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
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/cryptohash.h>
31 #include <linux/kref.h>
32 #include <linux/ktime.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
49 extern struct inet_hashinfo tcp_hashinfo;
51 extern struct percpu_counter tcp_orphan_count;
52 void tcp_time_wait(struct sock *sk, int state, int timeo);
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
61 #define MAX_TCP_WINDOW 32767U
63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 1024
69 /* probing interval, default to 10 minutes as per RFC4821 */
70 #define TCP_PROBE_INTERVAL 600
72 /* Specify interval when tcp mtu probing will stop */
73 #define TCP_PROBE_THRESHOLD 8
75 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
76 #define TCP_FASTRETRANS_THRESH 3
78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
79 #define TCP_MAX_QUICKACKS 16U
82 #define TCP_URG_VALID 0x0100
83 #define TCP_URG_NOTYET 0x0200
84 #define TCP_URG_READ 0x0400
86 #define TCP_RETR1 3 /*
87 * This is how many retries it does before it
88 * tries to figure out if the gateway is
89 * down. Minimal RFC value is 3; it corresponds
90 * to ~3sec-8min depending on RTO.
93 #define TCP_RETR2 15 /*
94 * This should take at least
95 * 90 minutes to time out.
96 * RFC1122 says that the limit is 100 sec.
97 * 15 is ~13-30min depending on RTO.
100 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
101 * when active opening a connection.
102 * RFC1122 says the minimum retry MUST
103 * be at least 180secs. Nevertheless
104 * this value is corresponding to
105 * 63secs of retransmission with the
106 * current initial RTO.
109 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
110 * when passive opening a connection.
111 * This is corresponding to 31secs of
112 * retransmission with the current
116 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
117 * state, about 60 seconds */
118 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
119 /* BSD style FIN_WAIT2 deadlock breaker.
120 * It used to be 3min, new value is 60sec,
121 * to combine FIN-WAIT-2 timeout with
125 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
127 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
128 #define TCP_ATO_MIN ((unsigned)(HZ/25))
130 #define TCP_DELACK_MIN 4U
131 #define TCP_ATO_MIN 4U
133 #define TCP_RTO_MAX ((unsigned)(120*HZ))
134 #define TCP_RTO_MIN ((unsigned)(HZ/5))
135 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
136 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
137 * used as a fallback RTO for the
138 * initial data transmission if no
139 * valid RTT sample has been acquired,
140 * most likely due to retrans in 3WHS.
143 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
144 * for local resources.
147 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
148 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
149 #define TCP_KEEPALIVE_INTVL (75*HZ)
151 #define MAX_TCP_KEEPIDLE 32767
152 #define MAX_TCP_KEEPINTVL 32767
153 #define MAX_TCP_KEEPCNT 127
154 #define MAX_TCP_SYNCNT 127
156 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
158 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
159 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
160 * after this time. It should be equal
161 * (or greater than) TCP_TIMEWAIT_LEN
162 * to provide reliability equal to one
163 * provided by timewait state.
165 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
166 * timestamps. It must be less than
167 * minimal timewait lifetime.
173 #define TCPOPT_NOP 1 /* Padding */
174 #define TCPOPT_EOL 0 /* End of options */
175 #define TCPOPT_MSS 2 /* Segment size negotiating */
176 #define TCPOPT_WINDOW 3 /* Window scaling */
177 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
178 #define TCPOPT_SACK 5 /* SACK Block */
179 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
180 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
181 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
182 #define TCPOPT_EXP 254 /* Experimental */
183 /* Magic number to be after the option value for sharing TCP
184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
186 #define TCPOPT_FASTOPEN_MAGIC 0xF989
192 #define TCPOLEN_MSS 4
193 #define TCPOLEN_WINDOW 3
194 #define TCPOLEN_SACK_PERM 2
195 #define TCPOLEN_TIMESTAMP 10
196 #define TCPOLEN_MD5SIG 18
197 #define TCPOLEN_FASTOPEN_BASE 2
198 #define TCPOLEN_EXP_FASTOPEN_BASE 4
200 /* But this is what stacks really send out. */
201 #define TCPOLEN_TSTAMP_ALIGNED 12
202 #define TCPOLEN_WSCALE_ALIGNED 4
203 #define TCPOLEN_SACKPERM_ALIGNED 4
204 #define TCPOLEN_SACK_BASE 2
205 #define TCPOLEN_SACK_BASE_ALIGNED 4
206 #define TCPOLEN_SACK_PERBLOCK 8
207 #define TCPOLEN_MD5SIG_ALIGNED 20
208 #define TCPOLEN_MSS_ALIGNED 4
210 /* Flags in tp->nonagle */
211 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
212 #define TCP_NAGLE_CORK 2 /* Socket is corked */
213 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
215 /* TCP thin-stream limits */
216 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
218 /* TCP initial congestion window as per rfc6928 */
219 #define TCP_INIT_CWND 10
221 /* Bit Flags for sysctl_tcp_fastopen */
222 #define TFO_CLIENT_ENABLE 1
223 #define TFO_SERVER_ENABLE 2
224 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
226 /* Accept SYN data w/o any cookie option */
227 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
229 /* Force enable TFO on all listeners, i.e., not requiring the
230 * TCP_FASTOPEN socket option.
