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/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.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 extern 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 reordering. */
74 #define TCP_MAX_REORDERING 127
76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
77 #define TCP_MAX_QUICKACKS 16U
80 #define TCP_URG_VALID 0x0100
81 #define TCP_URG_NOTYET 0x0200
82 #define TCP_URG_READ 0x0400
84 #define TCP_RETR1 3 /*
85 * This is how many retries it does before it
86 * tries to figure out if the gateway is
87 * down. Minimal RFC value is 3; it corresponds
88 * to ~3sec-8min depending on RTO.
91 #define TCP_RETR2 15 /*
92 * This should take at least
93 * 90 minutes to time out.
94 * RFC1122 says that the limit is 100 sec.
95 * 15 is ~13-30min depending on RTO.
98 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
99 * when active opening a connection.
100 * RFC1122 says the minimum retry MUST
101 * be at least 180secs. Nevertheless
102 * this value is corresponding to
103 * 63secs of retransmission with the
104 * current initial RTO.
107 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
108 * when passive opening a connection.
109 * This is corresponding to 31secs of
110 * retransmission with the current
114 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
115 * state, about 60 seconds */
116 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
117 /* BSD style FIN_WAIT2 deadlock breaker.
118 * It used to be 3min, new value is 60sec,
119 * to combine FIN-WAIT-2 timeout with
123 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
125 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
126 #define TCP_ATO_MIN ((unsigned)(HZ/25))
128 #define TCP_DELACK_MIN 4U
129 #define TCP_ATO_MIN 4U
131 #define TCP_RTO_MAX ((unsigned)(120*HZ))
132 #define TCP_RTO_MIN ((unsigned)(HZ/5))
133 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
134 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
135 * used as a fallback RTO for the
136 * initial data transmission if no
137 * valid RTT sample has been acquired,
138 * most likely due to retrans in 3WHS.
141 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
142 * for local resources.
145 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
146 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
147 #define TCP_KEEPALIVE_INTVL (75*HZ)
149 #define MAX_TCP_KEEPIDLE 32767
150 #define MAX_TCP_KEEPINTVL 32767
151 #define MAX_TCP_KEEPCNT 127
152 #define MAX_TCP_SYNCNT 127
154 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
156 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
157 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
158 * after this time. It should be equal
159 * (or greater than) TCP_TIMEWAIT_LEN
160 * to provide reliability equal to one
161 * provided by timewait state.
163 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
164 * timestamps. It must be less than
165 * minimal timewait lifetime.
171 #define TCPOPT_NOP 1 /* Padding */
172 #define TCPOPT_EOL 0 /* End of options */
173 #define TCPOPT_MSS 2 /* Segment size negotiating */
174 #define TCPOPT_WINDOW 3 /* Window scaling */
175 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
176 #define TCPOPT_SACK 5 /* SACK Block */
177 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
178 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
179 #define TCPOPT_EXP 254 /* Experimental */
180 /* Magic number to be after the option value for sharing TCP
181 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
183 #define TCPOPT_FASTOPEN_MAGIC 0xF989
189 #define TCPOLEN_MSS 4
190 #define TCPOLEN_WINDOW 3
191 #define TCPOLEN_SACK_PERM 2
192 #define TCPOLEN_TIMESTAMP 10
193 #define TCPOLEN_MD5SIG 18
194 #define TCPOLEN_EXP_FASTOPEN_BASE 4
196 /* But this is what stacks really send out. */
197 #define TCPOLEN_TSTAMP_ALIGNED 12
198 #define TCPOLEN_WSCALE_ALIGNED 4
199 #define TCPOLEN_SACKPERM_ALIGNED 4
200 #define TCPOLEN_SACK_BASE 2
201 #define TCPOLEN_SACK_BASE_ALIGNED 4
202 #define TCPOLEN_SACK_PERBLOCK 8
203 #define TCPOLEN_MD5SIG_ALIGNED 20
204 #define TCPOLEN_MSS_ALIGNED 4
206 /* Flags in tp->nonagle */
207 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
208 #define TCP_NAGLE_CORK 2 /* Socket is corked */
209 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
211 /* TCP thin-stream limits */
212 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
214 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
215 #define TCP_INIT_CWND 10
217 /* Bit Flags for sysctl_tcp_fastopen */
218 #define TFO_CLIENT_ENABLE 1
219 #define TFO_SERVER_ENABLE 2
220 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
222 /* Process SYN data but skip cookie validation */
223 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100
224 /* Accept SYN data w/o any cookie option */
225 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
227 /* Force enable TFO on all listeners, i.e., not requiring the
228 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
230 #define TFO_SERVER_WO_SOCKOPT1 0x400
231 #define TFO_SERVER_WO_SOCKOPT2 0x800
232 /* Always create TFO child sockets on a TFO listener even when
233 * cookie/data not present. (For testing purpose!)
