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_workaround_signed_windows;
266 extern int sysctl_tcp_slow_start_after_idle;
267 extern int sysctl_tcp_thin_linear_timeouts;
268 extern int sysctl_tcp_thin_dupack;
269 extern int sysctl_tcp_early_retrans;
270 extern int sysctl_tcp_limit_output_bytes;
271 extern int sysctl_tcp_challenge_ack_limit;
272 extern unsigned int sysctl_tcp_notsent_lowat;
273 extern int sysctl_tcp_min_tso_segs;
274 extern int sysctl_tcp_autocorking;
275 extern int sysctl_tcp_invalid_ratelimit;
277 extern atomic_long_t tcp_memory_allocated;
278 extern struct percpu_counter tcp_sockets_allocated;
279 extern int tcp_memory_pressure;
282 * The next routines deal with comparing 32 bit unsigned ints
283 * and worry about wraparound (automatic with unsigned arithmetic).
286 static inline bool before(__u32 seq1, __u32 seq2)
288 return (__s32)(seq1-seq2) < 0;
290 #define after(seq2, seq1) before(seq1, seq2)
292 /* is s2<=s1<=s3 ? */
293 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
295 return seq3 - seq2 >= seq1 - seq2;
298 static inline bool tcp_out_of_memory(struct sock *sk)
300 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
301 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
306 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
308 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
309 int orphans = percpu_counter_read_positive(ocp);
311 if (orphans << shift > sysctl_tcp_max_orphans) {
312 orphans = percpu_counter_sum_positive(ocp);
313 if (orphans << shift > sysctl_tcp_max_orphans)
319 bool tcp_check_oom(struct sock *sk, int shift);
321 /* syncookies: remember time of last synqueue overflow */
322 static inline void tcp_synq_overflow(struct sock *sk)
324 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
327 /* syncookies: no recent synqueue overflow on this listening socket? */
328 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
330 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
331 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
334 extern struct proto tcp_prot;
336 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
337 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
338 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
339 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
340 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
342 void tcp_tasklet_init(void);
344 void tcp_v4_err(struct sk_buff *skb, u32);
346 void tcp_shutdown(struct sock *sk, int how);
348 void tcp_v4_early_demux(struct sk_buff *skb);
349 int tcp_v4_rcv(struct sk_buff *skb);
351 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
352 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
353 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
355 void tcp_release_cb(struct sock *sk);
356 void tcp_wfree(struct sk_buff *skb);
357 void tcp_write_timer_handler(struct sock *sk);
358 void tcp_delack_timer_handler(struct sock *sk);
359 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
360 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
361 const struct tcphdr *th, unsigned int len);
362 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
363 const struct tcphdr *th, unsigned int len);
364 void tcp_rcv_space_adjust(struct sock *sk);
365 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
366 void tcp_twsk_destructor(struct sock *sk);
367 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
368 struct pipe_inode_info *pipe, size_t len,
371 static inline void tcp_dec_quickack_mode(struct sock *sk,
372 const unsigned int pkts)
374 struct inet_connection_sock *icsk = inet_csk(sk);
376 if (icsk->icsk_ack.quick) {
377 if (pkts >= icsk->icsk_ack.quick) {
378 icsk->icsk_ack.quick = 0;
379 /* Leaving quickack mode we deflate ATO. */
380 icsk->icsk_ack.ato = TCP_ATO_MIN;
382 icsk->icsk_ack.quick -= pkts;
387 #define TCP_ECN_QUEUE_CWR 2
388 #define TCP_ECN_DEMAND_CWR 4
389 #define TCP_ECN_SEEN 8
399 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
401 const struct tcphdr *th);
402 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
403 struct request_sock *req, struct request_sock **prev,
405 int tcp_child_process(struct sock *parent, struct sock *child,
406 struct sk_buff *skb);
407 void tcp_enter_loss(struct sock *sk);
408 void tcp_clear_retrans(struct tcp_sock *tp);
409 void tcp_update_metrics(struct sock *sk);
410 void tcp_init_metrics(struct sock *sk);
411 void tcp_metrics_init(void);
412 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
413 bool paws_check, bool timestamps);
414 bool tcp_remember_stamp(struct sock *sk);
415 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
416 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
417 void tcp_disable_fack(struct tcp_sock *tp);
418 void tcp_close(struct sock *sk, long timeout);
419 void tcp_init_sock(struct sock *sk);
420 unsigned int tcp_poll(struct file *file, struct socket *sock,
421 struct poll_table_struct *wait);
422 int tcp_getsockopt(struct sock *sk, int level, int optname,
423 char __user *optval, int __user *optlen);
424 int tcp_setsockopt(struct sock *sk, int level, int optname,
425 char __user *optval, unsigned int optlen);
426 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
427 char __user *optval, int __user *optlen);
428 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
429 char __user *optval, unsigned int optlen);
430 void tcp_set_keepalive(struct sock *sk, int val);
431 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
432 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
433 int flags, int *addr_len);
434 void tcp_parse_options(const struct sk_buff *skb,
435 struct tcp_options_received *opt_rx,
436 int estab, struct tcp_fastopen_cookie *foc);
437 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
440 * TCP v4 functions exported for the inet6 API
443 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
444 void tcp_v4_mtu_reduced(struct sock *sk);
445 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
446 struct sock *tcp_create_openreq_child(struct sock *sk,
447 struct request_sock *req,
448 struct sk_buff *skb);
449 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
450 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
451 struct request_sock *req,
452 struct dst_entry *dst);
453 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
454 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
455 int tcp_connect(struct sock *sk);
456 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
457 struct request_sock *req,
458 struct tcp_fastopen_cookie *foc);
459 int tcp_disconnect(struct sock *sk, int flags);
461 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
462 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
463 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
465 /* From syncookies.c */
466 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
468 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
469 #ifdef CONFIG_SYN_COOKIES
471 /* Syncookies use a monotonic timer which increments every 60 seconds.
