 { 1047 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 1048 } 1049 1050 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 1051 { 1052 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 1053 } 1054 1055 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 1056 static inline int tcp_skb_mss(const struct sk_buff *skb) 1057 { 1058 return TCP_SKB_CB(skb)->tcp_gso_size; 1059 } 1060 1061 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 1062 { 1063 return likely(!TCP_SKB_CB(skb)->eor); 1064 } 1065 1066 static inline bool tcp_skb_can_collapse(const struct sk_buff *to, 1067 const struct sk_buff *from) 1068 { 1069 /* skb_cmp_decrypted() not needed, use tcp_write_collapse_fence() */ 1070 return likely(tcp_skb_can_collapse_to(to) && 1071 mptcp_skb_can_collapse(to, from) && 1072 skb_pure_zcopy_same(to, from)); 1073 } 1074 1075 static inline bool tcp_skb_can_collapse_rx(const struct sk_buff *to, 1076 const struct sk_buff *from) 1077 { 1078 return likely(mptcp_skb_can_collapse(to, from) && 1079 !skb_cmp_decrypted(to, from)); 1080 } 1081 1082 /* Events passed to congestion control interface */ 1083 enum tcp_ca_event { 1084 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 1085 CA_EVENT_CWND_RESTART, /* congestion window restart */ 1086 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 1087 CA_EVENT_LOSS, /* loss timeout */ 1088 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 1089 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 1090 }; 1091 1092 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 1093 enum tcp_ca_ack_event_flags { 1094 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 1095 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 1096 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 1097 }; 1098 1099 /* 1100 * Interface for adding new TCP congestion control handlers 1101 */ 1102 #define TCP_CA_NAME_MAX 16 1103 #define TCP_CA_MAX 128 1104 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 1105 1106 #define TCP_CA_UNSPEC 0 1107 1108 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 1109 #define TCP_CONG_NON_RESTRICTED 0x1 1110 /* Requires ECN/ECT set on all packets */ 1111 #define TCP_CONG_NEEDS_ECN 0x2 1112 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) 1113 1114 union tcp_cc_info; 1115 1116 struct ack_sample { 1117 u32 pkts_acked; 1118 s32 rtt_us; 1119 u32 in_flight; 1120 }; 1121 1122 /* A rate sample measures the number of (original/retransmitted) data 1123 * packets delivered "delivered" over an interval of time "interval_us". 1124 * The tcp_rate.c code fills in the rate sample, and congestion 1125 * control modules that define a cong_control function to run at the end 1126 * of ACK processing can optionally chose to consult this sample when 1127 * setting cwnd and pacing rate. 1128 * A sample is invalid if "delivered" or "interval_us" is negative. 1129 */ 1130 struct rate_sample { 1131 u64 prior_mstamp; /* starting timestamp for interval */ 1132 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 1133 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ 1134 s32 delivered; /* number of packets delivered over interval */ 1135 s32 delivered_ce; /* number of packets delivered w/ CE marks*/ 1136 long interval_us; /* time for tp->delivered to incr "delivered" */ 1137 u32 snd_interval_us; /* snd interval for delivered packets */ 1138 u32 rcv_interval_us; /* rcv interval for delivered packets */ 1139 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 1140 int losses; /* number of packets marked lost upon ACK */ 1141 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 1142 u32 prior_in_flight; /* in flight before this ACK */ 1143 u32 last_end_seq; /* end_seq of most recently ACKed packet */ 1144 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 1145 bool is_retrans; /* is sample from retransmission? */ 1146 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 1147 }; 1148 1149 struct tcp_congestion_ops { 1150 /* fast path fields are put first to fill one cache line */ 1151 1152 /* return slow start threshold (required) */ 1153 u32 (*ssthresh)(struct sock *sk); 1154 1155 /* do new cwnd calculation (required) */ 1156 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 1157 1158 /* call before changing ca_state (optional) */ 1159 void (*set_state)(struct sock *sk, u8 new_state); 1160 1161 /* call when cwnd event occurs (optional) */ 1162 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 1163 1164 /* call when ack arrives (optional) */ 1165 void (*in_ack_event)(struct sock *sk, u32 flags); 1166 1167 /* hook for packet ack accounting (optional) */ 1168 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 1169 1170 /* override sysctl_tcp_min_tso_segs */ 1171 u32 (*min_tso_segs)(struct sock *sk); 1172 1173 /* call when packets are delivered to update cwnd and pacing rate, 1174 * after all the ca_state processing. (optional) 1175 */ 1176 void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs); 1177 1178 1179 /* new value of cwnd after loss (required) */ 1180 u32 (*undo_cwnd)(struct sock *sk); 1181 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 1182 u32 (*sndbuf_expand)(struct sock *sk); 1183 1184 /* control/slow paths put last */ 1185 /* get info for inet_diag (optional) */ 1186 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 1187 union tcp_cc_info *info); 1188 1189 char name[TCP_CA_NAME_MAX]; 1190 struct module *owner; 1191 struct list_head list; 1192 u32 key; 1193 u32 flags; 1194 1195 /* initialize private data (optional) */ 1196 void (*init)(struct sock *sk); 1197 /* cleanup private data (optional) */ 1198 void (*release)(struct sock *sk); 1199 } ____cacheline_aligned_in_smp; 1200 1201 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1202 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1203 int tcp_update_congestion_control(struct tcp_congestion_ops *type, 1204 struct tcp_congestion_ops *old_type); 1205 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca); 1206 1207 void tcp_assign_congestion_control(struct sock *sk); 1208 void tcp_init_congestion_control(struct sock *sk); 1209 void tcp_cleanup_congestion_control(struct sock *sk); 1210 int tcp_set_default_congestion_control(struct net *net, const char *name); 1211 void tcp_get_default_congestion_control(struct net *net, char *name); 1212 void tcp_get_available_congestion_control(char *buf, size_t len); 1213 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1214 int tcp_set_allowed_congestion_control(char *allowed); 1215 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, 1216 bool cap_net_admin); 1217 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1218 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1219 1220 u32 tcp_reno_ssthresh(struct sock *sk); 1221 u32 tcp_reno_undo_cwnd(struct sock *sk); 1222 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1223 extern struct tcp_congestion_ops tcp_reno; 1224 1225 struct tcp_congestion_ops *tcp_ca_find(const char *name); 1226 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1227 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1228 #ifdef CONFIG_INET 1229 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1230 #else 1231 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1232 { 1233 return NULL; 1234 } 1235 #endif 