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
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;
289 extern int sysctl_tcp_default_init_rwnd;
291 extern atomic_long_t tcp_memory_allocated;
292 extern struct percpu_counter tcp_sockets_allocated;
293 extern int tcp_memory_pressure;
295 /* optimized version of sk_under_memory_pressure() for TCP sockets */
296 static inline bool tcp_under_memory_pressure(const struct sock *sk)
298 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
299 return !!sk->sk_cgrp->memory_pressure;
301 return tcp_memory_pressure;
304 * The next routines deal with comparing 32 bit unsigned ints
305 * and worry about wraparound (automatic with unsigned arithmetic).
308 static inline bool before(__u32 seq1, __u32 seq2)
310 return (__s32)(seq1-seq2) < 0;
312 #define after(seq2, seq1) before(seq1, seq2)
314 /* is s2<=s1<=s3 ? */
315 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
317 return seq3 - seq2 >= seq1 - seq2;
320 static inline bool tcp_out_of_memory(struct sock *sk)
322 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
323 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
328 void sk_forced_mem_schedule(struct sock *sk, int size);
330 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
332 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
333 int orphans = percpu_counter_read_positive(ocp);
335 if (orphans << shift > sysctl_tcp_max_orphans) {
336 orphans = percpu_counter_sum_positive(ocp);
337 if (orphans << shift > sysctl_tcp_max_orphans)
343 bool tcp_check_oom(struct sock *sk, int shift);
346 extern struct proto tcp_prot;
348 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
349 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
350 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
351 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
352 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
354 void tcp_tasklet_init(void);
356 void tcp_v4_err(struct sk_buff *skb, u32);
358 void tcp_shutdown(struct sock *sk, int how);
360 void tcp_v4_early_demux(struct sk_buff *skb);
361 int tcp_v4_rcv(struct sk_buff *skb);
363 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
364 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
365 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
367 void tcp_release_cb(struct sock *sk);
368 void tcp_wfree(struct sk_buff *skb);
369 void tcp_write_timer_handler(struct sock *sk);
370 void tcp_delack_timer_handler(struct sock *sk);
371 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
372 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
373 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
374 const struct tcphdr *th, unsigned int len);
375 void tcp_rcv_space_adjust(struct sock *sk);
376 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
377 void tcp_twsk_destructor(struct sock *sk);
378 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
379 struct pipe_inode_info *pipe, size_t len,
382 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
383 static inline void tcp_dec_quickack_mode(struct sock *sk,
384 const unsigned int pkts)
386 struct inet_connection_sock *icsk = inet_csk(sk);
388 if (icsk->icsk_ack.quick) {
389 if (pkts >= icsk->icsk_ack.quick) {
390 icsk->icsk_ack.quick = 0;
391 /* Leaving quickack mode we deflate ATO. */
392 icsk->icsk_ack.ato = TCP_ATO_MIN;
394 icsk->icsk_ack.quick -= pkts;
399 #define TCP_ECN_QUEUE_CWR 2
400 #define TCP_ECN_DEMAND_CWR 4
401 #define TCP_ECN_SEEN 8
411 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
413 const struct tcphdr *th);
414 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
415 struct request_sock *req, bool fastopen);
416 int tcp_child_process(struct sock *parent, struct sock *child,
417 struct sk_buff *skb);
418 void tcp_enter_loss(struct sock *sk);
419 void tcp_clear_retrans(struct tcp_sock *tp);
420 void tcp_update_metrics(struct sock *sk);
421 void tcp_init_metrics(struct sock *sk);
422 void tcp_metrics_init(void);
423 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
424 bool paws_check, bool timestamps);
425 bool tcp_remember_stamp(struct sock *sk);
426 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
427 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
428 void tcp_disable_fack(struct tcp_sock *tp);
429 void tcp_close(struct sock *sk, long timeout);
430 void tcp_init_sock(struct sock *sk);
431 unsigned int tcp_poll(struct file *file, struct socket *sock,
432 struct poll_table_struct *wait);
433 int tcp_getsockopt(struct sock *sk, int level, int optname,
434 char __user *optval, int __user *optlen);
435 int tcp_setsockopt(struct sock *sk, int level, int optname,
436 char __user *optval, unsigned int optlen);
437 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
438 char __user *optval, int __user *optlen);
439 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
440 char __user *optval, unsigned int optlen);
441 void tcp_set_keepalive(struct sock *sk, int val);
442 void tcp_syn_ack_timeout(const struct request_sock *req);
443 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
444 int flags, int *addr_len);
445 void tcp_parse_options(const struct sk_buff *skb,
446 struct tcp_options_received *opt_rx,
447 int estab, struct tcp_fastopen_cookie *foc);
448 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
451 * TCP v4 functions exported for the inet6 API
454 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
455 void tcp_v4_mtu_reduced(struct sock *sk);
456 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
457 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
458 struct sock *tcp_create_openreq_child(const struct sock *sk,
459 struct request_sock *req,
460 struct sk_buff *skb);
461 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
462 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
463 struct request_sock *req,
464 struct dst_entry *dst,
465 struct request_sock *req_unhash,
467 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
468 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
469 int tcp_connect(struct sock *sk);
470 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
471 struct request_sock *req,
472 struct tcp_fastopen_cookie *foc,
474 int tcp_disconnect(struct sock *sk, int flags);
476 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
477 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
478 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
480 /* From syncookies.c */
481 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
482 struct request_sock *req,
483 struct dst_entry *dst);
484 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
486 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
487 #ifdef CONFIG_SYN_COOKIES
489 /* Syncookies use a monotonic timer which increments every 60 seconds.
