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 L1_CACHE_ALIGN(128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
57 #define TCP_MIN_SND_MSS 48
58 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
61 * Never offer a window over 32767 without using window scaling. Some
62 * poor stacks do signed 16bit maths!
64 #define MAX_TCP_WINDOW 32767U
66 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
67 #define TCP_MIN_MSS 88U
69 /* The least MTU to use for probing */
70 #define TCP_BASE_MSS 1024
72 /* probing interval, default to 10 minutes as per RFC4821 */
73 #define TCP_PROBE_INTERVAL 600
75 /* Specify interval when tcp mtu probing will stop */
76 #define TCP_PROBE_THRESHOLD 8
78 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
79 #define TCP_FASTRETRANS_THRESH 3
81 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
82 #define TCP_MAX_QUICKACKS 16U
85 #define TCP_URG_VALID 0x0100
86 #define TCP_URG_NOTYET 0x0200
87 #define TCP_URG_READ 0x0400
89 #define TCP_RETR1 3 /*
90 * This is how many retries it does before it
91 * tries to figure out if the gateway is
92 * down. Minimal RFC value is 3; it corresponds
93 * to ~3sec-8min depending on RTO.
96 #define TCP_RETR2 15 /*
97 * This should take at least
98 * 90 minutes to time out.
99 * RFC1122 says that the limit is 100 sec.
100 * 15 is ~13-30min depending on RTO.
103 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
104 * when active opening a connection.
105 * RFC1122 says the minimum retry MUST
106 * be at least 180secs. Nevertheless
107 * this value is corresponding to
108 * 63secs of retransmission with the
109 * current initial RTO.
112 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
113 * when passive opening a connection.
114 * This is corresponding to 31secs of
115 * retransmission with the current
119 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
120 * state, about 60 seconds */
121 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
122 /* BSD style FIN_WAIT2 deadlock breaker.
123 * It used to be 3min, new value is 60sec,
124 * to combine FIN-WAIT-2 timeout with
128 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
130 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
131 #define TCP_ATO_MIN ((unsigned)(HZ/25))
133 #define TCP_DELACK_MIN 4U
134 #define TCP_ATO_MIN 4U
136 #define TCP_RTO_MAX ((unsigned)(120*HZ))
137 #define TCP_RTO_MIN ((unsigned)(HZ/5))
138 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
139 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
140 * used as a fallback RTO for the
141 * initial data transmission if no
142 * valid RTT sample has been acquired,
143 * most likely due to retrans in 3WHS.
146 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
147 * for local resources.
150 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
151 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
152 #define TCP_KEEPALIVE_INTVL (75*HZ)
154 #define MAX_TCP_KEEPIDLE 32767
155 #define MAX_TCP_KEEPINTVL 32767
156 #define MAX_TCP_KEEPCNT 127
157 #define MAX_TCP_SYNCNT 127
159 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
161 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
162 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
163 * after this time. It should be equal
164 * (or greater than) TCP_TIMEWAIT_LEN
165 * to provide reliability equal to one
166 * provided by timewait state.
168 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
169 * timestamps. It must be less than
170 * minimal timewait lifetime.
176 #define TCPOPT_NOP 1 /* Padding */
177 #define TCPOPT_EOL 0 /* End of options */
178 #define TCPOPT_MSS 2 /* Segment size negotiating */
179 #define TCPOPT_WINDOW 3 /* Window scaling */
180 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
181 #define TCPOPT_SACK 5 /* SACK Block */
182 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
183 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
184 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
185 #define TCPOPT_EXP 254 /* Experimental */
186 /* Magic number to be after the option value for sharing TCP
187 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
189 #define TCPOPT_FASTOPEN_MAGIC 0xF989
195 #define TCPOLEN_MSS 4
196 #define TCPOLEN_WINDOW 3
197 #define TCPOLEN_SACK_PERM 2
198 #define TCPOLEN_TIMESTAMP 10
199 #define TCPOLEN_MD5SIG 18
200 #define TCPOLEN_FASTOPEN_BASE 2
201 #define TCPOLEN_EXP_FASTOPEN_BASE 4
203 /* But this is what stacks really send out. */
204 #define TCPOLEN_TSTAMP_ALIGNED 12
205 #define TCPOLEN_WSCALE_ALIGNED 4
206 #define TCPOLEN_SACKPERM_ALIGNED 4
207 #define TCPOLEN_SACK_BASE 2
208 #define TCPOLEN_SACK_BASE_ALIGNED 4
209 #define TCPOLEN_SACK_PERBLOCK 8
210 #define TCPOLEN_MD5SIG_ALIGNED 20
211 #define TCPOLEN_MSS_ALIGNED 4
213 /* Flags in tp->nonagle */
214 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
215 #define TCP_NAGLE_CORK 2 /* Socket is corked */
216 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
218 /* TCP thin-stream limits */
219 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
221 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
222 #define TCP_INIT_CWND 10
224 /* Bit Flags for sysctl_tcp_fastopen */
225 #define TFO_CLIENT_ENABLE 1
226 #define TFO_SERVER_ENABLE 2
227 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
229 /* Accept SYN data w/o any cookie option */
230 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
232 /* Force enable TFO on all listeners, i.e., not requiring the
233 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
235 #define TFO_SERVER_WO_SOCKOPT1 0x400
236 #define TFO_SERVER_WO_SOCKOPT2 0x800
238 extern struct inet_timewait_death_row tcp_death_row;
240 /* sysctl variables for tcp */
241 extern int sysctl_tcp_timestamps;
242 extern int sysctl_tcp_window_scaling;
243 extern int sysctl_tcp_sack;
244 extern int sysctl_tcp_fin_timeout;
245 extern int sysctl_tcp_keepalive_time;
246 extern int sysctl_tcp_keepalive_probes;
247 extern int sysctl_tcp_keepalive_intvl;
248 extern int sysctl_tcp_syn_retries;
249 extern int sysctl_tcp_synack_retries;
250 extern int sysctl_tcp_retries1;
251 extern int sysctl_tcp_retries2;
252 extern int sysctl_tcp_orphan_retries;
253 extern int sysctl_tcp_syncookies;
