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 /* Number of full MSS to receive before Acking RFC2581 */
147 #define TCP_DELACK_SEG 1
149 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
150 * for local resources.
153 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
154 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
155 #define TCP_KEEPALIVE_INTVL (75*HZ)
157 #define MAX_TCP_KEEPIDLE 32767
158 #define MAX_TCP_KEEPINTVL 32767
159 #define MAX_TCP_KEEPCNT 127
160 #define MAX_TCP_SYNCNT 127
162 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
164 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
165 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
166 * after this time. It should be equal
167 * (or greater than) TCP_TIMEWAIT_LEN
168 * to provide reliability equal to one
169 * provided by timewait state.
171 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
172 * timestamps. It must be less than
173 * minimal timewait lifetime.
179 #define TCPOPT_NOP 1 /* Padding */
180 #define TCPOPT_EOL 0 /* End of options */
181 #define TCPOPT_MSS 2 /* Segment size negotiating */
182 #define TCPOPT_WINDOW 3 /* Window scaling */
183 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
184 #define TCPOPT_SACK 5 /* SACK Block */
185 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
186 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
187 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
188 #define TCPOPT_EXP 254 /* Experimental */
189 /* Magic number to be after the option value for sharing TCP
190 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
192 #define TCPOPT_FASTOPEN_MAGIC 0xF989
198 #define TCPOLEN_MSS 4
199 #define TCPOLEN_WINDOW 3
200 #define TCPOLEN_SACK_PERM 2
201 #define TCPOLEN_TIMESTAMP 10
202 #define TCPOLEN_MD5SIG 18
203 #define TCPOLEN_FASTOPEN_BASE 2
204 #define TCPOLEN_EXP_FASTOPEN_BASE 4
206 /* But this is what stacks really send out. */
207 #define TCPOLEN_TSTAMP_ALIGNED 12
208 #define TCPOLEN_WSCALE_ALIGNED 4
209 #define TCPOLEN_SACKPERM_ALIGNED 4
210 #define TCPOLEN_SACK_BASE 2
211 #define TCPOLEN_SACK_BASE_ALIGNED 4
212 #define TCPOLEN_SACK_PERBLOCK 8
213 #define TCPOLEN_MD5SIG_ALIGNED 20
214 #define TCPOLEN_MSS_ALIGNED 4
216 /* Flags in tp->nonagle */
217 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
218 #define TCP_NAGLE_CORK 2 /* Socket is corked */
219 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
221 /* TCP thin-stream limits */
222 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
224 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
225 #define TCP_INIT_CWND 10
227 /* Bit Flags for sysctl_tcp_fastopen */
228 #define TFO_CLIENT_ENABLE 1
229 #define TFO_SERVER_ENABLE 2
230 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
232 /* Accept SYN data w/o any cookie option */
233 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
235 /* Force enable TFO on all listeners, i.e., not requiring the
236 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
238 #define TFO_SERVER_WO_SOCKOPT1 0x400
239 #define TFO_SERVER_WO_SOCKOPT2 0x800
241 extern struct inet_timewait_death_row tcp_death_row;
243 /* sysctl variables for tcp */
244 extern int sysctl_tcp_timestamps;
245 extern int sysctl_tcp_window_scaling;
246 extern int sysctl_tcp_sack;
247 extern int sysctl_tcp_fin_timeout;
248 extern int sysctl_tcp_keepalive_time;
249 extern int sysctl_tcp_keepalive_probes;
250 extern int sysctl_tcp_keepalive_intvl;
251 extern int sysctl_tcp_syn_retries;
252 extern int sysctl_tcp_synack_retries;
253 extern int sysctl_tcp_retries1;
254 extern int sysctl_tcp_retries2;
255 extern int sysctl_tcp_orphan_retries;
256 extern int sysctl_tcp_syncookies;
257 extern int sysctl_tcp_fastopen;
258 extern int sysctl_tcp_retrans_collapse;
259 extern int sysctl_tcp_stdurg;
260 extern int sysctl_tcp_rfc1337;
261 extern int sysctl_tcp_abort_on_overflow;
262 extern int sysctl_tcp_max_orphans;
263 extern int sysctl_tcp_fack;
264 extern int sysctl_tcp_reordering;
265 extern int sysctl_tcp_max_reordering;
266 extern int sysctl_tcp_dsack;
267 extern long sysctl_tcp_mem[3];
268 extern int sysctl_tcp_wmem[3];
269 extern int sysctl_tcp_rmem[3];
270 extern int sysctl_tcp_app_win;
271 extern int sysctl_tcp_adv_win_scale;
272 extern int sysctl_tcp_tw_reuse;
273 extern int sysctl_tcp_frto;
274 extern int sysctl_tcp_low_latency;
275 extern int sysctl_tcp_nometrics_save;
276 extern int sysctl_tcp_moderate_rcvbuf;
277 extern int sysctl_tcp_tso_win_divisor;
278 extern int sysctl_tcp_workaround_signed_windows;
279 extern int sysctl_tcp_slow_start_after_idle;
280 extern int sysctl_tcp_thin_linear_timeouts;
281 extern int sysctl_tcp_thin_dupack;
282 extern int sysctl_tcp_early_retrans;
283 extern int sysctl_tcp_limit_output_bytes;
284 extern int sysctl_tcp_challenge_ack_limit;
285 extern unsigned int sysctl_tcp_notsent_lowat;
286 extern int sysctl_tcp_min_tso_segs;
287 extern int sysctl_tcp_min_rtt_wlen;
288 extern int sysctl_tcp_autocorking;
289 extern int sysctl_tcp_invalid_ratelimit;
290 extern int sysctl_tcp_pacing_ss_ratio;
291 extern int sysctl_tcp_pacing_ca_ratio;
292 extern int sysctl_tcp_default_init_rwnd;
294 extern atomic_long_t tcp_memory_allocated;
296 /* sysctl variables for controlling various tcp parameters */
297 extern int sysctl_tcp_delack_seg;
298 extern int sysctl_tcp_use_userconfig;
300 extern struct percpu_counter tcp_sockets_allocated;
301 extern int tcp_memory_pressure;
303 /* optimized version of sk_under_memory_pressure() for TCP sockets */
304 static inline bool tcp_under_memory_pressure(const struct sock *sk)
306 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
307 return !!sk->sk_cgrp->memory_pressure;
309 return tcp_memory_pressure;
312 * The next routines deal with comparing 32 bit unsigned ints
313 * and worry about wraparound (automatic with unsigned arithmetic).