490 * This counter is used both as a hash input and partially encoded into
491 * the cookie value. A cookie is only validated further if the delta
492 * between the current counter value and the encoded one is less than this,
493 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
494 * the counter advances immediately after a cookie is generated).
496 #define MAX_SYNCOOKIE_AGE 2
497 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
498 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
500 /* syncookies: remember time of last synqueue overflow
501 * But do not dirty this field too often (once per second is enough)
502 * It is racy as we do not hold a lock, but race is very minor.
504 static inline void tcp_synq_overflow(const struct sock *sk)
506 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
507 unsigned long now = jiffies;
509 if (time_after(now, last_overflow + HZ))
510 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
513 /* syncookies: no recent synqueue overflow on this listening socket? */
514 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
516 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
518 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
521 static inline u32 tcp_cookie_time(void)
523 u64 val = get_jiffies_64();
525 do_div(val, TCP_SYNCOOKIE_PERIOD);
529 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
531 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
532 __u32 cookie_init_timestamp(struct request_sock *req);
533 bool cookie_timestamp_decode(struct tcp_options_received *opt);
534 bool cookie_ecn_ok(const struct tcp_options_received *opt,
535 const struct net *net, const struct dst_entry *dst);
537 /* From net/ipv6/syncookies.c */
538 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
540 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
542 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
543 const struct tcphdr *th, u16 *mssp);
544 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
548 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
550 bool tcp_may_send_now(struct sock *sk);
551 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
552 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
553 void tcp_retransmit_timer(struct sock *sk);
554 void tcp_xmit_retransmit_queue(struct sock *);
555 void tcp_simple_retransmit(struct sock *);
556 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
557 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
559 void tcp_send_probe0(struct sock *);
560 void tcp_send_partial(struct sock *);
561 int tcp_write_wakeup(struct sock *, int mib);
562 void tcp_send_fin(struct sock *sk);
563 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
564 int tcp_send_synack(struct sock *);
565 void tcp_push_one(struct sock *, unsigned int mss_now);
566 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
567 void tcp_send_ack(struct sock *sk);
568 void tcp_send_delayed_ack(struct sock *sk);
569 void tcp_send_loss_probe(struct sock *sk);
570 bool tcp_schedule_loss_probe(struct sock *sk);
573 void tcp_resume_early_retransmit(struct sock *sk);
574 void tcp_rearm_rto(struct sock *sk);
575 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
576 void tcp_reset(struct sock *sk);
577 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
580 void tcp_init_xmit_timers(struct sock *);
581 static inline void tcp_clear_xmit_timers(struct sock *sk)
583 inet_csk_clear_xmit_timers(sk);
586 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
587 unsigned int tcp_current_mss(struct sock *sk);
589 /* Bound MSS / TSO packet size with the half of the window */
590 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
594 /* When peer uses tiny windows, there is no use in packetizing
595 * to sub-MSS pieces for the sake of SWS or making sure there
596 * are enough packets in the pipe for fast recovery.
598 * On the other hand, for extremely large MSS devices, handling
599 * smaller than MSS windows in this way does make sense.
