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.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
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.
21 #define FASTRETRANS_DEBUG 1
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/crypto.h>
31 #include <linux/cryptohash.h>
32 #include <linux/kref.h>
33 #include <linux/ktime.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo;
52 extern struct percpu_counter tcp_orphan_count;
53 void tcp_time_wait(struct sock *sk, int state, int timeo);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 1024
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL 600
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD 8
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
102 * when active opening a connection.
103 * RFC1122 says the minimum retry MUST
104 * be at least 180secs. Nevertheless
105 * this value is corresponding to
106 * 63secs of retransmission with the
107 * current initial RTO.
110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
111 * when passive opening a connection.
112 * This is corresponding to 31secs of
113 * retransmission with the current
117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 * state, about 60 seconds */
119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
120 /* BSD style FIN_WAIT2 deadlock breaker.
121 * It used to be 3min, new value is 60sec,
122 * to combine FIN-WAIT-2 timeout with
126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
129 #define TCP_ATO_MIN ((unsigned)(HZ/25))
131 #define TCP_DELACK_MIN 4U
132 #define TCP_ATO_MIN 4U
134 #define TCP_RTO_MAX ((unsigned)(120*HZ))
135 #define TCP_RTO_MIN ((unsigned)(HZ/5))
136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
138 * used as a fallback RTO for the
139 * initial data transmission if no
140 * valid RTT sample has been acquired,
141 * most likely due to retrans in 3WHS.
144 /* Number of full MSS to receive before Acking RFC2581 */
145 #define TCP_DELACK_SEG 1
147 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
148 * for local resources.
151 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153 #define TCP_KEEPALIVE_INTVL (75*HZ)
155 #define MAX_TCP_KEEPIDLE 32767
156 #define MAX_TCP_KEEPINTVL 32767
157 #define MAX_TCP_KEEPCNT 127
158 #define MAX_TCP_SYNCNT 127
160 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
162 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
169 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
177 #define TCPOPT_NOP 1 /* Padding */
178 #define TCPOPT_EOL 0 /* End of options */
179 #define TCPOPT_MSS 2 /* Segment size negotiating */
180 #define TCPOPT_WINDOW 3 /* Window scaling */
181 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182 #define TCPOPT_SACK 5 /* SACK Block */
183 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
184 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
185 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
186 #define TCPOPT_EXP 254 /* Experimental */
187 /* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
190 #define TCPOPT_FASTOPEN_MAGIC 0xF989
196 #define TCPOLEN_MSS 4
197 #define TCPOLEN_WINDOW 3
198 #define TCPOLEN_SACK_PERM 2
199 #define TCPOLEN_TIMESTAMP 10
200 #define TCPOLEN_MD5SIG 18
201 #define TCPOLEN_FASTOPEN_BASE 2
202 #define TCPOLEN_EXP_FASTOPEN_BASE 4
204 /* But this is what stacks really send out. */
205 #define TCPOLEN_TSTAMP_ALIGNED 12
206 #define TCPOLEN_WSCALE_ALIGNED 4
207 #define TCPOLEN_SACKPERM_ALIGNED 4
208 #define TCPOLEN_SACK_BASE 2
209 #define TCPOLEN_SACK_BASE_ALIGNED 4
210 #define TCPOLEN_SACK_PERBLOCK 8
211 #define TCPOLEN_MD5SIG_ALIGNED 20
212 #define TCPOLEN_MSS_ALIGNED 4
214 /* Flags in tp->nonagle */
215 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
216 #define TCP_NAGLE_CORK 2 /* Socket is corked */
217 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
219 /* TCP thin-stream limits */
220 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
222 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
223 #define TCP_INIT_CWND 10
225 /* Bit Flags for sysctl_tcp_fastopen */
226 #define TFO_CLIENT_ENABLE 1
227 #define TFO_SERVER_ENABLE 2
228 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
230 /* Accept SYN data w/o any cookie option */
231 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
233 /* Force enable TFO on all listeners, i.e., not requiring the
234 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
236 #define TFO_SERVER_WO_SOCKOPT1 0x400
237 #define TFO_SERVER_WO_SOCKOPT2 0x800
239 extern struct inet_timewait_death_row tcp_death_row;
241 /* sysctl variables for tcp */
242 extern int sysctl_tcp_timestamps;
243 extern int sysctl_tcp_window_scaling;
244 extern int sysctl_tcp_sack;
245 extern int sysctl_tcp_fin_timeout;
246 extern int sysctl_tcp_keepalive_time;
247 extern int sysctl_tcp_keepalive_probes;
248 extern int sysctl_tcp_keepalive_intvl;
249 extern int sysctl_tcp_syn_retries;
250 extern int sysctl_tcp_synack_retries;
251 extern int sysctl_tcp_retries1;
252 extern int sysctl_tcp_retries2;
253 extern int sysctl_tcp_orphan_retries;
254 extern int sysctl_tcp_syncookies;
255 extern int sysctl_tcp_fastopen;
256 extern int sysctl_tcp_retrans_collapse;
257 extern int sysctl_tcp_stdurg;
258 extern int sysctl_tcp_rfc1337;
259 extern int sysctl_tcp_abort_on_overflow;
260 extern int sysctl_tcp_max_orphans;
261 extern int sysctl_tcp_fack;
262 extern int sysctl_tcp_reordering;
263 extern int sysctl_tcp_max_reordering;
264 extern int sysctl_tcp_dsack;
265 extern long sysctl_tcp_mem[3];
266 extern int sysctl_tcp_wmem[3];
267 extern int sysctl_tcp_rmem[3];
268 extern int sysctl_tcp_app_win;
269 extern int sysctl_tcp_adv_win_scale;
270 extern int sysctl_tcp_tw_reuse;
271 extern int sysctl_tcp_frto;
272 extern int sysctl_tcp_low_latency;
273 extern int sysctl_tcp_nometrics_save;
274 extern int sysctl_tcp_moderate_rcvbuf;
