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>
34 #include <linux/ktime.h>
36 #include <net/inet_connection_sock.h>
37 #include <net/inet_timewait_sock.h>
38 #include <net/inet_hashtables.h>
39 #include <net/checksum.h>
40 #include <net/request_sock.h>
44 #include <net/tcp_states.h>
45 #include <net/inet_ecn.h>
48 #include <linux/seq_file.h>
49 #include <linux/memcontrol.h>
51 extern struct inet_hashinfo tcp_hashinfo
;
53 extern struct percpu_counter tcp_orphan_count
;
54 void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
56 #define MAX_TCP_HEADER (128 + MAX_HEADER)
57 #define MAX_TCP_OPTION_SPACE 40
60 * Never offer a window over 32767 without using window scaling. Some
61 * poor stacks do signed 16bit maths!
63 #define MAX_TCP_WINDOW 32767U
65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66 #define TCP_MIN_MSS 88U
68 /* The least MTU to use for probing */
69 #define TCP_BASE_MSS 512
71 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
72 #define TCP_FASTRETRANS_THRESH 3
74 /* Maximal reordering. */
75 #define TCP_MAX_REORDERING 127
77 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
78 #define TCP_MAX_QUICKACKS 16U
81 #define TCP_URG_VALID 0x0100
82 #define TCP_URG_NOTYET 0x0200
83 #define TCP_URG_READ 0x0400
85 #define TCP_RETR1 3 /*
86 * This is how many retries it does before it
87 * tries to figure out if the gateway is
88 * down. Minimal RFC value is 3; it corresponds
89 * to ~3sec-8min depending on RTO.
92 #define TCP_RETR2 15 /*
93 * This should take at least
94 * 90 minutes to time out.
95 * RFC1122 says that the limit is 100 sec.
96 * 15 is ~13-30min depending on RTO.
99 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
100 * when active opening a connection.
101 * RFC1122 says the minimum retry MUST
102 * be at least 180secs. Nevertheless
103 * this value is corresponding to
104 * 63secs of retransmission with the
105 * current initial RTO.
108 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
109 * when passive opening a connection.
110 * This is corresponding to 31secs of
111 * retransmission with the current
115 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
116 * state, about 60 seconds */
117 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
118 /* BSD style FIN_WAIT2 deadlock breaker.
119 * It used to be 3min, new value is 60sec,
120 * to combine FIN-WAIT-2 timeout with
124 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
126 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
127 #define TCP_ATO_MIN ((unsigned)(HZ/25))
129 #define TCP_DELACK_MIN 4U
130 #define TCP_ATO_MIN 4U
132 #define TCP_RTO_MAX ((unsigned)(120*HZ))
133 #define TCP_RTO_MIN ((unsigned)(HZ/5))
134 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
135 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
136 * used as a fallback RTO for the
137 * initial data transmission if no
138 * valid RTT sample has been acquired,
139 * most likely due to retrans in 3WHS.
142 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
143 * for local resources.
146 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
147 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
148 #define TCP_KEEPALIVE_INTVL (75*HZ)
150 #define MAX_TCP_KEEPIDLE 32767
151 #define MAX_TCP_KEEPINTVL 32767
152 #define MAX_TCP_KEEPCNT 127
153 #define MAX_TCP_SYNCNT 127
155 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
157 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
158 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
159 * after this time. It should be equal
160 * (or greater than) TCP_TIMEWAIT_LEN
161 * to provide reliability equal to one
162 * provided by timewait state.
164 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
165 * timestamps. It must be less than
166 * minimal timewait lifetime.
172 #define TCPOPT_NOP 1 /* Padding */
173 #define TCPOPT_EOL 0 /* End of options */
174 #define TCPOPT_MSS 2 /* Segment size negotiating */
175 #define TCPOPT_WINDOW 3 /* Window scaling */
176 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
177 #define TCPOPT_SACK 5 /* SACK Block */
178 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
179 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
180 #define TCPOPT_EXP 254 /* Experimental */
181 /* Magic number to be after the option value for sharing TCP
182 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
184 #define TCPOPT_FASTOPEN_MAGIC 0xF989
190 #define TCPOLEN_MSS 4
191 #define TCPOLEN_WINDOW 3
192 #define TCPOLEN_SACK_PERM 2
193 #define TCPOLEN_TIMESTAMP 10
194 #define TCPOLEN_MD5SIG 18
195 #define TCPOLEN_EXP_FASTOPEN_BASE 4
197 /* But this is what stacks really send out. */
198 #define TCPOLEN_TSTAMP_ALIGNED 12
199 #define TCPOLEN_WSCALE_ALIGNED 4
200 #define TCPOLEN_SACKPERM_ALIGNED 4
201 #define TCPOLEN_SACK_BASE 2
202 #define TCPOLEN_SACK_BASE_ALIGNED 4
203 #define TCPOLEN_SACK_PERBLOCK 8
204 #define TCPOLEN_MD5SIG_ALIGNED 20
205 #define TCPOLEN_MSS_ALIGNED 4
207 /* Flags in tp->nonagle */
208 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
209 #define TCP_NAGLE_CORK 2 /* Socket is corked */
210 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
212 /* TCP thin-stream limits */
213 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
215 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
216 #define TCP_INIT_CWND 10
218 /* Bit Flags for sysctl_tcp_fastopen */
219 #define TFO_CLIENT_ENABLE 1
220 #define TFO_SERVER_ENABLE 2
221 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
223 /* Accept SYN data w/o any cookie option */
224 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
226 /* Force enable TFO on all listeners, i.e., not requiring the
227 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
229 #define TFO_SERVER_WO_SOCKOPT1 0x400
230 #define TFO_SERVER_WO_SOCKOPT2 0x800