601 if (tp->max_window >= 512)
602 cutoff = (tp->max_window >> 1);
604 cutoff = tp->max_window;
606 if (cutoff && pktsize > cutoff)
607 return max_t(int, cutoff, 68U - tp->tcp_header_len);
613 void tcp_get_info(struct sock *, struct tcp_info *);
615 /* Read 'sendfile()'-style from a TCP socket */
616 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
617 unsigned int, size_t);
618 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
619 sk_read_actor_t recv_actor);
621 void tcp_initialize_rcv_mss(struct sock *sk);
623 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
624 int tcp_mss_to_mtu(struct sock *sk, int mss);
625 void tcp_mtup_init(struct sock *sk);
626 void tcp_init_buffer_space(struct sock *sk);
628 static inline void tcp_bound_rto(const struct sock *sk)
630 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
631 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
634 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
636 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
639 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
641 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
642 ntohl(TCP_FLAG_ACK) |
646 static inline void tcp_fast_path_on(struct tcp_sock *tp)
648 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
651 static inline void tcp_fast_path_check(struct sock *sk)
653 struct tcp_sock *tp = tcp_sk(sk);
655 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
657 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
659 tcp_fast_path_on(tp);
662 /* Compute the actual rto_min value */
663 static inline u32 tcp_rto_min(struct sock *sk)
665 const struct dst_entry *dst = __sk_dst_get(sk);
666 u32 rto_min = TCP_RTO_MIN;
668 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
669 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
673 static inline u32 tcp_rto_min_us(struct sock *sk)
675 return jiffies_to_usecs(tcp_rto_min(sk));
678 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
680 return dst_metric_locked(dst, RTAX_CC_ALGO);
683 /* Minimum RTT in usec. ~0 means not available. */
684 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
686 return tp->rtt_min[0].rtt;
689 /* Compute the actual receive window we are currently advertising.
690 * Rcv_nxt can be after the window if our peer push more data
691 * than the offered window.
693 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
695 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
702 /* Choose a new window, without checks for shrinking, and without
703 * scaling applied to the result. The caller does these things
704 * if necessary. This is a "raw" window selection.
706 u32 __tcp_select_window(struct sock *sk);
708 void tcp_send_window_probe(struct sock *sk);
710 /* TCP timestamps are only 32-bits, this causes a slight
711 * complication on 64-bit systems since we store a snapshot
712 * of jiffies in the buffer control blocks below. We decided
713 * to use only the low 32-bits of jiffies and hide the ugly
714 * casts with the following macro.
716 #define tcp_time_stamp ((__u32)(jiffies))
718 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
720 return skb->skb_mstamp.stamp_jiffies;
724 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
726 #define TCPHDR_FIN 0x01
727 #define TCPHDR_SYN 0x02
728 #define TCPHDR_RST 0x04
729 #define TCPHDR_PSH 0x08
730 #define TCPHDR_ACK 0x10
731 #define TCPHDR_URG 0x20
732 #define TCPHDR_ECE 0x40
733 #define TCPHDR_CWR 0x80
735 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
737 /* This is what the send packet queuing engine uses to pass
738 * TCP per-packet control information to the transmission code.
739 * We also store the host-order sequence numbers in here too.
740 * This is 44 bytes if IPV6 is enabled.
741 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
744 __u32 seq; /* Starting sequence number */
745 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
747 /* Note : tcp_tw_isn is used in input path only
748 * (isn chosen by tcp_timewait_state_process())
750 * tcp_gso_segs/size are used in write queue only,
751 * cf tcp_skb_pcount()/tcp_skb_mss()
759 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
761 __u8 sacked; /* State flags for SACK/FACK. */
762 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
763 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
764 #define TCPCB_LOST 0x04 /* SKB is lost */
765 #define TCPCB_TAGBITS 0x07 /* All tag bits */
766 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
767 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
768 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
771 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
773 __u32 ack_seq; /* Sequence number ACK'd */
775 struct inet_skb_parm h4;
776 #if IS_ENABLED(CONFIG_IPV6)
777 struct inet6_skb_parm h6;
779 } header; /* For incoming frames */
782 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
785 #if IS_ENABLED(CONFIG_IPV6)
786 /* This is the variant of inet6_iif() that must be used by TCP,
787 * as TCP moves IP6CB into a different location in skb->cb[]
789 static inline int tcp_v6_iif(const struct sk_buff *skb)
791 return TCP_SKB_CB(skb)->header.h6.iif;
795 /* Due to TSO, an SKB can be composed of multiple actual
796 * packets. To keep these tracked properly, we use this.
