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 extern struct inet_hashinfo tcp_hashinfo
;
51 extern struct percpu_counter tcp_orphan_count
;
52 void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
61 #define MAX_TCP_WINDOW 32767U
63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 1024
69 /* probing interval, default to 10 minutes as per RFC4821 */
70 #define TCP_PROBE_INTERVAL 600
72 /* Specify interval when tcp mtu probing will stop */
73 #define TCP_PROBE_THRESHOLD 8
75 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
76 #define TCP_FASTRETRANS_THRESH 3
78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
79 #define TCP_MAX_QUICKACKS 16U
81 /* Maximal number of window scale according to RFC1323 */
82 #define TCP_MAX_WSCALE 14U
85 #define TCP_URG_VALID 0x0100
86 #define TCP_URG_NOTYET 0x0200
87 #define TCP_URG_READ 0x0400
89 #define TCP_RETR1 3 /*
90 * This is how many retries it does before it
91 * tries to figure out if the gateway is
92 * down. Minimal RFC value is 3; it corresponds
93 * to ~3sec-8min depending on RTO.
96 #define TCP_RETR2 15 /*
97 * This should take at least
98 * 90 minutes to time out.
99 * RFC1122 says that the limit is 100 sec.
100 * 15 is ~13-30min depending on RTO.
103 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
104 * when active opening a connection.
105 * RFC1122 says the minimum retry MUST
106 * be at least 180secs. Nevertheless
107 * this value is corresponding to
108 * 63secs of retransmission with the
109 * current initial RTO.
112 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
113 * when passive opening a connection.
114 * This is corresponding to 31secs of
115 * retransmission with the current
119 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
120 * state, about 60 seconds */
121 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
122 /* BSD style FIN_WAIT2 deadlock breaker.
123 * It used to be 3min, new value is 60sec,
124 * to combine FIN-WAIT-2 timeout with
128 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
130 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
131 #define TCP_ATO_MIN ((unsigned)(HZ/25))
133 #define TCP_DELACK_MIN 4U
134 #define TCP_ATO_MIN 4U
136 #define TCP_RTO_MAX ((unsigned)(120*HZ))
137 #define TCP_RTO_MIN ((unsigned)(HZ/5))
138 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
139 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
140 * used as a fallback RTO for the
141 * initial data transmission if no
142 * valid RTT sample has been acquired,
143 * most likely due to retrans in 3WHS.
146 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
147 * for local resources.
149 #define TCP_REO_TIMEOUT_MIN (2000) /* Min RACK reordering timeout in usec */
151 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153 #define TCP_KEEPALIVE_INTVL (75*HZ)
155 #define MAX_TCP_KEEPIDLE 32767
156 #define MAX_TCP_KEEPINTVL 32767
157 #define MAX_TCP_KEEPCNT 127
158 #define MAX_TCP_SYNCNT 127
160 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
162 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
169 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
177 #define TCPOPT_NOP 1 /* Padding */
178 #define TCPOPT_EOL 0 /* End of options */
179 #define TCPOPT_MSS 2 /* Segment size negotiating */
180 #define TCPOPT_WINDOW 3 /* Window scaling */
181 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182 #define TCPOPT_SACK 5 /* SACK Block */
183 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
184 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
185 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
186 #define TCPOPT_EXP 254 /* Experimental */
187 /* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
190 #define TCPOPT_FASTOPEN_MAGIC 0xF989
196 #define TCPOLEN_MSS 4
197 #define TCPOLEN_WINDOW 3
198 #define TCPOLEN_SACK_PERM 2
199 #define TCPOLEN_TIMESTAMP 10
200 #define TCPOLEN_MD5SIG 18
201 #define TCPOLEN_FASTOPEN_BASE 2
202 #define TCPOLEN_EXP_FASTOPEN_BASE 4
204 /* But this is what stacks really send out. */
205 #define TCPOLEN_TSTAMP_ALIGNED 12
206 #define TCPOLEN_WSCALE_ALIGNED 4
207 #define TCPOLEN_SACKPERM_ALIGNED 4
208 #define TCPOLEN_SACK_BASE 2
209 #define TCPOLEN_SACK_BASE_ALIGNED 4
210 #define TCPOLEN_SACK_PERBLOCK 8
211 #define TCPOLEN_MD5SIG_ALIGNED 20
212 #define TCPOLEN_MSS_ALIGNED 4
214 /* Flags in tp->nonagle */
215 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
216 #define TCP_NAGLE_CORK 2 /* Socket is corked */
217 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
219 /* TCP thin-stream limits */
220 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
222 /* TCP initial congestion window as per rfc6928 */
223 #define TCP_INIT_CWND 10
225 /* Bit Flags for sysctl_tcp_fastopen */
226 #define TFO_CLIENT_ENABLE 1
227 #define TFO_SERVER_ENABLE 2
228 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
230 /* Accept SYN data w/o any cookie option */
231 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
233 /* Force enable TFO on all listeners, i.e., not requiring the
234 * TCP_FASTOPEN socket option.
