2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/crypto.h>
31 #include <linux/cryptohash.h>
32 #include <linux/kref.h>
33 #include <linux/ktime.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo;
52 extern struct percpu_counter tcp_orphan_count;
53 void tcp_time_wait(struct sock *sk, int state, int timeo);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 1024
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL 600
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD 8
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
102 * when active opening a connection.
103 * RFC1122 says the minimum retry MUST
104 * be at least 180secs. Nevertheless
105 * this value is corresponding to
106 * 63secs of retransmission with the
107 * current initial RTO.
110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
111 * when passive opening a connection.
112 * This is corresponding to 31secs of
113 * retransmission with the current
117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 * state, about 60 seconds */
119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
120 /* BSD style FIN_WAIT2 deadlock breaker.
121 * It used to be 3min, new value is 60sec,
122 * to combine FIN-WAIT-2 timeout with
126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
129 #define TCP_ATO_MIN ((unsigned)(HZ/25))
131 #define TCP_DELACK_MIN 4U
132 #define TCP_ATO_MIN 4U
134 #define TCP_RTO_MAX ((unsigned)(120*HZ))
135 #define TCP_RTO_MIN ((unsigned)(HZ/5))
136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
138 * used as a fallback RTO for the
139 * initial data transmission if no
140 * valid RTT sample has been acquired,
141 * most likely due to retrans in 3WHS.
144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
145 * for local resources.
148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
150 #define TCP_KEEPALIVE_INTVL (75*HZ)
152 #define MAX_TCP_KEEPIDLE 32767
153 #define MAX_TCP_KEEPINTVL 32767
154 #define MAX_TCP_KEEPCNT 127
155 #define MAX_TCP_SYNCNT 127
157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
161 * after this time. It should be equal
162 * (or greater than) TCP_TIMEWAIT_LEN
163 * to provide reliability equal to one
164 * provided by timewait state.
166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
167 * timestamps. It must be less than
168 * minimal timewait lifetime.
174 #define TCPOPT_NOP 1 /* Padding */
175 #define TCPOPT_EOL 0 /* End of options */
176 #define TCPOPT_MSS 2 /* Segment size negotiating */
177 #define TCPOPT_WINDOW 3 /* Window scaling */
178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
179 #define TCPOPT_SACK 5 /* SACK Block */
180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
183 #define TCPOPT_EXP 254 /* Experimental */
184 /* Magic number to be after the option value for sharing TCP
185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
187 #define TCPOPT_FASTOPEN_MAGIC 0xF989
193 #define TCPOLEN_MSS 4
194 #define TCPOLEN_WINDOW 3
195 #define TCPOLEN_SACK_PERM 2
196 #define TCPOLEN_TIMESTAMP 10
197 #define TCPOLEN_MD5SIG 18
198 #define TCPOLEN_FASTOPEN_BASE 2
199 #define TCPOLEN_EXP_FASTOPEN_BASE 4
201 /* But this is what stacks really send out. */
202 #define TCPOLEN_TSTAMP_ALIGNED 12
203 #define TCPOLEN_WSCALE_ALIGNED 4
204 #define TCPOLEN_SACKPERM_ALIGNED 4
205 #define TCPOLEN_SACK_BASE 2
206 #define TCPOLEN_SACK_BASE_ALIGNED 4
207 #define TCPOLEN_SACK_PERBLOCK 8
208 #define TCPOLEN_MD5SIG_ALIGNED 20
209 #define TCPOLEN_MSS_ALIGNED 4
211 /* Flags in tp->nonagle */
212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
213 #define TCP_NAGLE_CORK 2 /* Socket is corked */
214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
216 /* TCP thin-stream limits */
217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
219 /* TCP initial congestion window as per rfc6928 */
220 #define TCP_INIT_CWND 10
222 /* Bit Flags for sysctl_tcp_fastopen */
223 #define TFO_CLIENT_ENABLE 1
224 #define TFO_SERVER_ENABLE 2
225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
227 /* Accept SYN data w/o any cookie option */
228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
230 /* Force enable TFO on all listeners, i.e., not requiring the
231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
233 #define TFO_SERVER_WO_SOCKOPT1 0x400
234 #define TFO_SERVER_WO_SOCKOPT2 0x800
236 extern struct inet_timewait_death_row tcp_death_row;
238 /* sysctl variables for tcp */
239 extern int sysctl_tcp_timestamps;
240 extern int sysctl_tcp_window_scaling;
241 extern int sysctl_tcp_sack;
242 extern int sysctl_tcp_fin_timeout;
243 extern int sysctl_tcp_retries1;
244 extern int sysctl_tcp_retries2;
245 extern int sysctl_tcp_orphan_retries;
246 extern int sysctl_tcp_fastopen;
247 extern int sysctl_tcp_retrans_collapse;
248 extern int sysctl_tcp_stdurg;
249 extern int sysctl_tcp_rfc1337;
250 extern int sysctl_tcp_abort_on_overflow;
251 extern int sysctl_tcp_max_orphans;
252 extern int sysctl_tcp_fack;
253 extern int sysctl_tcp_reordering;
254 extern int sysctl_tcp_max_reordering;
255 extern int sysctl_tcp_dsack;
256 extern long sysctl_tcp_mem[3];
257 extern int sysctl_tcp_wmem[3];
258 extern int sysctl_tcp_rmem[3];
259 extern int sysctl_tcp_app_win;
260 extern int sysctl_tcp_adv_win_scale;
261 extern int sysctl_tcp_tw_reuse;
262 extern int sysctl_tcp_frto;
263 extern int sysctl_tcp_low_latency;
264 extern int sysctl_tcp_nometrics_save;
265 extern int sysctl_tcp_moderate_rcvbuf;