798 static inline int tcp_skb_pcount(const struct sk_buff *skb)
800 return TCP_SKB_CB(skb)->tcp_gso_segs;
803 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
805 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
808 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
810 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
813 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
814 static inline int tcp_skb_mss(const struct sk_buff *skb)
816 return TCP_SKB_CB(skb)->tcp_gso_size;
819 /* Events passed to congestion control interface */
821 CA_EVENT_TX_START, /* first transmit when no packets in flight */
822 CA_EVENT_CWND_RESTART, /* congestion window restart */
823 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
824 CA_EVENT_LOSS, /* loss timeout */
825 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
826 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
829 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
830 enum tcp_ca_ack_event_flags {
831 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
832 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
833 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
837 * Interface for adding new TCP congestion control handlers
839 #define TCP_CA_NAME_MAX 16
840 #define TCP_CA_MAX 128
841 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
843 #define TCP_CA_UNSPEC 0
845 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
846 #define TCP_CONG_NON_RESTRICTED 0x1
847 /* Requires ECN/ECT set on all packets */
848 #define TCP_CONG_NEEDS_ECN 0x2
852 struct tcp_congestion_ops {
853 struct list_head list;
857 /* initialize private data (optional) */
858 void (*init)(struct sock *sk);
859 /* cleanup private data (optional) */
860 void (*release)(struct sock *sk);
862 /* return slow start threshold (required) */
863 u32 (*ssthresh)(struct sock *sk);
864 /* do new cwnd calculation (required) */
865 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
866 /* call before changing ca_state (optional) */
867 void (*set_state)(struct sock *sk, u8 new_state);
868 /* call when cwnd event occurs (optional) */
869 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
870 /* call when ack arrives (optional) */
871 void (*in_ack_event)(struct sock *sk, u32 flags);
872 /* new value of cwnd after loss (optional) */
873 u32 (*undo_cwnd)(struct sock *sk);
874 /* hook for packet ack accounting (optional) */
875 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
876 /* get info for inet_diag (optional) */
877 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
878 union tcp_cc_info *info);
880 char name[TCP_CA_NAME_MAX];
881 struct module *owner;
884 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
885 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
887 void tcp_assign_congestion_control(struct sock *sk);
888 void tcp_init_congestion_control(struct sock *sk);
889 void tcp_cleanup_congestion_control(struct sock *sk);
890 int tcp_set_default_congestion_control(const char *name);
891 void tcp_get_default_congestion_control(char *name);
892 void tcp_get_available_congestion_control(char *buf, size_t len);
893 void tcp_get_allowed_congestion_control(char *buf, size_t len);
894 int tcp_set_allowed_congestion_control(char *allowed);
895 int tcp_set_congestion_control(struct sock *sk, const char *name);
896 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
897 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
899 u32 tcp_reno_ssthresh(struct sock *sk);
900 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
901 extern struct tcp_congestion_ops tcp_reno;
903 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
904 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
906 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
908 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
914 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
916 const struct inet_connection_sock *icsk = inet_csk(sk);
918 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
921 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
923 struct inet_connection_sock *icsk = inet_csk(sk);
925 if (icsk->icsk_ca_ops->set_state)
926 icsk->icsk_ca_ops->set_state(sk, ca_state);
927 icsk->icsk_ca_state = ca_state;
930 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
932 const struct inet_connection_sock *icsk = inet_csk(sk);
934 if (icsk->icsk_ca_ops->cwnd_event)
935 icsk->icsk_ca_ops->cwnd_event(sk, event);
938 /* These functions determine how the current flow behaves in respect of SACK
939 * handling. SACK is negotiated with the peer, and therefore it can vary
940 * between different flows.
942 * tcp_is_sack - SACK enabled
943 * tcp_is_reno - No SACK
944 * tcp_is_fack - FACK enabled, implies SACK enabled
946 static inline int tcp_is_sack(const struct tcp_sock *tp)
948 return tp->rx_opt.sack_ok;
951 static inline bool tcp_is_reno(const struct tcp_sock *tp)
953 return !tcp_is_sack(tp);
956 static inline bool tcp_is_fack(const struct tcp_sock *tp)
958 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
961 static inline void tcp_enable_fack(struct tcp_sock *tp)
963 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
966 /* TCP early-retransmit (ER) is similar to but more conservative than
967 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
969 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
971 tp->do_early_retrans = sysctl_tcp_early_retrans &&
972 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
973 sysctl_tcp_reordering == 3;
976 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
978 tp->do_early_retrans = 0;
981 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
983 return tp->sacked_out + tp->lost_out;
986 /* This determines how many packets are "in the network" to the best
987 * of our knowledge. In many cases it is conservative, but where
988 * detailed information is available from the receiver (via SACK
989 * blocks etc.) we can make more aggressive calculations.
991 * Use this for decisions involving congestion control, use just
992 * tp->packets_out to determine if the send queue is empty or not.