232 extern struct inet_timewait_death_row tcp_death_row
;
234 /* sysctl variables for tcp */
235 extern int sysctl_tcp_timestamps
;
236 extern int sysctl_tcp_window_scaling
;
237 extern int sysctl_tcp_sack
;
238 extern int sysctl_tcp_fin_timeout
;
239 extern int sysctl_tcp_keepalive_time
;
240 extern int sysctl_tcp_keepalive_probes
;
241 extern int sysctl_tcp_keepalive_intvl
;
242 extern int sysctl_tcp_syn_retries
;
243 extern int sysctl_tcp_synack_retries
;
244 extern int sysctl_tcp_retries1
;
245 extern int sysctl_tcp_retries2
;
246 extern int sysctl_tcp_orphan_retries
;
247 extern int sysctl_tcp_syncookies
;
248 extern int sysctl_tcp_fastopen
;
249 extern int sysctl_tcp_retrans_collapse
;
250 extern int sysctl_tcp_stdurg
;
251 extern int sysctl_tcp_rfc1337
;
252 extern int sysctl_tcp_abort_on_overflow
;
253 extern int sysctl_tcp_max_orphans
;
254 extern int sysctl_tcp_fack
;
255 extern int sysctl_tcp_reordering
;
256 extern int sysctl_tcp_dsack
;
257 extern long sysctl_tcp_mem
[3];
258 extern int sysctl_tcp_wmem
[3];
259 extern int sysctl_tcp_rmem
[3];
260 extern int sysctl_tcp_app_win
;
261 extern int sysctl_tcp_adv_win_scale
;
262 extern int sysctl_tcp_tw_reuse
;
263 extern int sysctl_tcp_frto
;
264 extern int sysctl_tcp_low_latency
;
265 extern int sysctl_tcp_dma_copybreak
;
266 extern int sysctl_tcp_nometrics_save
;
267 extern int sysctl_tcp_moderate_rcvbuf
;
268 extern int sysctl_tcp_tso_win_divisor
;
269 extern int sysctl_tcp_mtu_probing
;
270 extern int sysctl_tcp_base_mss
;
271 extern int sysctl_tcp_workaround_signed_windows
;
272 extern int sysctl_tcp_slow_start_after_idle
;
273 extern int sysctl_tcp_thin_linear_timeouts
;
274 extern int sysctl_tcp_thin_dupack
;
275 extern int sysctl_tcp_early_retrans
;
276 extern int sysctl_tcp_limit_output_bytes
;
277 extern int sysctl_tcp_challenge_ack_limit
;
278 extern unsigned int sysctl_tcp_notsent_lowat
;
279 extern int sysctl_tcp_min_tso_segs
;
280 extern int sysctl_tcp_autocorking
;
282 extern atomic_long_t tcp_memory_allocated
;
283 extern struct percpu_counter tcp_sockets_allocated
;
284 extern int tcp_memory_pressure
;
287 * The next routines deal with comparing 32 bit unsigned ints
288 * and worry about wraparound (automatic with unsigned arithmetic).
291 static inline bool before(__u32 seq1
, __u32 seq2
)
293 return (__s32
)(seq1
-seq2
) < 0;
295 #define after(seq2, seq1) before(seq1, seq2)
297 /* is s2<=s1<=s3 ? */
298 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
300 return seq3
- seq2
>= seq1
- seq2
;
303 static inline bool tcp_out_of_memory(struct sock
*sk
)
305 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
306 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
311 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
313 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
314 int orphans
= percpu_counter_read_positive(ocp
);
316 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
317 orphans
= percpu_counter_sum_positive(ocp
);
318 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
324 bool tcp_check_oom(struct sock
*sk
, int shift
);
326 /* syncookies: remember time of last synqueue overflow */
327 static inline void tcp_synq_overflow(struct sock
*sk
)
329 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= jiffies
;
332 /* syncookies: no recent synqueue overflow on this listening socket? */
333 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
335 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
336 return time_after(jiffies
, last_overflow
+ TCP_TIMEOUT_FALLBACK
);
339 extern struct proto tcp_prot
;
341 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
342 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
343 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
344 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
345 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
347 void tcp_tasklet_init(void);
349 void tcp_v4_err(struct sk_buff
*skb
, u32
);
351 void tcp_shutdown(struct sock
*sk
, int how
);
353 void tcp_v4_early_demux(struct sk_buff
*skb
);
354 int tcp_v4_rcv(struct sk_buff
*skb
);
356 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
357 int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
359 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
361 void tcp_release_cb(struct sock
*sk
);
362 void tcp_wfree(struct sk_buff
*skb
);
363 void tcp_write_timer_handler(struct sock
*sk
);
364 void tcp_delack_timer_handler(struct sock
*sk
);
365 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
366 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
367 const struct tcphdr
*th
, unsigned int len
);
368 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
369 const struct tcphdr
*th
, unsigned int len
);
370 void tcp_rcv_space_adjust(struct sock
*sk
);
371 void tcp_cleanup_rbuf(struct sock
*sk
, int copied
);
372 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
373 void tcp_twsk_destructor(struct sock
*sk
);
374 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
375 struct pipe_inode_info
*pipe
, size_t len
,
378 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
379 const unsigned int pkts
)
381 struct inet_connection_sock
*icsk
= inet_csk(sk
);
383 if (icsk
->icsk_ack
.quick
) {
384 if (pkts
>= icsk
->icsk_ack
.quick
) {
385 icsk
->icsk_ack
.quick
= 0;
386 /* Leaving quickack mode we deflate ATO. */
387 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
389 icsk
->icsk_ack
.quick
-= pkts
;
394 #define TCP_ECN_QUEUE_CWR 2
395 #define TCP_ECN_DEMAND_CWR 4
396 #define TCP_ECN_SEEN 8
406 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
408 const struct tcphdr
*th
);
409 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
410 struct request_sock
*req
, struct request_sock
**prev
,
412 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
413 struct sk_buff
*skb
