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 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
145 * for local resources.
148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
150 #define TCP_KEEPALIVE_INTVL (75*HZ)
152 #define MAX_TCP_KEEPIDLE 32767
153 #define MAX_TCP_KEEPINTVL 32767
154 #define MAX_TCP_KEEPCNT 127
155 #define MAX_TCP_SYNCNT 127
157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
161 * after this time. It should be equal
162 * (or greater than) TCP_TIMEWAIT_LEN
163 * to provide reliability equal to one
164 * provided by timewait state.
166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
167 * timestamps. It must be less than
168 * minimal timewait lifetime.
174 #define TCPOPT_NOP 1 /* Padding */
175 #define TCPOPT_EOL 0 /* End of options */
176 #define TCPOPT_MSS 2 /* Segment size negotiating */
177 #define TCPOPT_WINDOW 3 /* Window scaling */
178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
179 #define TCPOPT_SACK 5 /* SACK Block */
180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
183 #define TCPOPT_EXP 254 /* Experimental */
184 /* Magic number to be after the option value for sharing TCP
185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
187 #define TCPOPT_FASTOPEN_MAGIC 0xF989
193 #define TCPOLEN_MSS 4
194 #define TCPOLEN_WINDOW 3
195 #define TCPOLEN_SACK_PERM 2
196 #define TCPOLEN_TIMESTAMP 10
197 #define TCPOLEN_MD5SIG 18
198 #define TCPOLEN_FASTOPEN_BASE 2
199 #define TCPOLEN_EXP_FASTOPEN_BASE 4
201 /* But this is what stacks really send out. */
202 #define TCPOLEN_TSTAMP_ALIGNED 12
203 #define TCPOLEN_WSCALE_ALIGNED 4
204 #define TCPOLEN_SACKPERM_ALIGNED 4
205 #define TCPOLEN_SACK_BASE 2
206 #define TCPOLEN_SACK_BASE_ALIGNED 4
207 #define TCPOLEN_SACK_PERBLOCK 8
208 #define TCPOLEN_MD5SIG_ALIGNED 20
209 #define TCPOLEN_MSS_ALIGNED 4
211 /* Flags in tp->nonagle */
212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
213 #define TCP_NAGLE_CORK 2 /* Socket is corked */
214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
216 /* TCP thin-stream limits */
217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
219 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
220 #define TCP_INIT_CWND 10
222 /* Bit Flags for sysctl_tcp_fastopen */
223 #define TFO_CLIENT_ENABLE 1
224 #define TFO_SERVER_ENABLE 2
225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
227 /* Accept SYN data w/o any cookie option */
228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
230 /* Force enable TFO on all listeners, i.e., not requiring the
231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
233 #define TFO_SERVER_WO_SOCKOPT1 0x400
234 #define TFO_SERVER_WO_SOCKOPT2 0x800
236 extern struct inet_timewait_death_row tcp_death_row;
238 /* sysctl variables for tcp */
239 extern int sysctl_tcp_timestamps;
240 extern int sysctl_tcp_window_scaling;
241 extern int sysctl_tcp_sack;
242 extern int sysctl_tcp_fin_timeout;
243 extern int sysctl_tcp_keepalive_time;
244 extern int sysctl_tcp_keepalive_probes;
245 extern int sysctl_tcp_keepalive_intvl;
246 extern int sysctl_tcp_syn_retries;
247 extern int sysctl_tcp_synack_retries;
248 extern int sysctl_tcp_retries1;
249 extern int sysctl_tcp_retries2;
250 extern int sysctl_tcp_orphan_retries;
251 extern int sysctl_tcp_syncookies;
252 extern int sysctl_tcp_fastopen;
253 extern int sysctl_tcp_retrans_collapse;
254 extern int sysctl_tcp_stdurg;
255 extern int sysctl_tcp_rfc1337;
256 extern int sysctl_tcp_abort_on_overflow;
257 extern int sysctl_tcp_max_orphans;
258 extern int sysctl_tcp_fack;
259 extern int sysctl_tcp_reordering;
260 extern int sysctl_tcp_max_reordering;
261 extern int sysctl_tcp_dsack;
262 extern long sysctl_tcp_mem[3];
263 extern int sysctl_tcp_wmem[3];
264 extern int sysctl_tcp_rmem[3];
265 extern int sysctl_tcp_app_win;
266 extern int sysctl_tcp_adv_win_scale;
267 extern int sysctl_tcp_tw_reuse;
268 extern int sysctl_tcp_frto;
269 extern int sysctl_tcp_low_latency;
270 extern int sysctl_tcp_nometrics_save;
271 extern int sysctl_tcp_moderate_rcvbuf;
272 extern int sysctl_tcp_tso_win_divisor;
273 extern int sysctl_tcp_workaround_signed_windows;
274 extern int sysctl_tcp_slow_start_after_idle;
275 extern int sysctl_tcp_thin_linear_timeouts;
276 extern int sysctl_tcp_thin_dupack;
277 extern int sysctl_tcp_early_retrans;
278 extern int sysctl_tcp_limit_output_bytes;
279 extern int sysctl_tcp_challenge_ack_limit;
280 extern unsigned int sysctl_tcp_notsent_lowat;
281 extern int sysctl_tcp_min_tso_segs;
282 extern int sysctl_tcp_min_rtt_wlen;
283 extern int sysctl_tcp_autocorking;
284 extern int sysctl_tcp_invalid_ratelimit;
285 extern int sysctl_tcp_pacing_ss_ratio;
286 extern int sysctl_tcp_pacing_ca_ratio;
287 extern int sysctl_tcp_default_init_rwnd;
289 extern atomic_long_t tcp_memory_allocated;
290 extern struct percpu_counter tcp_sockets_allocated;
291 extern int tcp_memory_pressure;
293 /* optimized version of sk_under_memory_pressure() for TCP sockets */
294 static inline bool tcp_under_memory_pressure(const struct sock *sk)
296 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
297 return !!sk->sk_cgrp->memory_pressure;
299 return tcp_memory_pressure;
302 * The next routines deal with comparing 32 bit unsigned ints
303 * and worry about wraparound (automatic with unsigned arithmetic).