1236 1237 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1238 { 1239 const struct inet_connection_sock *icsk = inet_csk(sk); 1240 1241 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1242 } 1243 1244 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1245 { 1246 const struct inet_connection_sock *icsk = inet_csk(sk); 1247 1248 if (icsk->icsk_ca_ops->cwnd_event) 1249 icsk->icsk_ca_ops->cwnd_event(sk, event); 1250 } 1251 1252 /* From tcp_cong.c */ 1253 void tcp_set_ca_state(struct sock *sk, const u8 ca_state); 1254 1255 /* From tcp_rate.c */ 1256 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1257 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1258 struct rate_sample *rs); 1259 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1260 bool is_sack_reneg, struct rate_sample *rs); 1261 void tcp_rate_check_app_limited(struct sock *sk); 1262 1263 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) 1264 { 1265 return t1 > t2 || (t1 == t2 && after(seq1, seq2)); 1266 } 1267 1268 /* These functions determine how the current flow behaves in respect of SACK 1269 * handling. SACK is negotiated with the peer, and therefore it can vary 1270 * between different flows. 1271 * 1272 * tcp_is_sack - SACK enabled 1273 * tcp_is_reno - No SACK 1274 */ 1275 static inline int tcp_is_sack(const struct tcp_sock *tp) 1276 { 1277 return likely(tp->rx_opt.sack_ok); 1278 } 1279 1280 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1281 { 1282 return !tcp_is_sack(tp); 1283 } 1284 1285 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1286 { 1287 return tp->sacked_out + tp->lost_out; 1288 } 1289 1290 /* This determines how many packets are "in the network" to the best 1291 * of our knowledge. In many cases it is conservative, but where 1292 * detailed information is available from the receiver (via SACK 1293 * blocks etc.) we can make more aggressive calculations. 1294 * 1295 * Use this for decisions involving congestion control, use just 1296 * tp->packets_out to determine if the send queue is empty or not. 1297 * 1298 * Read this equation as: 1299 * 1300 * "Packets sent once on transmission queue" MINUS 1301 * "Packets left network, but not honestly ACKed yet" PLUS 1302 * "Packets fast retransmitted" 1303 */ 1304 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1305 { 1306 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1307 } 1308 1309 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1310 1311 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp) 1312 { 1313 return tp->snd_cwnd; 1314 } 1315 1316 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val) 1317 { 1318 WARN_ON_ONCE((int)val <= 0); 1319 tp->snd_cwnd = val; 1320 } 1321 1322 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1323 { 1324 return tcp_snd_cwnd(tp) < tp->snd_ssthresh; 1325 } 1326 1327 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1328 { 1329 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1330 } 1331 1332 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1333 { 1334 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1335 (1 << inet_csk(sk)->icsk_ca_state); 1336 } 1337 1338 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1339 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1340 * ssthresh. 1341 */ 1342 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1343 { 1344 const struct tcp_sock *tp = tcp_sk(sk); 1345 1346 if (tcp_in_cwnd_reduction(sk)) 1347 return tp->snd_ssthresh; 1348 else 1349 return max(tp->snd_ssthresh, 1350 ((tcp_snd_cwnd(tp) >> 1) + 1351 (tcp_snd_cwnd(tp) >> 2))); 1352 } 1353 1354 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1355 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1356 1357 void tcp_enter_cwr(struct sock *sk); 1358 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1359 1360 /* The maximum number of MSS of available cwnd for which TSO defers 1361 * sending if not using sysctl_tcp_tso_win_divisor. 1362 */ 1363 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1364 { 1365 return 3; 1366 } 1367 1368 /* Returns end sequence number of the receiver's advertised window */ 1369 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1370 { 1371 return tp->snd_una + tp->snd_wnd; 1372 } 1373 1374 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1375 * flexible approach. The RFC suggests cwnd should not be raised unless 1376 * it was fully used previously. And that's exactly what we do in 1377 * congestion avoidance mode. But in slow start we allow cwnd to grow 1378 * as long as the application has used half the cwnd. 1379 * Example : 1380 * cwnd is 10 (IW10), but application sends 9 frames. 1381 * We allow cwnd to reach 18 when all frames are ACKed. 1382 * This check is safe because it's as aggressive as slow start which already 1383 * risks 100% overshoot. The advantage is that we discourage application to 1384 * either send more filler packets or data to artificially blow up the cwnd 1385 * usage, and allow application-limited process to probe bw more aggressively. 1386 */ 1387 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1388 { 1389 const struct tcp_sock *tp = tcp_sk(sk); 1390 1391 if (tp->is_cwnd_limited) 1392 return true; 1393 1394 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1395 if (tcp_in_slow_start(tp)) 1396 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out; 1397 1398 return false; 1399 } 1400 1401 /* BBR congestion control needs pacing. 1402 * Same remark for SO_MAX_PACING_RATE. 1403 * sch_fq packet scheduler is efficiently handling pacing, 1404 * but is not always installed/used. 1405 * Return true if TCP stack should pace packets itself. 1406 */ 1407 static inline bool tcp_needs_internal_pacing(const struct sock *sk) 1408 { 1409 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; 1410 } 1411 1412 /* Estimates in how many jiffies next packet for this flow can be sent. 1413 * Scheduling a retransmit timer too early would be silly. 1414 */ 1415 static inline unsigned long tcp_pacing_delay(const struct sock *sk) 1416 { 1417 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache; 1418 1419 return delay > 0 ? nsecs_to_jiffies(delay) : 0; 1420 } 1421 1422 static inline void tcp_reset_xmit_timer(struct sock *sk, 1423 const int what, 1424 unsigned long when, 1425 const unsigned long max_when) 1426 { 1427 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk), 1428 max_when); 1429 } 1430 1431 /* Something is really bad, we could not queue an additional packet, 1432 * because qdisc is full or receiver sent a 0 window, or we are paced. 