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/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. SOCKOPT1/2 determine how to set max_qlen.
232 #define TFO_SERVER_WO_SOCKOPT1 0x400
233 #define TFO_SERVER_WO_SOCKOPT2 0x800
235 extern struct inet_timewait_death_row tcp_death_row;
237 /* sysctl variables for tcp */
238 extern int sysctl_tcp_timestamps;
239 extern int sysctl_tcp_window_scaling;
240 extern int sysctl_tcp_sack;
241 extern int sysctl_tcp_fastopen;
242 extern int sysctl_tcp_retrans_collapse;
243 extern int sysctl_tcp_stdurg;
244 extern int sysctl_tcp_rfc1337;
245 extern int sysctl_tcp_abort_on_overflow;
246 extern int sysctl_tcp_max_orphans;
247 extern int sysctl_tcp_fack;
248 extern int sysctl_tcp_reordering;
249 extern int sysctl_tcp_max_reordering;
250 extern int sysctl_tcp_dsack;
251 extern long sysctl_tcp_mem[3];
252 extern int sysctl_tcp_wmem[3];
253 extern int sysctl_tcp_rmem[3];
254 extern int sysctl_tcp_app_win;
255 extern int sysctl_tcp_adv_win_scale;
256 extern int sysctl_tcp_tw_reuse;
257 extern int sysctl_tcp_frto;
258 extern int sysctl_tcp_low_latency;
259 extern int sysctl_tcp_nometrics_save;
260 extern int sysctl_tcp_moderate_rcvbuf;
261 extern int sysctl_tcp_tso_win_divisor;
262 extern int sysctl_tcp_workaround_signed_windows;
263 extern int sysctl_tcp_slow_start_after_idle;
264 extern int sysctl_tcp_thin_linear_timeouts;
265 extern int sysctl_tcp_thin_dupack;
266 extern int sysctl_tcp_early_retrans;
267 extern int sysctl_tcp_limit_output_bytes;
268 extern int sysctl_tcp_challenge_ack_limit;
269 extern int sysctl_tcp_min_tso_segs;
270 extern int sysctl_tcp_min_rtt_wlen;
271 extern int sysctl_tcp_autocorking;
272 extern int sysctl_tcp_invalid_ratelimit;
273 extern int sysctl_tcp_pacing_ss_ratio;
274 extern int sysctl_tcp_pacing_ca_ratio;
276 extern atomic_long_t tcp_memory_allocated;
277 extern struct percpu_counter tcp_sockets_allocated;
278 extern int tcp_memory_pressure;
280 /* optimized version of sk_under_memory_pressure() for TCP sockets */
281 static inline bool tcp_under_memory_pressure(const struct sock *sk)
283 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
284 mem_cgroup_under_socket_pressure(sk->sk_memcg))
287 return tcp_memory_pressure;
290 * The next routines deal with comparing 32 bit unsigned ints
291 * and worry about wraparound (automatic with unsigned arithmetic).
294 static inline bool before(__u32 seq1, __u32 seq2)
296 return (__s32)(seq1-seq2) < 0;
298 #define after(seq2, seq1) before(seq1, seq2)
300 /* is s2<=s1<=s3 ? */
301 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
303 return seq3 - seq2 >= seq1 - seq2;
306 static inline bool tcp_out_of_memory(struct sock *sk)
308 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
309 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
314 void sk_forced_mem_schedule(struct sock *sk, int size);
316 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
318 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
319 int orphans = percpu_counter_read_positive(ocp);
321 if (orphans << shift > sysctl_tcp_max_orphans) {
322 orphans = percpu_counter_sum_positive(ocp);
323 if (orphans << shift > sysctl_tcp_max_orphans)
329 bool tcp_check_oom(struct sock *sk, int shift);
332 extern struct proto tcp_prot;
334 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
335 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
336 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
337 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
339 void tcp_tasklet_init(void);
341 void tcp_v4_err(struct sk_buff *skb, u32);
343 void tcp_shutdown(struct sock *sk, int how);
345 void tcp_v4_early_demux(struct sk_buff *skb);
346 int tcp_v4_rcv(struct sk_buff *skb);
348 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
349 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
350 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
352 void tcp_release_cb(struct sock *sk);
353 void tcp_wfree(struct sk_buff *skb);
354 void tcp_write_timer_handler(struct sock *sk);
355 void tcp_delack_timer_handler(struct sock *sk);
356 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
357 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
358 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
359 const struct tcphdr *th, unsigned int len);
360 void tcp_rcv_space_adjust(struct sock *sk);
361 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
362 void tcp_twsk_destructor(struct sock *sk);
363 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
364 struct pipe_inode_info *pipe, size_t len,