232 #define TFO_SERVER_WO_SOCKOPT1 0x400
234 extern struct inet_timewait_death_row tcp_death_row;
236 /* sysctl variables for tcp */
237 extern int sysctl_tcp_timestamps;
238 extern int sysctl_tcp_window_scaling;
239 extern int sysctl_tcp_sack;
240 extern int sysctl_tcp_fastopen;
241 extern int sysctl_tcp_retrans_collapse;
242 extern int sysctl_tcp_stdurg;
243 extern int sysctl_tcp_rfc1337;
244 extern int sysctl_tcp_abort_on_overflow;
245 extern int sysctl_tcp_max_orphans;
246 extern int sysctl_tcp_fack;
247 extern int sysctl_tcp_reordering;
248 extern int sysctl_tcp_max_reordering;
249 extern int sysctl_tcp_dsack;
250 extern long sysctl_tcp_mem[3];
251 extern int sysctl_tcp_wmem[3];
252 extern int sysctl_tcp_rmem[3];
253 extern int sysctl_tcp_app_win;
254 extern int sysctl_tcp_adv_win_scale;
255 extern int sysctl_tcp_frto;
256 extern int sysctl_tcp_low_latency;
257 extern int sysctl_tcp_nometrics_save;
258 extern int sysctl_tcp_moderate_rcvbuf;
259 extern int sysctl_tcp_tso_win_divisor;
260 extern int sysctl_tcp_workaround_signed_windows;
261 extern int sysctl_tcp_slow_start_after_idle;
262 extern int sysctl_tcp_thin_linear_timeouts;
263 extern int sysctl_tcp_thin_dupack;
264 extern int sysctl_tcp_early_retrans;
265 extern int sysctl_tcp_limit_output_bytes;
266 extern int sysctl_tcp_challenge_ack_limit;
267 extern int sysctl_tcp_min_tso_segs;
268 extern int sysctl_tcp_min_rtt_wlen;
269 extern int sysctl_tcp_autocorking;
270 extern int sysctl_tcp_invalid_ratelimit;
271 extern int sysctl_tcp_pacing_ss_ratio;
272 extern int sysctl_tcp_pacing_ca_ratio;
274 extern atomic_long_t tcp_memory_allocated;
275 extern struct percpu_counter tcp_sockets_allocated;
276 extern int tcp_memory_pressure;
278 /* optimized version of sk_under_memory_pressure() for TCP sockets */
279 static inline bool tcp_under_memory_pressure(const struct sock *sk)
281 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
282 mem_cgroup_under_socket_pressure(sk->sk_memcg))
285 return tcp_memory_pressure;
288 * The next routines deal with comparing 32 bit unsigned ints
289 * and worry about wraparound (automatic with unsigned arithmetic).
292 static inline bool before(__u32 seq1, __u32 seq2)
294 return (__s32)(seq1-seq2) < 0;
296 #define after(seq2, seq1) before(seq1, seq2)
298 /* is s2<=s1<=s3 ? */
299 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
301 return seq3 - seq2 >= seq1 - seq2;
304 static inline bool tcp_out_of_memory(struct sock *sk)
306 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
307 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
312 void sk_forced_mem_schedule(struct sock *sk, int size);
314 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
316 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
317 int orphans = percpu_counter_read_positive(ocp);
319 if (orphans << shift > sysctl_tcp_max_orphans) {
320 orphans = percpu_counter_sum_positive(ocp);
321 if (orphans << shift > sysctl_tcp_max_orphans)
327 bool tcp_check_oom(struct sock *sk, int shift);
330 extern struct proto tcp_prot;
332 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
333 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
334 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
335 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
337 void tcp_tasklet_init(void);
339 void tcp_v4_err(struct sk_buff *skb, u32);
341 void tcp_shutdown(struct sock *sk, int how);
343 void tcp_v4_early_demux(struct sk_buff *skb);
344 int tcp_v4_rcv(struct sk_buff *skb);
346 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
347 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
348 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
350 void tcp_release_cb(struct sock *sk);
351 void tcp_wfree(struct sk_buff *skb);
352 void tcp_write_timer_handler(struct sock *sk);
353 void tcp_delack_timer_handler(struct sock *sk);
354 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
355 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
356 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
357 const struct tcphdr *th, unsigned int len);
358 void tcp_rcv_space_adjust(struct sock *sk);
359 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
360 void tcp_twsk_destructor(struct sock *sk);
361 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
362 struct pipe_inode_info *pipe, size_t len,
365 static inline void tcp_dec_quickack_mode(struct sock *sk,
366 const unsigned int pkts)
368 struct inet_connection_sock *icsk = inet_csk(sk);
370 if (icsk->icsk_ack.quick) {
371 if (pkts >= icsk->icsk_ack.quick) {
372 icsk->icsk_ack.quick = 0;
373 /* Leaving quickack mode we deflate ATO. */
374 icsk->icsk_ack.ato = TCP_ATO_MIN;
376 icsk->icsk_ack.quick -= pkts;
381 #define TCP_ECN_QUEUE_CWR 2
382 #define TCP_ECN_DEMAND_CWR 4
383 #define TCP_ECN_SEEN 8
393 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
395 const struct tcphdr *th);
396 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
397 struct request_sock *req, bool fastopen);
398 int tcp_child_process(struct sock *parent, struct sock *child,
399 struct sk_buff *skb);
400 void tcp_enter_loss(struct sock *sk);
401 void tcp_clear_retrans(struct tcp_sock *tp);
402 void tcp_update_metrics(struct sock *sk);
403 void tcp_init_metrics(struct sock *sk);
404 void tcp_metrics_init(void);
405 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
406 bool paws_check, bool timestamps);
407 bool tcp_remember_stamp(struct sock *sk);
408 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
409 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
410 void tcp_disable_fack(struct tcp_sock *tp);
411 void tcp_close(struct sock *sk, long timeout);
412 void tcp_init_sock(struct sock *sk);
413 unsigned int tcp_poll(struct file *file, struct socket *sock,
414 struct poll_table_struct *wait);
415 int tcp_getsockopt(struct sock *sk, int level, int optname,
416 char __user *optval, int __user *optlen);
417 int tcp_setsockopt(struct sock *sk, int level, int optname,
418 char __user *optval, unsigned int optlen);
419 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
420 char __user *optval, int __user *optlen);
421 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
422 char __user *optval, unsigned int optlen);
423 void tcp_set_keepalive(struct sock *sk, int val);
424 void tcp_syn_ack_timeout(const struct request_sock *req);
425 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
426 int flags, int *addr_len);
427 void tcp_parse_options(const struct sk_buff *skb,
428 struct tcp_options_received *opt_rx,
429 int estab, struct tcp_fastopen_cookie *foc);
430 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
433 * TCP v4 functions exported for the inet6 API
436 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
437 void tcp_v4_mtu_reduced(struct sock *sk);
438 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
439 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
440 struct sock *tcp_create_openreq_child(const struct sock *sk,
441 struct request_sock *req,
442 struct sk_buff *skb);
443 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
444 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
445 struct request_sock *req,
446 struct dst_entry *dst,
447 struct request_sock *req_unhash,
449 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
450 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
451 int tcp_connect(struct sock *sk);
452 enum tcp_synack_type {
457 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
458 struct request_sock *req,
459 struct tcp_fastopen_cookie *foc,
460 enum tcp_synack_type synack_type);
461 int tcp_disconnect(struct sock *sk, int flags);
463 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
464 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
465 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
467 /* From syncookies.c */
468 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
469 struct request_sock *req,
470 struct dst_entry *dst);
471 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
473 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
474 #ifdef CONFIG_SYN_COOKIES
476 /* Syncookies use a monotonic timer which increments every 60 seconds.