275 extern int sysctl_tcp_tso_win_divisor;
276 extern int sysctl_tcp_workaround_signed_windows;
277 extern int sysctl_tcp_slow_start_after_idle;
278 extern int sysctl_tcp_thin_linear_timeouts;
279 extern int sysctl_tcp_thin_dupack;
280 extern int sysctl_tcp_early_retrans;
281 extern int sysctl_tcp_limit_output_bytes;
282 extern int sysctl_tcp_challenge_ack_limit;
283 extern unsigned int sysctl_tcp_notsent_lowat;
284 extern int sysctl_tcp_min_tso_segs;
285 extern int sysctl_tcp_min_rtt_wlen;
286 extern int sysctl_tcp_autocorking;
287 extern int sysctl_tcp_invalid_ratelimit;
288 extern int sysctl_tcp_pacing_ss_ratio;
289 extern int sysctl_tcp_pacing_ca_ratio;
290 extern int sysctl_tcp_default_init_rwnd;
292 extern atomic_long_t tcp_memory_allocated;
294 /* sysctl variables for controlling various tcp parameters */
295 extern int sysctl_tcp_delack_seg;
296 extern int sysctl_tcp_use_userconfig;
298 extern struct percpu_counter tcp_sockets_allocated;
299 extern int tcp_memory_pressure;
301 /* optimized version of sk_under_memory_pressure() for TCP sockets */
302 static inline bool tcp_under_memory_pressure(const struct sock *sk)
304 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
305 return !!sk->sk_cgrp->memory_pressure;
307 return tcp_memory_pressure;
310 * The next routines deal with comparing 32 bit unsigned ints
311 * and worry about wraparound (automatic with unsigned arithmetic).
314 static inline bool before(__u32 seq1, __u32 seq2)
316 return (__s32)(seq1-seq2) < 0;
318 #define after(seq2, seq1) before(seq1, seq2)
320 /* is s2<=s1<=s3 ? */
321 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
323 return seq3 - seq2 >= seq1 - seq2;
326 static inline bool tcp_out_of_memory(struct sock *sk)
328 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
329 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
334 void sk_forced_mem_schedule(struct sock *sk, int size);
336 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
338 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
339 int orphans = percpu_counter_read_positive(ocp);
341 if (orphans << shift > sysctl_tcp_max_orphans) {
342 orphans = percpu_counter_sum_positive(ocp);
343 if (orphans << shift > sysctl_tcp_max_orphans)
349 bool tcp_check_oom(struct sock *sk, int shift);
352 extern struct proto tcp_prot;
354 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
355 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
356 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
357 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
358 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
360 void tcp_tasklet_init(void);
362 void tcp_v4_err(struct sk_buff *skb, u32);
364 void tcp_shutdown(struct sock *sk, int how);
366 void tcp_v4_early_demux(struct sk_buff *skb);
367 int tcp_v4_rcv(struct sk_buff *skb);
369 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
370 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
371 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
373 void tcp_release_cb(struct sock *sk);
374 void tcp_wfree(struct sk_buff *skb);
375 void tcp_write_timer_handler(struct sock *sk);
376 void tcp_delack_timer_handler(struct sock *sk);
377 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
378 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
379 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
380 const struct tcphdr *th, unsigned int len);
381 void tcp_rcv_space_adjust(struct sock *sk);
382 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
383 void tcp_twsk_destructor(struct sock *sk);
384 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
385 struct pipe_inode_info *pipe, size_t len,
388 /* sysctl master controller */
389 extern int tcp_use_userconfig_sysctl_handler(struct ctl_table *, int,
390 void __user *, size_t *, loff_t *);
391 extern int tcp_proc_delayed_ack_control(struct ctl_table *, int,
392 void __user *, size_t *, loff_t *);
394 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
396 static inline void tcp_dec_quickack_mode(struct sock *sk,
397 const unsigned int pkts)
399 struct inet_connection_sock *icsk = inet_csk(sk);
401 if (icsk->icsk_ack.quick) {
402 if (pkts >= icsk->icsk_ack.quick) {
403 icsk->icsk_ack.quick = 0;
404 /* Leaving quickack mode we deflate ATO. */
405 icsk->icsk_ack.ato = TCP_ATO_MIN;
407 icsk->icsk_ack.quick -= pkts;
412 #define TCP_ECN_QUEUE_CWR 2
413 #define TCP_ECN_DEMAND_CWR 4
414 #define TCP_ECN_SEEN 8
424 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
426 const struct tcphdr *th);
427 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
428 struct request_sock *req, bool fastopen);
429 int tcp_child_process(struct sock *parent, struct sock *child,
430 struct sk_buff *skb);
431 void tcp_enter_loss(struct sock *sk);
432 void tcp_clear_retrans(struct tcp_sock *tp);
433 void tcp_update_metrics(struct sock *sk);
434 void tcp_init_metrics(struct sock *sk);
435 void tcp_metrics_init(void);
436 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
437 bool paws_check, bool timestamps);
438 bool tcp_remember_stamp(struct sock *sk);
439 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
440 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
441 void tcp_disable_fack(struct tcp_sock *tp);
442 void tcp_close(struct sock *sk, long timeout);
443 void tcp_init_sock(struct sock *sk);
444 unsigned int tcp_poll(struct file *file, struct socket *sock,
445 struct poll_table_struct *wait);
446 int tcp_getsockopt(struct sock *sk, int level, int optname,
447 char __user *optval, int __user *optlen);
448 int tcp_setsockopt(struct sock *sk, int level, int optname,
449 char __user *optval, unsigned int optlen);
450 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
451 char __user *optval, int __user *optlen);
452 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
453 char __user *optval, unsigned int optlen);