306 static inline bool before(__u32 seq1, __u32 seq2)
308 return (__s32)(seq1-seq2) < 0;
310 #define after(seq2, seq1) before(seq1, seq2)
312 /* is s2<=s1<=s3 ? */
313 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
315 return seq3 - seq2 >= seq1 - seq2;
318 static inline bool tcp_out_of_memory(struct sock *sk)
320 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
321 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
326 void sk_forced_mem_schedule(struct sock *sk, int size);
328 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
330 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
331 int orphans = percpu_counter_read_positive(ocp);
333 if (orphans << shift > sysctl_tcp_max_orphans) {
334 orphans = percpu_counter_sum_positive(ocp);
335 if (orphans << shift > sysctl_tcp_max_orphans)
341 bool tcp_check_oom(struct sock *sk, int shift);
344 extern struct proto tcp_prot;
346 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
347 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
348 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
349 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
350 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
352 void tcp_tasklet_init(void);
354 void tcp_v4_err(struct sk_buff *skb, u32);
356 void tcp_shutdown(struct sock *sk, int how);
358 void tcp_v4_early_demux(struct sk_buff *skb);
359 int tcp_v4_rcv(struct sk_buff *skb);
361 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
362 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
363 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
365 void tcp_release_cb(struct sock *sk);
366 void tcp_wfree(struct sk_buff *skb);
367 void tcp_write_timer_handler(struct sock *sk);
368 void tcp_delack_timer_handler(struct sock *sk);
369 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
370 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
371 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
372 const struct tcphdr *th, unsigned int len);
373 void tcp_rcv_space_adjust(struct sock *sk);
374 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
375 void tcp_twsk_destructor(struct sock *sk);
376 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
377 struct pipe_inode_info *pipe, size_t len,
380 static inline void tcp_dec_quickack_mode(struct sock *sk,
381 const unsigned int pkts)
383 struct inet_connection_sock *icsk = inet_csk(sk);
385 if (icsk->icsk_ack.quick) {
386 if (pkts >= icsk->icsk_ack.quick) {
387 icsk->icsk_ack.quick = 0;
388 /* Leaving quickack mode we deflate ATO. */
389 icsk->icsk_ack.ato = TCP_ATO_MIN;
391 icsk->icsk_ack.quick -= pkts;
396 #define TCP_ECN_QUEUE_CWR 2
397 #define TCP_ECN_DEMAND_CWR 4
398 #define TCP_ECN_SEEN 8
408 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
410 const struct tcphdr *th);
411 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
412 struct request_sock *req, bool fastopen);
413 int tcp_child_process(struct sock *parent, struct sock *child,
414 struct sk_buff *skb);
415 void tcp_enter_loss(struct sock *sk);
416 void tcp_clear_retrans(struct tcp_sock *tp);
417 void tcp_update_metrics(struct sock *sk);
418 void tcp_init_metrics(struct sock *sk);
419 void tcp_metrics_init(void);
420 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
421 bool paws_check, bool timestamps);
422 bool tcp_remember_stamp(struct sock *sk);
423 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
424 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
425 void tcp_disable_fack(struct tcp_sock *tp);
426 void tcp_close(struct sock *sk, long timeout);
427 void tcp_init_sock(struct sock *sk);
428 unsigned int tcp_poll(struct file *file, struct socket *sock,
429 struct poll_table_struct *wait);
430 int tcp_getsockopt(struct sock *sk, int level, int optname,
431 char __user *optval, int __user *optlen);
432 int tcp_setsockopt(struct sock *sk, int level, int optname,
433 char __user *optval, unsigned int optlen);
434 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
435 char __user *optval, int __user *optlen);
436 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
437 char __user *optval, unsigned int optlen);
438 void tcp_set_keepalive(struct sock *sk, int val);
439 void tcp_syn_ack_timeout(const struct request_sock *req);
440 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
441 int flags, int *addr_len);
442 void tcp_parse_options(const struct sk_buff *skb,
443 struct tcp_options_received *opt_rx,
444 int estab, struct tcp_fastopen_cookie *foc);
445 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
448 * TCP v4 functions exported for the inet6 API
451 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
452 void tcp_v4_mtu_reduced(struct sock *sk);
453 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
454 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
455 struct sock *tcp_create_openreq_child(const struct sock *sk,
456 struct request_sock *req,
457 struct sk_buff *skb);
458 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
459 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
460 struct request_sock *req,
461 struct dst_entry *dst,
462 struct request_sock *req_unhash,
464 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
465 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
466 int tcp_connect(struct sock *sk);
467 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
468 struct request_sock *req,
469 struct tcp_fastopen_cookie *foc,
471 int tcp_disconnect(struct sock *sk, int flags);
473 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
474 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
475 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
477 /* From syncookies.c */
478 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
479 struct request_sock *req,
480 struct dst_entry *dst);
481 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
483 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
484 #ifdef CONFIG_SYN_COOKIES
486 /* Syncookies use a monotonic timer which increments every 60 seconds.