266 extern int sysctl_tcp_tso_win_divisor;
267 extern int sysctl_tcp_workaround_signed_windows;
268 extern int sysctl_tcp_slow_start_after_idle;
269 extern int sysctl_tcp_thin_linear_timeouts;
270 extern int sysctl_tcp_thin_dupack;
271 extern int sysctl_tcp_early_retrans;
272 extern int sysctl_tcp_limit_output_bytes;
273 extern int sysctl_tcp_challenge_ack_limit;
274 extern unsigned int sysctl_tcp_notsent_lowat;
275 extern int sysctl_tcp_min_tso_segs;
276 extern int sysctl_tcp_min_rtt_wlen;
277 extern int sysctl_tcp_autocorking;
278 extern int sysctl_tcp_invalid_ratelimit;
279 extern int sysctl_tcp_pacing_ss_ratio;
280 extern int sysctl_tcp_pacing_ca_ratio;
282 extern atomic_long_t tcp_memory_allocated;
283 extern struct percpu_counter tcp_sockets_allocated;
284 extern int tcp_memory_pressure;
286 /* optimized version of sk_under_memory_pressure() for TCP sockets */
287 static inline bool tcp_under_memory_pressure(const struct sock *sk)
289 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
290 mem_cgroup_under_socket_pressure(sk->sk_memcg))
293 return tcp_memory_pressure;
296 * The next routines deal with comparing 32 bit unsigned ints
297 * and worry about wraparound (automatic with unsigned arithmetic).
300 static inline bool before(__u32 seq1, __u32 seq2)
302 return (__s32)(seq1-seq2) < 0;
304 #define after(seq2, seq1) before(seq1, seq2)
306 /* is s2<=s1<=s3 ? */
307 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
309 return seq3 - seq2 >= seq1 - seq2;
312 static inline bool tcp_out_of_memory(struct sock *sk)
314 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
315 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
320 void sk_forced_mem_schedule(struct sock *sk, int size);
322 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
324 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
325 int orphans = percpu_counter_read_positive(ocp);
327 if (orphans << shift > sysctl_tcp_max_orphans) {
328 orphans = percpu_counter_sum_positive(ocp);
329 if (orphans << shift > sysctl_tcp_max_orphans)
335 bool tcp_check_oom(struct sock *sk, int shift);
338 extern struct proto tcp_prot;
340 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
341 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
342 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
343 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
344 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
346 void tcp_tasklet_init(void);
348 void tcp_v4_err(struct sk_buff *skb, u32);
350 void tcp_shutdown(struct sock *sk, int how);
352 void tcp_v4_early_demux(struct sk_buff *skb);
353 int tcp_v4_rcv(struct sk_buff *skb);
355 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
356 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
357 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
359 void tcp_release_cb(struct sock *sk);
360 void tcp_wfree(struct sk_buff *skb);
361 void tcp_write_timer_handler(struct sock *sk);
362 void tcp_delack_timer_handler(struct sock *sk);
363 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
364 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
365 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
366 const struct tcphdr *th, unsigned int len);
367 void tcp_rcv_space_adjust(struct sock *sk);
368 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
369 void tcp_twsk_destructor(struct sock *sk);
370 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
371 struct pipe_inode_info *pipe, size_t len,
374 static inline void tcp_dec_quickack_mode(struct sock *sk,
375 const unsigned int pkts)
377 struct inet_connection_sock *icsk = inet_csk(sk);
379 if (icsk->icsk_ack.quick) {
380 if (pkts >= icsk->icsk_ack.quick) {
381 icsk->icsk_ack.quick = 0;
382 /* Leaving quickack mode we deflate ATO. */
383 icsk->icsk_ack.ato = TCP_ATO_MIN;
385 icsk->icsk_ack.quick -= pkts;
390 #define TCP_ECN_QUEUE_CWR 2
391 #define TCP_ECN_DEMAND_CWR 4
392 #define TCP_ECN_SEEN 8
402 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
404 const struct tcphdr *th);
405 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
406 struct request_sock *req, bool fastopen);
407 int tcp_child_process(struct sock *parent, struct sock *child,
408 struct sk_buff *skb);
409 void tcp_enter_loss(struct sock *sk);
410 void tcp_clear_retrans(struct tcp_sock *tp);
411 void tcp_update_metrics(struct sock *sk);
412 void tcp_init_metrics(struct sock *sk);
413 void tcp_metrics_init(void);
414 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
415 bool paws_check, bool timestamps);
416 bool tcp_remember_stamp(struct sock *sk);
417 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
418 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
419 void tcp_disable_fack(struct tcp_sock *tp);
420 void tcp_close(struct sock *sk, long timeout);
421 void tcp_init_sock(struct sock *sk);
422 unsigned int tcp_poll(struct file *file, struct socket *sock,
423 struct poll_table_struct *wait);
424 int tcp_getsockopt(struct sock *sk, int level, int optname,
425 char __user *optval, int __user *optlen);
426 int tcp_setsockopt(struct sock *sk, int level, int optname,
427 char __user *optval, unsigned int optlen);
428 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
429 char __user *optval, int __user *optlen);
430 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
431 char __user *optval, unsigned int optlen);
432 void tcp_set_keepalive(struct sock *sk, int val);