236 #define TFO_SERVER_WO_SOCKOPT1 0x400
239 /* sysctl variables for tcp */
240 extern int sysctl_tcp_timestamps
;
241 extern int sysctl_tcp_fastopen
;
242 extern int sysctl_tcp_retrans_collapse
;
243 extern int sysctl_tcp_stdurg
;
244 extern int sysctl_tcp_rfc1337
;
245 extern int sysctl_tcp_abort_on_overflow
;
246 extern int sysctl_tcp_max_orphans
;
247 extern int sysctl_tcp_fack
;
248 extern int sysctl_tcp_reordering
;
249 extern int sysctl_tcp_max_reordering
;
250 extern int sysctl_tcp_dsack
;
251 extern long sysctl_tcp_mem
[3];
252 extern int sysctl_tcp_wmem
[3];
253 extern int sysctl_tcp_rmem
[3];
254 extern int sysctl_tcp_app_win
;
255 extern int sysctl_tcp_adv_win_scale
;
256 extern int sysctl_tcp_frto
;
257 extern int sysctl_tcp_low_latency
;
258 extern int sysctl_tcp_nometrics_save
;
259 extern int sysctl_tcp_moderate_rcvbuf
;
260 extern int sysctl_tcp_tso_win_divisor
;
261 extern int sysctl_tcp_workaround_signed_windows
;
262 extern int sysctl_tcp_slow_start_after_idle
;
263 extern int sysctl_tcp_thin_linear_timeouts
;
264 extern int sysctl_tcp_thin_dupack
;
265 extern int sysctl_tcp_early_retrans
;
266 extern int sysctl_tcp_recovery
;
267 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
269 extern int sysctl_tcp_limit_output_bytes
;
270 extern int sysctl_tcp_challenge_ack_limit
;
271 extern int sysctl_tcp_min_tso_segs
;
272 extern int sysctl_tcp_min_rtt_wlen
;
273 extern int sysctl_tcp_autocorking
;
274 extern int sysctl_tcp_invalid_ratelimit
;
275 extern int sysctl_tcp_pacing_ss_ratio
;
276 extern int sysctl_tcp_pacing_ca_ratio
;
278 extern atomic_long_t tcp_memory_allocated
;
279 extern struct percpu_counter tcp_sockets_allocated
;
280 extern int tcp_memory_pressure
;
282 /* optimized version of sk_under_memory_pressure() for TCP sockets */
283 static inline bool tcp_under_memory_pressure(const struct sock
*sk
)
285 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
286 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
289 return 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 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
318 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
320 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
321 int orphans
= percpu_counter_read_positive(ocp
);
323 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
324 orphans
= percpu_counter_sum_positive(ocp
);
325 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
331 bool tcp_check_oom(struct sock
*sk
, int shift
);
334 extern struct proto tcp_prot
;
336 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
337 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
338 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
339 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
341 void tcp_tasklet_init(void);
343 void tcp_v4_err(struct sk_buff
*skb
, u32
);
345 void tcp_shutdown(struct sock
*sk
, int how
);
347 void tcp_v4_early_demux(struct sk_buff
*skb
);
348 int tcp_v4_rcv(struct sk_buff
*skb
);
350 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
351 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
352 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
354 void tcp_release_cb(struct sock
*sk
);
355 void tcp_wfree(struct sk_buff
*skb
);
356 void tcp_write_timer_handler(struct sock
*sk
);
357 void tcp_delack_timer_handler(struct sock
*sk
);
358 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
359 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
);
360 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
361 const struct tcphdr
*th
, unsigned int len
);
362 void tcp_rcv_space_adjust(struct sock
*sk
);
363 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
364 void tcp_twsk_destructor(struct sock
*sk
);
365 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
366 struct pipe_inode_info
*pipe
, size_t len
,
369 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
370 const unsigned int pkts
)
372 struct inet_connection_sock
*icsk
= inet_csk(sk
);
374 if (icsk
->icsk_ack
.quick
) {
375 if (pkts
>= icsk
->icsk_ack
.quick
) {
376 icsk
->icsk_ack
.quick
= 0;
377 /* Leaving quickack mode we deflate ATO. */
378 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
380 icsk
->icsk_ack
.quick
-= pkts
;
385 #define TCP_ECN_QUEUE_CWR 2
386 #define TCP_ECN_DEMAND_CWR 4
387 #define TCP_ECN_SEEN 8
397 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
399 const struct tcphdr
*th
);
400 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
401 struct request_sock
*req
, bool fastopen
);
402 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
403 struct sk_buff
*skb
);
404 void tcp_enter_loss(struct sock
*sk
);
405 void tcp_cwnd_reduction(struct sock
*sk
, int newly_acked_sacked
, int flag
);
406 void tcp_clear_retrans(struct tcp_sock
*tp
);
407 void tcp_update_metrics(struct sock
*sk
);
408 void tcp_init_metrics(struct sock
*sk
);
409 void tcp_metrics_init(void);
410 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
);
411 void tcp_disable_fack(struct tcp_sock
*tp
);
412 void tcp_close(struct sock
*sk
, long timeout
);
413 void tcp_init_sock(struct sock
*sk
);
414 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
,
415 struct poll_table_struct
*wait
);
416 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
417 char __user
*optval
, int __user
*optlen
);
418 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
419 char __user
*optval
, unsigned int optlen
);
420 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
421 char __user
*optval
, int __user
*optlen
);
422 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
423 char __user
*optval
, unsigned int optlen
);
424 void tcp_set_keepalive(struct sock
*sk
, int val
);
425 void tcp_syn_ack_timeout(const struct request_sock
*req
);
426 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
427 int flags
, int *addr_len
);
428 void tcp_parse_options(const struct net
*net
, const struct sk_buff
*skb
,
429 struct tcp_options_received
*opt_rx
,
430 int estab
, struct tcp_fastopen_cookie
*foc
);
431 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
434 * TCP v4 functions exported for the inet6 API
437 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
438 void tcp_v4_mtu_reduced(struct sock
*sk
);
439 void tcp_req_err(struct sock
*sk
, u32 seq
, bool abort
);
440 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
441 struct sock
*tcp_create_openreq_child(const struct sock
*sk
,
442 struct request_sock
*req
,
443 struct sk_buff
*skb
);
444 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
445 struct sock
*tcp_v4_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
446 struct request_sock
*req
,
447 struct dst_entry
*dst
,
448 struct request_sock
*req_unhash
,
450 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
451 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
452 int tcp_connect(struct sock
*sk
);
453 enum tcp_synack_type
{
458 struct sk_buff
*tcp_make_synack(const struct sock
*sk
, struct dst_entry
*dst
,
459 struct request_sock
*req
,
460 struct tcp_fastopen_cookie
*foc
,
461 enum tcp_synack_type synack_type
);
462 int tcp_disconnect(struct sock
*sk
, int flags
);
464 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
465 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
466 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
468 /* From syncookies.c */
469 struct sock
*tcp_get_cookie_sock(struct sock
*sk
, struct sk_buff
*skb
,
470 struct request_sock
*req
,
471 struct dst_entry
*dst
, u32 tsoff
);
472 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
474 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
475 #ifdef CONFIG_SYN_COOKIES
477 /* Syncookies use a monotonic timer which increments every 60 seconds.
478 * This counter is used both as a hash input and partially encoded into
479 * the cookie value. A cookie is only validated further if the delta
480 * between the current counter value and the encoded one is less than this,
481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
482 * the counter advances immediately after a cookie is generated).
484 #define MAX_SYNCOOKIE_AGE 2
485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
488 /* syncookies: remember time of last synqueue overflow
489 * But do not dirty this field too often (once per second is enough)
490 * It is racy as we do not hold a lock, but race is very minor.
