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_autocorking;
283 extern int sysctl_tcp_invalid_ratelimit;
285 extern atomic_long_t tcp_memory_allocated;
286 extern struct percpu_counter tcp_sockets_allocated;
287 extern int tcp_memory_pressure;
289 /* optimized version of sk_under_memory_pressure() for TCP sockets */
290 static inline bool tcp_under_memory_pressure(const struct sock *sk)
292 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
293 return !!sk->sk_cgrp->memory_pressure;
295 return tcp_memory_pressure;
298 * The next routines deal with comparing 32 bit unsigned ints
299 * and worry about wraparound (automatic with unsigned arithmetic).
302 static inline bool before(__u32 seq1, __u32 seq2)
304 return (__s32)(seq1-seq2) < 0;
306 #define after(seq2, seq1) before(seq1, seq2)
308 /* is s2<=s1<=s3 ? */
309 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
311 return seq3 - seq2 >= seq1 - seq2;
314 static inline bool tcp_out_of_memory(struct sock *sk)
316 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
317 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
322 void sk_forced_mem_schedule(struct sock *sk, int size);
324 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
326 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
327 int orphans = percpu_counter_read_positive(ocp);
329 if (orphans << shift > sysctl_tcp_max_orphans) {
330 orphans = percpu_counter_sum_positive(ocp);
331 if (orphans << shift > sysctl_tcp_max_orphans)
337 bool tcp_check_oom(struct sock *sk, int shift);
340 extern struct proto tcp_prot;
342 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
343 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
344 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
345 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
346 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
348 void tcp_tasklet_init(void);
350 void tcp_v4_err(struct sk_buff *skb, u32);
352 void tcp_shutdown(struct sock *sk, int how);
354 void tcp_v4_early_demux(struct sk_buff *skb);
355 int tcp_v4_rcv(struct sk_buff *skb);
357 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
358 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
359 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
361 void tcp_release_cb(struct sock *sk);
362 void tcp_wfree(struct sk_buff *skb);
363 void tcp_write_timer_handler(struct sock *sk);
364 void tcp_delack_timer_handler(struct sock *sk);
365 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
366 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
367 const struct tcphdr *th, unsigned int len);
368 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
369 const struct tcphdr *th, unsigned int len);
370 void tcp_rcv_space_adjust(struct sock *sk);
371 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
372 void tcp_twsk_destructor(struct sock *sk);
373 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
374 struct pipe_inode_info *pipe, size_t len,
377 static inline void tcp_dec_quickack_mode(struct sock *sk,
378 const unsigned int pkts)
380 struct inet_connection_sock *icsk = inet_csk(sk);
382 if (icsk->icsk_ack.quick) {
383 if (pkts >= icsk->icsk_ack.quick) {
384 icsk->icsk_ack.quick = 0;
385 /* Leaving quickack mode we deflate ATO. */
386 icsk->icsk_ack.ato = TCP_ATO_MIN;
388 icsk->icsk_ack.quick -= pkts;
393 #define TCP_ECN_QUEUE_CWR 2
394 #define TCP_ECN_DEMAND_CWR 4
395 #define TCP_ECN_SEEN 8
405 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
407 const struct tcphdr *th);
408 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
409 struct request_sock *req, bool fastopen);
410 int tcp_child_process(struct sock *parent, struct sock *child,
411 struct sk_buff *skb);
412 void tcp_enter_loss(struct sock *sk);
413 void tcp_clear_retrans(struct tcp_sock *tp);
414 void tcp_update_metrics(struct sock *sk);
415 void tcp_init_metrics(struct sock *sk);
416 void tcp_metrics_init(void);
417 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
418 bool paws_check, bool timestamps);
419 bool tcp_remember_stamp(struct sock *sk);
420 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
421 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
422 void tcp_disable_fack(struct tcp_sock *tp);
423 void tcp_close(struct sock *sk, long timeout);
424 void tcp_init_sock(struct sock *sk);
425 unsigned int tcp_poll(struct file *file, struct socket *sock,
426 struct poll_table_struct *wait);
427 int tcp_getsockopt(struct sock *sk, int level, int optname,
428 char __user *optval, int __user *optlen);
429 int tcp_setsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, unsigned int optlen);
431 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
432 char __user *optval, int __user *optlen);
433 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
434 char __user *optval, unsigned int optlen);
435 void tcp_set_keepalive(struct sock *sk, int val);
436 void tcp_syn_ack_timeout(const struct request_sock *req);
437 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
438 int flags, int *addr_len);
439 void tcp_parse_options(const struct sk_buff *skb,
440 struct tcp_options_received *opt_rx,
441 int estab, struct tcp_fastopen_cookie *foc);
442 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
445 * TCP v4 functions exported for the inet6 API
448 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
449 void tcp_v4_mtu_reduced(struct sock *sk);
450 void tcp_req_err(struct sock *sk, u32 seq);
451 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
452 struct sock *tcp_create_openreq_child(struct sock *sk,
453 struct request_sock *req,
454 struct sk_buff *skb);
455 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
456 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
457 struct request_sock *req,
458 struct dst_entry *dst);
459 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
460 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
461 int tcp_connect(struct sock *sk);
462 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
463 struct request_sock *req,
464 struct tcp_fastopen_cookie *foc);
465 int tcp_disconnect(struct sock *sk, int flags);
467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
471 /* From syncookies.c */
472 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
473 struct request_sock *req,
474 struct dst_entry *dst);
475 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
477 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
478 #ifdef CONFIG_SYN_COOKIES
480 /* Syncookies use a monotonic timer which increments every 60 seconds.