433 void tcp_syn_ack_timeout(const struct request_sock *req);
434 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
435 int flags, int *addr_len);
436 void tcp_parse_options(const struct sk_buff *skb,
437 struct tcp_options_received *opt_rx,
438 int estab, struct tcp_fastopen_cookie *foc);
439 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
442 * TCP v4 functions exported for the inet6 API
445 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
446 void tcp_v4_mtu_reduced(struct sock *sk);
447 void tcp_req_err(struct sock *sk, u32 seq);
448 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
449 struct sock *tcp_create_openreq_child(const struct sock *sk,
450 struct request_sock *req,
451 struct sk_buff *skb);
452 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
453 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
454 struct request_sock *req,
455 struct dst_entry *dst,
456 struct request_sock *req_unhash,
458 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
459 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
460 int tcp_connect(struct sock *sk);
461 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
462 struct request_sock *req,
463 struct tcp_fastopen_cookie *foc,
465 int tcp_disconnect(struct sock *sk, int flags);
467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
471 /* From syncookies.c */
472 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
473 struct request_sock *req,
474 struct dst_entry *dst);
475 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
477 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
478 #ifdef CONFIG_SYN_COOKIES
480 /* Syncookies use a monotonic timer which increments every 60 seconds.
481 * This counter is used both as a hash input and partially encoded into
482 * the cookie value. A cookie is only validated further if the delta
483 * between the current counter value and the encoded one is less than this,
484 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
485 * the counter advances immediately after a cookie is generated).
487 #define MAX_SYNCOOKIE_AGE 2
488 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
489 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
491 /* syncookies: remember time of last synqueue overflow
492 * But do not dirty this field too often (once per second is enough)
493 * It is racy as we do not hold a lock, but race is very minor.
495 static inline void tcp_synq_overflow(const struct sock *sk)
497 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
498 unsigned long now = jiffies;
500 if (time_after(now, last_overflow + HZ))
501 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
504 /* syncookies: no recent synqueue overflow on this listening socket? */
505 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
507 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
509 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
512 static inline u32 tcp_cookie_time(void)
514 u64 val = get_jiffies_64();
516 do_div(val, TCP_SYNCOOKIE_PERIOD);
520 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
522 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
523 __u32 cookie_init_timestamp(struct request_sock *req);
524 bool cookie_timestamp_decode(struct tcp_options_received *opt);
525 bool cookie_ecn_ok(const struct tcp_options_received *opt,
526 const struct net *net, const struct dst_entry *dst);
528 /* From net/ipv6/syncookies.c */
529 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
531 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
533 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
534 const struct tcphdr *th, u16 *mssp);
535 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
539 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
541 bool tcp_may_send_now(struct sock *sk);
542 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
543 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
544 void tcp_retransmit_timer(struct sock *sk);
545 void tcp_xmit_retransmit_queue(struct sock *);
546 void tcp_simple_retransmit(struct sock *);
547 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
548 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
550 void tcp_send_probe0(struct sock *);
551 void tcp_send_partial(struct sock *);
552 int tcp_write_wakeup(struct sock *, int mib);
553 void tcp_send_fin(struct sock *sk);
554 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
555 int tcp_send_synack(struct sock *);
556 void tcp_push_one(struct sock *, unsigned int mss_now);
557 void tcp_send_ack(struct sock *sk);
558 void tcp_send_delayed_ack(struct sock *sk);
559 void tcp_send_loss_probe(struct sock *sk);
560 bool tcp_schedule_loss_probe(struct sock *sk);
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 >= 512)
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 */
764 __u32 ack_seq; /* Sequence number ACK'd */
766 struct inet_skb_parm h4;
767 #if IS_ENABLED(CONFIG_IPV6)
768 struct inet6_skb_parm h6;
770 } header; /* For incoming frames */
773 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
776 #if IS_ENABLED(CONFIG_IPV6)
777 /* This is the variant of inet6_iif() that must be used by TCP,
778 * as TCP moves IP6CB into a different location in skb->cb[]
780 static inline int tcp_v6_iif(const struct sk_buff *skb)
782 return TCP_SKB_CB(skb)->header.h6.iif;
786 /* Due to TSO, an SKB can be composed of multiple actual
787 * packets. To keep these tracked properly, we use this.
