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 512
70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
71 #define TCP_FASTRETRANS_THRESH 3
73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
74 #define TCP_MAX_QUICKACKS 16U
77 #define TCP_URG_VALID 0x0100
78 #define TCP_URG_NOTYET 0x0200
79 #define TCP_URG_READ 0x0400
81 #define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
88 #define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
95 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
96 * when active opening a connection.
97 * RFC1122 says the minimum retry MUST
98 * be at least 180secs. Nevertheless
99 * this value is corresponding to
100 * 63secs of retransmission with the
101 * current initial RTO.
104 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
105 * when passive opening a connection.
106 * This is corresponding to 31secs of
107 * retransmission with the current
111 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
112 * state, about 60 seconds */
113 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
114 /* BSD style FIN_WAIT2 deadlock breaker.
115 * It used to be 3min, new value is 60sec,
116 * to combine FIN-WAIT-2 timeout with
120 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
122 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
123 #define TCP_ATO_MIN ((unsigned)(HZ/25))
125 #define TCP_DELACK_MIN 4U
126 #define TCP_ATO_MIN 4U
128 #define TCP_RTO_MAX ((unsigned)(120*HZ))
129 #define TCP_RTO_MIN ((unsigned)(HZ/5))
130 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
131 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
132 * used as a fallback RTO for the
133 * initial data transmission if no
134 * valid RTT sample has been acquired,
135 * most likely due to retrans in 3WHS.
138 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
139 * for local resources.
142 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
143 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
144 #define TCP_KEEPALIVE_INTVL (75*HZ)
146 #define MAX_TCP_KEEPIDLE 32767
147 #define MAX_TCP_KEEPINTVL 32767
148 #define MAX_TCP_KEEPCNT 127
149 #define MAX_TCP_SYNCNT 127
151 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
153 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
154 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
155 * after this time. It should be equal
156 * (or greater than) TCP_TIMEWAIT_LEN
157 * to provide reliability equal to one
158 * provided by timewait state.
160 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
161 * timestamps. It must be less than
162 * minimal timewait lifetime.
168 #define TCPOPT_NOP 1 /* Padding */
169 #define TCPOPT_EOL 0 /* End of options */
170 #define TCPOPT_MSS 2 /* Segment size negotiating */
171 #define TCPOPT_WINDOW 3 /* Window scaling */
172 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
173 #define TCPOPT_SACK 5 /* SACK Block */
174 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
175 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
176 #define TCPOPT_EXP 254 /* Experimental */
177 /* Magic number to be after the option value for sharing TCP
178 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
180 #define TCPOPT_FASTOPEN_MAGIC 0xF989
186 #define TCPOLEN_MSS 4
187 #define TCPOLEN_WINDOW 3
188 #define TCPOLEN_SACK_PERM 2
189 #define TCPOLEN_TIMESTAMP 10
190 #define TCPOLEN_MD5SIG 18
191 #define TCPOLEN_EXP_FASTOPEN_BASE 4
193 /* But this is what stacks really send out. */
194 #define TCPOLEN_TSTAMP_ALIGNED 12
195 #define TCPOLEN_WSCALE_ALIGNED 4
196 #define TCPOLEN_SACKPERM_ALIGNED 4
197 #define TCPOLEN_SACK_BASE 2
198 #define TCPOLEN_SACK_BASE_ALIGNED 4
199 #define TCPOLEN_SACK_PERBLOCK 8
200 #define TCPOLEN_MD5SIG_ALIGNED 20
201 #define TCPOLEN_MSS_ALIGNED 4
203 /* Flags in tp->nonagle */
204 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
205 #define TCP_NAGLE_CORK 2 /* Socket is corked */
206 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
208 /* TCP thin-stream limits */
209 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
211 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
212 #define TCP_INIT_CWND 10
214 /* Bit Flags for sysctl_tcp_fastopen */
215 #define TFO_CLIENT_ENABLE 1
216 #define TFO_SERVER_ENABLE 2
217 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
219 /* Accept SYN data w/o any cookie option */
220 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
222 /* Force enable TFO on all listeners, i.e., not requiring the
223 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
225 #define TFO_SERVER_WO_SOCKOPT1 0x400
226 #define TFO_SERVER_WO_SOCKOPT2 0x800
228 extern struct inet_timewait_death_row tcp_death_row;
230 /* sysctl variables for tcp */
231 extern int sysctl_tcp_timestamps;
232 extern int sysctl_tcp_window_scaling;
233 extern int sysctl_tcp_sack;
234 extern int sysctl_tcp_fin_timeout;
235 extern int sysctl_tcp_keepalive_time;
236 extern int sysctl_tcp_keepalive_probes;
237 extern int sysctl_tcp_keepalive_intvl;
238 extern int sysctl_tcp_syn_retries;
239 extern int sysctl_tcp_synack_retries;
240 extern int sysctl_tcp_retries1;
241 extern int sysctl_tcp_retries2;
242 extern int sysctl_tcp_orphan_retries;
243 extern int sysctl_tcp_syncookies;
244 extern int sysctl_tcp_fastopen;
245 extern int sysctl_tcp_retrans_collapse;
246 extern int sysctl_tcp_stdurg;
247 extern int sysctl_tcp_rfc1337;
248 extern int sysctl_tcp_abort_on_overflow;
249 extern int sysctl_tcp_max_orphans;
250 extern int sysctl_tcp_fack;