487 * This counter is used both as a hash input and partially encoded into
488 * the cookie value. A cookie is only validated further if the delta
489 * between the current counter value and the encoded one is less than this,
490 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
491 * the counter advances immediately after a cookie is generated).
493 #define MAX_SYNCOOKIE_AGE 2
494 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
495 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
497 /* syncookies: remember time of last synqueue overflow
498 * But do not dirty this field too often (once per second is enough)
499 * It is racy as we do not hold a lock, but race is very minor.
501 static inline void tcp_synq_overflow(const struct sock *sk)
503 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
504 unsigned long now = jiffies;
506 if (time_after(now, last_overflow + HZ))
507 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
510 /* syncookies: no recent synqueue overflow on this listening socket? */
511 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
513 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
515 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
518 static inline u32 tcp_cookie_time(void)
520 u64 val = get_jiffies_64();
522 do_div(val, TCP_SYNCOOKIE_PERIOD);
526 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
528 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
529 __u32 cookie_init_timestamp(struct request_sock *req);
530 bool cookie_timestamp_decode(struct tcp_options_received *opt);
531 bool cookie_ecn_ok(const struct tcp_options_received *opt,
532 const struct net *net, const struct dst_entry *dst);
534 /* From net/ipv6/syncookies.c */
535 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
537 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
539 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
540 const struct tcphdr *th, u16 *mssp);
541 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
545 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
547 bool tcp_may_send_now(struct sock *sk);
548 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
549 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
550 void tcp_retransmit_timer(struct sock *sk);
551 void tcp_xmit_retransmit_queue(struct sock *);
552 void tcp_simple_retransmit(struct sock *);
553 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
554 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
556 void tcp_send_probe0(struct sock *);
557 void tcp_send_partial(struct sock *);
558 int tcp_write_wakeup(struct sock *, int mib);
559 void tcp_send_fin(struct sock *sk);
560 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
561 int tcp_send_synack(struct sock *);
562 void tcp_push_one(struct sock *, unsigned int mss_now);
563 void tcp_send_ack(struct sock *sk);
564 void tcp_send_delayed_ack(struct sock *sk);
565 void tcp_send_loss_probe(struct sock *sk);
566 bool tcp_schedule_loss_probe(struct sock *sk);
569 void tcp_resume_early_retransmit(struct sock *sk);
570 void tcp_rearm_rto(struct sock *sk);
571 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
572 void tcp_reset(struct sock *sk);
573 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
576 void tcp_init_xmit_timers(struct sock *);
577 static inline void tcp_clear_xmit_timers(struct sock *sk)
579 inet_csk_clear_xmit_timers(sk);
582 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
583 unsigned int tcp_current_mss(struct sock *sk);
585 /* Bound MSS / TSO packet size with the half of the window */
586 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
590 /* When peer uses tiny windows, there is no use in packetizing
591 * to sub-MSS pieces for the sake of SWS or making sure there
592 * are enough packets in the pipe for fast recovery.
594 * On the other hand, for extremely large MSS devices, handling
595 * smaller than MSS windows in this way does make sense.
597 if (tp->max_window >= 512)
598 cutoff = (tp->max_window >> 1);
600 cutoff = tp->max_window;
602 if (cutoff && pktsize > cutoff)
603 return max_t(int, cutoff, 68U - tp->tcp_header_len);
609 void tcp_get_info(struct sock *, struct tcp_info *);
611 /* Read 'sendfile()'-style from a TCP socket */
612 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
613 unsigned int, size_t);
614 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
615 sk_read_actor_t recv_actor);
617 void tcp_initialize_rcv_mss(struct sock *sk);
619 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
620 int tcp_mss_to_mtu(struct sock *sk, int mss);
621 void tcp_mtup_init(struct sock *sk);
622 void tcp_init_buffer_space(struct sock *sk);
624 static inline void tcp_bound_rto(const struct sock *sk)
626 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
627 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
630 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
632 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
635 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
637 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
638 ntohl(TCP_FLAG_ACK) |
642 static inline void tcp_fast_path_on(struct tcp_sock *tp)
644 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
647 static inline void tcp_fast_path_check(struct sock *sk)
649 struct tcp_sock *tp = tcp_sk(sk);
651 if (skb_queue_empty(&tp->out_of_order_queue) &&
653 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
655 tcp_fast_path_on(tp);
658 /* Compute the actual rto_min value */
659 static inline u32 tcp_rto_min(struct sock *sk)
661 const struct dst_entry *dst = __sk_dst_get(sk);
662 u32 rto_min = TCP_RTO_MIN;
664 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
665 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
669 static inline u32 tcp_rto_min_us(struct sock *sk)
671 return jiffies_to_usecs(tcp_rto_min(sk));
674 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
676 return dst_metric_locked(dst, RTAX_CC_ALGO);
679 /* Minimum RTT in usec. ~0 means not available. */
680 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
682 return tp->rtt_min[0].rtt;
685 /* Compute the actual receive window we are currently advertising.