477 * This counter is used both as a hash input and partially encoded into
478 * the cookie value. A cookie is only validated further if the delta
479 * between the current counter value and the encoded one is less than this,
480 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
481 * the counter advances immediately after a cookie is generated).
483 #define MAX_SYNCOOKIE_AGE 2
484 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
485 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
487 /* syncookies: remember time of last synqueue overflow
488 * But do not dirty this field too often (once per second is enough)
489 * It is racy as we do not hold a lock, but race is very minor.
491 static inline void tcp_synq_overflow(const struct sock *sk)
493 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
494 unsigned long now = jiffies;
496 if (time_after(now, last_overflow + HZ))
497 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
500 /* syncookies: no recent synqueue overflow on this listening socket? */
501 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
503 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
505 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
508 static inline u32 tcp_cookie_time(void)
510 u64 val = get_jiffies_64();
512 do_div(val, TCP_SYNCOOKIE_PERIOD);
516 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
518 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
519 __u32 cookie_init_timestamp(struct request_sock *req);
520 bool cookie_timestamp_decode(struct tcp_options_received *opt);
521 bool cookie_ecn_ok(const struct tcp_options_received *opt,
522 const struct net *net, const struct dst_entry *dst);
524 /* From net/ipv6/syncookies.c */
525 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
527 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
529 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
530 const struct tcphdr *th, u16 *mssp);
531 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
535 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
537 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
539 bool tcp_may_send_now(struct sock *sk);
540 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
541 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
542 void tcp_retransmit_timer(struct sock *sk);
543 void tcp_xmit_retransmit_queue(struct sock *);
544 void tcp_simple_retransmit(struct sock *);
545 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
546 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
548 void tcp_send_probe0(struct sock *);
549 void tcp_send_partial(struct sock *);
550 int tcp_write_wakeup(struct sock *, int mib);
551 void tcp_send_fin(struct sock *sk);
552 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
553 int tcp_send_synack(struct sock *);
554 void tcp_push_one(struct sock *, unsigned int mss_now);
555 void tcp_send_ack(struct sock *sk);
556 void tcp_send_delayed_ack(struct sock *sk);
557 void tcp_send_loss_probe(struct sock *sk);
558 bool tcp_schedule_loss_probe(struct sock *sk);
559 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
560 const struct sk_buff *next_skb);
563 void tcp_resume_early_retransmit(struct sock *sk);
564 void tcp_rearm_rto(struct sock *sk);
565 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
566 void tcp_reset(struct sock *sk);
567 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
568 void tcp_fin(struct sock *sk);
571 void tcp_init_xmit_timers(struct sock *);
572 static inline void tcp_clear_xmit_timers(struct sock *sk)
574 inet_csk_clear_xmit_timers(sk);
577 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
578 unsigned int tcp_current_mss(struct sock *sk);
580 /* Bound MSS / TSO packet size with the half of the window */
581 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
585 /* When peer uses tiny windows, there is no use in packetizing
586 * to sub-MSS pieces for the sake of SWS or making sure there
587 * are enough packets in the pipe for fast recovery.
589 * On the other hand, for extremely large MSS devices, handling
590 * smaller than MSS windows in this way does make sense.
592 if (tp->max_window > TCP_MSS_DEFAULT)
593 cutoff = (tp->max_window >> 1);
595 cutoff = tp->max_window;
597 if (cutoff && pktsize > cutoff)
598 return max_t(int, cutoff, 68U - tp->tcp_header_len);
604 void tcp_get_info(struct sock *, struct tcp_info *);
606 /* Read 'sendfile()'-style from a TCP socket */
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 (RB_EMPTY_ROOT(&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 /* Minimum RTT in usec. ~0 means not available. */
673 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
675 return minmax_get(&tp->rtt_min);
678 /* Compute the actual receive window we are currently advertising.
679 * Rcv_nxt can be after the window if our peer push more data
680 * than the offered window.
682 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
684 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
691 /* Choose a new window, without checks for shrinking, and without
692 * scaling applied to the result. The caller does these things
693 * if necessary. This is a "raw" window selection.
695 u32 __tcp_select_window(struct sock *sk);
697 void tcp_send_window_probe(struct sock *sk);
699 /* TCP timestamps are only 32-bits, this causes a slight
700 * complication on 64-bit systems since we store a snapshot
701 * of jiffies in the buffer control blocks below. We decided
702 * to use only the low 32-bits of jiffies and hide the ugly
703 * casts with the following macro.