);
414 void tcp_enter_loss(struct sock
*sk
);
415 void tcp_clear_retrans(struct tcp_sock
*tp
);
416 void tcp_update_metrics(struct sock
*sk
);
417 void tcp_init_metrics(struct sock
*sk
);
418 void tcp_metrics_init(void);
419 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
,
420 bool paws_check
, bool timestamps
);
421 bool tcp_remember_stamp(struct sock
*sk
);
422 bool tcp_tw_remember_stamp(struct inet_timewait_sock
*tw
);
423 void tcp_fetch_timewait_stamp(struct sock
*sk
, struct dst_entry
*dst
);
424 void tcp_disable_fack(struct tcp_sock
*tp
);
425 void tcp_close(struct sock
*sk
, long timeout
);
426 void tcp_init_sock(struct sock
*sk
);
427 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
,
428 struct poll_table_struct
*wait
);
429 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
430 char __user
*optval
, int __user
*optlen
);
431 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
432 char __user
*optval
, unsigned int optlen
);
433 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
434 char __user
*optval
, int __user
*optlen
);
435 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
436 char __user
*optval
, unsigned int optlen
);
437 void tcp_set_keepalive(struct sock
*sk
, int val
);
438 void tcp_syn_ack_timeout(struct sock
*sk
, struct request_sock
*req
);
439 int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
440 size_t len
, int nonblock
, int flags
, int *addr_len
);
441 void tcp_parse_options(const struct sk_buff
*skb
,
442 struct tcp_options_received
*opt_rx
,
443 int estab
, struct tcp_fastopen_cookie
*foc
);
444 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
447 * TCP v4 functions exported for the inet6 API
450 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
451 void tcp_v4_mtu_reduced(struct sock
*sk
);
452 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
453 struct sock
*tcp_create_openreq_child(struct sock
*sk
,
454 struct request_sock
*req
,
455 struct sk_buff
*skb
);
456 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
457 struct request_sock
*req
,
458 struct dst_entry
*dst
);
459 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
460 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
461 int tcp_connect(struct sock
*sk
);
462 struct sk_buff
*tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
463 struct request_sock
*req
,
464 struct tcp_fastopen_cookie
*foc
);
465 int tcp_disconnect(struct sock
*sk
, int flags
);
467 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
468 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
469 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
471 /* From syncookies.c */
472 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
474 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
475 struct ip_options
*opt
);
476 #ifdef CONFIG_SYN_COOKIES
478 /* Syncookies use a monotonic timer which increments every 60 seconds.
479 * This counter is used both as a hash input and partially encoded into
480 * the cookie value. A cookie is only validated further if the delta
481 * between the current counter value and the encoded one is less than this,
482 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
483 * the counter advances immediately after a cookie is generated).
485 #define MAX_SYNCOOKIE_AGE 2
487 static inline u32
tcp_cookie_time(void)
489 u64 val
= get_jiffies_64();
491 do_div(val
, 60 * HZ
);
495 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
497 __u32
cookie_v4_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
501 __u32
cookie_init_timestamp(struct request_sock
*req
);
502 bool cookie_check_timestamp(struct tcp_options_received
*opt
, struct net
*net
,
505 /* From net/ipv6/syncookies.c */
506 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
508 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
509 #ifdef CONFIG_SYN_COOKIES
510 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
511 const struct tcphdr
*th
, u16
*mssp
);
512 __u32
cookie_v6_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
517 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
519 bool tcp_may_send_now(struct sock
*sk
);
520 int __tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
521 int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
522 void tcp_retransmit_timer(struct sock
*sk
);
523 void tcp_xmit_retransmit_queue(struct sock
*);
524 void tcp_simple_retransmit(struct sock
*);
525 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
526 int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int, gfp_t
);
528 void tcp_send_probe0(struct sock
*);
529 void tcp_send_partial(struct sock
*);
530 int tcp_write_wakeup(struct sock
*);
531 void tcp_send_fin(struct sock
*sk
);
532 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
533 int tcp_send_synack(struct sock
*);
534 bool tcp_syn_flood_action(struct sock
*sk
, const struct sk_buff
*skb
,
536 void tcp_push_one(struct sock
*, unsigned int mss_now
);
537 void tcp_send_ack(struct sock
*sk
);
538 void tcp_send_delayed_ack(struct sock
*sk
);
539 void tcp_send_loss_probe(struct sock
*sk
);
540 bool tcp_schedule_loss_probe(struct sock
*sk
);
543 void tcp_resume_early_retransmit(struct sock
*sk
);
544 void tcp_rearm_rto(struct sock
*sk
);
545 void tcp_reset(struct sock
*sk
);
548 void tcp_init_xmit_timers(struct sock
*);
549 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
551 inet_csk_clear_xmit_timers(sk
);
554 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
555 unsigned int tcp_current_mss(struct sock
*sk
);
557 /* Bound MSS / TSO packet size with the half of the window */
558 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
562 /* When peer uses tiny windows, there is no use in packetizing
563 * to sub-MSS pieces for the sake of SWS or making sure there
564 * are enough packets in the pipe for fast recovery.