367 static inline void tcp_dec_quickack_mode(struct sock *sk,
368 const unsigned int pkts)
370 struct inet_connection_sock *icsk = inet_csk(sk);
372 if (icsk->icsk_ack.quick) {
373 if (pkts >= icsk->icsk_ack.quick) {
374 icsk->icsk_ack.quick = 0;
375 /* Leaving quickack mode we deflate ATO. */
376 icsk->icsk_ack.ato = TCP_ATO_MIN;
378 icsk->icsk_ack.quick -= pkts;
383 #define TCP_ECN_QUEUE_CWR 2
384 #define TCP_ECN_DEMAND_CWR 4
385 #define TCP_ECN_SEEN 8
395 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
397 const struct tcphdr *th);
398 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
399 struct request_sock *req, bool fastopen);
400 int tcp_child_process(struct sock *parent, struct sock *child,
401 struct sk_buff *skb);
402 void tcp_enter_loss(struct sock *sk);
403 void tcp_clear_retrans(struct tcp_sock *tp);
404 void tcp_update_metrics(struct sock *sk);
405 void tcp_init_metrics(struct sock *sk);
406 void tcp_metrics_init(void);
407 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
408 bool paws_check, bool timestamps);
409 bool tcp_remember_stamp(struct sock *sk);
410 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
411 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
412 void tcp_disable_fack(struct tcp_sock *tp);
413 void tcp_close(struct sock *sk, long timeout);
414 void tcp_init_sock(struct sock *sk);
415 unsigned int tcp_poll(struct file *file, struct socket *sock,
416 struct poll_table_struct *wait);
417 int tcp_getsockopt(struct sock *sk, int level, int optname,
418 char __user *optval, int __user *optlen);
419 int tcp_setsockopt(struct sock *sk, int level, int optname,
420 char __user *optval, unsigned int optlen);
421 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
422 char __user *optval, int __user *optlen);
423 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
424 char __user *optval, unsigned int optlen);
425 void tcp_set_keepalive(struct sock *sk, int val);
426 void tcp_syn_ack_timeout(const struct request_sock *req);
427 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
428 int flags, int *addr_len);
429 void tcp_parse_options(const struct sk_buff *skb,
430 struct tcp_options_received *opt_rx,
431 int estab, struct tcp_fastopen_cookie *foc);
432 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
435 * TCP v4 functions exported for the inet6 API
438 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
439 void tcp_v4_mtu_reduced(struct sock *sk);
440 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
441 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
442 struct sock *tcp_create_openreq_child(const struct sock *sk,
443 struct request_sock *req,
444 struct sk_buff *skb);
445 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
446 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
447 struct request_sock *req,
448 struct dst_entry *dst,
449 struct request_sock *req_unhash,
451 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
452 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
453 int tcp_connect(struct sock *sk);
454 enum tcp_synack_type {
459 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
460 struct request_sock *req,
461 struct tcp_fastopen_cookie *foc,
462 enum tcp_synack_type synack_type);
463 int tcp_disconnect(struct sock *sk, int flags);
465 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
466 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
467 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
469 /* From syncookies.c */
470 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
471 struct request_sock *req,
472 struct dst_entry *dst);
473 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
475 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
476 #ifdef CONFIG_SYN_COOKIES
478 /* Syncookies use a monotonic timer which increments every 60 seconds.
479 * This counter is used both as a hash input and partially encoded into
480 * the cookie value. A cookie is only validated further if the delta
481 * between the current counter value and the encoded one is less than this,
482 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
483 * the counter advances immediately after a cookie is generated).
485 #define MAX_SYNCOOKIE_AGE 2
486 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
487 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
489 /* syncookies: remember time of last synqueue overflow
490 * But do not dirty this field too often (once per second is enough)
491 * It is racy as we do not hold a lock, but race is very minor.
