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/cryptohash.h>
31 #include <linux/kref.h>
32 #include <linux/ktime.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
49 #include <linux/bpf.h>
50 #include <linux/filter.h>
51 #include <linux/bpf-cgroup.h>
53 extern struct inet_hashinfo tcp_hashinfo
;
55 extern struct percpu_counter tcp_orphan_count
;
56 void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
58 #define MAX_TCP_HEADER (128 + MAX_HEADER)
59 #define MAX_TCP_OPTION_SPACE 40
62 * Never offer a window over 32767 without using window scaling. Some
63 * poor stacks do signed 16bit maths!
65 #define MAX_TCP_WINDOW 32767U
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 1024
73 /* probing interval, default to 10 minutes as per RFC4821 */
74 #define TCP_PROBE_INTERVAL 600
76 /* Specify interval when tcp mtu probing will stop */
77 #define TCP_PROBE_THRESHOLD 8
79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
80 #define TCP_FASTRETRANS_THRESH 3
82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
83 #define TCP_MAX_QUICKACKS 16U
85 /* Maximal number of window scale according to RFC1323 */
86 #define TCP_MAX_WSCALE 14U
89 #define TCP_URG_VALID 0x0100
90 #define TCP_URG_NOTYET 0x0200
91 #define TCP_URG_READ 0x0400
93 #define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
100 #define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
132 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
134 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
135 #define TCP_ATO_MIN ((unsigned)(HZ/25))
137 #define TCP_DELACK_MIN 4U
138 #define TCP_ATO_MIN 4U
140 #define TCP_RTO_MAX ((unsigned)(120*HZ))
141 #define TCP_RTO_MIN ((unsigned)(HZ/5))
142 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
143 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
144 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
145 * used as a fallback RTO for the
146 * initial data transmission if no
147 * valid RTT sample has been acquired,
148 * most likely due to retrans in 3WHS.
151 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
152 * for local resources.
154 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
155 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
156 #define TCP_KEEPALIVE_INTVL (75*HZ)
158 #define MAX_TCP_KEEPIDLE 32767
159 #define MAX_TCP_KEEPINTVL 32767
160 #define MAX_TCP_KEEPCNT 127
161 #define MAX_TCP_SYNCNT 127
163 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
167 * after this time. It should be equal
168 * (or greater than) TCP_TIMEWAIT_LEN
169 * to provide reliability equal to one
170 * provided by timewait state.
172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
173 * timestamps. It must be less than
174 * minimal timewait lifetime.
180 #define TCPOPT_NOP 1 /* Padding */
181 #define TCPOPT_EOL 0 /* End of options */
182 #define TCPOPT_MSS 2 /* Segment size negotiating */
183 #define TCPOPT_WINDOW 3 /* Window scaling */
184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
185 #define TCPOPT_SACK 5 /* SACK Block */
186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
189 #define TCPOPT_EXP 254 /* Experimental */
190 /* Magic number to be after the option value for sharing TCP
191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
193 #define TCPOPT_FASTOPEN_MAGIC 0xF989
199 #define TCPOLEN_MSS 4
200 #define TCPOLEN_WINDOW 3
201 #define TCPOLEN_SACK_PERM 2
202 #define TCPOLEN_TIMESTAMP 10
203 #define TCPOLEN_MD5SIG 18
204 #define TCPOLEN_FASTOPEN_BASE 2
205 #define TCPOLEN_EXP_FASTOPEN_BASE 4
207 /* But this is what stacks really send out. */
208 #define TCPOLEN_TSTAMP_ALIGNED 12
209 #define TCPOLEN_WSCALE_ALIGNED 4
210 #define TCPOLEN_SACKPERM_ALIGNED 4
211 #define TCPOLEN_SACK_BASE 2
212 #define TCPOLEN_SACK_BASE_ALIGNED 4
213 #define TCPOLEN_SACK_PERBLOCK 8
214 #define TCPOLEN_MD5SIG_ALIGNED 20
215 #define TCPOLEN_MSS_ALIGNED 4
217 /* Flags in tp->nonagle */
218 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219 #define TCP_NAGLE_CORK 2 /* Socket is corked */
220 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
222 /* TCP thin-stream limits */
223 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
225 /* TCP initial congestion window as per rfc6928 */
226 #define TCP_INIT_CWND 10
228 /* Bit Flags for sysctl_tcp_fastopen */
229 #define TFO_CLIENT_ENABLE 1
230 #define TFO_SERVER_ENABLE 2
231 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
233 /* Accept SYN data w/o any cookie option */
234 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
236 /* Force enable TFO on all listeners, i.e., not requiring the
237 * TCP_FASTOPEN socket option.
239 #define TFO_SERVER_WO_SOCKOPT1 0x400
242 /* sysctl variables for tcp */
243 extern int sysctl_tcp_fastopen
;
244 extern int sysctl_tcp_retrans_collapse
;
245 extern int sysctl_tcp_stdurg
;
246 extern int sysctl_tcp_rfc1337
;
247 extern int sysctl_tcp_abort_on_overflow
;
248 extern int sysctl_tcp_max_orphans
;
249 extern int sysctl_tcp_fack
;
250 extern int sysctl_tcp_reordering
;
251 extern int sysctl_tcp_max_reordering
;
252 extern int sysctl_tcp_dsack
;
253 extern long sysctl_tcp_mem
[3];
254 extern int sysctl_tcp_wmem
[3];
255 extern int sysctl_tcp_rmem
[3];
256 extern int sysctl_tcp_app_win
;
257 extern int sysctl_tcp_adv_win_scale
;
258 extern int sysctl_tcp_frto
;
259 extern int sysctl_tcp_nometrics_save
;
260 extern int sysctl_tcp_moderate_rcvbuf
;
261 extern int sysctl_tcp_tso_win_divisor
;
262 extern int sysctl_tcp_workaround_signed_windows
;
263 extern int sysctl_tcp_slow_start_after_idle
;
264 extern int sysctl_tcp_thin_linear_timeouts
;
265 extern int sysctl_tcp_thin_dupack
;
266 extern int sysctl_tcp_early_retrans
;
267 extern int sysctl_tcp_recovery
;
268 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
270 extern int sysctl_tcp_limit_output_bytes
;
271 extern int sysctl_tcp_challenge_ack_limit
;
272 extern int sysctl_tcp_min_tso_segs
;
273 extern int sysctl_tcp_min_rtt_wlen
;
274 extern int sysctl_tcp_autocorking
;
275 extern int sysctl_tcp_invalid_ratelimit
;
276 extern int sysctl_tcp_pacing_ss_ratio
;
277 extern int sysctl_tcp_pacing_ca_ratio
;
279 extern atomic_long_t tcp_memory_allocated
;
280 extern struct percpu_counter tcp_sockets_allocated
;
281 extern unsigned long tcp_memory_pressure
;
283 /* optimized version of sk_under_memory_pressure() for TCP sockets */
284 static inline bool tcp_under_memory_pressure(const struct sock
*sk
)
286 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
287 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
290 return tcp_memory_pressure
;
293 * The next routines deal with comparing 32 bit unsigned ints
294 * and worry about wraparound (automatic with unsigned arithmetic).
