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;
277 extern int sysctl_tcp_invalid_ratelimit;
279 extern atomic_long_t tcp_memory_allocated;
280 extern struct percpu_counter tcp_sockets_allocated;
281 extern int tcp_memory_pressure;
284 * The next routines deal with comparing 32 bit unsigned ints
285 * and worry about wraparound (automatic with unsigned arithmetic).
288 static inline bool before(__u32 seq1, __u32 seq2)
290 return (__s32)(seq1-seq2) < 0;
292 #define after(seq2, seq1) before(seq1, seq2)
294 /* is s2<=s1<=s3 ? */
295 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
297 return seq3 - seq2 >= seq1 - seq2;
300 static inline bool tcp_out_of_memory(struct sock *sk)
302 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
303 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
308 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
310 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
311 int orphans = percpu_counter_read_positive(ocp);
313 if (orphans << shift > sysctl_tcp_max_orphans) {
314 orphans = percpu_counter_sum_positive(ocp);
315 if (orphans << shift > sysctl_tcp_max_orphans)
321 bool tcp_check_oom(struct sock *sk, int shift);
323 /* syncookies: remember time of last synqueue overflow */
324 static inline void tcp_synq_overflow(struct sock *sk)
326 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
329 /* syncookies: no recent synqueue overflow on this listening socket? */
330 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
332 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
333 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
336 extern struct proto tcp_prot;
338 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
339 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
340 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
341 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
342 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
344 void tcp_tasklet_init(void);
346 void tcp_v4_err(struct sk_buff *skb, u32);
348 void tcp_shutdown(struct sock *sk, int how);
350 void tcp_v4_early_demux(struct sk_buff *skb);
351 int tcp_v4_rcv(struct sk_buff *skb);
353 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
354 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
356 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
358 void tcp_release_cb(struct sock *sk);
359 void tcp_wfree(struct sk_buff *skb);
360 void tcp_write_timer_handler(struct sock *sk);
361 void tcp_delack_timer_handler(struct sock *sk);
362 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
363 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
364 const struct tcphdr *th, unsigned int len);
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, struct request_sock **prev,
408 int tcp_child_process(struct sock *parent, struct sock *child,
409 struct sk_buff *skb);
410 void tcp_enter_loss(struct sock *sk);
411 void tcp_clear_retrans(struct tcp_sock *tp);
412 void tcp_update_metrics(struct sock *sk);
413 void tcp_init_metrics(struct sock *sk);
414 void tcp_metrics_init(void);
415 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
416 bool paws_check, bool timestamps);
417 bool tcp_remember_stamp(struct sock *sk);
418 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
419 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
420 void tcp_disable_fack(struct tcp_sock *tp);
421 void tcp_close(struct sock *sk, long timeout);
422 void tcp_init_sock(struct sock *sk);
423 unsigned int tcp_poll(struct file *file, struct socket *sock,
424 struct poll_table_struct *wait);
425 int tcp_getsockopt(struct sock *sk, int level, int optname,
426 char __user *optval, int __user *optlen);
427 int tcp_setsockopt(struct sock *sk, int level, int optname,
428 char __user *optval, unsigned int optlen);
429 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, int __user *optlen);
431 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
432 char __user *optval, unsigned int optlen);
433 void tcp_set_keepalive(struct sock *sk, int val);
434 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
435 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
436 size_t len, int nonblock, int flags, int *addr_len);
437 void tcp_parse_options(const struct sk_buff *skb,
438 struct tcp_options_received *opt_rx,
439 int estab, struct tcp_fastopen_cookie *foc);
440 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
443 * TCP v4 functions exported for the inet6 API
446 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
447 void tcp_v4_mtu_reduced(struct sock *sk);
448 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
449 struct sock *tcp_create_openreq_child(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(struct sock *sk, struct sk_buff *skb,
454 struct request_sock *req,
455 struct dst_entry *dst);
456 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
457 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
458 int tcp_connect(struct sock *sk);
459 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
460 struct request_sock *req,
461 struct tcp_fastopen_cookie *foc);
462 int tcp_disconnect(struct sock *sk, int flags);
464 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
465 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
466 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
468 /* From syncookies.c */
469 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
471 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
472 #ifdef CONFIG_SYN_COOKIES
474 /* Syncookies use a monotonic timer which increments every 60 seconds.