705 #define tcp_time_stamp ((__u32)(jiffies))
707 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
709 return skb->skb_mstamp.stamp_jiffies;
713 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
715 #define TCPHDR_FIN 0x01
716 #define TCPHDR_SYN 0x02
717 #define TCPHDR_RST 0x04
718 #define TCPHDR_PSH 0x08
719 #define TCPHDR_ACK 0x10
720 #define TCPHDR_URG 0x20
721 #define TCPHDR_ECE 0x40
722 #define TCPHDR_CWR 0x80
724 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
726 /* This is what the send packet queuing engine uses to pass
727 * TCP per-packet control information to the transmission code.
728 * We also store the host-order sequence numbers in here too.
729 * This is 44 bytes if IPV6 is enabled.
730 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
733 __u32 seq; /* Starting sequence number */
734 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
736 /* Note : tcp_tw_isn is used in input path only
737 * (isn chosen by tcp_timewait_state_process())
739 * tcp_gso_segs/size are used in write queue only,
740 * cf tcp_skb_pcount()/tcp_skb_mss()
748 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
750 __u8 sacked; /* State flags for SACK/FACK. */
751 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
752 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
753 #define TCPCB_LOST 0x04 /* SKB is lost */
754 #define TCPCB_TAGBITS 0x07 /* All tag bits */
755 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
756 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
757 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
760 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
761 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
762 eor:1, /* Is skb MSG_EOR marked? */
764 __u32 ack_seq; /* Sequence number ACK'd */
767 /* There is space for up to 24 bytes */
768 __u32 in_flight:30,/* Bytes in flight at transmit */
769 is_app_limited:1, /* cwnd not fully used? */
771 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
773 /* start of send pipeline phase */
774 struct skb_mstamp first_tx_mstamp;
775 /* when we reached the "delivered" count */
776 struct skb_mstamp delivered_mstamp;
777 } tx; /* only used for outgoing skbs */
779 struct inet_skb_parm h4;
780 #if IS_ENABLED(CONFIG_IPV6)
781 struct inet6_skb_parm h6;
783 } header; /* For incoming skbs */
787 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
790 #if IS_ENABLED(CONFIG_IPV6)
791 /* This is the variant of inet6_iif() that must be used by TCP,
792 * as TCP moves IP6CB into a different location in skb->cb[]
794 static inline int tcp_v6_iif(const struct sk_buff *skb)
796 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
798 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
802 /* TCP_SKB_CB reference means this can not be used from early demux */
803 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
805 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
806 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
807 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
813 /* Due to TSO, an SKB can be composed of multiple actual
814 * packets. To keep these tracked properly, we use this.
816 static inline int tcp_skb_pcount(const struct sk_buff *skb)
818 return TCP_SKB_CB(skb)->tcp_gso_segs;
821 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
823 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
826 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
828 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
831 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
832 static inline int tcp_skb_mss(const struct sk_buff *skb)
834 return TCP_SKB_CB(skb)->tcp_gso_size;
837 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
839 return likely(!TCP_SKB_CB(skb)->eor);
842 /* Events passed to congestion control interface */
844 CA_EVENT_TX_START, /* first transmit when no packets in flight */
845 CA_EVENT_CWND_RESTART, /* congestion window restart */
846 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
847 CA_EVENT_LOSS, /* loss timeout */
848 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
849 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
850 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
851 CA_EVENT_NON_DELAYED_ACK,
854 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
855 enum tcp_ca_ack_event_flags {
856 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
857 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
858 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
862 * Interface for adding new TCP congestion control handlers
864 #define TCP_CA_NAME_MAX 16
865 #define TCP_CA_MAX 128
866 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
868 #define TCP_CA_UNSPEC 0
870 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
871 #define TCP_CONG_NON_RESTRICTED 0x1
872 /* Requires ECN/ECT set on all packets */
873 #define TCP_CONG_NEEDS_ECN 0x2
883 /* A rate sample measures the number of (original/retransmitted) data
884 * packets delivered "delivered" over an interval of time "interval_us".
885 * The tcp_rate.c code fills in the rate sample, and congestion
886 * control modules that define a cong_control function to run at the end
887 * of ACK processing can optionally chose to consult this sample when
888 * setting cwnd and pacing rate.
889 * A sample is invalid if "delivered" or "interval_us" is negative.