493 static inline void tcp_synq_overflow(const struct sock *sk)
495 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
496 unsigned long now = jiffies;
498 if (time_after(now, last_overflow + HZ))
499 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
502 /* syncookies: no recent synqueue overflow on this listening socket? */
503 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
505 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
507 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
510 static inline u32 tcp_cookie_time(void)
512 u64 val = get_jiffies_64();
514 do_div(val, TCP_SYNCOOKIE_PERIOD);
518 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
520 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
521 __u32 cookie_init_timestamp(struct request_sock *req);
522 bool cookie_timestamp_decode(struct tcp_options_received *opt);
523 bool cookie_ecn_ok(const struct tcp_options_received *opt,
524 const struct net *net, const struct dst_entry *dst);
526 /* From net/ipv6/syncookies.c */
527 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
529 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
531 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
532 const struct tcphdr *th, u16 *mssp);
533 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
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 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
608 unsigned int, size_t);
609 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
610 sk_read_actor_t recv_actor);
612 void tcp_initialize_rcv_mss(struct sock *sk);
614 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
615 int tcp_mss_to_mtu(struct sock *sk, int mss);
616 void tcp_mtup_init(struct sock *sk);
617 void tcp_init_buffer_space(struct sock *sk);
619 static inline void tcp_bound_rto(const struct sock *sk)
621 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
622 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
625 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
627 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
630 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
632 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
633 ntohl(TCP_FLAG_ACK) |
637 static inline void tcp_fast_path_on(struct tcp_sock *tp)
639 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
642 static inline void tcp_fast_path_check(struct sock *sk)
644 struct tcp_sock *tp = tcp_sk(sk);
646 if (skb_queue_empty(&tp->out_of_order_queue) &&
648 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
650 tcp_fast_path_on(tp);
653 /* Compute the actual rto_min value */
654 static inline u32 tcp_rto_min(struct sock *sk)
656 const struct dst_entry *dst = __sk_dst_get(sk);
657 u32 rto_min = TCP_RTO_MIN;
659 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
660 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
664 static inline u32 tcp_rto_min_us(struct sock *sk)
666 return jiffies_to_usecs(tcp_rto_min(sk));
669 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
671 return dst_metric_locked(dst, RTAX_CC_ALGO);
674 /* Minimum RTT in usec. ~0 means not available. */
675 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
677 return tp->rtt_min[0].rtt;
680 /* Compute the actual receive window we are currently advertising.
681 * Rcv_nxt can be after the window if our peer push more data
682 * than the offered window.
684 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
686 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
693 /* Choose a new window, without checks for shrinking, and without
694 * scaling applied to the result. The caller does these things
695 * if necessary. This is a "raw" window selection.
697 u32 __tcp_select_window(struct sock *sk);
699 void tcp_send_window_probe(struct sock *sk);
701 /* TCP timestamps are only 32-bits, this causes a slight
702 * complication on 64-bit systems since we store a snapshot
703 * of jiffies in the buffer control blocks below. We decided
704 * to use only the low 32-bits of jiffies and hide the ugly
705 * casts with the following macro.
707 #define tcp_time_stamp ((__u32)(jiffies))
709 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
711 return skb->skb_mstamp.stamp_jiffies;
715 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
717 #define TCPHDR_FIN 0x01
718 #define TCPHDR_SYN 0x02
719 #define TCPHDR_RST 0x04
720 #define TCPHDR_PSH 0x08
721 #define TCPHDR_ACK 0x10
722 #define TCPHDR_URG 0x20
723 #define TCPHDR_ECE 0x40
724 #define TCPHDR_CWR 0x80
726 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
728 /* This is what the send packet queuing engine uses to pass
729 * TCP per-packet control information to the transmission code.
730 * We also store the host-order sequence numbers in here too.
731 * This is 44 bytes if IPV6 is enabled.
732 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
735 __u32 seq; /* Starting sequence number */
736 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
738 /* Note : tcp_tw_isn is used in input path only
739 * (isn chosen by tcp_timewait_state_process())
741 * tcp_gso_segs/size are used in write queue only,
742 * cf tcp_skb_pcount()/tcp_skb_mss()
750 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
752 __u8 sacked; /* State flags for SACK/FACK. */
753 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
754 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
755 #define TCPCB_LOST 0x04 /* SKB is lost */
756 #define TCPCB_TAGBITS 0x07 /* All tag bits */
757 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
758 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
759 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
762 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
763 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
764 eor:1, /* Is skb MSG_EOR marked? */
766 __u32 ack_seq; /* Sequence number ACK'd */
769 /* There is space for up to 20 bytes */
770 __u32 in_flight;/* Bytes in flight when packet sent */
771 } tx; /* only used for outgoing skbs */
773 struct inet_skb_parm h4;
774 #if IS_ENABLED(CONFIG_IPV6)
775 struct inet6_skb_parm h6;
777 } header; /* For incoming skbs */
781 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
784 #if IS_ENABLED(CONFIG_IPV6)
785 /* This is the variant of inet6_iif() that must be used by TCP,
786 * as TCP moves IP6CB into a different location in skb->cb[]
788 static inline int tcp_v6_iif(const struct sk_buff *skb)
790 bool l3_slave = skb_l3mdev_slave(TCP_SKB_CB(skb)->header.h6.flags);
792 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
796 /* Due to TSO, an SKB can be composed of multiple actual
797 * packets. To keep these tracked properly, we use this.