297 static inline bool before(__u32 seq1
, __u32 seq2
)
299 return (__s32
)(seq1
-seq2
) < 0;
301 #define after(seq2, seq1) before(seq1, seq2)
303 /* is s2<=s1<=s3 ? */
304 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
306 return seq3
- seq2
>= seq1
- seq2
;
309 static inline bool tcp_out_of_memory(struct sock
*sk
)
311 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
312 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
317 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
319 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
321 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
322 int orphans
= percpu_counter_read_positive(ocp
);
324 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
325 orphans
= percpu_counter_sum_positive(ocp
);
326 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
332 bool tcp_check_oom(struct sock
*sk
, int shift
);
335 extern struct proto tcp_prot
;
337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
338 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
340 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
342 void tcp_tasklet_init(void);
344 void tcp_v4_err(struct sk_buff
*skb
, u32
);
346 void tcp_shutdown(struct sock
*sk
, int how
);
348 void tcp_v4_early_demux(struct sk_buff
*skb
);
349 int tcp_v4_rcv(struct sk_buff
*skb
);
351 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
352 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
353 int tcp_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
354 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
356 int tcp_sendpage_locked(struct sock
*sk
, struct page
*page
, int offset
,
357 size_t size
, int flags
);
358 ssize_t
do_tcp_sendpages(struct sock
*sk
, struct page
*page
, int offset
,
359 size_t size
, int flags
);
360 void tcp_release_cb(struct sock
*sk
);
361 void tcp_wfree(struct sk_buff
*skb
);
362 void tcp_write_timer_handler(struct sock
*sk
);
363 void tcp_delack_timer_handler(struct sock
*sk
);
364 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
365 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
);
366 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
367 const struct tcphdr
*th
);
368 void tcp_rcv_space_adjust(struct sock
*sk
);
369 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
370 void tcp_twsk_destructor(struct sock
*sk
);
371 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
372 struct pipe_inode_info
*pipe
, size_t len
,
375 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
376 const unsigned int pkts
)
378 struct inet_connection_sock
*icsk
= inet_csk(sk
);
380 if (icsk
->icsk_ack
.quick
) {
381 if (pkts
>= icsk
->icsk_ack
.quick
) {
382 icsk
->icsk_ack
.quick
= 0;
383 /* Leaving quickack mode we deflate ATO. */
384 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
386 icsk
->icsk_ack
.quick
-= pkts
;
391 #define TCP_ECN_QUEUE_CWR 2
392 #define TCP_ECN_DEMAND_CWR 4
393 #define TCP_ECN_SEEN 8
403 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
405 const struct tcphdr
*th
);
406 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
407 struct request_sock
*req
, bool fastopen
);
408 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
409 struct sk_buff
*skb
);
410 void tcp_enter_loss(struct sock
*sk
);
411 void tcp_cwnd_reduction(struct sock
*sk
, int newly_acked_sacked
, int flag
);
412 void tcp_clear_retrans(struct tcp_sock
*tp
);
413 void tcp_update_metrics(struct sock
*sk
);
414 void tcp_init_metrics(struct sock
*sk
);
415 void tcp_metrics_init(void);
416 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
);
417 void tcp_disable_fack(struct tcp_sock
*tp
);
418 void tcp_close(struct sock
*sk
, long timeout
);
419 void tcp_init_sock(struct sock
*sk
);
420 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
,
421 struct poll_table_struct
*wait
);
422 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
423 char __user
*optval
, int __user
*optlen
);
424 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
425 char __user
*optval
, unsigned int optlen
);
426 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
427 char __user
*optval
, int __user
*optlen
);
428 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
429 char __user
*optval
, unsigned int optlen
);
430 void tcp_set_keepalive(struct sock
*sk
, int val
);
431 void tcp_syn_ack_timeout(const struct request_sock
*req
);
432 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
433 int flags
, int *addr_len
);
434 void tcp_parse_options(const struct net
*net
, const struct sk_buff
*skb
,
435 struct tcp_options_received
*opt_rx
,
436 int estab
, struct tcp_fastopen_cookie
*foc
);
437 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
440 * TCP v4 functions exported for the inet6 API
443 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
444 void tcp_v4_mtu_reduced(struct sock
*sk
);
445 void tcp_req_err(struct sock
*sk
, u32 seq
, bool abort
);
446 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
447 struct sock
*tcp_create_openreq_child(const struct sock
*sk
,
448 struct request_sock
*req
,
449 struct sk_buff
*skb
);
450 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
451 struct sock
*tcp_v4_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
452 struct request_sock
*req
,
453 struct dst_entry
*dst
,
454 struct request_sock
*req_unhash
,
456 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
457 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
458 int tcp_connect(struct sock
*sk
);
459 enum tcp_synack_type
{
464 struct sk_buff
*tcp_make_synack(const struct sock
*sk
, struct dst_entry
*dst
,
465 struct request_sock
*req
,
466 struct tcp_fastopen_cookie
*foc
,
467 enum tcp_synack_type synack_type
);
468 int tcp_disconnect(struct sock
*sk
, int flags
);
470 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
471 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
472 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
474 /* From syncookies.c */
475 struct sock
*tcp_get_cookie_sock(struct sock
*sk
, struct sk_buff
*skb
,
476 struct request_sock
*req
,
477 struct dst_entry
*dst
, u32 tsoff
);
478 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
480 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
481 #ifdef CONFIG_SYN_COOKIES
483 /* Syncookies use a monotonic timer which increments every 60 seconds.
484 * This counter is used both as a hash input and partially encoded into
485 * the cookie value. A cookie is only validated further if the delta
486 * between the current counter value and the encoded one is less than this,
487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
488 * the counter advances immediately after a cookie is generated).
490 #define MAX_SYNCOOKIE_AGE 2
491 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
492 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
494 /* syncookies: remember time of last synqueue overflow
495 * But do not dirty this field too often (once per second is enough)
496 * It is racy as we do not hold a lock, but race is very minor.