454 void tcp_set_keepalive(struct sock *sk, int val);
455 void tcp_syn_ack_timeout(const struct request_sock *req);
456 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
457 int flags, int *addr_len);
458 void tcp_parse_options(const struct sk_buff *skb,
459 struct tcp_options_received *opt_rx,
460 int estab, struct tcp_fastopen_cookie *foc);
461 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
464 * TCP v4 functions exported for the inet6 API
467 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
468 void tcp_v4_mtu_reduced(struct sock *sk);
469 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
470 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
471 struct sock *tcp_create_openreq_child(const struct sock *sk,
472 struct request_sock *req,
473 struct sk_buff *skb);
474 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
475 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
476 struct request_sock *req,
477 struct dst_entry *dst,
478 struct request_sock *req_unhash,
480 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
481 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
482 int tcp_connect(struct sock *sk);
483 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
484 struct request_sock *req,
485 struct tcp_fastopen_cookie *foc,
487 int tcp_disconnect(struct sock *sk, int flags);
489 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
490 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
491 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
493 /* From syncookies.c */
494 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
495 struct request_sock *req,
496 struct dst_entry *dst);
497 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
499 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
500 #ifdef CONFIG_SYN_COOKIES
502 /* Syncookies use a monotonic timer which increments every 60 seconds.
503 * This counter is used both as a hash input and partially encoded into
504 * the cookie value. A cookie is only validated further if the delta
505 * between the current counter value and the encoded one is less than this,
506 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
507 * the counter advances immediately after a cookie is generated).
509 #define MAX_SYNCOOKIE_AGE 2
510 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
511 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
513 /* syncookies: remember time of last synqueue overflow
514 * But do not dirty this field too often (once per second is enough)
515 * It is racy as we do not hold a lock, but race is very minor.
517 static inline void tcp_synq_overflow(const struct sock *sk)
519 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
520 unsigned long now = jiffies;
522 if (time_after(now, last_overflow + HZ))
523 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
526 /* syncookies: no recent synqueue overflow on this listening socket? */
527 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
529 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
531 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
534 static inline u32 tcp_cookie_time(void)
536 u64 val = get_jiffies_64();
538 do_div(val, TCP_SYNCOOKIE_PERIOD);
542 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
544 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
545 __u32 cookie_init_timestamp(struct request_sock *req);
546 bool cookie_timestamp_decode(struct tcp_options_received *opt);
547 bool cookie_ecn_ok(const struct tcp_options_received *opt,
548 const struct net *net, const struct dst_entry *dst);
550 /* From net/ipv6/syncookies.c */
551 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
553 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
555 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
556 const struct tcphdr *th, u16 *mssp);
557 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
561 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
563 bool tcp_may_send_now(struct sock *sk);
564 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
565 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
566 void tcp_retransmit_timer(struct sock *sk);
567 void tcp_xmit_retransmit_queue(struct sock *);
568 void tcp_simple_retransmit(struct sock *);
569 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
570 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
572 void tcp_send_probe0(struct sock *);
573 void tcp_send_partial(struct sock *);
574 int tcp_write_wakeup(struct sock *, int mib);
575 void tcp_send_fin(struct sock *sk);
576 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
577 int tcp_send_synack(struct sock *);
578 void tcp_push_one(struct sock *, unsigned int mss_now);
579 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
580 void tcp_send_ack(struct sock *sk);
581 void tcp_send_delayed_ack(struct sock *sk);
582 void tcp_send_loss_probe(struct sock *sk);
583 bool tcp_schedule_loss_probe(struct sock *sk);
586 void tcp_resume_early_retransmit(struct sock *sk);
587 void tcp_rearm_rto(struct sock *sk);
588 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
589 void tcp_reset(struct sock *sk);
590 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
593 void tcp_init_xmit_timers(struct sock *);
594 static inline void tcp_clear_xmit_timers(struct sock *sk)
596 inet_csk_clear_xmit_timers(sk);
599 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
600 unsigned int tcp_current_mss(struct sock *sk);
602 /* Bound MSS / TSO packet size with the half of the window */
603 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
607 /* When peer uses tiny windows, there is no use in packetizing
608 * to sub-MSS pieces for the sake of SWS or making sure there
609 * are enough packets in the pipe for fast recovery.
611 * On the other hand, for extremely large MSS devices, handling
612 * smaller than MSS windows in this way does make sense.