789 static inline int tcp_skb_pcount(const struct sk_buff *skb)
791 return TCP_SKB_CB(skb)->tcp_gso_segs;
794 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
796 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
799 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
801 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
804 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
805 static inline int tcp_skb_mss(const struct sk_buff *skb)
807 return TCP_SKB_CB(skb)->tcp_gso_size;
810 /* Events passed to congestion control interface */
812 CA_EVENT_TX_START, /* first transmit when no packets in flight */
813 CA_EVENT_CWND_RESTART, /* congestion window restart */
814 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
815 CA_EVENT_LOSS, /* loss timeout */
816 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
817 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
818 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
819 CA_EVENT_NON_DELAYED_ACK,
822 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
823 enum tcp_ca_ack_event_flags {
824 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
825 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
826 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
830 * Interface for adding new TCP congestion control handlers
832 #define TCP_CA_NAME_MAX 16
833 #define TCP_CA_MAX 128
834 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
836 #define TCP_CA_UNSPEC 0
838 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
839 #define TCP_CONG_NON_RESTRICTED 0x1
840 /* Requires ECN/ECT set on all packets */
841 #define TCP_CONG_NEEDS_ECN 0x2
845 struct tcp_congestion_ops {
846 struct list_head list;
850 /* initialize private data (optional) */
851 void (*init)(struct sock *sk);
852 /* cleanup private data (optional) */
853 void (*release)(struct sock *sk);
855 /* return slow start threshold (required) */
856 u32 (*ssthresh)(struct sock *sk);
857 /* do new cwnd calculation (required) */
858 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
859 /* call before changing ca_state (optional) */
860 void (*set_state)(struct sock *sk, u8 new_state);
861 /* call when cwnd event occurs (optional) */
862 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
863 /* call when ack arrives (optional) */
864 void (*in_ack_event)(struct sock *sk, u32 flags);
865 /* new value of cwnd after loss (optional) */
866 u32 (*undo_cwnd)(struct sock *sk);
867 /* hook for packet ack accounting (optional) */
868 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
869 /* get info for inet_diag (optional) */
870 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
871 union tcp_cc_info *info);
873 char name[TCP_CA_NAME_MAX];
874 struct module *owner;
877 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
878 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
880 void tcp_assign_congestion_control(struct sock *sk);
881 void tcp_init_congestion_control(struct sock *sk);
882 void tcp_cleanup_congestion_control(struct sock *sk);
883 int tcp_set_default_congestion_control(const char *name);
884 void tcp_get_default_congestion_control(char *name);
885 void tcp_get_available_congestion_control(char *buf, size_t len);
886 void tcp_get_allowed_congestion_control(char *buf, size_t len);
887 int tcp_set_allowed_congestion_control(char *allowed);
888 int tcp_set_congestion_control(struct sock *sk, const char *name);
889 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
890 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
892 u32 tcp_reno_ssthresh(struct sock *sk);
893 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
894 extern struct tcp_congestion_ops tcp_reno;
896 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
897 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
899 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
901 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
907 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
909 const struct inet_connection_sock *icsk = inet_csk(sk);
911 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
914 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
916 struct inet_connection_sock *icsk = inet_csk(sk);
918 if (icsk->icsk_ca_ops->set_state)
919 icsk->icsk_ca_ops->set_state(sk, ca_state);
920 icsk->icsk_ca_state = ca_state;
923 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
925 const struct inet_connection_sock *icsk = inet_csk(sk);
927 if (icsk->icsk_ca_ops->cwnd_event)
928 icsk->icsk_ca_ops->cwnd_event(sk, event);
931 /* These functions determine how the current flow behaves in respect of SACK
932 * handling. SACK is negotiated with the peer, and therefore it can vary
933 * between different flows.