316 static inline bool before(__u32 seq1, __u32 seq2)
318 return (__s32)(seq1-seq2) < 0;
320 #define after(seq2, seq1) before(seq1, seq2)
322 /* is s2<=s1<=s3 ? */
323 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
325 return seq3 - seq2 >= seq1 - seq2;
328 static inline bool tcp_out_of_memory(struct sock *sk)
330 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
331 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
336 void sk_forced_mem_schedule(struct sock *sk, int size);
338 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
340 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
341 int orphans = percpu_counter_read_positive(ocp);
343 if (orphans << shift > sysctl_tcp_max_orphans) {
344 orphans = percpu_counter_sum_positive(ocp);
345 if (orphans << shift > sysctl_tcp_max_orphans)
351 bool tcp_check_oom(struct sock *sk, int shift);
354 extern struct proto tcp_prot;
356 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
357 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
358 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
359 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
360 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
362 void tcp_tasklet_init(void);
364 void tcp_v4_err(struct sk_buff *skb, u32);
366 void tcp_shutdown(struct sock *sk, int how);
368 void tcp_v4_early_demux(struct sk_buff *skb);
369 int tcp_v4_rcv(struct sk_buff *skb);
371 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
372 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
373 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
375 void tcp_release_cb(struct sock *sk);
376 void tcp_wfree(struct sk_buff *skb);
377 void tcp_write_timer_handler(struct sock *sk);
378 void tcp_delack_timer_handler(struct sock *sk);
379 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
380 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
381 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
382 const struct tcphdr *th, unsigned int len);
383 void tcp_rcv_space_adjust(struct sock *sk);
384 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
385 void tcp_twsk_destructor(struct sock *sk);
386 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
387 struct pipe_inode_info *pipe, size_t len,
390 /* sysctl master controller */
391 extern int tcp_use_userconfig_sysctl_handler(struct ctl_table *, int,
392 void __user *, size_t *, loff_t *);
393 extern int tcp_proc_delayed_ack_control(struct ctl_table *, int,
394 void __user *, size_t *, loff_t *);
396 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
398 static inline void tcp_dec_quickack_mode(struct sock *sk,
399 const unsigned int pkts)
401 struct inet_connection_sock *icsk = inet_csk(sk);
403 if (icsk->icsk_ack.quick) {
404 if (pkts >= icsk->icsk_ack.quick) {
405 icsk->icsk_ack.quick = 0;
406 /* Leaving quickack mode we deflate ATO. */
407 icsk->icsk_ack.ato = TCP_ATO_MIN;
409 icsk->icsk_ack.quick -= pkts;
414 #define TCP_ECN_QUEUE_CWR 2
415 #define TCP_ECN_DEMAND_CWR 4
416 #define TCP_ECN_SEEN 8
426 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
428 const struct tcphdr *th);
429 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
430 struct request_sock *req, bool fastopen);
431 int tcp_child_process(struct sock *parent, struct sock *child,
432 struct sk_buff *skb);
433 void tcp_enter_loss(struct sock *sk);
434 void tcp_clear_retrans(struct tcp_sock *tp);
435 void tcp_update_metrics(struct sock *sk);
436 void tcp_init_metrics(struct sock *sk);
437 void tcp_metrics_init(void);
438 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
439 bool paws_check, bool timestamps);
440 bool tcp_remember_stamp(struct sock *sk);
441 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
442 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
443 void tcp_disable_fack(struct tcp_sock *tp);
444 void tcp_close(struct sock *sk, long timeout);
445 void tcp_init_sock(struct sock *sk);
446 unsigned int tcp_poll(struct file *file, struct socket *sock,
447 struct poll_table_struct *wait);
448 int tcp_getsockopt(struct sock *sk, int level, int optname,
449 char __user *optval, int __user *optlen);
450 int tcp_setsockopt(struct sock *sk, int level, int optname,
451 char __user *optval, unsigned int optlen);
452 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
453 char __user *optval, int __user *optlen);
454 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
455 char __user *optval, unsigned int optlen);
456 void tcp_set_keepalive(struct sock *sk, int val);
457 void tcp_syn_ack_timeout(const struct request_sock *req);
458 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
459 int flags, int *addr_len);
460 void tcp_parse_options(const struct sk_buff *skb,
461 struct tcp_options_received *opt_rx,
462 int estab, struct tcp_fastopen_cookie *foc);
463 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
466 * TCP v4 functions exported for the inet6 API
469 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
470 void tcp_v4_mtu_reduced(struct sock *sk);
471 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
472 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
473 struct sock *tcp_create_openreq_child(const struct sock *sk,
474 struct request_sock *req,
475 struct sk_buff *skb);
476 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
477 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
478 struct request_sock *req,
479 struct dst_entry *dst,
480 struct request_sock *req_unhash,
482 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
483 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
484 int tcp_connect(struct sock *sk);
485 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
486 struct request_sock *req,
487 struct tcp_fastopen_cookie *foc,
489 int tcp_disconnect(struct sock *sk, int flags);
491 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
492 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
493 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
495 /* From syncookies.c */
496 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
497 struct request_sock *req,
498 struct dst_entry *dst);
499 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
501 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
502 #ifdef CONFIG_SYN_COOKIES
504 /* Syncookies use a monotonic timer which increments every 60 seconds.