251 extern int sysctl_tcp_reordering;
252 extern int sysctl_tcp_max_reordering;
253 extern int sysctl_tcp_dsack;
254 extern long sysctl_tcp_mem[3];
255 extern int sysctl_tcp_wmem[3];
256 extern int sysctl_tcp_rmem[3];
257 extern int sysctl_tcp_app_win;
258 extern int sysctl_tcp_adv_win_scale;
259 extern int sysctl_tcp_tw_reuse;
260 extern int sysctl_tcp_frto;
261 extern int sysctl_tcp_low_latency;
262 extern int sysctl_tcp_nometrics_save;
263 extern int sysctl_tcp_moderate_rcvbuf;
264 extern int sysctl_tcp_tso_win_divisor;
265 extern int sysctl_tcp_mtu_probing;
266 extern int sysctl_tcp_base_mss;
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_autocorking;
278 extern atomic_long_t tcp_memory_allocated;
279 extern struct percpu_counter tcp_sockets_allocated;
280 extern int tcp_memory_pressure;
283 * The next routines deal with comparing 32 bit unsigned ints
284 * and worry about wraparound (automatic with unsigned arithmetic).
287 static inline bool before(__u32 seq1, __u32 seq2)
289 return (__s32)(seq1-seq2) < 0;
291 #define after(seq2, seq1) before(seq1, seq2)
293 /* is s2<=s1<=s3 ? */
294 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
296 return seq3 - seq2 >= seq1 - seq2;
299 static inline bool tcp_out_of_memory(struct sock *sk)
301 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
302 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
307 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
309 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
310 int orphans = percpu_counter_read_positive(ocp);
312 if (orphans << shift > sysctl_tcp_max_orphans) {
313 orphans = percpu_counter_sum_positive(ocp);
314 if (orphans << shift > sysctl_tcp_max_orphans)
320 bool tcp_check_oom(struct sock *sk, int shift);
322 /* syncookies: remember time of last synqueue overflow */
323 static inline void tcp_synq_overflow(struct sock *sk)
325 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
328 /* syncookies: no recent synqueue overflow on this listening socket? */
329 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
331 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
332 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
335 extern struct proto tcp_prot;
337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
338 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
340 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
341 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
343 void tcp_tasklet_init(void);
345 void tcp_v4_err(struct sk_buff *skb, u32);
347 void tcp_shutdown(struct sock *sk, int how);
349 void tcp_v4_early_demux(struct sk_buff *skb);
350 int tcp_v4_rcv(struct sk_buff *skb);
352 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
353 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
355 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
357 void tcp_release_cb(struct sock *sk);
358 void tcp_wfree(struct sk_buff *skb);
359 void tcp_write_timer_handler(struct sock *sk);
360 void tcp_delack_timer_handler(struct sock *sk);
361 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
362 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
363 const struct tcphdr *th, unsigned int len);
364 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
365 const struct tcphdr *th, unsigned int len);
366 void tcp_rcv_space_adjust(struct sock *sk);
367 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
368 void tcp_twsk_destructor(struct sock *sk);
369 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
370 struct pipe_inode_info *pipe, size_t len,
373 static inline void tcp_dec_quickack_mode(struct sock *sk,
374 const unsigned int pkts)
376 struct inet_connection_sock *icsk = inet_csk(sk);
378 if (icsk->icsk_ack.quick) {
379 if (pkts >= icsk->icsk_ack.quick) {
380 icsk->icsk_ack.quick = 0;
381 /* Leaving quickack mode we deflate ATO. */
382 icsk->icsk_ack.ato = TCP_ATO_MIN;
384 icsk->icsk_ack.quick -= pkts;
389 #define TCP_ECN_QUEUE_CWR 2
390 #define TCP_ECN_DEMAND_CWR 4
391 #define TCP_ECN_SEEN 8
401 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
403 const struct tcphdr *th);
404 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
405 struct request_sock *req, struct request_sock **prev,
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(struct sock *sk, struct request_sock *req);
434 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
435 size_t len, int nonblock, 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 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
448 struct sock *tcp_create_openreq_child(struct sock *sk,
449 struct request_sock *req,
450 struct sk_buff *skb);
451 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
452 struct request_sock *req,
453 struct dst_entry *dst);
454 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
455 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
456 int tcp_connect(struct sock *sk);
457 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
458 struct request_sock *req,
459 struct tcp_fastopen_cookie *foc);
460 int tcp_disconnect(struct sock *sk, int flags);
462 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
463 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
464 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
466 /* From syncookies.c */
467 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
469 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
470 #ifdef CONFIG_SYN_COOKIES
472 /* Syncookies use a monotonic timer which increments every 60 seconds.