1433 * We do not want to add fuel to the fire, or abort too early, 1434 * so make sure the timer we arm now is at least 200ms in the future, 1435 * regardless of current icsk_rto value (as it could be ~2ms) 1436 */ 1437 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1438 { 1439 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1440 } 1441 1442 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1443 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1444 unsigned long max_when) 1445 { 1446 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1, 1447 inet_csk(sk)->icsk_backoff); 1448 u64 when = (u64)tcp_probe0_base(sk) << backoff; 1449 1450 return (unsigned long)min_t(u64, when, max_when); 1451 } 1452 1453 static inline void tcp_check_probe_timer(struct sock *sk) 1454 { 1455 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1456 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1457 tcp_probe0_base(sk), TCP_RTO_MAX); 1458 } 1459 1460 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1461 { 1462 tp->snd_wl1 = seq; 1463 } 1464 1465 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1466 { 1467 tp->snd_wl1 = seq; 1468 } 1469 1470 /* 1471 * Calculate(/check) TCP checksum 1472 */ 1473 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1474 __be32 daddr, __wsum base) 1475 { 1476 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); 1477 } 1478 1479 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1480 { 1481 return !skb_csum_unnecessary(skb) && 1482 __skb_checksum_complete(skb); 1483 } 1484 1485 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb, 1486 enum skb_drop_reason *reason); 1487 1488 1489 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1490 void tcp_set_state(struct sock *sk, int state); 1491 void tcp_done(struct sock *sk); 1492 int tcp_abort(struct sock *sk, int err); 1493 1494 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1495 { 1496 rx_opt->dsack = 0; 1497 rx_opt->num_sacks = 0; 1498 } 1499 1500 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1501 1502 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1503 { 1504 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1505 struct tcp_sock *tp = tcp_sk(sk); 1506 s32 delta; 1507 1508 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) || 1509 tp->packets_out || ca_ops->cong_control) 1510 return; 1511 delta = tcp_jiffies32 - tp->lsndtime; 1512 if (delta > inet_csk(sk)->icsk_rto) 1513 tcp_cwnd_restart(sk, delta); 1514 } 1515 1516 /* Determine a window scaling and initial window to offer. */ 1517 void tcp_select_initial_window(const struct sock *sk, int __space, 1518 __u32 mss, __u32 *rcv_wnd, 1519 __u32 *window_clamp, int wscale_ok, 1520 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1521 1522 static inline int __tcp_win_from_space(u8 scaling_ratio, int space) 1523 { 1524 s64 scaled_space = (s64)space * scaling_ratio; 1525 1526 return scaled_space >> TCP_RMEM_TO_WIN_SCALE; 1527 } 1528 1529 static inline int tcp_win_from_space(const struct sock *sk, int space) 1530 { 1531 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space); 1532 } 1533 1534 /* inverse of __tcp_win_from_space() */ 1535 static inline int __tcp_space_from_win(u8 scaling_ratio, int win) 1536 { 1537 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE; 1538 1539 do_div(val, scaling_ratio); 1540 return val; 1541 } 1542 1543 static inline int tcp_space_from_win(const struct sock *sk, int win) 1544 { 1545 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win); 1546 } 1547 1548 /* Assume a 50% default for skb->len/skb->truesize ratio. 1549 * This may be adjusted later in tcp_measure_rcv_mss(). 1550 */ 1551 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1)) 1552 1553 static inline void tcp_scaling_ratio_init(struct sock *sk) 1554 { 1555 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO; 1556 } 1557 1558 /* Note: caller must be prepared to deal with negative returns */ 1559 static inline int tcp_space(const struct sock *sk) 1560 { 1561 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - 1562 READ_ONCE(sk->sk_backlog.len) - 1563 atomic_read(&sk->sk_rmem_alloc)); 1564 } 1565 1566 static inline int tcp_full_space(const struct sock *sk) 1567 { 1568 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1569 } 1570 1571 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh) 1572 { 1573 int unused_mem = sk_unused_reserved_mem(sk); 1574 struct tcp_sock *tp = tcp_sk(sk); 1575 1576 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh); 1577 if (unused_mem) 1578 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh, 1579 tcp_win_from_space(sk, unused_mem)); 1580 } 1581 1582 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk) 1583 { 1584 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss); 1585 } 1586 1587 void tcp_cleanup_rbuf(struct sock *sk, int copied); 1588 void __tcp_cleanup_rbuf(struct sock *sk, int copied); 1589 1590 1591 /* We provision sk_rcvbuf around 200% of sk_rcvlowat. 1592 * If 87.5 % (7/8) of the space has been consumed, we want to override 1593 * SO_RCVLOWAT constraint, since we are receiving skbs with too small 1594 * len/truesize ratio. 1595 */ 1596 static inline bool tcp_rmem_pressure(const struct sock *sk) 1597 { 1598 int rcvbuf, threshold; 1599 1600 if (tcp_under_memory_pressure(sk)) 1601 return true; 1602 1603 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1604 threshold = rcvbuf - (rcvbuf >> 3); 1605 1606 return atomic_read(&sk->sk_rmem_alloc) > threshold; 1607 } 1608 1609 static inline bool tcp_epollin_ready(const struct sock *sk, int target) 1610 { 1611 const struct tcp_sock *tp = tcp_sk(sk); 1612 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq); 1613 1614 if (avail <= 0) 1615 return false; 1616 1617 return (avail >= target) || tcp_rmem_pressure(sk) || 1618 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss); 1619 } 1620 1621 extern void tcp_openreq_init_rwin(struct request_sock *req, 1622 const struct sock *sk_listener, 1623 const struct dst_entry *dst); 1624 1625 void tcp_enter_memory_pressure(struct sock *sk); 1626 void tcp_leave_memory_pressure(struct sock *sk); 1627 1628 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1629 { 1630 struct net *net = sock_net((struct sock *)tp); 1631 int val; 1632 1633 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl() 1634 * and do_tcp_setsockopt(). 1635 */ 1636 val = READ_ONCE(tp->keepalive_intvl); 1637 1638 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl); 1639 } 1640 1641 static inline int keepalive_time_when(const struct tcp_sock *tp) 1642 { 1643 struct net *net = sock_net((struct sock *)tp); 1644 int val; 1645 1646 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */ 1647 val = READ_ONCE(tp->keepalive_time); 1648 1649 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time); 1650 } 1651 1652 static inline int keepalive_probes(const struct tcp_sock *tp) 1653 { 1654 struct net *net = sock_net((struct sock *)tp); 1655 int val; 1656 1657 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt() 1658 * and do_tcp_setsockopt(). 1659 */ 1660 val = READ_ONCE(tp->keepalive_probes); 1661 1662 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes); 1663 } 1664 1665 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1666 { 1667 const struct inet_connection_sock *icsk = &tp->inet_conn; 1668 1669 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1670 tcp_jiffies32 - tp->rcv_tstamp); 1671 } 1672 1673 static inline int tcp_fin_time(const struct sock *sk) 1674 { 1675 int fin_timeout = tcp_sk(sk)->linger2 ? : 1676 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout); 1677 const int rto = inet_csk(sk)->icsk_rto; 1678 1679 if (fin_timeout < (rto << 2) - (rto >> 1)) 1680 fin_timeout = (rto << 2) - (rto >> 1); 1681 1682 return fin_timeout; 1683 } 1684 1685 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1686 int paws_win) 1687 { 1688 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1689 return true; 1690 if (unlikely(!time_before32(ktime_get_seconds(), 1691 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP))) 1692 return true; 1693 /* 1694 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1695 * then following tcp messages have valid values. Ignore 0 value, 1696 * or else 'negative' tsval might forbid us to accept their packets. 