935 * tcp_is_sack - SACK enabled
936 * tcp_is_reno - No SACK
937 * tcp_is_fack - FACK enabled, implies SACK enabled
939 static inline int tcp_is_sack(const struct tcp_sock *tp)
941 return tp->rx_opt.sack_ok;
944 static inline bool tcp_is_reno(const struct tcp_sock *tp)
946 return !tcp_is_sack(tp);
949 static inline bool tcp_is_fack(const struct tcp_sock *tp)
951 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
954 static inline void tcp_enable_fack(struct tcp_sock *tp)
956 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
959 /* TCP early-retransmit (ER) is similar to but more conservative than
960 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
962 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
964 tp->do_early_retrans = sysctl_tcp_early_retrans &&
965 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
966 sysctl_tcp_reordering == 3;
969 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
971 tp->do_early_retrans = 0;
974 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
976 return tp->sacked_out + tp->lost_out;
979 /* This determines how many packets are "in the network" to the best
980 * of our knowledge. In many cases it is conservative, but where
981 * detailed information is available from the receiver (via SACK
982 * blocks etc.) we can make more aggressive calculations.
984 * Use this for decisions involving congestion control, use just
985 * tp->packets_out to determine if the send queue is empty or not.
987 * Read this equation as:
989 * "Packets sent once on transmission queue" MINUS
990 * "Packets left network, but not honestly ACKed yet" PLUS
991 * "Packets fast retransmitted"
993 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
995 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
998 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1000 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1002 return tp->snd_cwnd < tp->snd_ssthresh;
1005 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1007 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1010 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1012 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1013 (1 << inet_csk(sk)->icsk_ca_state);
1016 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1017 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1020 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1022 const struct tcp_sock *tp = tcp_sk(sk);
1024 if (tcp_in_cwnd_reduction(sk))
1025 return tp->snd_ssthresh;
1027 return max(tp->snd_ssthresh,
1028 ((tp->snd_cwnd >> 1) +
1029 (tp->snd_cwnd >> 2)));
1032 /* Use define here intentionally to get WARN_ON location shown at the caller */
1033 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1035 void tcp_enter_cwr(struct sock *sk);
1036 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1038 /* The maximum number of MSS of available cwnd for which TSO defers
1039 * sending if not using sysctl_tcp_tso_win_divisor.
1041 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1046 /* Slow start with delack produces 3 packets of burst, so that
1047 * it is safe "de facto". This will be the default - same as
1048 * the default reordering threshold - but if reordering increases,
1049 * we must be able to allow cwnd to burst at least this much in order
1050 * to not pull it back when holes are filled.
1052 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1054 return tp->reordering;
1057 /* Returns end sequence number of the receiver's advertised window */
1058 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1060 return tp->snd_una + tp->snd_wnd;
1063 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1064 * flexible approach. The RFC suggests cwnd should not be raised unless
1065 * it was fully used previously. And that's exactly what we do in
1066 * congestion avoidance mode. But in slow start we allow cwnd to grow
1067 * as long as the application has used half the cwnd.
1069 * cwnd is 10 (IW10), but application sends 9 frames.
1070 * We allow cwnd to reach 18 when all frames are ACKed.
1071 * This check is safe because it's as aggressive as slow start which already
1072 * risks 100% overshoot. The advantage is that we discourage application to
1073 * either send more filler packets or data to artificially blow up the cwnd
1074 * usage, and allow application-limited process to probe bw more aggressively.
1076 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1078 const struct tcp_sock *tp = tcp_sk(sk);
1080 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1081 if (tcp_in_slow_start(tp))
1082 return tp->snd_cwnd < 2 * tp->max_packets_out;
1084 return tp->is_cwnd_limited;
1087 /* Something is really bad, we could not queue an additional packet,
1088 * because qdisc is full or receiver sent a 0 window.
1089 * We do not want to add fuel to the fire, or abort too early,
1090 * so make sure the timer we arm now is at least 200ms in the future,
1091 * regardless of current icsk_rto value (as it could be ~2ms)
1093 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1095 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1098 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1099 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1100 unsigned long max_when)
1102 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1104 return (unsigned long)min_t(u64, when, max_when);
1107 static inline void tcp_check_probe_timer(struct sock *sk)
1109 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1110 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1111 tcp_probe0_base(sk), TCP_RTO_MAX);
1114 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1119 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1125 * Calculate(/check) TCP checksum
1127 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1128 __be32 daddr, __wsum base)
1130 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1133 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1135 return __skb_checksum_complete(skb);
1138 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1140 return !skb_csum_unnecessary(skb) &&
1141 __tcp_checksum_complete(skb);
1144 /* Prequeue for VJ style copy to user, combined with checksumming. */
1146 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1148 tp->ucopy.task = NULL;
1150 tp->ucopy.memory = 0;
1151 skb_queue_head_init(&tp->ucopy.prequeue);
1154 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1159 static const char *statename[]={
1160 "Unused","Established","Syn Sent","Syn Recv",
1161 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1162 "Close Wait","Last ACK","Listen","Closing"
1165 void tcp_set_state(struct sock *sk, int state);
1167 void tcp_done(struct sock *sk);
1169 int tcp_abort(struct sock *sk, int err);
1171 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1174 rx_opt->num_sacks = 0;
1177 u32 tcp_default_init_rwnd(u32 mss);
1178 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1180 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1182 struct tcp_sock *tp = tcp_sk(sk);
1185 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1187 delta = tcp_time_stamp - tp->lsndtime;
1188 if (delta > inet_csk(sk)->icsk_rto)
1189 tcp_cwnd_restart(sk, delta);
1192 /* Determine a window scaling and initial window to offer. */
1193 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1194 __u32 *window_clamp, int wscale_ok,
1195 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1197 static inline int tcp_win_from_space(int space)
1199 return sysctl_tcp_adv_win_scale<=0 ?