235 #define TFO_SERVER_ALWAYS 0x1000
237 extern struct inet_timewait_death_row tcp_death_row
;
239 /* sysctl variables for tcp */
240 extern int sysctl_tcp_timestamps
;
241 extern int sysctl_tcp_window_scaling
;
242 extern int sysctl_tcp_sack
;
243 extern int sysctl_tcp_fin_timeout
;
244 extern int sysctl_tcp_keepalive_time
;
245 extern int sysctl_tcp_keepalive_probes
;
246 extern int sysctl_tcp_keepalive_intvl
;
247 extern int sysctl_tcp_syn_retries
;
248 extern int sysctl_tcp_synack_retries
;
249 extern int sysctl_tcp_retries1
;
250 extern int sysctl_tcp_retries2
;
251 extern int sysctl_tcp_orphan_retries
;
252 extern int sysctl_tcp_syncookies
;
253 extern int sysctl_tcp_fastopen
;
254 extern int sysctl_tcp_retrans_collapse
;
255 extern int sysctl_tcp_stdurg
;
256 extern int sysctl_tcp_rfc1337
;
257 extern int sysctl_tcp_abort_on_overflow
;
258 extern int sysctl_tcp_max_orphans
;
259 extern int sysctl_tcp_fack
;
260 extern int sysctl_tcp_reordering
;
261 extern int sysctl_tcp_dsack
;
262 extern int sysctl_tcp_wmem
[3];
263 extern int sysctl_tcp_rmem
[3];
264 extern int sysctl_tcp_app_win
;
265 extern int sysctl_tcp_adv_win_scale
;
266 extern int sysctl_tcp_tw_reuse
;
267 extern int sysctl_tcp_frto
;
268 extern int sysctl_tcp_low_latency
;
269 extern int sysctl_tcp_dma_copybreak
;
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_mtu_probing
;
274 extern int sysctl_tcp_base_mss
;
275 extern int sysctl_tcp_workaround_signed_windows
;
276 extern int sysctl_tcp_slow_start_after_idle
;
277 extern int sysctl_tcp_max_ssthresh
;
278 extern int sysctl_tcp_thin_linear_timeouts
;
279 extern int sysctl_tcp_thin_dupack
;
280 extern int sysctl_tcp_early_retrans
;
281 extern int sysctl_tcp_limit_output_bytes
;
282 extern int sysctl_tcp_challenge_ack_limit
;
283 extern unsigned int sysctl_tcp_notsent_lowat
;
284 extern int sysctl_tcp_min_tso_segs
;
286 extern atomic_long_t tcp_memory_allocated
;
287 extern struct percpu_counter tcp_sockets_allocated
;
288 extern int tcp_memory_pressure
;
291 * The next routines deal with comparing 32 bit unsigned ints
292 * and worry about wraparound (automatic with unsigned arithmetic).
295 static inline bool before(__u32 seq1
, __u32 seq2
)
297 return (__s32
)(seq1
-seq2
) < 0;
299 #define after(seq2, seq1) before(seq1, seq2)
301 /* is s2<=s1<=s3 ? */
302 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
304 return seq3
- seq2
>= seq1
- seq2
;
307 static inline bool tcp_out_of_memory(struct sock
*sk
)
309 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
310 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
315 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
317 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
318 int orphans
= percpu_counter_read_positive(ocp
);
320 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
321 orphans
= percpu_counter_sum_positive(ocp
);
322 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
328 extern bool tcp_check_oom(struct sock
*sk
, int shift
);
330 /* syncookies: remember time of last synqueue overflow */
331 static inline void tcp_synq_overflow(struct sock
*sk
)
333 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= jiffies
;
336 /* syncookies: no recent synqueue overflow on this listening socket? */
337 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
339 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
340 return time_after(jiffies
, last_overflow
+ TCP_TIMEOUT_FALLBACK
);
343 extern struct proto tcp_prot
;
345 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
346 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
347 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
348 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
349 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
351 extern void tcp_init_mem(struct net
*net
);
353 extern void tcp_tasklet_init(void);
355 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
357 extern void tcp_shutdown (struct sock
*sk
, int how
);
359 extern void tcp_v4_early_demux(struct sk_buff
*skb
);
360 extern int tcp_v4_rcv(struct sk_buff
*skb
);
362 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
363 extern int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
365 extern int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
366 size_t size
, int flags
);
367 extern void tcp_release_cb(struct sock
*sk
);
368 extern void tcp_wfree(struct sk_buff
*skb
);
369 extern void tcp_write_timer_handler(struct sock
*sk
);
370 extern void tcp_delack_timer_handler(struct sock
*sk
);
371 extern int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
372 extern int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
373 const struct tcphdr
*th
, unsigned int len
);
374 extern void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
375 const struct tcphdr
*th
, unsigned int len
);
376 extern void tcp_rcv_space_adjust(struct sock
*sk
);
377 extern void tcp_cleanup_rbuf(struct sock
*sk
, int copied
);
378 extern int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
379 extern void tcp_twsk_destructor(struct sock
*sk
);
380 extern ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
381 struct pipe_inode_info
*pipe
, size_t len
,
384 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
385 const unsigned int pkts
)
387 struct inet_connection_sock
*icsk
= inet_csk(sk
);
389 if (icsk
->icsk_ack
.quick
) {
390 if (pkts
>= icsk
->icsk_ack
.quick
) {
391 icsk
->icsk_ack
.quick
= 0;
392 /* Leaving quickack mode we deflate ATO. */
393 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
395 icsk
->icsk_ack
.quick
-= pkts
;
400 #define TCP_ECN_QUEUE_CWR 2
401 #define TCP_ECN_DEMAND_CWR 4
402 #define TCP_ECN_SEEN 8
412 extern enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
414 const struct tcphdr
*th
);
415 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
416 struct request_sock
*req
,
417 struct request_sock
**prev
,
419 extern int tcp_child_process(struct sock
*parent
, struct sock
*child
,
420 struct sk_buff
*skb
);
421 extern void tcp_enter_loss(struct sock
*sk
, int how
);
422 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
423 extern void tcp_update_metrics(struct sock
*sk
);
424 extern void tcp_init_metrics(struct sock
*sk
);
425 extern void tcp_metrics_init(void);