481 * This counter is used both as a hash input and partially encoded into
482 * the cookie value. A cookie is only validated further if the delta
483 * between the current counter value and the encoded one is less than this,
484 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
485 * the counter advances immediately after a cookie is generated).
487 #define MAX_SYNCOOKIE_AGE 2
488 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
489 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
491 /* syncookies: remember time of last synqueue overflow
492 * But do not dirty this field too often (once per second is enough)
494 static inline void tcp_synq_overflow(struct sock *sk)
496 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
497 unsigned long now = jiffies;
499 if (time_after(now, last_overflow + HZ))
500 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
503 /* syncookies: no recent synqueue overflow on this listening socket? */
504 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
506 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
508 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
511 static inline u32 tcp_cookie_time(void)
513 u64 val = get_jiffies_64();
515 do_div(val, TCP_SYNCOOKIE_PERIOD);
519 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
521 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
523 __u32 cookie_init_timestamp(struct request_sock *req);
524 bool cookie_timestamp_decode(struct tcp_options_received *opt);
525 bool cookie_ecn_ok(const struct tcp_options_received *opt,
526 const struct net *net, const struct dst_entry *dst);
528 /* From net/ipv6/syncookies.c */
529 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
531 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
533 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
534 const struct tcphdr *th, u16 *mssp);
535 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
540 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
542 bool tcp_may_send_now(struct sock *sk);
543 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
544 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
545 void tcp_retransmit_timer(struct sock *sk);
546 void tcp_xmit_retransmit_queue(struct sock *);
547 void tcp_simple_retransmit(struct sock *);
548 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
549 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
551 void tcp_send_probe0(struct sock *);
552 void tcp_send_partial(struct sock *);
553 int tcp_write_wakeup(struct sock *, int mib);
554 void tcp_send_fin(struct sock *sk);
555 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
556 int tcp_send_synack(struct sock *);
557 void tcp_push_one(struct sock *, unsigned int mss_now);
558 void tcp_send_ack(struct sock *sk);
559 void tcp_send_delayed_ack(struct sock *sk);
560 void tcp_send_loss_probe(struct sock *sk);
561 bool tcp_schedule_loss_probe(struct sock *sk);
564 void tcp_resume_early_retransmit(struct sock *sk);
565 void tcp_rearm_rto(struct sock *sk);
566 void tcp_reset(struct sock *sk);
569 void tcp_init_xmit_timers(struct sock *);
570 static inline void tcp_clear_xmit_timers(struct sock *sk)
572 inet_csk_clear_xmit_timers(sk);
575 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
576 unsigned int tcp_current_mss(struct sock *sk);
578 /* Bound MSS / TSO packet size with the half of the window */
579 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
583 /* When peer uses tiny windows, there is no use in packetizing
584 * to sub-MSS pieces for the sake of SWS or making sure there
585 * are enough packets in the pipe for fast recovery.
587 * On the other hand, for extremely large MSS devices, handling
588 * smaller than MSS windows in this way does make sense.