498 static inline void tcp_synq_overflow(const struct sock
*sk
)
500 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
501 unsigned long now
= jiffies
;
503 if (time_after(now
, last_overflow
+ HZ
))
504 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= now
;
507 /* syncookies: no recent synqueue overflow on this listening socket? */
508 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
510 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
512 return time_after(jiffies
, last_overflow
+ TCP_SYNCOOKIE_VALID
);
515 static inline u32
tcp_cookie_time(void)
517 u64 val
= get_jiffies_64();
519 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
523 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
525 __u32
cookie_v4_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
526 u64
cookie_init_timestamp(struct request_sock
*req
);
527 bool cookie_timestamp_decode(const struct net
*net
,
528 struct tcp_options_received
*opt
);
529 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
530 const struct net
*net
, const struct dst_entry
*dst
);
532 /* From net/ipv6/syncookies.c */
533 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
535 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
537 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
538 const struct tcphdr
*th
, u16
*mssp
);
539 __u32
cookie_v6_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
543 u32
tcp_tso_autosize(const struct sock
*sk
, unsigned int mss_now
,
545 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
547 bool tcp_may_send_now(struct sock
*sk
);
548 int __tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
549 int tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
550 void tcp_retransmit_timer(struct sock
*sk
);
551 void tcp_xmit_retransmit_queue(struct sock
*);
552 void tcp_simple_retransmit(struct sock
*);
553 void tcp_enter_recovery(struct sock
*sk
, bool ece_ack
);
554 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
555 int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int, gfp_t
);
557 void tcp_send_probe0(struct sock
*);
558 void tcp_send_partial(struct sock
*);
559 int tcp_write_wakeup(struct sock
*, int mib
);
560 void tcp_send_fin(struct sock
*sk
);
561 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
562 int tcp_send_synack(struct sock
*);
563 void tcp_push_one(struct sock
*, unsigned int mss_now
);
564 void tcp_send_ack(struct sock
*sk
);
565 void tcp_send_delayed_ack(struct sock
*sk
);
566 void tcp_send_loss_probe(struct sock
*sk
);
567 bool tcp_schedule_loss_probe(struct sock
*sk
);
568 void tcp_skb_collapse_tstamp(struct sk_buff
*skb
,
569 const struct sk_buff
*next_skb
);
572 void tcp_rearm_rto(struct sock
*sk
);
573 void tcp_synack_rtt_meas(struct sock
*sk
, struct request_sock
*req
);
574 void tcp_reset(struct sock
*sk
);
575 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock
*tp
, struct sk_buff
*skb
);
576 void tcp_fin(struct sock
*sk
);
579 void tcp_init_xmit_timers(struct sock
*);
580 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
582 hrtimer_cancel(&tcp_sk(sk
)->pacing_timer
);
583 inet_csk_clear_xmit_timers(sk
);
586 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
587 unsigned int tcp_current_mss(struct sock
*sk
);
589 /* Bound MSS / TSO packet size with the half of the window */
590 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
594 /* When peer uses tiny windows, there is no use in packetizing
595 * to sub-MSS pieces for the sake of SWS or making sure there
596 * are enough packets in the pipe for fast recovery.
598 * On the other hand, for extremely large MSS devices, handling
599 * smaller than MSS windows in this way does make sense.
601 if (tp
->max_window
> TCP_MSS_DEFAULT
)
602 cutoff
= (tp
->max_window
>> 1);
604 cutoff
= tp
->max_window
;
606 if (cutoff
&& pktsize
> cutoff
)
607 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
613 void tcp_get_info(struct sock
*, struct tcp_info
*);
615 /* Read 'sendfile()'-style from a TCP socket */
616 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
617 sk_read_actor_t recv_actor
);
619 void tcp_initialize_rcv_mss(struct sock
*sk
);
621 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
622 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
623 void tcp_mtup_init(struct sock
*sk
);
624 void tcp_init_buffer_space(struct sock
*sk
);
626 static inline void tcp_bound_rto(const struct sock
*sk
)
628 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
629 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
632 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
634 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
637 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
639 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
640 ntohl(TCP_FLAG_ACK
) |
644 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
646 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
649 static inline void tcp_fast_path_check(struct sock
*sk
)
651 struct tcp_sock
*tp
= tcp_sk(sk
);
653 if (RB_EMPTY_ROOT(&tp
->out_of_order_queue
) &&
655 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
657 tcp_fast_path_on(tp
);
660 /* Compute the actual rto_min value */
661 static inline u32
tcp_rto_min(struct sock
*sk
)
663 const struct dst_entry
*dst
= __sk_dst_get(sk
);
664 u32 rto_min
= TCP_RTO_MIN
;
666 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
667 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
671 static inline u32
tcp_rto_min_us(struct sock
*sk
)
673 return jiffies_to_usecs(tcp_rto_min(sk
));
676 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
678 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
681 /* Minimum RTT in usec. ~0 means not available. */
682 static inline u32
tcp_min_rtt(const struct tcp_sock
*tp
)
684 return minmax_get(&tp
->rtt_min
);
687 /* Compute the actual receive window we are currently advertising.
688 * Rcv_nxt can be after the window if our peer push more data
689 * than the offered window.
691 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
693 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
700 /* Choose a new window, without checks for shrinking, and without
701 * scaling applied to the result. The caller does these things
702 * if necessary. This is a "raw" window selection.
704 u32
__tcp_select_window(struct sock
*sk
);
706 void tcp_send_window_probe(struct sock
*sk
);
708 /* TCP uses 32bit jiffies to save some space.
709 * Note that this is different from tcp_time_stamp, which
710 * historically has been the same until linux-4.13.
712 #define tcp_jiffies32 ((u32)jiffies)
715 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
716 * It is no longer tied to jiffies, but to 1 ms clock.
717 * Note: double check if you want to use tcp_jiffies32 instead of this.
719 #define TCP_TS_HZ 1000
721 static inline u64
tcp_clock_ns(void)
723 return local_clock();
726 static inline u64
tcp_clock_us(void)
728 return div_u64(tcp_clock_ns(), NSEC_PER_USEC
);
731 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
732 static inline u32
tcp_time_stamp(const struct tcp_sock
*tp
)
734 return div_u64(tp
->tcp_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
737 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
738 static inline u32
tcp_time_stamp_raw(void)
740 return div_u64(tcp_clock_ns(), NSEC_PER_SEC
/ TCP_TS_HZ
);
744 /* Refresh 1us clock of a TCP socket,
745 * ensuring monotically increasing values.
747 static inline void tcp_mstamp_refresh(struct tcp_sock
*tp
)
749 u64 val
= tcp_clock_us();
751 if (val
> tp
->tcp_mstamp
)
752 tp
->tcp_mstamp
= val
;
755 static inline u32
tcp_stamp_us_delta(u64 t1
, u64 t0
)
757 return max_t(s64
, t1
- t0
, 0);
760 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
762 return div_u64(skb
->skb_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
766 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
768 #define TCPHDR_FIN 0x01
769 #define TCPHDR_SYN 0x02
770 #define TCPHDR_RST 0x04
771 #define TCPHDR_PSH 0x08
772 #define TCPHDR_ACK 0x10
773 #define TCPHDR_URG 0x20
774 #define TCPHDR_ECE 0x40
775 #define TCPHDR_CWR 0x80
777 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
779 /* This is what the send packet queuing engine uses to pass
780 * TCP per-packet control information to the transmission code.
781 * We also store the host-order sequence numbers in here too.
782 * This is 44 bytes if IPV6 is enabled.