686 * Rcv_nxt can be after the window if our peer push more data
687 * than the offered window.
689 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
691 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
698 /* Choose a new window, without checks for shrinking, and without
699 * scaling applied to the result. The caller does these things
700 * if necessary. This is a "raw" window selection.
702 u32 __tcp_select_window(struct sock *sk);
704 void tcp_send_window_probe(struct sock *sk);
706 /* TCP timestamps are only 32-bits, this causes a slight
707 * complication on 64-bit systems since we store a snapshot
708 * of jiffies in the buffer control blocks below. We decided
709 * to use only the low 32-bits of jiffies and hide the ugly
710 * casts with the following macro.
712 #define tcp_time_stamp ((__u32)(jiffies))
714 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
716 return skb->skb_mstamp.stamp_jiffies;
720 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
722 #define TCPHDR_FIN 0x01
723 #define TCPHDR_SYN 0x02
724 #define TCPHDR_RST 0x04
725 #define TCPHDR_PSH 0x08
726 #define TCPHDR_ACK 0x10
727 #define TCPHDR_URG 0x20
728 #define TCPHDR_ECE 0x40
729 #define TCPHDR_CWR 0x80
731 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
733 /* This is what the send packet queuing engine uses to pass
734 * TCP per-packet control information to the transmission code.
735 * We also store the host-order sequence numbers in here too.
736 * This is 44 bytes if IPV6 is enabled.
737 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
740 __u32 seq; /* Starting sequence number */
741 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
743 /* Note : tcp_tw_isn is used in input path only
744 * (isn chosen by tcp_timewait_state_process())
746 * tcp_gso_segs/size are used in write queue only,
747 * cf tcp_skb_pcount()/tcp_skb_mss()
755 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
757 __u8 sacked; /* State flags for SACK/FACK. */
758 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
759 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
760 #define TCPCB_LOST 0x04 /* SKB is lost */
761 #define TCPCB_TAGBITS 0x07 /* All tag bits */
762 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
763 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
764 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
767 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
769 __u32 ack_seq; /* Sequence number ACK'd */
771 struct inet_skb_parm h4;
772 #if IS_ENABLED(CONFIG_IPV6)
773 struct inet6_skb_parm h6;
775 } header; /* For incoming frames */
778 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
781 #if IS_ENABLED(CONFIG_IPV6)
782 /* This is the variant of inet6_iif() that must be used by TCP,
783 * as TCP moves IP6CB into a different location in skb->cb[]
785 static inline int tcp_v6_iif(const struct sk_buff *skb)
787 return TCP_SKB_CB(skb)->header.h6.iif;
791 /* Due to TSO, an SKB can be composed of multiple actual
792 * packets. To keep these tracked properly, we use this.
794 static inline int tcp_skb_pcount(const struct sk_buff *skb)
796 return TCP_SKB_CB(skb)->tcp_gso_segs;
799 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
801 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
804 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
806 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
809 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
810 static inline int tcp_skb_mss(const struct sk_buff *skb)
812 return TCP_SKB_CB(skb)->tcp_gso_size;
815 /* Events passed to congestion control interface */
817 CA_EVENT_TX_START, /* first transmit when no packets in flight */
818 CA_EVENT_CWND_RESTART, /* congestion window restart */
819 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
820 CA_EVENT_LOSS, /* loss timeout */
821 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
822 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
823 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
824 CA_EVENT_NON_DELAYED_ACK,
827 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
828 enum tcp_ca_ack_event_flags {
829 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
830 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
831 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
835 * Interface for adding new TCP congestion control handlers
837 #define TCP_CA_NAME_MAX 16
838 #define TCP_CA_MAX 128
839 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
841 #define TCP_CA_UNSPEC 0
843 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
844 #define TCP_CONG_NON_RESTRICTED 0x1
845 /* Requires ECN/ECT set on all packets */
846 #define TCP_CONG_NEEDS_ECN 0x2
850 struct tcp_congestion_ops {
851 struct list_head list;
855 /* initialize private data (optional) */
856 void (*init)(struct sock *sk);
857 /* cleanup private data (optional) */
858 void (*release)(struct sock *sk);
860 /* return slow start threshold (required) */
861 u32 (*ssthresh)(struct sock *sk);
862 /* do new cwnd calculation (required) */
863 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
864 /* call before changing ca_state (optional) */
865 void (*set_state)(struct sock *sk, u8 new_state);
866 /* call when cwnd event occurs (optional) */
867 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
868 /* call when ack arrives (optional) */
869 void (*in_ack_event)(struct sock *sk, u32 flags);
870 /* new value of cwnd after loss (optional) */
871 u32 (*undo_cwnd)(struct sock *sk);
872 /* hook for packet ack accounting (optional) */
873 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
874 /* get info for inet_diag (optional) */
875 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
876 union tcp_cc_info *info);
878 char name[TCP_CA_NAME_MAX];
879 struct module *owner;
882 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
883 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
885 void tcp_assign_congestion_control(struct sock *sk);
886 void tcp_init_congestion_control(struct sock *sk);
887 void tcp_cleanup_congestion_control(struct sock *sk);