254 extern int sysctl_tcp_fastopen;
255 extern int sysctl_tcp_retrans_collapse;
256 extern int sysctl_tcp_stdurg;
257 extern int sysctl_tcp_rfc1337;
258 extern int sysctl_tcp_abort_on_overflow;
259 extern int sysctl_tcp_max_orphans;
260 extern int sysctl_tcp_fack;
261 extern int sysctl_tcp_reordering;
262 extern int sysctl_tcp_max_reordering;
263 extern int sysctl_tcp_dsack;
264 extern long sysctl_tcp_mem[3];
265 extern int sysctl_tcp_wmem[3];
266 extern int sysctl_tcp_rmem[3];
267 extern int sysctl_tcp_app_win;
268 extern int sysctl_tcp_adv_win_scale;
269 extern int sysctl_tcp_tw_reuse;
270 extern int sysctl_tcp_frto;
271 extern int sysctl_tcp_low_latency;
272 extern int sysctl_tcp_nometrics_save;
273 extern int sysctl_tcp_moderate_rcvbuf;
274 extern int sysctl_tcp_tso_win_divisor;
275 extern int sysctl_tcp_workaround_signed_windows;
276 extern int sysctl_tcp_slow_start_after_idle;
277 extern int sysctl_tcp_thin_linear_timeouts;
278 extern int sysctl_tcp_thin_dupack;
279 extern int sysctl_tcp_early_retrans;
280 extern int sysctl_tcp_limit_output_bytes;
281 extern int sysctl_tcp_challenge_ack_limit;
282 extern unsigned int sysctl_tcp_notsent_lowat;
283 extern int sysctl_tcp_min_tso_segs;
284 extern int sysctl_tcp_min_rtt_wlen;
285 extern int sysctl_tcp_autocorking;
286 extern int sysctl_tcp_invalid_ratelimit;
287 extern int sysctl_tcp_pacing_ss_ratio;
288 extern int sysctl_tcp_pacing_ca_ratio;
290 extern atomic_long_t tcp_memory_allocated;
291 extern struct percpu_counter tcp_sockets_allocated;
292 extern int tcp_memory_pressure;
294 /* optimized version of sk_under_memory_pressure() for TCP sockets */
295 static inline bool tcp_under_memory_pressure(const struct sock *sk)
297 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
298 return !!sk->sk_cgrp->memory_pressure;
300 return tcp_memory_pressure;
303 * The next routines deal with comparing 32 bit unsigned ints
304 * and worry about wraparound (automatic with unsigned arithmetic).
307 static inline bool before(__u32 seq1, __u32 seq2)
309 return (__s32)(seq1-seq2) < 0;
311 #define after(seq2, seq1) before(seq1, seq2)
313 /* is s2<=s1<=s3 ? */
314 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
316 return seq3 - seq2 >= seq1 - seq2;
319 static inline bool tcp_out_of_memory(struct sock *sk)
321 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
322 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
327 void sk_forced_mem_schedule(struct sock *sk, int size);
329 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
331 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
332 int orphans = percpu_counter_read_positive(ocp);
334 if (orphans << shift > sysctl_tcp_max_orphans) {
335 orphans = percpu_counter_sum_positive(ocp);
336 if (orphans << shift > sysctl_tcp_max_orphans)
342 bool tcp_check_oom(struct sock *sk, int shift);
345 extern struct proto tcp_prot;
347 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
348 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
349 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
350 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
351 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
353 void tcp_tasklet_init(void);
355 void tcp_v4_err(struct sk_buff *skb, u32);
357 void tcp_shutdown(struct sock *sk, int how);
359 void tcp_v4_early_demux(struct sk_buff *skb);
360 int tcp_v4_rcv(struct sk_buff *skb);
362 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
363 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
364 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
366 void tcp_release_cb(struct sock *sk);
367 void tcp_wfree(struct sk_buff *skb);
368 void tcp_write_timer_handler(struct sock *sk);
369 void tcp_delack_timer_handler(struct sock *sk);
370 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
371 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
372 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
373 const struct tcphdr *th, unsigned int len);
374 void tcp_rcv_space_adjust(struct sock *sk);
375 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
376 void tcp_twsk_destructor(struct sock *sk);
377 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
378 struct pipe_inode_info *pipe, size_t len,
381 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
382 static inline void tcp_dec_quickack_mode(struct sock *sk,
383 const unsigned int pkts)
385 struct inet_connection_sock *icsk = inet_csk(sk);
387 if (icsk->icsk_ack.quick) {
388 if (pkts >= icsk->icsk_ack.quick) {
389 icsk->icsk_ack.quick = 0;
390 /* Leaving quickack mode we deflate ATO. */
391 icsk->icsk_ack.ato = TCP_ATO_MIN;
393 icsk->icsk_ack.quick -= pkts;
398 #define TCP_ECN_QUEUE_CWR 2
399 #define TCP_ECN_DEMAND_CWR 4
400 #define TCP_ECN_SEEN 8
410 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
412 const struct tcphdr *th);
413 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
414 struct request_sock *req, bool fastopen);
415 int tcp_child_process(struct sock *parent, struct sock *child,
416 struct sk_buff *skb);
417 void tcp_enter_loss(struct sock *sk);
418 void tcp_clear_retrans(struct tcp_sock *tp);
419 void tcp_update_metrics(struct sock *sk);
420 void tcp_init_metrics(struct sock *sk);
421 void tcp_metrics_init(void);
422 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
423 bool paws_check, bool timestamps);
424 bool tcp_remember_stamp(struct sock *sk);
425 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
426 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
427 void tcp_disable_fack(struct tcp_sock *tp);
428 void tcp_close(struct sock *sk, long timeout);
429 void tcp_init_sock(struct sock *sk);
430 unsigned int tcp_poll(struct file *file, struct socket *sock,
431 struct poll_table_struct *wait);
432 int tcp_getsockopt(struct sock *sk, int level, int optname,
433 char __user *optval, int __user *optlen);
434 int tcp_setsockopt(struct sock *sk, int level, int optname,
435 char __user *optval, unsigned int optlen);
436 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
437 char __user *optval, int __user *optlen);
438 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
439 char __user *optval, unsigned int optlen);