472 * This counter is used both as a hash input and partially encoded into
473 * the cookie value. A cookie is only validated further if the delta
474 * between the current counter value and the encoded one is less than this,
475 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
476 * the counter advances immediately after a cookie is generated).
478 #define MAX_SYNCOOKIE_AGE 2
480 static inline u32 tcp_cookie_time(void)
482 u64 val = get_jiffies_64();
484 do_div(val, 60 * HZ);
488 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
490 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
492 __u32 cookie_init_timestamp(struct request_sock *req);
493 bool cookie_timestamp_decode(struct tcp_options_received *opt);
494 bool cookie_ecn_ok(const struct tcp_options_received *opt,
495 const struct net *net, const struct dst_entry *dst);
497 /* From net/ipv6/syncookies.c */
498 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
500 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
502 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
503 const struct tcphdr *th, u16 *mssp);
504 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
509 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
511 bool tcp_may_send_now(struct sock *sk);
512 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
513 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
514 void tcp_retransmit_timer(struct sock *sk);
515 void tcp_xmit_retransmit_queue(struct sock *);
516 void tcp_simple_retransmit(struct sock *);
517 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
518 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
520 void tcp_send_probe0(struct sock *);
521 void tcp_send_partial(struct sock *);
522 int tcp_write_wakeup(struct sock *);
523 void tcp_send_fin(struct sock *sk);
524 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
525 int tcp_send_synack(struct sock *);
526 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
528 void tcp_push_one(struct sock *, unsigned int mss_now);
529 void tcp_send_ack(struct sock *sk);
530 void tcp_send_delayed_ack(struct sock *sk);
531 void tcp_send_loss_probe(struct sock *sk);
532 bool tcp_schedule_loss_probe(struct sock *sk);
535 void tcp_resume_early_retransmit(struct sock *sk);
536 void tcp_rearm_rto(struct sock *sk);
537 void tcp_reset(struct sock *sk);
540 void tcp_init_xmit_timers(struct sock *);
541 static inline void tcp_clear_xmit_timers(struct sock *sk)
543 inet_csk_clear_xmit_timers(sk);
546 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
547 unsigned int tcp_current_mss(struct sock *sk);
549 /* Bound MSS / TSO packet size with the half of the window */
550 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
554 /* When peer uses tiny windows, there is no use in packetizing
555 * to sub-MSS pieces for the sake of SWS or making sure there
556 * are enough packets in the pipe for fast recovery.
558 * On the other hand, for extremely large MSS devices, handling
559 * smaller than MSS windows in this way does make sense.