505 * This counter is used both as a hash input and partially encoded into
506 * the cookie value. A cookie is only validated further if the delta
507 * between the current counter value and the encoded one is less than this,
508 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
509 * the counter advances immediately after a cookie is generated).
511 #define MAX_SYNCOOKIE_AGE 2
512 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
513 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
515 /* syncookies: remember time of last synqueue overflow
516 * But do not dirty this field too often (once per second is enough)
517 * It is racy as we do not hold a lock, but race is very minor.
519 static inline void tcp_synq_overflow(const struct sock *sk)
521 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
522 unsigned long now = jiffies;
524 if (time_after(now, last_overflow + HZ))
525 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
528 /* syncookies: no recent synqueue overflow on this listening socket? */
529 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
531 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
533 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
536 static inline u32 tcp_cookie_time(void)
538 u64 val = get_jiffies_64();
540 do_div(val, TCP_SYNCOOKIE_PERIOD);
544 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
546 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
547 __u32 cookie_init_timestamp(struct request_sock *req);
548 bool cookie_timestamp_decode(struct tcp_options_received *opt);
549 bool cookie_ecn_ok(const struct tcp_options_received *opt,
550 const struct net *net, const struct dst_entry *dst);
552 /* From net/ipv6/syncookies.c */
553 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
555 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
557 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
558 const struct tcphdr *th, u16 *mssp);
559 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
563 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
565 bool tcp_may_send_now(struct sock *sk);
566 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
567 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
568 void tcp_retransmit_timer(struct sock *sk);
569 void tcp_xmit_retransmit_queue(struct sock *);
570 void tcp_simple_retransmit(struct sock *);
571 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
572 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
574 void tcp_send_probe0(struct sock *);
575 void tcp_send_partial(struct sock *);
576 int tcp_write_wakeup(struct sock *, int mib);
577 void tcp_send_fin(struct sock *sk);
578 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
579 int tcp_send_synack(struct sock *);
580 void tcp_push_one(struct sock *, unsigned int mss_now);
581 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
582 void tcp_send_ack(struct sock *sk);
583 void tcp_send_delayed_ack(struct sock *sk);
584 void tcp_send_loss_probe(struct sock *sk);
585 bool tcp_schedule_loss_probe(struct sock *sk);
588 void tcp_resume_early_retransmit(struct sock *sk);
589 void tcp_rearm_rto(struct sock *sk);
590 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
591 void tcp_reset(struct sock *sk);
592 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
595 void tcp_init_xmit_timers(struct sock *);
596 static inline void tcp_clear_xmit_timers(struct sock *sk)
598 inet_csk_clear_xmit_timers(sk);
601 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
602 unsigned int tcp_current_mss(struct sock *sk);
604 /* Bound MSS / TSO packet size with the half of the window */
605 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
609 /* When peer uses tiny windows, there is no use in packetizing
610 * to sub-MSS pieces for the sake of SWS or making sure there
611 * are enough packets in the pipe for fast recovery.
613 * On the other hand, for extremely large MSS devices, handling
614 * smaller than MSS windows in this way does make sense.
616 if (tp->max_window >= 512)
617 cutoff = (tp->max_window >> 1);
619 cutoff = tp->max_window;
621 if (cutoff && pktsize > cutoff)
622 return max_t(int, cutoff, 68U - tp->tcp_header_len);
628 void tcp_get_info(struct sock *, struct tcp_info *);
630 /* Read 'sendfile()'-style from a TCP socket */
631 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
632 unsigned int, size_t);
633 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
634 sk_read_actor_t recv_actor);
636 void tcp_initialize_rcv_mss(struct sock *sk);
638 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
639 int tcp_mss_to_mtu(struct sock *sk, int mss);
640 void tcp_mtup_init(struct sock *sk);
641 void tcp_init_buffer_space(struct sock *sk);
643 static inline void tcp_bound_rto(const struct sock *sk)
645 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
646 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
649 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
651 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
654 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
656 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
657 ntohl(TCP_FLAG_ACK) |
661 static inline void tcp_fast_path_on(struct tcp_sock *tp)
663 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
666 static inline void tcp_fast_path_check(struct sock *sk)
668 struct tcp_sock *tp = tcp_sk(sk);
670 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
672 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
674 tcp_fast_path_on(tp);
677 /* Compute the actual rto_min value */
678 static inline u32 tcp_rto_min(struct sock *sk)
680 const struct dst_entry *dst = __sk_dst_get(sk);
681 u32 rto_min = TCP_RTO_MIN;
683 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
684 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
688 static inline u32 tcp_rto_min_us(struct sock *sk)
690 return jiffies_to_usecs(tcp_rto_min(sk));
693 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
695 return dst_metric_locked(dst, RTAX_CC_ALGO);
698 /* Minimum RTT in usec. ~0 means not available. */
699 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
701 return tp->rtt_min[0].rtt;
704 /* Compute the actual receive window we are currently advertising.