783 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
786 __u32 seq
; /* Starting sequence number */
787 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
789 /* Note : tcp_tw_isn is used in input path only
790 * (isn chosen by tcp_timewait_state_process())
792 * tcp_gso_segs/size are used in write queue only,
793 * cf tcp_skb_pcount()/tcp_skb_mss()
801 /* Used to stash the receive timestamp while this skb is in the
802 * out of order queue, as skb->tstamp is overwritten by the
807 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
809 __u8 sacked
; /* State flags for SACK/FACK. */
810 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
811 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
812 #define TCPCB_LOST 0x04 /* SKB is lost */
813 #define TCPCB_TAGBITS 0x07 /* All tag bits */
814 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
815 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
816 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
819 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
820 __u8 txstamp_ack
:1, /* Record TX timestamp for ack? */
821 eor
:1, /* Is skb MSG_EOR marked? */
822 has_rxtstamp
:1, /* SKB has a RX timestamp */
824 __u32 ack_seq
; /* Sequence number ACK'd */
827 /* There is space for up to 24 bytes */
828 __u32 in_flight
:30,/* Bytes in flight at transmit */
829 is_app_limited
:1, /* cwnd not fully used? */
831 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
833 /* start of send pipeline phase */
835 /* when we reached the "delivered" count */
836 u64 delivered_mstamp
;
837 } tx
; /* only used for outgoing skbs */
839 struct inet_skb_parm h4
;
840 #if IS_ENABLED(CONFIG_IPV6)
841 struct inet6_skb_parm h6
;
843 } header
; /* For incoming skbs */
847 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
850 #if IS_ENABLED(CONFIG_IPV6)
851 /* This is the variant of inet6_iif() that must be used by TCP,
852 * as TCP moves IP6CB into a different location in skb->cb[]
854 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
856 bool l3_slave
= ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
);
858 return l3_slave
? skb
->skb_iif
: TCP_SKB_CB(skb
)->header
.h6
.iif
;
861 /* TCP_SKB_CB reference means this can not be used from early demux */
862 static inline int tcp_v6_sdif(const struct sk_buff
*skb
)
864 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
865 if (skb
&& ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
))
866 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
872 /* TCP_SKB_CB reference means this can not be used from early demux */
873 static inline bool inet_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
875 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
876 if (!net
->ipv4
.sysctl_tcp_l3mdev_accept
&&
877 skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
883 /* TCP_SKB_CB reference means this can not be used from early demux */
884 static inline int tcp_v4_sdif(struct sk_buff
*skb
)
886 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
887 if (skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
888 return TCP_SKB_CB(skb
)->header
.h4
.iif
;
893 /* Due to TSO, an SKB can be composed of multiple actual
894 * packets. To keep these tracked properly, we use this.
896 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
898 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
901 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
903 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
906 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
908 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
911 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
912 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
914 return TCP_SKB_CB(skb
)->tcp_gso_size
;
917 static inline bool tcp_skb_can_collapse_to(const struct sk_buff
*skb
)
919 return likely(!TCP_SKB_CB(skb
)->eor
);
922 /* Events passed to congestion control interface */
924 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
925 CA_EVENT_CWND_RESTART
, /* congestion window restart */
926 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
927 CA_EVENT_LOSS
, /* loss timeout */
928 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
929 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
930 CA_EVENT_DELAYED_ACK
, /* Delayed ack is sent */
931 CA_EVENT_NON_DELAYED_ACK
,
934 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
935 enum tcp_ca_ack_event_flags
{
936 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
937 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
938 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
942 * Interface for adding new TCP congestion control handlers
944 #define TCP_CA_NAME_MAX 16
945 #define TCP_CA_MAX 128
946 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
948 #define TCP_CA_UNSPEC 0
950 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
951 #define TCP_CONG_NON_RESTRICTED 0x1
952 /* Requires ECN/ECT set on all packets */
953 #define TCP_CONG_NEEDS_ECN 0x2
963 /* A rate sample measures the number of (original/retransmitted) data
964 * packets delivered "delivered" over an interval of time "interval_us".
965 * The tcp_rate.c code fills in the rate sample, and congestion
966 * control modules that define a cong_control function to run at the end
967 * of ACK processing can optionally chose to consult this sample when
968 * setting cwnd and pacing rate.
969 * A sample is invalid if "delivered" or "interval_us" is negative.
972 u64 prior_mstamp
; /* starting timestamp for interval */
973 u32 prior_delivered
; /* tp->delivered at "prior_mstamp" */
974 s32 delivered
; /* number of packets delivered over interval */
975 long interval_us
; /* time for tp->delivered to incr "delivered" */
976 long rtt_us
; /* RTT of last (S)ACKed packet (or -1) */
977 int losses
; /* number of packets marked lost upon ACK */
978 u32 acked_sacked
; /* number of packets newly (S)ACKed upon ACK */
979 u32 prior_in_flight
; /* in flight before this ACK */
980 bool is_app_limited
; /* is sample from packet with bubble in pipe? */
981 bool is_retrans
; /* is sample from retransmission? */
984 struct tcp_congestion_ops
{
985 struct list_head list
;
989 /* initialize private data (optional) */
990 void (*init
)(struct sock
*sk
);
991 /* cleanup private data (optional) */
992 void (*release
)(struct sock
*sk
);
994 /* return slow start threshold (required) */
995 u32 (*ssthresh
)(struct sock
*sk
);
996 /* do new cwnd calculation (required) */
997 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
998 /* call before changing ca_state (optional) */
999 void (*set_state
)(struct sock
*sk
, u8 new_state
);
1000 /* call when cwnd event occurs (optional) */
1001 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
1002 /* call when ack arrives (optional) */
1003 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
1004 /* new value of cwnd after loss (required) */
1005 u32 (*undo_cwnd
)(struct sock
*sk
);
1006 /* hook for packet ack accounting (optional) */
1007 void (*pkts_acked
)(struct sock
*sk
, const struct ack_sample
*sample
);
1008 /* suggest number of segments for each skb to transmit (optional) */
1009 u32 (*tso_segs_goal
)(struct sock
*sk
);
1010 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1011 u32 (*sndbuf_expand
)(struct sock
*sk
);
1012 /* call when packets are delivered to update cwnd and pacing rate,
1013 * after all the ca_state processing. (optional)
1015 void (*cong_control
)(struct sock
*sk
, const struct rate_sample
*rs
);
1016 /* get info for inet_diag (optional) */
1017 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
1018 union tcp_cc_info
*info
);
1020 char name
[TCP_CA_NAME_MAX
];
1021 struct module
*owner
;
1024 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1025 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1027 void tcp_assign_congestion_control(struct sock
*sk
);
1028 void tcp_init_congestion_control(struct sock
*sk
);
1029 void tcp_cleanup_congestion_control(struct sock
*sk
);
1030 int tcp_set_default_congestion_control(const char *name
);
1031 void tcp_get_default_congestion_control(char *name
);
1032 void tcp_get_available_congestion_control(char *buf
, size_t len
);
1033 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
1034 int tcp_set_allowed_congestion_control(char *allowed
);
1035 int tcp_set_congestion_control(struct sock
*sk
, const char *name
, bool load
);
1036 void tcp_reinit_congestion_control(struct sock
*sk
,
1037 const struct tcp_congestion_ops
*ca
);
1038 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
1039 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
1041 u32
tcp_reno_ssthresh(struct sock
*sk
);
1042 u32
tcp_reno_undo_cwnd(struct sock
*sk
);
1043 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
1044 extern struct tcp_congestion_ops tcp_reno
;
1046 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
1047 u32
tcp_ca_get_key_by_name(const char *name
, bool *ecn_ca
);
1049 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
1051 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
1057 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
1059 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1061 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
1064 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
1066 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1068 if (icsk
->icsk_ca_ops
->set_state
)
1069 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
1070 icsk
->icsk_ca_state
= ca_state
;
1073 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
1075 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1077 if (icsk
->icsk_ca_ops
->cwnd_event
)
1078 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
1081 /* From tcp_rate.c */
1082 void tcp_rate_skb_sent(struct sock
*sk
, struct sk_buff
*skb
);
1083 void tcp_rate_skb_delivered(struct sock
*sk
, struct sk_buff
*skb
,
1084 struct rate_sample
*rs
);
1085 void tcp_rate_gen(struct sock
*sk
, u32 delivered
, u32 lost
,
1086 struct rate_sample
*rs
);
1087 void tcp_rate_check_app_limited(struct sock
*sk
);
1089 /* These functions determine how the current flow behaves in respect of SACK
1090 * handling. SACK is negotiated with the peer, and therefore it can vary
1091 * between different flows.