799 static inline int tcp_skb_pcount(const struct sk_buff *skb)
801 return TCP_SKB_CB(skb)->tcp_gso_segs;
804 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
806 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
809 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
811 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
814 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
815 static inline int tcp_skb_mss(const struct sk_buff *skb)
817 return TCP_SKB_CB(skb)->tcp_gso_size;
820 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
822 return likely(!TCP_SKB_CB(skb)->eor);
825 /* Events passed to congestion control interface */
827 CA_EVENT_TX_START, /* first transmit when no packets in flight */
828 CA_EVENT_CWND_RESTART, /* congestion window restart */
829 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
830 CA_EVENT_LOSS, /* loss timeout */
831 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
832 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
833 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
834 CA_EVENT_NON_DELAYED_ACK,
837 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
838 enum tcp_ca_ack_event_flags {
839 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
840 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
841 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
845 * Interface for adding new TCP congestion control handlers
847 #define TCP_CA_NAME_MAX 16
848 #define TCP_CA_MAX 128
849 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
851 #define TCP_CA_UNSPEC 0
853 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
854 #define TCP_CONG_NON_RESTRICTED 0x1
855 /* Requires ECN/ECT set on all packets */
856 #define TCP_CONG_NEEDS_ECN 0x2
866 struct tcp_congestion_ops {
867 struct list_head list;
871 /* initialize private data (optional) */
872 void (*init)(struct sock *sk);
873 /* cleanup private data (optional) */
874 void (*release)(struct sock *sk);
876 /* return slow start threshold (required) */
877 u32 (*ssthresh)(struct sock *sk);
878 /* do new cwnd calculation (required) */
879 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
880 /* call before changing ca_state (optional) */
881 void (*set_state)(struct sock *sk, u8 new_state);
882 /* call when cwnd event occurs (optional) */
883 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
884 /* call when ack arrives (optional) */
885 void (*in_ack_event)(struct sock *sk, u32 flags);
886 /* new value of cwnd after loss (optional) */
887 u32 (*undo_cwnd)(struct sock *sk);
888 /* hook for packet ack accounting (optional) */
889 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
890 /* get info for inet_diag (optional) */
891 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
892 union tcp_cc_info *info);
894 char name[TCP_CA_NAME_MAX];
895 struct module *owner;
898 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
899 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
901 void tcp_assign_congestion_control(struct sock *sk);
902 void tcp_init_congestion_control(struct sock *sk);
903 void tcp_cleanup_congestion_control(struct sock *sk);
904 int tcp_set_default_congestion_control(const char *name);
905 void tcp_get_default_congestion_control(char *name);
906 void tcp_get_available_congestion_control(char *buf, size_t len);
907 void tcp_get_allowed_congestion_control(char *buf, size_t len);
908 int tcp_set_allowed_congestion_control(char *allowed);
909 int tcp_set_congestion_control(struct sock *sk, const char *name);
910 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
911 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
913 u32 tcp_reno_ssthresh(struct sock *sk);
914 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
915 extern struct tcp_congestion_ops tcp_reno;
917 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
918 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
920 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
922 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
928 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
930 const struct inet_connection_sock *icsk = inet_csk(sk);
932 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
935 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
937 struct inet_connection_sock *icsk = inet_csk(sk);
939 if (icsk->icsk_ca_ops->set_state)
940 icsk->icsk_ca_ops->set_state(sk, ca_state);
941 icsk->icsk_ca_state = ca_state;
944 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
946 const struct inet_connection_sock *icsk = inet_csk(sk);
948 if (icsk->icsk_ca_ops->cwnd_event)
949 icsk->icsk_ca_ops->cwnd_event(sk, event);
952 /* These functions determine how the current flow behaves in respect of SACK
953 * handling. SACK is negotiated with the peer, and therefore it can vary
954 * between different flows.
956 * tcp_is_sack - SACK enabled
957 * tcp_is_reno - No SACK
958 * tcp_is_fack - FACK enabled, implies SACK enabled
960 static inline int tcp_is_sack(const struct tcp_sock *tp)
962 return tp->rx_opt.sack_ok;
965 static inline bool tcp_is_reno(const struct tcp_sock *tp)
967 return !tcp_is_sack(tp);
970 static inline bool tcp_is_fack(const struct tcp_sock *tp)
972 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
975 static inline void tcp_enable_fack(struct tcp_sock *tp)
977 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
980 /* TCP early-retransmit (ER) is similar to but more conservative than
981 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
983 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
985 struct net *net = sock_net((struct sock *)tp);
987 tp->do_early_retrans = sysctl_tcp_early_retrans &&
988 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
989 net->ipv4.sysctl_tcp_reordering == 3;
992 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
994 tp->do_early_retrans = 0;
997 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
999 return tp->sacked_out + tp->lost_out;
1002 /* This determines how many packets are "in the network" to the best
1003 * of our knowledge. In many cases it is conservative, but where
1004 * detailed information is available from the receiver (via SACK
1005 * blocks etc.) we can make more aggressive calculations.
1007 * Use this for decisions involving congestion control, use just
1008 * tp->packets_out to determine if the send queue is empty or not.