590 if (tp->max_window >= 512)
591 cutoff = (tp->max_window >> 1);
593 cutoff = tp->max_window;
595 if (cutoff && pktsize > cutoff)
596 return max_t(int, cutoff, 68U - tp->tcp_header_len);
602 void tcp_get_info(struct sock *, struct tcp_info *);
604 /* Read 'sendfile()'-style from a TCP socket */
605 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
606 unsigned int, size_t);
607 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
608 sk_read_actor_t recv_actor);
610 void tcp_initialize_rcv_mss(struct sock *sk);
612 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
613 int tcp_mss_to_mtu(struct sock *sk, int mss);
614 void tcp_mtup_init(struct sock *sk);
615 void tcp_init_buffer_space(struct sock *sk);
617 static inline void tcp_bound_rto(const struct sock *sk)
619 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
620 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
623 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
625 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
628 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
630 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
631 ntohl(TCP_FLAG_ACK) |
635 static inline void tcp_fast_path_on(struct tcp_sock *tp)
637 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
640 static inline void tcp_fast_path_check(struct sock *sk)
642 struct tcp_sock *tp = tcp_sk(sk);
644 if (skb_queue_empty(&tp->out_of_order_queue) &&
646 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
648 tcp_fast_path_on(tp);
651 /* Compute the actual rto_min value */
652 static inline u32 tcp_rto_min(struct sock *sk)
654 const struct dst_entry *dst = __sk_dst_get(sk);
655 u32 rto_min = TCP_RTO_MIN;
657 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
658 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
662 static inline u32 tcp_rto_min_us(struct sock *sk)
664 return jiffies_to_usecs(tcp_rto_min(sk));
667 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
669 return dst_metric_locked(dst, RTAX_CC_ALGO);
672 /* Compute the actual receive window we are currently advertising.
673 * Rcv_nxt can be after the window if our peer push more data
674 * than the offered window.
676 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
678 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
685 /* Choose a new window, without checks for shrinking, and without
686 * scaling applied to the result. The caller does these things
687 * if necessary. This is a "raw" window selection.
689 u32 __tcp_select_window(struct sock *sk);
691 void tcp_send_window_probe(struct sock *sk);
693 /* TCP timestamps are only 32-bits, this causes a slight
694 * complication on 64-bit systems since we store a snapshot
695 * of jiffies in the buffer control blocks below. We decided
696 * to use only the low 32-bits of jiffies and hide the ugly
697 * casts with the following macro.
699 #define tcp_time_stamp ((__u32)(jiffies))
701 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
703 return skb->skb_mstamp.stamp_jiffies;
707 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
709 #define TCPHDR_FIN 0x01
710 #define TCPHDR_SYN 0x02
711 #define TCPHDR_RST 0x04
712 #define TCPHDR_PSH 0x08
713 #define TCPHDR_ACK 0x10
714 #define TCPHDR_URG 0x20
715 #define TCPHDR_ECE 0x40
716 #define TCPHDR_CWR 0x80
718 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
720 /* This is what the send packet queuing engine uses to pass
721 * TCP per-packet control information to the transmission code.
722 * We also store the host-order sequence numbers in here too.
723 * This is 44 bytes if IPV6 is enabled.
724 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
727 __u32 seq; /* Starting sequence number */
728 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
730 /* Note : tcp_tw_isn is used in input path only
731 * (isn chosen by tcp_timewait_state_process())
733 * tcp_gso_segs is used in write queue only,
734 * cf tcp_skb_pcount()
739 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
741 __u8 sacked; /* State flags for SACK/FACK. */
742 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
743 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
744 #define TCPCB_LOST 0x04 /* SKB is lost */
745 #define TCPCB_TAGBITS 0x07 /* All tag bits */
746 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
747 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
748 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
751 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
753 __u32 ack_seq; /* Sequence number ACK'd */
755 struct inet_skb_parm h4;
756 #if IS_ENABLED(CONFIG_IPV6)
757 struct inet6_skb_parm h6;
759 } header; /* For incoming frames */
762 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
765 #if IS_ENABLED(CONFIG_IPV6)
766 /* This is the variant of inet6_iif() that must be used by TCP,
767 * as TCP moves IP6CB into a different location in skb->cb[]
769 static inline int tcp_v6_iif(const struct sk_buff *skb)
771 return TCP_SKB_CB(skb)->header.h6.iif;
775 /* Due to TSO, an SKB can be composed of multiple actual
776 * packets. To keep these tracked properly, we use this.