888 int tcp_set_default_congestion_control(const char *name);
889 void tcp_get_default_congestion_control(char *name);
890 void tcp_get_available_congestion_control(char *buf, size_t len);
891 void tcp_get_allowed_congestion_control(char *buf, size_t len);
892 int tcp_set_allowed_congestion_control(char *allowed);
893 int tcp_set_congestion_control(struct sock *sk, const char *name);
894 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
895 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
897 u32 tcp_reno_ssthresh(struct sock *sk);
898 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
899 extern struct tcp_congestion_ops tcp_reno;
901 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
902 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
904 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
906 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
912 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
914 const struct inet_connection_sock *icsk = inet_csk(sk);
916 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
919 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
921 struct inet_connection_sock *icsk = inet_csk(sk);
923 if (icsk->icsk_ca_ops->set_state)
924 icsk->icsk_ca_ops->set_state(sk, ca_state);
925 icsk->icsk_ca_state = ca_state;
928 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
930 const struct inet_connection_sock *icsk = inet_csk(sk);
932 if (icsk->icsk_ca_ops->cwnd_event)
933 icsk->icsk_ca_ops->cwnd_event(sk, event);
936 /* These functions determine how the current flow behaves in respect of SACK
937 * handling. SACK is negotiated with the peer, and therefore it can vary
938 * between different flows.
940 * tcp_is_sack - SACK enabled
941 * tcp_is_reno - No SACK
942 * tcp_is_fack - FACK enabled, implies SACK enabled
944 static inline int tcp_is_sack(const struct tcp_sock *tp)
946 return tp->rx_opt.sack_ok;
949 static inline bool tcp_is_reno(const struct tcp_sock *tp)
951 return !tcp_is_sack(tp);
954 static inline bool tcp_is_fack(const struct tcp_sock *tp)
956 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
959 static inline void tcp_enable_fack(struct tcp_sock *tp)
961 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
964 /* TCP early-retransmit (ER) is similar to but more conservative than
965 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
967 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
969 tp->do_early_retrans = sysctl_tcp_early_retrans &&
970 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
971 sysctl_tcp_reordering == 3;
974 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
976 tp->do_early_retrans = 0;
979 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
981 return tp->sacked_out + tp->lost_out;
984 /* This determines how many packets are "in the network" to the best
985 * of our knowledge. In many cases it is conservative, but where
986 * detailed information is available from the receiver (via SACK
987 * blocks etc.) we can make more aggressive calculations.
989 * Use this for decisions involving congestion control, use just
990 * tp->packets_out to determine if the send queue is empty or not.
992 * Read this equation as:
994 * "Packets sent once on transmission queue" MINUS
995 * "Packets left network, but not honestly ACKed yet" PLUS
996 * "Packets fast retransmitted"
998 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1000 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1003 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1005 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1007 return tp->snd_cwnd < tp->snd_ssthresh;
1010 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1012 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1015 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1017 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1018 (1 << inet_csk(sk)->icsk_ca_state);
1021 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1022 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1025 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1027 const struct tcp_sock *tp = tcp_sk(sk);
1029 if (tcp_in_cwnd_reduction(sk))
1030 return tp->snd_ssthresh;
1032 return max(tp->snd_ssthresh,
1033 ((tp->snd_cwnd >> 1) +
1034 (tp->snd_cwnd >> 2)));
1037 /* Use define here intentionally to get WARN_ON location shown at the caller */
1038 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1040 void tcp_enter_cwr(struct sock *sk);
1041 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1043 /* The maximum number of MSS of available cwnd for which TSO defers
1044 * sending if not using sysctl_tcp_tso_win_divisor.
1046 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1051 /* Slow start with delack produces 3 packets of burst, so that
1052 * it is safe "de facto". This will be the default - same as
1053 * the default reordering threshold - but if reordering increases,
1054 * we must be able to allow cwnd to burst at least this much in order
1055 * to not pull it back when holes are filled.
1057 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1059 return tp->reordering;
1062 /* Returns end sequence number of the receiver's advertised window */
1063 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1065 return tp->snd_una + tp->snd_wnd;
1068 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1069 * flexible approach. The RFC suggests cwnd should not be raised unless
1070 * it was fully used previously. And that's exactly what we do in
1071 * congestion avoidance mode. But in slow start we allow cwnd to grow
1072 * as long as the application has used half the cwnd.
1074 * cwnd is 10 (IW10), but application sends 9 frames.
1075 * We allow cwnd to reach 18 when all frames are ACKed.
1076 * This check is safe because it's as aggressive as slow start which already
1077 * risks 100% overshoot. The advantage is that we discourage application to
1078 * either send more filler packets or data to artificially blow up the cwnd
1079 * usage, and allow application-limited process to probe bw more aggressively.