561 if (tp->max_window >= 512)
562 cutoff = (tp->max_window >> 1);
564 cutoff = tp->max_window;
566 if (cutoff && pktsize > cutoff)
567 return max_t(int, cutoff, 68U - tp->tcp_header_len);
573 void tcp_get_info(const struct sock *, struct tcp_info *);
575 /* Read 'sendfile()'-style from a TCP socket */
576 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
577 unsigned int, size_t);
578 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
579 sk_read_actor_t recv_actor);
581 void tcp_initialize_rcv_mss(struct sock *sk);
583 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
584 int tcp_mss_to_mtu(struct sock *sk, int mss);
585 void tcp_mtup_init(struct sock *sk);
586 void tcp_init_buffer_space(struct sock *sk);
588 static inline void tcp_bound_rto(const struct sock *sk)
590 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
591 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
594 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
596 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
599 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
601 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
602 ntohl(TCP_FLAG_ACK) |
606 static inline void tcp_fast_path_on(struct tcp_sock *tp)
608 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
611 static inline void tcp_fast_path_check(struct sock *sk)
613 struct tcp_sock *tp = tcp_sk(sk);
615 if (skb_queue_empty(&tp->out_of_order_queue) &&
617 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
619 tcp_fast_path_on(tp);
622 /* Compute the actual rto_min value */
623 static inline u32 tcp_rto_min(struct sock *sk)
625 const struct dst_entry *dst = __sk_dst_get(sk);
626 u32 rto_min = TCP_RTO_MIN;
628 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
629 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
633 static inline u32 tcp_rto_min_us(struct sock *sk)
635 return jiffies_to_usecs(tcp_rto_min(sk));
638 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
640 return dst_metric_locked(dst, RTAX_CC_ALGO);
643 /* Compute the actual receive window we are currently advertising.
644 * Rcv_nxt can be after the window if our peer push more data
645 * than the offered window.
647 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
649 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
656 /* Choose a new window, without checks for shrinking, and without
657 * scaling applied to the result. The caller does these things
658 * if necessary. This is a "raw" window selection.
660 u32 __tcp_select_window(struct sock *sk);
662 void tcp_send_window_probe(struct sock *sk);
664 /* TCP timestamps are only 32-bits, this causes a slight
665 * complication on 64-bit systems since we store a snapshot
666 * of jiffies in the buffer control blocks below. We decided
667 * to use only the low 32-bits of jiffies and hide the ugly
668 * casts with the following macro.
670 #define tcp_time_stamp ((__u32)(jiffies))
672 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
674 return skb->skb_mstamp.stamp_jiffies;
678 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
680 #define TCPHDR_FIN 0x01
681 #define TCPHDR_SYN 0x02
682 #define TCPHDR_RST 0x04
683 #define TCPHDR_PSH 0x08
684 #define TCPHDR_ACK 0x10
685 #define TCPHDR_URG 0x20
686 #define TCPHDR_ECE 0x40
687 #define TCPHDR_CWR 0x80
689 /* This is what the send packet queuing engine uses to pass
690 * TCP per-packet control information to the transmission code.
691 * We also store the host-order sequence numbers in here too.
692 * This is 44 bytes if IPV6 is enabled.
693 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
696 __u32 seq; /* Starting sequence number */
697 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
699 /* Note : tcp_tw_isn is used in input path only
700 * (isn chosen by tcp_timewait_state_process())
702 * tcp_gso_segs is used in write queue only,
703 * cf tcp_skb_pcount()
708 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
710 __u8 sacked; /* State flags for SACK/FACK. */
711 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
712 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
713 #define TCPCB_LOST 0x04 /* SKB is lost */
714 #define TCPCB_TAGBITS 0x07 /* All tag bits */
715 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
716 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
717 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
720 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
722 __u32 ack_seq; /* Sequence number ACK'd */
724 struct inet_skb_parm h4;
725 #if IS_ENABLED(CONFIG_IPV6)
726 struct inet6_skb_parm h6;
728 } header; /* For incoming frames */
731 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
734 #if IS_ENABLED(CONFIG_IPV6)
735 /* This is the variant of inet6_iif() that must be used by TCP,
736 * as TCP moves IP6CB into a different location in skb->cb[]
738 static inline int tcp_v6_iif(const struct sk_buff *skb)
740 return TCP_SKB_CB(skb)->header.h6.iif;
744 /* Due to TSO, an SKB can be composed of multiple actual
745 * packets. To keep these tracked properly, we use this.