426 extern bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
, bool paws_check
);
427 extern bool tcp_remember_stamp(struct sock
*sk
);
428 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock
*tw
);
429 extern void tcp_fetch_timewait_stamp(struct sock
*sk
, struct dst_entry
*dst
);
430 extern void tcp_disable_fack(struct tcp_sock
*tp
);
431 extern void tcp_close(struct sock
*sk
, long timeout
);
432 extern void tcp_init_sock(struct sock
*sk
);
433 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
,
434 struct poll_table_struct
*wait
);
435 extern int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
436 char __user
*optval
, int __user
*optlen
);
437 extern int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
438 char __user
*optval
, unsigned int optlen
);
439 extern int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
440 char __user
*optval
, int __user
*optlen
);
441 extern int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
442 char __user
*optval
, unsigned int optlen
);
443 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
444 extern void tcp_syn_ack_timeout(struct sock
*sk
, struct request_sock
*req
);
445 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
446 size_t len
, int nonblock
, int flags
, int *addr_len
);
447 extern void tcp_parse_options(const struct sk_buff
*skb
,
448 struct tcp_options_received
*opt_rx
,
449 int estab
, struct tcp_fastopen_cookie
*foc
);
450 extern const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
453 * TCP v4 functions exported for the inet6 API
456 extern void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
457 extern int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
458 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
459 struct request_sock
*req
,
460 struct sk_buff
*skb
);
461 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
462 struct request_sock
*req
,
463 struct dst_entry
*dst
);
464 extern int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
465 extern int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
467 extern int tcp_connect(struct sock
*sk
);
468 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
469 struct request_sock
*req
,
470 struct tcp_fastopen_cookie
*foc
);
471 extern int tcp_disconnect(struct sock
*sk
, int flags
);
473 void tcp_connect_init(struct sock
*sk
);
474 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
475 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
476 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
478 /* From syncookies.c */
479 extern __u32 syncookie_secret
[2][16-4+SHA_DIGEST_WORDS
];
480 extern int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
482 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
483 struct ip_options
*opt
);
484 #ifdef CONFIG_SYN_COOKIES
485 #include <linux/ktime.h>
487 /* Syncookies use a monotonic timer which increments every 60 seconds.
488 * This counter is used both as a hash input and partially encoded into
489 * the cookie value. A cookie is only validated further if the delta
490 * between the current counter value and the encoded one is less than this,
491 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
492 * the counter advances immediately after a cookie is generated).
494 #define MAX_SYNCOOKIE_AGE 2
496 static inline u32
tcp_cookie_time(void)
498 u64 val
= get_jiffies_64();
500 do_div(val
, 60 * HZ
);
504 extern u32
__cookie_v4_init_sequence(const struct iphdr
*iph
,
505 const struct tcphdr
*th
, u16
*mssp
);
506 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
509 static inline __u32
cookie_v4_init_sequence(struct sock
*sk
,
517 extern __u32
cookie_init_timestamp(struct request_sock
*req
);
518 extern bool cookie_check_timestamp(struct tcp_options_received
*opt
,
519 struct net
*net
, bool *ecn_ok
);
521 /* From net/ipv6/syncookies.c */
522 extern int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
524 extern struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
525 #ifdef CONFIG_SYN_COOKIES
526 extern u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
527 const struct tcphdr
*th
, u16
*mssp
);
528 extern __u32
cookie_v6_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
531 static inline __u32
cookie_v6_init_sequence(struct sock
*sk
,
540 extern void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
542 extern bool tcp_may_send_now(struct sock
*sk
);
543 extern int __tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
544 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
545 extern void tcp_retransmit_timer(struct sock
*sk
);
546 extern void tcp_xmit_retransmit_queue(struct sock
*);
547 extern void tcp_simple_retransmit(struct sock
*);
548 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
549 extern int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int);
551 extern void tcp_send_probe0(struct sock
*);
552 extern void tcp_send_partial(struct sock
*);
553 extern int tcp_write_wakeup(struct sock
*);
554 extern void tcp_send_fin(struct sock
*sk
);
555 extern void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
556 extern int tcp_send_synack(struct sock
*);
557 extern bool tcp_syn_flood_action(struct sock
*sk
,
558 const struct sk_buff
*skb
,
560 extern void tcp_push_one(struct sock
*, unsigned int mss_now
);
561 extern void tcp_send_ack(struct sock
*sk
);
562 extern void tcp_send_delayed_ack(struct sock
*sk
);
563 extern void tcp_send_loss_probe(struct sock
*sk
);
564 extern bool tcp_schedule_loss_probe(struct sock
*sk
);
567 extern void tcp_cwnd_application_limited(struct sock
*sk
);
568 extern void tcp_resume_early_retransmit(struct sock
*sk
);
569 extern void tcp_rearm_rto(struct sock
*sk
);
570 extern void tcp_reset(struct sock
*sk
);
573 extern void tcp_init_xmit_timers(struct sock
*);
574 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
576 inet_csk_clear_xmit_timers(sk
);
579 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
580 extern unsigned int tcp_current_mss(struct sock
*sk
);
582 /* Bound MSS / TSO packet size with the half of the window */
583 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
587 /* When peer uses tiny windows, there is no use in packetizing
588 * to sub-MSS pieces for the sake of SWS or making sure there
589 * are enough packets in the pipe for fast recovery.