440 void tcp_set_keepalive(struct sock *sk, int val);
441 void tcp_syn_ack_timeout(const struct request_sock *req);
442 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
443 int flags, int *addr_len);
444 void tcp_parse_options(const struct sk_buff *skb,
445 struct tcp_options_received *opt_rx,
446 int estab, struct tcp_fastopen_cookie *foc);
447 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
450 * TCP v4 functions exported for the inet6 API
453 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
454 void tcp_v4_mtu_reduced(struct sock *sk);
455 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
456 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
457 struct sock *tcp_create_openreq_child(const struct sock *sk,
458 struct request_sock *req,
459 struct sk_buff *skb);
460 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
461 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
462 struct request_sock *req,
463 struct dst_entry *dst,
464 struct request_sock *req_unhash,
466 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
467 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
468 int tcp_connect(struct sock *sk);
469 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
470 struct request_sock *req,
471 struct tcp_fastopen_cookie *foc,
473 int tcp_disconnect(struct sock *sk, int flags);
475 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
476 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
477 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
479 /* From syncookies.c */
480 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
481 struct request_sock *req,
482 struct dst_entry *dst);
483 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
485 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
486 #ifdef CONFIG_SYN_COOKIES
488 /* Syncookies use a monotonic timer which increments every 60 seconds.
489 * This counter is used both as a hash input and partially encoded into
490 * the cookie value. A cookie is only validated further if the delta
491 * between the current counter value and the encoded one is less than this,
492 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
493 * the counter advances immediately after a cookie is generated).
495 #define MAX_SYNCOOKIE_AGE 2
496 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
497 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
499 /* syncookies: remember time of last synqueue overflow
500 * But do not dirty this field too often (once per second is enough)
501 * It is racy as we do not hold a lock, but race is very minor.
503 static inline void tcp_synq_overflow(const struct sock *sk)
505 unsigned long last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
506 unsigned long now = jiffies;
508 if (!time_between32(now, last_overflow, last_overflow + HZ))
509 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
512 /* syncookies: no recent synqueue overflow on this listening socket? */
513 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
515 unsigned long last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
517 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
518 * then we're under synflood. However, we have to use
519 * 'last_overflow - HZ' as lower bound. That's because a concurrent
520 * tcp_synq_overflow() could update .ts_recent_stamp after we read
521 * jiffies but before we store .ts_recent_stamp into last_overflow,
522 * which could lead to rejecting a valid syncookie.
524 return !time_between32(jiffies, last_overflow - HZ,
525 last_overflow + TCP_SYNCOOKIE_VALID);
528 static inline u32 tcp_cookie_time(void)
530 u64 val = get_jiffies_64();
532 do_div(val, TCP_SYNCOOKIE_PERIOD);
536 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
538 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
539 __u32 cookie_init_timestamp(struct request_sock *req);
540 bool cookie_timestamp_decode(struct tcp_options_received *opt);
541 bool cookie_ecn_ok(const struct tcp_options_received *opt,
542 const struct net *net, const struct dst_entry *dst);
544 /* From net/ipv6/syncookies.c */
545 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
547 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
549 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
550 const struct tcphdr *th, u16 *mssp);
551 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
555 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
557 bool tcp_may_send_now(struct sock *sk);
558 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
559 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
560 void tcp_retransmit_timer(struct sock *sk);
561 void tcp_xmit_retransmit_queue(struct sock *);
562 void tcp_simple_retransmit(struct sock *);
563 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
564 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
566 void tcp_send_probe0(struct sock *);
567 void tcp_send_partial(struct sock *);
568 int tcp_write_wakeup(struct sock *, int mib);
569 void tcp_send_fin(struct sock *sk);
570 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
571 int tcp_send_synack(struct sock *);
572 void tcp_push_one(struct sock *, unsigned int mss_now);
573 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
574 void tcp_send_ack(struct sock *sk);
575 void tcp_send_delayed_ack(struct sock *sk);
576 void tcp_send_loss_probe(struct sock *sk);
577 bool tcp_schedule_loss_probe(struct sock *sk);
580 void tcp_resume_early_retransmit(struct sock *sk);
581 void tcp_rearm_rto(struct sock *sk);
582 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
583 void tcp_reset(struct sock *sk);
584 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
587 void tcp_init_xmit_timers(struct sock *);
588 static inline void tcp_clear_xmit_timers(struct sock *sk)
590 inet_csk_clear_xmit_timers(sk);
593 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
594 unsigned int tcp_current_mss(struct sock *sk);
596 /* Bound MSS / TSO packet size with the half of the window */
597 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
601 /* When peer uses tiny windows, there is no use in packetizing
602 * to sub-MSS pieces for the sake of SWS or making sure there
603 * are enough packets in the pipe for fast recovery.
