| 1 | /* |
| 2 | * INET An implementation of the TCP/IP protocol suite for the LINUX |
| 3 | * operating system. INET is implemented using the BSD Socket |
| 4 | * interface as the means of communication with the user level. |
| 5 | * |
| 6 | * Definitions for the TCP module. |
| 7 | * |
| 8 | * Version: @(#)tcp.h 1.0.5 05/23/93 |
| 9 | * |
| 10 | * Authors: Ross Biro |
| 11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
| 12 | * |
| 13 | * This program is free software; you can redistribute it and/or |
| 14 | * modify it under the terms of the GNU General Public License |
| 15 | * as published by the Free Software Foundation; either version |
| 16 | * 2 of the License, or (at your option) any later version. |
| 17 | */ |
| 18 | #ifndef _TCP_H |
| 19 | #define _TCP_H |
| 20 | |
| 21 | #define FASTRETRANS_DEBUG 1 |
| 22 | |
| 23 | #include <linux/list.h> |
| 24 | #include <linux/tcp.h> |
| 25 | #include <linux/bug.h> |
| 26 | #include <linux/slab.h> |
| 27 | #include <linux/cache.h> |
| 28 | #include <linux/percpu.h> |
| 29 | #include <linux/skbuff.h> |
| 30 | #include <linux/cryptohash.h> |
| 31 | #include <linux/kref.h> |
| 32 | #include <linux/ktime.h> |
| 33 | |
| 34 | #include <net/inet_connection_sock.h> |
| 35 | #include <net/inet_timewait_sock.h> |
| 36 | #include <net/inet_hashtables.h> |
| 37 | #include <net/checksum.h> |
| 38 | #include <net/request_sock.h> |
| 39 | #include <net/sock.h> |
| 40 | #include <net/snmp.h> |
| 41 | #include <net/ip.h> |
| 42 | #include <net/tcp_states.h> |
| 43 | #include <net/inet_ecn.h> |
| 44 | #include <net/dst.h> |
| 45 | |
| 46 | #include <linux/seq_file.h> |
| 47 | #include <linux/memcontrol.h> |
| 48 | |
| 49 | extern struct inet_hashinfo tcp_hashinfo; |
| 50 | |
| 51 | extern struct percpu_counter tcp_orphan_count; |
| 52 | void tcp_time_wait(struct sock *sk, int state, int timeo); |
| 53 | |
| 54 | #define MAX_TCP_HEADER (128 + MAX_HEADER) |
| 55 | #define MAX_TCP_OPTION_SPACE 40 |
| 56 | |
| 57 | /* |
| 58 | * Never offer a window over 32767 without using window scaling. Some |
| 59 | * poor stacks do signed 16bit maths! |
| 60 | */ |
| 61 | #define MAX_TCP_WINDOW 32767U |
| 62 | |
| 63 | /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ |
| 64 | #define TCP_MIN_MSS 88U |
| 65 | |
| 66 | /* The least MTU to use for probing */ |
| 67 | #define TCP_BASE_MSS 1024 |
| 68 | |
| 69 | /* probing interval, default to 10 minutes as per RFC4821 */ |
| 70 | #define TCP_PROBE_INTERVAL 600 |
| 71 | |
| 72 | /* Specify interval when tcp mtu probing will stop */ |
| 73 | #define TCP_PROBE_THRESHOLD 8 |
| 74 | |
| 75 | /* After receiving this amount of duplicate ACKs fast retransmit starts. */ |
| 76 | #define TCP_FASTRETRANS_THRESH 3 |
| 77 | |
| 78 | /* Maximal number of ACKs sent quickly to accelerate slow-start. */ |
| 79 | #define TCP_MAX_QUICKACKS 16U |
| 80 | |
| 81 | /* urg_data states */ |
| 82 | #define TCP_URG_VALID 0x0100 |
| 83 | #define TCP_URG_NOTYET 0x0200 |
| 84 | #define TCP_URG_READ 0x0400 |
| 85 | |
| 86 | #define TCP_RETR1 3 /* |
| 87 | * This is how many retries it does before it |
| 88 | * tries to figure out if the gateway is |
| 89 | * down. Minimal RFC value is 3; it corresponds |
| 90 | * to ~3sec-8min depending on RTO. |
| 91 | */ |
| 92 | |
| 93 | #define TCP_RETR2 15 /* |
| 94 | * This should take at least |
| 95 | * 90 minutes to time out. |
| 96 | * RFC1122 says that the limit is 100 sec. |
| 97 | * 15 is ~13-30min depending on RTO. |
| 98 | */ |
| 99 | |
| 100 | #define TCP_SYN_RETRIES 6 /* This is how many retries are done |
| 101 | * when active opening a connection. |
| 102 | * RFC1122 says the minimum retry MUST |
| 103 | * be at least 180secs. Nevertheless |
| 104 | * this value is corresponding to |
| 105 | * 63secs of retransmission with the |
| 106 | * current initial RTO. |
| 107 | */ |
| 108 | |
| 109 | #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done |
| 110 | * when passive opening a connection. |
| 111 | * This is corresponding to 31secs of |
| 112 | * retransmission with the current |
| 113 | * initial RTO. |
| 114 | */ |
| 115 | |
| 116 | #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT |
| 117 | * state, about 60 seconds */ |
| 118 | #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN |
| 119 | /* BSD style FIN_WAIT2 deadlock breaker. |
| 120 | * It used to be 3min, new value is 60sec, |
| 121 | * to combine FIN-WAIT-2 timeout with |
| 122 | * TIME-WAIT timer. |
| 123 | */ |
| 124 | |
| 125 | #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ |
| 126 | #if HZ >= 100 |
| 127 | #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ |
| 128 | #define TCP_ATO_MIN ((unsigned)(HZ/25)) |
| 129 | #else |
| 130 | #define TCP_DELACK_MIN 4U |
| 131 | #define TCP_ATO_MIN 4U |
| 132 | #endif |
| 133 | #define TCP_RTO_MAX ((unsigned)(120*HZ)) |
| 134 | #define TCP_RTO_MIN ((unsigned)(HZ/5)) |
| 135 | #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ |
| 136 | #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now |
| 137 | * used as a fallback RTO for the |
| 138 | * initial data transmission if no |
| 139 | * valid RTT sample has been acquired, |
| 140 | * most likely due to retrans in 3WHS. |
| 141 | */ |
| 142 | |
| 143 | #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes |
| 144 | * for local resources. |
| 145 | */ |
| 146 | |
| 147 | #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ |
| 148 | #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ |
| 149 | #define TCP_KEEPALIVE_INTVL (75*HZ) |
| 150 | |
| 151 | #define MAX_TCP_KEEPIDLE 32767 |
| 152 | #define MAX_TCP_KEEPINTVL 32767 |
| 153 | #define MAX_TCP_KEEPCNT 127 |
| 154 | #define MAX_TCP_SYNCNT 127 |
| 155 | |
| 156 | #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ |
| 157 | |
| 158 | #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) |
| 159 | #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated |
| 160 | * after this time. It should be equal |
| 161 | * (or greater than) TCP_TIMEWAIT_LEN |
| 162 | * to provide reliability equal to one |
| 163 | * provided by timewait state. |
| 164 | */ |
| 165 | #define TCP_PAWS_WINDOW 1 /* Replay window for per-host |
| 166 | * timestamps. It must be less than |
| 167 | * minimal timewait lifetime. |
| 168 | */ |
| 169 | /* |
| 170 | * TCP option |
| 171 | */ |
| 172 | |
| 173 | #define TCPOPT_NOP 1 /* Padding */ |
| 174 | #define TCPOPT_EOL 0 /* End of options */ |
| 175 | #define TCPOPT_MSS 2 /* Segment size negotiating */ |
| 176 | #define TCPOPT_WINDOW 3 /* Window scaling */ |
| 177 | #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ |
| 178 | #define TCPOPT_SACK 5 /* SACK Block */ |
| 179 | #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ |
| 180 | #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ |
| 181 | #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ |
| 182 | #define TCPOPT_EXP 254 /* Experimental */ |
| 183 | /* Magic number to be after the option value for sharing TCP |
| 184 | * experimental options. See draft-ietf-tcpm-experimental-options-00.txt |
| 185 | */ |
| 186 | #define TCPOPT_FASTOPEN_MAGIC 0xF989 |
| 187 | |
| 188 | /* |
| 189 | * TCP option lengths |
| 190 | */ |
| 191 | |
| 192 | #define TCPOLEN_MSS 4 |
| 193 | #define TCPOLEN_WINDOW 3 |
| 194 | #define TCPOLEN_SACK_PERM 2 |
| 195 | #define TCPOLEN_TIMESTAMP 10 |
| 196 | #define TCPOLEN_MD5SIG 18 |
| 197 | #define TCPOLEN_FASTOPEN_BASE 2 |
| 198 | #define TCPOLEN_EXP_FASTOPEN_BASE 4 |
| 199 | |
| 200 | /* But this is what stacks really send out. */ |
| 201 | #define TCPOLEN_TSTAMP_ALIGNED 12 |
| 202 | #define TCPOLEN_WSCALE_ALIGNED 4 |
| 203 | #define TCPOLEN_SACKPERM_ALIGNED 4 |
| 204 | #define TCPOLEN_SACK_BASE 2 |
| 205 | #define TCPOLEN_SACK_BASE_ALIGNED 4 |
| 206 | #define TCPOLEN_SACK_PERBLOCK 8 |
| 207 | #define TCPOLEN_MD5SIG_ALIGNED 20 |
| 208 | #define TCPOLEN_MSS_ALIGNED 4 |
| 209 | |
| 210 | /* Flags in tp->nonagle */ |
| 211 | #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ |
| 212 | #define TCP_NAGLE_CORK 2 /* Socket is corked */ |
| 213 | #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ |
| 214 | |
| 215 | /* TCP thin-stream limits */ |
| 216 | #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ |
| 217 | |
| 218 | /* TCP initial congestion window as per rfc6928 */ |
| 219 | #define TCP_INIT_CWND 10 |
| 220 | |
| 221 | /* Bit Flags for sysctl_tcp_fastopen */ |
| 222 | #define TFO_CLIENT_ENABLE 1 |
| 223 | #define TFO_SERVER_ENABLE 2 |
| 224 | #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ |
| 225 | |
| 226 | /* Accept SYN data w/o any cookie option */ |
| 227 | #define TFO_SERVER_COOKIE_NOT_REQD 0x200 |
| 228 | |
| 229 | /* Force enable TFO on all listeners, i.e., not requiring the |
| 230 | * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. |
| 231 | */ |
| 232 | #define TFO_SERVER_WO_SOCKOPT1 0x400 |
| 233 | #define TFO_SERVER_WO_SOCKOPT2 0x800 |
| 234 | |
| 235 | extern struct inet_timewait_death_row tcp_death_row; |