994 * Read this equation as:
996 * "Packets sent once on transmission queue" MINUS
997 * "Packets left network, but not honestly ACKed yet" PLUS
998 * "Packets fast retransmitted"
1000 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1002 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1005 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1007 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1009 return tp->snd_cwnd < tp->snd_ssthresh;
1012 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1014 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1017 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1019 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1020 (1 << inet_csk(sk)->icsk_ca_state);
1023 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1024 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1027 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1029 const struct tcp_sock *tp = tcp_sk(sk);
1031 if (tcp_in_cwnd_reduction(sk))
1032 return tp->snd_ssthresh;
1034 return max(tp->snd_ssthresh,
1035 ((tp->snd_cwnd >> 1) +
1036 (tp->snd_cwnd >> 2)));
1039 /* Use define here intentionally to get WARN_ON location shown at the caller */
1040 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1042 void tcp_enter_cwr(struct sock *sk);
1043 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1045 /* The maximum number of MSS of available cwnd for which TSO defers
1046 * sending if not using sysctl_tcp_tso_win_divisor.
1048 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1053 /* Slow start with delack produces 3 packets of burst, so that
1054 * it is safe "de facto". This will be the default - same as
1055 * the default reordering threshold - but if reordering increases,
1056 * we must be able to allow cwnd to burst at least this much in order
1057 * to not pull it back when holes are filled.
1059 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1061 return tp->reordering;
1064 /* Returns end sequence number of the receiver's advertised window */
1065 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1067 return tp->snd_una + tp->snd_wnd;
1070 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1071 * flexible approach. The RFC suggests cwnd should not be raised unless
1072 * it was fully used previously. And that's exactly what we do in
1073 * congestion avoidance mode. But in slow start we allow cwnd to grow
1074 * as long as the application has used half the cwnd.
1076 * cwnd is 10 (IW10), but application sends 9 frames.
1077 * We allow cwnd to reach 18 when all frames are ACKed.
1078 * This check is safe because it's as aggressive as slow start which already
1079 * risks 100% overshoot. The advantage is that we discourage application to
1080 * either send more filler packets or data to artificially blow up the cwnd
1081 * usage, and allow application-limited process to probe bw more aggressively.
1083 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1085 const struct tcp_sock *tp = tcp_sk(sk);
1087 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1088 if (tcp_in_slow_start(tp))
1089 return tp->snd_cwnd < 2 * tp->max_packets_out;
1091 return tp->is_cwnd_limited;
1094 /* Something is really bad, we could not queue an additional packet,
1095 * because qdisc is full or receiver sent a 0 window.
1096 * We do not want to add fuel to the fire, or abort too early,
1097 * so make sure the timer we arm now is at least 200ms in the future,
1098 * regardless of current icsk_rto value (as it could be ~2ms)
1100 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1102 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1105 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1106 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1107 unsigned long max_when)
1109 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1111 return (unsigned long)min_t(u64, when, max_when);
1114 static inline void tcp_check_probe_timer(struct sock *sk)
1116 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1117 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1118 tcp_probe0_base(sk), TCP_RTO_MAX);
1121 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1126 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1132 * Calculate(/check) TCP checksum
1134 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1135 __be32 daddr, __wsum base)
1137 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1140 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1142 return __skb_checksum_complete(skb);
1145 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1147 return !skb_csum_unnecessary(skb) &&
1148 __tcp_checksum_complete(skb);
1151 /* Prequeue for VJ style copy to user, combined with checksumming. */
1153 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1155 tp->ucopy.task = NULL;
1157 tp->ucopy.memory = 0;
1158 skb_queue_head_init(&tp->ucopy.prequeue);
1161 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1162 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1167 static const char *statename[]={
1168 "Unused","Established","Syn Sent","Syn Recv",
1169 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1170 "Close Wait","Last ACK","Listen","Closing"
1173 void tcp_set_state(struct sock *sk, int state);
1175 void tcp_done(struct sock *sk);
1177 int tcp_abort(struct sock *sk, int err);
1179 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1182 rx_opt->num_sacks = 0;
1185 u32 tcp_default_init_rwnd(u32 mss);
1186 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1188 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1190 struct tcp_sock *tp = tcp_sk(sk);
1193 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1195 delta = tcp_time_stamp - tp->lsndtime;
1196 if (delta > inet_csk(sk)->icsk_rto)
1197 tcp_cwnd_restart(sk, delta);
1200 /* Determine a window scaling and initial window to offer. */
1201 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1202 __u32 *window_clamp, int wscale_ok,
1203 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1205 static inline int tcp_win_from_space(int space)
1207 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1209 return tcp_adv_win_scale <= 0 ?