475 * This counter is used both as a hash input and partially encoded into
476 * the cookie value. A cookie is only validated further if the delta
477 * between the current counter value and the encoded one is less than this,
478 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
479 * the counter advances immediately after a cookie is generated).
481 #define MAX_SYNCOOKIE_AGE 2
483 static inline u32 tcp_cookie_time(void)
485 u64 val = get_jiffies_64();
487 do_div(val, 60 * HZ);
491 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
493 __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_timestamp_decode(struct tcp_options_received *opt);
497 bool cookie_ecn_ok(const struct tcp_options_received *opt,
498 const struct net *net, const struct dst_entry *dst);
500 /* From net/ipv6/syncookies.c */
501 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
503 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
505 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
506 const struct tcphdr *th, u16 *mssp);
507 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
512 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
514 bool tcp_may_send_now(struct sock *sk);
515 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
516 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
517 void tcp_retransmit_timer(struct sock *sk);
518 void tcp_xmit_retransmit_queue(struct sock *);
519 void tcp_simple_retransmit(struct sock *);
520 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
521 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
523 void tcp_send_probe0(struct sock *);
524 void tcp_send_partial(struct sock *);
525 int tcp_write_wakeup(struct sock *);
526 void tcp_send_fin(struct sock *sk);
527 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
528 int tcp_send_synack(struct sock *);
529 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
531 void tcp_push_one(struct sock *, unsigned int mss_now);
532 void tcp_send_ack(struct sock *sk);
533 void tcp_send_delayed_ack(struct sock *sk);
534 void tcp_send_loss_probe(struct sock *sk);
535 bool tcp_schedule_loss_probe(struct sock *sk);
538 void tcp_resume_early_retransmit(struct sock *sk);
539 void tcp_rearm_rto(struct sock *sk);
540 void tcp_reset(struct sock *sk);
543 void tcp_init_xmit_timers(struct sock *);
544 static inline void tcp_clear_xmit_timers(struct sock *sk)
546 inet_csk_clear_xmit_timers(sk);
549 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
550 unsigned int tcp_current_mss(struct sock *sk);
552 /* Bound MSS / TSO packet size with the half of the window */
553 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
557 /* When peer uses tiny windows, there is no use in packetizing
558 * to sub-MSS pieces for the sake of SWS or making sure there
559 * are enough packets in the pipe for fast recovery.
561 * On the other hand, for extremely large MSS devices, handling
562 * smaller than MSS windows in this way does make sense.
564 if (tp->max_window >= 512)
565 cutoff = (tp->max_window >> 1);
567 cutoff = tp->max_window;
569 if (cutoff && pktsize > cutoff)
570 return max_t(int, cutoff, 68U - tp->tcp_header_len);
576 void tcp_get_info(const struct sock *, struct tcp_info *);
578 /* Read 'sendfile()'-style from a TCP socket */
579 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
580 unsigned int, size_t);
581 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
582 sk_read_actor_t recv_actor);
584 void tcp_initialize_rcv_mss(struct sock *sk);
586 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
587 int tcp_mss_to_mtu(struct sock *sk, int mss);
588 void tcp_mtup_init(struct sock *sk);
589 void tcp_init_buffer_space(struct sock *sk);
591 static inline void tcp_bound_rto(const struct sock *sk)
593 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
594 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
597 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
599 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
602 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
604 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
605 ntohl(TCP_FLAG_ACK) |
609 static inline void tcp_fast_path_on(struct tcp_sock *tp)
611 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
614 static inline void tcp_fast_path_check(struct sock *sk)
616 struct tcp_sock *tp = tcp_sk(sk);
618 if (skb_queue_empty(&tp->out_of_order_queue) &&
620 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
622 tcp_fast_path_on(tp);
625 /* Compute the actual rto_min value */
626 static inline u32 tcp_rto_min(struct sock *sk)
628 const struct dst_entry *dst = __sk_dst_get(sk);
629 u32 rto_min = TCP_RTO_MIN;
631 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
632 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
636 static inline u32 tcp_rto_min_us(struct sock *sk)
638 return jiffies_to_usecs(tcp_rto_min(sk));
641 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
643 return dst_metric_locked(dst, RTAX_CC_ALGO);
646 /* Compute the actual receive window we are currently advertising.
647 * Rcv_nxt can be after the window if our peer push more data
648 * than the offered window.
650 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
652 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
659 /* Choose a new window, without checks for shrinking, and without
660 * scaling applied to the result. The caller does these things
661 * if necessary. This is a "raw" window selection.