566 * On the other hand, for extremely large MSS devices, handling
567 * smaller than MSS windows in this way does make sense.
569 if (tp
->max_window
>= 512)
570 cutoff
= (tp
->max_window
>> 1);
572 cutoff
= tp
->max_window
;
574 if (cutoff
&& pktsize
> cutoff
)
575 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
581 void tcp_get_info(const struct sock
*, struct tcp_info
*);
583 /* Read 'sendfile()'-style from a TCP socket */
584 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
585 unsigned int, size_t);
586 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
587 sk_read_actor_t recv_actor
);
589 void tcp_initialize_rcv_mss(struct sock
*sk
);
591 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
592 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
593 void tcp_mtup_init(struct sock
*sk
);
594 void tcp_init_buffer_space(struct sock
*sk
);
596 static inline void tcp_bound_rto(const struct sock
*sk
)
598 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
599 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
602 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
604 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
607 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
609 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
610 ntohl(TCP_FLAG_ACK
) |
614 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
616 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
619 static inline void tcp_fast_path_check(struct sock
*sk
)
621 struct tcp_sock
*tp
= tcp_sk(sk
);
623 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
625 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
627 tcp_fast_path_on(tp
);
630 /* Compute the actual rto_min value */
631 static inline u32
tcp_rto_min(struct sock
*sk
)
633 const struct dst_entry
*dst
= __sk_dst_get(sk
);
634 u32 rto_min
= TCP_RTO_MIN
;
636 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
637 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
641 static inline u32
tcp_rto_min_us(struct sock
*sk
)
643 return jiffies_to_usecs(tcp_rto_min(sk
));
646 /* Compute the actual receive window we are currently advertising.
647 * Rcv_nxt can be after the window if our peer push more data
648 * than the offered window.
650 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
652 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
659 /* Choose a new window, without checks for shrinking, and without
660 * scaling applied to the result. The caller does these things
661 * if necessary. This is a "raw" window selection.
663 u32
__tcp_select_window(struct sock
*sk
);
665 void tcp_send_window_probe(struct sock
*sk
);
667 /* TCP timestamps are only 32-bits, this causes a slight
668 * complication on 64-bit systems since we store a snapshot
669 * of jiffies in the buffer control blocks below. We decided
670 * to use only the low 32-bits of jiffies and hide the ugly
671 * casts with the following macro.
673 #define tcp_time_stamp ((__u32)(jiffies))
675 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
677 return skb
->skb_mstamp
.stamp_jiffies
;
681 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
683 #define TCPHDR_FIN 0x01
684 #define TCPHDR_SYN 0x02
685 #define TCPHDR_RST 0x04
686 #define TCPHDR_PSH 0x08
687 #define TCPHDR_ACK 0x10
688 #define TCPHDR_URG 0x20
689 #define TCPHDR_ECE 0x40
690 #define TCPHDR_CWR 0x80
692 /* This is what the send packet queuing engine uses to pass
693 * TCP per-packet control information to the transmission code.
694 * We also store the host-order sequence numbers in here too.
695 * This is 44 bytes if IPV6 is enabled.
696 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
699 __u32 seq
; /* Starting sequence number */
700 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
702 /* Note : tcp_tw_isn is used in input path only
703 * (isn chosen by tcp_timewait_state_process())
705 * tcp_gso_segs is used in write queue only,
706 * cf tcp_skb_pcount()
711 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
713 __u8 sacked
; /* State flags for SACK/FACK. */
714 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
715 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
716 #define TCPCB_LOST 0x04 /* SKB is lost */
717 #define TCPCB_TAGBITS 0x07 /* All tag bits */
718 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
719 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
720 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
723 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
725 __u32 ack_seq
; /* Sequence number ACK'd */
727 struct inet_skb_parm h4
;
728 #if IS_ENABLED(CONFIG_IPV6)
729 struct inet6_skb_parm h6
;
731 } header
; /* For incoming frames */
734 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
736 /* Due to TSO, an SKB can be composed of multiple actual
737 * packets. To keep these tracked properly, we use this.