747 static inline int tcp_skb_pcount(const struct sk_buff *skb)
749 return TCP_SKB_CB(skb)->tcp_gso_segs;
752 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
754 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
757 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
759 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
762 /* This is valid iff tcp_skb_pcount() > 1. */
763 static inline int tcp_skb_mss(const struct sk_buff *skb)
765 return skb_shinfo(skb)->gso_size;
768 /* Events passed to congestion control interface */
770 CA_EVENT_TX_START, /* first transmit when no packets in flight */
771 CA_EVENT_CWND_RESTART, /* congestion window restart */
772 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
773 CA_EVENT_LOSS, /* loss timeout */
774 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
775 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
776 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
777 CA_EVENT_NON_DELAYED_ACK,
780 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
781 enum tcp_ca_ack_event_flags {
782 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
783 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
784 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
788 * Interface for adding new TCP congestion control handlers
790 #define TCP_CA_NAME_MAX 16
791 #define TCP_CA_MAX 128
792 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
794 #define TCP_CA_UNSPEC 0
796 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
797 #define TCP_CONG_NON_RESTRICTED 0x1
798 /* Requires ECN/ECT set on all packets */
799 #define TCP_CONG_NEEDS_ECN 0x2
801 struct tcp_congestion_ops {
802 struct list_head list;
806 /* initialize private data (optional) */
807 void (*init)(struct sock *sk);
808 /* cleanup private data (optional) */
809 void (*release)(struct sock *sk);
811 /* return slow start threshold (required) */
812 u32 (*ssthresh)(struct sock *sk);
813 /* do new cwnd calculation (required) */
814 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
815 /* call before changing ca_state (optional) */
816 void (*set_state)(struct sock *sk, u8 new_state);
817 /* call when cwnd event occurs (optional) */
818 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
819 /* call when ack arrives (optional) */
820 void (*in_ack_event)(struct sock *sk, u32 flags);
821 /* new value of cwnd after loss (optional) */
822 u32 (*undo_cwnd)(struct sock *sk);
823 /* hook for packet ack accounting (optional) */
824 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
825 /* get info for inet_diag (optional) */
826 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
828 char name[TCP_CA_NAME_MAX];
829 struct module *owner;
832 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
833 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
835 void tcp_assign_congestion_control(struct sock *sk);
836 void tcp_init_congestion_control(struct sock *sk);
837 void tcp_cleanup_congestion_control(struct sock *sk);
838 int tcp_set_default_congestion_control(const char *name);
839 void tcp_get_default_congestion_control(char *name);
840 void tcp_get_available_congestion_control(char *buf, size_t len);
841 void tcp_get_allowed_congestion_control(char *buf, size_t len);
842 int tcp_set_allowed_congestion_control(char *allowed);
843 int tcp_set_congestion_control(struct sock *sk, const char *name);
844 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
845 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
847 u32 tcp_reno_ssthresh(struct sock *sk);
848 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
849 extern struct tcp_congestion_ops tcp_reno;
851 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
852 u32 tcp_ca_get_key_by_name(const char *name);
854 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
856 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
862 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
864 const struct inet_connection_sock *icsk = inet_csk(sk);
866 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
869 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
871 struct inet_connection_sock *icsk = inet_csk(sk);
873 if (icsk->icsk_ca_ops->set_state)
874 icsk->icsk_ca_ops->set_state(sk, ca_state);
875 icsk->icsk_ca_state = ca_state;
878 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
880 const struct inet_connection_sock *icsk = inet_csk(sk);
882 if (icsk->icsk_ca_ops->cwnd_event)
883 icsk->icsk_ca_ops->cwnd_event(sk, event);
886 /* These functions determine how the current flow behaves in respect of SACK
887 * handling. SACK is negotiated with the peer, and therefore it can vary
888 * between different flows.
890 * tcp_is_sack - SACK enabled
891 * tcp_is_reno - No SACK
892 * tcp_is_fack - FACK enabled, implies SACK enabled
894 static inline int tcp_is_sack(const struct tcp_sock *tp)
896 return tp->rx_opt.sack_ok;
899 static inline bool tcp_is_reno(const struct tcp_sock *tp)
901 return !tcp_is_sack(tp);
904 static inline bool tcp_is_fack(const struct tcp_sock *tp)
906 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
909 static inline void tcp_enable_fack(struct tcp_sock *tp)
911 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
914 /* TCP early-retransmit (ER) is similar to but more conservative than
915 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
917 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
919 tp->do_early_retrans = sysctl_tcp_early_retrans &&
920 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
921 sysctl_tcp_reordering == 3;
924 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
926 tp->do_early_retrans = 0;
929 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
931 return tp->sacked_out + tp->lost_out;
934 /* This determines how many packets are "in the network" to the best
935 * of our knowledge. In many cases it is conservative, but where
936 * detailed information is available from the receiver (via SACK
937 * blocks etc.) we can make more aggressive calculations.
939 * Use this for decisions involving congestion control, use just
940 * tp->packets_out to determine if the send queue is empty or not.