892 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
893 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
894 s32 delivered; /* number of packets delivered over interval */
895 long interval_us; /* time for tp->delivered to incr "delivered" */
896 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
897 int losses; /* number of packets marked lost upon ACK */
898 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
899 u32 prior_in_flight; /* in flight before this ACK */
900 bool is_app_limited; /* is sample from packet with bubble in pipe? */
901 bool is_retrans; /* is sample from retransmission? */
904 struct tcp_congestion_ops {
905 struct list_head list;
909 /* initialize private data (optional) */
910 void (*init)(struct sock *sk);
911 /* cleanup private data (optional) */
912 void (*release)(struct sock *sk);
914 /* return slow start threshold (required) */
915 u32 (*ssthresh)(struct sock *sk);
916 /* do new cwnd calculation (required) */
917 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
918 /* call before changing ca_state (optional) */
919 void (*set_state)(struct sock *sk, u8 new_state);
920 /* call when cwnd event occurs (optional) */
921 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
922 /* call when ack arrives (optional) */
923 void (*in_ack_event)(struct sock *sk, u32 flags);
924 /* new value of cwnd after loss (optional) */
925 u32 (*undo_cwnd)(struct sock *sk);
926 /* hook for packet ack accounting (optional) */
927 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
928 /* suggest number of segments for each skb to transmit (optional) */
929 u32 (*tso_segs_goal)(struct sock *sk);
930 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
931 u32 (*sndbuf_expand)(struct sock *sk);
932 /* call when packets are delivered to update cwnd and pacing rate,
933 * after all the ca_state processing. (optional)
935 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
936 /* get info for inet_diag (optional) */
937 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
938 union tcp_cc_info *info);
940 char name[TCP_CA_NAME_MAX];
941 struct module *owner;
944 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
945 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
947 void tcp_assign_congestion_control(struct sock *sk);
948 void tcp_init_congestion_control(struct sock *sk);
949 void tcp_cleanup_congestion_control(struct sock *sk);
950 int tcp_set_default_congestion_control(const char *name);
951 void tcp_get_default_congestion_control(char *name);
952 void tcp_get_available_congestion_control(char *buf, size_t len);
953 void tcp_get_allowed_congestion_control(char *buf, size_t len);
954 int tcp_set_allowed_congestion_control(char *allowed);
955 int tcp_set_congestion_control(struct sock *sk, const char *name);
956 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
957 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
959 u32 tcp_reno_ssthresh(struct sock *sk);
960 u32 tcp_reno_undo_cwnd(struct sock *sk);
961 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
962 extern struct tcp_congestion_ops tcp_reno;
964 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
965 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
967 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
969 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
975 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
977 const struct inet_connection_sock *icsk = inet_csk(sk);
979 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
982 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
984 struct inet_connection_sock *icsk = inet_csk(sk);
986 if (icsk->icsk_ca_ops->set_state)
987 icsk->icsk_ca_ops->set_state(sk, ca_state);
988 icsk->icsk_ca_state = ca_state;
991 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
993 const struct inet_connection_sock *icsk = inet_csk(sk);
995 if (icsk->icsk_ca_ops->cwnd_event)
996 icsk->icsk_ca_ops->cwnd_event(sk, event);
999 /* From tcp_rate.c */
1000 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1001 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1002 struct rate_sample *rs);
1003 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1004 struct skb_mstamp *now, struct rate_sample *rs);
1005 void tcp_rate_check_app_limited(struct sock *sk);
1007 /* These functions determine how the current flow behaves in respect of SACK
1008 * handling. SACK is negotiated with the peer, and therefore it can vary
1009 * between different flows.
1011 * tcp_is_sack - SACK enabled
1012 * tcp_is_reno - No SACK
1013 * tcp_is_fack - FACK enabled, implies SACK enabled
1015 static inline int tcp_is_sack(const struct tcp_sock *tp)
1017 return tp->rx_opt.sack_ok;
1020 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1022 return !tcp_is_sack(tp);
1025 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1027 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1030 static inline void tcp_enable_fack(struct tcp_sock *tp)
1032 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1035 /* TCP early-retransmit (ER) is similar to but more conservative than
1036 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
1038 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
1040 struct net *net = sock_net((struct sock *)tp);
1042 tp->do_early_retrans = sysctl_tcp_early_retrans &&
1043 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
1044 net->ipv4.sysctl_tcp_reordering == 3;
1047 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
1049 tp->do_early_retrans = 0;
1052 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1054 return tp->sacked_out + tp->lost_out;
1057 /* This determines how many packets are "in the network" to the best
1058 * of our knowledge. In many cases it is conservative, but where
1059 * detailed information is available from the receiver (via SACK
1060 * blocks etc.) we can make more aggressive calculations.
1062 * Use this for decisions involving congestion control, use just
1063 * tp->packets_out to determine if the send queue is empty or not.
1065 * Read this equation as:
1067 * "Packets sent once on transmission queue" MINUS
1068 * "Packets left network, but not honestly ACKed yet" PLUS
1069 * "Packets fast retransmitted"
1071 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1073 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1076 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1078 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1080 return tp->snd_cwnd < tp->snd_ssthresh;
1083 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1085 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1088 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1090 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1091 (1 << inet_csk(sk)->icsk_ca_state);
1094 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1095 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1098 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1100 const struct tcp_sock *tp = tcp_sk(sk);
1102 if (tcp_in_cwnd_reduction(sk))
1103 return tp->snd_ssthresh;
1105 return max(tp->snd_ssthresh,
1106 ((tp->snd_cwnd >> 1) +
1107 (tp->snd_cwnd >> 2)));
1110 /* Use define here intentionally to get WARN_ON location shown at the caller */
1111 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1113 void tcp_enter_cwr(struct sock *sk);
1114 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1116 /* The maximum number of MSS of available cwnd for which TSO defers
1117 * sending if not using sysctl_tcp_tso_win_divisor.
1119 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1124 /* Returns end sequence number of the receiver's advertised window */
1125 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1127 return tp->snd_una + tp->snd_wnd;
1130 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1131 * flexible approach. The RFC suggests cwnd should not be raised unless
1132 * it was fully used previously. And that's exactly what we do in
1133 * congestion avoidance mode. But in slow start we allow cwnd to grow
1134 * as long as the application has used half the cwnd.
1136 * cwnd is 10 (IW10), but application sends 9 frames.
1137 * We allow cwnd to reach 18 when all frames are ACKed.
1138 * This check is safe because it's as aggressive as slow start which already
1139 * risks 100% overshoot. The advantage is that we discourage application to
1140 * either send more filler packets or data to artificially blow up the cwnd
1141 * usage, and allow application-limited process to probe bw more aggressively.
1143 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1145 const struct tcp_sock *tp = tcp_sk(sk);
1147 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1148 if (tcp_in_slow_start(tp))
1149 return tp->snd_cwnd < 2 * tp->max_packets_out;
1151 return tp->is_cwnd_limited;
1154 /* Something is really bad, we could not queue an additional packet,
1155 * because qdisc is full or receiver sent a 0 window.