778 static inline int tcp_skb_pcount(const struct sk_buff *skb)
780 return TCP_SKB_CB(skb)->tcp_gso_segs;
783 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
785 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
788 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
790 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
793 /* This is valid iff tcp_skb_pcount() > 1. */
794 static inline int tcp_skb_mss(const struct sk_buff *skb)
796 return skb_shinfo(skb)->gso_size;
799 /* Events passed to congestion control interface */
801 CA_EVENT_TX_START, /* first transmit when no packets in flight */
802 CA_EVENT_CWND_RESTART, /* congestion window restart */
803 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
804 CA_EVENT_LOSS, /* loss timeout */
805 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
806 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
807 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
808 CA_EVENT_NON_DELAYED_ACK,
811 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
812 enum tcp_ca_ack_event_flags {
813 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
814 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
815 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
819 * Interface for adding new TCP congestion control handlers
821 #define TCP_CA_NAME_MAX 16
822 #define TCP_CA_MAX 128
823 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
825 #define TCP_CA_UNSPEC 0
827 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
828 #define TCP_CONG_NON_RESTRICTED 0x1
829 /* Requires ECN/ECT set on all packets */
830 #define TCP_CONG_NEEDS_ECN 0x2
834 struct tcp_congestion_ops {
835 struct list_head list;
839 /* initialize private data (optional) */
840 void (*init)(struct sock *sk);
841 /* cleanup private data (optional) */
842 void (*release)(struct sock *sk);
844 /* return slow start threshold (required) */
845 u32 (*ssthresh)(struct sock *sk);
846 /* do new cwnd calculation (required) */
847 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
848 /* call before changing ca_state (optional) */
849 void (*set_state)(struct sock *sk, u8 new_state);
850 /* call when cwnd event occurs (optional) */
851 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
852 /* call when ack arrives (optional) */
853 void (*in_ack_event)(struct sock *sk, u32 flags);
854 /* new value of cwnd after loss (optional) */
855 u32 (*undo_cwnd)(struct sock *sk);
856 /* hook for packet ack accounting (optional) */
857 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
858 /* get info for inet_diag (optional) */
859 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
860 union tcp_cc_info *info);
862 char name[TCP_CA_NAME_MAX];
863 struct module *owner;
866 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
867 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
869 void tcp_assign_congestion_control(struct sock *sk);
870 void tcp_init_congestion_control(struct sock *sk);
871 void tcp_cleanup_congestion_control(struct sock *sk);
872 int tcp_set_default_congestion_control(const char *name);
873 void tcp_get_default_congestion_control(char *name);
874 void tcp_get_available_congestion_control(char *buf, size_t len);
875 void tcp_get_allowed_congestion_control(char *buf, size_t len);
876 int tcp_set_allowed_congestion_control(char *allowed);
877 int tcp_set_congestion_control(struct sock *sk, const char *name);
878 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
879 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
881 u32 tcp_reno_ssthresh(struct sock *sk);
882 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
883 extern struct tcp_congestion_ops tcp_reno;
885 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
886 u32 tcp_ca_get_key_by_name(const char *name);
888 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
890 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
896 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
898 const struct inet_connection_sock *icsk = inet_csk(sk);
900 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
903 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
905 struct inet_connection_sock *icsk = inet_csk(sk);
907 if (icsk->icsk_ca_ops->set_state)
908 icsk->icsk_ca_ops->set_state(sk, ca_state);
909 icsk->icsk_ca_state = ca_state;
912 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
914 const struct inet_connection_sock *icsk = inet_csk(sk);
916 if (icsk->icsk_ca_ops->cwnd_event)
917 icsk->icsk_ca_ops->cwnd_event(sk, event);
920 /* These functions determine how the current flow behaves in respect of SACK
921 * handling. SACK is negotiated with the peer, and therefore it can vary
922 * between different flows.
924 * tcp_is_sack - SACK enabled
925 * tcp_is_reno - No SACK
926 * tcp_is_fack - FACK enabled, implies SACK enabled
928 static inline int tcp_is_sack(const struct tcp_sock *tp)
930 return tp->rx_opt.sack_ok;
933 static inline bool tcp_is_reno(const struct tcp_sock *tp)
935 return !tcp_is_sack(tp);
938 static inline bool tcp_is_fack(const struct tcp_sock *tp)
940 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
943 static inline void tcp_enable_fack(struct tcp_sock *tp)
945 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
948 /* TCP early-retransmit (ER) is similar to but more conservative than
949 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
951 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
953 tp->do_early_retrans = sysctl_tcp_early_retrans &&
954 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
955 sysctl_tcp_reordering == 3;
958 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
960 tp->do_early_retrans = 0;
963 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
965 return tp->sacked_out + tp->lost_out;
968 /* This determines how many packets are "in the network" to the best
969 * of our knowledge. In many cases it is conservative, but where
970 * detailed information is available from the receiver (via SACK
971 * blocks etc.) we can make more aggressive calculations.