| 236 | |
| 237 | /* sysctl variables for tcp */ |
| 238 | extern int sysctl_tcp_timestamps; |
| 239 | extern int sysctl_tcp_window_scaling; |
| 240 | extern int sysctl_tcp_sack; |
| 241 | extern int sysctl_tcp_fastopen; |
| 242 | extern int sysctl_tcp_retrans_collapse; |
| 243 | extern int sysctl_tcp_stdurg; |
| 244 | extern int sysctl_tcp_rfc1337; |
| 245 | extern int sysctl_tcp_abort_on_overflow; |
| 246 | extern int sysctl_tcp_max_orphans; |
| 247 | extern int sysctl_tcp_fack; |
| 248 | extern int sysctl_tcp_reordering; |
| 249 | extern int sysctl_tcp_max_reordering; |
| 250 | extern int sysctl_tcp_dsack; |
| 251 | extern long sysctl_tcp_mem[3]; |
| 252 | extern int sysctl_tcp_wmem[3]; |
| 253 | extern int sysctl_tcp_rmem[3]; |
| 254 | extern int sysctl_tcp_app_win; |
| 255 | extern int sysctl_tcp_adv_win_scale; |
| 256 | extern int sysctl_tcp_tw_reuse; |
| 257 | extern int sysctl_tcp_frto; |
| 258 | extern int sysctl_tcp_low_latency; |
| 259 | extern int sysctl_tcp_nometrics_save; |
| 260 | extern int sysctl_tcp_moderate_rcvbuf; |
| 261 | extern int sysctl_tcp_tso_win_divisor; |
| 262 | extern int sysctl_tcp_workaround_signed_windows; |
| 263 | extern int sysctl_tcp_slow_start_after_idle; |
| 264 | extern int sysctl_tcp_thin_linear_timeouts; |
| 265 | extern int sysctl_tcp_thin_dupack; |
| 266 | extern int sysctl_tcp_early_retrans; |
| 267 | extern int sysctl_tcp_limit_output_bytes; |
| 268 | extern int sysctl_tcp_challenge_ack_limit; |
| 269 | extern int sysctl_tcp_min_tso_segs; |
| 270 | extern int sysctl_tcp_min_rtt_wlen; |
| 271 | extern int sysctl_tcp_autocorking; |
| 272 | extern int sysctl_tcp_invalid_ratelimit; |
| 273 | extern int sysctl_tcp_pacing_ss_ratio; |
| 274 | extern int sysctl_tcp_pacing_ca_ratio; |
| 275 | |
| 276 | extern atomic_long_t tcp_memory_allocated; |
| 277 | extern struct percpu_counter tcp_sockets_allocated; |
| 278 | extern int tcp_memory_pressure; |
| 279 | |
| 280 | /* optimized version of sk_under_memory_pressure() for TCP sockets */ |
| 281 | static inline bool tcp_under_memory_pressure(const struct sock *sk) |
| 282 | { |
| 283 | if (mem_cgroup_sockets_enabled && sk->sk_memcg && |
| 284 | mem_cgroup_under_socket_pressure(sk->sk_memcg)) |
| 285 | return true; |
| 286 | |
| 287 | return tcp_memory_pressure; |
| 288 | } |
| 289 | /* |
| 290 | * The next routines deal with comparing 32 bit unsigned ints |
| 291 | * and worry about wraparound (automatic with unsigned arithmetic). |
| 292 | */ |
| 293 | |
| 294 | static inline bool before(__u32 seq1, __u32 seq2) |
| 295 | { |
| 296 | return (__s32)(seq1-seq2) < 0; |
| 297 | } |
| 298 | #define after(seq2, seq1) before(seq1, seq2) |
| 299 | |
| 300 | /* is s2<=s1<=s3 ? */ |
| 301 | static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) |
| 302 | { |
| 303 | return seq3 - seq2 >= seq1 - seq2; |
| 304 | } |
| 305 | |
| 306 | static inline bool tcp_out_of_memory(struct sock *sk) |
| 307 | { |
| 308 | if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && |
| 309 | sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) |
| 310 | return true; |
| 311 | return false; |
| 312 | } |
| 313 | |
| 314 | void sk_forced_mem_schedule(struct sock *sk, int size); |
| 315 | |
| 316 | static inline bool tcp_too_many_orphans(struct sock *sk, int shift) |
| 317 | { |
| 318 | struct percpu_counter *ocp = sk->sk_prot->orphan_count; |
| 319 | int orphans = percpu_counter_read_positive(ocp); |
| 320 | |
| 321 | if (orphans << shift > sysctl_tcp_max_orphans) { |
| 322 | orphans = percpu_counter_sum_positive(ocp); |
| 323 | if (orphans << shift > sysctl_tcp_max_orphans) |
| 324 | return true; |
| 325 | } |
| 326 | return false; |
| 327 | } |
| 328 | |
| 329 | bool tcp_check_oom(struct sock *sk, int shift); |
| 330 | |
| 331 | |
| 332 | extern struct proto tcp_prot; |
| 333 | |
| 334 | #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) |
| 335 | #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) |
| 336 | #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) |
| 337 | #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) |
| 338 | #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) |
| 339 | |
| 340 | void tcp_tasklet_init(void); |
| 341 | |
| 342 | void tcp_v4_err(struct sk_buff *skb, u32); |
| 343 | |
| 344 | void tcp_shutdown(struct sock *sk, int how); |
| 345 | |
| 346 | void tcp_v4_early_demux(struct sk_buff *skb); |
| 347 | int tcp_v4_rcv(struct sk_buff *skb); |
| 348 | |
| 349 | int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); |
| 350 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); |
| 351 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, |
| 352 | int flags); |
| 353 | void tcp_release_cb(struct sock *sk); |
| 354 | void tcp_wfree(struct sk_buff *skb); |
| 355 | void tcp_write_timer_handler(struct sock *sk); |
| 356 | void tcp_delack_timer_handler(struct sock *sk); |
| 357 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); |
| 358 | int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); |
| 359 | void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, |
| 360 | const struct tcphdr *th, unsigned int len); |
| 361 | void tcp_rcv_space_adjust(struct sock *sk); |
| 362 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); |
| 363 | void tcp_twsk_destructor(struct sock *sk); |
| 364 | ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, |
| 365 | struct pipe_inode_info *pipe, size_t len, |
| 366 | unsigned int flags); |
| 367 | |
| 368 | static inline void tcp_dec_quickack_mode(struct sock *sk, |
| 369 | const unsigned int pkts) |
| 370 | { |
| 371 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 372 | |
| 373 | if (icsk->icsk_ack.quick) { |
| 374 | if (pkts >= icsk->icsk_ack.quick) { |
| 375 | icsk->icsk_ack.quick = 0; |
| 376 | /* Leaving quickack mode we deflate ATO. */ |
| 377 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
| 378 | } else |
| 379 | icsk->icsk_ack.quick -= pkts; |
| 380 | } |
| 381 | } |
| 382 | |
| 383 | #define TCP_ECN_OK 1 |
| 384 | #define TCP_ECN_QUEUE_CWR 2 |
| 385 | #define TCP_ECN_DEMAND_CWR 4 |
| 386 | #define TCP_ECN_SEEN 8 |
| 387 | |
| 388 | enum tcp_tw_status { |
| 389 | TCP_TW_SUCCESS = 0, |
| 390 | TCP_TW_RST = 1, |
| 391 | TCP_TW_ACK = 2, |
| 392 | TCP_TW_SYN = 3 |
| 393 | }; |
| 394 | |
| 395 | |
| 396 | enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, |
| 397 | struct sk_buff *skb, |
| 398 | const struct tcphdr *th); |
| 399 | struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, |
| 400 | struct request_sock *req, bool fastopen); |
| 401 | int tcp_child_process(struct sock *parent, struct sock *child, |
| 402 | struct sk_buff *skb); |
| 403 | void tcp_enter_loss(struct sock *sk); |
| 404 | void tcp_clear_retrans(struct tcp_sock *tp); |
| 405 | void tcp_update_metrics(struct sock *sk); |
| 406 | void tcp_init_metrics(struct sock *sk); |
| 407 | void tcp_metrics_init(void); |
| 408 | bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, |
| 409 | bool paws_check, bool timestamps); |
| 410 | bool tcp_remember_stamp(struct sock *sk); |
| 411 | bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); |
| 412 | void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); |
| 413 | void tcp_disable_fack(struct tcp_sock *tp); |
| 414 | void tcp_close(struct sock *sk, long timeout); |
| 415 | void tcp_init_sock(struct sock *sk); |
| 416 | unsigned int tcp_poll(struct file *file, struct socket *sock, |
| 417 | struct poll_table_struct *wait); |
| 418 | int tcp_getsockopt(struct sock *sk, int level, int optname, |
| 419 | char __user *optval, int __user *optlen); |
| 420 | int tcp_setsockopt(struct sock *sk, int level, int optname, |
| 421 | char __user *optval, unsigned int optlen); |
| 422 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, |
| 423 | char __user *optval, int __user *optlen); |
| 424 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
| 425 | char __user *optval, unsigned int optlen); |
| 426 | void tcp_set_keepalive(struct sock *sk, int val); |
| 427 | void tcp_syn_ack_timeout(const struct request_sock *req); |
| 428 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, |
| 429 | int flags, int *addr_len); |
| 430 | void tcp_parse_options(const struct sk_buff *skb, |
| 431 | struct tcp_options_received *opt_rx, |
| 432 | int estab, struct tcp_fastopen_cookie *foc); |
| 433 | const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); |
| 434 | |
| 435 | /* |
| 436 | * TCP v4 functions exported for the inet6 API |
| 437 | */ |
| 438 | |
| 439 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); |
| 440 | void tcp_v4_mtu_reduced(struct sock *sk); |
| 441 | void tcp_req_err(struct sock *sk, u32 seq, bool abort); |
| 442 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); |
| 443 | struct sock *tcp_create_openreq_child(const struct sock *sk, |
| 444 | struct request_sock *req, |
| 445 | struct sk_buff *skb); |
| 446 | void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); |