663 u32 __tcp_select_window(struct sock *sk);
665 void tcp_send_window_probe(struct sock *sk);
667 /* TCP timestamps are only 32-bits, this causes a slight
668 * complication on 64-bit systems since we store a snapshot
669 * of jiffies in the buffer control blocks below. We decided
670 * to use only the low 32-bits of jiffies and hide the ugly
671 * casts with the following macro.
673 #define tcp_time_stamp ((__u32)(jiffies))
675 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
677 return skb->skb_mstamp.stamp_jiffies;
681 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
683 #define TCPHDR_FIN 0x01
684 #define TCPHDR_SYN 0x02
685 #define TCPHDR_RST 0x04
686 #define TCPHDR_PSH 0x08
687 #define TCPHDR_ACK 0x10
688 #define TCPHDR_URG 0x20
689 #define TCPHDR_ECE 0x40
690 #define TCPHDR_CWR 0x80
692 /* This is what the send packet queuing engine uses to pass
693 * TCP per-packet control information to the transmission code.
694 * We also store the host-order sequence numbers in here too.
695 * This is 44 bytes if IPV6 is enabled.
696 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
699 __u32 seq; /* Starting sequence number */
700 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
702 /* Note : tcp_tw_isn is used in input path only
703 * (isn chosen by tcp_timewait_state_process())
705 * tcp_gso_segs is used in write queue only,
706 * cf tcp_skb_pcount()
711 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
713 __u8 sacked; /* State flags for SACK/FACK. */
714 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
715 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
716 #define TCPCB_LOST 0x04 /* SKB is lost */
717 #define TCPCB_TAGBITS 0x07 /* All tag bits */
718 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
719 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
720 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
723 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
725 __u32 ack_seq; /* Sequence number ACK'd */
727 struct inet_skb_parm h4;
728 #if IS_ENABLED(CONFIG_IPV6)
729 struct inet6_skb_parm h6;
731 } header; /* For incoming frames */
734 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
737 #if IS_ENABLED(CONFIG_IPV6)
738 /* This is the variant of inet6_iif() that must be used by TCP,
739 * as TCP moves IP6CB into a different location in skb->cb[]
741 static inline int tcp_v6_iif(const struct sk_buff *skb)
743 return TCP_SKB_CB(skb)->header.h6.iif;
747 /* Due to TSO, an SKB can be composed of multiple actual
748 * packets. To keep these tracked properly, we use this.
750 static inline int tcp_skb_pcount(const struct sk_buff *skb)
752 return TCP_SKB_CB(skb)->tcp_gso_segs;
755 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
757 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
760 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
762 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
765 /* This is valid iff tcp_skb_pcount() > 1. */
766 static inline int tcp_skb_mss(const struct sk_buff *skb)
768 return skb_shinfo(skb)->gso_size;
771 /* Events passed to congestion control interface */
773 CA_EVENT_TX_START, /* first transmit when no packets in flight */
774 CA_EVENT_CWND_RESTART, /* congestion window restart */
775 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
776 CA_EVENT_LOSS, /* loss timeout */
777 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
778 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
779 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
780 CA_EVENT_NON_DELAYED_ACK,
783 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
784 enum tcp_ca_ack_event_flags {
785 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
786 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
787 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
791 * Interface for adding new TCP congestion control handlers
793 #define TCP_CA_NAME_MAX 16
794 #define TCP_CA_MAX 128
795 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
797 #define TCP_CA_UNSPEC 0
799 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
800 #define TCP_CONG_NON_RESTRICTED 0x1
801 /* Requires ECN/ECT set on all packets */
802 #define TCP_CONG_NEEDS_ECN 0x2
804 struct tcp_congestion_ops {
805 struct list_head list;
809 /* initialize private data (optional) */
810 void (*init)(struct sock *sk);
811 /* cleanup private data (optional) */
812 void (*release)(struct sock *sk);
814 /* return slow start threshold (required) */
815 u32 (*ssthresh)(struct sock *sk);
816 /* do new cwnd calculation (required) */
817 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
818 /* call before changing ca_state (optional) */
819 void (*set_state)(struct sock *sk, u8 new_state);
820 /* call when cwnd event occurs (optional) */
821 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
822 /* call when ack arrives (optional) */
823 void (*in_ack_event)(struct sock *sk, u32 flags);
824 /* new value of cwnd after loss (optional) */
825 u32 (*undo_cwnd)(struct sock *sk);
826 /* hook for packet ack accounting (optional) */
827 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
828 /* get info for inet_diag (optional) */
829 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
831 char name[TCP_CA_NAME_MAX];