973 * Use this for decisions involving congestion control, use just
974 * tp->packets_out to determine if the send queue is empty or not.
976 * Read this equation as:
978 * "Packets sent once on transmission queue" MINUS
979 * "Packets left network, but not honestly ACKed yet" PLUS
980 * "Packets fast retransmitted"
982 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
984 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
987 #define TCP_INFINITE_SSTHRESH 0x7fffffff
989 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
991 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
994 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
996 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
997 (1 << inet_csk(sk)->icsk_ca_state);
1000 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1001 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1004 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1006 const struct tcp_sock *tp = tcp_sk(sk);
1008 if (tcp_in_cwnd_reduction(sk))
1009 return tp->snd_ssthresh;
1011 return max(tp->snd_ssthresh,
1012 ((tp->snd_cwnd >> 1) +
1013 (tp->snd_cwnd >> 2)));
1016 /* Use define here intentionally to get WARN_ON location shown at the caller */
1017 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1019 void tcp_enter_cwr(struct sock *sk);
1020 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1022 /* The maximum number of MSS of available cwnd for which TSO defers
1023 * sending if not using sysctl_tcp_tso_win_divisor.
1025 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1030 /* Slow start with delack produces 3 packets of burst, so that
1031 * it is safe "de facto". This will be the default - same as
1032 * the default reordering threshold - but if reordering increases,
1033 * we must be able to allow cwnd to burst at least this much in order
1034 * to not pull it back when holes are filled.
1036 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1038 return tp->reordering;
1041 /* Returns end sequence number of the receiver's advertised window */
1042 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1044 return tp->snd_una + tp->snd_wnd;
1047 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1048 * flexible approach. The RFC suggests cwnd should not be raised unless
1049 * it was fully used previously. And that's exactly what we do in
1050 * congestion avoidance mode. But in slow start we allow cwnd to grow
1051 * as long as the application has used half the cwnd.
1053 * cwnd is 10 (IW10), but application sends 9 frames.
1054 * We allow cwnd to reach 18 when all frames are ACKed.
1055 * This check is safe because it's as aggressive as slow start which already
1056 * risks 100% overshoot. The advantage is that we discourage application to
1057 * either send more filler packets or data to artificially blow up the cwnd
1058 * usage, and allow application-limited process to probe bw more aggressively.
1060 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1062 const struct tcp_sock *tp = tcp_sk(sk);
1064 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1065 if (tp->snd_cwnd <= tp->snd_ssthresh)
1066 return tp->snd_cwnd < 2 * tp->max_packets_out;
1068 return tp->is_cwnd_limited;
1071 /* Something is really bad, we could not queue an additional packet,
1072 * because qdisc is full or receiver sent a 0 window.
1073 * We do not want to add fuel to the fire, or abort too early,
1074 * so make sure the timer we arm now is at least 200ms in the future,
1075 * regardless of current icsk_rto value (as it could be ~2ms)
1077 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1079 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1082 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1083 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1084 unsigned long max_when)
1086 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1088 return (unsigned long)min_t(u64, when, max_when);
1091 static inline void tcp_check_probe_timer(struct sock *sk)
1093 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1094 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1095 tcp_probe0_base(sk), TCP_RTO_MAX);
1098 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1103 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1109 * Calculate(/check) TCP checksum
1111 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1112 __be32 daddr, __wsum base)
1114 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1117 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1119 return __skb_checksum_complete(skb);
1122 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1124 return !skb_csum_unnecessary(skb) &&
1125 __tcp_checksum_complete(skb);
1128 /* Prequeue for VJ style copy to user, combined with checksumming. */
1130 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1132 tp->ucopy.task = NULL;
1134 tp->ucopy.memory = 0;
1135 skb_queue_head_init(&tp->ucopy.prequeue);
1138 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1143 static const char *statename[]={
1144 "Unused","Established","Syn Sent","Syn Recv",
1145 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1146 "Close Wait","Last ACK","Listen","Closing"
1149 void tcp_set_state(struct sock *sk, int state);
1151 void tcp_done(struct sock *sk);
1153 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1156 rx_opt->num_sacks = 0;
1159 u32 tcp_default_init_rwnd(u32 mss);
1161 /* Determine a window scaling and initial window to offer. */
1162 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1163 __u32 *window_clamp, int wscale_ok,
1164 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1166 static inline int tcp_win_from_space(int space)
1168 return sysctl_tcp_adv_win_scale<=0 ?