1200 (space>>(-sysctl_tcp_adv_win_scale)) :
1201 space - (space>>sysctl_tcp_adv_win_scale);
1204 /* Note: caller must be prepared to deal with negative returns */
1205 static inline int tcp_space(const struct sock *sk)
1207 return tcp_win_from_space(sk->sk_rcvbuf -
1208 atomic_read(&sk->sk_rmem_alloc));
1211 static inline int tcp_full_space(const struct sock *sk)
1213 return tcp_win_from_space(sk->sk_rcvbuf);
1216 extern void tcp_openreq_init_rwin(struct request_sock *req,
1217 const struct sock *sk_listener,
1218 const struct dst_entry *dst);
1220 void tcp_enter_memory_pressure(struct sock *sk);
1222 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1224 struct net *net = sock_net((struct sock *)tp);
1226 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1229 static inline int keepalive_time_when(const struct tcp_sock *tp)
1231 struct net *net = sock_net((struct sock *)tp);
1233 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1236 static inline int keepalive_probes(const struct tcp_sock *tp)
1238 struct net *net = sock_net((struct sock *)tp);
1240 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1243 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1245 const struct inet_connection_sock *icsk = &tp->inet_conn;
1247 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1248 tcp_time_stamp - tp->rcv_tstamp);
1251 static inline int tcp_fin_time(const struct sock *sk)
1253 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1254 const int rto = inet_csk(sk)->icsk_rto;
1256 if (fin_timeout < (rto << 2) - (rto >> 1))
1257 fin_timeout = (rto << 2) - (rto >> 1);
1262 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1265 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1267 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1270 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1271 * then following tcp messages have valid values. Ignore 0 value,
1272 * or else 'negative' tsval might forbid us to accept their packets.
1274 if (!rx_opt->ts_recent)
1279 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1282 if (tcp_paws_check(rx_opt, 0))
1285 /* RST segments are not recommended to carry timestamp,
1286 and, if they do, it is recommended to ignore PAWS because
1287 "their cleanup function should take precedence over timestamps."
1288 Certainly, it is mistake. It is necessary to understand the reasons
1289 of this constraint to relax it: if peer reboots, clock may go
1290 out-of-sync and half-open connections will not be reset.
1291 Actually, the problem would be not existing if all
1292 the implementations followed draft about maintaining clock
1293 via reboots. Linux-2.2 DOES NOT!
1295 However, we can relax time bounds for RST segments to MSL.
1297 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1302 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1303 int mib_idx, u32 *last_oow_ack_time);
1305 static inline void tcp_mib_init(struct net *net)
1308 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1309 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1310 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1311 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1315 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1317 tp->lost_skb_hint = NULL;
1320 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1322 tcp_clear_retrans_hints_partial(tp);
1323 tp->retransmit_skb_hint = NULL;
1329 union tcp_md5_addr {
1331 #if IS_ENABLED(CONFIG_IPV6)
1336 /* - key database */
1337 struct tcp_md5sig_key {
1338 struct hlist_node node;
1340 u8 family; /* AF_INET or AF_INET6 */
1341 union tcp_md5_addr addr;
1342 u8 key[TCP_MD5SIG_MAXKEYLEN];
1343 struct rcu_head rcu;
1347 struct tcp_md5sig_info {
1348 struct hlist_head head;
1349 struct rcu_head rcu;
1352 /* - pseudo header */
1353 struct tcp4_pseudohdr {
1361 struct tcp6_pseudohdr {
1362 struct in6_addr saddr;
1363 struct in6_addr daddr;
1365 __be32 protocol; /* including padding */
1368 union tcp_md5sum_block {
1369 struct tcp4_pseudohdr ip4;
1370 #if IS_ENABLED(CONFIG_IPV6)
1371 struct tcp6_pseudohdr ip6;
1375 /* - pool: digest algorithm, hash description and scratch buffer */
1376 struct tcp_md5sig_pool {
1377 struct hash_desc md5_desc;
1378 union tcp_md5sum_block md5_blk;
1382 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1383 const struct sock *sk, const struct sk_buff *skb);
1384 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1385 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1386 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1388 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1389 const struct sock *addr_sk);
1391 #ifdef CONFIG_TCP_MD5SIG
1392 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1393 const union tcp_md5_addr *addr,
1395 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1397 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1398 const union tcp_md5_addr *addr,