605 * On the other hand, for extremely large MSS devices, handling
606 * smaller than MSS windows in this way does make sense.
608 if (tp->max_window >= 512)
609 cutoff = (tp->max_window >> 1);
611 cutoff = tp->max_window;
613 if (cutoff && pktsize > cutoff)
614 return max_t(int, cutoff, 68U - tp->tcp_header_len);
620 void tcp_get_info(struct sock *, struct tcp_info *);
622 /* Read 'sendfile()'-style from a TCP socket */
623 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
624 unsigned int, size_t);
625 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
626 sk_read_actor_t recv_actor);
628 void tcp_initialize_rcv_mss(struct sock *sk);
630 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
631 int tcp_mss_to_mtu(struct sock *sk, int mss);
632 void tcp_mtup_init(struct sock *sk);
633 void tcp_init_buffer_space(struct sock *sk);
635 static inline void tcp_bound_rto(const struct sock *sk)
637 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
638 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
641 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
643 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
646 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
648 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
649 ntohl(TCP_FLAG_ACK) |
653 static inline void tcp_fast_path_on(struct tcp_sock *tp)
655 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
658 static inline void tcp_fast_path_check(struct sock *sk)
660 struct tcp_sock *tp = tcp_sk(sk);
662 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
664 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
666 tcp_fast_path_on(tp);
669 /* Compute the actual rto_min value */
670 static inline u32 tcp_rto_min(struct sock *sk)
672 const struct dst_entry *dst = __sk_dst_get(sk);
673 u32 rto_min = TCP_RTO_MIN;
675 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
676 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
680 static inline u32 tcp_rto_min_us(struct sock *sk)
682 return jiffies_to_usecs(tcp_rto_min(sk));
685 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
687 return dst_metric_locked(dst, RTAX_CC_ALGO);
690 /* Minimum RTT in usec. ~0 means not available. */
691 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
693 return tp->rtt_min[0].rtt;
696 /* Compute the actual receive window we are currently advertising.
697 * Rcv_nxt can be after the window if our peer push more data
698 * than the offered window.
700 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
702 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
709 /* Choose a new window, without checks for shrinking, and without
710 * scaling applied to the result. The caller does these things
711 * if necessary. This is a "raw" window selection.
713 u32 __tcp_select_window(struct sock *sk);
715 void tcp_send_window_probe(struct sock *sk);
717 /* TCP timestamps are only 32-bits, this causes a slight
718 * complication on 64-bit systems since we store a snapshot
719 * of jiffies in the buffer control blocks below. We decided
720 * to use only the low 32-bits of jiffies and hide the ugly
721 * casts with the following macro.
723 #define tcp_time_stamp ((__u32)(jiffies))
725 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
727 return skb->skb_mstamp.stamp_jiffies;
731 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
733 #define TCPHDR_FIN 0x01
734 #define TCPHDR_SYN 0x02
735 #define TCPHDR_RST 0x04
736 #define TCPHDR_PSH 0x08
737 #define TCPHDR_ACK 0x10
738 #define TCPHDR_URG 0x20
739 #define TCPHDR_ECE 0x40
740 #define TCPHDR_CWR 0x80
742 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
744 /* This is what the send packet queuing engine uses to pass
745 * TCP per-packet control information to the transmission code.
746 * We also store the host-order sequence numbers in here too.
747 * This is 44 bytes if IPV6 is enabled.
748 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
751 __u32 seq; /* Starting sequence number */
752 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
754 /* Note : tcp_tw_isn is used in input path only
755 * (isn chosen by tcp_timewait_state_process())
757 * tcp_gso_segs/size are used in write queue only,
758 * cf tcp_skb_pcount()/tcp_skb_mss()
766 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
768 __u8 sacked; /* State flags for SACK/FACK. */
769 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
770 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
771 #define TCPCB_LOST 0x04 /* SKB is lost */
772 #define TCPCB_TAGBITS 0x07 /* All tag bits */
773 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
774 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
775 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
778 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
780 __u32 ack_seq; /* Sequence number ACK'd */
782 struct inet_skb_parm h4;
783 #if IS_ENABLED(CONFIG_IPV6)
784 struct inet6_skb_parm h6;
786 } header; /* For incoming frames */
789 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
792 #if IS_ENABLED(CONFIG_IPV6)
793 /* This is the variant of inet6_iif() that must be used by TCP,
794 * as TCP moves IP6CB into a different location in skb->cb[]
796 static inline int tcp_v6_iif(const struct sk_buff *skb)
798 return TCP_SKB_CB(skb)->header.h6.iif;
802 /* Due to TSO, an SKB can be composed of multiple actual
803 * packets. To keep these tracked properly, we use this.