1093 * tcp_is_sack - SACK enabled
1094 * tcp_is_reno - No SACK
1095 * tcp_is_fack - FACK enabled, implies SACK enabled
1097 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
1099 return tp
->rx_opt
.sack_ok
;
1102 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
1104 return !tcp_is_sack(tp
);
1107 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
1109 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
1112 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
1114 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
1117 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
1119 return tp
->sacked_out
+ tp
->lost_out
;
1122 /* This determines how many packets are "in the network" to the best
1123 * of our knowledge. In many cases it is conservative, but where
1124 * detailed information is available from the receiver (via SACK
1125 * blocks etc.) we can make more aggressive calculations.
1127 * Use this for decisions involving congestion control, use just
1128 * tp->packets_out to determine if the send queue is empty or not.
1130 * Read this equation as:
1132 * "Packets sent once on transmission queue" MINUS
1133 * "Packets left network, but not honestly ACKed yet" PLUS
1134 * "Packets fast retransmitted"
1136 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1138 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
1141 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1143 static inline bool tcp_in_slow_start(const struct tcp_sock
*tp
)
1145 return tp
->snd_cwnd
< tp
->snd_ssthresh
;
1148 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
1150 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
1153 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
1155 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
1156 (1 << inet_csk(sk
)->icsk_ca_state
);
1159 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1160 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1163 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
1165 const struct tcp_sock
*tp
= tcp_sk(sk
);
1167 if (tcp_in_cwnd_reduction(sk
))
1168 return tp
->snd_ssthresh
;
1170 return max(tp
->snd_ssthresh
,
1171 ((tp
->snd_cwnd
>> 1) +
1172 (tp
->snd_cwnd
>> 2)));
1175 /* Use define here intentionally to get WARN_ON location shown at the caller */
1176 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1178 void tcp_enter_cwr(struct sock
*sk
);
1179 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1181 /* The maximum number of MSS of available cwnd for which TSO defers
1182 * sending if not using sysctl_tcp_tso_win_divisor.
1184 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1189 /* Returns end sequence number of the receiver's advertised window */
1190 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1192 return tp
->snd_una
+ tp
->snd_wnd
;
1195 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1196 * flexible approach. The RFC suggests cwnd should not be raised unless
1197 * it was fully used previously. And that's exactly what we do in
1198 * congestion avoidance mode. But in slow start we allow cwnd to grow
1199 * as long as the application has used half the cwnd.
1201 * cwnd is 10 (IW10), but application sends 9 frames.
1202 * We allow cwnd to reach 18 when all frames are ACKed.
1203 * This check is safe because it's as aggressive as slow start which already
1204 * risks 100% overshoot. The advantage is that we discourage application to
1205 * either send more filler packets or data to artificially blow up the cwnd
1206 * usage, and allow application-limited process to probe bw more aggressively.
1208 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1210 const struct tcp_sock
*tp
= tcp_sk(sk
);
1212 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1213 if (tcp_in_slow_start(tp
))
1214 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1216 return tp
->is_cwnd_limited
;
1219 /* Something is really bad, we could not queue an additional packet,
1220 * because qdisc is full or receiver sent a 0 window.
1221 * We do not want to add fuel to the fire, or abort too early,
1222 * so make sure the timer we arm now is at least 200ms in the future,
1223 * regardless of current icsk_rto value (as it could be ~2ms)
1225 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1227 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1230 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1231 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1232 unsigned long max_when
)
1234 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1236 return (unsigned long)min_t(u64
, when
, max_when
);
1239 static inline void tcp_check_probe_timer(struct sock
*sk
)
1241 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1242 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1243 tcp_probe0_base(sk
), TCP_RTO_MAX
);
1246 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1251 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1257 * Calculate(/check) TCP checksum
1259 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1260 __be32 daddr
, __wsum base
)
1262 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1265 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1267 return __skb_checksum_complete(skb
);
1270 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1272 return !skb_csum_unnecessary(skb
) &&
1273 __tcp_checksum_complete(skb
);
1276 bool tcp_add_backlog(struct sock
*sk
, struct sk_buff
*skb
);
1277 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
);
1282 static const char *statename
[]={
1283 "Unused","Established","Syn Sent","Syn Recv",
1284 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1285 "Close Wait","Last ACK","Listen","Closing"
1288 void tcp_set_state(struct sock
*sk
, int state
);
1290 void tcp_done(struct sock
*sk
);
1292 int tcp_abort(struct sock
*sk
, int err
);
1294 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1297 rx_opt
->num_sacks
= 0;
1300 u32
tcp_default_init_rwnd(u32 mss
);
1301 void tcp_cwnd_restart(struct sock
*sk
, s32 delta
);
1303 static inline void tcp_slow_start_after_idle_check(struct sock
*sk
)
1305 const struct tcp_congestion_ops
*ca_ops
= inet_csk(sk
)->icsk_ca_ops
;
1306 struct tcp_sock
*tp
= tcp_sk(sk
);
1309 if (!sysctl_tcp_slow_start_after_idle
|| tp
->packets_out
||
1310 ca_ops
->cong_control
)
1312 delta
= tcp_jiffies32
- tp
->lsndtime
;
1313 if (delta
> inet_csk(sk
)->icsk_rto
)
1314 tcp_cwnd_restart(sk
, delta
);
1317 /* Determine a window scaling and initial window to offer. */
1318 void tcp_select_initial_window(int __space
, __u32 mss
, __u32
*rcv_wnd
,
1319 __u32
*window_clamp
, int wscale_ok
,
1320 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1322 static inline int tcp_win_from_space(int space
)
1324 int tcp_adv_win_scale
= sysctl_tcp_adv_win_scale
;
1326 return tcp_adv_win_scale
<= 0 ?