591 * On the other hand, for extremely large MSS devices, handling
592 * smaller than MSS windows in this way does make sense.
594 if (tp
->max_window
>= 512)
595 cutoff
= (tp
->max_window
>> 1);
597 cutoff
= tp
->max_window
;
599 if (cutoff
&& pktsize
> cutoff
)
600 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
606 extern void tcp_get_info(const struct sock
*, struct tcp_info
*);
608 /* Read 'sendfile()'-style from a TCP socket */
609 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
610 unsigned int, size_t);
611 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
612 sk_read_actor_t recv_actor
);
614 extern void tcp_initialize_rcv_mss(struct sock
*sk
);
616 extern int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
617 extern int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
618 extern void tcp_mtup_init(struct sock
*sk
);
619 extern void tcp_init_buffer_space(struct sock
*sk
);
621 static inline void tcp_bound_rto(const struct sock
*sk
)
623 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
624 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
627 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
629 return (tp
->srtt
>> 3) + tp
->rttvar
;
632 extern void tcp_set_rto(struct sock
*sk
);
634 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
636 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
637 ntohl(TCP_FLAG_ACK
) |
641 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
643 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
646 static inline void tcp_fast_path_check(struct sock
*sk
)
648 struct tcp_sock
*tp
= tcp_sk(sk
);
650 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
652 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
654 tcp_fast_path_on(tp
);
657 /* Compute the actual rto_min value */
658 static inline u32
tcp_rto_min(struct sock
*sk
)
660 const struct dst_entry
*dst
= __sk_dst_get(sk
);
661 u32 rto_min
= TCP_RTO_MIN
;
663 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
664 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
668 /* Compute the actual receive window we are currently advertising.
669 * Rcv_nxt can be after the window if our peer push more data
670 * than the offered window.
672 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
674 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
681 /* Choose a new window, without checks for shrinking, and without
682 * scaling applied to the result. The caller does these things
683 * if necessary. This is a "raw" window selection.
685 extern u32
__tcp_select_window(struct sock
*sk
);
687 void tcp_send_window_probe(struct sock
*sk
);
689 /* TCP timestamps are only 32-bits, this causes a slight
690 * complication on 64-bit systems since we store a snapshot
691 * of jiffies in the buffer control blocks below. We decided
692 * to use only the low 32-bits of jiffies and hide the ugly
693 * casts with the following macro.
695 #define tcp_time_stamp ((__u32)(jiffies))
697 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
699 #define TCPHDR_FIN 0x01
700 #define TCPHDR_SYN 0x02
701 #define TCPHDR_RST 0x04
702 #define TCPHDR_PSH 0x08
703 #define TCPHDR_ACK 0x10
704 #define TCPHDR_URG 0x20
705 #define TCPHDR_ECE 0x40
706 #define TCPHDR_CWR 0x80
708 /* This is what the send packet queuing engine uses to pass
709 * TCP per-packet control information to the transmission code.
710 * We also store the host-order sequence numbers in here too.
711 * This is 44 bytes if IPV6 is enabled.
712 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
716 struct inet_skb_parm h4
;
717 #if IS_ENABLED(CONFIG_IPV6)
718 struct inet6_skb_parm h6
;
720 } header
; /* For incoming frames */
721 __u32 seq
; /* Starting sequence number */
722 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
723 __u32 when
; /* used to compute rtt's */
724 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
726 __u8 sacked
; /* State flags for SACK/FACK. */
727 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
728 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
729 #define TCPCB_LOST 0x04 /* SKB is lost */
730 #define TCPCB_TAGBITS 0x07 /* All tag bits */
731 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
732 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
734 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
736 __u32 ack_seq
; /* Sequence number ACK'd */
739 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
741 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
743 * If we receive a SYN packet with these bits set, it means a network is
744 * playing bad games with TOS bits. In order to avoid possible false congestion
745 * notifications, we disable TCP ECN negociation.
748 TCP_ECN_create_request(struct request_sock
*req
, const struct sk_buff
*skb
,
751 const struct tcphdr
*th
= tcp_hdr(skb
);
753 if (net
->ipv4
.sysctl_tcp_ecn
&& th
->ece
&& th
->cwr
&&
754 INET_ECN_is_not_ect(TCP_SKB_CB(skb
)->ip_dsfield
))
755 inet_rsk(req
)->ecn_ok
= 1;
758 /* Due to TSO, an SKB can be composed of multiple actual
759 * packets. To keep these tracked properly, we use this.