832 struct module *owner;
835 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
836 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
838 void tcp_assign_congestion_control(struct sock *sk);
839 void tcp_init_congestion_control(struct sock *sk);
840 void tcp_cleanup_congestion_control(struct sock *sk);
841 int tcp_set_default_congestion_control(const char *name);
842 void tcp_get_default_congestion_control(char *name);
843 void tcp_get_available_congestion_control(char *buf, size_t len);
844 void tcp_get_allowed_congestion_control(char *buf, size_t len);
845 int tcp_set_allowed_congestion_control(char *allowed);
846 int tcp_set_congestion_control(struct sock *sk, const char *name);
847 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
848 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
850 u32 tcp_reno_ssthresh(struct sock *sk);
851 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
852 extern struct tcp_congestion_ops tcp_reno;
854 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
855 u32 tcp_ca_get_key_by_name(const char *name);
857 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
859 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
865 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
867 const struct inet_connection_sock *icsk = inet_csk(sk);
869 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
872 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
874 struct inet_connection_sock *icsk = inet_csk(sk);
876 if (icsk->icsk_ca_ops->set_state)
877 icsk->icsk_ca_ops->set_state(sk, ca_state);
878 icsk->icsk_ca_state = ca_state;
881 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
883 const struct inet_connection_sock *icsk = inet_csk(sk);
885 if (icsk->icsk_ca_ops->cwnd_event)
886 icsk->icsk_ca_ops->cwnd_event(sk, event);
889 /* These functions determine how the current flow behaves in respect of SACK
890 * handling. SACK is negotiated with the peer, and therefore it can vary
891 * between different flows.
893 * tcp_is_sack - SACK enabled
894 * tcp_is_reno - No SACK
895 * tcp_is_fack - FACK enabled, implies SACK enabled
897 static inline int tcp_is_sack(const struct tcp_sock *tp)
899 return tp->rx_opt.sack_ok;
902 static inline bool tcp_is_reno(const struct tcp_sock *tp)
904 return !tcp_is_sack(tp);
907 static inline bool tcp_is_fack(const struct tcp_sock *tp)
909 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
912 static inline void tcp_enable_fack(struct tcp_sock *tp)
914 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
917 /* TCP early-retransmit (ER) is similar to but more conservative than
918 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
920 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
922 tp->do_early_retrans = sysctl_tcp_early_retrans &&
923 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
924 sysctl_tcp_reordering == 3;
927 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
929 tp->do_early_retrans = 0;
932 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
934 return tp->sacked_out + tp->lost_out;
937 /* This determines how many packets are "in the network" to the best
938 * of our knowledge. In many cases it is conservative, but where
939 * detailed information is available from the receiver (via SACK
940 * blocks etc.) we can make more aggressive calculations.
942 * Use this for decisions involving congestion control, use just
943 * tp->packets_out to determine if the send queue is empty or not.
945 * Read this equation as:
947 * "Packets sent once on transmission queue" MINUS
948 * "Packets left network, but not honestly ACKed yet" PLUS
949 * "Packets fast retransmitted"
951 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
953 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
956 #define TCP_INFINITE_SSTHRESH 0x7fffffff
958 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
960 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
963 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
965 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
966 (1 << inet_csk(sk)->icsk_ca_state);
969 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
970 * The exception is cwnd reduction phase, when cwnd is decreasing towards
973 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
975 const struct tcp_sock *tp = tcp_sk(sk);
977 if (tcp_in_cwnd_reduction(sk))
978 return tp->snd_ssthresh;
980 return max(tp->snd_ssthresh,
981 ((tp->snd_cwnd >> 1) +
982 (tp->snd_cwnd >> 2)));
985 /* Use define here intentionally to get WARN_ON location shown at the caller */
986 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
988 void tcp_enter_cwr(struct sock *sk);
989 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
991 /* The maximum number of MSS of available cwnd for which TSO defers
992 * sending if not using sysctl_tcp_tso_win_divisor.
994 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
999 /* Slow start with delack produces 3 packets of burst, so that
1000 * it is safe "de facto". This will be the default - same as
1001 * the default reordering threshold - but if reordering increases,
1002 * we must be able to allow cwnd to burst at least this much in order
1003 * to not pull it back when holes are filled.
