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 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 long sysctl_tcp_mem
[3];
263 extern int sysctl_tcp_wmem
[3];
264 extern int sysctl_tcp_rmem
[3];
265 extern int sysctl_tcp_app_win
;
266 extern int sysctl_tcp_adv_win_scale
;
267 extern int sysctl_tcp_tw_reuse
;
268 extern int sysctl_tcp_frto
;
269 extern int sysctl_tcp_low_latency
;
270 extern int sysctl_tcp_dma_copybreak
;
271 extern int sysctl_tcp_nometrics_save
;
272 extern int sysctl_tcp_moderate_rcvbuf
;
273 extern int sysctl_tcp_tso_win_divisor
;
274 extern int sysctl_tcp_mtu_probing
;
275 extern int sysctl_tcp_base_mss
;
276 extern int sysctl_tcp_workaround_signed_windows
;
277 extern int sysctl_tcp_slow_start_after_idle
;
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
;
285 extern int sysctl_tcp_autocorking
;
287 extern atomic_long_t tcp_memory_allocated
;
288 extern struct percpu_counter tcp_sockets_allocated
;
289 extern int tcp_memory_pressure
;
292 * The next routines deal with comparing 32 bit unsigned ints
293 * and worry about wraparound (automatic with unsigned arithmetic).
296 static inline bool before(__u32 seq1
, __u32 seq2
)
298 return (__s32
)(seq1
-seq2
) < 0;
300 #define after(seq2, seq1) before(seq1, seq2)
302 /* is s2<=s1<=s3 ? */
303 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
305 return seq3
- seq2
>= seq1
- seq2
;
308 static inline bool tcp_out_of_memory(struct sock
*sk
)
310 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
311 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
316 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
318 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
319 int orphans
= percpu_counter_read_positive(ocp
);
321 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
322 orphans
= percpu_counter_sum_positive(ocp
);
323 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
329 bool tcp_check_oom(struct sock
*sk
, int shift
);
331 /* syncookies: remember time of last synqueue overflow */
332 static inline void tcp_synq_overflow(struct sock
*sk
)
334 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= jiffies
;
337 /* syncookies: no recent synqueue overflow on this listening socket? */
338 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
340 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
341 return time_after(jiffies
, last_overflow
+ TCP_TIMEOUT_FALLBACK
);
344 extern struct proto tcp_prot
;
346 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
347 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
348 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
349 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
350 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
352 void tcp_tasklet_init(void);
354 void tcp_v4_err(struct sk_buff
*skb
, u32
);
356 void tcp_shutdown(struct sock
*sk
, int how
);
358 void tcp_v4_early_demux(struct sk_buff
*skb
);
359 int tcp_v4_rcv(struct sk_buff
*skb
);
361 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
362 int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
364 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
366 void tcp_release_cb(struct sock
*sk
);
367 void tcp_wfree(struct sk_buff
*skb
);
368 void tcp_write_timer_handler(struct sock
*sk
);
369 void tcp_delack_timer_handler(struct sock
*sk
);
370 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
371 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
372 const struct tcphdr
*th
, unsigned int len
);
373 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
374 const struct tcphdr
*th
, unsigned int len
);
375 void tcp_rcv_space_adjust(struct sock
*sk
);
376 void tcp_cleanup_rbuf(struct sock
*sk
, int copied
);
377 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
378 void tcp_twsk_destructor(struct sock
*sk
);
379 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
380 struct pipe_inode_info
*pipe
, size_t len
,
383 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
384 const unsigned int pkts
)
386 struct inet_connection_sock
*icsk
= inet_csk(sk
);
388 if (icsk
->icsk_ack
.quick
) {
389 if (pkts
>= icsk
->icsk_ack
.quick
) {
390 icsk
->icsk_ack
.quick
= 0;
391 /* Leaving quickack mode we deflate ATO. */
392 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
394 icsk
->icsk_ack
.quick
-= pkts
;
399 #define TCP_ECN_QUEUE_CWR 2
400 #define TCP_ECN_DEMAND_CWR 4
401 #define TCP_ECN_SEEN 8
411 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
413 const struct tcphdr
*th
);
414 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
415 struct request_sock
*req
, struct request_sock
**prev
,
417 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
418 struct sk_buff
*skb
);
419 void tcp_enter_loss(struct sock
*sk
, int how
);
420 void tcp_clear_retrans(struct tcp_sock
*tp
);
421 void tcp_update_metrics(struct sock
*sk
);
422 void tcp_init_metrics(struct sock
*sk
);
423 void tcp_metrics_init(void);
424 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
,
426 bool tcp_remember_stamp(struct sock
*sk
);
427 bool tcp_tw_remember_stamp(struct inet_timewait_sock
*tw
);
428 void tcp_fetch_timewait_stamp(struct sock
*sk
, struct dst_entry
*dst
);
429 void tcp_disable_fack(struct tcp_sock
*tp
);
430 void tcp_close(struct sock
*sk
, long timeout
);
431 void tcp_init_sock(struct sock
*sk
);
432 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