1156 * We do not want to add fuel to the fire, or abort too early,
1157 * so make sure the timer we arm now is at least 200ms in the future,
1158 * regardless of current icsk_rto value (as it could be ~2ms)
1160 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1162 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1165 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1166 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1167 unsigned long max_when)
1169 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1171 return (unsigned long)min_t(u64, when, max_when);
1174 static inline void tcp_check_probe_timer(struct sock *sk)
1176 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1177 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1178 tcp_probe0_base(sk), TCP_RTO_MAX);
1181 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1186 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1192 * Calculate(/check) TCP checksum
1194 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1195 __be32 daddr, __wsum base)
1197 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1200 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1202 return __skb_checksum_complete(skb);
1205 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1207 return !skb_csum_unnecessary(skb) &&
1208 __tcp_checksum_complete(skb);
1211 /* Prequeue for VJ style copy to user, combined with checksumming. */
1213 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1215 tp->ucopy.task = NULL;
1217 tp->ucopy.memory = 0;
1218 skb_queue_head_init(&tp->ucopy.prequeue);
1221 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1222 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1223 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1228 static const char *statename[]={
1229 "Unused","Established","Syn Sent","Syn Recv",
1230 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1231 "Close Wait","Last ACK","Listen","Closing"
1234 void tcp_set_state(struct sock *sk, int state);
1236 void tcp_done(struct sock *sk);
1238 int tcp_abort(struct sock *sk, int err);
1240 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1243 rx_opt->num_sacks = 0;
1246 u32 tcp_default_init_rwnd(u32 mss);
1247 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1249 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1251 struct tcp_sock *tp = tcp_sk(sk);
1254 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1256 delta = tcp_time_stamp - tp->lsndtime;
1257 if (delta > inet_csk(sk)->icsk_rto)
1258 tcp_cwnd_restart(sk, delta);
1261 /* Determine a window scaling and initial window to offer. */
1262 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1263 __u32 *window_clamp, int wscale_ok,
1264 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1266 static inline int tcp_win_from_space(int space)
1268 return sysctl_tcp_adv_win_scale<=0 ?
1269 (space>>(-sysctl_tcp_adv_win_scale)) :
1270 space - (space>>sysctl_tcp_adv_win_scale);
1273 /* Note: caller must be prepared to deal with negative returns */
1274 static inline int tcp_space(const struct sock *sk)
1276 return tcp_win_from_space(sk->sk_rcvbuf -
1277 atomic_read(&sk->sk_rmem_alloc));
1280 static inline int tcp_full_space(const struct sock *sk)
1282 return tcp_win_from_space(sk->sk_rcvbuf);
1285 extern void tcp_openreq_init_rwin(struct request_sock *req,
1286 const struct sock *sk_listener,
1287 const struct dst_entry *dst);
1289 void tcp_enter_memory_pressure(struct sock *sk);
1291 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1293 struct net *net = sock_net((struct sock *)tp);
1295 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1298 static inline int keepalive_time_when(const struct tcp_sock *tp)
1300 struct net *net = sock_net((struct sock *)tp);
1302 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1305 static inline int keepalive_probes(const struct tcp_sock *tp)
1307 struct net *net = sock_net((struct sock *)tp);
1309 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1312 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1314 const struct inet_connection_sock *icsk = &tp->inet_conn;
1316 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1317 tcp_time_stamp - tp->rcv_tstamp);
1320 static inline int tcp_fin_time(const struct sock *sk)
1322 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1323 const int rto = inet_csk(sk)->icsk_rto;
1325 if (fin_timeout < (rto << 2) - (rto >> 1))
1326 fin_timeout = (rto << 2) - (rto >> 1);
1331 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1334 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1336 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1339 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1340 * then following tcp messages have valid values. Ignore 0 value,
1341 * or else 'negative' tsval might forbid us to accept their packets.
1343 if (!rx_opt->ts_recent)
1348 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1351 if (tcp_paws_check(rx_opt, 0))
1354 /* RST segments are not recommended to carry timestamp,
1355 and, if they do, it is recommended to ignore PAWS because
1356 "their cleanup function should take precedence over timestamps."
1357 Certainly, it is mistake. It is necessary to understand the reasons
1358 of this constraint to relax it: if peer reboots, clock may go
1359 out-of-sync and half-open connections will not be reset.
1360 Actually, the problem would be not existing if all
1361 the implementations followed draft about maintaining clock
1362 via reboots. Linux-2.2 DOES NOT!