1327 (space
>>(-tcp_adv_win_scale
)) :
1328 space
- (space
>>tcp_adv_win_scale
);
1331 /* Note: caller must be prepared to deal with negative returns */
1332 static inline int tcp_space(const struct sock
*sk
)
1334 return tcp_win_from_space(sk
->sk_rcvbuf
-
1335 atomic_read(&sk
->sk_rmem_alloc
));
1338 static inline int tcp_full_space(const struct sock
*sk
)
1340 return tcp_win_from_space(sk
->sk_rcvbuf
);
1343 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1344 const struct sock
*sk_listener
,
1345 const struct dst_entry
*dst
);
1347 void tcp_enter_memory_pressure(struct sock
*sk
);
1348 void tcp_leave_memory_pressure(struct sock
*sk
);
1350 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1352 struct net
*net
= sock_net((struct sock
*)tp
);
1354 return tp
->keepalive_intvl
? : net
->ipv4
.sysctl_tcp_keepalive_intvl
;
1357 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1359 struct net
*net
= sock_net((struct sock
*)tp
);
1361 return tp
->keepalive_time
? : net
->ipv4
.sysctl_tcp_keepalive_time
;
1364 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1366 struct net
*net
= sock_net((struct sock
*)tp
);
1368 return tp
->keepalive_probes
? : net
->ipv4
.sysctl_tcp_keepalive_probes
;
1371 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1373 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1375 return min_t(u32
, tcp_jiffies32
- icsk
->icsk_ack
.lrcvtime
,
1376 tcp_jiffies32
- tp
->rcv_tstamp
);
1379 static inline int tcp_fin_time(const struct sock
*sk
)
1381 int fin_timeout
= tcp_sk(sk
)->linger2
? : sock_net(sk
)->ipv4
.sysctl_tcp_fin_timeout
;
1382 const int rto
= inet_csk(sk
)->icsk_rto
;
1384 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1385 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1390 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1393 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1395 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1398 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1399 * then following tcp messages have valid values. Ignore 0 value,
1400 * or else 'negative' tsval might forbid us to accept their packets.
1402 if (!rx_opt
->ts_recent
)
1407 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1410 if (tcp_paws_check(rx_opt
, 0))
1413 /* RST segments are not recommended to carry timestamp,
1414 and, if they do, it is recommended to ignore PAWS because
1415 "their cleanup function should take precedence over timestamps."
1416 Certainly, it is mistake. It is necessary to understand the reasons
1417 of this constraint to relax it: if peer reboots, clock may go
1418 out-of-sync and half-open connections will not be reset.
1419 Actually, the problem would be not existing if all
1420 the implementations followed draft about maintaining clock
1421 via reboots. Linux-2.2 DOES NOT!
1423 However, we can relax time bounds for RST segments to MSL.
1425 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1430 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1431 int mib_idx
, u32
*last_oow_ack_time
);
1433 static inline void tcp_mib_init(struct net
*net
)
1436 TCP_ADD_STATS(net
, TCP_MIB_RTOALGORITHM
, 1);
1437 TCP_ADD_STATS(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1438 TCP_ADD_STATS(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1439 TCP_ADD_STATS(net
, TCP_MIB_MAXCONN
, -1);
1443 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1445 tp
->lost_skb_hint
= NULL
;
1448 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1450 tcp_clear_retrans_hints_partial(tp
);
1451 tp
->retransmit_skb_hint
= NULL
;
1454 union tcp_md5_addr
{
1456 #if IS_ENABLED(CONFIG_IPV6)
1461 /* - key database */
1462 struct tcp_md5sig_key
{
1463 struct hlist_node node
;
1465 u8 family
; /* AF_INET or AF_INET6 */
1466 union tcp_md5_addr addr
;
1468 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1469 struct rcu_head rcu
;
1473 struct tcp_md5sig_info
{
1474 struct hlist_head head
;
1475 struct rcu_head rcu
;
1478 /* - pseudo header */
1479 struct tcp4_pseudohdr
{
1487 struct tcp6_pseudohdr
{
1488 struct in6_addr saddr
;
1489 struct in6_addr daddr
;
1491 __be32 protocol
; /* including padding */
1494 union tcp_md5sum_block
{
1495 struct tcp4_pseudohdr ip4
;
1496 #if IS_ENABLED(CONFIG_IPV6)
1497 struct tcp6_pseudohdr ip6
;
1501 /* - pool: digest algorithm, hash description and scratch buffer */
1502 struct tcp_md5sig_pool
{
1503 struct ahash_request
*md5_req
;
1508 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1509 const struct sock
*sk
, const struct sk_buff
*skb
);
1510 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1511 int family
, u8 prefixlen
, const u8
*newkey
, u8 newkeylen
,
1513 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1514 int family
, u8 prefixlen
);
1515 struct tcp_md5sig_key
*tcp_v4_md5_lookup(const struct sock
*sk
,
1516 const struct sock
*addr_sk
);
1518 #ifdef CONFIG_TCP_MD5SIG
1519 struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1520 const union tcp_md5_addr
*addr
,
1522 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1524 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1525 const union tcp_md5_addr
*addr
,
1530 #define tcp_twsk_md5_key(twsk) NULL
1533 bool tcp_alloc_md5sig_pool(void);
1535 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1536 static inline void tcp_put_md5sig_pool(void)
1541 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1542 unsigned int header_len
);
1543 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1544 const struct tcp_md5sig_key
*key
);
1546 /* From tcp_fastopen.c */
1547 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1548 struct tcp_fastopen_cookie
*cookie
, int *syn_loss
,
1549 unsigned long *last_syn_loss
);
1550 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1551 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1553 struct tcp_fastopen_request
{
1554 /* Fast Open cookie. Size 0 means a cookie request */
1555 struct tcp_fastopen_cookie cookie
;
1556 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1558 int copied
; /* queued in tcp_connect() */
1560 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1562 extern struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
1563 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
);
1564 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
);
1565 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1566 struct request_sock
*req
,
1567 struct tcp_fastopen_cookie
*foc
);
1568 void tcp_fastopen_init_key_once(bool publish
);
1569 bool tcp_fastopen_cookie_check(struct sock
*sk
, u16
*mss
,
1570 struct tcp_fastopen_cookie
*cookie
);
1571 bool tcp_fastopen_defer_connect(struct sock
*sk
, int *err
);
1572 #define TCP_FASTOPEN_KEY_LENGTH 16
1574 /* Fastopen key context */
1575 struct tcp_fastopen_context
{
1576 struct crypto_cipher
*tfm
;
1577 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1578 struct rcu_head rcu
;
1581 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout
;
1582 void tcp_fastopen_active_disable(struct sock
*sk
);
1583 bool tcp_fastopen_active_should_disable(struct sock
*sk
);
1584 void tcp_fastopen_active_disable_ofo_check(struct sock
*sk
);
1585 void tcp_fastopen_active_timeout_reset(void);
1587 /* Latencies incurred by various limits for a sender. They are
1588 * chronograph-like stats that are mutually exclusive.