,
433 struct poll_table_struct
*wait
);
434 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
435 char __user
*optval
, int __user
*optlen
);
436 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
437 char __user
*optval
, unsigned int optlen
);
438 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
439 char __user
*optval
, int __user
*optlen
);
440 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
441 char __user
*optval
, unsigned int optlen
);
442 void tcp_set_keepalive(struct sock
*sk
, int val
);
443 void tcp_syn_ack_timeout(struct sock
*sk
, struct request_sock
*req
);
444 int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
445 size_t len
, int nonblock
, int flags
, int *addr_len
);
446 void tcp_parse_options(const struct sk_buff
*skb
,
447 struct tcp_options_received
*opt_rx
,
448 int estab
, struct tcp_fastopen_cookie
*foc
);
449 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
452 * TCP v4 functions exported for the inet6 API
455 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
456 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
457 struct sock
*tcp_create_openreq_child(struct sock
*sk
,
458 struct request_sock
*req
,
459 struct sk_buff
*skb
);
460 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
461 struct request_sock
*req
,
462 struct dst_entry
*dst
);
463 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
464 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
465 int tcp_connect(struct sock
*sk
);
466 struct sk_buff
*tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
467 struct request_sock
*req
,
468 struct tcp_fastopen_cookie
*foc
);
469 int tcp_disconnect(struct sock
*sk
, int flags
);
471 void tcp_connect_init(struct sock
*sk
);
472 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
473 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
474 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
476 /* From syncookies.c */
477 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
479 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
480 struct ip_options
*opt
);
481 #ifdef CONFIG_SYN_COOKIES
482 #include <linux/ktime.h>
484 /* Syncookies use a monotonic timer which increments every 64 seconds.
485 * This counter is used both as a hash input and partially encoded into
486 * the cookie value. A cookie is only validated further if the delta
487 * between the current counter value and the encoded one is less than this,
488 * i.e. a sent cookie is valid only at most for 128 seconds (or less if
489 * the counter advances immediately after a cookie is generated).
491 #define MAX_SYNCOOKIE_AGE 2
493 static inline u32
tcp_cookie_time(void)
496 getnstimeofday(&now
);
497 return now
.tv_sec
>> 6; /* 64 seconds granularity */
500 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
502 __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
, __u16
*mss
);
504 static inline __u32
cookie_v4_init_sequence(struct sock
*sk
,
512 __u32
cookie_init_timestamp(struct request_sock
*req
);
513 bool cookie_check_timestamp(struct tcp_options_received
*opt
, struct net
*net
,
516 /* From net/ipv6/syncookies.c */
517 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
519 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
520 #ifdef CONFIG_SYN_COOKIES
521 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
522 const struct tcphdr
*th
, u16
*mssp
);
523 __u32
cookie_v6_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
526 static inline __u32
cookie_v6_init_sequence(struct sock
*sk
,
535 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
537 bool tcp_may_send_now(struct sock
*sk
);
538 int __tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
539 int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
540 void tcp_retransmit_timer(struct sock
*sk
);
541 void tcp_xmit_retransmit_queue(struct sock
*);
542 void tcp_simple_retransmit(struct sock
*);
543 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
544 int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int);
546 void tcp_send_probe0(struct sock
*);
547 void tcp_send_partial(struct sock
*);
548 int tcp_write_wakeup(struct sock
*);
549 void tcp_send_fin(struct sock
*sk
);
550 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
551 int tcp_send_synack(struct sock
*);
552 bool tcp_syn_flood_action(struct sock
*sk
, const struct sk_buff
*skb
,
554 void tcp_push_one(struct sock
*, unsigned int mss_now
);
555 void tcp_send_ack(struct sock
*sk
);
556 void tcp_send_delayed_ack(struct sock
*sk
);
557 void tcp_send_loss_probe(struct sock
*sk
);
558 bool tcp_schedule_loss_probe(struct sock
*sk
);
561 void tcp_cwnd_application_limited(struct sock
*sk
);
562 void tcp_resume_early_retransmit(struct sock
*sk
);
563 void tcp_rearm_rto(struct sock
*sk
);
564 void tcp_reset(struct sock
*sk
);
567 void tcp_init_xmit_timers(struct sock
*);
568 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
570 inet_csk_clear_xmit_timers(sk
);
573 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
574 unsigned int tcp_current_mss(struct sock
*sk
);
576 /* Bound MSS / TSO packet size with the half of the window */
577 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
581 /* When peer uses tiny windows, there is no use in packetizing
582 * to sub-MSS pieces for the sake of SWS or making sure there
583 * are enough packets in the pipe for fast recovery.