1005 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1007 return tp->reordering;
1010 /* Returns end sequence number of the receiver's advertised window */
1011 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1013 return tp->snd_una + tp->snd_wnd;
1016 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1017 * flexible approach. The RFC suggests cwnd should not be raised unless
1018 * it was fully used previously. And that's exactly what we do in
1019 * congestion avoidance mode. But in slow start we allow cwnd to grow
1020 * as long as the application has used half the cwnd.
1022 * cwnd is 10 (IW10), but application sends 9 frames.
1023 * We allow cwnd to reach 18 when all frames are ACKed.
1024 * This check is safe because it's as aggressive as slow start which already
1025 * risks 100% overshoot. The advantage is that we discourage application to
1026 * either send more filler packets or data to artificially blow up the cwnd
1027 * usage, and allow application-limited process to probe bw more aggressively.
1029 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1031 const struct tcp_sock *tp = tcp_sk(sk);
1033 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1034 if (tp->snd_cwnd <= tp->snd_ssthresh)
1035 return tp->snd_cwnd < 2 * tp->max_packets_out;
1037 return tp->is_cwnd_limited;
1040 static inline void tcp_check_probe_timer(struct sock *sk)
1042 const struct tcp_sock *tp = tcp_sk(sk);
1043 const struct inet_connection_sock *icsk = inet_csk(sk);
1045 if (!tp->packets_out && !icsk->icsk_pending)
1046 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1047 icsk->icsk_rto, TCP_RTO_MAX);
1050 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1055 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1061 * Calculate(/check) TCP checksum
1063 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1064 __be32 daddr, __wsum base)
1066 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1069 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1071 return __skb_checksum_complete(skb);
1074 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1076 return !skb_csum_unnecessary(skb) &&
1077 __tcp_checksum_complete(skb);
1080 /* Prequeue for VJ style copy to user, combined with checksumming. */
1082 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1084 tp->ucopy.task = NULL;
1086 tp->ucopy.memory = 0;
1087 skb_queue_head_init(&tp->ucopy.prequeue);
1090 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1095 static const char *statename[]={
1096 "Unused","Established","Syn Sent","Syn Recv",
1097 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1098 "Close Wait","Last ACK","Listen","Closing"
1101 void tcp_set_state(struct sock *sk, int state);
1103 void tcp_done(struct sock *sk);
1105 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1108 rx_opt->num_sacks = 0;
1111 u32 tcp_default_init_rwnd(u32 mss);
1113 /* Determine a window scaling and initial window to offer. */
1114 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1115 __u32 *window_clamp, int wscale_ok,
1116 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1118 static inline int tcp_win_from_space(int space)
1120 return sysctl_tcp_adv_win_scale<=0 ?
1121 (space>>(-sysctl_tcp_adv_win_scale)) :
1122 space - (space>>sysctl_tcp_adv_win_scale);
1125 /* Note: caller must be prepared to deal with negative returns */
1126 static inline int tcp_space(const struct sock *sk)
1128 return tcp_win_from_space(sk->sk_rcvbuf -
1129 atomic_read(&sk->sk_rmem_alloc));
1132 static inline int tcp_full_space(const struct sock *sk)
1134 return tcp_win_from_space(sk->sk_rcvbuf);
1137 static inline void tcp_openreq_init(struct request_sock *req,
1138 struct tcp_options_received *rx_opt,
1139 struct sk_buff *skb, struct sock *sk)
1141 struct inet_request_sock *ireq = inet_rsk(req);
1143 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1145 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1146 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1147 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1148 req->mss = rx_opt->mss_clamp;
1149 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1150 ireq->tstamp_ok = rx_opt->tstamp_ok;
1151 ireq->sack_ok = rx_opt->sack_ok;
1152 ireq->snd_wscale = rx_opt->snd_wscale;
1153 ireq->wscale_ok = rx_opt->wscale_ok;
1156 ireq->ir_rmt_port = tcp_hdr(skb)->source;
1157 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
1158 ireq->ir_mark = inet_request_mark(sk, skb);
1161 extern void tcp_openreq_init_rwin(struct request_sock *req,
1162 struct sock *sk, struct dst_entry *dst);
1164 void tcp_enter_memory_pressure(struct sock *sk);
1166 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1168 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1171 static inline int keepalive_time_when(const struct tcp_sock *tp)
1173 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1176 static inline int keepalive_probes(const struct tcp_sock *tp)
1178 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1181 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1183 const struct inet_connection_sock *icsk = &tp->inet_conn;
1185 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1186 tcp_time_stamp - tp->rcv_tstamp);
1189 static inline int tcp_fin_time(const struct sock *sk)
1191 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1192 const int rto = inet_csk(sk)->icsk_rto;
1194 if (fin_timeout < (rto << 2) - (rto >> 1))
1195 fin_timeout = (rto << 2) - (rto >> 1);
1200 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1203 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1205 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1208 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1209 * then following tcp messages have valid values. Ignore 0 value,
1210 * or else 'negative' tsval might forbid us to accept their packets.