1592 TCP_CHRONO_BUSY
, /* Actively sending data (non-empty write queue) */
1593 TCP_CHRONO_RWND_LIMITED
, /* Stalled by insufficient receive window */
1594 TCP_CHRONO_SNDBUF_LIMITED
, /* Stalled by insufficient send buffer */
1598 void tcp_chrono_start(struct sock
*sk
, const enum tcp_chrono type
);
1599 void tcp_chrono_stop(struct sock
*sk
, const enum tcp_chrono type
);
1601 /* write queue abstraction */
1602 static inline void tcp_write_queue_purge(struct sock
*sk
)
1604 struct sk_buff
*skb
;
1606 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1607 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1608 sk_wmem_free_skb(sk
, skb
);
1610 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1613 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1615 return skb_peek(&sk
->sk_write_queue
);
1618 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1620 return skb_peek_tail(&sk
->sk_write_queue
);
1623 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1624 const struct sk_buff
*skb
)
1626 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1629 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1630 const struct sk_buff
*skb
)
1632 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1635 #define tcp_for_write_queue(skb, sk) \
1636 skb_queue_walk(&(sk)->sk_write_queue, skb)
1638 #define tcp_for_write_queue_from(skb, sk) \
1639 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1641 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1642 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1644 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1646 return sk
->sk_send_head
;
1649 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1650 const struct sk_buff
*skb
)
1652 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1655 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1657 if (tcp_skb_is_last(sk
, skb
))
1658 sk
->sk_send_head
= NULL
;
1660 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1663 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1665 if (sk
->sk_send_head
== skb_unlinked
) {
1666 sk
->sk_send_head
= NULL
;
1667 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1669 if (tcp_sk(sk
)->highest_sack
== skb_unlinked
)
1670 tcp_sk(sk
)->highest_sack
= NULL
;
1673 static inline void tcp_init_send_head(struct sock
*sk
)
1675 sk
->sk_send_head
= NULL
;
1678 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1680 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1683 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1685 __tcp_add_write_queue_tail(sk
, skb
);
1687 /* Queue it, remembering where we must start sending. */
1688 if (sk
->sk_send_head
== NULL
) {
1689 sk
->sk_send_head
= skb
;
1690 tcp_chrono_start(sk
, TCP_CHRONO_BUSY
);
1692 if (tcp_sk(sk
)->highest_sack
== NULL
)
1693 tcp_sk(sk
)->highest_sack
= skb
;
1697 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1699 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1702 /* Insert buff after skb on the write queue of sk. */
1703 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1704 struct sk_buff
*buff
,
1707 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1710 /* Insert new before skb on the write queue of sk. */
1711 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1712 struct sk_buff
*skb
,
1715 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1717 if (sk
->sk_send_head
== skb
)
1718 sk
->sk_send_head
= new;
1721 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1723 __skb_unlink(skb
, &sk
->sk_write_queue
);
1726 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1728 return skb_queue_empty(&sk
->sk_write_queue
);
1731 static inline void tcp_push_pending_frames(struct sock
*sk
)
1733 if (tcp_send_head(sk
)) {
1734 struct tcp_sock
*tp
= tcp_sk(sk
);
1736 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1740 /* Start sequence of the skb just after the highest skb with SACKed
1741 * bit, valid only if sacked_out > 0 or when the caller has ensured
1742 * validity by itself.
1744 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1746 if (!tp
->sacked_out
)
1749 if (tp
->highest_sack
== NULL
)
1752 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1755 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1757 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1758 tcp_write_queue_next(sk
, skb
);
1761 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1763 return tcp_sk(sk
)->highest_sack
;
1766 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1768 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1771 /* Called when old skb is about to be deleted (to be combined with new skb) */
1772 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1773 struct sk_buff
*old
,
1774 struct sk_buff
*new)
1776 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1777 tcp_sk(sk
)->highest_sack
= new;
1780 /* This helper checks if socket has IP_TRANSPARENT set */
1781 static inline bool inet_sk_transparent(const struct sock
*sk
)
1783 switch (sk
->sk_state
) {
1785 return inet_twsk(sk
)->tw_transparent
;
1786 case TCP_NEW_SYN_RECV
:
1787 return inet_rsk(inet_reqsk(sk
))->no_srccheck
;
1789 return inet_sk(sk
)->transparent
;
1792 /* Determines whether this is a thin stream (which may suffer from
1793 * increased latency). Used to trigger latency-reducing mechanisms.
1795 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1797 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1801 enum tcp_seq_states
{
1802 TCP_SEQ_STATE_LISTENING
,
1803 TCP_SEQ_STATE_ESTABLISHED
,
1806 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1808 struct tcp_seq_afinfo
{
1811 const struct file_operations
*seq_fops
;
1812 struct seq_operations seq_ops
;
1815 struct tcp_iter_state
{
1816 struct seq_net_private p
;
1818 enum tcp_seq_states state
;
1819 struct sock
*syn_wait_sk
;
1820 int bucket
, offset
, sbucket
, num
;
1824 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1825 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1827 extern struct request_sock_ops tcp_request_sock_ops
;
1828 extern struct request_sock_ops tcp6_request_sock_ops
;
1830 void tcp_v4_destroy_sock(struct sock
*sk
);
1832 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1833 netdev_features_t features
);
1834 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1835 int tcp_gro_complete(struct sk_buff
*skb
);
1837 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1839 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1841 struct net
*net
= sock_net((struct sock
*)tp
);
1842 return tp
->notsent_lowat
?: net
->ipv4
.sysctl_tcp_notsent_lowat
;
1845 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1847 const struct tcp_sock
*tp
= tcp_sk(sk
);