1081 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1083 const struct tcp_sock *tp = tcp_sk(sk);
1085 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1086 if (tcp_in_slow_start(tp))
1087 return tp->snd_cwnd < 2 * tp->max_packets_out;
1089 return tp->is_cwnd_limited;
1092 /* Something is really bad, we could not queue an additional packet,
1093 * because qdisc is full or receiver sent a 0 window.
1094 * We do not want to add fuel to the fire, or abort too early,
1095 * so make sure the timer we arm now is at least 200ms in the future,
1096 * regardless of current icsk_rto value (as it could be ~2ms)
1098 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1100 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1103 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1104 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1105 unsigned long max_when)
1107 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1109 return (unsigned long)min_t(u64, when, max_when);
1112 static inline void tcp_check_probe_timer(struct sock *sk)
1114 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1115 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1116 tcp_probe0_base(sk), TCP_RTO_MAX);
1119 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1124 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1130 * Calculate(/check) TCP checksum
1132 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1133 __be32 daddr, __wsum base)
1135 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1138 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1140 return __skb_checksum_complete(skb);
1143 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1145 return !skb_csum_unnecessary(skb) &&
1146 __tcp_checksum_complete(skb);
1149 /* Prequeue for VJ style copy to user, combined with checksumming. */
1151 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1153 tp->ucopy.task = NULL;
1155 tp->ucopy.memory = 0;
1156 skb_queue_head_init(&tp->ucopy.prequeue);
1159 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1160 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1165 static const char *statename[]={
1166 "Unused","Established","Syn Sent","Syn Recv",
1167 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1168 "Close Wait","Last ACK","Listen","Closing"
1171 void tcp_set_state(struct sock *sk, int state);
1173 void tcp_done(struct sock *sk);
1175 int tcp_abort(struct sock *sk, int err);
1177 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1180 rx_opt->num_sacks = 0;
1183 u32 tcp_default_init_rwnd(u32 mss);
1184 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1186 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1188 struct tcp_sock *tp = tcp_sk(sk);
1191 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1193 delta = tcp_time_stamp - tp->lsndtime;
1194 if (delta > inet_csk(sk)->icsk_rto)
1195 tcp_cwnd_restart(sk, delta);
1198 /* Determine a window scaling and initial window to offer. */
1199 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1200 __u32 *window_clamp, int wscale_ok,
1201 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1203 static inline int tcp_win_from_space(int space)
1205 return sysctl_tcp_adv_win_scale<=0 ?
1206 (space>>(-sysctl_tcp_adv_win_scale)) :
1207 space - (space>>sysctl_tcp_adv_win_scale);
1210 /* Note: caller must be prepared to deal with negative returns */
1211 static inline int tcp_space(const struct sock *sk)
1213 return tcp_win_from_space(sk->sk_rcvbuf -
1214 atomic_read(&sk->sk_rmem_alloc));
1217 static inline int tcp_full_space(const struct sock *sk)
1219 return tcp_win_from_space(sk->sk_rcvbuf);
1222 extern void tcp_openreq_init_rwin(struct request_sock *req,
1223 const struct sock *sk_listener,
1224 const struct dst_entry *dst);
1226 void tcp_enter_memory_pressure(struct sock *sk);
1228 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1230 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1233 static inline int keepalive_time_when(const struct tcp_sock *tp)
1235 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1238 static inline int keepalive_probes(const struct tcp_sock *tp)
1240 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1243 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1245 const struct inet_connection_sock *icsk = &tp->inet_conn;
1247 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1248 tcp_time_stamp - tp->rcv_tstamp);
1251 static inline int tcp_fin_time(const struct sock *sk)
1253 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1254 const int rto = inet_csk(sk)->icsk_rto;
1256 if (fin_timeout < (rto << 2) - (rto >> 1))
1257 fin_timeout = (rto << 2) - (rto >> 1);
1262 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1265 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1267 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1270 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1271 * then following tcp messages have valid values. Ignore 0 value,
1272 * or else 'negative' tsval might forbid us to accept their packets.
1274 if (!rx_opt->ts_recent)
1279 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1282 if (tcp_paws_check(rx_opt, 0))
1285 /* RST segments are not recommended to carry timestamp,
1286 and, if they do, it is recommended to ignore PAWS because
1287 "their cleanup function should take precedence over timestamps."
1288 Certainly, it is mistake. It is necessary to understand the reasons
1289 of this constraint to relax it: if peer reboots, clock may go
1290 out-of-sync and half-open connections will not be reset.
1291 Actually, the problem would be not existing if all
1292 the implementations followed draft about maintaining clock
1293 via reboots. Linux-2.2 DOES NOT!