705 * Rcv_nxt can be after the window if our peer push more data
706 * than the offered window.
708 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
710 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
717 /* Choose a new window, without checks for shrinking, and without
718 * scaling applied to the result. The caller does these things
719 * if necessary. This is a "raw" window selection.
721 u32 __tcp_select_window(struct sock *sk);
723 void tcp_send_window_probe(struct sock *sk);
725 /* TCP timestamps are only 32-bits, this causes a slight
726 * complication on 64-bit systems since we store a snapshot
727 * of jiffies in the buffer control blocks below. We decided
728 * to use only the low 32-bits of jiffies and hide the ugly
729 * casts with the following macro.
731 #define tcp_time_stamp ((__u32)(jiffies))
733 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
735 return skb->skb_mstamp.stamp_jiffies;
739 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
741 #define TCPHDR_FIN 0x01
742 #define TCPHDR_SYN 0x02
743 #define TCPHDR_RST 0x04
744 #define TCPHDR_PSH 0x08
745 #define TCPHDR_ACK 0x10
746 #define TCPHDR_URG 0x20
747 #define TCPHDR_ECE 0x40
748 #define TCPHDR_CWR 0x80
750 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
752 /* This is what the send packet queuing engine uses to pass
753 * TCP per-packet control information to the transmission code.
754 * We also store the host-order sequence numbers in here too.
755 * This is 44 bytes if IPV6 is enabled.
756 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
759 __u32 seq; /* Starting sequence number */
760 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
762 /* Note : tcp_tw_isn is used in input path only
763 * (isn chosen by tcp_timewait_state_process())
765 * tcp_gso_segs/size are used in write queue only,
766 * cf tcp_skb_pcount()/tcp_skb_mss()
774 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
776 __u8 sacked; /* State flags for SACK/FACK. */
777 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
778 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
779 #define TCPCB_LOST 0x04 /* SKB is lost */
780 #define TCPCB_TAGBITS 0x07 /* All tag bits */
781 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
782 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
783 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
786 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
788 __u32 ack_seq; /* Sequence number ACK'd */
790 struct inet_skb_parm h4;
791 #if IS_ENABLED(CONFIG_IPV6)
792 struct inet6_skb_parm h6;
794 } header; /* For incoming frames */
797 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
800 #if IS_ENABLED(CONFIG_IPV6)
801 /* This is the variant of inet6_iif() that must be used by TCP,
802 * as TCP moves IP6CB into a different location in skb->cb[]
804 static inline int tcp_v6_iif(const struct sk_buff *skb)
806 return TCP_SKB_CB(skb)->header.h6.iif;
810 /* Due to TSO, an SKB can be composed of multiple actual
811 * packets. To keep these tracked properly, we use this.
813 static inline int tcp_skb_pcount(const struct sk_buff *skb)
815 return TCP_SKB_CB(skb)->tcp_gso_segs;
818 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
820 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
823 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
825 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
828 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
829 static inline int tcp_skb_mss(const struct sk_buff *skb)
831 return TCP_SKB_CB(skb)->tcp_gso_size;
834 /* Events passed to congestion control interface */
836 CA_EVENT_TX_START, /* first transmit when no packets in flight */
837 CA_EVENT_CWND_RESTART, /* congestion window restart */
838 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
839 CA_EVENT_LOSS, /* loss timeout */
840 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
841 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
844 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
845 enum tcp_ca_ack_event_flags {
846 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
847 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
848 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
852 * Interface for adding new TCP congestion control handlers
854 #define TCP_CA_NAME_MAX 16
855 #define TCP_CA_MAX 128
856 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
858 #define TCP_CA_UNSPEC 0
860 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
861 #define TCP_CONG_NON_RESTRICTED 0x1
862 /* Requires ECN/ECT set on all packets */
863 #define TCP_CONG_NEEDS_ECN 0x2
867 struct tcp_congestion_ops {
868 struct list_head list;
872 /* initialize private data (optional) */
873 void (*init)(struct sock *sk);
874 /* cleanup private data (optional) */
875 void (*release)(struct sock *sk);
877 /* return slow start threshold (required) */
878 u32 (*ssthresh)(struct sock *sk);
879 /* do new cwnd calculation (required) */
880 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
881 /* call before changing ca_state (optional) */
882 void (*set_state)(struct sock *sk, u8 new_state);
883 /* call when cwnd event occurs (optional) */
884 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
885 /* call when ack arrives (optional) */
886 void (*in_ack_event)(struct sock *sk, u32 flags);
887 /* new value of cwnd after loss (optional) */
888 u32 (*undo_cwnd)(struct sock *sk);
889 /* hook for packet ack accounting (optional) */
890 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
891 /* get info for inet_diag (optional) */
892 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
893 union tcp_cc_info *info);
895 char name[TCP_CA_NAME_MAX];
896 struct module *owner;
899 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
900 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
902 void tcp_assign_congestion_control(struct sock *sk);
903 void tcp_init_congestion_control(struct sock *sk);
904 void tcp_cleanup_congestion_control(struct sock *sk);
905 int tcp_set_default_congestion_control(const char *name);