1010 * Read this equation as:
1012 * "Packets sent once on transmission queue" MINUS
1013 * "Packets left network, but not honestly ACKed yet" PLUS
1014 * "Packets fast retransmitted"
1016 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1018 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1021 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1023 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1025 return tp->snd_cwnd < tp->snd_ssthresh;
1028 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1030 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1033 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1035 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1036 (1 << inet_csk(sk)->icsk_ca_state);
1039 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1040 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1043 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1045 const struct tcp_sock *tp = tcp_sk(sk);
1047 if (tcp_in_cwnd_reduction(sk))
1048 return tp->snd_ssthresh;
1050 return max(tp->snd_ssthresh,
1051 ((tp->snd_cwnd >> 1) +
1052 (tp->snd_cwnd >> 2)));
1055 /* Use define here intentionally to get WARN_ON location shown at the caller */
1056 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1058 void tcp_enter_cwr(struct sock *sk);
1059 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1061 /* The maximum number of MSS of available cwnd for which TSO defers
1062 * sending if not using sysctl_tcp_tso_win_divisor.
1064 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1069 /* Returns end sequence number of the receiver's advertised window */
1070 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1072 return tp->snd_una + tp->snd_wnd;
1075 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1076 * flexible approach. The RFC suggests cwnd should not be raised unless
1077 * it was fully used previously. And that's exactly what we do in
1078 * congestion avoidance mode. But in slow start we allow cwnd to grow
1079 * as long as the application has used half the cwnd.
1081 * cwnd is 10 (IW10), but application sends 9 frames.
1082 * We allow cwnd to reach 18 when all frames are ACKed.
1083 * This check is safe because it's as aggressive as slow start which already
1084 * risks 100% overshoot. The advantage is that we discourage application to
1085 * either send more filler packets or data to artificially blow up the cwnd
1086 * usage, and allow application-limited process to probe bw more aggressively.
1088 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1090 const struct tcp_sock *tp = tcp_sk(sk);
1092 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1093 if (tcp_in_slow_start(tp))
1094 return tp->snd_cwnd < 2 * tp->max_packets_out;
1096 return tp->is_cwnd_limited;
1099 /* Something is really bad, we could not queue an additional packet,
1100 * because qdisc is full or receiver sent a 0 window.
1101 * We do not want to add fuel to the fire, or abort too early,
1102 * so make sure the timer we arm now is at least 200ms in the future,
1103 * regardless of current icsk_rto value (as it could be ~2ms)
1105 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1107 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1110 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1111 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1112 unsigned long max_when)
1114 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1116 return (unsigned long)min_t(u64, when, max_when);
1119 static inline void tcp_check_probe_timer(struct sock *sk)
1121 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1122 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1123 tcp_probe0_base(sk), TCP_RTO_MAX);
1126 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1131 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1137 * Calculate(/check) TCP checksum
1139 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1140 __be32 daddr, __wsum base)
1142 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1145 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1147 return __skb_checksum_complete(skb);
1150 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1152 return !skb_csum_unnecessary(skb) &&
1153 __tcp_checksum_complete(skb);
1156 /* Prequeue for VJ style copy to user, combined with checksumming. */
1158 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1160 tp->ucopy.task = NULL;
1162 tp->ucopy.memory = 0;
1163 skb_queue_head_init(&tp->ucopy.prequeue);
1166 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1171 static const char *statename[]={
1172 "Unused","Established","Syn Sent","Syn Recv",
1173 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1174 "Close Wait","Last ACK","Listen","Closing"
1177 void tcp_set_state(struct sock *sk, int state);
1179 void tcp_done(struct sock *sk);
1181 int tcp_abort(struct sock *sk, int err);
1183 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1186 rx_opt->num_sacks = 0;
1189 u32 tcp_default_init_rwnd(u32 mss);
1190 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1192 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1194 struct tcp_sock *tp = tcp_sk(sk);
1197 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1199 delta = tcp_time_stamp - tp->lsndtime;
1200 if (delta > inet_csk(sk)->icsk_rto)
1201 tcp_cwnd_restart(sk, delta);
1204 /* Determine a window scaling and initial window to offer. */
1205 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1206 __u32 *window_clamp, int wscale_ok,
1207 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1209 static inline int tcp_win_from_space(int space)
1211 return sysctl_tcp_adv_win_scale<=0 ?