1212 if (!rx_opt->ts_recent)
1217 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1220 if (tcp_paws_check(rx_opt, 0))
1223 /* RST segments are not recommended to carry timestamp,
1224 and, if they do, it is recommended to ignore PAWS because
1225 "their cleanup function should take precedence over timestamps."
1226 Certainly, it is mistake. It is necessary to understand the reasons
1227 of this constraint to relax it: if peer reboots, clock may go
1228 out-of-sync and half-open connections will not be reset.
1229 Actually, the problem would be not existing if all
1230 the implementations followed draft about maintaining clock
1231 via reboots. Linux-2.2 DOES NOT!
1233 However, we can relax time bounds for RST segments to MSL.
1235 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1240 /* Return true if we're currently rate-limiting out-of-window ACKs and
1241 * thus shouldn't send a dupack right now. We rate-limit dupacks in
1242 * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
1243 * attacks that send repeated SYNs or ACKs for the same connection. To
1244 * do this, we do not send a duplicate SYNACK or ACK if the remote
1245 * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
1247 static inline bool tcp_oow_rate_limited(struct net *net,
1248 const struct sk_buff *skb,
1249 int mib_idx, u32 *last_oow_ack_time)
1251 /* Data packets without SYNs are not likely part of an ACK loop. */
1252 if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) &&
1254 goto not_rate_limited;
1256 if (*last_oow_ack_time) {
1257 s32 elapsed = (s32)(tcp_time_stamp - *last_oow_ack_time);
1259 if (0 <= elapsed && elapsed < sysctl_tcp_invalid_ratelimit) {
1260 NET_INC_STATS_BH(net, mib_idx);
1261 return true; /* rate-limited: don't send yet! */
1265 *last_oow_ack_time = tcp_time_stamp;
1268 return false; /* not rate-limited: go ahead, send dupack now! */
1271 static inline void tcp_mib_init(struct net *net)
1274 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1275 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1276 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1277 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1281 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1283 tp->lost_skb_hint = NULL;
1286 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1288 tcp_clear_retrans_hints_partial(tp);
1289 tp->retransmit_skb_hint = NULL;
1295 union tcp_md5_addr {
1297 #if IS_ENABLED(CONFIG_IPV6)
1302 /* - key database */
1303 struct tcp_md5sig_key {
1304 struct hlist_node node;
1306 u8 family; /* AF_INET or AF_INET6 */
1307 union tcp_md5_addr addr;
1308 u8 key[TCP_MD5SIG_MAXKEYLEN];
1309 struct rcu_head rcu;
1313 struct tcp_md5sig_info {
1314 struct hlist_head head;
1315 struct rcu_head rcu;
1318 /* - pseudo header */
1319 struct tcp4_pseudohdr {
1327 struct tcp6_pseudohdr {
1328 struct in6_addr saddr;
1329 struct in6_addr daddr;
1331 __be32 protocol; /* including padding */
1334 union tcp_md5sum_block {
1335 struct tcp4_pseudohdr ip4;
1336 #if IS_ENABLED(CONFIG_IPV6)
1337 struct tcp6_pseudohdr ip6;
1341 /* - pool: digest algorithm, hash description and scratch buffer */
1342 struct tcp_md5sig_pool {
1343 struct hash_desc md5_desc;
1344 union tcp_md5sum_block md5_blk;
1348 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1349 const struct sock *sk, const struct request_sock *req,
1350 const struct sk_buff *skb);
1351 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1352 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1353 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1355 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1356 struct sock *addr_sk);
1358 #ifdef CONFIG_TCP_MD5SIG
1359 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1360 const union tcp_md5_addr *addr,
1362 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1364 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1365 const union tcp_md5_addr *addr,
1370 #define tcp_twsk_md5_key(twsk) NULL
1373 bool tcp_alloc_md5sig_pool(void);
1375 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1376 static inline void tcp_put_md5sig_pool(void)
1381 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1382 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1383 unsigned int header_len);
1384 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1385 const struct tcp_md5sig_key *key);
1387 /* From tcp_fastopen.c */