614 if (tp->max_window >= 512)
615 cutoff = (tp->max_window >> 1);
617 cutoff = tp->max_window;
619 if (cutoff && pktsize > cutoff)
620 return max_t(int, cutoff, 68U - tp->tcp_header_len);
626 void tcp_get_info(struct sock *, struct tcp_info *);
628 /* Read 'sendfile()'-style from a TCP socket */
629 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
630 unsigned int, size_t);
631 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
632 sk_read_actor_t recv_actor);
634 void tcp_initialize_rcv_mss(struct sock *sk);
636 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
637 int tcp_mss_to_mtu(struct sock *sk, int mss);
638 void tcp_mtup_init(struct sock *sk);
639 void tcp_init_buffer_space(struct sock *sk);
641 static inline void tcp_bound_rto(const struct sock *sk)
643 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
644 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
647 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
649 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
652 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
654 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
655 ntohl(TCP_FLAG_ACK) |
659 static inline void tcp_fast_path_on(struct tcp_sock *tp)
661 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
664 static inline void tcp_fast_path_check(struct sock *sk)
666 struct tcp_sock *tp = tcp_sk(sk);
668 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
670 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
672 tcp_fast_path_on(tp);
675 /* Compute the actual rto_min value */
676 static inline u32 tcp_rto_min(struct sock *sk)
678 const struct dst_entry *dst = __sk_dst_get(sk);
679 u32 rto_min = TCP_RTO_MIN;
681 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
682 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
686 static inline u32 tcp_rto_min_us(struct sock *sk)
688 return jiffies_to_usecs(tcp_rto_min(sk));
691 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
693 return dst_metric_locked(dst, RTAX_CC_ALGO);
696 /* Minimum RTT in usec. ~0 means not available. */
697 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
699 return tp->rtt_min[0].rtt;
702 /* Compute the actual receive window we are currently advertising.
703 * Rcv_nxt can be after the window if our peer push more data
704 * than the offered window.
706 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
708 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
715 /* Choose a new window, without checks for shrinking, and without
716 * scaling applied to the result. The caller does these things
717 * if necessary. This is a "raw" window selection.
719 u32 __tcp_select_window(struct sock *sk);
721 void tcp_send_window_probe(struct sock *sk);
723 /* TCP timestamps are only 32-bits, this causes a slight
724 * complication on 64-bit systems since we store a snapshot
725 * of jiffies in the buffer control blocks below. We decided
726 * to use only the low 32-bits of jiffies and hide the ugly
727 * casts with the following macro.
729 #define tcp_time_stamp ((__u32)(jiffies))
731 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
733 return skb->skb_mstamp.stamp_jiffies;
737 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
739 #define TCPHDR_FIN 0x01
740 #define TCPHDR_SYN 0x02
741 #define TCPHDR_RST 0x04
742 #define TCPHDR_PSH 0x08
743 #define TCPHDR_ACK 0x10
744 #define TCPHDR_URG 0x20
745 #define TCPHDR_ECE 0x40
746 #define TCPHDR_CWR 0x80
748 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
750 /* This is what the send packet queuing engine uses to pass
751 * TCP per-packet control information to the transmission code.
752 * We also store the host-order sequence numbers in here too.
753 * This is 44 bytes if IPV6 is enabled.
754 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
757 __u32 seq; /* Starting sequence number */
758 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
760 /* Note : tcp_tw_isn is used in input path only
761 * (isn chosen by tcp_timewait_state_process())
763 * tcp_gso_segs/size are used in write queue only,
764 * cf tcp_skb_pcount()/tcp_skb_mss()
772 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
774 __u8 sacked; /* State flags for SACK/FACK. */
775 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
776 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
777 #define TCPCB_LOST 0x04 /* SKB is lost */
778 #define TCPCB_TAGBITS 0x07 /* All tag bits */
779 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
780 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
781 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
784 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
786 __u32 ack_seq; /* Sequence number ACK'd */
788 struct inet_skb_parm h4;
789 #if IS_ENABLED(CONFIG_IPV6)
790 struct inet6_skb_parm h6;
792 } header; /* For incoming frames */
795 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
798 #if IS_ENABLED(CONFIG_IPV6)
799 /* This is the variant of inet6_iif() that must be used by TCP,
800 * as TCP moves IP6CB into a different location in skb->cb[]
802 static inline int tcp_v6_iif(const struct sk_buff *skb)
804 return TCP_SKB_CB(skb)->header.h6.iif;
808 /* Due to TSO, an SKB can be composed of multiple actual
809 * packets. To keep these tracked properly, we use this.