1697 */ 1698 if (!rx_opt->ts_recent) 1699 return true; 1700 return false; 1701 } 1702 1703 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1704 int rst) 1705 { 1706 if (tcp_paws_check(rx_opt, 0)) 1707 return false; 1708 1709 /* RST segments are not recommended to carry timestamp, 1710 and, if they do, it is recommended to ignore PAWS because 1711 "their cleanup function should take precedence over timestamps." 1712 Certainly, it is mistake. It is necessary to understand the reasons 1713 of this constraint to relax it: if peer reboots, clock may go 1714 out-of-sync and half-open connections will not be reset. 1715 Actually, the problem would be not existing if all 1716 the implementations followed draft about maintaining clock 1717 via reboots. Linux-2.2 DOES NOT! 1718 1719 However, we can relax time bounds for RST segments to MSL. 1720 */ 1721 if (rst && !time_before32(ktime_get_seconds(), 1722 rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) 1723 return false; 1724 return true; 1725 } 1726 1727 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1728 int mib_idx, u32 *last_oow_ack_time); 1729 1730 static inline void tcp_mib_init(struct net *net) 1731 { 1732 /* See RFC 2012 */ 1733 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1734 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1735 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1736 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1737 } 1738 1739 /* from STCP */ 1740 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1741 { 1742 tp->lost_skb_hint = NULL; 1743 } 1744 1745 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1746 { 1747 tcp_clear_retrans_hints_partial(tp); 1748 tp->retransmit_skb_hint = NULL; 1749 } 1750 1751 #define tcp_md5_addr tcp_ao_addr 1752 1753 /* - key database */ 1754 struct tcp_md5sig_key { 1755 struct hlist_node node; 1756 u8 keylen; 1757 u8 family; /* AF_INET or AF_INET6 */ 1758 u8 prefixlen; 1759 u8 flags; 1760 union tcp_md5_addr addr; 1761 int l3index; /* set if key added with L3 scope */ 1762 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1763 struct rcu_head rcu; 1764 }; 1765 1766 /* - sock block */ 1767 struct tcp_md5sig_info { 1768 struct hlist_head head; 1769 struct rcu_head rcu; 1770 }; 1771 1772 /* - pseudo header */ 1773 struct tcp4_pseudohdr { 1774 __be32 saddr; 1775 __be32 daddr; 1776 __u8 pad; 1777 __u8 protocol; 1778 __be16 len; 1779 }; 1780 1781 struct tcp6_pseudohdr { 1782 struct in6_addr saddr; 1783 struct in6_addr daddr; 1784 __be32 len; 1785 __be32 protocol; /* including padding */ 1786 }; 1787 1788 union tcp_md5sum_block { 1789 struct tcp4_pseudohdr ip4; 1790 #if IS_ENABLED(CONFIG_IPV6) 1791 struct tcp6_pseudohdr ip6; 1792 #endif 1793 }; 1794 1795 /* 1796 * struct tcp_sigpool - per-CPU pool of ahash_requests 1797 * @scratch: per-CPU temporary area, that can be used between 1798 * tcp_sigpool_start() and tcp_sigpool_end() to perform 1799 * crypto request 1800 * @req: pre-allocated ahash request 1801 */ 1802 struct tcp_sigpool { 1803 void *scratch; 1804 struct ahash_request *req; 1805 }; 1806 1807 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size); 1808 void tcp_sigpool_get(unsigned int id); 1809 void tcp_sigpool_release(unsigned int id); 1810 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp, 1811 const struct sk_buff *skb, 1812 unsigned int header_len); 1813 1814 /** 1815 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash 1816 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash() 1817 * @c: returned tcp_sigpool for usage (uninitialized on failure) 1818 * 1819 * Returns 0 on success, error otherwise. 1820 */ 1821 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c); 1822 /** 1823 * tcp_sigpool_end - enable bh and stop using tcp_sigpool 1824 * @c: tcp_sigpool context that was returned by tcp_sigpool_start() 1825 */ 1826 void tcp_sigpool_end(struct tcp_sigpool *c); 1827 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len); 1828 /* - functions */ 1829 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1830 const struct sock *sk, const struct sk_buff *skb); 1831 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1832 int family, u8 prefixlen, int l3index, u8 flags, 1833 const u8 *newkey, u8 newkeylen); 1834 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr, 1835 int family, u8 prefixlen, int l3index, 1836 struct tcp_md5sig_key *key); 1837 1838 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1839 int family, u8 prefixlen, int l3index, u8 flags); 1840 void tcp_clear_md5_list(struct sock *sk); 1841 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1842 const struct sock *addr_sk); 1843 1844 #ifdef CONFIG_TCP_MD5SIG 1845 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index, 1846 const union tcp_md5_addr *addr, 1847 int family, bool any_l3index); 1848 static inline struct tcp_md5sig_key * 1849 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1850 const union tcp_md5_addr *addr, int family) 1851 { 1852 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1853 return NULL; 1854 return __tcp_md5_do_lookup(sk, l3index, addr, family, false); 1855 } 1856 1857 static inline struct tcp_md5sig_key * 1858 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1859 const union tcp_md5_addr *addr, int family) 1860 { 1861 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1862 return NULL; 1863 return __tcp_md5_do_lookup(sk, 0, addr, family, true); 1864 } 1865 1866 enum skb_drop_reason 1867 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1868 const void *saddr, const void *daddr, 1869 int family, int l3index, const __u8 *hash_location); 1870 1871 1872 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1873 #else 1874 static inline struct tcp_md5sig_key * 1875 tcp_md5_do_lookup(const struct sock *sk, int l3index, 1876 const union tcp_md5_addr *addr, int family) 1877 { 1878 return NULL; 1879 } 1880 1881 static inline struct tcp_md5sig_key * 1882 tcp_md5_do_lookup_any_l3index(const struct sock *sk, 1883 const union tcp_md5_addr *addr, int family) 1884 { 1885 return NULL; 1886 } 1887 1888 static inline enum skb_drop_reason 1889 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb, 1890 const void *saddr, const void *daddr, 1891 int family, int l3index, const __u8 *hash_location) 1892 { 1893 return SKB_NOT_DROPPED_YET; 1894 } 1895 #define tcp_twsk_md5_key(twsk) NULL 1896 #endif 1897 1898 int tcp_md5_alloc_sigpool(void); 1899 void tcp_md5_release_sigpool(void); 1900 void tcp_md5_add_sigpool(void); 1901 extern int tcp_md5_sigpool_id; 1902 1903 int tcp_md5_hash_key(struct tcp_sigpool *hp, 1904 const struct tcp_md5sig_key *key); 1905 1906 /* From tcp_fastopen.c */ 1907 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1908 struct tcp_fastopen_cookie *cookie); 1909 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1910 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1911 u16 try_exp); 1912 struct tcp_fastopen_request { 1913 /* Fast Open cookie. Size 0 means a cookie