906 void tcp_get_default_congestion_control(char *name);
907 void tcp_get_available_congestion_control(char *buf, size_t len);
908 void tcp_get_allowed_congestion_control(char *buf, size_t len);
909 int tcp_set_allowed_congestion_control(char *allowed);
910 int tcp_set_congestion_control(struct sock *sk, const char *name);
911 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
912 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
914 u32 tcp_reno_ssthresh(struct sock *sk);
915 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
916 extern struct tcp_congestion_ops tcp_reno;
918 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
919 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
921 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
923 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
929 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
931 const struct inet_connection_sock *icsk = inet_csk(sk);
933 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
936 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
938 struct inet_connection_sock *icsk = inet_csk(sk);
940 if (icsk->icsk_ca_ops->set_state)
941 icsk->icsk_ca_ops->set_state(sk, ca_state);
942 icsk->icsk_ca_state = ca_state;
945 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
947 const struct inet_connection_sock *icsk = inet_csk(sk);
949 if (icsk->icsk_ca_ops->cwnd_event)
950 icsk->icsk_ca_ops->cwnd_event(sk, event);
953 /* These functions determine how the current flow behaves in respect of SACK
954 * handling. SACK is negotiated with the peer, and therefore it can vary
955 * between different flows.
957 * tcp_is_sack - SACK enabled
958 * tcp_is_reno - No SACK
959 * tcp_is_fack - FACK enabled, implies SACK enabled
961 static inline int tcp_is_sack(const struct tcp_sock *tp)
963 return tp->rx_opt.sack_ok;
966 static inline bool tcp_is_reno(const struct tcp_sock *tp)
968 return !tcp_is_sack(tp);
971 static inline bool tcp_is_fack(const struct tcp_sock *tp)
973 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
976 static inline void tcp_enable_fack(struct tcp_sock *tp)
978 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
981 /* TCP early-retransmit (ER) is similar to but more conservative than
982 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
984 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
986 tp->do_early_retrans = sysctl_tcp_early_retrans &&
987 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
988 sysctl_tcp_reordering == 3;
991 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
993 tp->do_early_retrans = 0;
996 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
998 return tp->sacked_out + tp->lost_out;
1001 /* This determines how many packets are "in the network" to the best
1002 * of our knowledge. In many cases it is conservative, but where
1003 * detailed information is available from the receiver (via SACK
1004 * blocks etc.) we can make more aggressive calculations.
1006 * Use this for decisions involving congestion control, use just
1007 * tp->packets_out to determine if the send queue is empty or not.
1009 * Read this equation as:
1011 * "Packets sent once on transmission queue" MINUS
1012 * "Packets left network, but not honestly ACKed yet" PLUS
1013 * "Packets fast retransmitted"
1015 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1017 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1020 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1022 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1024 return tp->snd_cwnd < tp->snd_ssthresh;
1027 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1029 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1032 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1034 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1035 (1 << inet_csk(sk)->icsk_ca_state);
1038 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1039 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1042 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1044 const struct tcp_sock *tp = tcp_sk(sk);
1046 if (tcp_in_cwnd_reduction(sk))
1047 return tp->snd_ssthresh;
1049 return max(tp->snd_ssthresh,
1050 ((tp->snd_cwnd >> 1) +
1051 (tp->snd_cwnd >> 2)));
1054 /* Use define here intentionally to get WARN_ON location shown at the caller */
1055 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1057 void tcp_enter_cwr(struct sock *sk);
1058 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1060 /* The maximum number of MSS of available cwnd for which TSO defers
1061 * sending if not using sysctl_tcp_tso_win_divisor.
1063 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1068 /* Slow start with delack produces 3 packets of burst, so that
1069 * it is safe "de facto". This will be the default - same as
1070 * the default reordering threshold - but if reordering increases,
1071 * we must be able to allow cwnd to burst at least this much in order
1072 * to not pull it back when holes are filled.
1074 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1076 return tp->reordering;
1079 /* Returns end sequence number of the receiver's advertised window */
1080 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1082 return tp->snd_una + tp->snd_wnd;
1085 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1086 * flexible approach. The RFC suggests cwnd should not be raised unless
1087 * it was fully used previously. And that's exactly what we do in
1088 * congestion avoidance mode. But in slow start we allow cwnd to grow
1089 * as long as the application has used half the cwnd.
1091 * cwnd is 10 (IW10), but application sends 9 frames.
1092 * We allow cwnd to reach 18 when all frames are ACKed.
1093 * This check is safe because it's as aggressive as slow start which already
1094 * risks 100% overshoot. The advantage is that we discourage application to
1095 * either send more filler packets or data to artificially blow up the cwnd
1096 * usage, and allow application-limited process to probe bw more aggressively.