1210 (space>>(-tcp_adv_win_scale)) :
1211 space - (space>>tcp_adv_win_scale);
1214 /* Note: caller must be prepared to deal with negative returns */
1215 static inline int tcp_space(const struct sock *sk)
1217 return tcp_win_from_space(sk->sk_rcvbuf -
1218 atomic_read(&sk->sk_rmem_alloc));
1221 static inline int tcp_full_space(const struct sock *sk)
1223 return tcp_win_from_space(sk->sk_rcvbuf);
1226 extern void tcp_openreq_init_rwin(struct request_sock *req,
1227 const struct sock *sk_listener,
1228 const struct dst_entry *dst);
1230 void tcp_enter_memory_pressure(struct sock *sk);
1232 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1234 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1237 static inline int keepalive_time_when(const struct tcp_sock *tp)
1239 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1242 static inline int keepalive_probes(const struct tcp_sock *tp)
1244 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1247 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1249 const struct inet_connection_sock *icsk = &tp->inet_conn;
1251 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1252 tcp_time_stamp - tp->rcv_tstamp);
1255 static inline int tcp_fin_time(const struct sock *sk)
1257 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1258 const int rto = inet_csk(sk)->icsk_rto;
1260 if (fin_timeout < (rto << 2) - (rto >> 1))
1261 fin_timeout = (rto << 2) - (rto >> 1);
1266 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1269 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1271 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1274 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1275 * then following tcp messages have valid values. Ignore 0 value,
1276 * or else 'negative' tsval might forbid us to accept their packets.
1278 if (!rx_opt->ts_recent)
1283 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1286 if (tcp_paws_check(rx_opt, 0))
1289 /* RST segments are not recommended to carry timestamp,
1290 and, if they do, it is recommended to ignore PAWS because
1291 "their cleanup function should take precedence over timestamps."
1292 Certainly, it is mistake. It is necessary to understand the reasons
1293 of this constraint to relax it: if peer reboots, clock may go
1294 out-of-sync and half-open connections will not be reset.
1295 Actually, the problem would be not existing if all
1296 the implementations followed draft about maintaining clock
1297 via reboots. Linux-2.2 DOES NOT!
1299 However, we can relax time bounds for RST segments to MSL.
1301 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1306 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1307 int mib_idx, u32 *last_oow_ack_time);
1309 static inline void tcp_mib_init(struct net *net)
1312 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1313 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1314 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1315 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1319 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1321 tp->lost_skb_hint = NULL;
1324 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1326 tcp_clear_retrans_hints_partial(tp);
1327 tp->retransmit_skb_hint = NULL;
1333 union tcp_md5_addr {
1335 #if IS_ENABLED(CONFIG_IPV6)
1340 /* - key database */
1341 struct tcp_md5sig_key {
1342 struct hlist_node node;
1344 u8 family; /* AF_INET or AF_INET6 */
1345 union tcp_md5_addr addr;
1346 u8 key[TCP_MD5SIG_MAXKEYLEN];
1347 struct rcu_head rcu;
1351 struct tcp_md5sig_info {
1352 struct hlist_head head;
1353 struct rcu_head rcu;
1356 /* - pseudo header */
1357 struct tcp4_pseudohdr {
1365 struct tcp6_pseudohdr {
1366 struct in6_addr saddr;
1367 struct in6_addr daddr;
1369 __be32 protocol; /* including padding */
1372 union tcp_md5sum_block {
1373 struct tcp4_pseudohdr ip4;
1374 #if IS_ENABLED(CONFIG_IPV6)
1375 struct tcp6_pseudohdr ip6;
1379 /* - pool: digest algorithm, hash description and scratch buffer */
1380 struct tcp_md5sig_pool {
1381 struct hash_desc md5_desc;
1382 union tcp_md5sum_block md5_blk;
1386 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1387 const struct sock *sk, const struct sk_buff *skb);
1388 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1389 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1390 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1392 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1393 const struct sock *addr_sk);
1395 #ifdef CONFIG_TCP_MD5SIG
1396 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1397 const union tcp_md5_addr *addr,
1399 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1401 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1402 const union tcp_md5_addr *addr,
1407 #define tcp_twsk_md5_key(twsk) NULL
1410 bool tcp_alloc_md5sig_pool(void);
1412 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1413 static inline void tcp_put_md5sig_pool(void)
1418 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1419 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1420 unsigned int header_len);
1421 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1422 const struct tcp_md5sig_key *key);
1424 /* From tcp_fastopen.c */
1425 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1426 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1427 unsigned long *last_syn_loss);