492 static inline void tcp_synq_overflow(const struct sock
*sk
)
494 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
495 unsigned long now
= jiffies
;
497 if (time_after(now
, last_overflow
+ HZ
))
498 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= now
;
501 /* syncookies: no recent synqueue overflow on this listening socket? */
502 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
504 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
506 return time_after(jiffies
, last_overflow
+ TCP_SYNCOOKIE_VALID
);
509 static inline u32
tcp_cookie_time(void)
511 u64 val
= get_jiffies_64();
513 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
517 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
519 __u32
cookie_v4_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
520 u64
cookie_init_timestamp(struct request_sock
*req
);
521 bool cookie_timestamp_decode(const struct net
*net
,
522 struct tcp_options_received
*opt
);
523 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
524 const struct net
*net
, const struct dst_entry
*dst
);
526 /* From net/ipv6/syncookies.c */
527 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
529 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
531 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
532 const struct tcphdr
*th
, u16
*mssp
);
533 __u32
cookie_v6_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
537 u32
tcp_tso_autosize(const struct sock
*sk
, unsigned int mss_now
,
539 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
541 bool tcp_may_send_now(struct sock
*sk
);
542 int __tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
543 int tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
544 void tcp_retransmit_timer(struct sock
*sk
);
545 void tcp_xmit_retransmit_queue(struct sock
*);
546 void tcp_simple_retransmit(struct sock
*);
547 void tcp_enter_recovery(struct sock
*sk
, bool ece_ack
);
548 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
549 int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int, gfp_t
);
551 void tcp_send_probe0(struct sock
*);
552 void tcp_send_partial(struct sock
*);
553 int tcp_write_wakeup(struct sock
*, int mib
);
554 void tcp_send_fin(struct sock
*sk
);
555 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
556 int tcp_send_synack(struct sock
*);
557 void tcp_push_one(struct sock
*, unsigned int mss_now
);
558 void tcp_send_ack(struct sock
*sk
);
559 void tcp_send_delayed_ack(struct sock
*sk
);
560 void tcp_send_loss_probe(struct sock
*sk
);
561 bool tcp_schedule_loss_probe(struct sock
*sk
);
562 void tcp_skb_collapse_tstamp(struct sk_buff
*skb
,
563 const struct sk_buff
*next_skb
);
566 void tcp_rearm_rto(struct sock
*sk
);
567 void tcp_synack_rtt_meas(struct sock
*sk
, struct request_sock
*req
);
568 void tcp_reset(struct sock
*sk
);
569 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock
*tp
, struct sk_buff
*skb
);
570 void tcp_fin(struct sock
*sk
);
573 void tcp_init_xmit_timers(struct sock
*);
574 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
576 hrtimer_cancel(&tcp_sk(sk
)->pacing_timer
);
577 inet_csk_clear_xmit_timers(sk
);
580 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
581 unsigned int tcp_current_mss(struct sock
*sk
);
583 /* Bound MSS / TSO packet size with the half of the window */
584 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
588 /* When peer uses tiny windows, there is no use in packetizing
589 * to sub-MSS pieces for the sake of SWS or making sure there
590 * are enough packets in the pipe for fast recovery.
592 * On the other hand, for extremely large MSS devices, handling
593 * smaller than MSS windows in this way does make sense.
595 if (tp
->max_window
> TCP_MSS_DEFAULT
)
596 cutoff
= (tp
->max_window
>> 1);
598 cutoff
= tp
->max_window
;
600 if (cutoff
&& pktsize
> cutoff
)
601 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
607 void tcp_get_info(struct sock
*, struct tcp_info
*);
609 /* Read 'sendfile()'-style from a TCP socket */
610 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
611 sk_read_actor_t recv_actor
);
613 void tcp_initialize_rcv_mss(struct sock
*sk
);
615 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
616 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
617 void tcp_mtup_init(struct sock
*sk
);
618 void tcp_init_buffer_space(struct sock
*sk
);
620 static inline void tcp_bound_rto(const struct sock
*sk
)
622 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
623 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
626 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
628 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
631 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
633 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
634 ntohl(TCP_FLAG_ACK
) |
638 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
640 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
643 static inline void tcp_fast_path_check(struct sock
*sk
)
645 struct tcp_sock
*tp
= tcp_sk(sk
);
647 if (RB_EMPTY_ROOT(&tp
->out_of_order_queue
) &&
649 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
651 tcp_fast_path_on(tp
);
654 /* Compute the actual rto_min value */
655 static inline u32
tcp_rto_min(struct sock
*sk
)
657 const struct dst_entry
*dst
= __sk_dst_get(sk
);
658 u32 rto_min
= TCP_RTO_MIN
;
660 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
661 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
665 static inline u32
tcp_rto_min_us(struct sock
*sk
)
667 return jiffies_to_usecs(tcp_rto_min(sk
));
670 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
672 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
675 /* Minimum RTT in usec. ~0 means not available. */
676 static inline u32
tcp_min_rtt(const struct tcp_sock
*tp
)
678 return minmax_get(&tp
->rtt_min
);
681 /* Compute the actual receive window we are currently advertising.
682 * Rcv_nxt can be after the window if our peer push more data
683 * than the offered window.
685 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
687 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
694 /* Choose a new window, without checks for shrinking, and without
695 * scaling applied to the result. The caller does these things
696 * if necessary. This is a "raw" window selection.
698 u32
__tcp_select_window(struct sock
*sk
);
700 void tcp_send_window_probe(struct sock
*sk
);
702 /* TCP uses 32bit jiffies to save some space.
703 * Note that this is different from tcp_time_stamp, which
704 * historically has been the same until linux-4.13.
706 #define tcp_jiffies32 ((u32)jiffies)
709 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
710 * It is no longer tied to jiffies, but to 1 ms clock.
711 * Note: double check if you want to use tcp_jiffies32 instead of this.
713 #define TCP_TS_HZ 1000
715 static inline u64
tcp_clock_ns(void)
717 return local_clock();
720 static inline u64
tcp_clock_us(void)
722 return div_u64(tcp_clock_ns(), NSEC_PER_USEC
);
725 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
726 static inline u32
tcp_time_stamp(const struct tcp_sock
*tp
)
728 return div_u64(tp
->tcp_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
731 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
732 static inline u32
tcp_time_stamp_raw(void)
734 return div_u64(tcp_clock_ns(), NSEC_PER_SEC
/ TCP_TS_HZ
);
738 /* Refresh 1us clock of a TCP socket,
739 * ensuring monotically increasing values.