1098 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1100 const struct tcp_sock *tp = tcp_sk(sk);
1102 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1103 if (tcp_in_slow_start(tp))
1104 return tp->snd_cwnd < 2 * tp->max_packets_out;
1106 return tp->is_cwnd_limited;
1109 /* Something is really bad, we could not queue an additional packet,
1110 * because qdisc is full or receiver sent a 0 window.
1111 * We do not want to add fuel to the fire, or abort too early,
1112 * so make sure the timer we arm now is at least 200ms in the future,
1113 * regardless of current icsk_rto value (as it could be ~2ms)
1115 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1117 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1120 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1121 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1122 unsigned long max_when)
1124 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1126 return (unsigned long)min_t(u64, when, max_when);
1129 static inline void tcp_check_probe_timer(struct sock *sk)
1131 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1132 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1133 tcp_probe0_base(sk), TCP_RTO_MAX);
1136 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1141 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1147 * Calculate(/check) TCP checksum
1149 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1150 __be32 daddr, __wsum base)
1152 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1155 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1157 return __skb_checksum_complete(skb);
1160 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1162 return !skb_csum_unnecessary(skb) &&
1163 __tcp_checksum_complete(skb);
1166 /* Prequeue for VJ style copy to user, combined with checksumming. */
1168 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1170 tp->ucopy.task = NULL;
1172 tp->ucopy.memory = 0;
1173 skb_queue_head_init(&tp->ucopy.prequeue);
1176 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1177 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1182 static const char *statename[]={
1183 "Unused","Established","Syn Sent","Syn Recv",
1184 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1185 "Close Wait","Last ACK","Listen","Closing"
1188 void tcp_set_state(struct sock *sk, int state);
1190 void tcp_done(struct sock *sk);
1192 int tcp_abort(struct sock *sk, int err);
1194 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1197 rx_opt->num_sacks = 0;
1200 u32 tcp_default_init_rwnd(u32 mss);
1201 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1203 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1205 struct tcp_sock *tp = tcp_sk(sk);
1208 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1210 delta = tcp_time_stamp - tp->lsndtime;
1211 if (delta > inet_csk(sk)->icsk_rto)
1212 tcp_cwnd_restart(sk, delta);
1215 /* Determine a window scaling and initial window to offer. */
1216 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1217 __u32 *window_clamp, int wscale_ok,
1218 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1220 static inline int tcp_win_from_space(int space)
1222 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1224 return tcp_adv_win_scale <= 0 ?
1225 (space>>(-tcp_adv_win_scale)) :
1226 space - (space>>tcp_adv_win_scale);
1229 /* Note: caller must be prepared to deal with negative returns */
1230 static inline int tcp_space(const struct sock *sk)
1232 return tcp_win_from_space(sk->sk_rcvbuf -
1233 atomic_read(&sk->sk_rmem_alloc));
1236 static inline int tcp_full_space(const struct sock *sk)
1238 return tcp_win_from_space(sk->sk_rcvbuf);
1241 extern void tcp_openreq_init_rwin(struct request_sock *req,
1242 const struct sock *sk_listener,
1243 const struct dst_entry *dst);
1245 void tcp_enter_memory_pressure(struct sock *sk);
1247 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1249 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1252 static inline int keepalive_time_when(const struct tcp_sock *tp)
1254 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1257 static inline int keepalive_probes(const struct tcp_sock *tp)
1259 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1262 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1264 const struct inet_connection_sock *icsk = &tp->inet_conn;
1266 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1267 tcp_time_stamp - tp->rcv_tstamp);
1270 static inline int tcp_fin_time(const struct sock *sk)
1272 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1273 const int rto = inet_csk(sk)->icsk_rto;
1275 if (fin_timeout < (rto << 2) - (rto >> 1))
1276 fin_timeout = (rto << 2) - (rto >> 1);
1281 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1284 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1286 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1289 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1290 * then following tcp messages have valid values. Ignore 0 value,
1291 * or else 'negative' tsval might forbid us to accept their packets.
1293 if (!rx_opt->ts_recent)
1298 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1301 if (tcp_paws_check(rx_opt, 0))
1304 /* RST segments are not recommended to carry timestamp,
1305 and, if they do, it is recommended to ignore PAWS because
1306 "their cleanup function should take precedence over timestamps."
1307 Certainly, it is mistake. It is necessary to understand the reasons
1308 of this constraint to relax it: if peer reboots, clock may go
1309 out-of-sync and half-open connections will not be reset.
1310 Actually, the problem would be not existing if all
1311 the implementations followed draft about maintaining clock
1312 via reboots. Linux-2.2 DOES NOT!