761 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
763 return skb_shinfo(skb
)->gso_segs
;
766 /* This is valid iff tcp_skb_pcount() > 1. */
767 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
769 return skb_shinfo(skb
)->gso_size
;
772 /* Events passed to congestion control interface */
774 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
775 CA_EVENT_CWND_RESTART
, /* congestion window restart */
776 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
777 CA_EVENT_LOSS
, /* loss timeout */
778 CA_EVENT_FAST_ACK
, /* in sequence ack */
779 CA_EVENT_SLOW_ACK
, /* other ack */
783 * Interface for adding new TCP congestion control handlers
785 #define TCP_CA_NAME_MAX 16
786 #define TCP_CA_MAX 128
787 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
789 #define TCP_CONG_NON_RESTRICTED 0x1
790 #define TCP_CONG_RTT_STAMP 0x2
792 struct tcp_congestion_ops
{
793 struct list_head list
;
796 /* initialize private data (optional) */
797 void (*init
)(struct sock
*sk
);
798 /* cleanup private data (optional) */
799 void (*release
)(struct sock
*sk
);
801 /* return slow start threshold (required) */
802 u32 (*ssthresh
)(struct sock
*sk
);
803 /* lower bound for congestion window (optional) */
804 u32 (*min_cwnd
)(const struct sock
*sk
);
805 /* do new cwnd calculation (required) */
806 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 in_flight
);
807 /* call before changing ca_state (optional) */
808 void (*set_state
)(struct sock
*sk
, u8 new_state
);
809 /* call when cwnd event occurs (optional) */
810 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
811 /* new value of cwnd after loss (optional) */
812 u32 (*undo_cwnd
)(struct sock
*sk
);
813 /* hook for packet ack accounting (optional) */
814 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
815 /* get info for inet_diag (optional) */
816 void (*get_info
)(struct sock
*sk
, u32 ext
, struct sk_buff
*skb
);
818 char name
[TCP_CA_NAME_MAX
];
819 struct module
*owner
;
822 extern int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
823 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
825 extern void tcp_init_congestion_control(struct sock
*sk
);
826 extern void tcp_cleanup_congestion_control(struct sock
*sk
);
827 extern int tcp_set_default_congestion_control(const char *name
);
828 extern void tcp_get_default_congestion_control(char *name
);
829 extern void tcp_get_available_congestion_control(char *buf
, size_t len
);
830 extern void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
831 extern int tcp_set_allowed_congestion_control(char *allowed
);
832 extern int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
833 extern void tcp_slow_start(struct tcp_sock
*tp
);
834 extern void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
);
836 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
837 extern u32
tcp_reno_ssthresh(struct sock
*sk
);
838 extern void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 in_flight
);
839 extern u32
tcp_reno_min_cwnd(const struct sock
*sk
);
840 extern struct tcp_congestion_ops tcp_reno
;
842 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
844 struct inet_connection_sock
*icsk
= inet_csk(sk
);
846 if (icsk
->icsk_ca_ops
->set_state
)
847 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
848 icsk
->icsk_ca_state
= ca_state
;
851 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
853 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
855 if (icsk
->icsk_ca_ops
->cwnd_event
)
856 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
859 /* These functions determine how the current flow behaves in respect of SACK
860 * handling. SACK is negotiated with the peer, and therefore it can vary
861 * between different flows.
863 * tcp_is_sack - SACK enabled
864 * tcp_is_reno - No SACK
865 * tcp_is_fack - FACK enabled, implies SACK enabled
867 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
869 return tp
->rx_opt
.sack_ok
;
872 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
874 return !tcp_is_sack(tp
);
877 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
879 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
882 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
884 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
887 /* TCP early-retransmit (ER) is similar to but more conservative than
888 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
890 static inline void tcp_enable_early_retrans(struct tcp_sock
*tp
)
892 tp
->do_early_retrans
= sysctl_tcp_early_retrans
&&
893 sysctl_tcp_early_retrans
< 4 && !sysctl_tcp_thin_dupack
&&
894 sysctl_tcp_reordering
== 3;
897 static inline void tcp_disable_early_retrans(struct tcp_sock
*tp
)
899 tp
->do_early_retrans
= 0;
902 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
904 return tp
->sacked_out
+ tp
->lost_out
;
907 /* This determines how many packets are "in the network" to the best
908 * of our knowledge. In many cases it is conservative, but where
909 * detailed information is available from the receiver (via SACK
910 * blocks etc.) we can make more aggressive calculations.
912 * Use this for decisions involving congestion control, use just
913 * tp->packets_out to determine if the send queue is empty or not.