1295 However, we can relax time bounds for RST segments to MSL.
1297 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1302 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1303 int mib_idx, u32 *last_oow_ack_time);
1305 static inline void tcp_mib_init(struct net *net)
1308 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1309 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1310 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1311 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1315 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1317 tp->lost_skb_hint = NULL;
1320 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1322 tcp_clear_retrans_hints_partial(tp);
1323 tp->retransmit_skb_hint = NULL;
1329 union tcp_md5_addr {
1331 #if IS_ENABLED(CONFIG_IPV6)
1336 /* - key database */
1337 struct tcp_md5sig_key {
1338 struct hlist_node node;
1340 u8 family; /* AF_INET or AF_INET6 */
1341 union tcp_md5_addr addr;
1342 u8 key[TCP_MD5SIG_MAXKEYLEN];
1343 struct rcu_head rcu;
1347 struct tcp_md5sig_info {
1348 struct hlist_head head;
1349 struct rcu_head rcu;
1352 /* - pseudo header */
1353 struct tcp4_pseudohdr {
1361 struct tcp6_pseudohdr {
1362 struct in6_addr saddr;
1363 struct in6_addr daddr;
1365 __be32 protocol; /* including padding */
1368 union tcp_md5sum_block {
1369 struct tcp4_pseudohdr ip4;
1370 #if IS_ENABLED(CONFIG_IPV6)
1371 struct tcp6_pseudohdr ip6;
1375 /* - pool: digest algorithm, hash description and scratch buffer */
1376 struct tcp_md5sig_pool {
1377 struct hash_desc md5_desc;
1378 union tcp_md5sum_block md5_blk;
1382 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1383 const struct sock *sk, const struct sk_buff *skb);
1384 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1385 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1386 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1388 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1389 const struct sock *addr_sk);
1391 #ifdef CONFIG_TCP_MD5SIG
1392 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1393 const union tcp_md5_addr *addr,
1395 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1397 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1398 const union tcp_md5_addr *addr,
1403 #define tcp_twsk_md5_key(twsk) NULL
1406 bool tcp_alloc_md5sig_pool(void);
1408 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1409 static inline void tcp_put_md5sig_pool(void)
1414 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1415 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1416 unsigned int header_len);
1417 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1418 const struct tcp_md5sig_key *key);
1420 /* From tcp_fastopen.c */
1421 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1422 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1423 unsigned long *last_syn_loss);
1424 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1425 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1427 struct tcp_fastopen_request {
1428 /* Fast Open cookie. Size 0 means a cookie request */
1429 struct tcp_fastopen_cookie cookie;
1430 struct msghdr *data; /* data in MSG_FASTOPEN */
1432 int copied; /* queued in tcp_connect() */
1434 void tcp_free_fastopen_req(struct tcp_sock *tp);
1436 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1437 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1438 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1439 struct request_sock *req,
1440 struct tcp_fastopen_cookie *foc,
1441 struct dst_entry *dst);
1442 void tcp_fastopen_init_key_once(bool publish);
1443 #define TCP_FASTOPEN_KEY_LENGTH 16
1445 /* Fastopen key context */
1446 struct tcp_fastopen_context {
1447 struct crypto_cipher *tfm;
1448 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1449 struct rcu_head rcu;
1452 /* write queue abstraction */
1453 static inline void tcp_write_queue_purge(struct sock *sk)
1455 struct sk_buff *skb;
1457 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1458 sk_wmem_free_skb(sk, skb);
1460 tcp_clear_all_retrans_hints(tcp_sk(sk));
1463 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1465 return skb_peek(&sk->sk_write_queue);
1468 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1470 return skb_peek_tail(&sk->sk_write_queue);
1473 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1474 const struct sk_buff *skb)
1476 return skb_queue_next(&sk->sk_write_queue, skb);
1479 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1480 const struct sk_buff *skb)
1482 return skb_queue_prev(&sk->sk_write_queue, skb);
1485 #define tcp_for_write_queue(skb, sk) \
1486 skb_queue_walk(&(sk)->sk_write_queue, skb)
1488 #define tcp_for_write_queue_from(skb, sk) \
1489 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1491 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1492 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1494 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1496 return sk->sk_send_head;
1499 static inline bool tcp_skb_is_last(const struct sock *sk,
1500 const struct sk_buff *skb)
1502 return skb_queue_is_last(&sk->sk_write_queue, skb);
1505 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1507 if (tcp_skb_is_last(sk, skb))
1508 sk->sk_send_head = NULL;
1510 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1513 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1515 if (sk->sk_send_head == skb_unlinked)
1516 sk->sk_send_head = NULL;
1517 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1518 tcp_sk(sk)->highest_sack = NULL;
1521 static inline void tcp_init_send_head(struct sock *sk)
1523 sk->sk_send_head = NULL;
1526 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1528 __skb_queue_tail(&sk->sk_write_queue, skb);
1531 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1533 __tcp_add_write_queue_tail(sk, skb);
1535 /* Queue it, remembering where we must start sending. */
1536 if (sk->sk_send_head == NULL) {
1537 sk->sk_send_head = skb;
1539 if (tcp_sk(sk)->highest_sack == NULL)
1540 tcp_sk(sk)->highest_sack = skb;
1544 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1546 __skb_queue_head(&sk->sk_write_queue, skb);
1549 /* Insert buff after skb on the write queue of sk. */
1550 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1551 struct sk_buff *buff,
1554 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1557 /* Insert new before skb on the write queue of sk. */
1558 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1559 struct sk_buff *skb,
1562 __skb_queue_before(&sk->sk_write_queue, skb, new);
1564 if (sk->sk_send_head == skb)
1565 sk->sk_send_head = new;
1568 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1570 __skb_unlink(skb, &sk->sk_write_queue);
1573 static inline bool tcp_write_queue_empty(struct sock *sk)
1575 return skb_queue_empty(&sk->sk_write_queue);
1578 static inline void tcp_push_pending_frames(struct sock *sk)
1580 if (tcp_send_head(sk)) {
1581 struct tcp_sock *tp = tcp_sk(sk);
1583 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1587 /* Start sequence of the skb just after the highest skb with SACKed
1588 * bit, valid only if sacked_out > 0 or when the caller has ensured
1589 * validity by itself.