473 * This counter is used both as a hash input and partially encoded into
474 * the cookie value. A cookie is only validated further if the delta
475 * between the current counter value and the encoded one is less than this,
476 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
477 * the counter advances immediately after a cookie is generated).
479 #define MAX_SYNCOOKIE_AGE 2
481 static inline u32 tcp_cookie_time(void)
483 u64 val = get_jiffies_64();
485 do_div(val, 60 * HZ);
489 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
491 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
495 __u32 cookie_init_timestamp(struct request_sock *req);
496 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net,
499 /* From net/ipv6/syncookies.c */
500 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
502 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
503 #ifdef CONFIG_SYN_COOKIES
504 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
505 const struct tcphdr *th, u16 *mssp);
506 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
511 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
513 bool tcp_may_send_now(struct sock *sk);
514 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
515 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
516 void tcp_retransmit_timer(struct sock *sk);
517 void tcp_xmit_retransmit_queue(struct sock *);
518 void tcp_simple_retransmit(struct sock *);
519 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
520 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
522 void tcp_send_probe0(struct sock *);
523 void tcp_send_partial(struct sock *);
524 int tcp_write_wakeup(struct sock *);
525 void tcp_send_fin(struct sock *sk);
526 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
527 int tcp_send_synack(struct sock *);
528 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
530 void tcp_push_one(struct sock *, unsigned int mss_now);
531 void tcp_send_ack(struct sock *sk);
532 void tcp_send_delayed_ack(struct sock *sk);
533 void tcp_send_loss_probe(struct sock *sk);
534 bool tcp_schedule_loss_probe(struct sock *sk);
537 void tcp_resume_early_retransmit(struct sock *sk);
538 void tcp_rearm_rto(struct sock *sk);
539 void tcp_reset(struct sock *sk);
542 void tcp_init_xmit_timers(struct sock *);
543 static inline void tcp_clear_xmit_timers(struct sock *sk)
545 inet_csk_clear_xmit_timers(sk);
548 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
549 unsigned int tcp_current_mss(struct sock *sk);
551 /* Bound MSS / TSO packet size with the half of the window */
552 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
556 /* When peer uses tiny windows, there is no use in packetizing
557 * to sub-MSS pieces for the sake of SWS or making sure there
558 * are enough packets in the pipe for fast recovery.
560 * On the other hand, for extremely large MSS devices, handling
561 * smaller than MSS windows in this way does make sense.
563 if (tp->max_window >= 512)
564 cutoff = (tp->max_window >> 1);
566 cutoff = tp->max_window;
568 if (cutoff && pktsize > cutoff)
569 return max_t(int, cutoff, 68U - tp->tcp_header_len);
575 void tcp_get_info(const struct sock *, struct tcp_info *);
577 /* Read 'sendfile()'-style from a TCP socket */
578 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
579 unsigned int, size_t);
580 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
581 sk_read_actor_t recv_actor);
583 void tcp_initialize_rcv_mss(struct sock *sk);
585 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
586 int tcp_mss_to_mtu(struct sock *sk, int mss);
587 void tcp_mtup_init(struct sock *sk);
588 void tcp_init_buffer_space(struct sock *sk);
590 static inline void tcp_bound_rto(const struct sock *sk)
592 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
593 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
596 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
598 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
601 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
603 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
604 ntohl(TCP_FLAG_ACK) |
608 static inline void tcp_fast_path_on(struct tcp_sock *tp)
610 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
613 static inline void tcp_fast_path_check(struct sock *sk)
615 struct tcp_sock *tp = tcp_sk(sk);
617 if (skb_queue_empty(&tp->out_of_order_queue) &&
619 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
621 tcp_fast_path_on(tp);
624 /* Compute the actual rto_min value */
625 static inline u32 tcp_rto_min(struct sock *sk)
627 const struct dst_entry *dst = __sk_dst_get(sk);
628 u32 rto_min = TCP_RTO_MIN;
630 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
631 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
635 static inline u32 tcp_rto_min_us(struct sock *sk)
637 return jiffies_to_usecs(tcp_rto_min(sk));
640 /* Compute the actual receive window we are currently advertising.
641 * Rcv_nxt can be after the window if our peer push more data
642 * than the offered window.
644 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
646 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
653 /* Choose a new window, without checks for shrinking, and without
654 * scaling applied to the result. The caller does these things
655 * if necessary. This is a "raw" window selection.
657 u32 __tcp_select_window(struct sock *sk);
659 void tcp_send_window_probe(struct sock *sk);
661 /* TCP timestamps are only 32-bits, this causes a slight
662 * complication on 64-bit systems since we store a snapshot
663 * of jiffies in the buffer control blocks below. We decided
664 * to use only the low 32-bits of jiffies and hide the ugly
665 * casts with the following macro.