1212 (space>>(-sysctl_tcp_adv_win_scale)) :
1213 space - (space>>sysctl_tcp_adv_win_scale);
1216 /* Note: caller must be prepared to deal with negative returns */
1217 static inline int tcp_space(const struct sock *sk)
1219 return tcp_win_from_space(sk->sk_rcvbuf -
1220 atomic_read(&sk->sk_rmem_alloc));
1223 static inline int tcp_full_space(const struct sock *sk)
1225 return tcp_win_from_space(sk->sk_rcvbuf);
1228 extern void tcp_openreq_init_rwin(struct request_sock *req,
1229 const struct sock *sk_listener,
1230 const struct dst_entry *dst);
1232 void tcp_enter_memory_pressure(struct sock *sk);
1234 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1236 struct net *net = sock_net((struct sock *)tp);
1238 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1241 static inline int keepalive_time_when(const struct tcp_sock *tp)
1243 struct net *net = sock_net((struct sock *)tp);
1245 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1248 static inline int keepalive_probes(const struct tcp_sock *tp)
1250 struct net *net = sock_net((struct sock *)tp);
1252 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1255 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1257 const struct inet_connection_sock *icsk = &tp->inet_conn;
1259 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1260 tcp_time_stamp - tp->rcv_tstamp);
1263 static inline int tcp_fin_time(const struct sock *sk)
1265 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1266 const int rto = inet_csk(sk)->icsk_rto;
1268 if (fin_timeout < (rto << 2) - (rto >> 1))
1269 fin_timeout = (rto << 2) - (rto >> 1);
1274 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1277 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1279 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1282 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1283 * then following tcp messages have valid values. Ignore 0 value,
1284 * or else 'negative' tsval might forbid us to accept their packets.
1286 if (!rx_opt->ts_recent)
1291 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1294 if (tcp_paws_check(rx_opt, 0))
1297 /* RST segments are not recommended to carry timestamp,
1298 and, if they do, it is recommended to ignore PAWS because
1299 "their cleanup function should take precedence over timestamps."
1300 Certainly, it is mistake. It is necessary to understand the reasons
1301 of this constraint to relax it: if peer reboots, clock may go
1302 out-of-sync and half-open connections will not be reset.
1303 Actually, the problem would be not existing if all
1304 the implementations followed draft about maintaining clock
1305 via reboots. Linux-2.2 DOES NOT!
1307 However, we can relax time bounds for RST segments to MSL.
1309 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1314 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1315 int mib_idx, u32 *last_oow_ack_time);
1317 static inline void tcp_mib_init(struct net *net)
1320 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1321 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1322 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1323 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1327 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1329 tp->lost_skb_hint = NULL;
1332 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1334 tcp_clear_retrans_hints_partial(tp);
1335 tp->retransmit_skb_hint = NULL;
1338 union tcp_md5_addr {
1340 #if IS_ENABLED(CONFIG_IPV6)
1345 /* - key database */
1346 struct tcp_md5sig_key {
1347 struct hlist_node node;
1349 u8 family; /* AF_INET or AF_INET6 */
1350 union tcp_md5_addr addr;
1351 u8 key[TCP_MD5SIG_MAXKEYLEN];
1352 struct rcu_head rcu;
1356 struct tcp_md5sig_info {
1357 struct hlist_head head;
1358 struct rcu_head rcu;
1361 /* - pseudo header */
1362 struct tcp4_pseudohdr {
1370 struct tcp6_pseudohdr {
1371 struct in6_addr saddr;
1372 struct in6_addr daddr;
1374 __be32 protocol; /* including padding */
1377 union tcp_md5sum_block {
1378 struct tcp4_pseudohdr ip4;
1379 #if IS_ENABLED(CONFIG_IPV6)
1380 struct tcp6_pseudohdr ip6;
1384 /* - pool: digest algorithm, hash description and scratch buffer */
1385 struct tcp_md5sig_pool {
1386 struct ahash_request *md5_req;
1391 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1392 const struct sock *sk, const struct sk_buff *skb);
1393 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1394 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1395 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1397 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1398 const struct sock *addr_sk);
1400 #ifdef CONFIG_TCP_MD5SIG
1401 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1402 const union tcp_md5_addr *addr,
1404 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1406 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1407 const union tcp_md5_addr *addr,
1412 #define tcp_twsk_md5_key(twsk) NULL
1415 bool tcp_alloc_md5sig_pool(void);
1417 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1418 static inline void tcp_put_md5sig_pool(void)
1423 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1424 unsigned int header_len);
1425 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1426 const struct tcp_md5sig_key *key);
1428 /* From tcp_fastopen.c */
1429 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1430 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1431 unsigned long *last_syn_loss);
1432 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1433 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1435 struct tcp_fastopen_request {