1314 However, we can relax time bounds for RST segments to MSL.
1316 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1321 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1322 int mib_idx, u32 *last_oow_ack_time);
1324 static inline void tcp_mib_init(struct net *net)
1327 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1328 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1329 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1330 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1334 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1336 tp->lost_skb_hint = NULL;
1339 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1341 tcp_clear_retrans_hints_partial(tp);
1342 tp->retransmit_skb_hint = NULL;
1348 union tcp_md5_addr {
1350 #if IS_ENABLED(CONFIG_IPV6)
1355 /* - key database */
1356 struct tcp_md5sig_key {
1357 struct hlist_node node;
1359 u8 family; /* AF_INET or AF_INET6 */
1360 union tcp_md5_addr addr;
1361 u8 key[TCP_MD5SIG_MAXKEYLEN];
1362 struct rcu_head rcu;
1366 struct tcp_md5sig_info {
1367 struct hlist_head head;
1368 struct rcu_head rcu;
1371 /* - pseudo header */
1372 struct tcp4_pseudohdr {
1380 struct tcp6_pseudohdr {
1381 struct in6_addr saddr;
1382 struct in6_addr daddr;
1384 __be32 protocol; /* including padding */
1387 union tcp_md5sum_block {
1388 struct tcp4_pseudohdr ip4;
1389 #if IS_ENABLED(CONFIG_IPV6)
1390 struct tcp6_pseudohdr ip6;
1394 /* - pool: digest algorithm, hash description and scratch buffer */
1395 struct tcp_md5sig_pool {
1396 struct hash_desc md5_desc;
1397 union tcp_md5sum_block md5_blk;
1401 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1402 const struct sock *sk, const struct sk_buff *skb);
1403 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1404 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1405 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1407 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1408 const struct sock *addr_sk);
1410 #ifdef CONFIG_TCP_MD5SIG
1411 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1412 const union tcp_md5_addr *addr,
1414 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1416 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1417 const union tcp_md5_addr *addr,
1422 #define tcp_twsk_md5_key(twsk) NULL
1425 bool tcp_alloc_md5sig_pool(void);
1427 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1428 static inline void tcp_put_md5sig_pool(void)
1433 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1434 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1435 unsigned int header_len);
1436 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1437 const struct tcp_md5sig_key *key);
1439 /* From tcp_fastopen.c */
1440 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1441 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1442 unsigned long *last_syn_loss);
1443 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1444 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1446 struct tcp_fastopen_request {
1447 /* Fast Open cookie. Size 0 means a cookie request */
1448 struct tcp_fastopen_cookie cookie;
1449 struct msghdr *data; /* data in MSG_FASTOPEN */
1451 int copied; /* queued in tcp_connect() */
1453 void tcp_free_fastopen_req(struct tcp_sock *tp);
1455 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1456 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1457 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1458 struct request_sock *req,
1459 struct tcp_fastopen_cookie *foc,
1460 struct dst_entry *dst);
1461 void tcp_fastopen_init_key_once(bool publish);
1462 #define TCP_FASTOPEN_KEY_LENGTH 16
1464 static inline void tcp_init_send_head(struct sock *sk)
1466 sk->sk_send_head = NULL;
1469 /* Fastopen key context */
1470 struct tcp_fastopen_context {
1471 struct crypto_cipher *tfm;
1472 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1473 struct rcu_head rcu;
1476 /* write queue abstraction */
1477 static inline void tcp_write_queue_purge(struct sock *sk)
1479 struct sk_buff *skb;
1481 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1482 sk_wmem_free_skb(sk, skb);
1484 tcp_clear_all_retrans_hints(tcp_sk(sk));
1485 tcp_init_send_head(sk);
1486 inet_csk(sk)->icsk_backoff = 0;
1489 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1491 return skb_peek(&sk->sk_write_queue);
1494 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1496 return skb_peek_tail(&sk->sk_write_queue);
1499 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1500 const struct sk_buff *skb)
1502 return skb_queue_next(&sk->sk_write_queue, skb);
1505 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1506 const struct sk_buff *skb)
1508 return skb_queue_prev(&sk->sk_write_queue, skb);
1511 #define tcp_for_write_queue(skb, sk) \
1512 skb_queue_walk(&(sk)->sk_write_queue, skb)
1514 #define tcp_for_write_queue_from(skb, sk) \
1515 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1517 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1518 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1520 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1522 return sk->sk_send_head;
1525 static inline bool tcp_skb_is_last(const struct sock *sk,
1526 const struct sk_buff *skb)
1528 return skb_queue_is_last(&sk->sk_write_queue, skb);
1531 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1533 if (tcp_skb_is_last(sk, skb))
1534 sk->sk_send_head = NULL;
1536 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1539 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1541 if (sk->sk_send_head == skb_unlinked)
1542 sk->sk_send_head = NULL;
1543 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1544 tcp_sk(sk)->highest_sack = NULL;
1547 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1549 __skb_queue_tail(&sk->sk_write_queue, skb);
1552 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1554 __tcp_add_write_queue_tail(sk, skb);
1556 /* Queue it, remembering where we must start sending. */
1557 if (sk->sk_send_head == NULL) {
1558 sk->sk_send_head = skb;
1560 if (tcp_sk(sk)->highest_sack == NULL)
1561 tcp_sk(sk)->highest_sack = skb;
1565 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1567 __skb_queue_head(&sk->sk_write_queue, skb);
1570 /* Insert buff after skb on the write queue of sk. */
1571 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1572 struct sk_buff *buff,
1575 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1578 /* Insert new before skb on the write queue of sk. */
1579 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1580 struct sk_buff *skb,
1583 __skb_queue_before(&sk->sk_write_queue, skb, new);
1585 if (sk->sk_send_head == skb)
1586 sk->sk_send_head = new;
1589 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1591 __skb_unlink(skb, &sk->sk_write_queue);
1594 static inline bool tcp_write_queue_empty(struct sock *sk)
1596 return skb_queue_empty(&sk->sk_write_queue);
1599 static inline void tcp_push_pending_frames(struct sock *sk)
1601 if (tcp_send_head(sk)) {
1602 struct tcp_sock *tp = tcp_sk(sk);
1604 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1608 /* Start sequence of the skb just after the highest skb with SACKed
1609 * bit, valid only if sacked_out > 0 or when the caller has ensured
1610 * validity by itself.