1169 (space>>(-sysctl_tcp_adv_win_scale)) :
1170 space - (space>>sysctl_tcp_adv_win_scale);
1173 /* Note: caller must be prepared to deal with negative returns */
1174 static inline int tcp_space(const struct sock *sk)
1176 return tcp_win_from_space(sk->sk_rcvbuf -
1177 atomic_read(&sk->sk_rmem_alloc));
1180 static inline int tcp_full_space(const struct sock *sk)
1182 return tcp_win_from_space(sk->sk_rcvbuf);
1185 extern void tcp_openreq_init_rwin(struct request_sock *req,
1186 struct sock *sk, struct dst_entry *dst);
1188 void tcp_enter_memory_pressure(struct sock *sk);
1190 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1192 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1195 static inline int keepalive_time_when(const struct tcp_sock *tp)
1197 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1200 static inline int keepalive_probes(const struct tcp_sock *tp)
1202 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1205 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1207 const struct inet_connection_sock *icsk = &tp->inet_conn;
1209 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1210 tcp_time_stamp - tp->rcv_tstamp);
1213 static inline int tcp_fin_time(const struct sock *sk)
1215 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1216 const int rto = inet_csk(sk)->icsk_rto;
1218 if (fin_timeout < (rto << 2) - (rto >> 1))
1219 fin_timeout = (rto << 2) - (rto >> 1);
1224 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1227 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1229 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1232 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1233 * then following tcp messages have valid values. Ignore 0 value,
1234 * or else 'negative' tsval might forbid us to accept their packets.
1236 if (!rx_opt->ts_recent)
1241 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1244 if (tcp_paws_check(rx_opt, 0))
1247 /* RST segments are not recommended to carry timestamp,
1248 and, if they do, it is recommended to ignore PAWS because
1249 "their cleanup function should take precedence over timestamps."
1250 Certainly, it is mistake. It is necessary to understand the reasons
1251 of this constraint to relax it: if peer reboots, clock may go
1252 out-of-sync and half-open connections will not be reset.
1253 Actually, the problem would be not existing if all
1254 the implementations followed draft about maintaining clock
1255 via reboots. Linux-2.2 DOES NOT!
1257 However, we can relax time bounds for RST segments to MSL.
1259 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1264 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1265 int mib_idx, u32 *last_oow_ack_time);
1267 static inline void tcp_mib_init(struct net *net)
1270 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1271 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1272 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1273 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1277 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1279 tp->lost_skb_hint = NULL;
1282 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1284 tcp_clear_retrans_hints_partial(tp);
1285 tp->retransmit_skb_hint = NULL;
1291 union tcp_md5_addr {
1293 #if IS_ENABLED(CONFIG_IPV6)
1298 /* - key database */
1299 struct tcp_md5sig_key {
1300 struct hlist_node node;
1302 u8 family; /* AF_INET or AF_INET6 */
1303 union tcp_md5_addr addr;
1304 u8 key[TCP_MD5SIG_MAXKEYLEN];
1305 struct rcu_head rcu;
1309 struct tcp_md5sig_info {
1310 struct hlist_head head;
1311 struct rcu_head rcu;
1314 /* - pseudo header */
1315 struct tcp4_pseudohdr {
1323 struct tcp6_pseudohdr {
1324 struct in6_addr saddr;
1325 struct in6_addr daddr;
1327 __be32 protocol; /* including padding */
1330 union tcp_md5sum_block {
1331 struct tcp4_pseudohdr ip4;
1332 #if IS_ENABLED(CONFIG_IPV6)
1333 struct tcp6_pseudohdr ip6;
1337 /* - pool: digest algorithm, hash description and scratch buffer */
1338 struct tcp_md5sig_pool {
1339 struct hash_desc md5_desc;
1340 union tcp_md5sum_block md5_blk;
1344 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1345 const struct sock *sk, const struct sk_buff *skb);
1346 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1347 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1348 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1350 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1351 const struct sock *addr_sk);
1353 #ifdef CONFIG_TCP_MD5SIG
1354 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1355 const union tcp_md5_addr *addr,
1357 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1359 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1360 const union tcp_md5_addr *addr,
1365 #define tcp_twsk_md5_key(twsk) NULL
1368 bool tcp_alloc_md5sig_pool(void);