1436 /* Fast Open cookie. Size 0 means a cookie request */
1437 struct tcp_fastopen_cookie cookie;
1438 struct msghdr *data; /* data in MSG_FASTOPEN */
1440 int copied; /* queued in tcp_connect() */
1442 void tcp_free_fastopen_req(struct tcp_sock *tp);
1444 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1445 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1446 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1447 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1448 struct request_sock *req,
1449 struct tcp_fastopen_cookie *foc,
1450 struct dst_entry *dst);
1451 void tcp_fastopen_init_key_once(bool publish);
1452 #define TCP_FASTOPEN_KEY_LENGTH 16
1454 /* Fastopen key context */
1455 struct tcp_fastopen_context {
1456 struct crypto_cipher *tfm;
1457 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1458 struct rcu_head rcu;
1461 /* write queue abstraction */
1462 static inline void tcp_write_queue_purge(struct sock *sk)
1464 struct sk_buff *skb;
1466 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1467 sk_wmem_free_skb(sk, skb);
1469 tcp_clear_all_retrans_hints(tcp_sk(sk));
1472 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1474 return skb_peek(&sk->sk_write_queue);
1477 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1479 return skb_peek_tail(&sk->sk_write_queue);
1482 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1483 const struct sk_buff *skb)
1485 return skb_queue_next(&sk->sk_write_queue, skb);
1488 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1489 const struct sk_buff *skb)
1491 return skb_queue_prev(&sk->sk_write_queue, skb);
1494 #define tcp_for_write_queue(skb, sk) \
1495 skb_queue_walk(&(sk)->sk_write_queue, skb)
1497 #define tcp_for_write_queue_from(skb, sk) \
1498 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1500 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1501 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1503 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1505 return sk->sk_send_head;
1508 static inline bool tcp_skb_is_last(const struct sock *sk,
1509 const struct sk_buff *skb)
1511 return skb_queue_is_last(&sk->sk_write_queue, skb);
1514 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1516 if (tcp_skb_is_last(sk, skb))
1517 sk->sk_send_head = NULL;
1519 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1522 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1524 if (sk->sk_send_head == skb_unlinked)
1525 sk->sk_send_head = NULL;
1528 static inline void tcp_init_send_head(struct sock *sk)
1530 sk->sk_send_head = NULL;
1533 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1535 __skb_queue_tail(&sk->sk_write_queue, skb);
1538 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1540 __tcp_add_write_queue_tail(sk, skb);
1542 /* Queue it, remembering where we must start sending. */
1543 if (sk->sk_send_head == NULL) {
1544 sk->sk_send_head = skb;
1546 if (tcp_sk(sk)->highest_sack == NULL)
1547 tcp_sk(sk)->highest_sack = skb;
1551 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1553 __skb_queue_head(&sk->sk_write_queue, skb);
1556 /* Insert buff after skb on the write queue of sk. */
1557 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1558 struct sk_buff *buff,
1561 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1564 /* Insert new before skb on the write queue of sk. */
1565 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1566 struct sk_buff *skb,
1569 __skb_queue_before(&sk->sk_write_queue, skb, new);
1571 if (sk->sk_send_head == skb)
1572 sk->sk_send_head = new;
1575 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1577 __skb_unlink(skb, &sk->sk_write_queue);
1580 static inline bool tcp_write_queue_empty(struct sock *sk)
1582 return skb_queue_empty(&sk->sk_write_queue);
1585 static inline void tcp_push_pending_frames(struct sock *sk)
1587 if (tcp_send_head(sk)) {
1588 struct tcp_sock *tp = tcp_sk(sk);
1590 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1594 /* Start sequence of the skb just after the highest skb with SACKed
1595 * bit, valid only if sacked_out > 0 or when the caller has ensured
1596 * validity by itself.
1598 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1600 if (!tp->sacked_out)
1603 if (tp->highest_sack == NULL)
1606 return TCP_SKB_CB(tp->highest_sack)->seq;
1609 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1611 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1612 tcp_write_queue_next(sk, skb);
1615 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1617 return tcp_sk(sk)->highest_sack;
1620 static inline void tcp_highest_sack_reset(struct sock *sk)
1622 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1625 /* Called when old skb is about to be deleted (to be combined with new skb) */
1626 static inline void tcp_highest_sack_combine(struct sock *sk,
1627 struct sk_buff *old,
1628 struct sk_buff *new)
1630 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1631 tcp_sk(sk)->highest_sack = new;
1634 /* This helper checks if socket has IP_TRANSPARENT set */
1635 static inline bool inet_sk_transparent(const struct sock *sk)
1637 switch (sk->sk_state) {
1639 return inet_twsk(sk)->tw_transparent;
1640 case TCP_NEW_SYN_RECV:
1641 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1643 return inet_sk(sk)->transparent;
1646 /* Determines whether this is a thin stream (which may suffer from
1647 * increased latency). Used to trigger latency-reducing mechanisms.