request */ 1914 struct tcp_fastopen_cookie cookie; 1915 struct msghdr *data; /* data in MSG_FASTOPEN */ 1916 size_t size; 1917 int copied; /* queued in tcp_connect() */ 1918 struct ubuf_info *uarg; 1919 }; 1920 void tcp_free_fastopen_req(struct tcp_sock *tp); 1921 void tcp_fastopen_destroy_cipher(struct sock *sk); 1922 void tcp_fastopen_ctx_destroy(struct net *net); 1923 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1924 void *primary_key, void *backup_key); 1925 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, 1926 u64 *key); 1927 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1928 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1929 struct request_sock *req, 1930 struct tcp_fastopen_cookie *foc, 1931 const struct dst_entry *dst); 1932 void tcp_fastopen_init_key_once(struct net *net); 1933 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1934 struct tcp_fastopen_cookie *cookie); 1935 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1936 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) 1937 #define TCP_FASTOPEN_KEY_MAX 2 1938 #define TCP_FASTOPEN_KEY_BUF_LENGTH \ 1939 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) 1940 1941 /* Fastopen key context */ 1942 struct tcp_fastopen_context { 1943 siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; 1944 int num; 1945 struct rcu_head rcu; 1946 }; 1947 1948 void tcp_fastopen_active_disable(struct sock *sk); 1949 bool tcp_fastopen_active_should_disable(struct sock *sk); 1950 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1951 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); 1952 1953 /* Caller needs to wrap with rcu_read_(un)lock() */ 1954 static inline 1955 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) 1956 { 1957 struct tcp_fastopen_context *ctx; 1958 1959 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); 1960 if (!ctx) 1961 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); 1962 return ctx; 1963 } 1964 1965 static inline 1966 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, 1967 const struct tcp_fastopen_cookie *orig) 1968 { 1969 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && 1970 orig->len == foc->len && 1971 !memcmp(orig->val, foc->val, foc->len)) 1972 return true; 1973 return false; 1974 } 1975 1976 static inline 1977 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) 1978 { 1979 return ctx->num; 1980 } 1981 1982 /* Latencies incurred by various limits for a sender. They are 1983 * chronograph-like stats that are mutually exclusive. 1984 */ 1985 enum tcp_chrono { 1986 TCP_CHRONO_UNSPEC, 1987 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1988 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1989 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1990 __TCP_CHRONO_MAX, 1991 }; 1992 1993 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1994 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1995 1996 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1997 * the same memory storage than skb->destructor/_skb_refdst 1998 */ 1999 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 2000 { 2001 skb->destructor = NULL; 2002 skb->_skb_refdst = 0UL; 2003 } 2004 2005 #define tcp_skb_tsorted_save(skb) { \ 2006 unsigned long _save = skb->_skb_refdst; \ 2007 skb->_skb_refdst = 0UL; 2008 2009 #define tcp_skb_tsorted_restore(skb) \ 2010 skb->_skb_refdst = _save; \ 2011 } 2012 2013 void tcp_write_queue_purge(struct sock *sk); 2014 2015 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 2016 { 2017 return skb_rb_first(&sk->tcp_rtx_queue); 2018 } 2019 2020 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) 2021 { 2022 return skb_rb_last(&sk->tcp_rtx_queue); 2023 } 2024 2025 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 2026 { 2027 return skb_peek_tail(&sk->sk_write_queue); 2028 } 2029 2030 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 2031 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 2032 2033 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 2034 { 2035 return skb_peek(&sk->sk_write_queue); 2036 } 2037 2038 static inline bool tcp_skb_is_last(const struct sock *sk, 2039 const struct sk_buff *skb) 2040 { 2041 return skb_queue_is_last(&sk->sk_write_queue, skb); 2042 } 2043 2044 /** 2045 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue 2046 * @sk: socket 2047 * 2048 * Since the write queue can have a temporary empty skb in it, 2049 * we must not use "return skb_queue_empty(&sk->sk_write_queue)" 2050 */ 2051 static inline bool tcp_write_queue_empty(const struct sock *sk) 2052 { 2053 const struct tcp_sock *tp = tcp_sk(sk); 2054 2055 return tp->write_seq == tp->snd_nxt; 2056 } 2057 2058 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 2059 { 2060 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 2061 } 2062 2063 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 2064 { 2065 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 2066 } 2067 2068 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 2069 { 2070 __skb_queue_tail(&sk->sk_write_queue, skb); 2071 2072 /* Queue it, remembering where we must start sending. */ 2073 if (sk->sk_write_queue.next == skb) 2074 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 2075 } 2076 2077 /* Insert new before skb on the write queue of sk. */ 2078 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 2079 struct sk_buff *skb, 2080 struct sock *sk) 2081 { 2082 __skb_queue_before(&sk->sk_write_queue, skb, new); 2083 } 2084 2085 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 2086 { 2087 tcp_skb_tsorted_anchor_cleanup(skb); 2088 __skb_unlink(skb, &sk->sk_write_queue); 2089 } 2090 2091 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 2092 2093 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 2094 { 2095 tcp_skb_tsorted_anchor_cleanup(skb); 2096 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 2097 } 2098 2099 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 2100 { 2101 list_del(&skb->tcp_tsorted_anchor); 2102 tcp_rtx_queue_unlink(skb, sk); 2103 tcp_wmem_free_skb(sk, skb); 2104 } 2105 2106 static inline void tcp_write_collapse_fence(struct sock *sk) 2107 { 2108 struct sk_buff *skb = tcp_write_queue_tail(sk); 2109 2110 if (skb) 2111 TCP_SKB_CB(skb)->eor = 1; 2112 } 2113 2114 static inline void tcp_push_pending_frames(struct sock *sk) 2115 { 2116 if (tcp_send_head(sk)) { 2117 struct tcp_sock *tp = tcp_sk(sk); 2118 2119 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 2120 } 2121 } 2122 2123 /* Start sequence of the skb just after the highest skb with SACKed 2124 * bit, valid only if sacked_out > 0 or when the caller has ensured 2125 * validity by itself. 