667 #define tcp_time_stamp ((__u32)(jiffies))
669 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
671 return skb->skb_mstamp.stamp_jiffies;
675 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
677 #define TCPHDR_FIN 0x01
678 #define TCPHDR_SYN 0x02
679 #define TCPHDR_RST 0x04
680 #define TCPHDR_PSH 0x08
681 #define TCPHDR_ACK 0x10
682 #define TCPHDR_URG 0x20
683 #define TCPHDR_ECE 0x40
684 #define TCPHDR_CWR 0x80
686 /* This is what the send packet queuing engine uses to pass
687 * TCP per-packet control information to the transmission code.
688 * We also store the host-order sequence numbers in here too.
689 * This is 44 bytes if IPV6 is enabled.
690 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
693 __u32 seq; /* Starting sequence number */
694 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
696 /* Note : tcp_tw_isn is used in input path only
697 * (isn chosen by tcp_timewait_state_process())
699 * tcp_gso_segs is used in write queue only,
700 * cf tcp_skb_pcount()
705 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
707 __u8 sacked; /* State flags for SACK/FACK. */
708 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
709 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
710 #define TCPCB_LOST 0x04 /* SKB is lost */
711 #define TCPCB_TAGBITS 0x07 /* All tag bits */
712 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
713 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
714 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
717 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
719 __u32 ack_seq; /* Sequence number ACK'd */
721 struct inet_skb_parm h4;
722 #if IS_ENABLED(CONFIG_IPV6)
723 struct inet6_skb_parm h6;
725 } header; /* For incoming frames */
728 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
731 #if IS_ENABLED(CONFIG_IPV6)
732 /* This is the variant of inet6_iif() that must be used by TCP,
733 * as TCP moves IP6CB into a different location in skb->cb[]
735 static inline int tcp_v6_iif(const struct sk_buff *skb)
737 return TCP_SKB_CB(skb)->header.h6.iif;
741 /* Due to TSO, an SKB can be composed of multiple actual
742 * packets. To keep these tracked properly, we use this.
744 static inline int tcp_skb_pcount(const struct sk_buff *skb)
746 return TCP_SKB_CB(skb)->tcp_gso_segs;
749 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
751 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
754 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
756 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
759 /* This is valid iff tcp_skb_pcount() > 1. */
760 static inline int tcp_skb_mss(const struct sk_buff *skb)
762 return skb_shinfo(skb)->gso_size;
765 /* Events passed to congestion control interface */
767 CA_EVENT_TX_START, /* first transmit when no packets in flight */
768 CA_EVENT_CWND_RESTART, /* congestion window restart */
769 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
770 CA_EVENT_LOSS, /* loss timeout */
771 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
772 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
773 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
774 CA_EVENT_NON_DELAYED_ACK,
777 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
778 enum tcp_ca_ack_event_flags {
779 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
780 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
781 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
785 * Interface for adding new TCP congestion control handlers
787 #define TCP_CA_NAME_MAX 16
788 #define TCP_CA_MAX 128
789 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
791 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
792 #define TCP_CONG_NON_RESTRICTED 0x1
793 /* Requires ECN/ECT set on all packets */
794 #define TCP_CONG_NEEDS_ECN 0x2
796 struct tcp_congestion_ops {
797 struct list_head list;
800 /* initialize private data (optional) */
801 void (*init)(struct sock *sk);
802 /* cleanup private data (optional) */
803 void (*release)(struct sock *sk);
805 /* return slow start threshold (required) */
806 u32 (*ssthresh)(struct sock *sk);
807 /* do new cwnd calculation (required) */
808 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
809 /* call before changing ca_state (optional) */
810 void (*set_state)(struct sock *sk, u8 new_state);
811 /* call when cwnd event occurs (optional) */
812 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
813 /* call when ack arrives (optional) */
814 void (*in_ack_event)(struct sock *sk, u32 flags);
815 /* new value of cwnd after loss (optional) */
816 u32 (*undo_cwnd)(struct sock *sk);
817 /* hook for packet ack accounting (optional) */
818 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
819 /* get info for inet_diag (optional) */
820 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
822 char name[TCP_CA_NAME_MAX];
823 struct module *owner;
826 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
827 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
829 void tcp_assign_congestion_control(struct sock *sk);
830 void tcp_init_congestion_control(struct sock *sk);
831 void tcp_cleanup_congestion_control(struct sock *sk);
832 int tcp_set_default_congestion_control(const char *name);
833 void tcp_get_default_congestion_control(char *name);
834 void tcp_get_available_congestion_control(char *buf, size_t len);
835 void tcp_get_allowed_congestion_control(char *buf, size_t len);
836 int tcp_set_allowed_congestion_control(char *allowed);
837 int tcp_set_congestion_control(struct sock *sk, const char *name);
838 void tcp_slow_start(struct tcp_sock *tp, u32 acked);
839 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
841 u32 tcp_reno_ssthresh(struct sock *sk);
842 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
843 extern struct tcp_congestion_ops tcp_reno;
845 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
847 const struct inet_connection_sock *icsk = inet_csk(sk);
849 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
852 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
854 struct inet_connection_sock *icsk = inet_csk(sk);
856 if (icsk->icsk_ca_ops->set_state)
857 icsk->icsk_ca_ops->set_state(sk, ca_state);
858 icsk->icsk_ca_state = ca_state;
861 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
863 const struct inet_connection_sock *icsk = inet_csk(sk);
865 if (icsk->icsk_ca_ops->cwnd_event)
866 icsk->icsk_ca_ops->cwnd_event(sk, event);
869 /* These functions determine how the current flow behaves in respect of SACK
870 * handling. SACK is negotiated with the peer, and therefore it can vary
871 * between different flows.