739 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
741 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
744 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
746 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
749 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
751 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
754 /* This is valid iff tcp_skb_pcount() > 1. */
755 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
757 return skb_shinfo(skb
)->gso_size
;
760 /* Events passed to congestion control interface */
762 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
763 CA_EVENT_CWND_RESTART
, /* congestion window restart */
764 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
765 CA_EVENT_LOSS
, /* loss timeout */
766 CA_EVENT_FAST_ACK
, /* in sequence ack */
767 CA_EVENT_SLOW_ACK
, /* other ack */
771 * Interface for adding new TCP congestion control handlers
773 #define TCP_CA_NAME_MAX 16
774 #define TCP_CA_MAX 128
775 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
777 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
778 #define TCP_CONG_NON_RESTRICTED 0x1
779 /* Requires ECN/ECT set on all packets */
780 #define TCP_CONG_NEEDS_ECN 0x2
782 struct tcp_congestion_ops
{
783 struct list_head list
;
786 /* initialize private data (optional) */
787 void (*init
)(struct sock
*sk
);
788 /* cleanup private data (optional) */
789 void (*release
)(struct sock
*sk
);
791 /* return slow start threshold (required) */
792 u32 (*ssthresh
)(struct sock
*sk
);
793 /* do new cwnd calculation (required) */
794 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
795 /* call before changing ca_state (optional) */
796 void (*set_state
)(struct sock
*sk
, u8 new_state
);
797 /* call when cwnd event occurs (optional) */
798 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
799 /* new value of cwnd after loss (optional) */
800 u32 (*undo_cwnd
)(struct sock
*sk
);
801 /* hook for packet ack accounting (optional) */
802 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
803 /* get info for inet_diag (optional) */
804 void (*get_info
)(struct sock
*sk
, u32 ext
, struct sk_buff
*skb
);
806 char name
[TCP_CA_NAME_MAX
];
807 struct module
*owner
;
810 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
811 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
813 void tcp_assign_congestion_control(struct sock
*sk
);
814 void tcp_init_congestion_control(struct sock
*sk
);
815 void tcp_cleanup_congestion_control(struct sock
*sk
);
816 int tcp_set_default_congestion_control(const char *name
);
817 void tcp_get_default_congestion_control(char *name
);
818 void tcp_get_available_congestion_control(char *buf
, size_t len
);
819 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
820 int tcp_set_allowed_congestion_control(char *allowed
);
821 int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
822 int tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
823 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
);
825 u32
tcp_reno_ssthresh(struct sock
*sk
);
826 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
827 extern struct tcp_congestion_ops tcp_reno
;
829 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
831 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
833 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
836 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
838 struct inet_connection_sock
*icsk
= inet_csk(sk
);
840 if (icsk
->icsk_ca_ops
->set_state
)
841 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
842 icsk
->icsk_ca_state
= ca_state
;
845 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
847 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
849 if (icsk
->icsk_ca_ops
->cwnd_event
)
850 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
853 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
855 * If we receive a SYN packet with these bits set, it means a
856 * network is playing bad games with TOS bits. In order to
857 * avoid possible false congestion notifications, we disable
858 * TCP ECN negociation.
860 * Exception: tcp_ca wants ECN. This is required for DCTCP
861 * congestion control; it requires setting ECT on all packets,
862 * including SYN. We inverse the test in this case: If our
863 * local socket wants ECN, but peer only set ece/cwr (but not
864 * ECT in IP header) its probably a non-DCTCP aware sender.
867 TCP_ECN_create_request(struct request_sock
*req
, const struct sk_buff
*skb
,
868 const struct sock
*listen_sk
)
870 const struct tcphdr
*th
= tcp_hdr(skb
);
871 const struct net
*net
= sock_net(listen_sk
);
872 bool th_ecn
= th
->ece
&& th
->cwr
;
878 ect
= !INET_ECN_is_not_ect(TCP_SKB_CB(skb
)->ip_dsfield
);
879 need_ecn
= tcp_ca_needs_ecn(listen_sk
);
881 if (!ect
&& !need_ecn
&& net
->ipv4
.sysctl_tcp_ecn
)
882 inet_rsk(req
)->ecn_ok
= 1;
883 else if (ect
&& need_ecn
)
884 inet_rsk(req
)->ecn_ok
= 1;
887 /* These functions determine how the current flow behaves in respect of SACK
888 * handling. SACK is negotiated with the peer, and therefore it can vary
889 * between different flows.
891 * tcp_is_sack - SACK enabled
892 * tcp_is_reno - No SACK
893 * tcp_is_fack - FACK enabled, implies SACK enabled
895 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
897 return tp
->rx_opt
.sack_ok
;
900 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
902 return !tcp_is_sack(tp
);
905 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
907 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
910 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
912 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
915 /* TCP early-retransmit (ER) is similar to but more conservative than
916 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
918 static inline void tcp_enable_early_retrans(struct tcp_sock
*tp
)
920 tp
->do_early_retrans
= sysctl_tcp_early_retrans
&&
921 sysctl_tcp_early_retrans
< 4 && !sysctl_tcp_thin_dupack
&&
922 sysctl_tcp_reordering
== 3;
925 static inline void tcp_disable_early_retrans(struct tcp_sock
*tp
)
927 tp
->do_early_retrans
= 0;
930 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
932 return tp
->sacked_out
+ tp
->lost_out
;
935 /* This determines how many packets are "in the network" to the best
936 * of our knowledge. In many cases it is conservative, but where
937 * detailed information is available from the receiver (via SACK
938 * blocks etc.) we can make more aggressive calculations.
940 * Use this for decisions involving congestion control, use just
941 * tp->packets_out to determine if the send queue is empty or not.
943 * Read this equation as:
945 * "Packets sent once on transmission queue" MINUS
946 * "Packets left network, but not honestly ACKed yet" PLUS
947 * "Packets fast retransmitted"
949 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
951 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
954 #define TCP_INFINITE_SSTHRESH 0x7fffffff
956 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
958 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
961 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
963 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
964 (1 << inet_csk(sk
)->icsk_ca_state
);
967 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
968 * The exception is cwnd reduction phase, when cwnd is decreasing towards
971 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
973 const struct tcp_sock
*tp
= tcp_sk(sk
);
975 if (tcp_in_cwnd_reduction(sk
))
976 return tp
->snd_ssthresh
;
978 return max(tp
->snd_ssthresh
,
979 ((tp
->snd_cwnd
>> 1) +
980 (tp
->snd_cwnd
>> 2)));
983 /* Use define here intentionally to get WARN_ON location shown at the caller */
984 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
986 void tcp_enter_cwr(struct sock
*sk
);
987 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
989 /* The maximum number of MSS of available cwnd for which TSO defers
990 * sending if not using sysctl_tcp_tso_win_divisor.