811 static inline int tcp_skb_pcount(const struct sk_buff *skb)
813 return TCP_SKB_CB(skb)->tcp_gso_segs;
816 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
818 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
821 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
823 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
826 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
827 static inline int tcp_skb_mss(const struct sk_buff *skb)
829 return TCP_SKB_CB(skb)->tcp_gso_size;
832 /* Events passed to congestion control interface */
834 CA_EVENT_TX_START, /* first transmit when no packets in flight */
835 CA_EVENT_CWND_RESTART, /* congestion window restart */
836 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
837 CA_EVENT_LOSS, /* loss timeout */
838 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
839 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
842 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
843 enum tcp_ca_ack_event_flags {
844 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
845 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
846 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
850 * Interface for adding new TCP congestion control handlers
852 #define TCP_CA_NAME_MAX 16
853 #define TCP_CA_MAX 128
854 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
856 #define TCP_CA_UNSPEC 0
858 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
859 #define TCP_CONG_NON_RESTRICTED 0x1
860 /* Requires ECN/ECT set on all packets */
861 #define TCP_CONG_NEEDS_ECN 0x2
865 struct tcp_congestion_ops {
866 struct list_head list;
870 /* initialize private data (optional) */
871 void (*init)(struct sock *sk);
872 /* cleanup private data (optional) */
873 void (*release)(struct sock *sk);
875 /* return slow start threshold (required) */
876 u32 (*ssthresh)(struct sock *sk);
877 /* do new cwnd calculation (required) */
878 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
879 /* call before changing ca_state (optional) */
880 void (*set_state)(struct sock *sk, u8 new_state);
881 /* call when cwnd event occurs (optional) */
882 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
883 /* call when ack arrives (optional) */
884 void (*in_ack_event)(struct sock *sk, u32 flags);
885 /* new value of cwnd after loss (optional) */
886 u32 (*undo_cwnd)(struct sock *sk);
887 /* hook for packet ack accounting (optional) */
888 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
889 /* get info for inet_diag (optional) */
890 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
891 union tcp_cc_info *info);
893 char name[TCP_CA_NAME_MAX];
894 struct module *owner;
897 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
898 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
900 void tcp_assign_congestion_control(struct sock *sk);
901 void tcp_init_congestion_control(struct sock *sk);
902 void tcp_cleanup_congestion_control(struct sock *sk);
903 int tcp_set_default_congestion_control(const char *name);
904 void tcp_get_default_congestion_control(char *name);
905 void tcp_get_available_congestion_control(char *buf, size_t len);
906 void tcp_get_allowed_congestion_control(char *buf, size_t len);
907 int tcp_set_allowed_congestion_control(char *allowed);
908 int tcp_set_congestion_control(struct sock *sk, const char *name);
909 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
910 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
912 u32 tcp_reno_ssthresh(struct sock *sk);
913 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
914 extern struct tcp_congestion_ops tcp_reno;
916 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
917 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
919 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
921 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
927 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
929 const struct inet_connection_sock *icsk = inet_csk(sk);
931 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
934 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
936 struct inet_connection_sock *icsk = inet_csk(sk);
938 if (icsk->icsk_ca_ops->set_state)
939 icsk->icsk_ca_ops->set_state(sk, ca_state);
940 icsk->icsk_ca_state = ca_state;
943 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
945 const struct inet_connection_sock *icsk = inet_csk(sk);
947 if (icsk->icsk_ca_ops->cwnd_event)
948 icsk->icsk_ca_ops->cwnd_event(sk, event);
951 /* These functions determine how the current flow behaves in respect of SACK
952 * handling. SACK is negotiated with the peer, and therefore it can vary
953 * between different flows.
955 * tcp_is_sack - SACK enabled
956 * tcp_is_reno - No SACK
957 * tcp_is_fack - FACK enabled, implies SACK enabled
959 static inline int tcp_is_sack(const struct tcp_sock *tp)
961 return tp->rx_opt.sack_ok;
964 static inline bool tcp_is_reno(const struct tcp_sock *tp)
966 return !tcp_is_sack(tp);
969 static inline bool tcp_is_fack(const struct tcp_sock *tp)
971 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
974 static inline void tcp_enable_fack(struct tcp_sock *tp)
976 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
979 /* TCP early-retransmit (ER) is similar to but more conservative than
980 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
982 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
984 tp->do_early_retrans = sysctl_tcp_early_retrans &&
985 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
986 sysctl_tcp_reordering == 3;
989 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
991 tp->do_early_retrans = 0;
994 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
996 return tp->sacked_out + tp->lost_out;
999 /* This determines how many packets are "in the network" to the best
1000 * of our knowledge. In many cases it is conservative, but where
1001 * detailed information is available from the receiver (via SACK
1002 * blocks etc.) we can make more aggressive calculations.
1004 * Use this for decisions involving congestion control, use just
1005 * tp->packets_out to determine if the send queue is empty or not.
1007 * Read this equation as:
1009 * "Packets sent once on transmission queue" MINUS
1010 * "Packets left network, but not honestly ACKed yet" PLUS
1011 * "Packets fast retransmitted"
1013 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1015 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1018 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1020 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1022 return tp->snd_cwnd < tp->snd_ssthresh;
1025 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1027 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1030 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1032 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1033 (1 << inet_csk(sk)->icsk_ca_state);
1036 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1037 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1040 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1042 const struct tcp_sock *tp = tcp_sk(sk);
1044 if (tcp_in_cwnd_reduction(sk))
1045 return tp->snd_ssthresh;
1047 return max(tp->snd_ssthresh,
1048 ((tp->snd_cwnd >> 1) +
1049 (tp->snd_cwnd >> 2)));
1052 /* Use define here intentionally to get WARN_ON location shown at the caller */
1053 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1055 void tcp_enter_cwr(struct sock *sk);
1056 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1058 /* The maximum number of MSS of available cwnd for which TSO defers
1059 * sending if not using sysctl_tcp_tso_win_divisor.
1061 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1066 /* Slow start with delack produces 3 packets of burst, so that
1067 * it is safe "de facto". This will be the default - same as
1068 * the default reordering threshold - but if reordering increases,
1069 * we must be able to allow cwnd to burst at least this much in order
1070 * to not pull it back when holes are filled.
1072 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1074 return tp->reordering;
1077 /* Returns end sequence number of the receiver's advertised window */
1078 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1080 return tp->snd_una + tp->snd_wnd;
1083 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1084 * flexible approach. The RFC suggests cwnd should not be raised unless
1085 * it was fully used previously. And that's exactly what we do in
1086 * congestion avoidance mode. But in slow start we allow cwnd to grow
1087 * as long as the application has used half the cwnd.