1428 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1429 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1431 struct tcp_fastopen_request {
1432 /* Fast Open cookie. Size 0 means a cookie request */
1433 struct tcp_fastopen_cookie cookie;
1434 struct msghdr *data; /* data in MSG_FASTOPEN */
1436 int copied; /* queued in tcp_connect() */
1438 void tcp_free_fastopen_req(struct tcp_sock *tp);
1440 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1441 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1442 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1443 struct request_sock *req,
1444 struct tcp_fastopen_cookie *foc,
1445 struct dst_entry *dst);
1446 void tcp_fastopen_init_key_once(bool publish);
1447 #define TCP_FASTOPEN_KEY_LENGTH 16
1449 /* Fastopen key context */
1450 struct tcp_fastopen_context {
1451 struct crypto_cipher *tfm;
1452 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1453 struct rcu_head rcu;
1456 /* write queue abstraction */
1457 static inline void tcp_write_queue_purge(struct sock *sk)
1459 struct sk_buff *skb;
1461 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1462 sk_wmem_free_skb(sk, skb);
1464 tcp_clear_all_retrans_hints(tcp_sk(sk));
1465 inet_csk(sk)->icsk_backoff = 0;
1468 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1470 return skb_peek(&sk->sk_write_queue);
1473 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1475 return skb_peek_tail(&sk->sk_write_queue);
1478 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1479 const struct sk_buff *skb)
1481 return skb_queue_next(&sk->sk_write_queue, skb);
1484 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1485 const struct sk_buff *skb)
1487 return skb_queue_prev(&sk->sk_write_queue, skb);
1490 #define tcp_for_write_queue(skb, sk) \
1491 skb_queue_walk(&(sk)->sk_write_queue, skb)
1493 #define tcp_for_write_queue_from(skb, sk) \
1494 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1496 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1497 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1499 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1501 return sk->sk_send_head;
1504 static inline bool tcp_skb_is_last(const struct sock *sk,
1505 const struct sk_buff *skb)
1507 return skb_queue_is_last(&sk->sk_write_queue, skb);
1510 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1512 if (tcp_skb_is_last(sk, skb))
1513 sk->sk_send_head = NULL;
1515 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1518 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1520 if (sk->sk_send_head == skb_unlinked)
1521 sk->sk_send_head = NULL;
1522 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1523 tcp_sk(sk)->highest_sack = NULL;
1526 static inline void tcp_init_send_head(struct sock *sk)
1528 sk->sk_send_head = NULL;
1531 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1533 __skb_queue_tail(&sk->sk_write_queue, skb);
1536 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1538 __tcp_add_write_queue_tail(sk, skb);
1540 /* Queue it, remembering where we must start sending. */
1541 if (sk->sk_send_head == NULL) {
1542 sk->sk_send_head = skb;
1544 if (tcp_sk(sk)->highest_sack == NULL)
1545 tcp_sk(sk)->highest_sack = skb;
1549 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1551 __skb_queue_head(&sk->sk_write_queue, skb);
1554 /* Insert buff after skb on the write queue of sk. */
1555 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1556 struct sk_buff *buff,
1559 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1562 /* Insert new before skb on the write queue of sk. */
1563 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1564 struct sk_buff *skb,
1567 __skb_queue_before(&sk->sk_write_queue, skb, new);
1569 if (sk->sk_send_head == skb)
1570 sk->sk_send_head = new;
1573 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1575 __skb_unlink(skb, &sk->sk_write_queue);
1578 static inline bool tcp_write_queue_empty(struct sock *sk)
1580 return skb_queue_empty(&sk->sk_write_queue);
1583 static inline void tcp_push_pending_frames(struct sock *sk)
1585 if (tcp_send_head(sk)) {
1586 struct tcp_sock *tp = tcp_sk(sk);
1588 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1592 /* Start sequence of the skb just after the highest skb with SACKed
1593 * bit, valid only if sacked_out > 0 or when the caller has ensured
1594 * validity by itself.
1596 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1598 if (!tp->sacked_out)
1601 if (tp->highest_sack == NULL)
1604 return TCP_SKB_CB(tp->highest_sack)->seq;
1607 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1609 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1610 tcp_write_queue_next(sk, skb);
1613 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1615 return tcp_sk(sk)->highest_sack;
1618 static inline void tcp_highest_sack_reset(struct sock *sk)
1620 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1623 /* Called when old skb is about to be deleted and replaced by new skb */
1624 static inline void tcp_highest_sack_replace(struct sock *sk,
1625 struct sk_buff *old,
1626 struct sk_buff *new)
1628 if (old == tcp_highest_sack(sk))
1629 tcp_sk(sk)->highest_sack = new;
1632 /* Determines whether this is a thin stream (which may suffer from
1633 * increased latency). Used to trigger latency-reducing mechanisms.