873 * tcp_is_sack - SACK enabled
874 * tcp_is_reno - No SACK
875 * tcp_is_fack - FACK enabled, implies SACK enabled
877 static inline int tcp_is_sack(const struct tcp_sock *tp)
879 return tp->rx_opt.sack_ok;
882 static inline bool tcp_is_reno(const struct tcp_sock *tp)
884 return !tcp_is_sack(tp);
887 static inline bool tcp_is_fack(const struct tcp_sock *tp)
889 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
892 static inline void tcp_enable_fack(struct tcp_sock *tp)
894 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
897 /* TCP early-retransmit (ER) is similar to but more conservative than
898 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
900 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
902 tp->do_early_retrans = sysctl_tcp_early_retrans &&
903 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
904 sysctl_tcp_reordering == 3;
907 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
909 tp->do_early_retrans = 0;
912 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
914 return tp->sacked_out + tp->lost_out;
917 /* This determines how many packets are "in the network" to the best
918 * of our knowledge. In many cases it is conservative, but where
919 * detailed information is available from the receiver (via SACK
920 * blocks etc.) we can make more aggressive calculations.
922 * Use this for decisions involving congestion control, use just
923 * tp->packets_out to determine if the send queue is empty or not.
925 * Read this equation as:
927 * "Packets sent once on transmission queue" MINUS
928 * "Packets left network, but not honestly ACKed yet" PLUS
929 * "Packets fast retransmitted"
931 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
933 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
936 #define TCP_INFINITE_SSTHRESH 0x7fffffff
938 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
940 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
943 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
945 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
946 (1 << inet_csk(sk)->icsk_ca_state);
949 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
950 * The exception is cwnd reduction phase, when cwnd is decreasing towards
953 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
955 const struct tcp_sock *tp = tcp_sk(sk);
957 if (tcp_in_cwnd_reduction(sk))
958 return tp->snd_ssthresh;
960 return max(tp->snd_ssthresh,
961 ((tp->snd_cwnd >> 1) +
962 (tp->snd_cwnd >> 2)));
965 /* Use define here intentionally to get WARN_ON location shown at the caller */
966 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
968 void tcp_enter_cwr(struct sock *sk);
969 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
971 /* The maximum number of MSS of available cwnd for which TSO defers
972 * sending if not using sysctl_tcp_tso_win_divisor.
974 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
979 /* Slow start with delack produces 3 packets of burst, so that
980 * it is safe "de facto". This will be the default - same as
981 * the default reordering threshold - but if reordering increases,
982 * we must be able to allow cwnd to burst at least this much in order
983 * to not pull it back when holes are filled.
985 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
987 return tp->reordering;
990 /* Returns end sequence number of the receiver's advertised window */
991 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
993 return tp->snd_una + tp->snd_wnd;
996 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
997 * flexible approach. The RFC suggests cwnd should not be raised unless
998 * it was fully used previously. And that's exactly what we do in
999 * congestion avoidance mode. But in slow start we allow cwnd to grow
1000 * as long as the application has used half the cwnd.
1002 * cwnd is 10 (IW10), but application sends 9 frames.
1003 * We allow cwnd to reach 18 when all frames are ACKed.
1004 * This check is safe because it's as aggressive as slow start which already
1005 * risks 100% overshoot. The advantage is that we discourage application to
1006 * either send more filler packets or data to artificially blow up the cwnd
1007 * usage, and allow application-limited process to probe bw more aggressively.