741 static inline void tcp_mstamp_refresh(struct tcp_sock
*tp
)
743 u64 val
= tcp_clock_us();
745 if (val
> tp
->tcp_mstamp
)
746 tp
->tcp_mstamp
= val
;
749 static inline u32
tcp_stamp_us_delta(u64 t1
, u64 t0
)
751 return max_t(s64
, t1
- t0
, 0);
754 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
756 return div_u64(skb
->skb_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
760 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
762 #define TCPHDR_FIN 0x01
763 #define TCPHDR_SYN 0x02
764 #define TCPHDR_RST 0x04
765 #define TCPHDR_PSH 0x08
766 #define TCPHDR_ACK 0x10
767 #define TCPHDR_URG 0x20
768 #define TCPHDR_ECE 0x40
769 #define TCPHDR_CWR 0x80
771 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
773 /* This is what the send packet queuing engine uses to pass
774 * TCP per-packet control information to the transmission code.
775 * We also store the host-order sequence numbers in here too.
776 * This is 44 bytes if IPV6 is enabled.
777 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
780 __u32 seq
; /* Starting sequence number */
781 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
783 /* Note : tcp_tw_isn is used in input path only
784 * (isn chosen by tcp_timewait_state_process())
786 * tcp_gso_segs/size are used in write queue only,
787 * cf tcp_skb_pcount()/tcp_skb_mss()
795 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
797 __u8 sacked
; /* State flags for SACK/FACK. */
798 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
799 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
800 #define TCPCB_LOST 0x04 /* SKB is lost */
801 #define TCPCB_TAGBITS 0x07 /* All tag bits */
802 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
803 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
804 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
807 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
808 __u8 txstamp_ack
:1, /* Record TX timestamp for ack? */
809 eor
:1, /* Is skb MSG_EOR marked? */
811 __u32 ack_seq
; /* Sequence number ACK'd */
814 /* There is space for up to 24 bytes */
815 __u32 in_flight
:30,/* Bytes in flight at transmit */
816 is_app_limited
:1, /* cwnd not fully used? */
818 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
820 /* start of send pipeline phase */
822 /* when we reached the "delivered" count */
823 u64 delivered_mstamp
;
824 } tx
; /* only used for outgoing skbs */
826 struct inet_skb_parm h4
;
827 #if IS_ENABLED(CONFIG_IPV6)
828 struct inet6_skb_parm h6
;
830 } header
; /* For incoming skbs */
834 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
837 #if IS_ENABLED(CONFIG_IPV6)
838 /* This is the variant of inet6_iif() that must be used by TCP,
839 * as TCP moves IP6CB into a different location in skb->cb[]
841 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
843 bool l3_slave
= ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
);
845 return l3_slave
? skb
->skb_iif
: TCP_SKB_CB(skb
)->header
.h6
.iif
;
849 /* TCP_SKB_CB reference means this can not be used from early demux */
850 static inline bool inet_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
852 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
853 if (!net
->ipv4
.sysctl_tcp_l3mdev_accept
&&
854 skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
860 /* Due to TSO, an SKB can be composed of multiple actual
861 * packets. To keep these tracked properly, we use this.
863 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
865 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
868 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
870 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
873 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
875 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
878 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
879 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
881 return TCP_SKB_CB(skb
)->tcp_gso_size
;
884 static inline bool tcp_skb_can_collapse_to(const struct sk_buff
*skb
)
886 return likely(!TCP_SKB_CB(skb
)->eor
);
889 /* Events passed to congestion control interface */
891 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
892 CA_EVENT_CWND_RESTART
, /* congestion window restart */
893 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
894 CA_EVENT_LOSS
, /* loss timeout */
895 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
896 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
897 CA_EVENT_DELAYED_ACK
, /* Delayed ack is sent */
898 CA_EVENT_NON_DELAYED_ACK
,
901 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
902 enum tcp_ca_ack_event_flags
{
903 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
904 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
905 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
909 * Interface for adding new TCP congestion control handlers
911 #define TCP_CA_NAME_MAX 16
912 #define TCP_CA_MAX 128
913 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
915 #define TCP_CA_UNSPEC 0
917 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
918 #define TCP_CONG_NON_RESTRICTED 0x1
919 /* Requires ECN/ECT set on all packets */
920 #define TCP_CONG_NEEDS_ECN 0x2
930 /* A rate sample measures the number of (original/retransmitted) data
931 * packets delivered "delivered" over an interval of time "interval_us".
932 * The tcp_rate.c code fills in the rate sample, and congestion
933 * control modules that define a cong_control function to run at the end
934 * of ACK processing can optionally chose to consult this sample when
935 * setting cwnd and pacing rate.
936 * A sample is invalid if "delivered" or "interval_us" is negative.