2126 */ 2127 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 2128 { 2129 if (!tp->sacked_out) 2130 return tp->snd_una; 2131 2132 if (tp->highest_sack == NULL) 2133 return tp->snd_nxt; 2134 2135 return TCP_SKB_CB(tp->highest_sack)->seq; 2136 } 2137 2138 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 2139 { 2140 tcp_sk(sk)->highest_sack = skb_rb_next(skb); 2141 } 2142 2143 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 2144 { 2145 return tcp_sk(sk)->highest_sack; 2146 } 2147 2148 static inline void tcp_highest_sack_reset(struct sock *sk) 2149 { 2150 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); 2151 } 2152 2153 /* Called when old skb is about to be deleted and replaced by new skb */ 2154 static inline void tcp_highest_sack_replace(struct sock *sk, 2155 struct sk_buff *old, 2156 struct sk_buff *new) 2157 { 2158 if (old == tcp_highest_sack(sk)) 2159 tcp_sk(sk)->highest_sack = new; 2160 } 2161 2162 /* This helper checks if socket has IP_TRANSPARENT set */ 2163 static inline bool inet_sk_transparent(const struct sock *sk) 2164 { 2165 switch (sk->sk_state) { 2166 case TCP_TIME_WAIT: 2167 return inet_twsk(sk)->tw_transparent; 2168 case TCP_NEW_SYN_RECV: 2169 return inet_rsk(inet_reqsk(sk))->no_srccheck; 2170 } 2171 return inet_test_bit(TRANSPARENT, sk); 2172 } 2173 2174 /* Determines whether this is a thin stream (which may suffer from 2175 * increased latency). Used to trigger latency-reducing mechanisms. 2176 */ 2177 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 2178 { 2179 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 2180 } 2181 2182 /* /proc */ 2183 enum tcp_seq_states { 2184 TCP_SEQ_STATE_LISTENING, 2185 TCP_SEQ_STATE_ESTABLISHED, 2186 }; 2187 2188 void *tcp_seq_start(struct seq_file *seq, loff_t *pos); 2189 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); 2190 void tcp_seq_stop(struct seq_file *seq, void *v); 2191 2192 struct tcp_seq_afinfo { 2193 sa_family_t family; 2194 }; 2195 2196 struct tcp_iter_state { 2197 struct seq_net_private p; 2198 enum tcp_seq_states state; 2199 struct sock *syn_wait_sk; 2200 int bucket, offset, sbucket, num; 2201 loff_t last_pos; 2202 }; 2203 2204 extern struct request_sock_ops tcp_request_sock_ops; 2205 extern struct request_sock_ops tcp6_request_sock_ops; 2206 2207 void tcp_v4_destroy_sock(struct sock *sk); 2208 2209 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 2210 netdev_features_t features); 2211 struct tcphdr *tcp_gro_pull_header(struct sk_buff *skb); 2212 struct sk_buff *tcp_gro_lookup(struct list_head *head, struct tcphdr *th); 2213 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb, 2214 struct tcphdr *th); 2215 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff)); 2216 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)); 2217 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff)); 2218 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb)); 2219 #ifdef CONFIG_INET 2220 void tcp_gro_complete(struct sk_buff *skb); 2221 #else 2222 static inline void tcp_gro_complete(struct sk_buff *skb) { } 2223 #endif 2224 2225 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 2226 2227 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 2228 { 2229 struct net *net = sock_net((struct sock *)tp); 2230 u32 val; 2231 2232 val = READ_ONCE(tp->notsent_lowat); 2233 2234 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat); 2235 } 2236 2237 bool tcp_stream_memory_free(const struct sock *sk, int wake); 2238 2239 #ifdef CONFIG_PROC_FS 2240 int tcp4_proc_init(void); 2241 void tcp4_proc_exit(void); 2242 #endif 2243 2244 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 2245 int tcp_conn_request(struct request_sock_ops *rsk_ops, 2246 const struct tcp_request_sock_ops *af_ops, 2247 struct sock *sk, struct sk_buff *skb); 2248 2249 /* TCP af-specific functions */ 2250 struct tcp_sock_af_ops { 2251 #ifdef CONFIG_TCP_MD5SIG 2252 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 2253 const struct sock *addr_sk); 2254 int (*calc_md5_hash)(char *location, 2255 const struct tcp_md5sig_key *md5, 2256 const struct sock *sk, 2257 const struct sk_buff *skb); 2258 int (*md5_parse)(struct sock *sk, 2259 int optname, 2260 sockptr_t optval, 2261 int optlen); 2262 #endif 2263 #ifdef CONFIG_TCP_AO 2264 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen); 2265 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2266 struct sock *addr_sk, 2267 int sndid, int rcvid); 2268 int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key, 2269 const struct sock *sk, 2270 __be32 sisn, __be32 disn, bool send); 2271 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao, 2272 const struct sock *sk, const struct sk_buff *skb, 2273 const u8 *tkey, int hash_offset, u32 sne); 2274 #endif 2275 }; 2276 2277 struct tcp_request_sock_ops { 2278 u16 mss_clamp; 2279 #ifdef CONFIG_TCP_MD5SIG 2280 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 2281 const struct sock *addr_sk); 2282 int (*calc_md5_hash) (char *location, 2283 const struct tcp_md5sig_key *md5, 2284 const struct sock *sk, 2285 const struct sk_buff *skb); 2286 #endif 2287 #ifdef CONFIG_TCP_AO 2288 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk, 2289 struct request_sock *req, 2290 int sndid, int rcvid); 2291 int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk); 2292 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt, 2293 struct request_sock *req, const struct sk_buff *skb, 2294 int hash_offset, u32 sne); 2295 #endif 2296 #ifdef CONFIG_SYN_COOKIES 2297 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 2298 __u16 *mss); 2299 #endif 2300 struct dst_entry *(*route_req)(const struct sock *sk, 2301 struct sk_buff *skb, 2302 struct flowi *fl, 2303 struct request_sock *req, 2304 u32 tw_isn); 2305 u32 (*init_seq)(const struct sk_buff *skb); 2306 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 2307 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 2308 struct flowi *fl, struct request_sock *req, 2309 struct tcp_fastopen_cookie *foc, 2310 enum tcp_synack_type synack_type, 2311 struct sk_buff *syn_skb); 2312 }; 2313 2314 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; 2315 #if IS_ENABLED(CONFIG_IPV6) 2316 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops; 2317 #endif 2318 2319 #ifdef CONFIG_SYN_COOKIES 2320 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2321 const struct sock *sk, struct sk_buff *skb, 2322 __u16 *mss) 2323 { 2324 tcp_synq_overflow(sk); 2325 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 2326 return ops->cookie_init_seq(skb, mss); 2327 } 2328 #else 2329 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 2330 const struct sock *sk, struct sk_buff *skb, 2331 __u16 *mss) 2332 { 2333 return 0; 2334 } 2335 #endif 2336 2337 struct tcp_key { 2338 union { 2339 struct { 2340 struct tcp_ao_key *ao_key; 2341 char *traffic_key; 2342 u32 sne; 2343 u8 rcv_next; 2344 }; 2345 struct tcp_md5sig_key *md5_key; 2346 }; 2347 enum { 2348 TCP_KEY_NONE = 0, 2349 TCP_KEY_MD5, 2350 TCP_KEY_AO, 2351 } type; 2352 }; 2353 2354 static inline void tcp_get_current_key(const struct sock *sk, 2355 struct tcp_key *out) 2356 { 2357 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG) 2358 const struct tcp_sock *tp = tcp_sk(sk); 2359 #endif 2360 2361 #ifdef CONFIG_TCP_AO 2362 if (static_branch_unlikely(&tcp_ao_needed.key)) { 2363 struct tcp_ao_info *ao; 2364 2365 ao = rcu_dereference_protected(tp->ao_info, 2366 lockdep_sock_is_held(sk)); 2367 if (ao) { 2368 out->ao_key = READ_ONCE(ao->current_key); 2369 out->type = TCP_KEY_AO; 2370 return; 2371 } 2372 } 2373 #endif 2374 #ifdef CONFIG_TCP_MD5SIG 2375 if (static_branch_unlikely(&tcp_md5_needed.key) && 2376 rcu_access_pointer(tp->md5sig_info)) { 2377 out->md5_key = tp->af_specific->md5_lookup(sk, sk); 2378 if (out->md5_key) { 2379 out->type = TCP_KEY_MD5; 2380 return; 2381 } 2382 } 2383 #endif 2384 out->type = TCP_KEY_NONE; 2385 } 2386 