1089 * cwnd is 10 (IW10), but application sends 9 frames.
1090 * We allow cwnd to reach 18 when all frames are ACKed.
1091 * This check is safe because it's as aggressive as slow start which already
1092 * risks 100% overshoot. The advantage is that we discourage application to
1093 * either send more filler packets or data to artificially blow up the cwnd
1094 * usage, and allow application-limited process to probe bw more aggressively.
1096 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1098 const struct tcp_sock *tp = tcp_sk(sk);
1100 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1101 if (tcp_in_slow_start(tp))
1102 return tp->snd_cwnd < 2 * tp->max_packets_out;
1104 return tp->is_cwnd_limited;
1107 /* Something is really bad, we could not queue an additional packet,
1108 * because qdisc is full or receiver sent a 0 window.
1109 * We do not want to add fuel to the fire, or abort too early,
1110 * so make sure the timer we arm now is at least 200ms in the future,
1111 * regardless of current icsk_rto value (as it could be ~2ms)
1113 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1115 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1118 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1119 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1120 unsigned long max_when)
1122 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1124 return (unsigned long)min_t(u64, when, max_when);
1127 static inline void tcp_check_probe_timer(struct sock *sk)
1129 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1130 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1131 tcp_probe0_base(sk), TCP_RTO_MAX);
1134 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1139 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1145 * Calculate(/check) TCP checksum
1147 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1148 __be32 daddr, __wsum base)
1150 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1153 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1155 return __skb_checksum_complete(skb);
1158 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1160 return !skb_csum_unnecessary(skb) &&
1161 __tcp_checksum_complete(skb);
1164 /* Prequeue for VJ style copy to user, combined with checksumming. */
1166 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1168 tp->ucopy.task = NULL;
1170 tp->ucopy.memory = 0;
1171 skb_queue_head_init(&tp->ucopy.prequeue);
1174 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1175 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1180 static const char *statename[]={
1181 "Unused","Established","Syn Sent","Syn Recv",
1182 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1183 "Close Wait","Last ACK","Listen","Closing"
1186 void tcp_set_state(struct sock *sk, int state);
1188 void tcp_done(struct sock *sk);
1190 int tcp_abort(struct sock *sk, int err);
1192 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1195 rx_opt->num_sacks = 0;
1198 u32 tcp_default_init_rwnd(u32 mss);
1199 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1201 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1203 struct tcp_sock *tp = tcp_sk(sk);
1206 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1208 delta = tcp_time_stamp - tp->lsndtime;
1209 if (delta > inet_csk(sk)->icsk_rto)
1210 tcp_cwnd_restart(sk, delta);
1213 /* Determine a window scaling and initial window to offer. */
1214 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1215 __u32 *window_clamp, int wscale_ok,
1216 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1218 static inline int tcp_win_from_space(int space)
1220 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1222 return tcp_adv_win_scale <= 0 ?
1223 (space>>(-tcp_adv_win_scale)) :
1224 space - (space>>tcp_adv_win_scale);
1227 /* Note: caller must be prepared to deal with negative returns */
1228 static inline int tcp_space(const struct sock *sk)
1230 return tcp_win_from_space(sk->sk_rcvbuf -
1231 atomic_read(&sk->sk_rmem_alloc));
1234 static inline int tcp_full_space(const struct sock *sk)
1236 return tcp_win_from_space(sk->sk_rcvbuf);
1239 extern void tcp_openreq_init_rwin(struct request_sock *req,
1240 const struct sock *sk_listener,
1241 const struct dst_entry *dst);
1243 void tcp_enter_memory_pressure(struct sock *sk);
1245 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1247 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1250 static inline int keepalive_time_when(const struct tcp_sock *tp)
1252 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1255 static inline int keepalive_probes(const struct tcp_sock *tp)
1257 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1260 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1262 const struct inet_connection_sock *icsk = &tp->inet_conn;
1264 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1265 tcp_time_stamp - tp->rcv_tstamp);
1268 static inline int tcp_fin_time(const struct sock *sk)
1270 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1271 const int rto = inet_csk(sk)->icsk_rto;
1273 if (fin_timeout < (rto << 2) - (rto >> 1))
1274 fin_timeout = (rto << 2) - (rto >> 1);
1279 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1282 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1284 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1287 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1288 * then following tcp messages have valid values. Ignore 0 value,
1289 * or else 'negative' tsval might forbid us to accept their packets.
1291 if (!rx_opt->ts_recent)
1296 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1299 if (tcp_paws_check(rx_opt, 0))
1302 /* RST segments are not recommended to carry timestamp,
1303 and, if they do, it is recommended to ignore PAWS because
1304 "their cleanup function should take precedence over timestamps."
1305 Certainly, it is mistake. It is necessary to understand the reasons
1306 of this constraint to relax it: if peer reboots, clock may go
1307 out-of-sync and half-open connections will not be reset.
1308 Actually, the problem would be not existing if all
1309 the implementations followed draft about maintaining clock
1310 via reboots. Linux-2.2 DOES NOT!