805 static inline int tcp_skb_pcount(const struct sk_buff *skb)
807 return TCP_SKB_CB(skb)->tcp_gso_segs;
810 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
812 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
815 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
817 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
820 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
821 static inline int tcp_skb_mss(const struct sk_buff *skb)
823 return TCP_SKB_CB(skb)->tcp_gso_size;
826 /* Events passed to congestion control interface */
828 CA_EVENT_TX_START, /* first transmit when no packets in flight */
829 CA_EVENT_CWND_RESTART, /* congestion window restart */
830 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
831 CA_EVENT_LOSS, /* loss timeout */
832 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
833 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
836 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
837 enum tcp_ca_ack_event_flags {
838 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
839 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
840 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
844 * Interface for adding new TCP congestion control handlers
846 #define TCP_CA_NAME_MAX 16
847 #define TCP_CA_MAX 128
848 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
850 #define TCP_CA_UNSPEC 0
852 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
853 #define TCP_CONG_NON_RESTRICTED 0x1
854 /* Requires ECN/ECT set on all packets */
855 #define TCP_CONG_NEEDS_ECN 0x2
859 struct tcp_congestion_ops {
860 struct list_head list;
864 /* initialize private data (optional) */
865 void (*init)(struct sock *sk);
866 /* cleanup private data (optional) */
867 void (*release)(struct sock *sk);
869 /* return slow start threshold (required) */
870 u32 (*ssthresh)(struct sock *sk);
871 /* do new cwnd calculation (required) */
872 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
873 /* call before changing ca_state (optional) */
874 void (*set_state)(struct sock *sk, u8 new_state);
875 /* call when cwnd event occurs (optional) */
876 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
877 /* call when ack arrives (optional) */
878 void (*in_ack_event)(struct sock *sk, u32 flags);
879 /* new value of cwnd after loss (optional) */
880 u32 (*undo_cwnd)(struct sock *sk);
881 /* hook for packet ack accounting (optional) */
882 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
883 /* get info for inet_diag (optional) */
884 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
885 union tcp_cc_info *info);
887 char name[TCP_CA_NAME_MAX];
888 struct module *owner;
891 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
892 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
894 void tcp_assign_congestion_control(struct sock *sk);
895 void tcp_init_congestion_control(struct sock *sk);
896 void tcp_cleanup_congestion_control(struct sock *sk);
897 int tcp_set_default_congestion_control(const char *name);
898 void tcp_get_default_congestion_control(char *name);
899 void tcp_get_available_congestion_control(char *buf, size_t len);
900 void tcp_get_allowed_congestion_control(char *buf, size_t len);
901 int tcp_set_allowed_congestion_control(char *allowed);
902 int tcp_set_congestion_control(struct sock *sk, const char *name);
903 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
904 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
906 u32 tcp_reno_ssthresh(struct sock *sk);
907 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
908 extern struct tcp_congestion_ops tcp_reno;
910 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
911 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
913 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
915 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
921 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
923 const struct inet_connection_sock *icsk = inet_csk(sk);
925 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
928 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
930 struct inet_connection_sock *icsk = inet_csk(sk);
932 if (icsk->icsk_ca_ops->set_state)
933 icsk->icsk_ca_ops->set_state(sk, ca_state);
934 icsk->icsk_ca_state = ca_state;
937 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
939 const struct inet_connection_sock *icsk = inet_csk(sk);
941 if (icsk->icsk_ca_ops->cwnd_event)
942 icsk->icsk_ca_ops->cwnd_event(sk, event);
945 /* These functions determine how the current flow behaves in respect of SACK
946 * handling. SACK is negotiated with the peer, and therefore it can vary
947 * between different flows.
949 * tcp_is_sack - SACK enabled
950 * tcp_is_reno - No SACK
951 * tcp_is_fack - FACK enabled, implies SACK enabled
953 static inline int tcp_is_sack(const struct tcp_sock *tp)
955 return tp->rx_opt.sack_ok;
958 static inline bool tcp_is_reno(const struct tcp_sock *tp)
960 return !tcp_is_sack(tp);
963 static inline bool tcp_is_fack(const struct tcp_sock *tp)
965 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
968 static inline void tcp_enable_fack(struct tcp_sock *tp)
970 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
973 /* TCP early-retransmit (ER) is similar to but more conservative than
974 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
976 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
978 tp->do_early_retrans = sysctl_tcp_early_retrans &&
979 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
980 sysctl_tcp_reordering == 3;
983 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
985 tp->do_early_retrans = 0;
988 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
990 return tp->sacked_out + tp->lost_out;
993 /* This determines how many packets are "in the network" to the best
994 * of our knowledge. In many cases it is conservative, but where
995 * detailed information is available from the receiver (via SACK
996 * blocks etc.) we can make more aggressive calculations.
998 * Use this for decisions involving congestion control, use just
999 * tp->packets_out to determine if the send queue is empty or not.
1001 * Read this equation as:
1003 * "Packets sent once on transmission queue" MINUS
1004 * "Packets left network, but not honestly ACKed yet" PLUS
1005 * "Packets fast retransmitted"
1007 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1009 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1012 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1014 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1016 return tp->snd_cwnd < tp->snd_ssthresh;
1019 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1021 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1024 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1026 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1027 (1 << inet_csk(sk)->icsk_ca_state);
1030 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1031 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1034 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1036 const struct tcp_sock *tp = tcp_sk(sk);
1038 if (tcp_in_cwnd_reduction(sk))
1039 return tp->snd_ssthresh;
1041 return max(tp->snd_ssthresh,
1042 ((tp->snd_cwnd >> 1) +
1043 (tp->snd_cwnd >> 2)));
1046 /* Use define here intentionally to get WARN_ON location shown at the caller */
1047 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1049 void tcp_enter_cwr(struct sock *sk);
1050 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1052 /* The maximum number of MSS of available cwnd for which TSO defers
1053 * sending if not using sysctl_tcp_tso_win_divisor.
1055 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1060 /* Slow start with delack produces 3 packets of burst, so that
1061 * it is safe "de facto". This will be the default - same as
1062 * the default reordering threshold - but if reordering increases,
1063 * we must be able to allow cwnd to burst at least this much in order
1064 * to not pull it back when holes are filled.