1370 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1371 static inline void tcp_put_md5sig_pool(void)
1376 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1377 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1378 unsigned int header_len);
1379 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1380 const struct tcp_md5sig_key *key);
1382 /* From tcp_fastopen.c */
1383 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1384 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1385 unsigned long *last_syn_loss);
1386 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1387 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1389 struct tcp_fastopen_request {
1390 /* Fast Open cookie. Size 0 means a cookie request */
1391 struct tcp_fastopen_cookie cookie;
1392 struct msghdr *data; /* data in MSG_FASTOPEN */
1394 int copied; /* queued in tcp_connect() */
1396 void tcp_free_fastopen_req(struct tcp_sock *tp);
1398 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1399 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1400 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1401 struct request_sock *req,
1402 struct tcp_fastopen_cookie *foc,
1403 struct dst_entry *dst);
1404 void tcp_fastopen_init_key_once(bool publish);
1405 #define TCP_FASTOPEN_KEY_LENGTH 16
1407 /* Fastopen key context */
1408 struct tcp_fastopen_context {
1409 struct crypto_cipher *tfm;
1410 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1411 struct rcu_head rcu;
1414 /* write queue abstraction */
1415 static inline void tcp_write_queue_purge(struct sock *sk)
1417 struct sk_buff *skb;
1419 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1420 sk_wmem_free_skb(sk, skb);
1422 tcp_clear_all_retrans_hints(tcp_sk(sk));
1425 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1427 return skb_peek(&sk->sk_write_queue);
1430 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1432 return skb_peek_tail(&sk->sk_write_queue);
1435 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1436 const struct sk_buff *skb)
1438 return skb_queue_next(&sk->sk_write_queue, skb);
1441 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1442 const struct sk_buff *skb)
1444 return skb_queue_prev(&sk->sk_write_queue, skb);
1447 #define tcp_for_write_queue(skb, sk) \
1448 skb_queue_walk(&(sk)->sk_write_queue, skb)
1450 #define tcp_for_write_queue_from(skb, sk) \
1451 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1453 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1454 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1456 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1458 return sk->sk_send_head;
1461 static inline bool tcp_skb_is_last(const struct sock *sk,
1462 const struct sk_buff *skb)
1464 return skb_queue_is_last(&sk->sk_write_queue, skb);
1467 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1469 if (tcp_skb_is_last(sk, skb))
1470 sk->sk_send_head = NULL;
1472 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1475 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1477 if (sk->sk_send_head == skb_unlinked)
1478 sk->sk_send_head = NULL;
1481 static inline void tcp_init_send_head(struct sock *sk)
1483 sk->sk_send_head = NULL;
1486 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1488 __skb_queue_tail(&sk->sk_write_queue, skb);
1491 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1493 __tcp_add_write_queue_tail(sk, skb);
1495 /* Queue it, remembering where we must start sending. */
1496 if (sk->sk_send_head == NULL) {
1497 sk->sk_send_head = skb;
1499 if (tcp_sk(sk)->highest_sack == NULL)
1500 tcp_sk(sk)->highest_sack = skb;
1504 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1506 __skb_queue_head(&sk->sk_write_queue, skb);
1509 /* Insert buff after skb on the write queue of sk. */
1510 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1511 struct sk_buff *buff,
1514 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1517 /* Insert new before skb on the write queue of sk. */
1518 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1519 struct sk_buff *skb,
1522 __skb_queue_before(&sk->sk_write_queue, skb, new);
1524 if (sk->sk_send_head == skb)
1525 sk->sk_send_head = new;
1528 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1530 __skb_unlink(skb, &sk->sk_write_queue);
1533 static inline bool tcp_write_queue_empty(struct sock *sk)
1535 return skb_queue_empty(&sk->sk_write_queue);
1538 static inline void tcp_push_pending_frames(struct sock *sk)
1540 if (tcp_send_head(sk)) {
1541 struct tcp_sock *tp = tcp_sk(sk);
1543 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1547 /* Start sequence of the skb just after the highest skb with SACKed
1548 * bit, valid only if sacked_out > 0 or when the caller has ensured
1549 * validity by itself.