| 447 | struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, |
| 448 | struct request_sock *req, |
| 449 | struct dst_entry *dst, |
| 450 | struct request_sock *req_unhash, |
| 451 | bool *own_req); |
| 452 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); |
| 453 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); |
| 454 | int tcp_connect(struct sock *sk); |
| 455 | struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, |
| 456 | struct request_sock *req, |
| 457 | struct tcp_fastopen_cookie *foc, |
| 458 | bool attach_req); |
| 459 | int tcp_disconnect(struct sock *sk, int flags); |
| 460 | |
| 461 | void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); |
| 462 | int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); |
| 463 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); |
| 464 | |
| 465 | /* From syncookies.c */ |
| 466 | struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, |
| 467 | struct request_sock *req, |
| 468 | struct dst_entry *dst); |
| 469 | int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, |
| 470 | u32 cookie); |
| 471 | struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); |
| 472 | #ifdef CONFIG_SYN_COOKIES |
| 473 | |
| 474 | /* Syncookies use a monotonic timer which increments every 60 seconds. |
| 475 | * This counter is used both as a hash input and partially encoded into |
| 476 | * the cookie value. A cookie is only validated further if the delta |
| 477 | * between the current counter value and the encoded one is less than this, |
| 478 | * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if |
| 479 | * the counter advances immediately after a cookie is generated). |
| 480 | */ |
| 481 | #define MAX_SYNCOOKIE_AGE 2 |
| 482 | #define TCP_SYNCOOKIE_PERIOD (60 * HZ) |
| 483 | #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) |
| 484 | |
| 485 | /* syncookies: remember time of last synqueue overflow |
| 486 | * But do not dirty this field too often (once per second is enough) |
| 487 | * It is racy as we do not hold a lock, but race is very minor. |
| 488 | */ |
| 489 | static inline void tcp_synq_overflow(const struct sock *sk) |
| 490 | { |
| 491 | unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; |
| 492 | unsigned long now = jiffies; |
| 493 | |
| 494 | if (time_after(now, last_overflow + HZ)) |
| 495 | tcp_sk(sk)->rx_opt.ts_recent_stamp = now; |
| 496 | } |
| 497 | |
| 498 | /* syncookies: no recent synqueue overflow on this listening socket? */ |
| 499 | static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) |
| 500 | { |
| 501 | unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; |
| 502 | |
| 503 | return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); |
| 504 | } |
| 505 | |
| 506 | static inline u32 tcp_cookie_time(void) |
| 507 | { |
| 508 | u64 val = get_jiffies_64(); |
| 509 | |
| 510 | do_div(val, TCP_SYNCOOKIE_PERIOD); |
| 511 | return val; |
| 512 | } |
| 513 | |
| 514 | u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, |
| 515 | u16 *mssp); |
| 516 | __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); |
| 517 | __u32 cookie_init_timestamp(struct request_sock *req); |
| 518 | bool cookie_timestamp_decode(struct tcp_options_received *opt); |
| 519 | bool cookie_ecn_ok(const struct tcp_options_received *opt, |
| 520 | const struct net *net, const struct dst_entry *dst); |
| 521 | |
| 522 | /* From net/ipv6/syncookies.c */ |
| 523 | int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, |
| 524 | u32 cookie); |
| 525 | struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); |
| 526 | |
| 527 | u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, |
| 528 | const struct tcphdr *th, u16 *mssp); |
| 529 | __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); |
| 530 | #endif |
| 531 | /* tcp_output.c */ |
| 532 | |
| 533 | void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, |
| 534 | int nonagle); |
| 535 | bool tcp_may_send_now(struct sock *sk); |
| 536 | int __tcp_retransmit_skb(struct sock *, struct sk_buff *); |
| 537 | int tcp_retransmit_skb(struct sock *, struct sk_buff *); |
| 538 | void tcp_retransmit_timer(struct sock *sk); |
| 539 | void tcp_xmit_retransmit_queue(struct sock *); |
| 540 | void tcp_simple_retransmit(struct sock *); |
| 541 | int tcp_trim_head(struct sock *, struct sk_buff *, u32); |
| 542 | int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); |
| 543 | |
| 544 | void tcp_send_probe0(struct sock *); |
| 545 | void tcp_send_partial(struct sock *); |
| 546 | int tcp_write_wakeup(struct sock *, int mib); |
| 547 | void tcp_send_fin(struct sock *sk); |
| 548 | void tcp_send_active_reset(struct sock *sk, gfp_t priority); |
| 549 | int tcp_send_synack(struct sock *); |
| 550 | void tcp_push_one(struct sock *, unsigned int mss_now); |
| 551 | void tcp_send_ack(struct sock *sk); |
| 552 | void tcp_send_delayed_ack(struct sock *sk); |
| 553 | void tcp_send_loss_probe(struct sock *sk); |
| 554 | bool tcp_schedule_loss_probe(struct sock *sk); |
| 555 | void tcp_skb_collapse_tstamp(struct sk_buff *skb, |
| 556 | const struct sk_buff *next_skb); |
| 557 | |
| 558 | /* tcp_input.c */ |
| 559 | void tcp_resume_early_retransmit(struct sock *sk); |
| 560 | void tcp_rearm_rto(struct sock *sk); |
| 561 | void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); |
| 562 | void tcp_reset(struct sock *sk); |
| 563 | void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); |
| 564 | void tcp_fin(struct sock *sk); |
| 565 | |
| 566 | /* tcp_timer.c */ |
| 567 | void tcp_init_xmit_timers(struct sock *); |
| 568 | static inline void tcp_clear_xmit_timers(struct sock *sk) |
| 569 | { |
| 570 | inet_csk_clear_xmit_timers(sk); |
| 571 | } |
| 572 | |
| 573 | unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); |
| 574 | unsigned int tcp_current_mss(struct sock *sk); |
| 575 | |
| 576 | /* Bound MSS / TSO packet size with the half of the window */ |
| 577 | static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) |
| 578 | { |
| 579 | int cutoff; |
| 580 | |
| 581 | /* When peer uses tiny windows, there is no use in packetizing |
| 582 | * to sub-MSS pieces for the sake of SWS or making sure there |
| 583 | * are enough packets in the pipe for fast recovery. |
| 584 | * |
| 585 | * On the other hand, for extremely large MSS devices, handling |
| 586 | * smaller than MSS windows in this way does make sense. |
| 587 | */ |
| 588 | if (tp->max_window >= 512) |
| 589 | cutoff = (tp->max_window >> 1); |
| 590 | else |
| 591 | cutoff = tp->max_window; |
| 592 | |
| 593 | if (cutoff && pktsize > cutoff) |
| 594 | return max_t(int, cutoff, 68U - tp->tcp_header_len); |
| 595 | else |
| 596 | return pktsize; |
| 597 | } |
| 598 | |
| 599 | /* tcp.c */ |
| 600 | void tcp_get_info(struct sock *, struct tcp_info *); |
| 601 | |
| 602 | /* Read 'sendfile()'-style from a TCP socket */ |
| 603 | typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, |
| 604 | unsigned int, size_t); |
| 605 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, |
| 606 | sk_read_actor_t recv_actor); |
| 607 | |
| 608 | void tcp_initialize_rcv_mss(struct sock *sk); |
| 609 | |
| 610 | int tcp_mtu_to_mss(struct sock *sk, int pmtu); |
| 611 | int tcp_mss_to_mtu(struct sock *sk, int mss); |
| 612 | void tcp_mtup_init(struct sock *sk); |
| 613 | void tcp_init_buffer_space(struct sock *sk); |
| 614 | |
| 615 | static inline void tcp_bound_rto(const struct sock *sk) |
| 616 | { |
| 617 | if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) |
| 618 | inet_csk(sk)->icsk_rto = TCP_RTO_MAX; |
| 619 | } |
| 620 | |
| 621 | static inline u32 __tcp_set_rto(const struct tcp_sock *tp) |
| 622 | { |
| 623 | return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); |
| 624 | } |
| 625 | |
| 626 | static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) |
| 627 | { |
| 628 | tp->pred_flags = htonl((tp->tcp_header_len << 26) | |
| 629 | ntohl(TCP_FLAG_ACK) | |
| 630 | snd_wnd); |
| 631 | } |
| 632 | |
| 633 | static inline void tcp_fast_path_on(struct tcp_sock *tp) |
| 634 | { |
| 635 | __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); |
| 636 | } |
| 637 | |
| 638 | static inline void tcp_fast_path_check(struct sock *sk) |
| 639 | { |
| 640 | struct tcp_sock *tp = tcp_sk(sk); |
| 641 | |
| 642 | if (skb_queue_empty(&tp->out_of_order_queue) && |
| 643 | tp->rcv_wnd && |
| 644 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && |
| 645 | !tp->urg_data) |
| 646 | tcp_fast_path_on(tp); |
| 647 | } |
| 648 | |
| 649 | /* Compute the actual rto_min value */ |
| 650 | static inline u32 tcp_rto_min(struct sock *sk) |
| 651 | { |
| 652 | const struct dst_entry *dst = __sk_dst_get(sk); |
| 653 | u32 rto_min = TCP_RTO_MIN; |
| 654 | |
| 655 | if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) |
| 656 | rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); |
| 657 | return rto_min; |
| 658 | } |
| 659 | |