1066 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1068 return tp->reordering;
1071 /* Returns end sequence number of the receiver's advertised window */
1072 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1074 return tp->snd_una + tp->snd_wnd;
1077 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1078 * flexible approach. The RFC suggests cwnd should not be raised unless
1079 * it was fully used previously. And that's exactly what we do in
1080 * congestion avoidance mode. But in slow start we allow cwnd to grow
1081 * as long as the application has used half the cwnd.
1083 * cwnd is 10 (IW10), but application sends 9 frames.
1084 * We allow cwnd to reach 18 when all frames are ACKed.
1085 * This check is safe because it's as aggressive as slow start which already
1086 * risks 100% overshoot. The advantage is that we discourage application to
1087 * either send more filler packets or data to artificially blow up the cwnd
1088 * usage, and allow application-limited process to probe bw more aggressively.
1090 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1092 const struct tcp_sock *tp = tcp_sk(sk);
1094 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1095 if (tcp_in_slow_start(tp))
1096 return tp->snd_cwnd < 2 * tp->max_packets_out;
1098 return tp->is_cwnd_limited;
1101 /* Something is really bad, we could not queue an additional packet,
1102 * because qdisc is full or receiver sent a 0 window.
1103 * We do not want to add fuel to the fire, or abort too early,
1104 * so make sure the timer we arm now is at least 200ms in the future,
1105 * regardless of current icsk_rto value (as it could be ~2ms)
1107 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1109 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1112 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1113 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1114 unsigned long max_when)
1116 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1118 return (unsigned long)min_t(u64, when, max_when);
1121 static inline void tcp_check_probe_timer(struct sock *sk)
1123 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1124 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1125 tcp_probe0_base(sk), TCP_RTO_MAX);
1128 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1133 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1139 * Calculate(/check) TCP checksum
1141 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1142 __be32 daddr, __wsum base)
1144 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1147 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1149 return __skb_checksum_complete(skb);
1152 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1154 return !skb_csum_unnecessary(skb) &&
1155 __tcp_checksum_complete(skb);
1158 /* Prequeue for VJ style copy to user, combined with checksumming. */
1160 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1162 tp->ucopy.task = NULL;
1164 tp->ucopy.memory = 0;
1165 skb_queue_head_init(&tp->ucopy.prequeue);
1168 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1169 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1174 static const char *statename[]={
1175 "Unused","Established","Syn Sent","Syn Recv",
1176 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1177 "Close Wait","Last ACK","Listen","Closing"
1180 void tcp_set_state(struct sock *sk, int state);
1182 void tcp_done(struct sock *sk);
1184 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1187 rx_opt->num_sacks = 0;
1190 u32 tcp_default_init_rwnd(u32 mss);
1191 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1193 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1195 struct tcp_sock *tp = tcp_sk(sk);
1198 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1200 delta = tcp_time_stamp - tp->lsndtime;
1201 if (delta > inet_csk(sk)->icsk_rto)
1202 tcp_cwnd_restart(sk, delta);
1205 /* Determine a window scaling and initial window to offer. */
1206 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1207 __u32 *window_clamp, int wscale_ok,
1208 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1210 static inline int tcp_win_from_space(int space)
1212 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1214 return tcp_adv_win_scale <= 0 ?
1215 (space>>(-tcp_adv_win_scale)) :
1216 space - (space>>tcp_adv_win_scale);
1219 /* Note: caller must be prepared to deal with negative returns */
1220 static inline int tcp_space(const struct sock *sk)
1222 return tcp_win_from_space(sk->sk_rcvbuf -
1223 atomic_read(&sk->sk_rmem_alloc));
1226 static inline int tcp_full_space(const struct sock *sk)
1228 return tcp_win_from_space(sk->sk_rcvbuf);
1231 extern void tcp_openreq_init_rwin(struct request_sock *req,
1232 const struct sock *sk_listener,
1233 const struct dst_entry *dst);
1235 void tcp_enter_memory_pressure(struct sock *sk);
1237 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1239 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1242 static inline int keepalive_time_when(const struct tcp_sock *tp)
1244 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1247 static inline int keepalive_probes(const struct tcp_sock *tp)
1249 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1252 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1254 const struct inet_connection_sock *icsk = &tp->inet_conn;
1256 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1257 tcp_time_stamp - tp->rcv_tstamp);
1260 static inline int tcp_fin_time(const struct sock *sk)
1262 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1263 const int rto = inet_csk(sk)->icsk_rto;
1265 if (fin_timeout < (rto << 2) - (rto >> 1))
1266 fin_timeout = (rto << 2) - (rto >> 1);
1271 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1274 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1276 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1279 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1280 * then following tcp messages have valid values. Ignore 0 value,
1281 * or else 'negative' tsval might forbid us to accept their packets.
1283 if (!rx_opt->ts_recent)
1288 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1291 if (tcp_paws_check(rx_opt, 0))
1294 /* RST segments are not recommended to carry timestamp,
1295 and, if they do, it is recommended to ignore PAWS because
1296 "their cleanup function should take precedence over timestamps."
1297 Certainly, it is mistake. It is necessary to understand the reasons
1298 of this constraint to relax it: if peer reboots, clock may go
1299 out-of-sync and half-open connections will not be reset.
1300 Actually, the problem would be not existing if all
1301 the implementations followed draft about maintaining clock
1302 via reboots. Linux-2.2 DOES NOT!