992 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
997 /* Slow start with delack produces 3 packets of burst, so that
998 * it is safe "de facto". This will be the default - same as
999 * the default reordering threshold - but if reordering increases,
1000 * we must be able to allow cwnd to burst at least this much in order
1001 * to not pull it back when holes are filled.
1003 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
1005 return tp
->reordering
;
1008 /* Returns end sequence number of the receiver's advertised window */
1009 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1011 return tp
->snd_una
+ tp
->snd_wnd
;
1014 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1015 * flexible approach. The RFC suggests cwnd should not be raised unless
1016 * it was fully used previously. And that's exactly what we do in
1017 * congestion avoidance mode. But in slow start we allow cwnd to grow
1018 * as long as the application has used half the cwnd.
1020 * cwnd is 10 (IW10), but application sends 9 frames.
1021 * We allow cwnd to reach 18 when all frames are ACKed.
1022 * This check is safe because it's as aggressive as slow start which already
1023 * risks 100% overshoot. The advantage is that we discourage application to
1024 * either send more filler packets or data to artificially blow up the cwnd
1025 * usage, and allow application-limited process to probe bw more aggressively.
1027 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1029 const struct tcp_sock
*tp
= tcp_sk(sk
);
1031 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1032 if (tp
->snd_cwnd
<= tp
->snd_ssthresh
)
1033 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1035 return tp
->is_cwnd_limited
;
1038 static inline void tcp_check_probe_timer(struct sock
*sk
)
1040 const struct tcp_sock
*tp
= tcp_sk(sk
);
1041 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1043 if (!tp
->packets_out
&& !icsk
->icsk_pending
)
1044 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1045 icsk
->icsk_rto
, TCP_RTO_MAX
);
1048 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1053 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1059 * Calculate(/check) TCP checksum
1061 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1062 __be32 daddr
, __wsum base
)
1064 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1067 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1069 return __skb_checksum_complete(skb
);
1072 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1074 return !skb_csum_unnecessary(skb
) &&
1075 __tcp_checksum_complete(skb
);
1078 /* Prequeue for VJ style copy to user, combined with checksumming. */
1080 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
1082 tp
->ucopy
.task
= NULL
;
1084 tp
->ucopy
.memory
= 0;
1085 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1086 #ifdef CONFIG_NET_DMA
1087 tp
->ucopy
.dma_chan
= NULL
;
1088 tp
->ucopy
.wakeup
= 0;
1089 tp
->ucopy
.pinned_list
= NULL
;
1090 tp
->ucopy
.dma_cookie
= 0;
1094 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
);
1099 static const char *statename
[]={
1100 "Unused","Established","Syn Sent","Syn Recv",
1101 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1102 "Close Wait","Last ACK","Listen","Closing"
1105 void tcp_set_state(struct sock
*sk
, int state
);
1107 void tcp_done(struct sock
*sk
);
1109 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1112 rx_opt
->num_sacks
= 0;
1115 u32
tcp_default_init_rwnd(u32 mss
);
1117 /* Determine a window scaling and initial window to offer. */
1118 void tcp_select_initial_window(int __space
, __u32 mss
, __u32
*rcv_wnd
,
1119 __u32
*window_clamp
, int wscale_ok
,
1120 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1122 static inline int tcp_win_from_space(int space
)
1124 return sysctl_tcp_adv_win_scale
<=0 ?
1125 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1126 space
- (space
>>sysctl_tcp_adv_win_scale
);
1129 /* Note: caller must be prepared to deal with negative returns */
1130 static inline int tcp_space(const struct sock
*sk
)
1132 return tcp_win_from_space(sk
->sk_rcvbuf
-
1133 atomic_read(&sk
->sk_rmem_alloc
));
1136 static inline int tcp_full_space(const struct sock
*sk
)
1138 return tcp_win_from_space(sk
->sk_rcvbuf
);
1141 static inline void tcp_openreq_init(struct request_sock
*req
,
1142 struct tcp_options_received
*rx_opt
,
1143 struct sk_buff
*skb
, struct sock
*sk
)
1145 struct inet_request_sock
*ireq
= inet_rsk(req
);
1147 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1149 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1150 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->seq
+ 1;
1151 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1152 req
->mss
= rx_opt
->mss_clamp
;
1153 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1154 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1155 ireq
->sack_ok
= rx_opt
->sack_ok
;
1156 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1157 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1160 ireq
->ir_rmt_port
= tcp_hdr(skb
)->source
;
1161 ireq
->ir_num
= ntohs(tcp_hdr(skb
)->dest
);
1162 ireq
->ir_mark
= inet_request_mark(sk
, skb
);
1165 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1166 struct sock
*sk
, struct dst_entry
*dst
);
1168 void tcp_enter_memory_pressure(struct sock
*sk
);
1170 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1172 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1175 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1177 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1180 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1182 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
1185 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1187 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1189 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
1190 tcp_time_stamp
- tp
->rcv_tstamp
);
1193 static inline int tcp_fin_time(const struct sock
*sk
)
1195 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1196 const int rto
= inet_csk(sk
)->icsk_rto
;
1198 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1199 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1204 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1207 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1209 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1212 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1213 * then following tcp messages have valid values. Ignore 0 value,
1214 * or else 'negative' tsval might forbid us to accept their packets.