939 u64 prior_mstamp
; /* starting timestamp for interval */
940 u32 prior_delivered
; /* tp->delivered at "prior_mstamp" */
941 s32 delivered
; /* number of packets delivered over interval */
942 long interval_us
; /* time for tp->delivered to incr "delivered" */
943 long rtt_us
; /* RTT of last (S)ACKed packet (or -1) */
944 int losses
; /* number of packets marked lost upon ACK */
945 u32 acked_sacked
; /* number of packets newly (S)ACKed upon ACK */
946 u32 prior_in_flight
; /* in flight before this ACK */
947 bool is_app_limited
; /* is sample from packet with bubble in pipe? */
948 bool is_retrans
; /* is sample from retransmission? */
951 struct tcp_congestion_ops
{
952 struct list_head list
;
956 /* initialize private data (optional) */
957 void (*init
)(struct sock
*sk
);
958 /* cleanup private data (optional) */
959 void (*release
)(struct sock
*sk
);
961 /* return slow start threshold (required) */
962 u32 (*ssthresh
)(struct sock
*sk
);
963 /* do new cwnd calculation (required) */
964 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
965 /* call before changing ca_state (optional) */
966 void (*set_state
)(struct sock
*sk
, u8 new_state
);
967 /* call when cwnd event occurs (optional) */
968 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
969 /* call when ack arrives (optional) */
970 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
971 /* new value of cwnd after loss (required) */
972 u32 (*undo_cwnd
)(struct sock
*sk
);
973 /* hook for packet ack accounting (optional) */
974 void (*pkts_acked
)(struct sock
*sk
, const struct ack_sample
*sample
);
975 /* suggest number of segments for each skb to transmit (optional) */
976 u32 (*tso_segs_goal
)(struct sock
*sk
);
977 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
978 u32 (*sndbuf_expand
)(struct sock
*sk
);
979 /* call when packets are delivered to update cwnd and pacing rate,
980 * after all the ca_state processing. (optional)
982 void (*cong_control
)(struct sock
*sk
, const struct rate_sample
*rs
);
983 /* get info for inet_diag (optional) */
984 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
985 union tcp_cc_info
*info
);
987 char name
[TCP_CA_NAME_MAX
];
988 struct module
*owner
;
991 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
992 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
994 void tcp_assign_congestion_control(struct sock
*sk
);
995 void tcp_init_congestion_control(struct sock
*sk
);
996 void tcp_cleanup_congestion_control(struct sock
*sk
);
997 int tcp_set_default_congestion_control(const char *name
);
998 void tcp_get_default_congestion_control(char *name
);
999 void tcp_get_available_congestion_control(char *buf
, size_t len
);
1000 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
1001 int tcp_set_allowed_congestion_control(char *allowed
);
1002 int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
1003 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
1004 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
1006 u32
tcp_reno_ssthresh(struct sock
*sk
);
1007 u32
tcp_reno_undo_cwnd(struct sock
*sk
);
1008 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
1009 extern struct tcp_congestion_ops tcp_reno
;
1011 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
1012 u32
tcp_ca_get_key_by_name(const char *name
, bool *ecn_ca
);
1014 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
1016 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
1022 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
1024 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1026 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
1029 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
1031 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1033 if (icsk
->icsk_ca_ops
->set_state
)
1034 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
1035 icsk
->icsk_ca_state
= ca_state
;
1038 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
1040 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1042 if (icsk
->icsk_ca_ops
->cwnd_event
)
1043 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
1046 /* From tcp_rate.c */
1047 void tcp_rate_skb_sent(struct sock
*sk
, struct sk_buff
*skb
);
1048 void tcp_rate_skb_delivered(struct sock
*sk
, struct sk_buff
*skb
,
1049 struct rate_sample
*rs
);
1050 void tcp_rate_gen(struct sock
*sk
, u32 delivered
, u32 lost
,
1051 struct rate_sample
*rs
);
1052 void tcp_rate_check_app_limited(struct sock
*sk
);
1054 /* These functions determine how the current flow behaves in respect of SACK
1055 * handling. SACK is negotiated with the peer, and therefore it can vary
1056 * between different flows.
1058 * tcp_is_sack - SACK enabled
1059 * tcp_is_reno - No SACK
1060 * tcp_is_fack - FACK enabled, implies SACK enabled
1062 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
1064 return tp
->rx_opt
.sack_ok
;
1067 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
1069 return !tcp_is_sack(tp
);
1072 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
1074 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
1077 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
1079 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
1082 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
1084 return tp
->sacked_out
+ tp
->lost_out
;
1087 /* This determines how many packets are "in the network" to the best
1088 * of our knowledge. In many cases it is conservative, but where
1089 * detailed information is available from the receiver (via SACK
1090 * blocks etc.) we can make more aggressive calculations.
1092 * Use this for decisions involving congestion control, use just
1093 * tp->packets_out to determine if the send queue is empty or not.
1095 * Read this equation as:
1097 * "Packets sent once on transmission queue" MINUS
1098 * "Packets left network, but not honestly ACKed yet" PLUS
1099 * "Packets fast retransmitted"
1101 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1103 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
1106 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1108 static inline bool tcp_in_slow_start(const struct tcp_sock
*tp
)
1110 return tp
->snd_cwnd
< tp
->snd_ssthresh
;
1113 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
1115 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
1118 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
1120 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
1121 (1 << inet_csk(sk
)->icsk_ca_state
);
1124 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1125 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1128 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
1130 const struct tcp_sock
*tp
= tcp_sk(sk
);
1132 if (tcp_in_cwnd_reduction(sk
))
1133 return tp
->snd_ssthresh
;
1135 return max(tp
->snd_ssthresh
,
1136 ((tp
->snd_cwnd
>> 1) +
1137 (tp
->snd_cwnd
>> 2)));
1140 /* Use define here intentionally to get WARN_ON location shown at the caller */
1141 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1143 void tcp_enter_cwr(struct sock
*sk
);
1144 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1146 /* The maximum number of MSS of available cwnd for which TSO defers
1147 * sending if not using sysctl_tcp_tso_win_divisor.
1149 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1154 /* Returns end sequence number of the receiver's advertised window */
1155 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1157 return tp
->snd_una
+ tp
->snd_wnd
;
1160 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1161 * flexible approach. The RFC suggests cwnd should not be raised unless
1162 * it was fully used previously. And that's exactly what we do in
1163 * congestion avoidance mode. But in slow start we allow cwnd to grow
1164 * as long as the application has used half the cwnd.
1166 * cwnd is 10 (IW10), but application sends 9 frames.
1167 * We allow cwnd to reach 18 when all frames are ACKed.
1168 * This check is safe because it's as aggressive as slow start which already
1169 * risks 100% overshoot. The advantage is that we discourage application to
1170 * either send more filler packets or data to artificially blow up the cwnd
1171 * usage, and allow application-limited process to probe bw more aggressively.
1173 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1175 const struct tcp_sock
*tp
= tcp_sk(sk
);
1177 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1178 if (tcp_in_slow_start(tp
))
1179 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1181 return tp
->is_cwnd_limited
;
1184 /* Something is really bad, we could not queue an additional packet,
1185 * because qdisc is full or receiver sent a 0 window.