915 * Read this equation as:
917 * "Packets sent once on transmission queue" MINUS
918 * "Packets left network, but not honestly ACKed yet" PLUS
919 * "Packets fast retransmitted"
921 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
923 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
926 #define TCP_INFINITE_SSTHRESH 0x7fffffff
928 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
930 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
933 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
935 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
936 (1 << inet_csk(sk
)->icsk_ca_state
);
939 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
940 * The exception is cwnd reduction phase, when cwnd is decreasing towards
943 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
945 const struct tcp_sock
*tp
= tcp_sk(sk
);
947 if (tcp_in_cwnd_reduction(sk
))
948 return tp
->snd_ssthresh
;
950 return max(tp
->snd_ssthresh
,
951 ((tp
->snd_cwnd
>> 1) +
952 (tp
->snd_cwnd
>> 2)));
955 /* Use define here intentionally to get WARN_ON location shown at the caller */
956 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
958 extern void tcp_enter_cwr(struct sock
*sk
, const int set_ssthresh
);
959 extern __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
961 /* The maximum number of MSS of available cwnd for which TSO defers
962 * sending if not using sysctl_tcp_tso_win_divisor.
964 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
969 /* Slow start with delack produces 3 packets of burst, so that
970 * it is safe "de facto". This will be the default - same as
971 * the default reordering threshold - but if reordering increases,
972 * we must be able to allow cwnd to burst at least this much in order
973 * to not pull it back when holes are filled.
975 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
977 return tp
->reordering
;
980 /* Returns end sequence number of the receiver's advertised window */
981 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
983 return tp
->snd_una
+ tp
->snd_wnd
;
985 extern bool tcp_is_cwnd_limited(const struct sock
*sk
, u32 in_flight
);
987 static inline void tcp_minshall_update(struct tcp_sock
*tp
, unsigned int mss
,
988 const struct sk_buff
*skb
)
991 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
994 static inline void tcp_check_probe_timer(struct sock
*sk
)
996 const struct tcp_sock
*tp
= tcp_sk(sk
);
997 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
999 if (!tp
->packets_out
&& !icsk
->icsk_pending
)
1000 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1001 icsk
->icsk_rto
, TCP_RTO_MAX
);
1004 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1009 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1015 * Calculate(/check) TCP checksum
1017 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1018 __be32 daddr
, __wsum base
)
1020 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1023 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1025 return __skb_checksum_complete(skb
);
1028 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1030 return !skb_csum_unnecessary(skb
) &&
1031 __tcp_checksum_complete(skb
);
1034 /* Prequeue for VJ style copy to user, combined with checksumming. */
1036 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
1038 tp
->ucopy
.task
= NULL
;
1040 tp
->ucopy
.memory
= 0;
1041 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1042 #ifdef CONFIG_NET_DMA
1043 tp
->ucopy
.dma_chan
= NULL
;
1044 tp
->ucopy
.wakeup
= 0;
1045 tp
->ucopy
.pinned_list
= NULL
;
1046 tp
->ucopy
.dma_cookie
= 0;
1050 extern bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
);
1055 static const char *statename
[]={
1056 "Unused","Established","Syn Sent","Syn Recv",
1057 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1058 "Close Wait","Last ACK","Listen","Closing"
1061 extern void tcp_set_state(struct sock
*sk
, int state
);
1063 extern void tcp_done(struct sock
*sk
);
1065 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1068 rx_opt
->num_sacks
= 0;
1071 extern u32
tcp_default_init_rwnd(u32 mss
);
1073 /* Determine a window scaling and initial window to offer. */
1074 extern void tcp_select_initial_window(int __space
, __u32 mss
,
1075 __u32
*rcv_wnd
, __u32
*window_clamp
,
1076 int wscale_ok
, __u8
*rcv_wscale
,
1077 __u32 init_rcv_wnd
);
1079 static inline int tcp_win_from_space(int space
)
1081 return sysctl_tcp_adv_win_scale
<=0 ?
1082 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1083 space
- (space
>>sysctl_tcp_adv_win_scale
);
1086 /* Note: caller must be prepared to deal with negative returns */
1087 static inline int tcp_space(const struct sock
*sk
)
1089 return tcp_win_from_space(sk
->sk_rcvbuf
-
1090 atomic_read(&sk
->sk_rmem_alloc
));
1093 static inline int tcp_full_space(const struct sock
*sk
)
1095 return tcp_win_from_space(sk
->sk_rcvbuf
);
1098 static inline void tcp_openreq_init(struct request_sock
*req
,
1099 struct tcp_options_received
*rx_opt
,
1100 struct sk_buff
*skb
)
1102 struct inet_request_sock
*ireq
= inet_rsk(req
);
1104 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1106 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1107 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->seq
+ 1;
1108 tcp_rsk(req
)->snt_synack
= 0;
1109 req
->mss
= rx_opt
->mss_clamp
;
1110 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1111 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1112 ireq
->sack_ok
= rx_opt
->sack_ok
;
1113 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1114 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1117 ireq
->rmt_port
= tcp_hdr(skb
)->source
;
1118 ireq
->loc_port
= tcp_hdr(skb
)->dest
;
1121 extern void tcp_enter_memory_pressure(struct sock
*sk
);
1123 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1125 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1128 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1130 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1133 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1135 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
1138 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1140 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1142 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
1143 tcp_time_stamp
- tp
->rcv_tstamp
);
1146 static inline int tcp_fin_time(const struct sock
*sk
)
1148 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1149 const int rto
= inet_csk(sk
)->icsk_rto
;
1151 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1152 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1157 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1160 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1162 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1165 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1166 * then following tcp messages have valid values. Ignore 0 value,
1167 * or else 'negative' tsval might forbid us to accept their packets.