1364 However, we can relax time bounds for RST segments to MSL.
1366 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1371 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1372 int mib_idx, u32 *last_oow_ack_time);
1374 static inline void tcp_mib_init(struct net *net)
1377 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1378 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1379 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1380 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1384 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1386 tp->lost_skb_hint = NULL;
1389 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1391 tcp_clear_retrans_hints_partial(tp);
1392 tp->retransmit_skb_hint = NULL;
1395 union tcp_md5_addr {
1397 #if IS_ENABLED(CONFIG_IPV6)
1402 /* - key database */
1403 struct tcp_md5sig_key {
1404 struct hlist_node node;
1406 u8 family; /* AF_INET or AF_INET6 */
1407 union tcp_md5_addr addr;
1408 u8 key[TCP_MD5SIG_MAXKEYLEN];
1409 struct rcu_head rcu;
1413 struct tcp_md5sig_info {
1414 struct hlist_head head;
1415 struct rcu_head rcu;
1418 /* - pseudo header */
1419 struct tcp4_pseudohdr {
1427 struct tcp6_pseudohdr {
1428 struct in6_addr saddr;
1429 struct in6_addr daddr;
1431 __be32 protocol; /* including padding */
1434 union tcp_md5sum_block {
1435 struct tcp4_pseudohdr ip4;
1436 #if IS_ENABLED(CONFIG_IPV6)
1437 struct tcp6_pseudohdr ip6;
1441 /* - pool: digest algorithm, hash description and scratch buffer */
1442 struct tcp_md5sig_pool {
1443 struct ahash_request *md5_req;
1448 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1449 const struct sock *sk, const struct sk_buff *skb);
1450 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1451 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1452 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1454 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1455 const struct sock *addr_sk);
1457 #ifdef CONFIG_TCP_MD5SIG
1458 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1459 const union tcp_md5_addr *addr,
1461 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1463 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1464 const union tcp_md5_addr *addr,
1469 #define tcp_twsk_md5_key(twsk) NULL
1472 bool tcp_alloc_md5sig_pool(void);
1474 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1475 static inline void tcp_put_md5sig_pool(void)
1480 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1481 unsigned int header_len);
1482 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1483 const struct tcp_md5sig_key *key);
1485 /* From tcp_fastopen.c */
1486 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1487 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1488 unsigned long *last_syn_loss);
1489 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1490 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1492 struct tcp_fastopen_request {
1493 /* Fast Open cookie. Size 0 means a cookie request */
1494 struct tcp_fastopen_cookie cookie;
1495 struct msghdr *data; /* data in MSG_FASTOPEN */
1497 int copied; /* queued in tcp_connect() */
1499 void tcp_free_fastopen_req(struct tcp_sock *tp);
1501 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1502 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1503 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1504 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1505 struct request_sock *req,
1506 struct tcp_fastopen_cookie *foc,
1507 struct dst_entry *dst);
1508 void tcp_fastopen_init_key_once(bool publish);
1509 #define TCP_FASTOPEN_KEY_LENGTH 16
1511 /* Fastopen key context */
1512 struct tcp_fastopen_context {
1513 struct crypto_cipher *tfm;
1514 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1515 struct rcu_head rcu;
1518 /* Latencies incurred by various limits for a sender. They are
1519 * chronograph-like stats that are mutually exclusive.
1523 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1524 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1525 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1529 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1530 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1532 /* write queue abstraction */
1533 static inline void tcp_write_queue_purge(struct sock *sk)
1535 struct sk_buff *skb;
1537 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1538 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1539 sk_wmem_free_skb(sk, skb);
1541 tcp_clear_all_retrans_hints(tcp_sk(sk));
1544 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1546 return skb_peek(&sk->sk_write_queue);
1549 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1551 return skb_peek_tail(&sk->sk_write_queue);
1554 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1555 const struct sk_buff *skb)
1557 return skb_queue_next(&sk->sk_write_queue, skb);
1560 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1561 const struct sk_buff *skb)
1563 return skb_queue_prev(&sk->sk_write_queue, skb);
1566 #define tcp_for_write_queue(skb, sk) \
1567 skb_queue_walk(&(sk)->sk_write_queue, skb)
1569 #define tcp_for_write_queue_from(skb, sk) \
1570 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1572 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1573 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1575 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1577 return sk->sk_send_head;
1580 static inline bool tcp_skb_is_last(const struct sock *sk,
1581 const struct sk_buff *skb)
1583 return skb_queue_is_last(&sk->sk_write_queue, skb);
1586 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1588 if (tcp_skb_is_last(sk, skb))
1589 sk->sk_send_head = NULL;
1591 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1594 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1596 if (sk->sk_send_head == skb_unlinked) {
1597 sk->sk_send_head = NULL;
1598 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1600 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1601 tcp_sk(sk)->highest_sack = NULL;
1604 static inline void tcp_init_send_head(struct sock *sk)
1606 sk->sk_send_head = NULL;
1609 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1611 __skb_queue_tail(&sk->sk_write_queue, skb);
1614 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1616 __tcp_add_write_queue_tail(sk, skb);
1618 /* Queue it, remembering where we must start sending. */
1619 if (sk->sk_send_head == NULL) {
1620 sk->sk_send_head = skb;
1621 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1623 if (tcp_sk(sk)->highest_sack == NULL)
1624 tcp_sk(sk)->highest_sack = skb;
1628 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1630 __skb_queue_head(&sk->sk_write_queue, skb);
1633 /* Insert buff after skb on the write queue of sk. */
1634 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1635 struct sk_buff *buff,
1638 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1641 /* Insert new before skb on the write queue of sk. */
1642 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1643 struct sk_buff *skb,
1646 __skb_queue_before(&sk->sk_write_queue, skb, new);
1648 if (sk->sk_send_head == skb)
1649 sk->sk_send_head = new;
1652 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1654 __skb_unlink(skb, &sk->sk_write_queue);
1657 static inline bool tcp_write_queue_empty(struct sock *sk)
1659 return skb_queue_empty(&sk->sk_write_queue);
1662 static inline void tcp_push_pending_frames(struct sock *sk)
1664 if (tcp_send_head(sk)) {
1665 struct tcp_sock *tp = tcp_sk(sk);
1667 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1671 /* Start sequence of the skb just after the highest skb with SACKed
1672 * bit, valid only if sacked_out > 0 or when the caller has ensured
1673 * validity by itself.
1675 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1677 if (!tp->sacked_out)
1680 if (tp->highest_sack == NULL)
1683 return TCP_SKB_CB(tp->highest_sack)->seq;
1686 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1688 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1689 tcp_write_queue_next(sk, skb);
1692 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1694 return tcp_sk(sk)->highest_sack;
1697 static inline void tcp_highest_sack_reset(struct sock *sk)
1699 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1702 /* Called when old skb is about to be deleted (to be combined with new skb) */
1703 static inline void tcp_highest_sack_combine(struct sock *sk,
1704 struct sk_buff *old,
1705 struct sk_buff *new)
1707 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1708 tcp_sk(sk)->highest_sack = new;
1711 /* This helper checks if socket has IP_TRANSPARENT set */
1712 static inline bool inet_sk_transparent(const struct sock *sk)
1714 switch (sk->sk_state) {
1716 return inet_twsk(sk)->tw_transparent;
1717 case TCP_NEW_SYN_RECV:
1718 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1720 return inet_sk(sk)->transparent;
1723 /* Determines whether this is a thin stream (which may suffer from
1724 * increased latency). Used to trigger latency-reducing mechanisms.