1186 * We do not want to add fuel to the fire, or abort too early,
1187 * so make sure the timer we arm now is at least 200ms in the future,
1188 * regardless of current icsk_rto value (as it could be ~2ms)
1190 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1192 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1195 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1196 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1197 unsigned long max_when
)
1199 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1201 return (unsigned long)min_t(u64
, when
, max_when
);
1204 static inline void tcp_check_probe_timer(struct sock
*sk
)
1206 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1207 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1208 tcp_probe0_base(sk
), TCP_RTO_MAX
);
1211 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1216 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1222 * Calculate(/check) TCP checksum
1224 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1225 __be32 daddr
, __wsum base
)
1227 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1230 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1232 return __skb_checksum_complete(skb
);
1235 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1237 return !skb_csum_unnecessary(skb
) &&
1238 __tcp_checksum_complete(skb
);
1241 /* Prequeue for VJ style copy to user, combined with checksumming. */
1243 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
1245 tp
->ucopy
.task
= NULL
;
1247 tp
->ucopy
.memory
= 0;
1248 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1251 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
);
1252 bool tcp_add_backlog(struct sock
*sk
, struct sk_buff
*skb
);
1253 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
);
1258 static const char *statename
[]={
1259 "Unused","Established","Syn Sent","Syn Recv",
1260 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1261 "Close Wait","Last ACK","Listen","Closing"
1264 void tcp_set_state(struct sock
*sk
, int state
);
1266 void tcp_done(struct sock
*sk
);
1268 int tcp_abort(struct sock
*sk
, int err
);
1270 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1273 rx_opt
->num_sacks
= 0;
1276 u32
tcp_default_init_rwnd(u32 mss
);
1277 void tcp_cwnd_restart(struct sock
*sk
, s32 delta
);
1279 static inline void tcp_slow_start_after_idle_check(struct sock
*sk
)
1281 const struct tcp_congestion_ops
*ca_ops
= inet_csk(sk
)->icsk_ca_ops
;
1282 struct tcp_sock
*tp
= tcp_sk(sk
);
1285 if (!sysctl_tcp_slow_start_after_idle
|| tp
->packets_out
||
1286 ca_ops
->cong_control
)
1288 delta
= tcp_jiffies32
- tp
->lsndtime
;
1289 if (delta
> inet_csk(sk
)->icsk_rto
)
1290 tcp_cwnd_restart(sk
, delta
);
1293 /* Determine a window scaling and initial window to offer. */
1294 void tcp_select_initial_window(int __space
, __u32 mss
, __u32
*rcv_wnd
,
1295 __u32
*window_clamp
, int wscale_ok
,
1296 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1298 static inline int tcp_win_from_space(int space
)
1300 int tcp_adv_win_scale
= sysctl_tcp_adv_win_scale
;
1302 return tcp_adv_win_scale
<= 0 ?
1303 (space
>>(-tcp_adv_win_scale
)) :
1304 space
- (space
>>tcp_adv_win_scale
);
1307 /* Note: caller must be prepared to deal with negative returns */
1308 static inline int tcp_space(const struct sock
*sk
)
1310 return tcp_win_from_space(sk
->sk_rcvbuf
-
1311 atomic_read(&sk
->sk_rmem_alloc
));
1314 static inline int tcp_full_space(const struct sock
*sk
)
1316 return tcp_win_from_space(sk
->sk_rcvbuf
);
1319 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1320 const struct sock
*sk_listener
,
1321 const struct dst_entry
*dst
);
1323 void tcp_enter_memory_pressure(struct sock
*sk
);
1325 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1327 struct net
*net
= sock_net((struct sock
*)tp
);
1329 return tp
->keepalive_intvl
? : net
->ipv4
.sysctl_tcp_keepalive_intvl
;
1332 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1334 struct net
*net
= sock_net((struct sock
*)tp
);
1336 return tp
->keepalive_time
? : net
->ipv4
.sysctl_tcp_keepalive_time
;
1339 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1341 struct net
*net
= sock_net((struct sock
*)tp
);
1343 return tp
->keepalive_probes
? : net
->ipv4
.sysctl_tcp_keepalive_probes
;
1346 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1348 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1350 return min_t(u32
, tcp_jiffies32
- icsk
->icsk_ack
.lrcvtime
,
1351 tcp_jiffies32
- tp
->rcv_tstamp
);
1354 static inline int tcp_fin_time(const struct sock
*sk
)
1356 int fin_timeout
= tcp_sk(sk
)->linger2
? : sock_net(sk
)->ipv4
.sysctl_tcp_fin_timeout
;
1357 const int rto
= inet_csk(sk
)->icsk_rto
;
1359 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1360 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1365 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1368 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1370 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1373 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1374 * then following tcp messages have valid values. Ignore 0 value,
1375 * or else 'negative' tsval might forbid us to accept their packets.
1377 if (!rx_opt
->ts_recent
)
1382 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1385 if (tcp_paws_check(rx_opt
, 0))
1388 /* RST segments are not recommended to carry timestamp,
1389 and, if they do, it is recommended to ignore PAWS because
1390 "their cleanup function should take precedence over timestamps."
1391 Certainly, it is mistake. It is necessary to understand the reasons
1392 of this constraint to relax it: if peer reboots, clock may go
1393 out-of-sync and half-open connections will not be reset.
1394 Actually, the problem would be not existing if all
1395 the implementations followed draft about maintaining clock
1396 via reboots. Linux-2.2 DOES NOT!