| 660 | static inline u32 tcp_rto_min_us(struct sock *sk) |
| 661 | { |
| 662 | return jiffies_to_usecs(tcp_rto_min(sk)); |
| 663 | } |
| 664 | |
| 665 | static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) |
| 666 | { |
| 667 | return dst_metric_locked(dst, RTAX_CC_ALGO); |
| 668 | } |
| 669 | |
| 670 | /* Minimum RTT in usec. ~0 means not available. */ |
| 671 | static inline u32 tcp_min_rtt(const struct tcp_sock *tp) |
| 672 | { |
| 673 | return tp->rtt_min[0].rtt; |
| 674 | } |
| 675 | |
| 676 | /* Compute the actual receive window we are currently advertising. |
| 677 | * Rcv_nxt can be after the window if our peer push more data |
| 678 | * than the offered window. |
| 679 | */ |
| 680 | static inline u32 tcp_receive_window(const struct tcp_sock *tp) |
| 681 | { |
| 682 | s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; |
| 683 | |
| 684 | if (win < 0) |
| 685 | win = 0; |
| 686 | return (u32) win; |
| 687 | } |
| 688 | |
| 689 | /* Choose a new window, without checks for shrinking, and without |
| 690 | * scaling applied to the result. The caller does these things |
| 691 | * if necessary. This is a "raw" window selection. |
| 692 | */ |
| 693 | u32 __tcp_select_window(struct sock *sk); |
| 694 | |
| 695 | void tcp_send_window_probe(struct sock *sk); |
| 696 | |
| 697 | /* TCP timestamps are only 32-bits, this causes a slight |
| 698 | * complication on 64-bit systems since we store a snapshot |
| 699 | * of jiffies in the buffer control blocks below. We decided |
| 700 | * to use only the low 32-bits of jiffies and hide the ugly |
| 701 | * casts with the following macro. |
| 702 | */ |
| 703 | #define tcp_time_stamp ((__u32)(jiffies)) |
| 704 | |
| 705 | static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) |
| 706 | { |
| 707 | return skb->skb_mstamp.stamp_jiffies; |
| 708 | } |
| 709 | |
| 710 | |
| 711 | #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) |
| 712 | |
| 713 | #define TCPHDR_FIN 0x01 |
| 714 | #define TCPHDR_SYN 0x02 |
| 715 | #define TCPHDR_RST 0x04 |
| 716 | #define TCPHDR_PSH 0x08 |
| 717 | #define TCPHDR_ACK 0x10 |
| 718 | #define TCPHDR_URG 0x20 |
| 719 | #define TCPHDR_ECE 0x40 |
| 720 | #define TCPHDR_CWR 0x80 |
| 721 | |
| 722 | #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) |
| 723 | |
| 724 | /* This is what the send packet queuing engine uses to pass |
| 725 | * TCP per-packet control information to the transmission code. |
| 726 | * We also store the host-order sequence numbers in here too. |
| 727 | * This is 44 bytes if IPV6 is enabled. |
| 728 | * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. |
| 729 | */ |
| 730 | struct tcp_skb_cb { |
| 731 | __u32 seq; /* Starting sequence number */ |
| 732 | __u32 end_seq; /* SEQ + FIN + SYN + datalen */ |
| 733 | union { |
| 734 | /* Note : tcp_tw_isn is used in input path only |
| 735 | * (isn chosen by tcp_timewait_state_process()) |
| 736 | * |
| 737 | * tcp_gso_segs/size are used in write queue only, |
| 738 | * cf tcp_skb_pcount()/tcp_skb_mss() |
| 739 | */ |
| 740 | __u32 tcp_tw_isn; |
| 741 | struct { |
| 742 | u16 tcp_gso_segs; |
| 743 | u16 tcp_gso_size; |
| 744 | }; |
| 745 | }; |
| 746 | __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ |
| 747 | |
| 748 | __u8 sacked; /* State flags for SACK/FACK. */ |
| 749 | #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ |
| 750 | #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ |
| 751 | #define TCPCB_LOST 0x04 /* SKB is lost */ |
| 752 | #define TCPCB_TAGBITS 0x07 /* All tag bits */ |
| 753 | #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ |
| 754 | #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ |
| 755 | #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ |
| 756 | TCPCB_REPAIRED) |
| 757 | |
| 758 | __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ |
| 759 | /* 1 byte hole */ |
| 760 | __u32 ack_seq; /* Sequence number ACK'd */ |
| 761 | union { |
| 762 | struct inet_skb_parm h4; |
| 763 | #if IS_ENABLED(CONFIG_IPV6) |
| 764 | struct inet6_skb_parm h6; |
| 765 | #endif |
| 766 | } header; /* For incoming frames */ |
| 767 | }; |
| 768 | |
| 769 | #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) |
| 770 | |
| 771 | |
| 772 | #if IS_ENABLED(CONFIG_IPV6) |
| 773 | /* This is the variant of inet6_iif() that must be used by TCP, |
| 774 | * as TCP moves IP6CB into a different location in skb->cb[] |
| 775 | */ |
| 776 | static inline int tcp_v6_iif(const struct sk_buff *skb) |
| 777 | { |
| 778 | return TCP_SKB_CB(skb)->header.h6.iif; |
| 779 | } |
| 780 | #endif |
| 781 | |
| 782 | /* Due to TSO, an SKB can be composed of multiple actual |
| 783 | * packets. To keep these tracked properly, we use this. |
| 784 | */ |
| 785 | static inline int tcp_skb_pcount(const struct sk_buff *skb) |
| 786 | { |
| 787 | return TCP_SKB_CB(skb)->tcp_gso_segs; |
| 788 | } |
| 789 | |
| 790 | static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) |
| 791 | { |
| 792 | TCP_SKB_CB(skb)->tcp_gso_segs = segs; |
| 793 | } |
| 794 | |
| 795 | static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) |
| 796 | { |
| 797 | TCP_SKB_CB(skb)->tcp_gso_segs += segs; |
| 798 | } |
| 799 | |
| 800 | /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ |
| 801 | static inline int tcp_skb_mss(const struct sk_buff *skb) |
| 802 | { |
| 803 | return TCP_SKB_CB(skb)->tcp_gso_size; |
| 804 | } |
| 805 | |
| 806 | /* Events passed to congestion control interface */ |
| 807 | enum tcp_ca_event { |
| 808 | CA_EVENT_TX_START, /* first transmit when no packets in flight */ |
| 809 | CA_EVENT_CWND_RESTART, /* congestion window restart */ |
| 810 | CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ |
| 811 | CA_EVENT_LOSS, /* loss timeout */ |
| 812 | CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ |
| 813 | CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ |
| 814 | CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ |
| 815 | CA_EVENT_NON_DELAYED_ACK, |
| 816 | }; |
| 817 | |
| 818 | /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ |
| 819 | enum tcp_ca_ack_event_flags { |
| 820 | CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ |
| 821 | CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ |
| 822 | CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ |
| 823 | }; |
| 824 | |
| 825 | /* |
| 826 | * Interface for adding new TCP congestion control handlers |
| 827 | */ |
| 828 | #define TCP_CA_NAME_MAX 16 |
| 829 | #define TCP_CA_MAX 128 |
| 830 | #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) |
| 831 | |
| 832 | #define TCP_CA_UNSPEC 0 |
| 833 | |
| 834 | /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ |
| 835 | #define TCP_CONG_NON_RESTRICTED 0x1 |
| 836 | /* Requires ECN/ECT set on all packets */ |
| 837 | #define TCP_CONG_NEEDS_ECN 0x2 |
| 838 | |
| 839 | union tcp_cc_info; |
| 840 | |
| 841 | struct tcp_congestion_ops { |
| 842 | struct list_head list; |
| 843 | u32 key; |
| 844 | u32 flags; |
| 845 | |
| 846 | /* initialize private data (optional) */ |
| 847 | void (*init)(struct sock *sk); |
| 848 | /* cleanup private data (optional) */ |
| 849 | void (*release)(struct sock *sk); |
| 850 | |
| 851 | /* return slow start threshold (required) */ |
| 852 | u32 (*ssthresh)(struct sock *sk); |
| 853 | /* do new cwnd calculation (required) */ |
| 854 | void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); |
| 855 | /* call before changing ca_state (optional) */ |
| 856 | void (*set_state)(struct sock *sk, u8 new_state); |
| 857 | /* call when cwnd event occurs (optional) */ |
| 858 | void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); |
| 859 | /* call when ack arrives (optional) */ |
| 860 | void (*in_ack_event)(struct sock *sk, u32 flags); |
| 861 | /* new value of cwnd after loss (optional) */ |
| 862 | u32 (*undo_cwnd)(struct sock *sk); |
| 863 | /* hook for packet ack accounting (optional) */ |
| 864 | void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); |
| 865 | /* get info for inet_diag (optional) */ |
| 866 | size_t (*get_info)(struct sock *sk, u32 ext, int *attr, |
| 867 | union tcp_cc_info *info); |
| 868 | |
| 869 | char name[TCP_CA_NAME_MAX]; |
| 870 | struct module *owner; |
| 871 | }; |
| 872 | |
| 873 | int tcp_register_congestion_control(struct tcp_congestion_ops *type); |
| 874 | void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); |
| 875 | |
| 876 | void tcp_assign_congestion_control(struct sock *sk); |
| 877 | void tcp_init_congestion_control(struct sock *sk); |
| 878 | void tcp_cleanup_congestion_control(struct sock *sk); |
| 879 | int tcp_set_default_congestion_control(const char *name); |
| 880 | void tcp_get_default_congestion_control(char *name); |
| 881 | void tcp_get_available_congestion_control(char *buf, size_t len); |
| 882 | void tcp_get_allowed_congestion_control(char *buf, size_t len); |
| 883 | int tcp_set_allowed_congestion_control(char *allowed); |