2387 static inline bool tcp_key_is_md5(const struct tcp_key *key) 2388 { 2389 #ifdef CONFIG_TCP_MD5SIG 2390 if (static_branch_unlikely(&tcp_md5_needed.key) && 2391 key->type == TCP_KEY_MD5) 2392 return true; 2393 #endif 2394 return false; 2395 } 2396 2397 static inline bool tcp_key_is_ao(const struct tcp_key *key) 2398 { 2399 #ifdef CONFIG_TCP_AO 2400 if (static_branch_unlikely(&tcp_ao_needed.key) && 2401 key->type == TCP_KEY_AO) 2402 return true; 2403 #endif 2404 return false; 2405 } 2406 2407 int tcpv4_offload_init(void); 2408 2409 void tcp_v4_init(void); 2410 void tcp_init(void); 2411 2412 /* tcp_recovery.c */ 2413 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); 2414 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); 2415 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, 2416 u32 reo_wnd); 2417 extern bool tcp_rack_mark_lost(struct sock *sk); 2418 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 2419 u64 xmit_time); 2420 extern void tcp_rack_reo_timeout(struct sock *sk); 2421 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 2422 2423 /* tcp_plb.c */ 2424 2425 /* 2426 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state 2427 * expects cong_ratio which represents fraction of traffic that experienced 2428 * congestion over a single RTT. In order to avoid floating point operations, 2429 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in. 2430 */ 2431 #define TCP_PLB_SCALE 8 2432 2433 /* State for PLB (Protective Load Balancing) for a single TCP connection. */ 2434 struct tcp_plb_state { 2435 u8 consec_cong_rounds:5, /* consecutive congested rounds */ 2436 unused:3; 2437 u32 pause_until; /* jiffies32 when PLB can resume rerouting */ 2438 }; 2439 2440 static inline void tcp_plb_init(const struct sock *sk, 2441 struct tcp_plb_state *plb) 2442 { 2443 plb->consec_cong_rounds = 0; 2444 plb->pause_until = 0; 2445 } 2446 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb, 2447 const int cong_ratio); 2448 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb); 2449 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb); 2450 2451 /* At how many usecs into the future should the RTO fire? */ 2452 static inline s64 tcp_rto_delta_us(const struct sock *sk) 2453 { 2454 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 2455 u32 rto = inet_csk(sk)->icsk_rto; 2456 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); 2457 2458 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 2459 } 2460 2461 /* 2462 * Save and compile IPv4 options, return a pointer to it 2463 */ 2464 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 2465 struct sk_buff *skb) 2466 { 2467 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 2468 struct ip_options_rcu *dopt = NULL; 2469 2470 if (opt->optlen) { 2471 int opt_size = sizeof(*dopt) + opt->optlen; 2472 2473 dopt = kmalloc(opt_size, GFP_ATOMIC); 2474 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 2475 kfree(dopt); 2476 dopt = NULL; 2477 } 2478 } 2479 return dopt; 2480 } 2481 2482 /* locally generated TCP pure ACKs have skb->truesize == 2 2483 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 2484 * This is much faster than dissecting the packet to find out. 2485 * (Think of GRE encapsulations, IPv4, IPv6, ...) 2486 */ 2487 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 2488 { 2489 return skb->truesize == 2; 2490 } 2491 2492 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 2493 { 2494 skb->truesize = 2; 2495 } 2496 2497 static inline int tcp_inq(struct sock *sk) 2498 { 2499 struct tcp_sock *tp = tcp_sk(sk); 2500 int answ; 2501 2502 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 2503 answ = 0; 2504 } else if (sock_flag(sk, SOCK_URGINLINE) || 2505 !tp->urg_data || 2506 before(tp->urg_seq, tp->copied_seq) || 2507 !before(tp->urg_seq, tp->rcv_nxt)) { 2508 2509 answ = tp->rcv_nxt - tp->copied_seq; 2510 2511 /* Subtract 1, if FIN was received */ 2512 if (answ && sock_flag(sk, SOCK_DONE)) 2513 answ--; 2514 } else { 2515 answ = tp->urg_seq - tp->copied_seq; 2516 } 2517 2518 return answ; 2519 } 2520 2521 int tcp_peek_len(struct socket *sock); 2522 2523 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 2524 { 2525 u16 segs_in; 2526 2527 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2528 2529 /* We update these fields while other threads might 2530 * read them from tcp_get_info() 2531 */ 2532 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in); 2533 if (skb->len > tcp_hdrlen(skb)) 2534 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in); 2535 } 2536 2537 /* 2538 * TCP listen path runs lockless. 2539 * We forced "struct sock" to be const qualified to make sure 2540 * we don't modify one of its field by mistake. 2541 * Here, we increment sk_drops which is an atomic_t, so we can safely 2542 * make sock writable again. 2543 */ 2544 static inline void tcp_listendrop(const struct sock *sk) 2545 { 2546 atomic_inc(&((struct sock *)sk)->sk_drops); 2547 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2548 } 2549 2550 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2551 2552 /* 2553 * Interface for adding Upper Level Protocols over TCP 2554 */ 2555 2556 #define TCP_ULP_NAME_MAX 16 2557 #define TCP_ULP_MAX 128 2558 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2559 2560 struct tcp_ulp_ops { 2561 struct list_head list; 2562 2563 /* initialize ulp */ 2564 int (*init)(struct sock *sk); 2565 /* update ulp */ 2566 void (*update)(struct sock *sk, struct proto *p, 2567 void (*write_space)(struct sock *sk)); 2568 /* cleanup ulp */ 2569 void (*release)(struct sock *sk); 2570 /* diagnostic */ 2571 int (*get_info)(struct sock *sk, struct sk_buff *skb); 2572 size_t (*get_info_size)(const struct sock *sk); 2573 /* clone ulp */ 2574 void (*clone)(const struct request_sock *req, struct sock *newsk, 2575 const gfp_t priority); 2576 2577 char name[TCP_ULP_NAME_MAX]; 2578 struct module *owner; 2579 }; 2580 int tcp_register_ulp(struct tcp_ulp_ops *type); 2581 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2582 int tcp_set_ulp(struct sock *sk, const char *name); 2583 void tcp_get_available_ulp(char *buf, size_t len); 2584 void tcp_cleanup_ulp(struct sock *sk); 2585 void tcp_update_ulp(struct sock *sk, struct proto *p, 2586 void (*write_space)(struct sock *sk)); 2587 2588 #define MODULE_ALIAS_TCP_ULP(name) \ 2589 __MODULE_INFO(alias, alias_userspace, name); \ 2590 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) 2591 2592 #ifdef CONFIG_NET_SOCK_MSG 2593 struct sk_msg; 2594 struct sk_psock; 2595 2596 #ifdef CONFIG_BPF_SYSCALL 2597 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore); 2598 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk); 2599 #endif /* CONFIG_BPF_SYSCALL */ 2600 2601 #ifdef CONFIG_INET 2602 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb); 2603 #else 2604 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 2605 { 2606 } 2607 #endif 2608 2609 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 2610 struct sk_msg *msg, u32 bytes, int flags); 2611 #endif /* CONFIG_NET_SOCK_MSG */ 2612 2613 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG) 2614 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 2615 { 2616 } 2617 #endif 2618 2619 #ifdef CONFIG_CGROUP_BPF 2620 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2621 struct sk_buff *skb, 2622 unsigned int end_offset) 2623 { 2624 skops->skb = skb; 2625 skops->skb_data_end = skb->data + end_offset; 2626 } 2627 #else 2628 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, 2629 struct sk_buff *skb, 2630 unsigned int end_offset) 2631 { 2632 } 2633 #endif 2634 2635 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2636 * is < 0, then the BPF op failed (for example if the loaded BPF 2637 * program does not support the chosen operation or there is no BPF 2638 * program loaded). 