1398 However, we can relax time bounds for RST segments to MSL.
1400 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1405 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1406 int mib_idx
, u32
*last_oow_ack_time
);
1408 static inline void tcp_mib_init(struct net
*net
)
1411 TCP_ADD_STATS(net
, TCP_MIB_RTOALGORITHM
, 1);
1412 TCP_ADD_STATS(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1413 TCP_ADD_STATS(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1414 TCP_ADD_STATS(net
, TCP_MIB_MAXCONN
, -1);
1418 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1420 tp
->lost_skb_hint
= NULL
;
1423 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1425 tcp_clear_retrans_hints_partial(tp
);
1426 tp
->retransmit_skb_hint
= NULL
;
1429 union tcp_md5_addr
{
1431 #if IS_ENABLED(CONFIG_IPV6)
1436 /* - key database */
1437 struct tcp_md5sig_key
{
1438 struct hlist_node node
;
1440 u8 family
; /* AF_INET or AF_INET6 */
1441 union tcp_md5_addr addr
;
1442 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1443 struct rcu_head rcu
;
1447 struct tcp_md5sig_info
{
1448 struct hlist_head head
;
1449 struct rcu_head rcu
;
1452 /* - pseudo header */
1453 struct tcp4_pseudohdr
{
1461 struct tcp6_pseudohdr
{
1462 struct in6_addr saddr
;
1463 struct in6_addr daddr
;
1465 __be32 protocol
; /* including padding */
1468 union tcp_md5sum_block
{
1469 struct tcp4_pseudohdr ip4
;
1470 #if IS_ENABLED(CONFIG_IPV6)
1471 struct tcp6_pseudohdr ip6
;
1475 /* - pool: digest algorithm, hash description and scratch buffer */
1476 struct tcp_md5sig_pool
{
1477 struct ahash_request
*md5_req
;
1482 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1483 const struct sock
*sk
, const struct sk_buff
*skb
);
1484 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1485 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
);
1486 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1488 struct tcp_md5sig_key
*tcp_v4_md5_lookup(const struct sock
*sk
,
1489 const struct sock
*addr_sk
);
1491 #ifdef CONFIG_TCP_MD5SIG
1492 struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1493 const union tcp_md5_addr
*addr
,
1495 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1497 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1498 const union tcp_md5_addr
*addr
,
1503 #define tcp_twsk_md5_key(twsk) NULL
1506 bool tcp_alloc_md5sig_pool(void);
1508 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1509 static inline void tcp_put_md5sig_pool(void)
1514 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1515 unsigned int header_len
);
1516 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1517 const struct tcp_md5sig_key
*key
);
1519 /* From tcp_fastopen.c */
1520 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1521 struct tcp_fastopen_cookie
*cookie
, int *syn_loss
,
1522 unsigned long *last_syn_loss
);
1523 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1524 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1526 struct tcp_fastopen_request
{
1527 /* Fast Open cookie. Size 0 means a cookie request */
1528 struct tcp_fastopen_cookie cookie
;
1529 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1531 int copied
; /* queued in tcp_connect() */
1533 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1535 extern struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
1536 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
);
1537 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
);
1538 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1539 struct request_sock
*req
,
1540 struct tcp_fastopen_cookie
*foc
,
1541 struct dst_entry
*dst
);
1542 void tcp_fastopen_init_key_once(bool publish
);
1543 bool tcp_fastopen_cookie_check(struct sock
*sk
, u16
*mss
,
1544 struct tcp_fastopen_cookie
*cookie
);
1545 bool tcp_fastopen_defer_connect(struct sock
*sk
, int *err
);
1546 #define TCP_FASTOPEN_KEY_LENGTH 16
1548 /* Fastopen key context */
1549 struct tcp_fastopen_context
{
1550 struct crypto_cipher
*tfm
;
1551 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1552 struct rcu_head rcu
;
1555 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout
;
1556 void tcp_fastopen_active_disable(struct sock
*sk
);
1557 bool tcp_fastopen_active_should_disable(struct sock
*sk
);
1558 void tcp_fastopen_active_disable_ofo_check(struct sock
*sk
);
1559 void tcp_fastopen_active_timeout_reset(void);
1561 /* Latencies incurred by various limits for a sender. They are
1562 * chronograph-like stats that are mutually exclusive.
1566 TCP_CHRONO_BUSY
, /* Actively sending data (non-empty write queue) */
1567 TCP_CHRONO_RWND_LIMITED
, /* Stalled by insufficient receive window */
1568 TCP_CHRONO_SNDBUF_LIMITED
, /* Stalled by insufficient send buffer */
1572 void tcp_chrono_start(struct sock
*sk
, const enum tcp_chrono type
);
1573 void tcp_chrono_stop(struct sock
*sk
, const enum tcp_chrono type
);
1575 /* write queue abstraction */
1576 static inline void tcp_write_queue_purge(struct sock
*sk
)
1578 struct sk_buff
*skb
;
1580 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1581 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1582 sk_wmem_free_skb(sk
, skb
);
1584 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1587 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1589 return skb_peek(&sk
->sk_write_queue
);
1592 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1594 return skb_peek_tail(&sk
->sk_write_queue
);
1597 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1598 const struct sk_buff
*skb
)
1600 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1603 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1604 const struct sk_buff
*skb
)
1606 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1609 #define tcp_for_write_queue(skb, sk) \
1610 skb_queue_walk(&(sk)->sk_write_queue, skb)
1612 #define tcp_for_write_queue_from(skb, sk) \
1613 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1615 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1616 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1618 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1620 return sk
->sk_send_head
;
1623 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1624 const struct sk_buff
*skb
)
1626 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1629 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1631 if (tcp_skb_is_last(sk
, skb
))
1632 sk
->sk_send_head
= NULL
;
1634 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1637 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1639 if (sk
->sk_send_head
== skb_unlinked
) {
1640 sk
->sk_send_head
= NULL
;
1641 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1643 if (tcp_sk(sk
)->highest_sack
== skb_unlinked
)
1644 tcp_sk(sk
)->highest_sack
= NULL
;
1647 static inline void tcp_init_send_head(struct sock
*sk
)
1649 sk
->sk_send_head
= NULL
;
1652 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1654 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1657 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1659 __tcp_add_write_queue_tail(sk
, skb
);
1661 /* Queue it, remembering where we must start sending. */
1662 if (sk
->sk_send_head
== NULL
) {
1663 sk
->sk_send_head
= skb
;
1664 tcp_chrono_start(sk
, TCP_CHRONO_BUSY
);
1666 if (tcp_sk(sk
)->highest_sack
== NULL
)
1667 tcp_sk(sk
)->highest_sack
= skb
;
1671 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1673 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1676 /* Insert buff after skb on the write queue of sk. */
1677 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1678 struct sk_buff
*buff
,
1681 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1684 /* Insert new before skb on the write queue of sk. */
1685 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1686 struct sk_buff
*skb
,
1689 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1691 if (sk
->sk_send_head
== skb
)
1692 sk
->sk_send_head
= new;
1695 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1697 __skb_unlink(skb
, &sk
->sk_write_queue
);
1700 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1702 return skb_queue_empty(&sk
->sk_write_queue
);
1705 static inline void tcp_push_pending_frames(struct sock
*sk
)
1707 if (tcp_send_head(sk
)) {
1708 struct tcp_sock
*tp
= tcp_sk(sk
);
1710 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1714 /* Start sequence of the skb just after the highest skb with SACKed
1715 * bit, valid only if sacked_out > 0 or when the caller has ensured
1716 * validity by itself.