1649 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1651 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1655 enum tcp_seq_states {
1656 TCP_SEQ_STATE_LISTENING,
1657 TCP_SEQ_STATE_ESTABLISHED,
1660 int tcp_seq_open(struct inode *inode, struct file *file);
1662 struct tcp_seq_afinfo {
1665 const struct file_operations *seq_fops;
1666 struct seq_operations seq_ops;
1669 struct tcp_iter_state {
1670 struct seq_net_private p;
1672 enum tcp_seq_states state;
1673 struct sock *syn_wait_sk;
1674 int bucket, offset, sbucket, num;
1678 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1679 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1681 extern struct request_sock_ops tcp_request_sock_ops;
1682 extern struct request_sock_ops tcp6_request_sock_ops;
1684 void tcp_v4_destroy_sock(struct sock *sk);
1686 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1687 netdev_features_t features);
1688 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1689 int tcp_gro_complete(struct sk_buff *skb);
1691 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1693 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1695 struct net *net = sock_net((struct sock *)tp);
1696 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1699 static inline bool tcp_stream_memory_free(const struct sock *sk)
1701 const struct tcp_sock *tp = tcp_sk(sk);
1702 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1704 return notsent_bytes < tcp_notsent_lowat(tp);
1707 #ifdef CONFIG_PROC_FS
1708 int tcp4_proc_init(void);
1709 void tcp4_proc_exit(void);
1712 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1713 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1714 const struct tcp_request_sock_ops *af_ops,
1715 struct sock *sk, struct sk_buff *skb);
1717 /* TCP af-specific functions */
1718 struct tcp_sock_af_ops {
1719 #ifdef CONFIG_TCP_MD5SIG
1720 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1721 const struct sock *addr_sk);
1722 int (*calc_md5_hash)(char *location,
1723 const struct tcp_md5sig_key *md5,
1724 const struct sock *sk,
1725 const struct sk_buff *skb);
1726 int (*md5_parse)(struct sock *sk,
1727 char __user *optval,
1732 struct tcp_request_sock_ops {
1734 #ifdef CONFIG_TCP_MD5SIG
1735 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1736 const struct sock *addr_sk);
1737 int (*calc_md5_hash) (char *location,
1738 const struct tcp_md5sig_key *md5,
1739 const struct sock *sk,
1740 const struct sk_buff *skb);
1742 void (*init_req)(struct request_sock *req,
1743 const struct sock *sk_listener,
1744 struct sk_buff *skb);
1745 #ifdef CONFIG_SYN_COOKIES
1746 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1749 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1750 const struct request_sock *req,
1752 __u32 (*init_seq)(const struct sk_buff *skb);
1753 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1754 struct flowi *fl, struct request_sock *req,
1755 struct tcp_fastopen_cookie *foc,
1756 enum tcp_synack_type synack_type);
1759 #ifdef CONFIG_SYN_COOKIES
1760 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1761 const struct sock *sk, struct sk_buff *skb,
1764 tcp_synq_overflow(sk);
1765 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1766 return ops->cookie_init_seq(skb, mss);
1769 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1770 const struct sock *sk, struct sk_buff *skb,
1777 int tcpv4_offload_init(void);
1779 void tcp_v4_init(void);
1780 void tcp_init(void);
1782 /* tcp_recovery.c */
1784 /* Flags to enable various loss recovery features. See below */
1785 extern int sysctl_tcp_recovery;
1787 /* Use TCP RACK to detect (some) tail and retransmit losses */
1788 #define TCP_RACK_LOST_RETRANS 0x1
1790 extern int tcp_rack_mark_lost(struct sock *sk);
1792 extern void tcp_rack_advance(struct tcp_sock *tp,
1793 const struct skb_mstamp *xmit_time, u8 sacked);
1796 * Save and compile IPv4 options, return a pointer to it
1798 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1800 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1801 struct ip_options_rcu *dopt = NULL;
1804 int opt_size = sizeof(*dopt) + opt->optlen;
1806 dopt = kmalloc(opt_size, GFP_ATOMIC);
1807 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1815 /* locally generated TCP pure ACKs have skb->truesize == 2
1816 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1817 * This is much faster than dissecting the packet to find out.
1818 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1820 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1822 return skb->truesize == 2;
1825 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1830 static inline int tcp_inq(struct sock *sk)
1832 struct tcp_sock *tp = tcp_sk(sk);
1835 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1837 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1839 before(tp->urg_seq, tp->copied_seq) ||
1840 !before(tp->urg_seq, tp->rcv_nxt)) {
1842 answ = tp->rcv_nxt - tp->copied_seq;
1844 /* Subtract 1, if FIN was received */
1845 if (answ && sock_flag(sk, SOCK_DONE))
1848 answ = tp->urg_seq - tp->copied_seq;
1854 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1858 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1859 tp->segs_in += segs_in;
1860 if (skb->len > tcp_hdrlen(skb))
1861 tp->data_segs_in += segs_in;
1865 * TCP listen path runs lockless.
1866 * We forced "struct sock" to be const qualified to make sure
1867 * we don't modify one of its field by mistake.
1868 * Here, we increment sk_drops which is an atomic_t, so we can safely
1869 * make sock writable again.
1871 static inline void tcp_listendrop(const struct sock *sk)
1873 atomic_inc(&((struct sock *)sk)->sk_drops);
1874 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);