585 * On the other hand, for extremely large MSS devices, handling
586 * smaller than MSS windows in this way does make sense.
588 if (tp
->max_window
>= 512)
589 cutoff
= (tp
->max_window
>> 1);
591 cutoff
= tp
->max_window
;
593 if (cutoff
&& pktsize
> cutoff
)
594 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
600 void tcp_get_info(const struct sock
*, struct tcp_info
*);
602 /* Read 'sendfile()'-style from a TCP socket */
603 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
604 unsigned int, size_t);
605 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
606 sk_read_actor_t recv_actor
);
608 void tcp_initialize_rcv_mss(struct sock
*sk
);
610 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
611 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
612 void tcp_mtup_init(struct sock
*sk
);
613 void tcp_init_buffer_space(struct sock
*sk
);
615 static inline void tcp_bound_rto(const struct sock
*sk
)
617 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
618 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
621 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
623 return (tp
->srtt
>> 3) + tp
->rttvar
;
626 void tcp_set_rto(struct sock
*sk
);
628 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
630 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
631 ntohl(TCP_FLAG_ACK
) |
635 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
637 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
640 static inline void tcp_fast_path_check(struct sock
*sk
)
642 struct tcp_sock
*tp
= tcp_sk(sk
);
644 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
646 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
648 tcp_fast_path_on(tp
);
651 /* Compute the actual rto_min value */
652 static inline u32
tcp_rto_min(struct sock
*sk
)
654 const struct dst_entry
*dst
= __sk_dst_get(sk
);
655 u32 rto_min
= TCP_RTO_MIN
;
657 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
658 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
662 /* Compute the actual receive window we are currently advertising.
663 * Rcv_nxt can be after the window if our peer push more data
664 * than the offered window.
666 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
668 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
675 /* Choose a new window, without checks for shrinking, and without
676 * scaling applied to the result. The caller does these things
677 * if necessary. This is a "raw" window selection.
679 u32
__tcp_select_window(struct sock
*sk
);
681 void tcp_send_window_probe(struct sock
*sk
);
683 /* TCP timestamps are only 32-bits, this causes a slight
684 * complication on 64-bit systems since we store a snapshot
685 * of jiffies in the buffer control blocks below. We decided
686 * to use only the low 32-bits of jiffies and hide the ugly
687 * casts with the following macro.
689 #define tcp_time_stamp ((__u32)(jiffies))
691 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
693 #define TCPHDR_FIN 0x01
694 #define TCPHDR_SYN 0x02
695 #define TCPHDR_RST 0x04
696 #define TCPHDR_PSH 0x08
697 #define TCPHDR_ACK 0x10
698 #define TCPHDR_URG 0x20
699 #define TCPHDR_ECE 0x40
700 #define TCPHDR_CWR 0x80
702 /* This is what the send packet queuing engine uses to pass
703 * TCP per-packet control information to the transmission code.
704 * We also store the host-order sequence numbers in here too.