| 884 | int tcp_set_congestion_control(struct sock *sk, const char *name); |
| 885 | u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); |
| 886 | void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); |
| 887 | |
| 888 | u32 tcp_reno_ssthresh(struct sock *sk); |
| 889 | void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); |
| 890 | extern struct tcp_congestion_ops tcp_reno; |
| 891 | |
| 892 | struct tcp_congestion_ops *tcp_ca_find_key(u32 key); |
| 893 | u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); |
| 894 | #ifdef CONFIG_INET |
| 895 | char *tcp_ca_get_name_by_key(u32 key, char *buffer); |
| 896 | #else |
| 897 | static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) |
| 898 | { |
| 899 | return NULL; |
| 900 | } |
| 901 | #endif |
| 902 | |
| 903 | static inline bool tcp_ca_needs_ecn(const struct sock *sk) |
| 904 | { |
| 905 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 906 | |
| 907 | return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; |
| 908 | } |
| 909 | |
| 910 | static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) |
| 911 | { |
| 912 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 913 | |
| 914 | if (icsk->icsk_ca_ops->set_state) |
| 915 | icsk->icsk_ca_ops->set_state(sk, ca_state); |
| 916 | icsk->icsk_ca_state = ca_state; |
| 917 | } |
| 918 | |
| 919 | static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) |
| 920 | { |
| 921 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 922 | |
| 923 | if (icsk->icsk_ca_ops->cwnd_event) |
| 924 | icsk->icsk_ca_ops->cwnd_event(sk, event); |
| 925 | } |
| 926 | |
| 927 | /* These functions determine how the current flow behaves in respect of SACK |
| 928 | * handling. SACK is negotiated with the peer, and therefore it can vary |
| 929 | * between different flows. |
| 930 | * |
| 931 | * tcp_is_sack - SACK enabled |
| 932 | * tcp_is_reno - No SACK |
| 933 | * tcp_is_fack - FACK enabled, implies SACK enabled |
| 934 | */ |
| 935 | static inline int tcp_is_sack(const struct tcp_sock *tp) |
| 936 | { |
| 937 | return tp->rx_opt.sack_ok; |
| 938 | } |
| 939 | |
| 940 | static inline bool tcp_is_reno(const struct tcp_sock *tp) |
| 941 | { |
| 942 | return !tcp_is_sack(tp); |
| 943 | } |
| 944 | |
| 945 | static inline bool tcp_is_fack(const struct tcp_sock *tp) |
| 946 | { |
| 947 | return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; |
| 948 | } |
| 949 | |
| 950 | static inline void tcp_enable_fack(struct tcp_sock *tp) |
| 951 | { |
| 952 | tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; |
| 953 | } |
| 954 | |
| 955 | /* TCP early-retransmit (ER) is similar to but more conservative than |
| 956 | * the thin-dupack feature. Enable ER only if thin-dupack is disabled. |
| 957 | */ |
| 958 | static inline void tcp_enable_early_retrans(struct tcp_sock *tp) |
| 959 | { |
| 960 | struct net *net = sock_net((struct sock *)tp); |
| 961 | |
| 962 | tp->do_early_retrans = sysctl_tcp_early_retrans && |
| 963 | sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && |
| 964 | net->ipv4.sysctl_tcp_reordering == 3; |
| 965 | } |
| 966 | |
| 967 | static inline void tcp_disable_early_retrans(struct tcp_sock *tp) |
| 968 | { |
| 969 | tp->do_early_retrans = 0; |
| 970 | } |
| 971 | |
| 972 | static inline unsigned int tcp_left_out(const struct tcp_sock *tp) |
| 973 | { |
| 974 | return tp->sacked_out + tp->lost_out; |
| 975 | } |
| 976 | |
| 977 | /* This determines how many packets are "in the network" to the best |
| 978 | * of our knowledge. In many cases it is conservative, but where |
| 979 | * detailed information is available from the receiver (via SACK |
| 980 | * blocks etc.) we can make more aggressive calculations. |
| 981 | * |
| 982 | * Use this for decisions involving congestion control, use just |
| 983 | * tp->packets_out to determine if the send queue is empty or not. |
| 984 | * |
| 985 | * Read this equation as: |
| 986 | * |
| 987 | * "Packets sent once on transmission queue" MINUS |
| 988 | * "Packets left network, but not honestly ACKed yet" PLUS |
| 989 | * "Packets fast retransmitted" |
| 990 | */ |
| 991 | static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) |
| 992 | { |
| 993 | return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; |
| 994 | } |
| 995 | |
| 996 | #define TCP_INFINITE_SSTHRESH 0x7fffffff |
| 997 | |
| 998 | static inline bool tcp_in_slow_start(const struct tcp_sock *tp) |
| 999 | { |
| 1000 | return tp->snd_cwnd < tp->snd_ssthresh; |
| 1001 | } |
| 1002 | |
| 1003 | static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) |
| 1004 | { |
| 1005 | return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; |
| 1006 | } |
| 1007 | |
| 1008 | static inline bool tcp_in_cwnd_reduction(const struct sock *sk) |
| 1009 | { |
| 1010 | return (TCPF_CA_CWR | TCPF_CA_Recovery) & |
| 1011 | (1 << inet_csk(sk)->icsk_ca_state); |
| 1012 | } |
| 1013 | |
| 1014 | /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. |
| 1015 | * The exception is cwnd reduction phase, when cwnd is decreasing towards |
| 1016 | * ssthresh. |
| 1017 | */ |
| 1018 | static inline __u32 tcp_current_ssthresh(const struct sock *sk) |
| 1019 | { |
| 1020 | const struct tcp_sock *tp = tcp_sk(sk); |
| 1021 | |
| 1022 | if (tcp_in_cwnd_reduction(sk)) |
| 1023 | return tp->snd_ssthresh; |
| 1024 | else |
| 1025 | return max(tp->snd_ssthresh, |
| 1026 | ((tp->snd_cwnd >> 1) + |
| 1027 | (tp->snd_cwnd >> 2))); |
| 1028 | } |
| 1029 | |
| 1030 | /* Use define here intentionally to get WARN_ON location shown at the caller */ |
| 1031 | #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) |
| 1032 | |
| 1033 | void tcp_enter_cwr(struct sock *sk); |
| 1034 | __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); |
| 1035 | |
| 1036 | /* The maximum number of MSS of available cwnd for which TSO defers |
| 1037 | * sending if not using sysctl_tcp_tso_win_divisor. |
| 1038 | */ |
| 1039 | static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) |
| 1040 | { |
| 1041 | return 3; |
| 1042 | } |
| 1043 | |
| 1044 | /* Slow start with delack produces 3 packets of burst, so that |
| 1045 | * it is safe "de facto". This will be the default - same as |
| 1046 | * the default reordering threshold - but if reordering increases, |
| 1047 | * we must be able to allow cwnd to burst at least this much in order |
| 1048 | * to not pull it back when holes are filled. |
| 1049 | */ |
| 1050 | static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) |
| 1051 | { |
| 1052 | return tp->reordering; |
| 1053 | } |
| 1054 | |
| 1055 | /* Returns end sequence number of the receiver's advertised window */ |
| 1056 | static inline u32 tcp_wnd_end(const struct tcp_sock *tp) |
| 1057 | { |
| 1058 | return tp->snd_una + tp->snd_wnd; |
| 1059 | } |
| 1060 | |
| 1061 | /* We follow the spirit of RFC2861 to validate cwnd but implement a more |
| 1062 | * flexible approach. The RFC suggests cwnd should not be raised unless |
| 1063 | * it was fully used previously. And that's exactly what we do in |
| 1064 | * congestion avoidance mode. But in slow start we allow cwnd to grow |
| 1065 | * as long as the application has used half the cwnd. |
| 1066 | * Example : |
| 1067 | * cwnd is 10 (IW10), but application sends 9 frames. |
| 1068 | * We allow cwnd to reach 18 when all frames are ACKed. |
| 1069 | * This check is safe because it's as aggressive as slow start which already |
| 1070 | * risks 100% overshoot. The advantage is that we discourage application to |
| 1071 | * either send more filler packets or data to artificially blow up the cwnd |
| 1072 | * usage, and allow application-limited process to probe bw more aggressively. |
| 1073 | */ |
| 1074 | static inline bool tcp_is_cwnd_limited(const struct sock *sk) |
| 1075 | { |
| 1076 | const struct tcp_sock *tp = tcp_sk(sk); |
| 1077 | |
| 1078 | /* If in slow start, ensure cwnd grows to twice what was ACKed. */ |
| 1079 | if (tcp_in_slow_start(tp)) |
| 1080 | return tp->snd_cwnd < 2 * tp->max_packets_out; |
| 1081 | |
| 1082 | return tp->is_cwnd_limited; |
| 1083 | } |
| 1084 | |
| 1085 | /* Something is really bad, we could not queue an additional packet, |
| 1086 | * because qdisc is full or receiver sent a 0 window. |
| 1087 | * We do not want to add fuel to the fire, or abort too early, |
| 1088 | * so make sure the timer we arm now is at least 200ms in the future, |
| 1089 | * regardless of current icsk_rto value (as it could be ~2ms) |
| 1090 | */ |
| 1091 | static inline unsigned long tcp_probe0_base(const struct sock *sk) |
| 1092 | { |
| 1093 | return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); |
| 1094 | } |
| 1095 | |
| 1096 | /* Variant of inet_csk_rto_backoff() used for zero window probes */ |
| 1097 | static inline unsigned long tcp_probe0_when(const struct sock *sk, |
| 1098 | unsigned long max_when) |
| 1099 | { |
| 1100 | u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; |
| 1101 | |