1304 However, we can relax time bounds for RST segments to MSL.
1306 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1311 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1312 int mib_idx, u32 *last_oow_ack_time);
1314 static inline void tcp_mib_init(struct net *net)
1317 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1318 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1319 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1320 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1324 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1326 tp->lost_skb_hint = NULL;
1329 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1331 tcp_clear_retrans_hints_partial(tp);
1332 tp->retransmit_skb_hint = NULL;
1338 union tcp_md5_addr {
1340 #if IS_ENABLED(CONFIG_IPV6)
1345 /* - key database */
1346 struct tcp_md5sig_key {
1347 struct hlist_node node;
1349 u8 family; /* AF_INET or AF_INET6 */
1350 union tcp_md5_addr addr;
1351 u8 key[TCP_MD5SIG_MAXKEYLEN];
1352 struct rcu_head rcu;
1356 struct tcp_md5sig_info {
1357 struct hlist_head head;
1358 struct rcu_head rcu;
1361 /* - pseudo header */
1362 struct tcp4_pseudohdr {
1370 struct tcp6_pseudohdr {
1371 struct in6_addr saddr;
1372 struct in6_addr daddr;
1374 __be32 protocol; /* including padding */
1377 union tcp_md5sum_block {
1378 struct tcp4_pseudohdr ip4;
1379 #if IS_ENABLED(CONFIG_IPV6)
1380 struct tcp6_pseudohdr ip6;
1384 /* - pool: digest algorithm, hash description and scratch buffer */
1385 struct tcp_md5sig_pool {
1386 struct hash_desc md5_desc;
1387 union tcp_md5sum_block md5_blk;
1391 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1392 const struct sock *sk, const struct sk_buff *skb);
1393 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1394 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1395 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1397 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1398 const struct sock *addr_sk);
1400 #ifdef CONFIG_TCP_MD5SIG
1401 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1402 const union tcp_md5_addr *addr,
1404 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1406 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1407 const union tcp_md5_addr *addr,
1412 #define tcp_twsk_md5_key(twsk) NULL
1415 bool tcp_alloc_md5sig_pool(void);
1417 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1418 static inline void tcp_put_md5sig_pool(void)
1423 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1424 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1425 unsigned int header_len);
1426 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1427 const struct tcp_md5sig_key *key);
1429 /* From tcp_fastopen.c */
1430 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1431 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1432 unsigned long *last_syn_loss);
1433 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1434 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1436 struct tcp_fastopen_request {
1437 /* Fast Open cookie. Size 0 means a cookie request */
1438 struct tcp_fastopen_cookie cookie;
1439 struct msghdr *data; /* data in MSG_FASTOPEN */
1441 int copied; /* queued in tcp_connect() */
1443 void tcp_free_fastopen_req(struct tcp_sock *tp);
1445 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1446 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1447 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1448 struct request_sock *req,
1449 struct tcp_fastopen_cookie *foc,
1450 struct dst_entry *dst);
1451 void tcp_fastopen_init_key_once(bool publish);
1452 #define TCP_FASTOPEN_KEY_LENGTH 16
1454 static inline void tcp_init_send_head(struct sock *sk)
1456 sk->sk_send_head = NULL;
1459 /* Fastopen key context */
1460 struct tcp_fastopen_context {
1461 struct crypto_cipher *tfm;
1462 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1463 struct rcu_head rcu;
1466 /* write queue abstraction */
1467 static inline void tcp_write_queue_purge(struct sock *sk)
1469 struct sk_buff *skb;
1471 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1472 sk_wmem_free_skb(sk, skb);
1474 tcp_clear_all_retrans_hints(tcp_sk(sk));
1475 tcp_init_send_head(sk);
1476 inet_csk(sk)->icsk_backoff = 0;
1479 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1481 return skb_peek(&sk->sk_write_queue);
1484 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1486 return skb_peek_tail(&sk->sk_write_queue);
1489 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1490 const struct sk_buff *skb)
1492 return skb_queue_next(&sk->sk_write_queue, skb);
1495 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1496 const struct sk_buff *skb)
1498 return skb_queue_prev(&sk->sk_write_queue, skb);
1501 #define tcp_for_write_queue(skb, sk) \
1502 skb_queue_walk(&(sk)->sk_write_queue, skb)
1504 #define tcp_for_write_queue_from(skb, sk) \
1505 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1507 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1508 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1510 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1512 return sk->sk_send_head;
1515 static inline bool tcp_skb_is_last(const struct sock *sk,
1516 const struct sk_buff *skb)
1518 return skb_queue_is_last(&sk->sk_write_queue, skb);
1521 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1523 if (tcp_skb_is_last(sk, skb))
1524 sk->sk_send_head = NULL;
1526 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1529 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1531 if (sk->sk_send_head == skb_unlinked)
1532 sk->sk_send_head = NULL;
1533 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1534 tcp_sk(sk)->highest_sack = NULL;
1537 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1539 struct sk_buff *skb = tcp_write_queue_head(sk);
1541 if (skb == tcp_send_head(sk))
1547 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1549 struct sk_buff *skb = tcp_send_head(sk);
1551 /* empty retransmit queue, for example due to zero window */
1552 if (skb == tcp_write_queue_head(sk))
1555 return skb ? tcp_write_queue_prev(sk, skb) : tcp_write_queue_tail(sk);
1558 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1560 __skb_queue_tail(&sk->sk_write_queue, skb);
1563 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1565 __tcp_add_write_queue_tail(sk, skb);
1567 /* Queue it, remembering where we must start sending. */
1568 if (sk->sk_send_head == NULL) {
1569 sk->sk_send_head = skb;
1571 if (tcp_sk(sk)->highest_sack == NULL)
1572 tcp_sk(sk)->highest_sack = skb;
1576 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1578 __skb_queue_head(&sk->sk_write_queue, skb);
1581 /* Insert buff after skb on the write queue of sk. */
1582 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1583 struct sk_buff *buff,
1586 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1589 /* Insert new before skb on the write queue of sk. */
1590 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1591 struct sk_buff *skb,
1594 __skb_queue_before(&sk->sk_write_queue, skb, new);
1596 if (sk->sk_send_head == skb)
1597 sk->sk_send_head = new;
1600 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1602 __skb_unlink(skb, &sk->sk_write_queue);
1605 static inline bool tcp_write_queue_empty(struct sock *sk)
1607 return skb_queue_empty(&sk->sk_write_queue);
1610 static inline void tcp_push_pending_frames(struct sock *sk)
1612 if (tcp_send_head(sk)) {
1613 struct tcp_sock *tp = tcp_sk(sk);
1615 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1619 /* Start sequence of the skb just after the highest skb with SACKed
1620 * bit, valid only if sacked_out > 0 or when the caller has ensured
1621 * validity by itself.