942 * Read this equation as:
944 * "Packets sent once on transmission queue" MINUS
945 * "Packets left network, but not honestly ACKed yet" PLUS
946 * "Packets fast retransmitted"
948 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
950 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
953 #define TCP_INFINITE_SSTHRESH 0x7fffffff
955 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
957 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
960 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
962 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
963 (1 << inet_csk(sk)->icsk_ca_state);
966 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
967 * The exception is cwnd reduction phase, when cwnd is decreasing towards
970 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
972 const struct tcp_sock *tp = tcp_sk(sk);
974 if (tcp_in_cwnd_reduction(sk))
975 return tp->snd_ssthresh;
977 return max(tp->snd_ssthresh,
978 ((tp->snd_cwnd >> 1) +
979 (tp->snd_cwnd >> 2)));
982 /* Use define here intentionally to get WARN_ON location shown at the caller */
983 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
985 void tcp_enter_cwr(struct sock *sk);
986 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
988 /* The maximum number of MSS of available cwnd for which TSO defers
989 * sending if not using sysctl_tcp_tso_win_divisor.
991 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
996 /* Slow start with delack produces 3 packets of burst, so that
997 * it is safe "de facto". This will be the default - same as
998 * the default reordering threshold - but if reordering increases,
999 * we must be able to allow cwnd to burst at least this much in order
1000 * to not pull it back when holes are filled.
1002 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1004 return tp->reordering;
1007 /* Returns end sequence number of the receiver's advertised window */
1008 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1010 return tp->snd_una + tp->snd_wnd;
1013 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1014 * flexible approach. The RFC suggests cwnd should not be raised unless
1015 * it was fully used previously. And that's exactly what we do in
1016 * congestion avoidance mode. But in slow start we allow cwnd to grow
1017 * as long as the application has used half the cwnd.
1019 * cwnd is 10 (IW10), but application sends 9 frames.
1020 * We allow cwnd to reach 18 when all frames are ACKed.
1021 * This check is safe because it's as aggressive as slow start which already
1022 * risks 100% overshoot. The advantage is that we discourage application to
1023 * either send more filler packets or data to artificially blow up the cwnd
1024 * usage, and allow application-limited process to probe bw more aggressively.
1026 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1028 const struct tcp_sock *tp = tcp_sk(sk);
1030 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1031 if (tp->snd_cwnd <= tp->snd_ssthresh)
1032 return tp->snd_cwnd < 2 * tp->max_packets_out;
1034 return tp->is_cwnd_limited;
1037 static inline void tcp_check_probe_timer(struct sock *sk)
1039 const struct tcp_sock *tp = tcp_sk(sk);
1040 const struct inet_connection_sock *icsk = inet_csk(sk);
1042 if (!tp->packets_out && !icsk->icsk_pending)
1043 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1044 icsk->icsk_rto, TCP_RTO_MAX);
1047 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1052 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1058 * Calculate(/check) TCP checksum
1060 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1061 __be32 daddr, __wsum base)
1063 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1066 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1068 return __skb_checksum_complete(skb);
1071 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1073 return !skb_csum_unnecessary(skb) &&
1074 __tcp_checksum_complete(skb);
1077 /* Prequeue for VJ style copy to user, combined with checksumming. */
1079 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1081 tp->ucopy.task = NULL;
1083 tp->ucopy.memory = 0;
1084 skb_queue_head_init(&tp->ucopy.prequeue);
1087 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1092 static const char *statename[]={
1093 "Unused","Established","Syn Sent","Syn Recv",
1094 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1095 "Close Wait","Last ACK","Listen","Closing"
1098 void tcp_set_state(struct sock *sk, int state);
1100 void tcp_done(struct sock *sk);
1102 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1105 rx_opt->num_sacks = 0;
1108 u32 tcp_default_init_rwnd(u32 mss);
1110 /* Determine a window scaling and initial window to offer. */
1111 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1112 __u32 *window_clamp, int wscale_ok,
1113 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1115 static inline int tcp_win_from_space(int space)
1117 return sysctl_tcp_adv_win_scale<=0 ?