1009 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1011 const struct tcp_sock *tp = tcp_sk(sk);
1013 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1014 if (tp->snd_cwnd <= tp->snd_ssthresh)
1015 return tp->snd_cwnd < 2 * tp->max_packets_out;
1017 return tp->is_cwnd_limited;
1020 static inline void tcp_check_probe_timer(struct sock *sk)
1022 const struct tcp_sock *tp = tcp_sk(sk);
1023 const struct inet_connection_sock *icsk = inet_csk(sk);
1025 if (!tp->packets_out && !icsk->icsk_pending)
1026 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1027 icsk->icsk_rto, TCP_RTO_MAX);
1030 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1035 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1041 * Calculate(/check) TCP checksum
1043 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1044 __be32 daddr, __wsum base)
1046 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1049 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1051 return __skb_checksum_complete(skb);
1054 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1056 return !skb_csum_unnecessary(skb) &&
1057 __tcp_checksum_complete(skb);
1060 /* Prequeue for VJ style copy to user, combined with checksumming. */
1062 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1064 tp->ucopy.task = NULL;
1066 tp->ucopy.memory = 0;
1067 skb_queue_head_init(&tp->ucopy.prequeue);
1070 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1075 static const char *statename[]={
1076 "Unused","Established","Syn Sent","Syn Recv",
1077 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1078 "Close Wait","Last ACK","Listen","Closing"
1081 void tcp_set_state(struct sock *sk, int state);
1083 void tcp_done(struct sock *sk);
1085 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1088 rx_opt->num_sacks = 0;
1091 u32 tcp_default_init_rwnd(u32 mss);
1093 /* Determine a window scaling and initial window to offer. */
1094 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1095 __u32 *window_clamp, int wscale_ok,
1096 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1098 static inline int tcp_win_from_space(int space)
1100 return sysctl_tcp_adv_win_scale<=0 ?
1101 (space>>(-sysctl_tcp_adv_win_scale)) :
1102 space - (space>>sysctl_tcp_adv_win_scale);
1105 /* Note: caller must be prepared to deal with negative returns */
1106 static inline int tcp_space(const struct sock *sk)
1108 return tcp_win_from_space(sk->sk_rcvbuf -
1109 atomic_read(&sk->sk_rmem_alloc));
1112 static inline int tcp_full_space(const struct sock *sk)
1114 return tcp_win_from_space(sk->sk_rcvbuf);
1117 static inline void tcp_openreq_init(struct request_sock *req,
1118 struct tcp_options_received *rx_opt,
1119 struct sk_buff *skb, struct sock *sk)
1121 struct inet_request_sock *ireq = inet_rsk(req);
1123 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1125 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1126 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1127 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1128 req->mss = rx_opt->mss_clamp;
1129 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1130 ireq->tstamp_ok = rx_opt->tstamp_ok;
1131 ireq->sack_ok = rx_opt->sack_ok;
1132 ireq->snd_wscale = rx_opt->snd_wscale;
1133 ireq->wscale_ok = rx_opt->wscale_ok;
1136 ireq->ir_rmt_port = tcp_hdr(skb)->source;
1137 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
1138 ireq->ir_mark = inet_request_mark(sk, skb);
1141 extern void tcp_openreq_init_rwin(struct request_sock *req,
1142 struct sock *sk, struct dst_entry *dst);
1144 void tcp_enter_memory_pressure(struct sock *sk);
1146 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1148 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1151 static inline int keepalive_time_when(const struct tcp_sock *tp)
1153 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1156 static inline int keepalive_probes(const struct tcp_sock *tp)
1158 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1161 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1163 const struct inet_connection_sock *icsk = &tp->inet_conn;
1165 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1166 tcp_time_stamp - tp->rcv_tstamp);
1169 static inline int tcp_fin_time(const struct sock *sk)
1171 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1172 const int rto = inet_csk(sk)->icsk_rto;
1174 if (fin_timeout < (rto << 2) - (rto >> 1))
1175 fin_timeout = (rto << 2) - (rto >> 1);
1180 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1183 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1185 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1188 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1189 * then following tcp messages have valid values. Ignore 0 value,
1190 * or else 'negative' tsval might forbid us to accept their packets.
1192 if (!rx_opt->ts_recent)
1197 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1200 if (tcp_paws_check(rx_opt, 0))
1203 /* RST segments are not recommended to carry timestamp,
1204 and, if they do, it is recommended to ignore PAWS because
1205 "their cleanup function should take precedence over timestamps."
1206 Certainly, it is mistake. It is necessary to understand the reasons
1207 of this constraint to relax it: if peer reboots, clock may go
1208 out-of-sync and half-open connections will not be reset.
1209 Actually, the problem would be not existing if all
1210 the implementations followed draft about maintaining clock
1211 via reboots. Linux-2.2 DOES NOT!