1726 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1728 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1732 enum tcp_seq_states {
1733 TCP_SEQ_STATE_LISTENING,
1734 TCP_SEQ_STATE_ESTABLISHED,
1737 int tcp_seq_open(struct inode *inode, struct file *file);
1739 struct tcp_seq_afinfo {
1742 const struct file_operations *seq_fops;
1743 struct seq_operations seq_ops;
1746 struct tcp_iter_state {
1747 struct seq_net_private p;
1749 enum tcp_seq_states state;
1750 struct sock *syn_wait_sk;
1751 int bucket, offset, sbucket, num;
1755 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1756 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1758 extern struct request_sock_ops tcp_request_sock_ops;
1759 extern struct request_sock_ops tcp6_request_sock_ops;
1761 void tcp_v4_destroy_sock(struct sock *sk);
1763 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1764 netdev_features_t features);
1765 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1766 int tcp_gro_complete(struct sk_buff *skb);
1768 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1770 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1772 struct net *net = sock_net((struct sock *)tp);
1773 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1776 static inline bool tcp_stream_memory_free(const struct sock *sk)
1778 const struct tcp_sock *tp = tcp_sk(sk);
1779 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1781 return notsent_bytes < tcp_notsent_lowat(tp);
1784 #ifdef CONFIG_PROC_FS
1785 int tcp4_proc_init(void);
1786 void tcp4_proc_exit(void);
1789 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1790 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1791 const struct tcp_request_sock_ops *af_ops,
1792 struct sock *sk, struct sk_buff *skb);
1794 /* TCP af-specific functions */
1795 struct tcp_sock_af_ops {
1796 #ifdef CONFIG_TCP_MD5SIG
1797 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1798 const struct sock *addr_sk);
1799 int (*calc_md5_hash)(char *location,
1800 const struct tcp_md5sig_key *md5,
1801 const struct sock *sk,
1802 const struct sk_buff *skb);
1803 int (*md5_parse)(struct sock *sk,
1804 char __user *optval,
1809 struct tcp_request_sock_ops {
1811 #ifdef CONFIG_TCP_MD5SIG
1812 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1813 const struct sock *addr_sk);
1814 int (*calc_md5_hash) (char *location,
1815 const struct tcp_md5sig_key *md5,
1816 const struct sock *sk,
1817 const struct sk_buff *skb);
1819 void (*init_req)(struct request_sock *req,
1820 const struct sock *sk_listener,
1821 struct sk_buff *skb);
1822 #ifdef CONFIG_SYN_COOKIES
1823 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1826 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1827 const struct request_sock *req,
1829 __u32 (*init_seq)(const struct sk_buff *skb, u32 *tsoff);
1830 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1831 struct flowi *fl, struct request_sock *req,
1832 struct tcp_fastopen_cookie *foc,
1833 enum tcp_synack_type synack_type);
1836 #ifdef CONFIG_SYN_COOKIES
1837 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1838 const struct sock *sk, struct sk_buff *skb,
1841 tcp_synq_overflow(sk);
1842 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1843 return ops->cookie_init_seq(skb, mss);
1846 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1847 const struct sock *sk, struct sk_buff *skb,
1854 int tcpv4_offload_init(void);
1856 void tcp_v4_init(void);
1857 void tcp_init(void);
1859 /* tcp_recovery.c */
1861 /* Flags to enable various loss recovery features. See below */
1862 extern int sysctl_tcp_recovery;
1864 /* Use TCP RACK to detect (some) tail and retransmit losses */
1865 #define TCP_RACK_LOST_RETRANS 0x1
1867 extern int tcp_rack_mark_lost(struct sock *sk);
1869 extern void tcp_rack_advance(struct tcp_sock *tp,
1870 const struct skb_mstamp *xmit_time, u8 sacked);
1873 * Save and compile IPv4 options, return a pointer to it
1875 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1877 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1878 struct ip_options_rcu *dopt = NULL;
1881 int opt_size = sizeof(*dopt) + opt->optlen;
1883 dopt = kmalloc(opt_size, GFP_ATOMIC);
1884 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1892 /* locally generated TCP pure ACKs have skb->truesize == 2
1893 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1894 * This is much faster than dissecting the packet to find out.
1895 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1897 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1899 return skb->truesize == 2;
1902 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1907 static inline int tcp_inq(struct sock *sk)
1909 struct tcp_sock *tp = tcp_sk(sk);
1912 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1914 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1916 before(tp->urg_seq, tp->copied_seq) ||
1917 !before(tp->urg_seq, tp->rcv_nxt)) {
1919 answ = tp->rcv_nxt - tp->copied_seq;
1921 /* Subtract 1, if FIN was received */
1922 if (answ && sock_flag(sk, SOCK_DONE))
1925 answ = tp->urg_seq - tp->copied_seq;
1931 int tcp_peek_len(struct socket *sock);
1933 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1937 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1938 tp->segs_in += segs_in;
1939 if (skb->len > tcp_hdrlen(skb))
1940 tp->data_segs_in += segs_in;
1944 * TCP listen path runs lockless.
1945 * We forced "struct sock" to be const qualified to make sure
1946 * we don't modify one of its field by mistake.
1947 * Here, we increment sk_drops which is an atomic_t, so we can safely
1948 * make sock writable again.
1950 static inline void tcp_listendrop(const struct sock *sk)
1952 atomic_inc(&((struct sock *)sk)->sk_drops);
1953 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);