1118 (space>>(-sysctl_tcp_adv_win_scale)) :
1119 space - (space>>sysctl_tcp_adv_win_scale);
1122 /* Note: caller must be prepared to deal with negative returns */
1123 static inline int tcp_space(const struct sock *sk)
1125 return tcp_win_from_space(sk->sk_rcvbuf -
1126 atomic_read(&sk->sk_rmem_alloc));
1129 static inline int tcp_full_space(const struct sock *sk)
1131 return tcp_win_from_space(sk->sk_rcvbuf);
1134 static inline void tcp_openreq_init(struct request_sock *req,
1135 struct tcp_options_received *rx_opt,
1136 struct sk_buff *skb, struct sock *sk)
1138 struct inet_request_sock *ireq = inet_rsk(req);
1140 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1142 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1143 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1144 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1145 tcp_rsk(req)->last_oow_ack_time = 0;
1146 req->mss = rx_opt->mss_clamp;
1147 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1148 ireq->tstamp_ok = rx_opt->tstamp_ok;
1149 ireq->sack_ok = rx_opt->sack_ok;
1150 ireq->snd_wscale = rx_opt->snd_wscale;
1151 ireq->wscale_ok = rx_opt->wscale_ok;
1154 ireq->ir_rmt_port = tcp_hdr(skb)->source;
1155 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
1156 ireq->ir_mark = inet_request_mark(sk, skb);
1159 extern void tcp_openreq_init_rwin(struct request_sock *req,
1160 struct sock *sk, struct dst_entry *dst);
1162 void tcp_enter_memory_pressure(struct sock *sk);
1164 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1166 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1169 static inline int keepalive_time_when(const struct tcp_sock *tp)
1171 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1174 static inline int keepalive_probes(const struct tcp_sock *tp)
1176 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1179 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1181 const struct inet_connection_sock *icsk = &tp->inet_conn;
1183 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1184 tcp_time_stamp - tp->rcv_tstamp);
1187 static inline int tcp_fin_time(const struct sock *sk)
1189 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1190 const int rto = inet_csk(sk)->icsk_rto;
1192 if (fin_timeout < (rto << 2) - (rto >> 1))
1193 fin_timeout = (rto << 2) - (rto >> 1);
1198 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1201 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1203 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1206 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1207 * then following tcp messages have valid values. Ignore 0 value,
1208 * or else 'negative' tsval might forbid us to accept their packets.
1210 if (!rx_opt->ts_recent)
1215 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1218 if (tcp_paws_check(rx_opt, 0))
1221 /* RST segments are not recommended to carry timestamp,
1222 and, if they do, it is recommended to ignore PAWS because
1223 "their cleanup function should take precedence over timestamps."
1224 Certainly, it is mistake. It is necessary to understand the reasons
1225 of this constraint to relax it: if peer reboots, clock may go
1226 out-of-sync and half-open connections will not be reset.
1227 Actually, the problem would be not existing if all
1228 the implementations followed draft about maintaining clock
1229 via reboots. Linux-2.2 DOES NOT!
1231 However, we can relax time bounds for RST segments to MSL.
1233 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1238 /* Return true if we're currently rate-limiting out-of-window ACKs and
1239 * thus shouldn't send a dupack right now. We rate-limit dupacks in
1240 * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
1241 * attacks that send repeated SYNs or ACKs for the same connection. To
1242 * do this, we do not send a duplicate SYNACK or ACK if the remote
1243 * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
1245 static inline bool tcp_oow_rate_limited(struct net *net,
1246 const struct sk_buff *skb,
1247 int mib_idx, u32 *last_oow_ack_time)
1249 /* Data packets without SYNs are not likely part of an ACK loop. */
1250 if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) &&
1252 goto not_rate_limited;
1254 if (*last_oow_ack_time) {
1255 s32 elapsed = (s32)(tcp_time_stamp - *last_oow_ack_time);
1257 if (0 <= elapsed && elapsed < sysctl_tcp_invalid_ratelimit) {
1258 NET_INC_STATS_BH(net, mib_idx);
1259 return true; /* rate-limited: don't send yet! */
1263 *last_oow_ack_time = tcp_time_stamp;
1266 return false; /* not rate-limited: go ahead, send dupack now! */
1269 static inline void tcp_mib_init(struct net *net)
1272 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1273 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1274 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1275 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1279 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1281 tp->lost_skb_hint = NULL;
1284 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1286 tcp_clear_retrans_hints_partial(tp);
1287 tp->retransmit_skb_hint = NULL;
1293 union tcp_md5_addr {
1295 #if IS_ENABLED(CONFIG_IPV6)
1300 /* - key database */
1301 struct tcp_md5sig_key {
1302 struct hlist_node node;
1304 u8 family; /* AF_INET or AF_INET6 */
1305 union tcp_md5_addr addr;
1306 u8 key[TCP_MD5SIG_MAXKEYLEN];
1307 struct rcu_head rcu;
1311 struct tcp_md5sig_info {
1312 struct hlist_head head;
1313 struct rcu_head rcu;
1316 /* - pseudo header */
1317 struct tcp4_pseudohdr {
1325 struct tcp6_pseudohdr {
1326 struct in6_addr saddr;
1327 struct in6_addr daddr;
1329 __be32 protocol; /* including padding */
1332 union tcp_md5sum_block {
1333 struct tcp4_pseudohdr ip4;
1334 #if IS_ENABLED(CONFIG_IPV6)