1213 However, we can relax time bounds for RST segments to MSL.
1215 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1220 static inline void tcp_mib_init(struct net *net)
1223 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1224 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1225 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1226 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1230 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1232 tp->lost_skb_hint = NULL;
1235 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1237 tcp_clear_retrans_hints_partial(tp);
1238 tp->retransmit_skb_hint = NULL;
1244 union tcp_md5_addr {
1246 #if IS_ENABLED(CONFIG_IPV6)
1251 /* - key database */
1252 struct tcp_md5sig_key {
1253 struct hlist_node node;
1255 u8 family; /* AF_INET or AF_INET6 */
1256 union tcp_md5_addr addr;
1257 u8 key[TCP_MD5SIG_MAXKEYLEN];
1258 struct rcu_head rcu;
1262 struct tcp_md5sig_info {
1263 struct hlist_head head;
1264 struct rcu_head rcu;
1267 /* - pseudo header */
1268 struct tcp4_pseudohdr {
1276 struct tcp6_pseudohdr {
1277 struct in6_addr saddr;
1278 struct in6_addr daddr;
1280 __be32 protocol; /* including padding */
1283 union tcp_md5sum_block {
1284 struct tcp4_pseudohdr ip4;
1285 #if IS_ENABLED(CONFIG_IPV6)
1286 struct tcp6_pseudohdr ip6;
1290 /* - pool: digest algorithm, hash description and scratch buffer */
1291 struct tcp_md5sig_pool {
1292 struct hash_desc md5_desc;
1293 union tcp_md5sum_block md5_blk;
1297 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1298 const struct sock *sk, const struct request_sock *req,
1299 const struct sk_buff *skb);
1300 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1301 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1302 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1304 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1305 struct sock *addr_sk);
1307 #ifdef CONFIG_TCP_MD5SIG
1308 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1309 const union tcp_md5_addr *addr,
1311 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1313 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1314 const union tcp_md5_addr *addr,
1319 #define tcp_twsk_md5_key(twsk) NULL
1322 bool tcp_alloc_md5sig_pool(void);
1324 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1325 static inline void tcp_put_md5sig_pool(void)
1330 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1331 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1332 unsigned int header_len);
1333 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1334 const struct tcp_md5sig_key *key);
1336 /* From tcp_fastopen.c */
1337 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1338 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1339 unsigned long *last_syn_loss);
1340 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1341 struct tcp_fastopen_cookie *cookie, bool syn_lost);
1342 struct tcp_fastopen_request {
1343 /* Fast Open cookie. Size 0 means a cookie request */
1344 struct tcp_fastopen_cookie cookie;
1345 struct msghdr *data; /* data in MSG_FASTOPEN */
1347 int copied; /* queued in tcp_connect() */
1349 void tcp_free_fastopen_req(struct tcp_sock *tp);
1351 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1352 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1353 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1354 struct request_sock *req,
1355 struct tcp_fastopen_cookie *foc,
1356 struct dst_entry *dst);
1357 void tcp_fastopen_init_key_once(bool publish);
1358 #define TCP_FASTOPEN_KEY_LENGTH 16
1360 /* Fastopen key context */
1361 struct tcp_fastopen_context {
1362 struct crypto_cipher *tfm;
1363 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1364 struct rcu_head rcu;
1367 /* write queue abstraction */
1368 static inline void tcp_write_queue_purge(struct sock *sk)
1370 struct sk_buff *skb;
1372 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1373 sk_wmem_free_skb(sk, skb);
1375 tcp_clear_all_retrans_hints(tcp_sk(sk));
1378 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1380 return skb_peek(&sk->sk_write_queue);
1383 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1385 return skb_peek_tail(&sk->sk_write_queue);
1388 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1389 const struct sk_buff *skb)
1391 return skb_queue_next(&sk->sk_write_queue, skb);
1394 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1395 const struct sk_buff *skb)
1397 return skb_queue_prev(&sk->sk_write_queue, skb);
1400 #define tcp_for_write_queue(skb, sk) \
1401 skb_queue_walk(&(sk)->sk_write_queue, skb)
1403 #define tcp_for_write_queue_from(skb, sk) \
1404 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1406 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1407 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1409 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1411 return sk->sk_send_head;
1414 static inline bool tcp_skb_is_last(const struct sock *sk,
1415 const struct sk_buff *skb)
1417 return skb_queue_is_last(&sk->sk_write_queue, skb);
1420 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1422 if (tcp_skb_is_last(sk, skb))
1423 sk->sk_send_head = NULL;
1425 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1428 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1430 if (sk->sk_send_head == skb_unlinked)
1431 sk->sk_send_head = NULL;
1434 static inline void tcp_init_send_head(struct sock *sk)
1436 sk->sk_send_head = NULL;
1439 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1441 __skb_queue_tail(&sk->sk_write_queue, skb);
1444 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1446 __tcp_add_write_queue_tail(sk, skb);
1448 /* Queue it, remembering where we must start sending. */
1449 if (sk->sk_send_head == NULL) {
1450 sk->sk_send_head = skb;
1452 if (tcp_sk(sk)->highest_sack == NULL)
1453 tcp_sk(sk)->highest_sack = skb;
1457 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1459 __skb_queue_head(&sk->sk_write_queue, skb);
1462 /* Insert buff after skb on the write queue of sk. */
1463 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1464 struct sk_buff *buff,
1467 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1470 /* Insert new before skb on the write queue of sk. */
1471 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1472 struct sk_buff *skb,
1475 __skb_queue_before(&sk->sk_write_queue, skb, new);
1477 if (sk->sk_send_head == skb)
1478 sk->sk_send_head = new;
1481 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1483 __skb_unlink(skb, &sk->sk_write_queue);
1486 static inline bool tcp_write_queue_empty(struct sock *sk)
1488 return skb_queue_empty(&sk->sk_write_queue);
1491 static inline void tcp_push_pending_frames(struct sock *sk)
1493 if (tcp_send_head(sk)) {
1494 struct tcp_sock *tp = tcp_sk(sk);
1496 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1500 /* Start sequence of the skb just after the highest skb with SACKed
1501 * bit, valid only if sacked_out > 0 or when the caller has ensured
1502 * validity by itself.