1169 if (!rx_opt
->ts_recent
)
1174 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1177 if (tcp_paws_check(rx_opt
, 0))
1180 /* RST segments are not recommended to carry timestamp,
1181 and, if they do, it is recommended to ignore PAWS because
1182 "their cleanup function should take precedence over timestamps."
1183 Certainly, it is mistake. It is necessary to understand the reasons
1184 of this constraint to relax it: if peer reboots, clock may go
1185 out-of-sync and half-open connections will not be reset.
1186 Actually, the problem would be not existing if all
1187 the implementations followed draft about maintaining clock
1188 via reboots. Linux-2.2 DOES NOT!
1190 However, we can relax time bounds for RST segments to MSL.
1192 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1197 static inline void tcp_mib_init(struct net
*net
)
1200 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1201 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1202 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1203 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1207 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1209 tp
->lost_skb_hint
= NULL
;
1212 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1214 tcp_clear_retrans_hints_partial(tp
);
1215 tp
->retransmit_skb_hint
= NULL
;
1221 union tcp_md5_addr
{
1223 #if IS_ENABLED(CONFIG_IPV6)
1228 /* - key database */
1229 struct tcp_md5sig_key
{
1230 struct hlist_node node
;
1232 u8 family
; /* AF_INET or AF_INET6 */
1233 union tcp_md5_addr addr
;
1234 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1235 struct rcu_head rcu
;
1239 struct tcp_md5sig_info
{
1240 struct hlist_head head
;
1241 struct rcu_head rcu
;
1244 /* - pseudo header */
1245 struct tcp4_pseudohdr
{
1253 struct tcp6_pseudohdr
{
1254 struct in6_addr saddr
;
1255 struct in6_addr daddr
;
1257 __be32 protocol
; /* including padding */
1260 union tcp_md5sum_block
{
1261 struct tcp4_pseudohdr ip4
;
1262 #if IS_ENABLED(CONFIG_IPV6)
1263 struct tcp6_pseudohdr ip6
;
1267 /* - pool: digest algorithm, hash description and scratch buffer */
1268 struct tcp_md5sig_pool
{
1269 struct hash_desc md5_desc
;
1270 union tcp_md5sum_block md5_blk
;
1274 extern int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1275 const struct sock
*sk
,
1276 const struct request_sock
*req
,
1277 const struct sk_buff
*skb
);
1278 extern int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1279 int family
, const u8
*newkey
,
1280 u8 newkeylen
, gfp_t gfp
);
1281 extern int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1283 extern struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
1284 struct sock
*addr_sk
);
1286 #ifdef CONFIG_TCP_MD5SIG
1287 extern struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1288 const union tcp_md5_addr
*addr
, int family
);
1289 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1291 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1292 const union tcp_md5_addr
*addr
,
1297 #define tcp_twsk_md5_key(twsk) NULL
1300 extern bool tcp_alloc_md5sig_pool(void);
1302 extern struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1303 static inline void tcp_put_md5sig_pool(void)
1308 extern int tcp_md5_hash_header(struct tcp_md5sig_pool
*, const struct tcphdr
*);
1309 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1310 unsigned int header_len
);
1311 extern int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1312 const struct tcp_md5sig_key
*key
);
1314 /* From tcp_fastopen.c */
1315 extern void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1316 struct tcp_fastopen_cookie
*cookie
,
1317 int *syn_loss
, unsigned long *last_syn_loss
);
1318 extern void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1319 struct tcp_fastopen_cookie
*cookie
,
1321 struct tcp_fastopen_request
{
1322 /* Fast Open cookie. Size 0 means a cookie request */
1323 struct tcp_fastopen_cookie cookie
;
1324 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1326 int copied
; /* queued in tcp_connect() */
1328 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1330 extern struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
1331 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
);
1332 extern void tcp_fastopen_cookie_gen(__be32 src
, __be32 dst
,
1333 struct tcp_fastopen_cookie
*foc
);
1335 #define TCP_FASTOPEN_KEY_LENGTH 16
1337 /* Fastopen key context */
1338 struct tcp_fastopen_context
{
1339 struct crypto_cipher
*tfm
;
1340 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1341 struct rcu_head rcu
;
1344 /* write queue abstraction */
1345 static inline void tcp_write_queue_purge(struct sock
*sk
)
1347 struct sk_buff
*skb
;
1349 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1350 sk_wmem_free_skb(sk
, skb
);
1352 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1355 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1357 return skb_peek(&sk
->sk_write_queue
);
1360 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1362 return skb_peek_tail(&sk
->sk_write_queue
);
1365 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1366 const struct sk_buff
*skb
)
1368 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1371 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1372 const struct sk_buff
*skb
)
1374 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1377 #define tcp_for_write_queue(skb, sk) \
1378 skb_queue_walk(&(sk)->sk_write_queue, skb)
1380 #define tcp_for_write_queue_from(skb, sk) \
1381 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1383 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1384 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1386 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1388 return sk