1335 struct tcp6_pseudohdr ip6;
1339 /* - pool: digest algorithm, hash description and scratch buffer */
1340 struct tcp_md5sig_pool {
1341 struct hash_desc md5_desc;
1342 union tcp_md5sum_block md5_blk;
1346 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1347 const struct sock *sk, const struct request_sock *req,
1348 const struct sk_buff *skb);
1349 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1350 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1351 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1353 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1354 struct sock *addr_sk);
1356 #ifdef CONFIG_TCP_MD5SIG
1357 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1358 const union tcp_md5_addr *addr,
1360 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1362 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1363 const union tcp_md5_addr *addr,
1368 #define tcp_twsk_md5_key(twsk) NULL
1371 bool tcp_alloc_md5sig_pool(void);
1373 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1374 static inline void tcp_put_md5sig_pool(void)
1379 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1380 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1381 unsigned int header_len);
1382 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1383 const struct tcp_md5sig_key *key);
1385 /* From tcp_fastopen.c */
1386 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1387 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1388 unsigned long *last_syn_loss);
1389 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1390 struct tcp_fastopen_cookie *cookie, bool syn_lost);
1391 struct tcp_fastopen_request {
1392 /* Fast Open cookie. Size 0 means a cookie request */
1393 struct tcp_fastopen_cookie cookie;
1394 struct msghdr *data; /* data in MSG_FASTOPEN */
1396 int copied; /* queued in tcp_connect() */
1398 void tcp_free_fastopen_req(struct tcp_sock *tp);
1400 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1401 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1402 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1403 struct request_sock *req,
1404 struct tcp_fastopen_cookie *foc,
1405 struct dst_entry *dst);
1406 void tcp_fastopen_init_key_once(bool publish);
1407 #define TCP_FASTOPEN_KEY_LENGTH 16
1409 /* Fastopen key context */
1410 struct tcp_fastopen_context {
1411 struct crypto_cipher *tfm;
1412 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1413 struct rcu_head rcu;
1416 /* write queue abstraction */
1417 static inline void tcp_write_queue_purge(struct sock *sk)
1419 struct sk_buff *skb;
1421 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1422 sk_wmem_free_skb(sk, skb);
1424 tcp_clear_all_retrans_hints(tcp_sk(sk));
1427 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1429 return skb_peek(&sk->sk_write_queue);
1432 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1434 return skb_peek_tail(&sk->sk_write_queue);
1437 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1438 const struct sk_buff *skb)
1440 return skb_queue_next(&sk->sk_write_queue, skb);
1443 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1444 const struct sk_buff *skb)
1446 return skb_queue_prev(&sk->sk_write_queue, skb);
1449 #define tcp_for_write_queue(skb, sk) \
1450 skb_queue_walk(&(sk)->sk_write_queue, skb)
1452 #define tcp_for_write_queue_from(skb, sk) \
1453 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1455 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1456 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1458 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1460 return sk->sk_send_head;
1463 static inline bool tcp_skb_is_last(const struct sock *sk,
1464 const struct sk_buff *skb)
1466 return skb_queue_is_last(&sk->sk_write_queue, skb);
1469 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1471 if (tcp_skb_is_last(sk, skb))
1472 sk->sk_send_head = NULL;
1474 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1477 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1479 if (sk->sk_send_head == skb_unlinked)
1480 sk->sk_send_head = NULL;
1483 static inline void tcp_init_send_head(struct sock *sk)
1485 sk->sk_send_head = NULL;
1488 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1490 __skb_queue_tail(&sk->sk_write_queue, skb);
1493 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1495 __tcp_add_write_queue_tail(sk, skb);
1497 /* Queue it, remembering where we must start sending. */
1498 if (sk->sk_send_head == NULL) {
1499 sk->sk_send_head = skb;
1501 if (tcp_sk(sk)->highest_sack == NULL)
1502 tcp_sk(sk)->highest_sack = skb;
1506 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1508 __skb_queue_head(&sk->sk_write_queue, skb);
1511 /* Insert buff after skb on the write queue of sk. */
1512 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1513 struct sk_buff *buff,
1516 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1519 /* Insert new before skb on the write queue of sk. */
1520 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1521 struct sk_buff *skb,
1524 __skb_queue_before(&sk->sk_write_queue, skb, new);
1526 if (sk->sk_send_head == skb)
1527 sk->sk_send_head = new;
1530 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1532 __skb_unlink(skb, &sk->sk_write_queue);
1535 static inline bool tcp_write_queue_empty(struct sock *sk)
1537 return skb_queue_empty(&sk->sk_write_queue);
1540 static inline void tcp_push_pending_frames(struct sock *sk)
1542 if (tcp_send_head(sk)) {
1543 struct tcp_sock *tp = tcp_sk(sk);
1545 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1549 /* Start sequence of the skb just after the highest skb with SACKed
1550 * bit, valid only if sacked_out > 0 or when the caller has ensured
1551 * validity by itself.