1312 However, we can relax time bounds for RST segments to MSL.
1314 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1319 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1320 int mib_idx, u32 *last_oow_ack_time);
1322 static inline void tcp_mib_init(struct net *net)
1325 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1326 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1327 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1328 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1332 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1334 tp->lost_skb_hint = NULL;
1337 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1339 tcp_clear_retrans_hints_partial(tp);
1340 tp->retransmit_skb_hint = NULL;
1346 union tcp_md5_addr {
1348 #if IS_ENABLED(CONFIG_IPV6)
1353 /* - key database */
1354 struct tcp_md5sig_key {
1355 struct hlist_node node;
1357 u8 family; /* AF_INET or AF_INET6 */
1358 union tcp_md5_addr addr;
1359 u8 key[TCP_MD5SIG_MAXKEYLEN];
1360 struct rcu_head rcu;
1364 struct tcp_md5sig_info {
1365 struct hlist_head head;
1366 struct rcu_head rcu;
1369 /* - pseudo header */
1370 struct tcp4_pseudohdr {
1378 struct tcp6_pseudohdr {
1379 struct in6_addr saddr;
1380 struct in6_addr daddr;
1382 __be32 protocol; /* including padding */
1385 union tcp_md5sum_block {
1386 struct tcp4_pseudohdr ip4;
1387 #if IS_ENABLED(CONFIG_IPV6)
1388 struct tcp6_pseudohdr ip6;
1392 /* - pool: digest algorithm, hash description and scratch buffer */
1393 struct tcp_md5sig_pool {
1394 struct hash_desc md5_desc;
1395 union tcp_md5sum_block md5_blk;
1399 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1400 const struct sock *sk, const struct sk_buff *skb);
1401 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1402 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1403 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1405 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1406 const struct sock *addr_sk);
1408 #ifdef CONFIG_TCP_MD5SIG
1409 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1410 const union tcp_md5_addr *addr,
1412 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1414 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1415 const union tcp_md5_addr *addr,
1420 #define tcp_twsk_md5_key(twsk) NULL
1423 bool tcp_alloc_md5sig_pool(void);
1425 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1426 static inline void tcp_put_md5sig_pool(void)
1431 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1432 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1433 unsigned int header_len);
1434 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1435 const struct tcp_md5sig_key *key);
1437 /* From tcp_fastopen.c */
1438 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1439 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1440 unsigned long *last_syn_loss);
1441 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1442 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1444 struct tcp_fastopen_request {
1445 /* Fast Open cookie. Size 0 means a cookie request */
1446 struct tcp_fastopen_cookie cookie;
1447 struct msghdr *data; /* data in MSG_FASTOPEN */
1449 int copied; /* queued in tcp_connect() */
1451 void tcp_free_fastopen_req(struct tcp_sock *tp);
1453 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1454 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1455 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1456 struct request_sock *req,
1457 struct tcp_fastopen_cookie *foc,
1458 struct dst_entry *dst);
1459 void tcp_fastopen_init_key_once(bool publish);
1460 #define TCP_FASTOPEN_KEY_LENGTH 16
1462 /* Fastopen key context */
1463 struct tcp_fastopen_context {
1464 struct crypto_cipher *tfm;
1465 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1466 struct rcu_head rcu;
1469 /* write queue abstraction */
1470 static inline void tcp_write_queue_purge(struct sock *sk)
1472 struct sk_buff *skb;
1474 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1475 sk_wmem_free_skb(sk, skb);
1477 tcp_clear_all_retrans_hints(tcp_sk(sk));
1480 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1482 return skb_peek(&sk->sk_write_queue);
1485 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1487 return skb_peek_tail(&sk->sk_write_queue);
1490 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1491 const struct sk_buff *skb)
1493 return skb_queue_next(&sk->sk_write_queue, skb);
1496 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1497 const struct sk_buff *skb)
1499 return skb_queue_prev(&sk->sk_write_queue, skb);
1502 #define tcp_for_write_queue(skb, sk) \
1503 skb_queue_walk(&(sk)->sk_write_queue, skb)
1505 #define tcp_for_write_queue_from(skb, sk) \
1506 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1508 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1509 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1511 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1513 return sk->sk_send_head;
1516 static inline bool tcp_skb_is_last(const struct sock *sk,
1517 const struct sk_buff *skb)
1519 return skb_queue_is_last(&sk->sk_write_queue, skb);
1522 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1524 if (tcp_skb_is_last(sk, skb))
1525 sk->sk_send_head = NULL;
1527 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1530 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1532 if (sk->sk_send_head == skb_unlinked)
1533 sk->sk_send_head = NULL;
1534 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1535 tcp_sk(sk)->highest_sack = NULL;
1538 static inline void tcp_init_send_head(struct sock *sk)
1540 sk->sk_send_head = NULL;
1543 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1545 __skb_queue_tail(&sk->sk_write_queue, skb);
1548 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1550 __tcp_add_write_queue_tail(sk, skb);
1552 /* Queue it, remembering where we must start sending. */
1553 if (sk->sk_send_head == NULL) {
1554 sk->sk_send_head = skb;
1556 if (tcp_sk(sk)->highest_sack == NULL)
1557 tcp_sk(sk)->highest_sack = skb;
1561 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1563 __skb_queue_head(&sk->sk_write_queue, skb);
1566 /* Insert buff after skb on the write queue of sk. */
1567 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1568 struct sk_buff *buff,
1571 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1574 /* Insert new before skb on the write queue of sk. */
1575 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1576 struct sk_buff *skb,
1579 __skb_queue_before(&sk->sk_write_queue, skb, new);
1581 if (sk->sk_send_head == skb)
1582 sk->sk_send_head = new;
1585 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1587 __skb_unlink(skb, &sk->sk_write_queue);
1590 static inline bool tcp_write_queue_empty(struct sock *sk)
1592 return skb_queue_empty(&sk->sk_write_queue);
1595 static inline void tcp_push_pending_frames(struct sock *sk)
1597 if (tcp_send_head(sk)) {
1598 struct tcp_sock *tp = tcp_sk(sk);
1600 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1604 /* Start sequence of the skb just after the highest skb with SACKed
1605 * bit, valid only if sacked_out > 0 or when the caller has ensured
1606 * validity by itself.