1403 #define tcp_twsk_md5_key(twsk) NULL
1406 bool tcp_alloc_md5sig_pool(void);
1408 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1409 static inline void tcp_put_md5sig_pool(void)
1414 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1415 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1416 unsigned int header_len);
1417 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1418 const struct tcp_md5sig_key *key);
1420 /* From tcp_fastopen.c */
1421 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1422 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1423 unsigned long *last_syn_loss);
1424 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1425 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1427 struct tcp_fastopen_request {
1428 /* Fast Open cookie. Size 0 means a cookie request */
1429 struct tcp_fastopen_cookie cookie;
1430 struct msghdr *data; /* data in MSG_FASTOPEN */
1432 int copied; /* queued in tcp_connect() */
1434 void tcp_free_fastopen_req(struct tcp_sock *tp);
1436 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1437 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1438 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1439 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1440 struct request_sock *req,
1441 struct tcp_fastopen_cookie *foc,
1442 struct dst_entry *dst);
1443 void tcp_fastopen_init_key_once(bool publish);
1444 #define TCP_FASTOPEN_KEY_LENGTH 16
1446 /* Fastopen key context */
1447 struct tcp_fastopen_context {
1448 struct crypto_cipher *tfm;
1449 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1450 struct rcu_head rcu;
1453 /* write queue abstraction */
1454 static inline void tcp_write_queue_purge(struct sock *sk)
1456 struct sk_buff *skb;
1458 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1459 sk_wmem_free_skb(sk, skb);
1461 tcp_clear_all_retrans_hints(tcp_sk(sk));
1464 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1466 return skb_peek(&sk->sk_write_queue);
1469 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1471 return skb_peek_tail(&sk->sk_write_queue);
1474 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1475 const struct sk_buff *skb)
1477 return skb_queue_next(&sk->sk_write_queue, skb);
1480 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1481 const struct sk_buff *skb)
1483 return skb_queue_prev(&sk->sk_write_queue, skb);
1486 #define tcp_for_write_queue(skb, sk) \
1487 skb_queue_walk(&(sk)->sk_write_queue, skb)
1489 #define tcp_for_write_queue_from(skb, sk) \
1490 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1492 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1493 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1495 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1497 return sk->sk_send_head;
1500 static inline bool tcp_skb_is_last(const struct sock *sk,
1501 const struct sk_buff *skb)
1503 return skb_queue_is_last(&sk->sk_write_queue, skb);
1506 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1508 if (tcp_skb_is_last(sk, skb))
1509 sk->sk_send_head = NULL;
1511 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1514 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1516 if (sk->sk_send_head == skb_unlinked)
1517 sk->sk_send_head = NULL;
1520 static inline void tcp_init_send_head(struct sock *sk)
1522 sk->sk_send_head = NULL;
1525 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1527 __skb_queue_tail(&sk->sk_write_queue, skb);
1530 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1532 __tcp_add_write_queue_tail(sk, skb);
1534 /* Queue it, remembering where we must start sending. */
1535 if (sk->sk_send_head == NULL) {
1536 sk->sk_send_head = skb;
1538 if (tcp_sk(sk)->highest_sack == NULL)
1539 tcp_sk(sk)->highest_sack = skb;
1543 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1545 __skb_queue_head(&sk->sk_write_queue, skb);
1548 /* Insert buff after skb on the write queue of sk. */
1549 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1550 struct sk_buff *buff,
1553 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1556 /* Insert new before skb on the write queue of sk. */
1557 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1558 struct sk_buff *skb,
1561 __skb_queue_before(&sk->sk_write_queue, skb, new);
1563 if (sk->sk_send_head == skb)
1564 sk->sk_send_head = new;
1567 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1569 __skb_unlink(skb, &sk->sk_write_queue);
1572 static inline bool tcp_write_queue_empty(struct sock *sk)
1574 return skb_queue_empty(&sk->sk_write_queue);
1577 static inline void tcp_push_pending_frames(struct sock *sk)
1579 if (tcp_send_head(sk)) {
1580 struct tcp_sock *tp = tcp_sk(sk);
1582 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1586 /* Start sequence of the skb just after the highest skb with SACKed
1587 * bit, valid only if sacked_out > 0 or when the caller has ensured
1588 * validity by itself.