->sk_send_head
;
1391 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1392 const struct sk_buff
*skb
)
1394 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1397 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1399 if (tcp_skb_is_last(sk
, skb
))
1400 sk
->sk_send_head
= NULL
;
1402 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1405 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1407 if (sk
->sk_send_head
== skb_unlinked
)
1408 sk
->sk_send_head
= NULL
;
1411 static inline void tcp_init_send_head(struct sock
*sk
)
1413 sk
->sk_send_head
= NULL
;
1416 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1418 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1421 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1423 __tcp_add_write_queue_tail(sk
, skb
);
1425 /* Queue it, remembering where we must start sending. */
1426 if (sk
->sk_send_head
== NULL
) {
1427 sk
->sk_send_head
= skb
;
1429 if (tcp_sk(sk
)->highest_sack
== NULL
)
1430 tcp_sk(sk
)->highest_sack
= skb
;
1434 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1436 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1439 /* Insert buff after skb on the write queue of sk. */
1440 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1441 struct sk_buff
*buff
,
1444 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1447 /* Insert new before skb on the write queue of sk. */
1448 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1449 struct sk_buff
*skb
,
1452 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1454 if (sk
->sk_send_head
== skb
)
1455 sk
->sk_send_head
= new;
1458 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1460 __skb_unlink(skb
, &sk
->sk_write_queue
);
1463 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1465 return skb_queue_empty(&sk
->sk_write_queue
);
1468 static inline void tcp_push_pending_frames(struct sock
*sk
)
1470 if (tcp_send_head(sk
)) {
1471 struct tcp_sock
*tp
= tcp_sk(sk
);
1473 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1477 /* Start sequence of the skb just after the highest skb with SACKed
1478 * bit, valid only if sacked_out > 0 or when the caller has ensured
1479 * validity by itself.
1481 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1483 if (!tp
->sacked_out
)
1486 if (tp
->highest_sack
== NULL
)
1489 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1492 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1494 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1495 tcp_write_queue_next(sk
, skb
);
1498 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1500 return tcp_sk(sk
)->highest_sack
;
1503 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1505 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1508 /* Called when old skb is about to be deleted (to be combined with new skb) */
1509 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1510 struct sk_buff
*old
,
1511 struct sk_buff
*new)
1513 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1514 tcp_sk(sk
)->highest_sack
= new;
1517 /* Determines whether this is a thin stream (which may suffer from
1518 * increased latency). Used to trigger latency-reducing mechanisms.
1520 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1522 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1526 enum tcp_seq_states
{
1527 TCP_SEQ_STATE_LISTENING
,
1528 TCP_SEQ_STATE_OPENREQ
,
1529 TCP_SEQ_STATE_ESTABLISHED
,
1530 TCP_SEQ_STATE_TIME_WAIT
,
1533 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1535 struct tcp_seq_afinfo
{
1538 const struct file_operations
*seq_fops
;
1539 struct seq_operations seq_ops
;
1542 struct tcp_iter_state
{
1543 struct seq_net_private p
;
1545 enum tcp_seq_states state
;
1546 struct sock
*syn_wait_sk
;
1547 int bucket
, offset
, sbucket
, num
;
1552 extern int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1553 extern void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1555 extern struct request_sock_ops tcp_request_sock_ops
;
1556 extern struct request_sock_ops tcp6_request_sock_ops
;
1558 extern void tcp_v4_destroy_sock(struct sock
*sk
);
1560 extern struct sk_buff
*tcp_tso_segment(struct sk_buff
*skb
,
1561 netdev_features_t features
);
1562 extern struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
,
1563 struct sk_buff
*skb
);
1564 extern int tcp_gro_complete(struct sk_buff
*skb
);
1566 extern void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
,
1569 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1571 return tp
->notsent_lowat
?: sysctl_tcp_notsent_lowat
;
1574 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1576 const struct tcp_sock
*tp
= tcp_sk(sk
);
1577 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1579 return notsent_bytes
< tcp_notsent_lowat(tp
);
1582 #ifdef CONFIG_PROC_FS
1583 extern int tcp4_proc_init(void);
1584 extern void tcp4_proc_exit(void);
1587 /* TCP af-specific functions */
1588 struct tcp_sock_af_ops
{
1589 #ifdef CONFIG_TCP_MD5SIG
1590 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1591 struct sock
*addr_sk
);
1592 int (*calc_md5_hash
) (char *location
,
1593 struct tcp_md5sig_key
*md5
,
1594 const struct sock
*sk
,
1595 const struct request_sock
*req
,
1596 const struct sk_buff
*skb
);
1597 int (*md5_parse
) (struct sock
*sk
,
1598 char __user
*optval
,
1603 struct tcp_request_sock_ops
{
1604 #ifdef CONFIG_TCP_MD5SIG
1605 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1606 struct request_sock
*req
);
1607 int (*calc_md5_hash
) (char *location
,
1608 struct tcp_md5sig_key
*md5
,
1609 const struct sock
*sk
,
1610 const struct request_sock
*req
,
1611 const struct sk_buff
*skb
);
1615 extern int tcpv4_offload_init(void);
1617 extern void tcp_v4_init(void);
1618 extern void tcp_init(void);