1635 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1637 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1641 enum tcp_seq_states {
1642 TCP_SEQ_STATE_LISTENING,
1643 TCP_SEQ_STATE_ESTABLISHED,
1646 int tcp_seq_open(struct inode *inode, struct file *file);
1648 struct tcp_seq_afinfo {
1651 const struct file_operations *seq_fops;
1652 struct seq_operations seq_ops;
1655 struct tcp_iter_state {
1656 struct seq_net_private p;
1658 enum tcp_seq_states state;
1659 struct sock *syn_wait_sk;
1660 int bucket, offset, sbucket, num;
1664 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1665 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1667 extern struct request_sock_ops tcp_request_sock_ops;
1668 extern struct request_sock_ops tcp6_request_sock_ops;
1670 void tcp_v4_destroy_sock(struct sock *sk);
1672 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1673 netdev_features_t features);
1674 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1675 int tcp_gro_complete(struct sk_buff *skb);
1677 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1679 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1681 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1684 static inline bool tcp_stream_memory_free(const struct sock *sk)
1686 const struct tcp_sock *tp = tcp_sk(sk);
1687 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1689 return notsent_bytes < tcp_notsent_lowat(tp);
1692 #ifdef CONFIG_PROC_FS
1693 int tcp4_proc_init(void);
1694 void tcp4_proc_exit(void);
1697 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1698 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1699 const struct tcp_request_sock_ops *af_ops,
1700 struct sock *sk, struct sk_buff *skb);
1702 /* TCP af-specific functions */
1703 struct tcp_sock_af_ops {
1704 #ifdef CONFIG_TCP_MD5SIG
1705 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1706 const struct sock *addr_sk);
1707 int (*calc_md5_hash)(char *location,
1708 const struct tcp_md5sig_key *md5,
1709 const struct sock *sk,
1710 const struct sk_buff *skb);
1711 int (*md5_parse)(struct sock *sk,
1712 char __user *optval,
1717 struct tcp_request_sock_ops {
1719 #ifdef CONFIG_TCP_MD5SIG
1720 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1721 const struct sock *addr_sk);
1722 int (*calc_md5_hash) (char *location,
1723 const struct tcp_md5sig_key *md5,
1724 const struct sock *sk,
1725 const struct sk_buff *skb);
1727 void (*init_req)(struct request_sock *req,
1728 const struct sock *sk_listener,
1729 struct sk_buff *skb);
1730 #ifdef CONFIG_SYN_COOKIES
1731 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1734 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1735 const struct request_sock *req,
1737 __u32 (*init_seq)(const struct sk_buff *skb);
1738 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1739 struct flowi *fl, struct request_sock *req,
1740 struct tcp_fastopen_cookie *foc,
1744 #ifdef CONFIG_SYN_COOKIES
1745 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1746 const struct sock *sk, struct sk_buff *skb,
1749 tcp_synq_overflow(sk);
1750 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1751 return ops->cookie_init_seq(skb, mss);
1754 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1755 const struct sock *sk, struct sk_buff *skb,
1762 int tcpv4_offload_init(void);
1764 void tcp_v4_init(void);
1765 void tcp_init(void);
1767 /* tcp_recovery.c */
1769 /* Flags to enable various loss recovery features. See below */
1770 extern int sysctl_tcp_recovery;
1772 /* Use TCP RACK to detect (some) tail and retransmit losses */
1773 #define TCP_RACK_LOST_RETRANS 0x1
1775 extern int tcp_rack_mark_lost(struct sock *sk);
1777 extern void tcp_rack_advance(struct tcp_sock *tp,
1778 const struct skb_mstamp *xmit_time, u8 sacked);
1781 * Save and compile IPv4 options, return a pointer to it
1783 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1785 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1786 struct ip_options_rcu *dopt = NULL;
1789 int opt_size = sizeof(*dopt) + opt->optlen;
1791 dopt = kmalloc(opt_size, GFP_ATOMIC);
1792 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1800 /* locally generated TCP pure ACKs have skb->truesize == 2
1801 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1802 * This is much faster than dissecting the packet to find out.
1803 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1805 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1807 return skb->truesize == 2;
1810 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)