1623 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1625 if (!tp->sacked_out)
1628 if (tp->highest_sack == NULL)
1631 return TCP_SKB_CB(tp->highest_sack)->seq;
1634 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1636 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1637 tcp_write_queue_next(sk, skb);
1640 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1642 return tcp_sk(sk)->highest_sack;
1645 static inline void tcp_highest_sack_reset(struct sock *sk)
1647 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1650 /* Called when old skb is about to be deleted and replaced by new skb */
1651 static inline void tcp_highest_sack_replace(struct sock *sk,
1652 struct sk_buff *old,
1653 struct sk_buff *new)
1655 if (old == tcp_highest_sack(sk))
1656 tcp_sk(sk)->highest_sack = new;
1659 /* Determines whether this is a thin stream (which may suffer from
1660 * increased latency). Used to trigger latency-reducing mechanisms.
1662 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1664 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1668 enum tcp_seq_states {
1669 TCP_SEQ_STATE_LISTENING,
1670 TCP_SEQ_STATE_ESTABLISHED,
1673 int tcp_seq_open(struct inode *inode, struct file *file);
1675 struct tcp_seq_afinfo {
1678 const struct file_operations *seq_fops;
1679 struct seq_operations seq_ops;
1682 struct tcp_iter_state {
1683 struct seq_net_private p;
1685 enum tcp_seq_states state;
1686 struct sock *syn_wait_sk;
1687 int bucket, offset, sbucket, num;
1691 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1692 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1694 extern struct request_sock_ops tcp_request_sock_ops;
1695 extern struct request_sock_ops tcp6_request_sock_ops;
1697 void tcp_v4_destroy_sock(struct sock *sk);
1699 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1700 netdev_features_t features);
1701 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1702 int tcp_gro_complete(struct sk_buff *skb);
1704 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1706 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1708 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1711 static inline bool tcp_stream_memory_free(const struct sock *sk)
1713 const struct tcp_sock *tp = tcp_sk(sk);
1714 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1716 return notsent_bytes < tcp_notsent_lowat(tp);
1719 #ifdef CONFIG_PROC_FS
1720 int tcp4_proc_init(void);
1721 void tcp4_proc_exit(void);
1724 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1725 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1726 const struct tcp_request_sock_ops *af_ops,
1727 struct sock *sk, struct sk_buff *skb);
1729 /* TCP af-specific functions */
1730 struct tcp_sock_af_ops {
1731 #ifdef CONFIG_TCP_MD5SIG
1732 struct tcp_md5sig_key *(*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);
1738 int (*md5_parse)(struct sock *sk,
1739 char __user *optval,
1744 struct tcp_request_sock_ops {
1746 #ifdef CONFIG_TCP_MD5SIG
1747 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1748 const struct sock *addr_sk);
1749 int (*calc_md5_hash) (char *location,
1750 const struct tcp_md5sig_key *md5,
1751 const struct sock *sk,
1752 const struct sk_buff *skb);
1754 void (*init_req)(struct request_sock *req,
1755 const struct sock *sk_listener,
1756 struct sk_buff *skb);
1757 #ifdef CONFIG_SYN_COOKIES
1758 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1761 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1762 const struct request_sock *req,
1764 __u32 (*init_seq)(const struct sk_buff *skb);
1765 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1766 struct flowi *fl, struct request_sock *req,
1767 struct tcp_fastopen_cookie *foc,
1771 #ifdef CONFIG_SYN_COOKIES
1772 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1773 const struct sock *sk, struct sk_buff *skb,
1776 tcp_synq_overflow(sk);
1777 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1778 return ops->cookie_init_seq(skb, mss);
1781 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1782 const struct sock *sk, struct sk_buff *skb,
1789 int tcpv4_offload_init(void);
1791 void tcp_v4_init(void);
1792 void tcp_init(void);
1794 /* tcp_recovery.c */
1796 /* Flags to enable various loss recovery features. See below */
1797 extern int sysctl_tcp_recovery;
1799 /* Use TCP RACK to detect (some) tail and retransmit losses */
1800 #define TCP_RACK_LOST_RETRANS 0x1
1802 extern int tcp_rack_mark_lost(struct sock *sk);
1804 extern void tcp_rack_advance(struct tcp_sock *tp,
1805 const struct skb_mstamp *xmit_time, u8 sacked);
1808 * Save and compile IPv4 options, return a pointer to it
1810 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1812 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1813 struct ip_options_rcu *dopt = NULL;
1816 int opt_size = sizeof(*dopt) + opt->optlen;
1818 dopt = kmalloc(opt_size, GFP_ATOMIC);
1819 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1827 /* locally generated TCP pure ACKs have skb->truesize == 2
1828 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1829 * This is much faster than dissecting the packet to find out.
1830 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1832 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1834 return skb->truesize == 2;
1837 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)