705 * This is 44 bytes if IPV6 is enabled.
706 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
710 struct inet_skb_parm h4
;
711 #if IS_ENABLED(CONFIG_IPV6)
712 struct inet6_skb_parm h6
;
714 } header
; /* For incoming frames */
715 __u32 seq
; /* Starting sequence number */
716 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
717 __u32 when
; /* used to compute rtt's */
718 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
720 __u8 sacked
; /* State flags for SACK/FACK. */
721 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
722 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
723 #define TCPCB_LOST 0x04 /* SKB is lost */
724 #define TCPCB_TAGBITS 0x07 /* All tag bits */
725 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
726 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
728 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
730 __u32 ack_seq
; /* Sequence number ACK'd */
733 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
735 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
737 * If we receive a SYN packet with these bits set, it means a network is
738 * playing bad games with TOS bits. In order to avoid possible false congestion
739 * notifications, we disable TCP ECN negociation.
742 TCP_ECN_create_request(struct request_sock
*req
, const struct sk_buff
*skb
,
745 const struct tcphdr
*th
= tcp_hdr(skb
);
747 if (net
->ipv4
.sysctl_tcp_ecn
&& th
->ece
&& th
->cwr
&&
748 INET_ECN_is_not_ect(TCP_SKB_CB(skb
)->ip_dsfield
))
749 inet_rsk(req
)->ecn_ok
= 1;
752 /* Due to TSO, an SKB can be composed of multiple actual
753 * packets. To keep these tracked properly, we use this.
755 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
757 return skb_shinfo(skb
)->gso_segs
;
760 /* This is valid iff tcp_skb_pcount() > 1. */
761 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
763 return skb_shinfo(skb
)->gso_size
;
766 /* Events passed to congestion control interface */
768 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
769 CA_EVENT_CWND_RESTART
, /* congestion window restart */
770 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
771 CA_EVENT_LOSS
, /* loss timeout */
772 CA_EVENT_FAST_ACK
, /* in sequence ack */
773 CA_EVENT_SLOW_ACK
, /* other ack */
777 * Interface for adding new TCP congestion control handlers
779 #define TCP_CA_NAME_MAX 16
780 #define TCP_CA_MAX 128
781 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
783 #define TCP_CONG_NON_RESTRICTED 0x1
784 #define TCP_CONG_RTT_STAMP 0x2
786 struct tcp_congestion_ops
{
787 struct list_head list
;
790 /* initialize private data (optional) */
791 void (*init
)(struct sock
*sk
);
792 /* cleanup private data (optional) */
793 void (*release
)(struct sock
*sk
);
795 /* return slow start threshold (required) */
796 u32 (*ssthresh
)(struct sock
*sk
);
797 /* lower bound for congestion window (optional) */
798 u32 (*min_cwnd
)(const struct sock
*sk
);
799 /* do new cwnd calculation (required) */
800 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
, u32 in_flight
);
801 /* call before changing ca_state (optional) */
802 void (*set_state
)(struct sock
*sk
, u8 new_state
);
803 /* call when cwnd event occurs (optional) */
804 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
805 /* new value of cwnd after loss (optional) */
806 u32 (*undo_cwnd
)(struct sock
*sk
);
807 /* hook for packet ack accounting (optional) */
808 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
809 /* get info for inet_diag (optional) */
810 void (*get_info
)(struct sock
*sk
, u32 ext
, struct sk_buff
*skb
);
812 char name
[TCP_CA_NAME_MAX
];
813 struct module
*owner
;
816 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
817 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
819 void tcp_init_congestion_control(struct sock
*sk
);
820 void tcp_cleanup_congestion_control(struct sock
*sk
);
821 int tcp_set_default_congestion_control(const char *name
);
822 void tcp_get_default_congestion_control(char *name
);
823 void tcp_get_available_congestion_control(char *buf
, size_t len
);
824 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
825 int tcp_set_allowed_congestion_control(char *allowed
);
826 int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
827 int tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
828 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
);
830 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
831 u32
tcp_reno_ssthresh(struct sock
*sk
);
832 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
, u32 in_flight
);
833 u32
tcp_reno_min_cwnd(const struct sock
*sk
);
834 extern struct tcp_congestion_ops tcp_reno
;
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 /* These functions determine how the current flow behaves in respect of SACK
854 * handling. SACK is negotiated with the peer, and therefore it can vary
855 * between different flows.