1848 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1850 return notsent_bytes
< tcp_notsent_lowat(tp
);
1853 #ifdef CONFIG_PROC_FS
1854 int tcp4_proc_init(void);
1855 void tcp4_proc_exit(void);
1858 int tcp_rtx_synack(const struct sock
*sk
, struct request_sock
*req
);
1859 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1860 const struct tcp_request_sock_ops
*af_ops
,
1861 struct sock
*sk
, struct sk_buff
*skb
);
1863 /* TCP af-specific functions */
1864 struct tcp_sock_af_ops
{
1865 #ifdef CONFIG_TCP_MD5SIG
1866 struct tcp_md5sig_key
*(*md5_lookup
) (const struct sock
*sk
,
1867 const struct sock
*addr_sk
);
1868 int (*calc_md5_hash
)(char *location
,
1869 const struct tcp_md5sig_key
*md5
,
1870 const struct sock
*sk
,
1871 const struct sk_buff
*skb
);
1872 int (*md5_parse
)(struct sock
*sk
,
1874 char __user
*optval
,
1879 struct tcp_request_sock_ops
{
1881 #ifdef CONFIG_TCP_MD5SIG
1882 struct tcp_md5sig_key
*(*req_md5_lookup
)(const struct sock
*sk
,
1883 const struct sock
*addr_sk
);
1884 int (*calc_md5_hash
) (char *location
,
1885 const struct tcp_md5sig_key
*md5
,
1886 const struct sock
*sk
,
1887 const struct sk_buff
*skb
);
1889 void (*init_req
)(struct request_sock
*req
,
1890 const struct sock
*sk_listener
,
1891 struct sk_buff
*skb
);
1892 #ifdef CONFIG_SYN_COOKIES
1893 __u32 (*cookie_init_seq
)(const struct sk_buff
*skb
,
1896 struct dst_entry
*(*route_req
)(const struct sock
*sk
, struct flowi
*fl
,
1897 const struct request_sock
*req
);
1898 u32 (*init_seq
)(const struct sk_buff
*skb
);
1899 u32 (*init_ts_off
)(const struct net
*net
, const struct sk_buff
*skb
);
1900 int (*send_synack
)(const struct sock
*sk
, struct dst_entry
*dst
,
1901 struct flowi
*fl
, struct request_sock
*req
,
1902 struct tcp_fastopen_cookie
*foc
,
1903 enum tcp_synack_type synack_type
);
1906 #ifdef CONFIG_SYN_COOKIES
1907 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1908 const struct sock
*sk
, struct sk_buff
*skb
,
1911 tcp_synq_overflow(sk
);
1912 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_SYNCOOKIESSENT
);
1913 return ops
->cookie_init_seq(skb
, mss
);
1916 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1917 const struct sock
*sk
, struct sk_buff
*skb
,
1924 int tcpv4_offload_init(void);
1926 void tcp_v4_init(void);
1927 void tcp_init(void);
1929 /* tcp_recovery.c */
1930 extern void tcp_rack_mark_lost(struct sock
*sk
);
1931 extern void tcp_rack_advance(struct tcp_sock
*tp
, u8 sacked
, u32 end_seq
,
1933 extern void tcp_rack_reo_timeout(struct sock
*sk
);
1935 /* At how many usecs into the future should the RTO fire? */
1936 static inline s64
tcp_rto_delta_us(const struct sock
*sk
)
1938 const struct sk_buff
*skb
= tcp_write_queue_head(sk
);
1939 u32 rto
= inet_csk(sk
)->icsk_rto
;
1940 u64 rto_time_stamp_us
= skb
->skb_mstamp
+ jiffies_to_usecs(rto
);
1942 return rto_time_stamp_us
- tcp_sk(sk
)->tcp_mstamp
;
1946 * Save and compile IPv4 options, return a pointer to it
1948 static inline struct ip_options_rcu
*tcp_v4_save_options(struct net
*net
,
1949 struct sk_buff
*skb
)
1951 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
1952 struct ip_options_rcu
*dopt
= NULL
;
1955 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
1957 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
1958 if (dopt
&& __ip_options_echo(net
, &dopt
->opt
, skb
, opt
)) {
1966 /* locally generated TCP pure ACKs have skb->truesize == 2
1967 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1968 * This is much faster than dissecting the packet to find out.
1969 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1971 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
1973 return skb
->truesize
== 2;
1976 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)
1981 static inline int tcp_inq(struct sock
*sk
)
1983 struct tcp_sock
*tp
= tcp_sk(sk
);
1986 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
1988 } else if (sock_flag(sk
, SOCK_URGINLINE
) ||
1990 before(tp
->urg_seq
, tp
->copied_seq
) ||
1991 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
1993 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
1995 /* Subtract 1, if FIN was received */
1996 if (answ
&& sock_flag(sk
, SOCK_DONE
))
1999 answ
= tp
->urg_seq
- tp
->copied_seq
;
2005 int tcp_peek_len(struct socket
*sock
);
2007 static inline void tcp_segs_in(struct tcp_sock
*tp
, const struct sk_buff
*skb
)
2011 segs_in
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
2012 tp
->segs_in
+= segs_in
;
2013 if (skb
->len
> tcp_hdrlen(skb
))
2014 tp
->data_segs_in
+= segs_in
;
2018 * TCP listen path runs lockless.
2019 * We forced "struct sock" to be const qualified to make sure
2020 * we don't modify one of its field by mistake.
2021 * Here, we increment sk_drops which is an atomic_t, so we can safely
2022 * make sock writable again.
2024 static inline void tcp_listendrop(const struct sock
*sk
)
2026 atomic_inc(&((struct sock
*)sk
)->sk_drops
);
2027 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
2030 enum hrtimer_restart
tcp_pace_kick(struct hrtimer
*timer
);
2033 * Interface for adding Upper Level Protocols over TCP
2036 #define TCP_ULP_NAME_MAX 16
2037 #define TCP_ULP_MAX 128
2038 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2040 struct tcp_ulp_ops
{
2041 struct list_head list
;
2043 /* initialize ulp */
2044 int (*init
)(struct sock
*sk
);
2046 void (*release
)(struct sock
*sk
);
2048 char name
[TCP_ULP_NAME_MAX
];
2049 struct module
*owner
;
2051 int tcp_register_ulp(struct tcp_ulp_ops
*type
);
2052 void tcp_unregister_ulp(struct tcp_ulp_ops
*type
);
2053 int tcp_set_ulp(struct sock
*sk
, const char *name
);
2054 void tcp_get_available_ulp(char *buf
, size_t len
);
2055 void tcp_cleanup_ulp(struct sock
*sk
);
2057 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2058 * is < 0, then the BPF op failed (for example if the loaded BPF
2059 * program does not support the chosen operation or there is no BPF
2063 static inline int tcp_call_bpf(struct sock
*sk
, int op
)
2065 struct bpf_sock_ops_kern sock_ops
;
2068 if (sk_fullsock(sk
))
2069 sock_owned_by_me(sk
);
2071 memset(&sock_ops
, 0, sizeof(sock_ops
));
2075 ret
= BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops
);
2077 ret
= sock_ops
.reply
;
2083 static inline int tcp_call_bpf(struct sock
*sk
, int op
)
2089 static inline u32
tcp_timeout_init(struct sock
*sk
)
2093 timeout
= tcp_call_bpf(sk
, BPF_SOCK_OPS_TIMEOUT_INIT
);
2096 timeout
= TCP_TIMEOUT_INIT
;
2100 static inline u32
tcp_rwnd_init_bpf(struct sock
*sk
)
2104 rwnd
= tcp_call_bpf(sk
, BPF_SOCK_OPS_RWND_INIT
);
2111 static inline bool tcp_bpf_ca_needs_ecn(struct sock
*sk
)
2113 return (tcp_call_bpf(sk
, BPF_SOCK_OPS_NEEDS_ECN
) == 1);