1551 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1553 if (!tp->sacked_out)
1556 if (tp->highest_sack == NULL)
1559 return TCP_SKB_CB(tp->highest_sack)->seq;
1562 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1564 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1565 tcp_write_queue_next(sk, skb);
1568 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1570 return tcp_sk(sk)->highest_sack;
1573 static inline void tcp_highest_sack_reset(struct sock *sk)
1575 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1578 /* Called when old skb is about to be deleted (to be combined with new skb) */
1579 static inline void tcp_highest_sack_combine(struct sock *sk,
1580 struct sk_buff *old,
1581 struct sk_buff *new)
1583 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1584 tcp_sk(sk)->highest_sack = new;
1587 /* Determines whether this is a thin stream (which may suffer from
1588 * increased latency). Used to trigger latency-reducing mechanisms.
1590 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1592 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1596 enum tcp_seq_states {
1597 TCP_SEQ_STATE_LISTENING,
1598 TCP_SEQ_STATE_OPENREQ,
1599 TCP_SEQ_STATE_ESTABLISHED,
1602 int tcp_seq_open(struct inode *inode, struct file *file);
1604 struct tcp_seq_afinfo {
1607 const struct file_operations *seq_fops;
1608 struct seq_operations seq_ops;
1611 struct tcp_iter_state {
1612 struct seq_net_private p;
1614 enum tcp_seq_states state;
1615 struct sock *syn_wait_sk;
1616 int bucket, offset, sbucket, num;
1621 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1622 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1624 extern struct request_sock_ops tcp_request_sock_ops;
1625 extern struct request_sock_ops tcp6_request_sock_ops;
1627 void tcp_v4_destroy_sock(struct sock *sk);
1629 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1630 netdev_features_t features);
1631 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1632 int tcp_gro_complete(struct sk_buff *skb);
1634 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1636 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1638 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1641 static inline bool tcp_stream_memory_free(const struct sock *sk)
1643 const struct tcp_sock *tp = tcp_sk(sk);
1644 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1646 return notsent_bytes < tcp_notsent_lowat(tp);
1649 #ifdef CONFIG_PROC_FS
1650 int tcp4_proc_init(void);
1651 void tcp4_proc_exit(void);
1654 int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
1655 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1656 const struct tcp_request_sock_ops *af_ops,
1657 struct sock *sk, struct sk_buff *skb);
1659 /* TCP af-specific functions */
1660 struct tcp_sock_af_ops {
1661 #ifdef CONFIG_TCP_MD5SIG
1662 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1663 const struct sock *addr_sk);
1664 int (*calc_md5_hash)(char *location,
1665 const struct tcp_md5sig_key *md5,
1666 const struct sock *sk,
1667 const struct sk_buff *skb);
1668 int (*md5_parse)(struct sock *sk,
1669 char __user *optval,
1674 struct tcp_request_sock_ops {
1676 #ifdef CONFIG_TCP_MD5SIG
1677 struct tcp_md5sig_key *(*req_md5_lookup)(struct sock *sk,
1678 const struct sock *addr_sk);
1679 int (*calc_md5_hash) (char *location,
1680 const struct tcp_md5sig_key *md5,
1681 const struct sock *sk,
1682 const struct sk_buff *skb);
1684 void (*init_req)(struct request_sock *req, struct sock *sk,
1685 struct sk_buff *skb);
1686 #ifdef CONFIG_SYN_COOKIES
1687 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1690 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1691 const struct request_sock *req,
1693 __u32 (*init_seq)(const struct sk_buff *skb);
1694 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1695 struct flowi *fl, struct request_sock *req,
1696 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
1697 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1698 const unsigned long timeout);
1701 #ifdef CONFIG_SYN_COOKIES
1702 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1703 struct sock *sk, struct sk_buff *skb,
1706 return ops->cookie_init_seq(sk, skb, mss);
1709 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1710 struct sock *sk, struct sk_buff *skb,
1717 int tcpv4_offload_init(void);
1719 void tcp_v4_init(void);
1720 void tcp_init(void);
1723 * Save and compile IPv4 options, return a pointer to it
1725 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1727 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1728 struct ip_options_rcu *dopt = NULL;
1731 int opt_size = sizeof(*dopt) + opt->optlen;
1733 dopt = kmalloc(opt_size, GFP_ATOMIC);
1734 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1742 /* locally generated TCP pure ACKs have skb->truesize == 2
1743 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1744 * This is much faster than dissecting the packet to find out.
1745 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1747 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1749 return skb->truesize == 2;
1752 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)