857 * tcp_is_sack - SACK enabled
858 * tcp_is_reno - No SACK
859 * tcp_is_fack - FACK enabled, implies SACK enabled
861 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
863 return tp
->rx_opt
.sack_ok
;
866 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
868 return !tcp_is_sack(tp
);
871 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
873 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
876 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
878 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
881 /* TCP early-retransmit (ER) is similar to but more conservative than
882 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
884 static inline void tcp_enable_early_retrans(struct tcp_sock
*tp
)
886 tp
->do_early_retrans
= sysctl_tcp_early_retrans
&&
887 sysctl_tcp_early_retrans
< 4 && !sysctl_tcp_thin_dupack
&&
888 sysctl_tcp_reordering
== 3;
891 static inline void tcp_disable_early_retrans(struct tcp_sock
*tp
)
893 tp
->do_early_retrans
= 0;
896 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
898 return tp
->sacked_out
+ tp
->lost_out
;
901 /* This determines how many packets are "in the network" to the best
902 * of our knowledge. In many cases it is conservative, but where
903 * detailed information is available from the receiver (via SACK
904 * blocks etc.) we can make more aggressive calculations.
906 * Use this for decisions involving congestion control, use just
907 * tp->packets_out to determine if the send queue is empty or not.
909 * Read this equation as:
911 * "Packets sent once on transmission queue" MINUS
912 * "Packets left network, but not honestly ACKed yet" PLUS
913 * "Packets fast retransmitted"
915 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
917 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
920 #define TCP_INFINITE_SSTHRESH 0x7fffffff
922 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
924 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
927 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
929 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
930 (1 << inet_csk(sk
)->icsk_ca_state
);
933 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
934 * The exception is cwnd reduction phase, when cwnd is decreasing towards
937 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
939 const struct tcp_sock
*tp
= tcp_sk(sk
);
941 if (tcp_in_cwnd_reduction(sk
))
942 return tp
->snd_ssthresh
;
944 return max(tp
->snd_ssthresh
,
945 ((tp
->snd_cwnd
>> 1) +
946 (tp
->snd_cwnd
>> 2)));
949 /* Use define here intentionally to get WARN_ON location shown at the caller */
950 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
952 void tcp_enter_cwr(struct sock
*sk
, const int set_ssthresh
);
953 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
955 /* The maximum number of MSS of available cwnd for which TSO defers
956 * sending if not using sysctl_tcp_tso_win_divisor.
958 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
963 /* Slow start with delack produces 3 packets of burst, so that
964 * it is safe "de facto". This will be the default - same as
965 * the default reordering threshold - but if reordering increases,
966 * we must be able to allow cwnd to burst at least this much in order
967 * to not pull it back when holes are filled.
969 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
971 return tp
->reordering
;
974 /* Returns end sequence number of the receiver's advertised window */
975 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
977 return tp
->snd_una
+ tp
->snd_wnd
;
979 bool tcp_is_cwnd_limited(const struct sock
*sk
, u32 in_flight
);
981 static inline void tcp_minshall_update(struct tcp_sock
*tp
, unsigned int mss
,
982 const struct sk_buff
*skb
)
985 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
988 static inline void tcp_check_probe_timer(struct sock
*sk
)
990 const struct tcp_sock
*tp
= tcp_sk(sk
);
991 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
993 if (!tp
->packets_out
&& !icsk
->icsk_pending
)
994 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
995 icsk
->icsk_rto
, TCP_RTO_MAX
);
998 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1003 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1009 * Calculate(/check) TCP checksum
1011 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1012 __be32 daddr
, __wsum base
)
1014 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1017 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1019 return __skb_checksum_complete(skb
);
1022 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1024 return !skb_csum_unnecessary(skb
) &&
1025 __tcp_checksum_complete(skb
);
1028 /* Prequeue for VJ style copy to user, combined with checksumming. */
1030 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
1032 tp
->ucopy
.task
= NULL
;
1034 tp
->ucopy
.memory
= 0;
1035 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1036 #ifdef CONFIG_NET_DMA
1037 tp
->ucopy
.dma_chan
= NULL
;
1038 tp
->ucopy
.wakeup
= 0;
1039 tp
->ucopy
.pinned_list
= NULL
;
1040 tp
->ucopy
.dma_cookie
= 0;
1044 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
);
1049 static const char *statename
[]={
1050 "Unused","Established","Syn Sent","Syn Recv",
1051 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1052 "Close Wait","Last ACK","Listen","Closing"
1055 void tcp_set_state(struct sock
*sk
, int state
);
1057 void tcp_done(struct sock
*sk
);
1059 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1062 rx_opt
->num_sacks
= 0;
1065 u32
tcp_default_init_rwnd(u32 mss
);
1067 /* Determine a window scaling and initial window to offer. */
1068 void tcp_select_initial_window(int __space
, __u32 mss
, __u32
*rcv_wnd
,
1069 __u32
*window_clamp
, int wscale_ok
,
1070 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1072 static inline int tcp_win_from_space(int space
)
1074 return sysctl_tcp_adv_win_scale
<=0 ?