1590 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1592 if (!tp->sacked_out)
1595 if (tp->highest_sack == NULL)
1598 return TCP_SKB_CB(tp->highest_sack)->seq;
1601 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1603 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1604 tcp_write_queue_next(sk, skb);
1607 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1609 return tcp_sk(sk)->highest_sack;
1612 static inline void tcp_highest_sack_reset(struct sock *sk)
1614 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1617 /* Called when old skb is about to be deleted (to be combined with new skb) */
1618 static inline void tcp_highest_sack_combine(struct sock *sk,
1619 struct sk_buff *old,
1620 struct sk_buff *new)
1622 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1623 tcp_sk(sk)->highest_sack = new;
1626 /* This helper checks if socket has IP_TRANSPARENT set */
1627 static inline bool inet_sk_transparent(const struct sock *sk)
1629 switch (sk->sk_state) {
1631 return inet_twsk(sk)->tw_transparent;
1632 case TCP_NEW_SYN_RECV:
1633 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1635 return inet_sk(sk)->transparent;
1638 /* Determines whether this is a thin stream (which may suffer from
1639 * increased latency). Used to trigger latency-reducing mechanisms.
1641 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1643 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1647 enum tcp_seq_states {
1648 TCP_SEQ_STATE_LISTENING,
1649 TCP_SEQ_STATE_ESTABLISHED,
1652 int tcp_seq_open(struct inode *inode, struct file *file);
1654 struct tcp_seq_afinfo {
1657 const struct file_operations *seq_fops;
1658 struct seq_operations seq_ops;
1661 struct tcp_iter_state {
1662 struct seq_net_private p;
1664 enum tcp_seq_states state;
1665 struct sock *syn_wait_sk;
1666 int bucket, offset, sbucket, num;
1670 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1671 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1673 extern struct request_sock_ops tcp_request_sock_ops;
1674 extern struct request_sock_ops tcp6_request_sock_ops;
1676 void tcp_v4_destroy_sock(struct sock *sk);
1678 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1679 netdev_features_t features);
1680 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1681 int tcp_gro_complete(struct sk_buff *skb);
1683 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1685 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1687 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1690 static inline bool tcp_stream_memory_free(const struct sock *sk)
1692 const struct tcp_sock *tp = tcp_sk(sk);
1693 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1695 return notsent_bytes < tcp_notsent_lowat(tp);
1698 #ifdef CONFIG_PROC_FS
1699 int tcp4_proc_init(void);
1700 void tcp4_proc_exit(void);
1703 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1704 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1705 const struct tcp_request_sock_ops *af_ops,
1706 struct sock *sk, struct sk_buff *skb);
1708 /* TCP af-specific functions */
1709 struct tcp_sock_af_ops {
1710 #ifdef CONFIG_TCP_MD5SIG
1711 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1712 const struct sock *addr_sk);
1713 int (*calc_md5_hash)(char *location,
1714 const struct tcp_md5sig_key *md5,
1715 const struct sock *sk,
1716 const struct sk_buff *skb);
1717 int (*md5_parse)(struct sock *sk,
1718 char __user *optval,
1723 struct tcp_request_sock_ops {
1725 #ifdef CONFIG_TCP_MD5SIG
1726 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1727 const struct sock *addr_sk);
1728 int (*calc_md5_hash) (char *location,
1729 const struct tcp_md5sig_key *md5,
1730 const struct sock *sk,
1731 const struct sk_buff *skb);
1733 void (*init_req)(struct request_sock *req,
1734 const struct sock *sk_listener,
1735 struct sk_buff *skb);
1736 #ifdef CONFIG_SYN_COOKIES
1737 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1740 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1741 const struct request_sock *req,
1743 __u32 (*init_seq)(const struct sk_buff *skb);
1744 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1745 struct flowi *fl, struct request_sock *req,
1746 struct tcp_fastopen_cookie *foc,
1750 #ifdef CONFIG_SYN_COOKIES
1751 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1752 const struct sock *sk, struct sk_buff *skb,
1755 tcp_synq_overflow(sk);
1756 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1757 return ops->cookie_init_seq(skb, mss);
1760 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1761 const struct sock *sk, struct sk_buff *skb,
1768 int tcpv4_offload_init(void);
1770 void tcp_v4_init(void);
1771 void tcp_init(void);
1773 /* tcp_recovery.c */
1775 /* Flags to enable various loss recovery features. See below */
1776 extern int sysctl_tcp_recovery;
1778 /* Use TCP RACK to detect (some) tail and retransmit losses */
1779 #define TCP_RACK_LOST_RETRANS 0x1
1781 extern int tcp_rack_mark_lost(struct sock *sk);
1783 extern void tcp_rack_advance(struct tcp_sock *tp,
1784 const struct skb_mstamp *xmit_time, u8 sacked);
1787 * Save and compile IPv4 options, return a pointer to it
1789 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1791 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1792 struct ip_options_rcu *dopt = NULL;
1795 int opt_size = sizeof(*dopt) + opt->optlen;
1797 dopt = kmalloc(opt_size, GFP_ATOMIC);
1798 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1806 /* locally generated TCP pure ACKs have skb->truesize == 2
1807 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1808 * This is much faster than dissecting the packet to find out.
1809 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1811 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1813 return skb->truesize == 2;
1816 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)