1612 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1614 if (!tp->sacked_out)
1617 if (tp->highest_sack == NULL)
1620 return TCP_SKB_CB(tp->highest_sack)->seq;
1623 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1625 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1626 tcp_write_queue_next(sk, skb);
1629 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1631 return tcp_sk(sk)->highest_sack;
1634 static inline void tcp_highest_sack_reset(struct sock *sk)
1636 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1639 /* Called when old skb is about to be deleted and replaced by new skb */
1640 static inline void tcp_highest_sack_replace(struct sock *sk,
1641 struct sk_buff *old,
1642 struct sk_buff *new)
1644 if (old == tcp_highest_sack(sk))
1645 tcp_sk(sk)->highest_sack = new;
1648 /* Determines whether this is a thin stream (which may suffer from
1649 * increased latency). Used to trigger latency-reducing mechanisms.
1651 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1653 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1657 enum tcp_seq_states {
1658 TCP_SEQ_STATE_LISTENING,
1659 TCP_SEQ_STATE_ESTABLISHED,
1662 int tcp_seq_open(struct inode *inode, struct file *file);
1664 struct tcp_seq_afinfo {
1667 const struct file_operations *seq_fops;
1668 struct seq_operations seq_ops;
1671 struct tcp_iter_state {
1672 struct seq_net_private p;
1674 enum tcp_seq_states state;
1675 struct sock *syn_wait_sk;
1676 int bucket, offset, sbucket, num;
1680 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1681 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1683 extern struct request_sock_ops tcp_request_sock_ops;
1684 extern struct request_sock_ops tcp6_request_sock_ops;
1686 void tcp_v4_destroy_sock(struct sock *sk);
1688 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1689 netdev_features_t features);
1690 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1691 int tcp_gro_complete(struct sk_buff *skb);
1693 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1695 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1697 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1700 static inline bool tcp_stream_memory_free(const struct sock *sk)
1702 const struct tcp_sock *tp = tcp_sk(sk);
1703 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1705 return notsent_bytes < tcp_notsent_lowat(tp);
1708 #ifdef CONFIG_PROC_FS
1709 int tcp4_proc_init(void);
1710 void tcp4_proc_exit(void);
1713 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1714 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1715 const struct tcp_request_sock_ops *af_ops,
1716 struct sock *sk, struct sk_buff *skb);
1718 /* TCP af-specific functions */
1719 struct tcp_sock_af_ops {
1720 #ifdef CONFIG_TCP_MD5SIG
1721 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1722 const struct sock *addr_sk);
1723 int (*calc_md5_hash)(char *location,
1724 const struct tcp_md5sig_key *md5,
1725 const struct sock *sk,
1726 const struct sk_buff *skb);
1727 int (*md5_parse)(struct sock *sk,
1728 char __user *optval,
1733 struct tcp_request_sock_ops {
1735 #ifdef CONFIG_TCP_MD5SIG
1736 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1737 const struct sock *addr_sk);
1738 int (*calc_md5_hash) (char *location,
1739 const struct tcp_md5sig_key *md5,
1740 const struct sock *sk,
1741 const struct sk_buff *skb);
1743 void (*init_req)(struct request_sock *req,
1744 const struct sock *sk_listener,
1745 struct sk_buff *skb);
1746 #ifdef CONFIG_SYN_COOKIES
1747 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1750 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1751 const struct request_sock *req,
1753 __u32 (*init_seq)(const struct sk_buff *skb);
1754 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1755 struct flowi *fl, struct request_sock *req,
1756 struct tcp_fastopen_cookie *foc,
1760 #ifdef CONFIG_SYN_COOKIES
1761 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1762 const struct sock *sk, struct sk_buff *skb,
1765 tcp_synq_overflow(sk);
1766 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1767 return ops->cookie_init_seq(skb, mss);
1770 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1771 const struct sock *sk, struct sk_buff *skb,
1778 int tcpv4_offload_init(void);
1780 void tcp_v4_init(void);
1781 void tcp_init(void);
1783 /* tcp_recovery.c */
1785 /* Flags to enable various loss recovery features. See below */
1786 extern int sysctl_tcp_recovery;
1788 /* Use TCP RACK to detect (some) tail and retransmit losses */
1789 #define TCP_RACK_LOST_RETRANS 0x1
1791 extern int tcp_rack_mark_lost(struct sock *sk);
1793 extern void tcp_rack_advance(struct tcp_sock *tp,
1794 const struct skb_mstamp *xmit_time, u8 sacked);
1797 * Save and compile IPv4 options, return a pointer to it
1799 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1801 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1802 struct ip_options_rcu *dopt = NULL;
1805 int opt_size = sizeof(*dopt) + opt->optlen;
1807 dopt = kmalloc(opt_size, GFP_ATOMIC);
1808 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1816 /* locally generated TCP pure ACKs have skb->truesize == 2
1817 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1818 * This is much faster than dissecting the packet to find out.
1819 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1821 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1823 return skb->truesize == 2;
1826 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)