1075 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1076 space
- (space
>>sysctl_tcp_adv_win_scale
);
1079 /* Note: caller must be prepared to deal with negative returns */
1080 static inline int tcp_space(const struct sock
*sk
)
1082 return tcp_win_from_space(sk
->sk_rcvbuf
-
1083 atomic_read(&sk
->sk_rmem_alloc
));
1086 static inline int tcp_full_space(const struct sock
*sk
)
1088 return tcp_win_from_space(sk
->sk_rcvbuf
);
1091 static inline void tcp_openreq_init(struct request_sock
*req
,
1092 struct tcp_options_received
*rx_opt
,
1093 struct sk_buff
*skb
)
1095 struct inet_request_sock
*ireq
= inet_rsk(req
);
1097 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1099 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1100 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->seq
+ 1;
1101 tcp_rsk(req
)->snt_synack
= 0;
1102 req
->mss
= rx_opt
->mss_clamp
;
1103 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1104 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1105 ireq
->sack_ok
= rx_opt
->sack_ok
;
1106 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1107 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1110 ireq
->ir_rmt_port
= tcp_hdr(skb
)->source
;
1111 ireq
->ir_num
= ntohs(tcp_hdr(skb
)->dest
);
1114 void tcp_enter_memory_pressure(struct sock
*sk
);
1116 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1118 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1121 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1123 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1126 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1128 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
1131 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1133 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1135 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
1136 tcp_time_stamp
- tp
->rcv_tstamp
);
1139 static inline int tcp_fin_time(const struct sock
*sk
)
1141 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1142 const int rto
= inet_csk(sk
)->icsk_rto
;
1144 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1145 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1150 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1153 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1155 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1158 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1159 * then following tcp messages have valid values. Ignore 0 value,
1160 * or else 'negative' tsval might forbid us to accept their packets.
1162 if (!rx_opt
->ts_recent
)
1167 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1170 if (tcp_paws_check(rx_opt
, 0))
1173 /* RST segments are not recommended to carry timestamp,
1174 and, if they do, it is recommended to ignore PAWS because
1175 "their cleanup function should take precedence over timestamps."
1176 Certainly, it is mistake. It is necessary to understand the reasons
1177 of this constraint to relax it: if peer reboots, clock may go
1178 out-of-sync and half-open connections will not be reset.
1179 Actually, the problem would be not existing if all
1180 the implementations followed draft about maintaining clock
1181 via reboots. Linux-2.2 DOES NOT!