1216 if (!rx_opt
->ts_recent
)
1221 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1224 if (tcp_paws_check(rx_opt
, 0))
1227 /* RST segments are not recommended to carry timestamp,
1228 and, if they do, it is recommended to ignore PAWS because
1229 "their cleanup function should take precedence over timestamps."
1230 Certainly, it is mistake. It is necessary to understand the reasons
1231 of this constraint to relax it: if peer reboots, clock may go
1232 out-of-sync and half-open connections will not be reset.
1233 Actually, the problem would be not existing if all
1234 the implementations followed draft about maintaining clock
1235 via reboots. Linux-2.2 DOES NOT!
1237 However, we can relax time bounds for RST segments to MSL.
1239 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1244 static inline void tcp_mib_init(struct net
*net
)
1247 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1248 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1249 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1250 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1254 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1256 tp
->lost_skb_hint
= NULL
;
1259 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1261 tcp_clear_retrans_hints_partial(tp
);
1262 tp
->retransmit_skb_hint
= NULL
;
1268 union tcp_md5_addr
{
1270 #if IS_ENABLED(CONFIG_IPV6)
1275 /* - key database */
1276 struct tcp_md5sig_key
{
1277 struct hlist_node node
;
1279 u8 family
; /* AF_INET or AF_INET6 */
1280 union tcp_md5_addr addr
;
1281 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1282 struct rcu_head rcu
;
1286 struct tcp_md5sig_info
{
1287 struct hlist_head head
;
1288 struct rcu_head rcu
;
1291 /* - pseudo header */
1292 struct tcp4_pseudohdr
{
1300 struct tcp6_pseudohdr
{
1301 struct in6_addr saddr
;
1302 struct in6_addr daddr
;
1304 __be32 protocol
; /* including padding */
1307 union tcp_md5sum_block
{
1308 struct tcp4_pseudohdr ip4
;
1309 #if IS_ENABLED(CONFIG_IPV6)
1310 struct tcp6_pseudohdr ip6
;
1314 /* - pool: digest algorithm, hash description and scratch buffer */
1315 struct tcp_md5sig_pool
{
1316 struct hash_desc md5_desc
;
1317 union tcp_md5sum_block md5_blk
;
1321 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1322 const struct sock
*sk
, const struct request_sock
*req
,
1323 const struct sk_buff
*skb
);
1324 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1325 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
);
1326 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1328 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
1329 struct sock
*addr_sk
);
1331 #ifdef CONFIG_TCP_MD5SIG
1332 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1333 const union tcp_md5_addr
*addr
,
1335 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1337 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1338 const union tcp_md5_addr
*addr
,
1343 #define tcp_twsk_md5_key(twsk) NULL
1346 bool tcp_alloc_md5sig_pool(void);
1348 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1349 static inline void tcp_put_md5sig_pool(void)
1354 int tcp_md5_hash_header(struct tcp_md5sig_pool
*, const struct tcphdr
*);
1355 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1356 unsigned int header_len
);
1357 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1358 const struct tcp_md5sig_key
*key
);
1360 /* From tcp_fastopen.c */
1361 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1362 struct tcp_fastopen_cookie
*cookie
, int *syn_loss
,
1363 unsigned long *last_syn_loss
);
1364 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1365 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
);
1366 struct tcp_fastopen_request
{
1367 /* Fast Open cookie. Size 0 means a cookie request */
1368 struct tcp_fastopen_cookie cookie
;
1369 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1371 int copied
; /* queued in tcp_connect() */
1373 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1375 extern struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
1376 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
);
1377 bool tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1378 struct request_sock
*req
,
1379 struct tcp_fastopen_cookie
*foc
,
1380 struct dst_entry
*dst
);
1381 void tcp_fastopen_init_key_once(bool publish
);
1382 #define TCP_FASTOPEN_KEY_LENGTH 16
1384 /* Fastopen key context */
1385 struct tcp_fastopen_context
{
1386 struct crypto_cipher
*tfm
;
1387 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1388 struct rcu_head rcu
;
1391 /* write queue abstraction */
1392 static inline void tcp_write_queue_purge(struct sock
*sk
)
1394 struct sk_buff
*skb
;
1396 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1397 sk_wmem_free_skb(sk
, skb
);
1399 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1402 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1404 return skb_peek(&sk
->sk_write_queue
);
1407 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1409 return skb_peek_tail(&sk
->sk_write_queue
);
1412 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1413 const struct sk_buff
*skb
)
1415 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1418 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1419 const struct sk_buff
*skb
)
1421 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1424 #define tcp_for_write_queue(skb, sk) \
1425 skb_queue_walk(&(sk)->sk_write_queue, skb)
1427 #define tcp_for_write_queue_from(skb, sk) \
1428 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1430 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1431 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1433 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1435 return sk
->sk_send_head
;
1438 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1439 const struct sk_buff
*skb
)
1441 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1444 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1446 if (tcp_skb_is_last(sk
, skb
))
1447 sk
->sk_send_head
= NULL
;
1449 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1452 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1454 if (sk
->sk_send_head
== skb_unlinked
)
1455 sk
->sk_send_head
= NULL
;
1458 static inline void tcp_init_send_head(struct sock
*sk
)
1460 sk
->sk_send_head
= NULL
;
1463 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1465 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1468 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1470 __tcp_add_write_queue_tail(sk
, skb
);
1472 /* Queue it, remembering where we must start sending. */
1473 if (sk
->sk_send_head
== NULL
) {
1474 sk
->sk_send_head
= skb
;
1476 if (tcp_sk(sk
)->highest_sack
== NULL
)
1477 tcp_sk(sk
)->highest_sack
= skb
;
1481 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1483 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1486 /* Insert buff after skb on the write queue of sk. */
1487 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1488 struct sk_buff
*buff
,
1491 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1494 /* Insert new before skb on the write queue of sk. */
1495 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1496 struct sk_buff
*skb
,
1499 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1501 if (sk
->sk_send_head
== skb
)
1502 sk
->sk_send_head
= new;
1505 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1507 __skb_unlink(skb
, &sk
->sk_write_queue
);
1510 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1512 return skb_queue_empty(&sk
->sk_write_queue
);
1515 static inline void tcp_push_pending_frames(struct sock
*sk
)
1517 if (tcp_send_head(sk
)) {
1518 struct tcp_sock
*tp
= tcp_sk(sk
);
1520 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1524 /* Start sequence of the skb just after the highest skb with SACKed
1525 * bit, valid only if sacked_out > 0 or when the caller has ensured
1526 * validity by itself.
1528 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1530 if (!tp
->sacked_out
)
1533 if (tp
->highest_sack
== NULL
)
1536 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1539 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1541 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1542 tcp_write_queue_next(sk
, skb
);
1545 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1547 return tcp_sk(sk
)->highest_sack
;
1550 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1552 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1555 /* Called when old skb is about to be deleted (to be combined with new skb) */
1556 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1557 struct sk_buff
*old
,
1558 struct sk_buff
*new)
1560 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1561 tcp_sk(sk
)->highest_sack
= new;
1564 /* Determines whether this is a thin stream (which may suffer from
1565 * increased latency). Used to trigger latency-reducing mechanisms.
1567 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1569 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1573 enum tcp_seq_states
{
1574 TCP_SEQ_STATE_LISTENING
,
1575 TCP_SEQ_STATE_OPENREQ
,
1576 TCP_SEQ_STATE_ESTABLISHED
,
1579 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1581 struct tcp_seq_afinfo
{
1584 const struct file_operations
*seq_fops
;
1585 struct seq_operations seq_ops
;
1588 struct tcp_iter_state
{
1589 struct seq_net_private p
;
1591 enum tcp_seq_states state
;
1592 struct sock
*syn_wait_sk
;
1593 int bucket
, offset
, sbucket
, num
;
1598 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1599 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1601 extern struct request_sock_ops tcp_request_sock_ops
;
1602 extern struct request_sock_ops tcp6_request_sock_ops
;
1604 void tcp_v4_destroy_sock(struct sock
*sk
);
1606 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1607 netdev_features_t features
);
1608 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1609 int tcp_gro_complete(struct sk_buff
*skb
);
1611 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1613 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1615 return tp
->notsent_lowat
?: sysctl_tcp_notsent_lowat
;
1618 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1620 const struct tcp_sock
*tp
= tcp_sk(sk
);
1621 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1623 return notsent_bytes
< tcp_notsent_lowat(tp
);
1626 #ifdef CONFIG_PROC_FS
1627 int tcp4_proc_init(void);
1628 void tcp4_proc_exit(void);
1631 int tcp_rtx_synack(struct sock
*sk
, struct request_sock
*req
);
1632 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1633 const struct tcp_request_sock_ops
*af_ops
,
1634 struct sock
*sk
, struct sk_buff
*skb
);
1636 /* TCP af-specific functions */
1637 struct tcp_sock_af_ops
{
1638 #ifdef CONFIG_TCP_MD5SIG
1639 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1640 struct sock
*addr_sk
);
1641 int (*calc_md5_hash
) (char *location
,
1642 struct tcp_md5sig_key
*md5
,
1643 const struct sock
*sk
,
1644 const struct request_sock
*req
,
1645 const struct sk_buff
*skb
);
1646 int (*md5_parse
) (struct sock
*sk
,
1647 char __user
*optval
,
1652 struct tcp_request_sock_ops
{
1654 #ifdef CONFIG_TCP_MD5SIG
1655 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1656 struct request_sock
*req
);
1657 int (*calc_md5_hash
) (char *location
,
1658 struct tcp_md5sig_key
*md5
,
1659 const struct sock
*sk
,
1660 const struct request_sock
*req
,
1661 const struct sk_buff
*skb
);
1663 void (*init_req
)(struct request_sock
*req
, struct sock
*sk
,
1664 struct sk_buff
*skb
);
1665 #ifdef CONFIG_SYN_COOKIES
1666 __u32 (*cookie_init_seq
)(struct sock
*sk
, const struct sk_buff
*skb
,
1669 struct dst_entry
*(*route_req
)(struct sock
*sk
, struct flowi
*fl
,
1670 const struct request_sock
*req
,
1672 __u32 (*init_seq
)(const struct sk_buff
*skb
);
1673 int (*send_synack
)(struct sock
*sk
, struct dst_entry
*dst
,
1674 struct flowi
*fl
, struct request_sock
*req
,
1675 u16 queue_mapping
, struct tcp_fastopen_cookie
*foc
);
1676 void (*queue_hash_add
)(struct sock
*sk
, struct request_sock
*req
,
1677 const unsigned long timeout
);
1680 #ifdef CONFIG_SYN_COOKIES
1681 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1682 struct sock
*sk
, struct sk_buff
*skb
,
1685 return ops
->cookie_init_seq(sk
, skb
, mss
);
1688 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1689 struct sock
*sk
, struct sk_buff
*skb
,
1696 int tcpv4_offload_init(void);
1698 void tcp_v4_init(void);
1699 void tcp_init(void);