1388 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1389 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1390 unsigned long *last_syn_loss);
1391 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1392 struct tcp_fastopen_cookie *cookie, bool syn_lost);
1393 struct tcp_fastopen_request {
1394 /* Fast Open cookie. Size 0 means a cookie request */
1395 struct tcp_fastopen_cookie cookie;
1396 struct msghdr *data; /* data in MSG_FASTOPEN */
1398 int copied; /* queued in tcp_connect() */
1400 void tcp_free_fastopen_req(struct tcp_sock *tp);
1402 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1403 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1404 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1405 struct request_sock *req,
1406 struct tcp_fastopen_cookie *foc,
1407 struct dst_entry *dst);
1408 void tcp_fastopen_init_key_once(bool publish);
1409 #define TCP_FASTOPEN_KEY_LENGTH 16
1411 /* Fastopen key context */
1412 struct tcp_fastopen_context {
1413 struct crypto_cipher *tfm;
1414 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1415 struct rcu_head rcu;
1418 /* write queue abstraction */
1419 static inline void tcp_write_queue_purge(struct sock *sk)
1421 struct sk_buff *skb;
1423 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1424 sk_wmem_free_skb(sk, skb);
1426 tcp_clear_all_retrans_hints(tcp_sk(sk));
1429 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1431 return skb_peek(&sk->sk_write_queue);
1434 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1436 return skb_peek_tail(&sk->sk_write_queue);
1439 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1440 const struct sk_buff *skb)
1442 return skb_queue_next(&sk->sk_write_queue, skb);
1445 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1446 const struct sk_buff *skb)
1448 return skb_queue_prev(&sk->sk_write_queue, skb);
1451 #define tcp_for_write_queue(skb, sk) \
1452 skb_queue_walk(&(sk)->sk_write_queue, skb)
1454 #define tcp_for_write_queue_from(skb, sk) \
1455 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1457 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1458 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1460 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1462 return sk->sk_send_head;
1465 static inline bool tcp_skb_is_last(const struct sock *sk,
1466 const struct sk_buff *skb)
1468 return skb_queue_is_last(&sk->sk_write_queue, skb);
1471 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1473 if (tcp_skb_is_last(sk, skb))
1474 sk->sk_send_head = NULL;
1476 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1479 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1481 if (sk->sk_send_head == skb_unlinked)
1482 sk->sk_send_head = NULL;
1485 static inline void tcp_init_send_head(struct sock *sk)
1487 sk->sk_send_head = NULL;
1490 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1492 __skb_queue_tail(&sk->sk_write_queue, skb);
1495 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1497 __tcp_add_write_queue_tail(sk, skb);
1499 /* Queue it, remembering where we must start sending. */
1500 if (sk->sk_send_head == NULL) {
1501 sk->sk_send_head = skb;
1503 if (tcp_sk(sk)->highest_sack == NULL)
1504 tcp_sk(sk)->highest_sack = skb;
1508 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1510 __skb_queue_head(&sk->sk_write_queue, skb);
1513 /* Insert buff after skb on the write queue of sk. */
1514 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1515 struct sk_buff *buff,
1518 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1521 /* Insert new before skb on the write queue of sk. */
1522 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1523 struct sk_buff *skb,
1526 __skb_queue_before(&sk->sk_write_queue, skb, new);
1528 if (sk->sk_send_head == skb)
1529 sk->sk_send_head = new;
1532 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1534 __skb_unlink(skb, &sk->sk_write_queue);
1537 static inline bool tcp_write_queue_empty(struct sock *sk)
1539 return skb_queue_empty(&sk->sk_write_queue);
1542 static inline void tcp_push_pending_frames(struct sock *sk)
1544 if (tcp_send_head(sk)) {
1545 struct tcp_sock *tp = tcp_sk(sk);
1547 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1551 /* Start sequence of the skb just after the highest skb with SACKed
1552 * bit, valid only if sacked_out > 0 or when the caller has ensured
1553 * validity by itself.