| 1102 | return (unsigned long)min_t(u64, when, max_when); |
| 1103 | } |
| 1104 | |
| 1105 | static inline void tcp_check_probe_timer(struct sock *sk) |
| 1106 | { |
| 1107 | if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) |
| 1108 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, |
| 1109 | tcp_probe0_base(sk), TCP_RTO_MAX); |
| 1110 | } |
| 1111 | |
| 1112 | static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) |
| 1113 | { |
| 1114 | tp->snd_wl1 = seq; |
| 1115 | } |
| 1116 | |
| 1117 | static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) |
| 1118 | { |
| 1119 | tp->snd_wl1 = seq; |
| 1120 | } |
| 1121 | |
| 1122 | /* |
| 1123 | * Calculate(/check) TCP checksum |
| 1124 | */ |
| 1125 | static inline __sum16 tcp_v4_check(int len, __be32 saddr, |
| 1126 | __be32 daddr, __wsum base) |
| 1127 | { |
| 1128 | return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); |
| 1129 | } |
| 1130 | |
| 1131 | static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) |
| 1132 | { |
| 1133 | return __skb_checksum_complete(skb); |
| 1134 | } |
| 1135 | |
| 1136 | static inline bool tcp_checksum_complete(struct sk_buff *skb) |
| 1137 | { |
| 1138 | return !skb_csum_unnecessary(skb) && |
| 1139 | __tcp_checksum_complete(skb); |
| 1140 | } |
| 1141 | |
| 1142 | /* Prequeue for VJ style copy to user, combined with checksumming. */ |
| 1143 | |
| 1144 | static inline void tcp_prequeue_init(struct tcp_sock *tp) |
| 1145 | { |
| 1146 | tp->ucopy.task = NULL; |
| 1147 | tp->ucopy.len = 0; |
| 1148 | tp->ucopy.memory = 0; |
| 1149 | skb_queue_head_init(&tp->ucopy.prequeue); |
| 1150 | } |
| 1151 | |
| 1152 | bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); |
| 1153 | |
| 1154 | #undef STATE_TRACE |
| 1155 | |
| 1156 | #ifdef STATE_TRACE |
| 1157 | static const char *statename[]={ |
| 1158 | "Unused","Established","Syn Sent","Syn Recv", |
| 1159 | "Fin Wait 1","Fin Wait 2","Time Wait", "Close", |
| 1160 | "Close Wait","Last ACK","Listen","Closing" |
| 1161 | }; |
| 1162 | #endif |
| 1163 | void tcp_set_state(struct sock *sk, int state); |
| 1164 | |
| 1165 | void tcp_done(struct sock *sk); |
| 1166 | |
| 1167 | int tcp_abort(struct sock *sk, int err); |
| 1168 | |
| 1169 | static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) |
| 1170 | { |
| 1171 | rx_opt->dsack = 0; |
| 1172 | rx_opt->num_sacks = 0; |
| 1173 | } |
| 1174 | |
| 1175 | u32 tcp_default_init_rwnd(u32 mss); |
| 1176 | void tcp_cwnd_restart(struct sock *sk, s32 delta); |
| 1177 | |
| 1178 | static inline void tcp_slow_start_after_idle_check(struct sock *sk) |
| 1179 | { |
| 1180 | struct tcp_sock *tp = tcp_sk(sk); |
| 1181 | s32 delta; |
| 1182 | |
| 1183 | if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) |
| 1184 | return; |
| 1185 | delta = tcp_time_stamp - tp->lsndtime; |
| 1186 | if (delta > inet_csk(sk)->icsk_rto) |
| 1187 | tcp_cwnd_restart(sk, delta); |
| 1188 | } |
| 1189 | |
| 1190 | /* Determine a window scaling and initial window to offer. */ |
| 1191 | void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, |
| 1192 | __u32 *window_clamp, int wscale_ok, |
| 1193 | __u8 *rcv_wscale, __u32 init_rcv_wnd); |
| 1194 | |
| 1195 | static inline int tcp_win_from_space(int space) |
| 1196 | { |
| 1197 | return sysctl_tcp_adv_win_scale<=0 ? |
| 1198 | (space>>(-sysctl_tcp_adv_win_scale)) : |
| 1199 | space - (space>>sysctl_tcp_adv_win_scale); |
| 1200 | } |
| 1201 | |
| 1202 | /* Note: caller must be prepared to deal with negative returns */ |
| 1203 | static inline int tcp_space(const struct sock *sk) |
| 1204 | { |
| 1205 | return tcp_win_from_space(sk->sk_rcvbuf - |
| 1206 | atomic_read(&sk->sk_rmem_alloc)); |
| 1207 | } |
| 1208 | |
| 1209 | static inline int tcp_full_space(const struct sock *sk) |
| 1210 | { |
| 1211 | return tcp_win_from_space(sk->sk_rcvbuf); |
| 1212 | } |
| 1213 | |
| 1214 | extern void tcp_openreq_init_rwin(struct request_sock *req, |
| 1215 | const struct sock *sk_listener, |
| 1216 | const struct dst_entry *dst); |
| 1217 | |
| 1218 | void tcp_enter_memory_pressure(struct sock *sk); |
| 1219 | |
| 1220 | static inline int keepalive_intvl_when(const struct tcp_sock *tp) |
| 1221 | { |
| 1222 | struct net *net = sock_net((struct sock *)tp); |
| 1223 | |
| 1224 | return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; |
| 1225 | } |
| 1226 | |
| 1227 | static inline int keepalive_time_when(const struct tcp_sock *tp) |
| 1228 | { |
| 1229 | struct net *net = sock_net((struct sock *)tp); |
| 1230 | |
| 1231 | return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; |
| 1232 | } |
| 1233 | |
| 1234 | static inline int keepalive_probes(const struct tcp_sock *tp) |
| 1235 | { |
| 1236 | struct net *net = sock_net((struct sock *)tp); |
| 1237 | |
| 1238 | return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; |
| 1239 | } |
| 1240 | |
| 1241 | static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) |
| 1242 | { |
| 1243 | const struct inet_connection_sock *icsk = &tp->inet_conn; |
| 1244 | |
| 1245 | return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, |
| 1246 | tcp_time_stamp - tp->rcv_tstamp); |
| 1247 | } |
| 1248 | |
| 1249 | static inline int tcp_fin_time(const struct sock *sk) |
| 1250 | { |
| 1251 | int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; |
| 1252 | const int rto = inet_csk(sk)->icsk_rto; |
| 1253 | |
| 1254 | if (fin_timeout < (rto << 2) - (rto >> 1)) |
| 1255 | fin_timeout = (rto << 2) - (rto >> 1); |
| 1256 | |
| 1257 | return fin_timeout; |
| 1258 | } |
| 1259 | |
| 1260 | static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, |
| 1261 | int paws_win) |
| 1262 | { |
| 1263 | if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) |
| 1264 | return true; |
| 1265 | if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) |
| 1266 | return true; |
| 1267 | /* |
| 1268 | * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, |
| 1269 | * then following tcp messages have valid values. Ignore 0 value, |
| 1270 | * or else 'negative' tsval might forbid us to accept their packets. |
| 1271 | */ |
| 1272 | if (!rx_opt->ts_recent) |
| 1273 | return true; |
| 1274 | return false; |
| 1275 | } |
| 1276 | |
| 1277 | static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, |
| 1278 | int rst) |
| 1279 | { |
| 1280 | if (tcp_paws_check(rx_opt, 0)) |
| 1281 | return false; |
| 1282 | |
| 1283 | /* RST segments are not recommended to carry timestamp, |
| 1284 | and, if they do, it is recommended to ignore PAWS because |
| 1285 | "their cleanup function should take precedence over timestamps." |
| 1286 | Certainly, it is mistake. It is necessary to understand the reasons |
| 1287 | of this constraint to relax it: if peer reboots, clock may go |
| 1288 | out-of-sync and half-open connections will not be reset. |
| 1289 | Actually, the problem would be not existing if all |
| 1290 | the implementations followed draft about maintaining clock |
| 1291 | via reboots. Linux-2.2 DOES NOT! |
| 1292 | |
| 1293 | However, we can relax time bounds for RST segments to MSL. |
| 1294 | */ |
| 1295 | if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) |
| 1296 | return false; |
| 1297 | return true; |
| 1298 | } |
| 1299 | |
| 1300 | bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, |
| 1301 | int mib_idx, u32 *last_oow_ack_time); |
| 1302 | |
| 1303 | static inline void tcp_mib_init(struct net *net) |
| 1304 | { |
| 1305 | /* See RFC 2012 */ |
| 1306 | TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); |
| 1307 | TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); |
| 1308 | TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); |
| 1309 | TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); |
| 1310 | } |
| 1311 | |
| 1312 | /* from STCP */ |
| 1313 | static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) |
| 1314 | { |
| 1315 | tp->lost_skb_hint = NULL; |
| 1316 | } |
| 1317 | |
| 1318 | static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) |
| 1319 | { |
| 1320 | tcp_clear_retrans_hints_partial(tp); |
| 1321 | tp->retransmit_skb_hint = NULL; |
| 1322 | } |
| 1323 | |
| 1324 | union tcp_md5_addr { |
| 1325 | struct in_addr a4; |
| 1326 | #if IS_ENABLED(CONFIG_IPV6) |
| 1327 | struct in6_addr a6; |
| 1328 | #endif |
| 1329 | }; |
| 1330 | |
| 1331 | /* - key database */ |
| 1332 | struct tcp_md5sig_key { |
| 1333 | struct hlist_node node; |
| 1334 | u8 keylen; |
| 1335 | u8 family; /* AF_INET or AF_INET6 */ |
| 1336 | union tcp_md5_addr addr; |
| 1337 | u8 key[TCP_MD5SIG_MAXKEYLEN]; |
| 1338 | struct rcu_head rcu; |
| 1339 | }; |
| 1340 | |
| 1341 | /* - sock block */ |
| 1342 | struct tcp_md5sig_info { |
| 1343 | struct hlist_head head; |
| 1344 | struct rcu_head rcu; |
| 1345 | }; |
| 1346 | |
| 1347 | /* - pseudo header */ |
| 1348 | struct tcp4_pseudohdr { |
| 1349 | __be32 saddr; |
| 1350 | __be32 daddr; |
| 1351 | __u8 pad; |
| 1352 | __u8 protocol; |
| 1353 | __be16 len; |
| 1354 | }; |
| 1355 | |