1553 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1555 if (!tp->sacked_out)
1558 if (tp->highest_sack == NULL)
1561 return TCP_SKB_CB(tp->highest_sack)->seq;
1564 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1566 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1567 tcp_write_queue_next(sk, skb);
1570 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1572 return tcp_sk(sk)->highest_sack;
1575 static inline void tcp_highest_sack_reset(struct sock *sk)
1577 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1580 /* Called when old skb is about to be deleted (to be combined with new skb) */
1581 static inline void tcp_highest_sack_combine(struct sock *sk,
1582 struct sk_buff *old,
1583 struct sk_buff *new)
1585 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1586 tcp_sk(sk)->highest_sack = new;
1589 /* Determines whether this is a thin stream (which may suffer from
1590 * increased latency). Used to trigger latency-reducing mechanisms.
1592 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1594 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1598 enum tcp_seq_states {
1599 TCP_SEQ_STATE_LISTENING,
1600 TCP_SEQ_STATE_OPENREQ,
1601 TCP_SEQ_STATE_ESTABLISHED,
1604 int tcp_seq_open(struct inode *inode, struct file *file);
1606 struct tcp_seq_afinfo {
1609 const struct file_operations *seq_fops;
1610 struct seq_operations seq_ops;
1613 struct tcp_iter_state {
1614 struct seq_net_private p;
1616 enum tcp_seq_states state;
1617 struct sock *syn_wait_sk;
1618 int bucket, offset, sbucket, num;
1623 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1624 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1626 extern struct request_sock_ops tcp_request_sock_ops;
1627 extern struct request_sock_ops tcp6_request_sock_ops;
1629 void tcp_v4_destroy_sock(struct sock *sk);
1631 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1632 netdev_features_t features);
1633 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1634 int tcp_gro_complete(struct sk_buff *skb);
1636 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1638 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1640 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1643 static inline bool tcp_stream_memory_free(const struct sock *sk)
1645 const struct tcp_sock *tp = tcp_sk(sk);
1646 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1648 return notsent_bytes < tcp_notsent_lowat(tp);
1651 #ifdef CONFIG_PROC_FS
1652 int tcp4_proc_init(void);
1653 void tcp4_proc_exit(void);
1656 int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
1657 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1658 const struct tcp_request_sock_ops *af_ops,
1659 struct sock *sk, struct sk_buff *skb);
1661 /* TCP af-specific functions */
1662 struct tcp_sock_af_ops {
1663 #ifdef CONFIG_TCP_MD5SIG
1664 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1665 struct sock *addr_sk);
1666 int (*calc_md5_hash) (char *location,
1667 struct tcp_md5sig_key *md5,
1668 const struct sock *sk,
1669 const struct request_sock *req,
1670 const struct sk_buff *skb);
1671 int (*md5_parse) (struct sock *sk,
1672 char __user *optval,
1677 struct tcp_request_sock_ops {
1679 #ifdef CONFIG_TCP_MD5SIG
1680 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1681 struct request_sock *req);
1682 int (*calc_md5_hash) (char *location,
1683 struct tcp_md5sig_key *md5,
1684 const struct sock *sk,
1685 const struct request_sock *req,
1686 const struct sk_buff *skb);
1688 void (*init_req)(struct request_sock *req, struct sock *sk,
1689 struct sk_buff *skb);
1690 #ifdef CONFIG_SYN_COOKIES
1691 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1694 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1695 const struct request_sock *req,
1697 __u32 (*init_seq)(const struct sk_buff *skb);
1698 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1699 struct flowi *fl, struct request_sock *req,
1700 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
1701 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1702 const unsigned long timeout);
1705 #ifdef CONFIG_SYN_COOKIES
1706 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1707 struct sock *sk, struct sk_buff *skb,
1710 return ops->cookie_init_seq(sk, skb, mss);
1713 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1714 struct sock *sk, struct sk_buff *skb,
1721 int tcpv4_offload_init(void);
1723 void tcp_v4_init(void);
1724 void tcp_init(void);
1727 * Save and compile IPv4 options, return a pointer to it
1729 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1731 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1732 struct ip_options_rcu *dopt = NULL;
1735 int opt_size = sizeof(*dopt) + opt->optlen;
1737 dopt = kmalloc(opt_size, GFP_ATOMIC);
1738 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1746 /* locally generated TCP pure ACKs have skb->truesize == 2
1747 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1748 * This is much faster than dissecting the packet to find out.
1749 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1751 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1753 return skb->truesize == 2;
1756 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)