2639 */ 2640 #ifdef CONFIG_BPF 2641 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2642 { 2643 struct bpf_sock_ops_kern sock_ops; 2644 int ret; 2645 2646 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 2647 if (sk_fullsock(sk)) { 2648 sock_ops.is_fullsock = 1; 2649 sock_owned_by_me(sk); 2650 } 2651 2652 sock_ops.sk = sk; 2653 sock_ops.op = op; 2654 if (nargs > 0) 2655 memcpy(sock_ops.args, args, nargs * sizeof(*args)); 2656 2657 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2658 if (ret == 0) 2659 ret = sock_ops.reply; 2660 else 2661 ret = -1; 2662 return ret; 2663 } 2664 2665 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2666 { 2667 u32 args[2] = {arg1, arg2}; 2668 2669 return tcp_call_bpf(sk, op, 2, args); 2670 } 2671 2672 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2673 u32 arg3) 2674 { 2675 u32 args[3] = {arg1, arg2, arg3}; 2676 2677 return tcp_call_bpf(sk, op, 3, args); 2678 } 2679 2680 #else 2681 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) 2682 { 2683 return -EPERM; 2684 } 2685 2686 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) 2687 { 2688 return -EPERM; 2689 } 2690 2691 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, 2692 u32 arg3) 2693 { 2694 return -EPERM; 2695 } 2696 2697 #endif 2698 2699 static inline u32 tcp_timeout_init(struct sock *sk) 2700 { 2701 int timeout; 2702 2703 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); 2704 2705 if (timeout <= 0) 2706 timeout = TCP_TIMEOUT_INIT; 2707 return min_t(int, timeout, TCP_RTO_MAX); 2708 } 2709 2710 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2711 { 2712 int rwnd; 2713 2714 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); 2715 2716 if (rwnd < 0) 2717 rwnd = 0; 2718 return rwnd; 2719 } 2720 2721 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2722 { 2723 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); 2724 } 2725 2726 static inline void tcp_bpf_rtt(struct sock *sk, long mrtt, u32 srtt) 2727 { 2728 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) 2729 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_RTT_CB, mrtt, srtt); 2730 } 2731 2732 #if IS_ENABLED(CONFIG_SMC) 2733 extern struct static_key_false tcp_have_smc; 2734 #endif 2735 2736 #if IS_ENABLED(CONFIG_TLS_DEVICE) 2737 void clean_acked_data_enable(struct inet_connection_sock *icsk, 2738 void (*cad)(struct sock *sk, u32 ack_seq)); 2739 void clean_acked_data_disable(struct inet_connection_sock *icsk); 2740 void clean_acked_data_flush(void); 2741 #endif 2742 2743 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); 2744 static inline void tcp_add_tx_delay(struct sk_buff *skb, 2745 const struct tcp_sock *tp) 2746 { 2747 if (static_branch_unlikely(&tcp_tx_delay_enabled)) 2748 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; 2749 } 2750 2751 /* Compute Earliest Departure Time for some control packets 2752 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. 2753 */ 2754 static inline u64 tcp_transmit_time(const struct sock *sk) 2755 { 2756 if (static_branch_unlikely(&tcp_tx_delay_enabled)) { 2757 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? 2758 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; 2759 2760 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; 2761 } 2762 return 0; 2763 } 2764 2765 static inline int tcp_parse_auth_options(const struct tcphdr *th, 2766 const u8 **md5_hash, const struct tcp_ao_hdr **aoh) 2767 { 2768 const u8 *md5_tmp, *ao_tmp; 2769 int ret; 2770 2771 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp); 2772 if (ret) 2773 return ret; 2774 2775 if (md5_hash) 2776 *md5_hash = md5_tmp; 2777 2778 if (aoh) { 2779 if (!ao_tmp) 2780 *aoh = NULL; 2781 else 2782 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2); 2783 } 2784 2785 return 0; 2786 } 2787 2788 static inline bool tcp_ao_required(struct sock *sk, const void *saddr, 2789 int family, int l3index, bool stat_inc) 2790 { 2791 #ifdef CONFIG_TCP_AO 2792 struct tcp_ao_info *ao_info; 2793 struct tcp_ao_key *ao_key; 2794 2795 if (!static_branch_unlikely(&tcp_ao_needed.key)) 2796 return false; 2797 2798 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info, 2799 lockdep_sock_is_held(sk)); 2800 if (!ao_info) 2801 return false; 2802 2803 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1); 2804 if (ao_info->ao_required || ao_key) { 2805 if (stat_inc) { 2806 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED); 2807 atomic64_inc(&ao_info->counters.ao_required); 2808 } 2809 return true; 2810 } 2811 #endif 2812 return false; 2813 } 2814 2815 /* Called with rcu_read_lock() */ 2816 static inline enum skb_drop_reason 2817 tcp_inbound_hash(struct sock *sk, const struct request_sock *req, 2818 const struct sk_buff *skb, 2819 const void *saddr, const void *daddr, 2820 int family, int dif, int sdif) 2821 { 2822 const struct tcphdr *th = tcp_hdr(skb); 2823 const struct tcp_ao_hdr *aoh; 2824 const __u8 *md5_location; 2825 int l3index; 2826 2827 /* Invalid option or two times meet any of auth options */ 2828 if (tcp_parse_auth_options(th, &md5_location, &aoh)) { 2829 tcp_hash_fail("TCP segment has incorrect auth options set", 2830 family, skb, ""); 2831 return SKB_DROP_REASON_TCP_AUTH_HDR; 2832 } 2833 2834 if (req) { 2835 if (tcp_rsk_used_ao(req) != !!aoh) { 2836 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD); 2837 tcp_hash_fail("TCP connection can't start/end using TCP-AO", 2838 family, skb, "%s", 2839 !aoh ? "missing AO" : "AO signed"); 2840 return SKB_DROP_REASON_TCP_AOFAILURE; 2841 } 2842 } 2843 2844 /* sdif set, means packet ingressed via a device 2845 * in an L3 domain and dif is set to the l3mdev 2846 */ 2847 l3index = sdif ? dif : 0; 2848 2849 /* Fast path: unsigned segments */ 2850 if (likely(!md5_location && !aoh)) { 2851 /* Drop if there's TCP-MD5 or TCP-AO key with any rcvid/sndid 2852 * for the remote peer. On TCP-AO established connection 2853 * the last key is impossible to remove, so there's 2854 * always at least one current_key. 2855 */ 2856 if (tcp_ao_required(sk, saddr, family, l3index, true)) { 2857 tcp_hash_fail("AO hash is required, but not found", 2858 family, skb, "L3 index %d", l3index); 2859 return SKB_DROP_REASON_TCP_AONOTFOUND; 2860 } 2861 if (unlikely(tcp_md5_do_lookup(sk, l3index, saddr, family))) { 2862 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); 2863 tcp_hash_fail("MD5 Hash not found", 2864 family, skb, "L3 index %d", l3index); 2865 return SKB_DROP_REASON_TCP_MD5NOTFOUND; 2866 } 2867 return SKB_NOT_DROPPED_YET; 2868 } 2869 2870 if (aoh) 2871 return tcp_inbound_ao_hash(sk, skb, family, req, l3index, aoh); 2872 2873 return tcp_inbound_md5_hash(sk, skb, saddr, daddr, family, 2874 l3index, md5_location); 2875 } 2876 2877 #endif /* _TCP_H */ 2878