1504 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1506 if (!tp->sacked_out)
1509 if (tp->highest_sack == NULL)
1512 return TCP_SKB_CB(tp->highest_sack)->seq;
1515 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1517 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1518 tcp_write_queue_next(sk, skb);
1521 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1523 return tcp_sk(sk)->highest_sack;
1526 static inline void tcp_highest_sack_reset(struct sock *sk)
1528 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1531 /* Called when old skb is about to be deleted (to be combined with new skb) */
1532 static inline void tcp_highest_sack_combine(struct sock *sk,
1533 struct sk_buff *old,
1534 struct sk_buff *new)
1536 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1537 tcp_sk(sk)->highest_sack = new;
1540 /* Determines whether this is a thin stream (which may suffer from
1541 * increased latency). Used to trigger latency-reducing mechanisms.
1543 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1545 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1549 enum tcp_seq_states {
1550 TCP_SEQ_STATE_LISTENING,
1551 TCP_SEQ_STATE_OPENREQ,
1552 TCP_SEQ_STATE_ESTABLISHED,
1555 int tcp_seq_open(struct inode *inode, struct file *file);
1557 struct tcp_seq_afinfo {
1560 const struct file_operations *seq_fops;
1561 struct seq_operations seq_ops;
1564 struct tcp_iter_state {
1565 struct seq_net_private p;
1567 enum tcp_seq_states state;
1568 struct sock *syn_wait_sk;
1569 int bucket, offset, sbucket, num;
1574 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1575 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1577 extern struct request_sock_ops tcp_request_sock_ops;
1578 extern struct request_sock_ops tcp6_request_sock_ops;
1580 void tcp_v4_destroy_sock(struct sock *sk);
1582 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1583 netdev_features_t features);
1584 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1585 int tcp_gro_complete(struct sk_buff *skb);
1587 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1589 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1591 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1594 static inline bool tcp_stream_memory_free(const struct sock *sk)
1596 const struct tcp_sock *tp = tcp_sk(sk);
1597 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1599 return notsent_bytes < tcp_notsent_lowat(tp);
1602 #ifdef CONFIG_PROC_FS
1603 int tcp4_proc_init(void);
1604 void tcp4_proc_exit(void);
1607 int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
1608 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1609 const struct tcp_request_sock_ops *af_ops,
1610 struct sock *sk, struct sk_buff *skb);
1612 /* TCP af-specific functions */
1613 struct tcp_sock_af_ops {
1614 #ifdef CONFIG_TCP_MD5SIG
1615 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1616 struct sock *addr_sk);
1617 int (*calc_md5_hash) (char *location,
1618 struct tcp_md5sig_key *md5,
1619 const struct sock *sk,
1620 const struct request_sock *req,
1621 const struct sk_buff *skb);
1622 int (*md5_parse) (struct sock *sk,
1623 char __user *optval,
1628 struct tcp_request_sock_ops {
1630 #ifdef CONFIG_TCP_MD5SIG
1631 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1632 struct request_sock *req);
1633 int (*calc_md5_hash) (char *location,
1634 struct tcp_md5sig_key *md5,
1635 const struct sock *sk,
1636 const struct request_sock *req,
1637 const struct sk_buff *skb);
1639 void (*init_req)(struct request_sock *req, struct sock *sk,
1640 struct sk_buff *skb);
1641 #ifdef CONFIG_SYN_COOKIES
1642 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1645 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1646 const struct request_sock *req,
1648 __u32 (*init_seq)(const struct sk_buff *skb);
1649 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1650 struct flowi *fl, struct request_sock *req,
1651 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
1652 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1653 const unsigned long timeout);
1656 #ifdef CONFIG_SYN_COOKIES
1657 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1658 struct sock *sk, struct sk_buff *skb,
1661 return ops->cookie_init_seq(sk, skb, mss);
1664 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1665 struct sock *sk, struct sk_buff *skb,
1672 int tcpv4_offload_init(void);
1674 void tcp_v4_init(void);
1675 void tcp_init(void);
1678 * Save and compile IPv4 options, return a pointer to it
1680 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1682 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1683 struct ip_options_rcu *dopt = NULL;
1686 int opt_size = sizeof(*dopt) + opt->optlen;
1688 dopt = kmalloc(opt_size, GFP_ATOMIC);
1689 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {