2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/crypto.h>
31 #include <linux/cryptohash.h>
32 #include <linux/kref.h>
33 #include <linux/ktime.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo
;
52 extern struct percpu_counter tcp_orphan_count
;
53 void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 1024
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL 600
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD 8
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
102 * when active opening a connection.
103 * RFC1122 says the minimum retry MUST
104 * be at least 180secs. Nevertheless
105 * this value is corresponding to
106 * 63secs of retransmission with the
107 * current initial RTO.
110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
111 * when passive opening a connection.
112 * This is corresponding to 31secs of
113 * retransmission with the current
117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 * state, about 60 seconds */
119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
120 /* BSD style FIN_WAIT2 deadlock breaker.
121 * It used to be 3min, new value is 60sec,
122 * to combine FIN-WAIT-2 timeout with
126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
129 #define TCP_ATO_MIN ((unsigned)(HZ/25))
131 #define TCP_DELACK_MIN 4U
132 #define TCP_ATO_MIN 4U
134 #define TCP_RTO_MAX ((unsigned)(120*HZ))
135 #define TCP_RTO_MIN ((unsigned)(HZ/5))
136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
138 * used as a fallback RTO for the
139 * initial data transmission if no
140 * valid RTT sample has been acquired,
141 * most likely due to retrans in 3WHS.
144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
145 * for local resources.
148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
150 #define TCP_KEEPALIVE_INTVL (75*HZ)
152 #define MAX_TCP_KEEPIDLE 32767
153 #define MAX_TCP_KEEPINTVL 32767
154 #define MAX_TCP_KEEPCNT 127
155 #define MAX_TCP_SYNCNT 127
157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
161 * after this time. It should be equal
162 * (or greater than) TCP_TIMEWAIT_LEN
163 * to provide reliability equal to one
164 * provided by timewait state.
166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
167 * timestamps. It must be less than
168 * minimal timewait lifetime.
174 #define TCPOPT_NOP 1 /* Padding */
175 #define TCPOPT_EOL 0 /* End of options */
176 #define TCPOPT_MSS 2 /* Segment size negotiating */
177 #define TCPOPT_WINDOW 3 /* Window scaling */
178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
179 #define TCPOPT_SACK 5 /* SACK Block */
180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
183 #define TCPOPT_EXP 254 /* Experimental */
184 /* Magic number to be after the option value for sharing TCP
185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
187 #define TCPOPT_FASTOPEN_MAGIC 0xF989
193 #define TCPOLEN_MSS 4
194 #define TCPOLEN_WINDOW 3
195 #define TCPOLEN_SACK_PERM 2
196 #define TCPOLEN_TIMESTAMP 10
197 #define TCPOLEN_MD5SIG 18
198 #define TCPOLEN_FASTOPEN_BASE 2
199 #define TCPOLEN_EXP_FASTOPEN_BASE 4
201 /* But this is what stacks really send out. */
202 #define TCPOLEN_TSTAMP_ALIGNED 12
203 #define TCPOLEN_WSCALE_ALIGNED 4
204 #define TCPOLEN_SACKPERM_ALIGNED 4
205 #define TCPOLEN_SACK_BASE 2
206 #define TCPOLEN_SACK_BASE_ALIGNED 4
207 #define TCPOLEN_SACK_PERBLOCK 8
208 #define TCPOLEN_MD5SIG_ALIGNED 20
209 #define TCPOLEN_MSS_ALIGNED 4
211 /* Flags in tp->nonagle */
212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
213 #define TCP_NAGLE_CORK 2 /* Socket is corked */
214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
216 /* TCP thin-stream limits */
217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
219 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
220 #define TCP_INIT_CWND 10
222 /* Bit Flags for sysctl_tcp_fastopen */
223 #define TFO_CLIENT_ENABLE 1
224 #define TFO_SERVER_ENABLE 2
225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
227 /* Accept SYN data w/o any cookie option */
228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
230 /* Force enable TFO on all listeners, i.e., not requiring the
231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
233 #define TFO_SERVER_WO_SOCKOPT1 0x400
234 #define TFO_SERVER_WO_SOCKOPT2 0x800
236 extern struct inet_timewait_death_row tcp_death_row
;
238 /* sysctl variables for tcp */
239 extern int sysctl_tcp_timestamps
;
240 extern int sysctl_tcp_window_scaling
;
241 extern int sysctl_tcp_sack
;
242 extern int sysctl_tcp_fin_timeout
;
243 extern int sysctl_tcp_keepalive_time
;
244 extern int sysctl_tcp_keepalive_probes
;
245 extern int sysctl_tcp_keepalive_intvl
;
246 extern int sysctl_tcp_syn_retries
;
247 extern int sysctl_tcp_synack_retries
;
248 extern int sysctl_tcp_retries1
;
249 extern int sysctl_tcp_retries2
;
250 extern int sysctl_tcp_orphan_retries
;
251 extern int sysctl_tcp_syncookies
;
252 extern int sysctl_tcp_fastopen
;
253 extern int sysctl_tcp_retrans_collapse
;
254 extern int sysctl_tcp_stdurg
;
255 extern int sysctl_tcp_rfc1337
;
256 extern int sysctl_tcp_abort_on_overflow
;
257 extern int sysctl_tcp_max_orphans
;
258 extern int sysctl_tcp_fack
;
259 extern int sysctl_tcp_reordering
;
260 extern int sysctl_tcp_max_reordering
;
261 extern int sysctl_tcp_dsack
;
262 extern long sysctl_tcp_mem
[3];
263 extern int sysctl_tcp_wmem
[3];
264 extern int sysctl_tcp_rmem
[3];
265 extern int sysctl_tcp_app_win
;
266 extern int sysctl_tcp_adv_win_scale
;
267 extern int sysctl_tcp_tw_reuse
;
268 extern int sysctl_tcp_frto
;
269 extern int sysctl_tcp_low_latency
;
270 extern int sysctl_tcp_nometrics_save
;
271 extern int sysctl_tcp_moderate_rcvbuf
;
272 extern int sysctl_tcp_tso_win_divisor
;
273 extern int sysctl_tcp_workaround_signed_windows
;
274 extern int sysctl_tcp_slow_start_after_idle
;
275 extern int sysctl_tcp_thin_linear_timeouts
;
276 extern int sysctl_tcp_thin_dupack
;
277 extern int sysctl_tcp_early_retrans
;
278 extern int sysctl_tcp_limit_output_bytes
;
279 extern int sysctl_tcp_challenge_ack_limit
;
280 extern unsigned int sysctl_tcp_notsent_lowat
;
281 extern int sysctl_tcp_min_tso_segs
;
282 extern int sysctl_tcp_autocorking
;
283 extern int sysctl_tcp_invalid_ratelimit
;
285 extern atomic_long_t tcp_memory_allocated
;
286 extern struct percpu_counter tcp_sockets_allocated
;
287 extern int tcp_memory_pressure
;
290 * The next routines deal with comparing 32 bit unsigned ints
291 * and worry about wraparound (automatic with unsigned arithmetic).
294 static inline bool before(__u32 seq1
, __u32 seq2
)
296 return (__s32
)(seq1
-seq2
) < 0;
298 #define after(seq2, seq1) before(seq1, seq2)
300 /* is s2<=s1<=s3 ? */
301 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
303 return seq3
- seq2
>= seq1
- seq2
;
306 static inline bool tcp_out_of_memory(struct sock
*sk
)
308 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
309 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
314 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
316 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
318 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
319 int orphans
= percpu_counter_read_positive(ocp
);
321 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
322 orphans
= percpu_counter_sum_positive(ocp
);
323 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
329 bool tcp_check_oom(struct sock
*sk
, int shift
);
332 extern struct proto tcp_prot
;
334 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
335 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
336 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
337 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
338 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
340 void tcp_tasklet_init(void);
342 void tcp_v4_err(struct sk_buff
*skb
, u32
);
344 void tcp_shutdown(struct sock
*sk
, int how
);
346 void tcp_v4_early_demux(struct sk_buff
*skb
);
347 int tcp_v4_rcv(struct sk_buff
*skb
);
349 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
350 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
351 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
353 void tcp_release_cb(struct sock
*sk
);
354 void tcp_wfree(struct sk_buff
*skb
);
355 void tcp_write_timer_handler(struct sock
*sk
);
356 void tcp_delack_timer_handler(struct sock
*sk
);
357 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
358 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
359 const struct tcphdr
*th
, unsigned int len
);
360 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
361 const struct tcphdr
*th
, unsigned int len
);
362 void tcp_rcv_space_adjust(struct sock
*sk
);
363 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
364 void tcp_twsk_destructor(struct sock
*sk
);
365 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
366 struct pipe_inode_info
*pipe
, size_t len
,
369 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
370 const unsigned int pkts
)
372 struct inet_connection_sock
*icsk
= inet_csk(sk
);
374 if (icsk
->icsk_ack
.quick
) {
375 if (pkts
>= icsk
->icsk_ack
.quick
) {
376 icsk
->icsk_ack
.quick
= 0;
377 /* Leaving quickack mode we deflate ATO. */
378 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
380 icsk
->icsk_ack
.quick
-= pkts
;
385 #define TCP_ECN_QUEUE_CWR 2
386 #define TCP_ECN_DEMAND_CWR 4
387 #define TCP_ECN_SEEN 8
397 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
399 const struct tcphdr
*th
);
400 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
401 struct request_sock
*req
, bool fastopen
);
402 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
403 struct sk_buff
*skb
);
404 void tcp_enter_loss(struct sock
*sk
);
405 void tcp_clear_retrans(struct tcp_sock
*tp
);
406 void tcp_update_metrics(struct sock
*sk
);
407 void tcp_init_metrics(struct sock
*sk
);
408 void tcp_metrics_init(void);
409 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
,
410 bool paws_check
, bool timestamps
);
411 bool tcp_remember_stamp(struct sock
*sk
);
412 bool tcp_tw_remember_stamp(struct inet_timewait_sock
*tw
);
413 void tcp_fetch_timewait_stamp(struct sock
*sk
, struct dst_entry
*dst
);
414 void tcp_disable_fack(struct tcp_sock
*tp
);
415 void tcp_close(struct sock
*sk
, long timeout
);
416 void tcp_init_sock(struct sock
*sk
);
417 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
,
418 struct poll_table_struct
*wait
);
419 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
420 char __user
*optval
, int __user
*optlen
);
421 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
422 char __user
*optval
, unsigned int optlen
);
423 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
424 char __user
*optval
, int __user
*optlen
);
425 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
426 char __user
*optval
, unsigned int optlen
);
427 void tcp_set_keepalive(struct sock
*sk
, int val
);
428 void tcp_syn_ack_timeout(const struct request_sock
*req
);
429 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
430 int flags
, int *addr_len
);
431 void tcp_parse_options(const struct sk_buff
*skb
,
432 struct tcp_options_received
*opt_rx
,
433 int estab
, struct tcp_fastopen_cookie
*foc
);
434 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
437 * TCP v4 functions exported for the inet6 API
440 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
441 void tcp_v4_mtu_reduced(struct sock
*sk
);
442 void tcp_req_err(struct sock
*sk
, u32 seq
);
443 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
444 struct sock
*tcp_create_openreq_child(struct sock
*sk
,
445 struct request_sock
*req
,
446 struct sk_buff
*skb
);
447 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
448 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
449 struct request_sock
*req
,
450 struct dst_entry
*dst
);
451 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
452 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
453 int tcp_connect(struct sock
*sk
);
454 struct sk_buff
*tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
455 struct request_sock
*req
,
456 struct tcp_fastopen_cookie
*foc
);
457 int tcp_disconnect(struct sock
*sk
, int flags
);
459 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
460 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
461 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
463 /* From syncookies.c */
464 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
466 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
467 #ifdef CONFIG_SYN_COOKIES
469 /* Syncookies use a monotonic timer which increments every 60 seconds.
470 * This counter is used both as a hash input and partially encoded into
471 * the cookie value. A cookie is only validated further if the delta
472 * between the current counter value and the encoded one is less than this,
473 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
474 * the counter advances immediately after a cookie is generated).
476 #define MAX_SYNCOOKIE_AGE 2
477 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
478 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
480 /* syncookies: remember time of last synqueue overflow
481 * But do not dirty this field too often (once per second is enough)
483 static inline void tcp_synq_overflow(struct sock
*sk
)
485 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
486 unsigned long now
= jiffies
;
488 if (time_after(now
, last_overflow
+ HZ
))
489 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= now
;
492 /* syncookies: no recent synqueue overflow on this listening socket? */
493 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
495 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
497 return time_after(jiffies
, last_overflow
+ TCP_SYNCOOKIE_VALID
);
500 static inline u32
tcp_cookie_time(void)
502 u64 val
= get_jiffies_64();
504 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
508 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
510 __u32
cookie_v4_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
512 __u32
cookie_init_timestamp(struct request_sock
*req
);
513 bool cookie_timestamp_decode(struct tcp_options_received
*opt
);
514 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
515 const struct net
*net
, const struct dst_entry
*dst
);
517 /* From net/ipv6/syncookies.c */
518 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
520 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
522 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
523 const struct tcphdr
*th
, u16
*mssp
);
524 __u32
cookie_v6_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
529 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
531 bool tcp_may_send_now(struct sock
*sk
);
532 int __tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
533 int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
534 void tcp_retransmit_timer(struct sock
*sk
);
535 void tcp_xmit_retransmit_queue(struct sock
*);
536 void tcp_simple_retransmit(struct sock
*);
537 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
538 int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int, gfp_t
);
540 void tcp_send_probe0(struct sock
*);
541 void tcp_send_partial(struct sock
*);
542 int tcp_write_wakeup(struct sock
*, int mib
);
543 void tcp_send_fin(struct sock
*sk
);
544 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
545 int tcp_send_synack(struct sock
*);
546 void tcp_push_one(struct sock
*, unsigned int mss_now
);
547 void tcp_send_ack(struct sock
*sk
);
548 void tcp_send_delayed_ack(struct sock
*sk
);
549 void tcp_send_loss_probe(struct sock
*sk
);
550 bool tcp_schedule_loss_probe(struct sock
*sk
);
553 void tcp_resume_early_retransmit(struct sock
*sk
);
554 void tcp_rearm_rto(struct sock
*sk
);
555 void tcp_reset(struct sock
*sk
);
558 void tcp_init_xmit_timers(struct sock
*);
559 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
561 inet_csk_clear_xmit_timers(sk
);
564 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
565 unsigned int tcp_current_mss(struct sock
*sk
);
567 /* Bound MSS / TSO packet size with the half of the window */
568 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
572 /* When peer uses tiny windows, there is no use in packetizing
573 * to sub-MSS pieces for the sake of SWS or making sure there
574 * are enough packets in the pipe for fast recovery.
576 * On the other hand, for extremely large MSS devices, handling
577 * smaller than MSS windows in this way does make sense.
579 if (tp
->max_window
>= 512)
580 cutoff
= (tp
->max_window
>> 1);
582 cutoff
= tp
->max_window
;
584 if (cutoff
&& pktsize
> cutoff
)
585 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
591 void tcp_get_info(struct sock
*, struct tcp_info
*);
593 /* Read 'sendfile()'-style from a TCP socket */
594 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
595 unsigned int, size_t);
596 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
597 sk_read_actor_t recv_actor
);
599 void tcp_initialize_rcv_mss(struct sock
*sk
);
601 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
602 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
603 void tcp_mtup_init(struct sock
*sk
);
604 void tcp_init_buffer_space(struct sock
*sk
);
606 static inline void tcp_bound_rto(const struct sock
*sk
)
608 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
609 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
612 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
614 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
617 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
619 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
620 ntohl(TCP_FLAG_ACK
) |
624 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
626 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
629 static inline void tcp_fast_path_check(struct sock
*sk
)
631 struct tcp_sock
*tp
= tcp_sk(sk
);
633 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
635 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
637 tcp_fast_path_on(tp
);
640 /* Compute the actual rto_min value */
641 static inline u32
tcp_rto_min(struct sock
*sk
)
643 const struct dst_entry
*dst
= __sk_dst_get(sk
);
644 u32 rto_min
= TCP_RTO_MIN
;
646 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
647 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
651 static inline u32
tcp_rto_min_us(struct sock
*sk
)
653 return jiffies_to_usecs(tcp_rto_min(sk
));
656 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
658 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
661 /* Compute the actual receive window we are currently advertising.
662 * Rcv_nxt can be after the window if our peer push more data
663 * than the offered window.
665 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
667 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
674 /* Choose a new window, without checks for shrinking, and without
675 * scaling applied to the result. The caller does these things
676 * if necessary. This is a "raw" window selection.
678 u32
__tcp_select_window(struct sock
*sk
);
680 void tcp_send_window_probe(struct sock
*sk
);
682 /* TCP timestamps are only 32-bits, this causes a slight
683 * complication on 64-bit systems since we store a snapshot
684 * of jiffies in the buffer control blocks below. We decided
685 * to use only the low 32-bits of jiffies and hide the ugly
686 * casts with the following macro.
688 #define tcp_time_stamp ((__u32)(jiffies))
690 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
692 return skb
->skb_mstamp
.stamp_jiffies
;
696 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
698 #define TCPHDR_FIN 0x01
699 #define TCPHDR_SYN 0x02
700 #define TCPHDR_RST 0x04
701 #define TCPHDR_PSH 0x08
702 #define TCPHDR_ACK 0x10
703 #define TCPHDR_URG 0x20
704 #define TCPHDR_ECE 0x40
705 #define TCPHDR_CWR 0x80
707 /* This is what the send packet queuing engine uses to pass
708 * TCP per-packet control information to the transmission code.
709 * We also store the host-order sequence numbers in here too.
710 * This is 44 bytes if IPV6 is enabled.
711 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
714 __u32 seq
; /* Starting sequence number */
715 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
717 /* Note : tcp_tw_isn is used in input path only
718 * (isn chosen by tcp_timewait_state_process())
720 * tcp_gso_segs is used in write queue only,
721 * cf tcp_skb_pcount()
726 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
728 __u8 sacked
; /* State flags for SACK/FACK. */
729 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
730 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
731 #define TCPCB_LOST 0x04 /* SKB is lost */
732 #define TCPCB_TAGBITS 0x07 /* All tag bits */
733 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
734 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
735 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
738 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
740 __u32 ack_seq
; /* Sequence number ACK'd */
742 struct inet_skb_parm h4
;
743 #if IS_ENABLED(CONFIG_IPV6)
744 struct inet6_skb_parm h6
;
746 } header
; /* For incoming frames */
749 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
752 #if IS_ENABLED(CONFIG_IPV6)
753 /* This is the variant of inet6_iif() that must be used by TCP,
754 * as TCP moves IP6CB into a different location in skb->cb[]
756 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
758 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
762 /* Due to TSO, an SKB can be composed of multiple actual
763 * packets. To keep these tracked properly, we use this.
765 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
767 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
770 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
772 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
775 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
777 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
780 /* This is valid iff tcp_skb_pcount() > 1. */
781 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
783 return skb_shinfo(skb
)->gso_size
;
786 /* Events passed to congestion control interface */
788 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
789 CA_EVENT_CWND_RESTART
, /* congestion window restart */
790 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
791 CA_EVENT_LOSS
, /* loss timeout */
792 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
793 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
794 CA_EVENT_DELAYED_ACK
, /* Delayed ack is sent */
795 CA_EVENT_NON_DELAYED_ACK
,
798 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
799 enum tcp_ca_ack_event_flags
{
800 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
801 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
802 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
806 * Interface for adding new TCP congestion control handlers
808 #define TCP_CA_NAME_MAX 16
809 #define TCP_CA_MAX 128
810 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
812 #define TCP_CA_UNSPEC 0
814 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
815 #define TCP_CONG_NON_RESTRICTED 0x1
816 /* Requires ECN/ECT set on all packets */
817 #define TCP_CONG_NEEDS_ECN 0x2
821 struct tcp_congestion_ops
{
822 struct list_head list
;
826 /* initialize private data (optional) */
827 void (*init
)(struct sock
*sk
);
828 /* cleanup private data (optional) */
829 void (*release
)(struct sock
*sk
);
831 /* return slow start threshold (required) */
832 u32 (*ssthresh
)(struct sock
*sk
);
833 /* do new cwnd calculation (required) */
834 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
835 /* call before changing ca_state (optional) */
836 void (*set_state
)(struct sock
*sk
, u8 new_state
);
837 /* call when cwnd event occurs (optional) */
838 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
839 /* call when ack arrives (optional) */
840 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
841 /* new value of cwnd after loss (optional) */
842 u32 (*undo_cwnd
)(struct sock
*sk
);
843 /* hook for packet ack accounting (optional) */
844 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
845 /* get info for inet_diag (optional) */
846 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
847 union tcp_cc_info
*info
);
849 char name
[TCP_CA_NAME_MAX
];
850 struct module
*owner
;
853 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
854 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
856 void tcp_assign_congestion_control(struct sock
*sk
);
857 void tcp_init_congestion_control(struct sock
*sk
);
858 void tcp_cleanup_congestion_control(struct sock
*sk
);
859 int tcp_set_default_congestion_control(const char *name
);
860 void tcp_get_default_congestion_control(char *name
);
861 void tcp_get_available_congestion_control(char *buf
, size_t len
);
862 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
863 int tcp_set_allowed_congestion_control(char *allowed
);
864 int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
865 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
866 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
868 u32
tcp_reno_ssthresh(struct sock
*sk
);
869 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
870 extern struct tcp_congestion_ops tcp_reno
;
872 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
873 u32
tcp_ca_get_key_by_name(const char *name
);
875 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
877 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
883 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
885 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
887 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
890 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
892 struct inet_connection_sock
*icsk
= inet_csk(sk
);
894 if (icsk
->icsk_ca_ops
->set_state
)
895 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
896 icsk
->icsk_ca_state
= ca_state
;
899 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
901 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
903 if (icsk
->icsk_ca_ops
->cwnd_event
)
904 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
907 /* These functions determine how the current flow behaves in respect of SACK
908 * handling. SACK is negotiated with the peer, and therefore it can vary
909 * between different flows.
911 * tcp_is_sack - SACK enabled
912 * tcp_is_reno - No SACK
913 * tcp_is_fack - FACK enabled, implies SACK enabled
915 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
917 return tp
->rx_opt
.sack_ok
;
920 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
922 return !tcp_is_sack(tp
);
925 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
927 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
930 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
932 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
935 /* TCP early-retransmit (ER) is similar to but more conservative than
936 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
938 static inline void tcp_enable_early_retrans(struct tcp_sock
*tp
)
940 tp
->do_early_retrans
= sysctl_tcp_early_retrans
&&
941 sysctl_tcp_early_retrans
< 4 && !sysctl_tcp_thin_dupack
&&
942 sysctl_tcp_reordering
== 3;
945 static inline void tcp_disable_early_retrans(struct tcp_sock
*tp
)
947 tp
->do_early_retrans
= 0;
950 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
952 return tp
->sacked_out
+ tp
->lost_out
;
955 /* This determines how many packets are "in the network" to the best
956 * of our knowledge. In many cases it is conservative, but where
957 * detailed information is available from the receiver (via SACK
958 * blocks etc.) we can make more aggressive calculations.
960 * Use this for decisions involving congestion control, use just
961 * tp->packets_out to determine if the send queue is empty or not.
963 * Read this equation as:
965 * "Packets sent once on transmission queue" MINUS
966 * "Packets left network, but not honestly ACKed yet" PLUS
967 * "Packets fast retransmitted"
969 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
971 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
974 #define TCP_INFINITE_SSTHRESH 0x7fffffff
976 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
978 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
981 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
983 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
984 (1 << inet_csk(sk
)->icsk_ca_state
);
987 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
988 * The exception is cwnd reduction phase, when cwnd is decreasing towards
991 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
993 const struct tcp_sock
*tp
= tcp_sk(sk
);
995 if (tcp_in_cwnd_reduction(sk
))
996 return tp
->snd_ssthresh
;
998 return max(tp
->snd_ssthresh
,
999 ((tp
->snd_cwnd
>> 1) +
1000 (tp
->snd_cwnd
>> 2)));
1003 /* Use define here intentionally to get WARN_ON location shown at the caller */
1004 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1006 void tcp_enter_cwr(struct sock
*sk
);
1007 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1009 /* The maximum number of MSS of available cwnd for which TSO defers
1010 * sending if not using sysctl_tcp_tso_win_divisor.
1012 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1017 /* Slow start with delack produces 3 packets of burst, so that
1018 * it is safe "de facto". This will be the default - same as
1019 * the default reordering threshold - but if reordering increases,
1020 * we must be able to allow cwnd to burst at least this much in order
1021 * to not pull it back when holes are filled.
1023 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
1025 return tp
->reordering
;
1028 /* Returns end sequence number of the receiver's advertised window */
1029 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1031 return tp
->snd_una
+ tp
->snd_wnd
;
1034 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1035 * flexible approach. The RFC suggests cwnd should not be raised unless
1036 * it was fully used previously. And that's exactly what we do in
1037 * congestion avoidance mode. But in slow start we allow cwnd to grow
1038 * as long as the application has used half the cwnd.
1040 * cwnd is 10 (IW10), but application sends 9 frames.
1041 * We allow cwnd to reach 18 when all frames are ACKed.
1042 * This check is safe because it's as aggressive as slow start which already
1043 * risks 100% overshoot. The advantage is that we discourage application to
1044 * either send more filler packets or data to artificially blow up the cwnd
1045 * usage, and allow application-limited process to probe bw more aggressively.
1047 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1049 const struct tcp_sock
*tp
= tcp_sk(sk
);
1051 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1052 if (tp
->snd_cwnd
<= tp
->snd_ssthresh
)
1053 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1055 return tp
->is_cwnd_limited
;
1058 /* Something is really bad, we could not queue an additional packet,
1059 * because qdisc is full or receiver sent a 0 window.
1060 * We do not want to add fuel to the fire, or abort too early,
1061 * so make sure the timer we arm now is at least 200ms in the future,
1062 * regardless of current icsk_rto value (as it could be ~2ms)
1064 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1066 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1069 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1070 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1071 unsigned long max_when
)
1073 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1075 return (unsigned long)min_t(u64
, when
, max_when
);
1078 static inline void tcp_check_probe_timer(struct sock
*sk
)
1080 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1081 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1082 tcp_probe0_base(sk
), TCP_RTO_MAX
);
1085 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1090 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1096 * Calculate(/check) TCP checksum
1098 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1099 __be32 daddr
, __wsum base
)
1101 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1104 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1106 return __skb_checksum_complete(skb
);
1109 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1111 return !skb_csum_unnecessary(skb
) &&
1112 __tcp_checksum_complete(skb
);
1115 /* Prequeue for VJ style copy to user, combined with checksumming. */
1117 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
1119 tp
->ucopy
.task
= NULL
;
1121 tp
->ucopy
.memory
= 0;
1122 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1125 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
);
1130 static const char *statename
[]={
1131 "Unused","Established","Syn Sent","Syn Recv",
1132 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1133 "Close Wait","Last ACK","Listen","Closing"
1136 void tcp_set_state(struct sock
*sk
, int state
);
1138 void tcp_done(struct sock
*sk
);
1140 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1143 rx_opt
->num_sacks
= 0;
1146 u32
tcp_default_init_rwnd(u32 mss
);
1148 /* Determine a window scaling and initial window to offer. */
1149 void tcp_select_initial_window(int __space
, __u32 mss
, __u32
*rcv_wnd
,
1150 __u32
*window_clamp
, int wscale_ok
,
1151 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1153 static inline int tcp_win_from_space(int space
)
1155 return sysctl_tcp_adv_win_scale
<=0 ?
1156 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1157 space
- (space
>>sysctl_tcp_adv_win_scale
);
1160 /* Note: caller must be prepared to deal with negative returns */
1161 static inline int tcp_space(const struct sock
*sk
)
1163 return tcp_win_from_space(sk
->sk_rcvbuf
-
1164 atomic_read(&sk
->sk_rmem_alloc
));
1167 static inline int tcp_full_space(const struct sock
*sk
)
1169 return tcp_win_from_space(sk
->sk_rcvbuf
);
1172 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1173 struct sock
*sk
, struct dst_entry
*dst
);
1175 void tcp_enter_memory_pressure(struct sock
*sk
);
1177 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1179 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1182 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1184 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1187 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1189 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
1192 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1194 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1196 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
1197 tcp_time_stamp
- tp
->rcv_tstamp
);
1200 static inline int tcp_fin_time(const struct sock
*sk
)
1202 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1203 const int rto
= inet_csk(sk
)->icsk_rto
;
1205 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1206 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1211 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1214 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1216 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1219 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1220 * then following tcp messages have valid values. Ignore 0 value,
1221 * or else 'negative' tsval might forbid us to accept their packets.
1223 if (!rx_opt
->ts_recent
)
1228 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1231 if (tcp_paws_check(rx_opt
, 0))
1234 /* RST segments are not recommended to carry timestamp,
1235 and, if they do, it is recommended to ignore PAWS because
1236 "their cleanup function should take precedence over timestamps."
1237 Certainly, it is mistake. It is necessary to understand the reasons
1238 of this constraint to relax it: if peer reboots, clock may go
1239 out-of-sync and half-open connections will not be reset.
1240 Actually, the problem would be not existing if all
1241 the implementations followed draft about maintaining clock
1242 via reboots. Linux-2.2 DOES NOT!
1244 However, we can relax time bounds for RST segments to MSL.
1246 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1251 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1252 int mib_idx
, u32
*last_oow_ack_time
);
1254 static inline void tcp_mib_init(struct net
*net
)
1257 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1258 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1259 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1260 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1264 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1266 tp
->lost_skb_hint
= NULL
;
1269 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1271 tcp_clear_retrans_hints_partial(tp
);
1272 tp
->retransmit_skb_hint
= NULL
;
1278 union tcp_md5_addr
{
1280 #if IS_ENABLED(CONFIG_IPV6)
1285 /* - key database */
1286 struct tcp_md5sig_key
{
1287 struct hlist_node node
;
1289 u8 family
; /* AF_INET or AF_INET6 */
1290 union tcp_md5_addr addr
;
1291 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1292 struct rcu_head rcu
;
1296 struct tcp_md5sig_info
{
1297 struct hlist_head head
;
1298 struct rcu_head rcu
;
1301 /* - pseudo header */
1302 struct tcp4_pseudohdr
{
1310 struct tcp6_pseudohdr
{
1311 struct in6_addr saddr
;
1312 struct in6_addr daddr
;
1314 __be32 protocol
; /* including padding */
1317 union tcp_md5sum_block
{
1318 struct tcp4_pseudohdr ip4
;
1319 #if IS_ENABLED(CONFIG_IPV6)
1320 struct tcp6_pseudohdr ip6
;
1324 /* - pool: digest algorithm, hash description and scratch buffer */
1325 struct tcp_md5sig_pool
{
1326 struct hash_desc md5_desc
;
1327 union tcp_md5sum_block md5_blk
;
1331 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1332 const struct sock
*sk
, const struct sk_buff
*skb
);
1333 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1334 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
);
1335 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1337 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
1338 const struct sock
*addr_sk
);
1340 #ifdef CONFIG_TCP_MD5SIG
1341 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1342 const union tcp_md5_addr
*addr
,
1344 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1346 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1347 const union tcp_md5_addr
*addr
,
1352 #define tcp_twsk_md5_key(twsk) NULL
1355 bool tcp_alloc_md5sig_pool(void);
1357 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1358 static inline void tcp_put_md5sig_pool(void)
1363 int tcp_md5_hash_header(struct tcp_md5sig_pool
*, const struct tcphdr
*);
1364 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1365 unsigned int header_len
);
1366 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1367 const struct tcp_md5sig_key
*key
);
1369 /* From tcp_fastopen.c */
1370 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1371 struct tcp_fastopen_cookie
*cookie
, int *syn_loss
,
1372 unsigned long *last_syn_loss
);
1373 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1374 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1376 struct tcp_fastopen_request
{
1377 /* Fast Open cookie. Size 0 means a cookie request */
1378 struct tcp_fastopen_cookie cookie
;
1379 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1381 int copied
; /* queued in tcp_connect() */
1383 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1385 extern struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
1386 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
);
1387 bool tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1388 struct request_sock
*req
,
1389 struct tcp_fastopen_cookie
*foc
,
1390 struct dst_entry
*dst
);
1391 void tcp_fastopen_init_key_once(bool publish
);
1392 #define TCP_FASTOPEN_KEY_LENGTH 16
1394 /* Fastopen key context */
1395 struct tcp_fastopen_context
{
1396 struct crypto_cipher
*tfm
;
1397 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1398 struct rcu_head rcu
;
1401 /* write queue abstraction */
1402 static inline void tcp_write_queue_purge(struct sock
*sk
)
1404 struct sk_buff
*skb
;
1406 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1407 sk_wmem_free_skb(sk
, skb
);
1409 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1412 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1414 return skb_peek(&sk
->sk_write_queue
);
1417 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1419 return skb_peek_tail(&sk
->sk_write_queue
);
1422 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1423 const struct sk_buff
*skb
)
1425 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1428 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1429 const struct sk_buff
*skb
)
1431 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1434 #define tcp_for_write_queue(skb, sk) \
1435 skb_queue_walk(&(sk)->sk_write_queue, skb)
1437 #define tcp_for_write_queue_from(skb, sk) \
1438 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1440 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1441 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1443 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1445 return sk
->sk_send_head
;
1448 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1449 const struct sk_buff
*skb
)
1451 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1454 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1456 if (tcp_skb_is_last(sk
, skb
))
1457 sk
->sk_send_head
= NULL
;
1459 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1462 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1464 if (sk
->sk_send_head
== skb_unlinked
)
1465 sk
->sk_send_head
= NULL
;
1468 static inline void tcp_init_send_head(struct sock
*sk
)
1470 sk
->sk_send_head
= NULL
;
1473 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1475 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1478 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1480 __tcp_add_write_queue_tail(sk
, skb
);
1482 /* Queue it, remembering where we must start sending. */
1483 if (sk
->sk_send_head
== NULL
) {
1484 sk
->sk_send_head
= skb
;
1486 if (tcp_sk(sk
)->highest_sack
== NULL
)
1487 tcp_sk(sk
)->highest_sack
= skb
;
1491 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1493 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1496 /* Insert buff after skb on the write queue of sk. */
1497 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1498 struct sk_buff
*buff
,
1501 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1504 /* Insert new before skb on the write queue of sk. */
1505 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1506 struct sk_buff
*skb
,
1509 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1511 if (sk
->sk_send_head
== skb
)
1512 sk
->sk_send_head
= new;
1515 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1517 __skb_unlink(skb
, &sk
->sk_write_queue
);
1520 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1522 return skb_queue_empty(&sk
->sk_write_queue
);
1525 static inline void tcp_push_pending_frames(struct sock
*sk
)
1527 if (tcp_send_head(sk
)) {
1528 struct tcp_sock
*tp
= tcp_sk(sk
);
1530 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1534 /* Start sequence of the skb just after the highest skb with SACKed
1535 * bit, valid only if sacked_out > 0 or when the caller has ensured
1536 * validity by itself.