1608 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1610 if (!tp->sacked_out)
1613 if (tp->highest_sack == NULL)
1616 return TCP_SKB_CB(tp->highest_sack)->seq;
1619 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1621 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1622 tcp_write_queue_next(sk, skb);
1625 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1627 return tcp_sk(sk)->highest_sack;
1630 static inline void tcp_highest_sack_reset(struct sock *sk)
1632 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1635 /* Called when old skb is about to be deleted and replaced by new skb */
1636 static inline void tcp_highest_sack_replace(struct sock *sk,
1637 struct sk_buff *old,
1638 struct sk_buff *new)
1640 if (old == tcp_highest_sack(sk))
1641 tcp_sk(sk)->highest_sack = new;
1644 /* Determines whether this is a thin stream (which may suffer from
1645 * increased latency). Used to trigger latency-reducing mechanisms.
1647 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1649 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1653 enum tcp_seq_states {
1654 TCP_SEQ_STATE_LISTENING,
1655 TCP_SEQ_STATE_ESTABLISHED,
1658 int tcp_seq_open(struct inode *inode, struct file *file);
1660 struct tcp_seq_afinfo {
1663 const struct file_operations *seq_fops;
1664 struct seq_operations seq_ops;
1667 struct tcp_iter_state {
1668 struct seq_net_private p;
1670 enum tcp_seq_states state;
1671 struct sock *syn_wait_sk;
1672 int bucket, offset, sbucket, num;
1676 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1677 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1679 extern struct request_sock_ops tcp_request_sock_ops;
1680 extern struct request_sock_ops tcp6_request_sock_ops;
1682 void tcp_v4_destroy_sock(struct sock *sk);
1684 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1685 netdev_features_t features);
1686 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1687 int tcp_gro_complete(struct sk_buff *skb);
1689 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1691 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1693 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1696 static inline bool tcp_stream_memory_free(const struct sock *sk)
1698 const struct tcp_sock *tp = tcp_sk(sk);
1699 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1701 return notsent_bytes < tcp_notsent_lowat(tp);
1704 #ifdef CONFIG_PROC_FS
1705 int tcp4_proc_init(void);
1706 void tcp4_proc_exit(void);
1709 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1710 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1711 const struct tcp_request_sock_ops *af_ops,
1712 struct sock *sk, struct sk_buff *skb);
1714 /* TCP af-specific functions */
1715 struct tcp_sock_af_ops {
1716 #ifdef CONFIG_TCP_MD5SIG
1717 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1718 const struct sock *addr_sk);
1719 int (*calc_md5_hash)(char *location,
1720 const struct tcp_md5sig_key *md5,
1721 const struct sock *sk,
1722 const struct sk_buff *skb);
1723 int (*md5_parse)(struct sock *sk,
1724 char __user *optval,
1729 struct tcp_request_sock_ops {
1731 #ifdef CONFIG_TCP_MD5SIG
1732 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1733 const struct sock *addr_sk);
1734 int (*calc_md5_hash) (char *location,
1735 const struct tcp_md5sig_key *md5,
1736 const struct sock *sk,
1737 const struct sk_buff *skb);
1739 void (*init_req)(struct request_sock *req,
1740 const struct sock *sk_listener,
1741 struct sk_buff *skb);
1742 #ifdef CONFIG_SYN_COOKIES
1743 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1746 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1747 const struct request_sock *req,
1749 __u32 (*init_seq)(const struct sk_buff *skb);
1750 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1751 struct flowi *fl, struct request_sock *req,
1752 struct tcp_fastopen_cookie *foc,
1756 #ifdef CONFIG_SYN_COOKIES
1757 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1758 const struct sock *sk, struct sk_buff *skb,
1761 tcp_synq_overflow(sk);
1762 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1763 return ops->cookie_init_seq(skb, mss);
1766 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1767 const struct sock *sk, struct sk_buff *skb,
1774 int tcpv4_offload_init(void);
1776 void tcp_v4_init(void);
1777 void tcp_init(void);
1779 /* tcp_recovery.c */
1781 /* Flags to enable various loss recovery features. See below */
1782 extern int sysctl_tcp_recovery;
1784 /* Use TCP RACK to detect (some) tail and retransmit losses */
1785 #define TCP_RACK_LOST_RETRANS 0x1
1787 extern int tcp_rack_mark_lost(struct sock *sk);
1789 extern void tcp_rack_advance(struct tcp_sock *tp,
1790 const struct skb_mstamp *xmit_time, u8 sacked);
1793 * Save and compile IPv4 options, return a pointer to it
1795 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1797 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1798 struct ip_options_rcu *dopt = NULL;
1801 int opt_size = sizeof(*dopt) + opt->optlen;
1803 dopt = kmalloc(opt_size, GFP_ATOMIC);
1804 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1812 /* locally generated TCP pure ACKs have skb->truesize == 2
1813 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1814 * This is much faster than dissecting the packet to find out.
1815 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1817 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1819 return skb->truesize == 2;
1822 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)