1183 However, we can relax time bounds for RST segments to MSL.
1185 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1190 static inline void tcp_mib_init(struct net
*net
)
1193 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1194 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1195 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1196 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1200 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1202 tp
->lost_skb_hint
= NULL
;
1205 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1207 tcp_clear_retrans_hints_partial(tp
);
1208 tp
->retransmit_skb_hint
= NULL
;
1214 union tcp_md5_addr
{
1216 #if IS_ENABLED(CONFIG_IPV6)
1221 /* - key database */
1222 struct tcp_md5sig_key
{
1223 struct hlist_node node
;
1225 u8 family
; /* AF_INET or AF_INET6 */
1226 union tcp_md5_addr addr
;
1227 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1228 struct rcu_head rcu
;
1232 struct tcp_md5sig_info
{
1233 struct hlist_head head
;
1234 struct rcu_head rcu
;
1237 /* - pseudo header */
1238 struct tcp4_pseudohdr
{
1246 struct tcp6_pseudohdr
{
1247 struct in6_addr saddr
;
1248 struct in6_addr daddr
;
1250 __be32 protocol
; /* including padding */
1253 union tcp_md5sum_block
{
1254 struct tcp4_pseudohdr ip4
;
1255 #if IS_ENABLED(CONFIG_IPV6)
1256 struct tcp6_pseudohdr ip6
;
1260 /* - pool: digest algorithm, hash description and scratch buffer */
1261 struct tcp_md5sig_pool
{
1262 struct hash_desc md5_desc
;
1263 union tcp_md5sum_block md5_blk
;
1267 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1268 const struct sock
*sk
, const struct request_sock
*req
,
1269 const struct sk_buff
*skb
);
1270 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1271 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
);
1272 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1274 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
1275 struct sock
*addr_sk
);
1277 #ifdef CONFIG_TCP_MD5SIG
1278 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1279 const union tcp_md5_addr
*addr
,
1281 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1283 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1284 const union tcp_md5_addr
*addr
,
1289 #define tcp_twsk_md5_key(twsk) NULL
1292 bool tcp_alloc_md5sig_pool(void);
1294 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1295 static inline void tcp_put_md5sig_pool(void)
1300 int tcp_md5_hash_header(struct tcp_md5sig_pool
*, const struct tcphdr
*);
1301 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1302 unsigned int header_len
);
1303 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1304 const struct tcp_md5sig_key
*key
);
1306 /* From tcp_fastopen.c */
1307 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1308 struct tcp_fastopen_cookie
*cookie
, int *syn_loss
,
1309 unsigned long *last_syn_loss
);
1310 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1311 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
);
1312 struct tcp_fastopen_request
{
1313 /* Fast Open cookie. Size 0 means a cookie request */
1314 struct tcp_fastopen_cookie cookie
;
1315 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1316 u16 copied
; /* queued in tcp_connect() */
1318 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1320 extern struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
1321 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
);
1322 void tcp_fastopen_cookie_gen(__be32 src
, __be32 dst
,
1323 struct tcp_fastopen_cookie
*foc
);
1324 void tcp_fastopen_init_key_once(bool publish
);
1325 #define TCP_FASTOPEN_KEY_LENGTH 16
1327 /* Fastopen key context */
1328 struct tcp_fastopen_context
{
1329 struct crypto_cipher
*tfm
;
1330 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1331 struct rcu_head rcu
;
1334 /* write queue abstraction */
1335 static inline void tcp_write_queue_purge(struct sock
*sk
)
1337 struct sk_buff
*skb
;
1339 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1340 sk_wmem_free_skb(sk
, skb
);
1342 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1345 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1347 return skb_peek(&sk
->sk_write_queue
);
1350 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1352 return skb_peek_tail(&sk
->sk_write_queue
);
1355 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1356 const struct sk_buff
*skb
)
1358 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1361 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1362 const struct sk_buff
*skb
)
1364 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1367 #define tcp_for_write_queue(skb, sk) \
1368 skb_queue_walk(&(sk)->sk_write_queue, skb)
1370 #define tcp_for_write_queue_from(skb, sk) \
1371 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1373 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1374 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1376 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1378 return sk
->sk_send_head
;
1381 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1382 const struct sk_buff
*skb
)
1384 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1387 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1389 if (tcp_skb_is_last(sk
, skb
))
1390 sk
->sk_send_head
= NULL
;
1392 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1395 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1397 if (sk
->sk_send_head
== skb_unlinked
)
1398 sk
->sk_send_head
= NULL
;
1401 static inline void tcp_init_send_head(struct sock
*sk
)
1403 sk
->sk_send_head
= NULL
;
1406 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1408 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1411 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1413 __tcp_add_write_queue_tail(sk
, skb
);
1415 /* Queue it, remembering where we must start sending. */
1416 if (sk
->sk_send_head
== NULL
) {
1417 sk
->sk_send_head
= skb
;
1419 if (tcp_sk(sk
)->highest_sack
== NULL
)
1420 tcp_sk(sk
)->highest_sack
= skb
;
1424 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1426 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1429 /* Insert buff after skb on the write queue of sk. */
1430 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1431 struct sk_buff
*buff
,
1434 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1437 /* Insert new before skb on the write queue of sk. */
1438 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1439 struct sk_buff
*skb
,
1442 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1444 if (sk
->sk_send_head
== skb
)
1445 sk
->sk_send_head
= new;
1448 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1450 __skb_unlink(skb
, &sk
->sk_write_queue
);
1453 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1455 return skb_queue_empty(&sk
->sk_write_queue
);
1458 static inline void tcp_push_pending_frames(struct sock
*sk
)
1460 if (tcp_send_head(sk
)) {
1461 struct tcp_sock
*tp
= tcp_sk(sk
);
1463 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1467 /* Start sequence of the skb just after the highest skb with SACKed
1468 * bit, valid only if sacked_out > 0 or when the caller has ensured
1469 * validity by itself.