1591 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1593 if (!tp->sacked_out)
1596 if (tp->highest_sack == NULL)
1599 return TCP_SKB_CB(tp->highest_sack)->seq;
1602 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1604 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1605 tcp_write_queue_next(sk, skb);
1608 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1610 return tcp_sk(sk)->highest_sack;
1613 static inline void tcp_highest_sack_reset(struct sock *sk)
1615 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1618 /* Called when old skb is about to be deleted (to be combined with new skb) */
1619 static inline void tcp_highest_sack_combine(struct sock *sk,
1620 struct sk_buff *old,
1621 struct sk_buff *new)
1623 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1624 tcp_sk(sk)->highest_sack = new;
1627 /* Determines whether this is a thin stream (which may suffer from
1628 * increased latency). Used to trigger latency-reducing mechanisms.
1630 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1632 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1636 enum tcp_seq_states {
1637 TCP_SEQ_STATE_LISTENING,
1638 TCP_SEQ_STATE_ESTABLISHED,
1641 int tcp_seq_open(struct inode *inode, struct file *file);
1643 struct tcp_seq_afinfo {
1646 const struct file_operations *seq_fops;
1647 struct seq_operations seq_ops;
1650 struct tcp_iter_state {
1651 struct seq_net_private p;
1653 enum tcp_seq_states state;
1654 struct sock *syn_wait_sk;
1655 int bucket, offset, sbucket, num;
1659 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1660 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1662 extern struct request_sock_ops tcp_request_sock_ops;
1663 extern struct request_sock_ops tcp6_request_sock_ops;
1665 void tcp_v4_destroy_sock(struct sock *sk);
1667 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1668 netdev_features_t features);
1669 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1670 int tcp_gro_complete(struct sk_buff *skb);
1672 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1674 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1676 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1679 static inline bool tcp_stream_memory_free(const struct sock *sk)
1681 const struct tcp_sock *tp = tcp_sk(sk);
1682 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1684 return notsent_bytes < tcp_notsent_lowat(tp);
1687 #ifdef CONFIG_PROC_FS
1688 int tcp4_proc_init(void);
1689 void tcp4_proc_exit(void);
1692 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1693 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1694 const struct tcp_request_sock_ops *af_ops,
1695 struct sock *sk, struct sk_buff *skb);
1697 /* TCP af-specific functions */
1698 struct tcp_sock_af_ops {
1699 #ifdef CONFIG_TCP_MD5SIG
1700 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1701 const struct sock *addr_sk);
1702 int (*calc_md5_hash)(char *location,
1703 const struct tcp_md5sig_key *md5,
1704 const struct sock *sk,
1705 const struct sk_buff *skb);
1706 int (*md5_parse)(struct sock *sk,
1707 char __user *optval,
1712 struct tcp_request_sock_ops {
1714 #ifdef CONFIG_TCP_MD5SIG
1715 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1716 const struct sock *addr_sk);
1717 int (*calc_md5_hash) (char *location,
1718 const struct tcp_md5sig_key *md5,
1719 const struct sock *sk,
1720 const struct sk_buff *skb);
1722 void (*init_req)(struct request_sock *req,
1723 const struct sock *sk_listener,
1724 struct sk_buff *skb);
1725 #ifdef CONFIG_SYN_COOKIES
1726 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1729 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1730 const struct request_sock *req,
1732 __u32 (*init_seq)(const struct sk_buff *skb);
1733 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1734 struct flowi *fl, struct request_sock *req,
1735 struct tcp_fastopen_cookie *foc,
1739 #ifdef CONFIG_SYN_COOKIES
1740 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1741 const struct sock *sk, struct sk_buff *skb,
1744 tcp_synq_overflow(sk);
1745 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1746 return ops->cookie_init_seq(skb, mss);
1749 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1750 const struct sock *sk, struct sk_buff *skb,
1757 int tcpv4_offload_init(void);
1759 void tcp_v4_init(void);
1760 void tcp_init(void);
1762 /* tcp_recovery.c */
1764 /* Flags to enable various loss recovery features. See below */
1765 extern int sysctl_tcp_recovery;
1767 /* Use TCP RACK to detect (some) tail and retransmit losses */
1768 #define TCP_RACK_LOST_RETRANS 0x1
1770 extern int tcp_rack_mark_lost(struct sock *sk);
1772 extern void tcp_rack_advance(struct tcp_sock *tp,
1773 const struct skb_mstamp *xmit_time, u8 sacked);
1776 * Save and compile IPv4 options, return a pointer to it
1778 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1780 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1781 struct ip_options_rcu *dopt = NULL;
1784 int opt_size = sizeof(*dopt) + opt->optlen;
1786 dopt = kmalloc(opt_size, GFP_ATOMIC);
1787 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1795 /* locally generated TCP pure ACKs have skb->truesize == 2
1796 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1797 * This is much faster than dissecting the packet to find out.
1798 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1800 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1802 return skb->truesize == 2;
1805 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)