1555 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1557 if (!tp->sacked_out)
1560 if (tp->highest_sack == NULL)
1563 return TCP_SKB_CB(tp->highest_sack)->seq;
1566 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1568 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1569 tcp_write_queue_next(sk, skb);
1572 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1574 return tcp_sk(sk)->highest_sack;
1577 static inline void tcp_highest_sack_reset(struct sock *sk)
1579 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1582 /* Called when old skb is about to be deleted (to be combined with new skb) */
1583 static inline void tcp_highest_sack_combine(struct sock *sk,
1584 struct sk_buff *old,
1585 struct sk_buff *new)
1587 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1588 tcp_sk(sk)->highest_sack = new;
1591 /* Determines whether this is a thin stream (which may suffer from
1592 * increased latency). Used to trigger latency-reducing mechanisms.
1594 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1596 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1600 enum tcp_seq_states {
1601 TCP_SEQ_STATE_LISTENING,
1602 TCP_SEQ_STATE_OPENREQ,
1603 TCP_SEQ_STATE_ESTABLISHED,
1606 int tcp_seq_open(struct inode *inode, struct file *file);
1608 struct tcp_seq_afinfo {
1611 const struct file_operations *seq_fops;
1612 struct seq_operations seq_ops;
1615 struct tcp_iter_state {
1616 struct seq_net_private p;
1618 enum tcp_seq_states state;
1619 struct sock *syn_wait_sk;
1620 int bucket, offset, sbucket, num;
1625 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1626 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1628 extern struct request_sock_ops tcp_request_sock_ops;
1629 extern struct request_sock_ops tcp6_request_sock_ops;
1631 void tcp_v4_destroy_sock(struct sock *sk);
1633 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1634 netdev_features_t features);
1635 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1636 int tcp_gro_complete(struct sk_buff *skb);
1638 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1640 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1642 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1645 static inline bool tcp_stream_memory_free(const struct sock *sk)
1647 const struct tcp_sock *tp = tcp_sk(sk);
1648 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1650 return notsent_bytes < tcp_notsent_lowat(tp);
1653 #ifdef CONFIG_PROC_FS
1654 int tcp4_proc_init(void);
1655 void tcp4_proc_exit(void);
1658 int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
1659 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1660 const struct tcp_request_sock_ops *af_ops,
1661 struct sock *sk, struct sk_buff *skb);
1663 /* TCP af-specific functions */
1664 struct tcp_sock_af_ops {
1665 #ifdef CONFIG_TCP_MD5SIG
1666 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1667 struct sock *addr_sk);
1668 int (*calc_md5_hash) (char *location,
1669 struct tcp_md5sig_key *md5,
1670 const struct sock *sk,
1671 const struct request_sock *req,
1672 const struct sk_buff *skb);
1673 int (*md5_parse) (struct sock *sk,
1674 char __user *optval,
1679 struct tcp_request_sock_ops {
1681 #ifdef CONFIG_TCP_MD5SIG
1682 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1683 struct request_sock *req);
1684 int (*calc_md5_hash) (char *location,
1685 struct tcp_md5sig_key *md5,
1686 const struct sock *sk,
1687 const struct request_sock *req,
1688 const struct sk_buff *skb);
1690 void (*init_req)(struct request_sock *req, struct sock *sk,
1691 struct sk_buff *skb);
1692 #ifdef CONFIG_SYN_COOKIES
1693 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1696 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1697 const struct request_sock *req,
1699 __u32 (*init_seq)(const struct sk_buff *skb);
1700 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1701 struct flowi *fl, struct request_sock *req,
1702 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
1703 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1704 const unsigned long timeout);
1707 #ifdef CONFIG_SYN_COOKIES
1708 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1709 struct sock *sk, struct sk_buff *skb,
1712 return ops->cookie_init_seq(sk, skb, mss);
1715 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1716 struct sock *sk, struct sk_buff *skb,
1723 int tcpv4_offload_init(void);
1725 void tcp_v4_init(void);
1726 void tcp_init(void);
1729 * Save and compile IPv4 options, return a pointer to it
1731 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1733 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1734 struct ip_options_rcu *dopt = NULL;
1737 int opt_size = sizeof(*dopt) + opt->optlen;
1739 dopt = kmalloc(opt_size, GFP_ATOMIC);
1740 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1748 /* locally generated TCP pure ACKs have skb->truesize == 2
1749 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1750 * This is much faster than dissecting the packet to find out.
1751 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1753 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1755 return skb->truesize == 2;
1758 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)