1471 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1473 if (!tp
->sacked_out
)
1476 if (tp
->highest_sack
== NULL
)
1479 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1482 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1484 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1485 tcp_write_queue_next(sk
, skb
);
1488 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1490 return tcp_sk(sk
)->highest_sack
;
1493 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1495 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1498 /* Called when old skb is about to be deleted (to be combined with new skb) */
1499 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1500 struct sk_buff
*old
,
1501 struct sk_buff
*new)
1503 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1504 tcp_sk(sk
)->highest_sack
= new;
1507 /* Determines whether this is a thin stream (which may suffer from
1508 * increased latency). Used to trigger latency-reducing mechanisms.
1510 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1512 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1516 enum tcp_seq_states
{
1517 TCP_SEQ_STATE_LISTENING
,
1518 TCP_SEQ_STATE_OPENREQ
,
1519 TCP_SEQ_STATE_ESTABLISHED
,
1522 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1524 struct tcp_seq_afinfo
{
1527 const struct file_operations
*seq_fops
;
1528 struct seq_operations seq_ops
;
1531 struct tcp_iter_state
{
1532 struct seq_net_private p
;
1534 enum tcp_seq_states state
;
1535 struct sock
*syn_wait_sk
;
1536 int bucket
, offset
, sbucket
, num
;
1541 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1542 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1544 extern struct request_sock_ops tcp_request_sock_ops
;
1545 extern struct request_sock_ops tcp6_request_sock_ops
;
1547 void tcp_v4_destroy_sock(struct sock
*sk
);
1549 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1550 netdev_features_t features
);
1551 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1552 int tcp_gro_complete(struct sk_buff
*skb
);
1554 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1556 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1558 return tp
->notsent_lowat
?: sysctl_tcp_notsent_lowat
;
1561 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1563 const struct tcp_sock
*tp
= tcp_sk(sk
);
1564 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1566 return notsent_bytes
< tcp_notsent_lowat(tp
);
1569 #ifdef CONFIG_PROC_FS
1570 int tcp4_proc_init(void);
1571 void tcp4_proc_exit(void);
1574 /* TCP af-specific functions */
1575 struct tcp_sock_af_ops
{
1576 #ifdef CONFIG_TCP_MD5SIG
1577 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1578 struct sock
*addr_sk
);
1579 int (*calc_md5_hash
) (char *location
,
1580 struct tcp_md5sig_key
*md5
,
1581 const struct sock
*sk
,
1582 const struct request_sock
*req
,
1583 const struct sk_buff
*skb
);
1584 int (*md5_parse
) (struct sock
*sk
,
1585 char __user
*optval
,
1590 struct tcp_request_sock_ops
{
1591 #ifdef CONFIG_TCP_MD5SIG
1592 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1593 struct request_sock
*req
);
1594 int (*calc_md5_hash
) (char *location
,
1595 struct tcp_md5sig_key
*md5
,
1596 const struct sock
*sk
,
1597 const struct request_sock
*req
,
1598 const struct sk_buff
*skb
);
1602 int tcpv4_offload_init(void);
1604 void tcp_v4_init(void);
1605 void tcp_init(void);