| 1356 | struct tcp6_pseudohdr { |
| 1357 | struct in6_addr saddr; |
| 1358 | struct in6_addr daddr; |
| 1359 | __be32 len; |
| 1360 | __be32 protocol; /* including padding */ |
| 1361 | }; |
| 1362 | |
| 1363 | union tcp_md5sum_block { |
| 1364 | struct tcp4_pseudohdr ip4; |
| 1365 | #if IS_ENABLED(CONFIG_IPV6) |
| 1366 | struct tcp6_pseudohdr ip6; |
| 1367 | #endif |
| 1368 | }; |
| 1369 | |
| 1370 | /* - pool: digest algorithm, hash description and scratch buffer */ |
| 1371 | struct tcp_md5sig_pool { |
| 1372 | struct ahash_request *md5_req; |
| 1373 | union tcp_md5sum_block md5_blk; |
| 1374 | }; |
| 1375 | |
| 1376 | /* - functions */ |
| 1377 | int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, |
| 1378 | const struct sock *sk, const struct sk_buff *skb); |
| 1379 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, |
| 1380 | int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); |
| 1381 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, |
| 1382 | int family); |
| 1383 | struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, |
| 1384 | const struct sock *addr_sk); |
| 1385 | |
| 1386 | #ifdef CONFIG_TCP_MD5SIG |
| 1387 | struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, |
| 1388 | const union tcp_md5_addr *addr, |
| 1389 | int family); |
| 1390 | #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) |
| 1391 | #else |
| 1392 | static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, |
| 1393 | const union tcp_md5_addr *addr, |
| 1394 | int family) |
| 1395 | { |
| 1396 | return NULL; |
| 1397 | } |
| 1398 | #define tcp_twsk_md5_key(twsk) NULL |
| 1399 | #endif |
| 1400 | |
| 1401 | bool tcp_alloc_md5sig_pool(void); |
| 1402 | |
| 1403 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); |
| 1404 | static inline void tcp_put_md5sig_pool(void) |
| 1405 | { |
| 1406 | local_bh_enable(); |
| 1407 | } |
| 1408 | |
| 1409 | int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); |
| 1410 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, |
| 1411 | unsigned int header_len); |
| 1412 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, |
| 1413 | const struct tcp_md5sig_key *key); |
| 1414 | |
| 1415 | /* From tcp_fastopen.c */ |
| 1416 | void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, |
| 1417 | struct tcp_fastopen_cookie *cookie, int *syn_loss, |
| 1418 | unsigned long *last_syn_loss); |
| 1419 | void tcp_fastopen_cache_set(struct sock *sk, u16 mss, |
| 1420 | struct tcp_fastopen_cookie *cookie, bool syn_lost, |
| 1421 | u16 try_exp); |
| 1422 | struct tcp_fastopen_request { |
| 1423 | /* Fast Open cookie. Size 0 means a cookie request */ |
| 1424 | struct tcp_fastopen_cookie cookie; |
| 1425 | struct msghdr *data; /* data in MSG_FASTOPEN */ |
| 1426 | size_t size; |
| 1427 | int copied; /* queued in tcp_connect() */ |
| 1428 | }; |
| 1429 | void tcp_free_fastopen_req(struct tcp_sock *tp); |
| 1430 | |
| 1431 | extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; |
| 1432 | int tcp_fastopen_reset_cipher(void *key, unsigned int len); |
| 1433 | void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); |
| 1434 | struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, |
| 1435 | struct request_sock *req, |
| 1436 | struct tcp_fastopen_cookie *foc, |
| 1437 | struct dst_entry *dst); |
| 1438 | void tcp_fastopen_init_key_once(bool publish); |
| 1439 | #define TCP_FASTOPEN_KEY_LENGTH 16 |
| 1440 | |
| 1441 | /* Fastopen key context */ |
| 1442 | struct tcp_fastopen_context { |
| 1443 | struct crypto_cipher *tfm; |
| 1444 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; |
| 1445 | struct rcu_head rcu; |
| 1446 | }; |
| 1447 | |
| 1448 | /* write queue abstraction */ |
| 1449 | static inline void tcp_write_queue_purge(struct sock *sk) |
| 1450 | { |
| 1451 | struct sk_buff *skb; |
| 1452 | |
| 1453 | while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) |
| 1454 | sk_wmem_free_skb(sk, skb); |
| 1455 | sk_mem_reclaim(sk); |
| 1456 | tcp_clear_all_retrans_hints(tcp_sk(sk)); |
| 1457 | } |
| 1458 | |
| 1459 | static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) |
| 1460 | { |
| 1461 | return skb_peek(&sk->sk_write_queue); |
| 1462 | } |
| 1463 | |
| 1464 | static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) |
| 1465 | { |
| 1466 | return skb_peek_tail(&sk->sk_write_queue); |
| 1467 | } |
| 1468 | |
| 1469 | static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, |
| 1470 | const struct sk_buff *skb) |
| 1471 | { |
| 1472 | return skb_queue_next(&sk->sk_write_queue, skb); |
| 1473 | } |
| 1474 | |
| 1475 | static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, |
| 1476 | const struct sk_buff *skb) |
| 1477 | { |
| 1478 | return skb_queue_prev(&sk->sk_write_queue, skb); |
| 1479 | } |
| 1480 | |
| 1481 | #define tcp_for_write_queue(skb, sk) \ |
| 1482 | skb_queue_walk(&(sk)->sk_write_queue, skb) |
| 1483 | |
| 1484 | #define tcp_for_write_queue_from(skb, sk) \ |
| 1485 | skb_queue_walk_from(&(sk)->sk_write_queue, skb) |
| 1486 | |
| 1487 | #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ |
| 1488 | skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) |
| 1489 | |
| 1490 | static inline struct sk_buff *tcp_send_head(const struct sock *sk) |
| 1491 | { |
| 1492 | return sk->sk_send_head; |
| 1493 | } |
| 1494 | |
| 1495 | static inline bool tcp_skb_is_last(const struct sock *sk, |
| 1496 | const struct sk_buff *skb) |
| 1497 | { |
| 1498 | return skb_queue_is_last(&sk->sk_write_queue, skb); |
| 1499 | } |
| 1500 | |
| 1501 | static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) |
| 1502 | { |
| 1503 | if (tcp_skb_is_last(sk, skb)) |
| 1504 | sk->sk_send_head = NULL; |
| 1505 | else |
| 1506 | sk->sk_send_head = tcp_write_queue_next(sk, skb); |
| 1507 | } |
| 1508 | |
| 1509 | static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) |
| 1510 | { |
| 1511 | if (sk->sk_send_head == skb_unlinked) |
| 1512 | sk->sk_send_head = NULL; |
| 1513 | } |
| 1514 | |
| 1515 | static inline void tcp_init_send_head(struct sock *sk) |
| 1516 | { |
| 1517 | sk->sk_send_head = NULL; |
| 1518 | } |
| 1519 | |
| 1520 | static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) |
| 1521 | { |
| 1522 | __skb_queue_tail(&sk->sk_write_queue, skb); |
| 1523 | } |
| 1524 | |
| 1525 | static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) |
| 1526 | { |
| 1527 | __tcp_add_write_queue_tail(sk, skb); |
| 1528 | |
| 1529 | /* Queue it, remembering where we must start sending. */ |
| 1530 | if (sk->sk_send_head == NULL) { |
| 1531 | sk->sk_send_head = skb; |
| 1532 | |
| 1533 | if (tcp_sk(sk)->highest_sack == NULL) |
| 1534 | tcp_sk(sk)->highest_sack = skb; |
| 1535 | } |
| 1536 | } |
| 1537 | |
| 1538 | static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) |
| 1539 | { |
| 1540 | __skb_queue_head(&sk->sk_write_queue, skb); |
| 1541 | } |
| 1542 | |
| 1543 | /* Insert buff after skb on the write queue of sk. */ |
| 1544 | static inline void tcp_insert_write_queue_after(struct sk_buff *skb, |
| 1545 | struct sk_buff *buff, |
| 1546 | struct sock *sk) |
| 1547 | { |
| 1548 | __skb_queue_after(&sk->sk_write_queue, skb, buff); |
| 1549 | } |
| 1550 | |
| 1551 | /* Insert new before skb on the write queue of sk. */ |
| 1552 | static inline void tcp_insert_write_queue_before(struct sk_buff *new, |
| 1553 | struct sk_buff *skb, |
| 1554 | struct sock *sk) |
| 1555 | { |
| 1556 | __skb_queue_before(&sk->sk_write_queue, skb, new); |
| 1557 | |
| 1558 | if (sk->sk_send_head == skb) |
| 1559 | sk->sk_send_head = new; |
| 1560 | } |
| 1561 | |
| 1562 | static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) |
| 1563 | { |
| 1564 | __skb_unlink(skb, &sk->sk_write_queue); |
| 1565 | } |
| 1566 | |
| 1567 | static inline bool tcp_write_queue_empty(struct sock *sk) |
| 1568 | { |
| 1569 | return skb_queue_empty(&sk->sk_write_queue); |
| 1570 | } |
| 1571 | |
| 1572 | static inline void tcp_push_pending_frames(struct sock *sk) |
| 1573 | { |
| 1574 | if (tcp_send_head(sk)) { |
| 1575 | struct tcp_sock *tp = tcp_sk(sk); |
| 1576 | |
| 1577 | __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); |
| 1578 | } |
| 1579 | } |
| 1580 | |
| 1581 | /* Start sequence of the skb just after the highest skb with SACKed |
| 1582 | * bit, valid only if sacked_out > 0 or when the caller has ensured |
| 1583 | * validity by itself. |
| 1584 | */ |
| 1585 | static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) |
| 1586 | { |
| 1587 | if (!tp->sacked_out) |
| 1588 | return tp->snd_una; |
| 1589 | |
| 1590 | if (tp->highest_sack == NULL) |
| 1591 | return tp->snd_nxt; |
| 1592 | |
| 1593 | return TCP_SKB_CB(tp->highest_sack)->seq; |
| 1594 | } |
| 1595 | |
| 1596 | static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) |
| 1597 | { |
| 1598 | tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : |
| 1599 | tcp_write_queue_next(sk, skb); |
| 1600 | } |
| 1601 | |
| 1602 | static inline struct sk_buff *tcp_highest_sack(struct sock *sk) |
| 1603 | { |
| 1604 | return tcp_sk(sk)->highest_sack; |
| 1605 | } |
| 1606 | |
| 1607 | static inline void tcp_highest_sack_reset(struct sock *sk) |
| 1608 | { |
| 1609 | tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); |
| 1610 | } |
| 1611 | |
| 1612 | /* Called when old skb is about to be deleted (to be combined with new skb) */ |
| 1613 | static inline void tcp_highest_sack_combine(struct sock *sk, |
| 1614 | struct sk_buff *old, |
| 1615 | struct sk_buff *new) |
| 1616 | { |
| 1617 | if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) |
| 1618 | tcp_sk(sk)->highest_sack = new; |
| 1619 | } |
| 1620 | |
| 1621 | /* This helper checks if socket has IP_TRANSPARENT set */ |
| 1622 | static inline bool inet_sk_transparent(const struct sock *sk) |
| 1623 | { |
| 1624 | switch (sk->sk_state) { |
| 1625 | case TCP_TIME_WAIT: |
| 1626 | return inet_twsk(sk)->tw_transparent; |
| 1627 | case TCP_NEW_SYN_RECV: |
| 1628 | return inet_rsk(inet_reqsk(sk))->no_srccheck; |
| 1629 | } |
| 1630 | return inet_sk(sk)->transparent; |
| 1631 | } |
| 1632 | |
| 1633 | /* Determines whether this is a thin stream (which may suffer from |
| 1634 | * increased latency). Used to trigger latency-reducing mechanisms. |
| 1635 | */ |
| 1636 | static inline bool tcp_stream_is_thin(struct tcp_sock *tp) |
| 1637 | { |
| 1638 | return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); |
| 1639 | } |
| 1640 | |
| 1641 | /* /proc */ |
| 1642 | enum tcp_seq_states { |
| 1643 | TCP_SEQ_STATE_LISTENING, |
| 1644 | TCP_SEQ_STATE_ESTABLISHED, |
| 1645 | }; |
| 1646 | |
| 1647 | int tcp_seq_open(struct inode *inode, struct file *file); |
| 1648 | |
| 1649 | struct tcp_seq_afinfo { |
| 1650 | char *name; |
| 1651 | sa_family_t family; |
| 1652 | const struct file_operations *seq_fops; |
| 1653 | struct seq_operations seq_ops; |
| 1654 | }; |
| 1655 | |
| 1656 | struct tcp_iter_state { |
| 1657 | struct seq_net_private p; |
| 1658 | sa_family_t family; |
| 1659 | enum tcp_seq_states state; |
| 1660 | struct sock *syn_wait_sk; |
| 1661 | int bucket, offset, sbucket, num; |
| 1662 | loff_t last_pos; |
| 1663 | }; |
| 1664 | |
| 1665 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); |
| 1666 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); |
| 1667 | |
| 1668 | extern struct request_sock_ops tcp_request_sock_ops; |
| 1669 | extern struct request_sock_ops tcp6_request_sock_ops; |
| 1670 | |
| 1671 | void tcp_v4_destroy_sock(struct sock *sk); |
| 1672 | |
| 1673 | struct sk_buff *tcp_gso_segment(struct sk_buff *skb, |
| 1674 | netdev_features_t features); |
| 1675 | struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); |
| 1676 | int tcp_gro_complete(struct sk_buff *skb); |
| 1677 | |
| 1678 | void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); |
| 1679 | |
| 1680 | static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) |
| 1681 | { |
| 1682 | struct net *net = sock_net((struct sock *)tp); |
| 1683 | return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; |
| 1684 | } |
| 1685 | |
| 1686 | static inline bool tcp_stream_memory_free(const struct sock *sk) |
| 1687 | { |
| 1688 | const struct tcp_sock *tp = tcp_sk(sk); |
| 1689 | u32 notsent_bytes = tp->write_seq - tp->snd_nxt; |
| 1690 | |
| 1691 | return notsent_bytes < tcp_notsent_lowat(tp); |
| 1692 | } |
| 1693 | |
| 1694 | #ifdef CONFIG_PROC_FS |
| 1695 | int tcp4_proc_init(void); |
| 1696 | void tcp4_proc_exit(void); |
| 1697 | #endif |
| 1698 | |
| 1699 | int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); |
| 1700 | int tcp_conn_request(struct request_sock_ops *rsk_ops, |
| 1701 | const struct tcp_request_sock_ops *af_ops, |
| 1702 | struct sock *sk, struct sk_buff *skb); |
| 1703 | |
| 1704 | /* TCP af-specific functions */ |
| 1705 | struct tcp_sock_af_ops { |
| 1706 | #ifdef CONFIG_TCP_MD5SIG |
| 1707 | struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, |
| 1708 | const struct sock *addr_sk); |
| 1709 | int (*calc_md5_hash)(char *location, |
| 1710 | const struct tcp_md5sig_key *md5, |
| 1711 | const struct sock *sk, |
| 1712 | const struct sk_buff *skb); |
| 1713 | int (*md5_parse)(struct sock *sk, |
| 1714 | char __user *optval, |
| 1715 | int optlen); |
| 1716 | #endif |
| 1717 | }; |
| 1718 | |
| 1719 | struct tcp_request_sock_ops { |
| 1720 | u16 mss_clamp; |
| 1721 | #ifdef CONFIG_TCP_MD5SIG |
| 1722 | struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, |
| 1723 | const struct sock *addr_sk); |
| 1724 | int (*calc_md5_hash) (char *location, |
| 1725 | const struct tcp_md5sig_key *md5, |
| 1726 | const struct sock *sk, |
| 1727 | const struct sk_buff *skb); |
| 1728 | #endif |
| 1729 | void (*init_req)(struct request_sock *req, |
| 1730 | const struct sock *sk_listener, |
| 1731 | struct sk_buff *skb); |
| 1732 | #ifdef CONFIG_SYN_COOKIES |
| 1733 | __u32 (*cookie_init_seq)(const struct sk_buff *skb, |
| 1734 | __u16 *mss); |
| 1735 | #endif |
| 1736 | struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, |
| 1737 | const struct request_sock *req, |
| 1738 | bool *strict); |
| 1739 | __u32 (*init_seq)(const struct sk_buff *skb); |
| 1740 | int (*send_synack)(const struct sock *sk, struct dst_entry *dst, |
| 1741 | struct flowi *fl, struct request_sock *req, |
| 1742 | struct tcp_fastopen_cookie *foc, |
| 1743 | bool attach_req); |
| 1744 | }; |
| 1745 | |
| 1746 | #ifdef CONFIG_SYN_COOKIES |
| 1747 | static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, |
| 1748 | const struct sock *sk, struct sk_buff *skb, |
| 1749 | __u16 *mss) |
| 1750 | { |
| 1751 | tcp_synq_overflow(sk); |
| 1752 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); |
| 1753 | return ops->cookie_init_seq(skb, mss); |
| 1754 | } |
| 1755 | #else |
| 1756 | static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, |
| 1757 | const struct sock *sk, struct sk_buff *skb, |
| 1758 | __u16 *mss) |
| 1759 | { |
| 1760 | return 0; |
| 1761 | } |
| 1762 | #endif |
| 1763 | |
| 1764 | int tcpv4_offload_init(void); |
| 1765 | |
| 1766 | void tcp_v4_init(void); |
| 1767 | void tcp_init(void); |
| 1768 | |
| 1769 | /* tcp_recovery.c */ |
| 1770 | |
| 1771 | /* Flags to enable various loss recovery features. See below */ |
| 1772 | extern int sysctl_tcp_recovery; |
| 1773 | |
| 1774 | /* Use TCP RACK to detect (some) tail and retransmit losses */ |
| 1775 | #define TCP_RACK_LOST_RETRANS 0x1 |
| 1776 | |
| 1777 | extern int tcp_rack_mark_lost(struct sock *sk); |
| 1778 | |
| 1779 | extern void tcp_rack_advance(struct tcp_sock *tp, |
| 1780 | const struct skb_mstamp *xmit_time, u8 sacked); |
| 1781 | |
| 1782 | /* |
| 1783 | * Save and compile IPv4 options, return a pointer to it |
| 1784 | */ |
| 1785 | static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) |
| 1786 | { |
| 1787 | const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; |
| 1788 | struct ip_options_rcu *dopt = NULL; |
| 1789 | |
| 1790 | if (opt->optlen) { |
| 1791 | int opt_size = sizeof(*dopt) + opt->optlen; |
| 1792 | |
| 1793 | dopt = kmalloc(opt_size, GFP_ATOMIC); |
| 1794 | if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { |
| 1795 | kfree(dopt); |
| 1796 | dopt = NULL; |
| 1797 | } |
| 1798 | } |
| 1799 | return dopt; |
| 1800 | } |
| 1801 | |
| 1802 | /* locally generated TCP pure ACKs have skb->truesize == 2 |
| 1803 | * (check tcp_send_ack() in net/ipv4/tcp_output.c ) |
| 1804 | * This is much faster than dissecting the packet to find out. |
| 1805 | * (Think of GRE encapsulations, IPv4, IPv6, ...) |
| 1806 | */ |
| 1807 | static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) |
| 1808 | { |
| 1809 | return skb->truesize == 2; |
| 1810 | } |
| 1811 | |
| 1812 | static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) |
| 1813 | { |
| 1814 | skb->truesize = 2; |
| 1815 | } |
| 1816 | |
| 1817 | static inline int tcp_inq(struct sock *sk) |
| 1818 | { |
| 1819 | struct tcp_sock *tp = tcp_sk(sk); |
| 1820 | int answ; |
| 1821 | |
| 1822 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { |
| 1823 | answ = 0; |
| 1824 | } else if (sock_flag(sk, SOCK_URGINLINE) || |
| 1825 | !tp->urg_data || |
| 1826 | before(tp->urg_seq, tp->copied_seq) || |
| 1827 | !before(tp->urg_seq, tp->rcv_nxt)) { |
| 1828 | |
| 1829 | answ = tp->rcv_nxt - tp->copied_seq; |
| 1830 | |
| 1831 | /* Subtract 1, if FIN was received */ |
| 1832 | if (answ && sock_flag(sk, SOCK_DONE)) |
| 1833 | answ--; |
| 1834 | } else { |
| 1835 | answ = tp->urg_seq - tp->copied_seq; |
| 1836 | } |
| 1837 | |
| 1838 | return answ; |
| 1839 | } |
| 1840 | |
| 1841 | static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) |
| 1842 | { |
| 1843 | u16 segs_in; |
| 1844 | |
| 1845 | segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); |
| 1846 | tp->segs_in += segs_in; |
| 1847 | if (skb->len > tcp_hdrlen(skb)) |
| 1848 | tp->data_segs_in += segs_in; |
| 1849 | } |
| 1850 | |
| 1851 | #endif /* _TCP_H */ |