1538 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1540 if (!tp
->sacked_out
)
1543 if (tp
->highest_sack
== NULL
)
1546 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1549 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1551 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1552 tcp_write_queue_next(sk
, skb
);
1555 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1557 return tcp_sk(sk
)->highest_sack
;
1560 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1562 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1565 /* Called when old skb is about to be deleted (to be combined with new skb) */
1566 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1567 struct sk_buff
*old
,
1568 struct sk_buff
*new)
1570 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1571 tcp_sk(sk
)->highest_sack
= new;
1574 /* Determines whether this is a thin stream (which may suffer from
1575 * increased latency). Used to trigger latency-reducing mechanisms.
1577 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1579 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1583 enum tcp_seq_states
{
1584 TCP_SEQ_STATE_LISTENING
,
1585 TCP_SEQ_STATE_OPENREQ
,
1586 TCP_SEQ_STATE_ESTABLISHED
,
1589 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1591 struct tcp_seq_afinfo
{
1594 const struct file_operations
*seq_fops
;
1595 struct seq_operations seq_ops
;
1598 struct tcp_iter_state
{
1599 struct seq_net_private p
;
1601 enum tcp_seq_states state
;
1602 struct sock
*syn_wait_sk
;
1603 int bucket
, offset
, sbucket
, num
;
1608 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1609 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1611 extern struct request_sock_ops tcp_request_sock_ops
;
1612 extern struct request_sock_ops tcp6_request_sock_ops
;
1614 void tcp_v4_destroy_sock(struct sock
*sk
);
1616 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1617 netdev_features_t features
);
1618 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1619 int tcp_gro_complete(struct sk_buff
*skb
);
1621 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1623 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1625 return tp
->notsent_lowat
?: sysctl_tcp_notsent_lowat
;
1628 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1630 const struct tcp_sock
*tp
= tcp_sk(sk
);
1631 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1633 return notsent_bytes
< tcp_notsent_lowat(tp
);
1636 #ifdef CONFIG_PROC_FS
1637 int tcp4_proc_init(void);
1638 void tcp4_proc_exit(void);
1641 int tcp_rtx_synack(struct sock
*sk
, struct request_sock
*req
);
1642 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1643 const struct tcp_request_sock_ops
*af_ops
,
1644 struct sock
*sk
, struct sk_buff
*skb
);
1646 /* TCP af-specific functions */
1647 struct tcp_sock_af_ops
{
1648 #ifdef CONFIG_TCP_MD5SIG
1649 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1650 const struct sock
*addr_sk
);
1651 int (*calc_md5_hash
)(char *location
,
1652 const struct tcp_md5sig_key
*md5
,
1653 const struct sock
*sk
,
1654 const struct sk_buff
*skb
);
1655 int (*md5_parse
)(struct sock
*sk
,
1656 char __user
*optval
,
1661 struct tcp_request_sock_ops
{
1663 #ifdef CONFIG_TCP_MD5SIG
1664 struct tcp_md5sig_key
*(*req_md5_lookup
)(struct sock
*sk
,
1665 const struct sock
*addr_sk
);
1666 int (*calc_md5_hash
) (char *location
,
1667 const struct tcp_md5sig_key
*md5
,
1668 const struct sock
*sk
,
1669 const struct sk_buff
*skb
);
1671 void (*init_req
)(struct request_sock
*req
, struct sock
*sk
,
1672 struct sk_buff
*skb
);
1673 #ifdef CONFIG_SYN_COOKIES
1674 __u32 (*cookie_init_seq
)(struct sock
*sk
, const struct sk_buff
*skb
,
1677 struct dst_entry
*(*route_req
)(struct sock
*sk
, struct flowi
*fl
,
1678 const struct request_sock
*req
,
1680 __u32 (*init_seq
)(const struct sk_buff
*skb
);
1681 int (*send_synack
)(struct sock
*sk
, struct dst_entry
*dst
,
1682 struct flowi
*fl
, struct request_sock
*req
,
1683 u16 queue_mapping
, struct tcp_fastopen_cookie
*foc
);
1684 void (*queue_hash_add
)(struct sock
*sk
, struct request_sock
*req
,
1685 const unsigned long timeout
);
1688 #ifdef CONFIG_SYN_COOKIES
1689 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1690 struct sock
*sk
, struct sk_buff
*skb
,
1693 return ops
->cookie_init_seq(sk
, skb
, mss
);
1696 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1697 struct sock
*sk
, struct sk_buff
*skb
,
1704 int tcpv4_offload_init(void);
1706 void tcp_v4_init(void);
1707 void tcp_init(void);
1710 * Save and compile IPv4 options, return a pointer to it
1712 static inline struct ip_options_rcu
*tcp_v4_save_options(struct sk_buff
*skb
)
1714 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
1715 struct ip_options_rcu
*dopt
= NULL
;
1718 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
1720 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
1721 if (dopt
&& __ip_options_echo(&dopt
->opt
, skb
, opt
)) {
1729 /* locally generated TCP pure ACKs have skb->truesize == 2
1730 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1731 * This is much faster than dissecting the packet to find out.
1732 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1734 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
1736 return skb
->truesize
== 2;
1739 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)