1718 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1720 if (!tp
->sacked_out
)
1723 if (tp
->highest_sack
== NULL
)
1726 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1729 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1731 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1732 tcp_write_queue_next(sk
, skb
);
1735 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1737 return tcp_sk(sk
)->highest_sack
;
1740 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1742 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1745 /* Called when old skb is about to be deleted (to be combined with new skb) */
1746 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1747 struct sk_buff
*old
,
1748 struct sk_buff
*new)
1750 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1751 tcp_sk(sk
)->highest_sack
= new;
1754 /* This helper checks if socket has IP_TRANSPARENT set */
1755 static inline bool inet_sk_transparent(const struct sock
*sk
)
1757 switch (sk
->sk_state
) {
1759 return inet_twsk(sk
)->tw_transparent
;
1760 case TCP_NEW_SYN_RECV
:
1761 return inet_rsk(inet_reqsk(sk
))->no_srccheck
;
1763 return inet_sk(sk
)->transparent
;
1766 /* Determines whether this is a thin stream (which may suffer from
1767 * increased latency). Used to trigger latency-reducing mechanisms.
1769 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1771 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1775 enum tcp_seq_states
{
1776 TCP_SEQ_STATE_LISTENING
,
1777 TCP_SEQ_STATE_ESTABLISHED
,
1780 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1782 struct tcp_seq_afinfo
{
1785 const struct file_operations
*seq_fops
;
1786 struct seq_operations seq_ops
;
1789 struct tcp_iter_state
{
1790 struct seq_net_private p
;
1792 enum tcp_seq_states state
;
1793 struct sock
*syn_wait_sk
;
1794 int bucket
, offset
, sbucket
, num
;
1798 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1799 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1801 extern struct request_sock_ops tcp_request_sock_ops
;
1802 extern struct request_sock_ops tcp6_request_sock_ops
;
1804 void tcp_v4_destroy_sock(struct sock
*sk
);
1806 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1807 netdev_features_t features
);
1808 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1809 int tcp_gro_complete(struct sk_buff
*skb
);
1811 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1813 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1815 struct net
*net
= sock_net((struct sock
*)tp
);
1816 return tp
->notsent_lowat
?: net
->ipv4
.sysctl_tcp_notsent_lowat
;
1819 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1821 const struct tcp_sock
*tp
= tcp_sk(sk
);
1822 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1824 return notsent_bytes
< tcp_notsent_lowat(tp
);
1827 #ifdef CONFIG_PROC_FS
1828 int tcp4_proc_init(void);
1829 void tcp4_proc_exit(void);
1832 int tcp_rtx_synack(const struct sock
*sk
, struct request_sock
*req
);
1833 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1834 const struct tcp_request_sock_ops
*af_ops
,
1835 struct sock
*sk
, struct sk_buff
*skb
);
1837 /* TCP af-specific functions */
1838 struct tcp_sock_af_ops
{
1839 #ifdef CONFIG_TCP_MD5SIG
1840 struct tcp_md5sig_key
*(*md5_lookup
) (const struct sock
*sk
,
1841 const struct sock
*addr_sk
);
1842 int (*calc_md5_hash
)(char *location
,
1843 const struct tcp_md5sig_key
*md5
,
1844 const struct sock
*sk
,
1845 const struct sk_buff
*skb
);
1846 int (*md5_parse
)(struct sock
*sk
,
1847 char __user
*optval
,
1852 struct tcp_request_sock_ops
{
1854 #ifdef CONFIG_TCP_MD5SIG
1855 struct tcp_md5sig_key
*(*req_md5_lookup
)(const struct sock
*sk
,
1856 const struct sock
*addr_sk
);
1857 int (*calc_md5_hash
) (char *location
,
1858 const struct tcp_md5sig_key
*md5
,
1859 const struct sock
*sk
,
1860 const struct sk_buff
*skb
);
1862 void (*init_req
)(struct request_sock
*req
,
1863 const struct sock
*sk_listener
,
1864 struct sk_buff
*skb
);
1865 #ifdef CONFIG_SYN_COOKIES
1866 __u32 (*cookie_init_seq
)(const struct sk_buff
*skb
,
1869 struct dst_entry
*(*route_req
)(const struct sock
*sk
, struct flowi
*fl
,
1870 const struct request_sock
*req
);
1871 u32 (*init_seq
)(const struct sk_buff
*skb
);
1872 u32 (*init_ts_off
)(const struct sk_buff
*skb
);
1873 int (*send_synack
)(const struct sock
*sk
, struct dst_entry
*dst
,
1874 struct flowi
*fl
, struct request_sock
*req
,
1875 struct tcp_fastopen_cookie
*foc
,
1876 enum tcp_synack_type synack_type
);
1879 #ifdef CONFIG_SYN_COOKIES
1880 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1881 const struct sock
*sk
, struct sk_buff
*skb
,
1884 tcp_synq_overflow(sk
);
1885 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_SYNCOOKIESSENT
);
1886 return ops
->cookie_init_seq(skb
, mss
);
1889 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1890 const struct sock
*sk
, struct sk_buff
*skb
,
1897 int tcpv4_offload_init(void);
1899 void tcp_v4_init(void);
1900 void tcp_init(void);
1902 /* tcp_recovery.c */
1903 extern void tcp_rack_mark_lost(struct sock
*sk
);
1904 extern void tcp_rack_advance(struct tcp_sock
*tp
, u8 sacked
, u32 end_seq
,
1906 extern void tcp_rack_reo_timeout(struct sock
*sk
);
1909 * Save and compile IPv4 options, return a pointer to it
1911 static inline struct ip_options_rcu
*tcp_v4_save_options(struct sk_buff
*skb
)
1913 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
1914 struct ip_options_rcu
*dopt
= NULL
;
1917 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
1919 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
1920 if (dopt
&& __ip_options_echo(&dopt
->opt
, skb
, opt
)) {
1928 /* locally generated TCP pure ACKs have skb->truesize == 2
1929 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1930 * This is much faster than dissecting the packet to find out.
1931 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1933 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
1935 return skb
->truesize
== 2;
1938 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)
1943 static inline int tcp_inq(struct sock
*sk
)
1945 struct tcp_sock
*tp
= tcp_sk(sk
);
1948 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
1950 } else if (sock_flag(sk
, SOCK_URGINLINE
) ||
1952 before(tp
->urg_seq
, tp
->copied_seq
) ||
1953 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
1955 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
1957 /* Subtract 1, if FIN was received */
1958 if (answ
&& sock_flag(sk
, SOCK_DONE
))
1961 answ
= tp
->urg_seq
- tp
->copied_seq
;
1967 int tcp_peek_len(struct socket
*sock
);
1969 static inline void tcp_segs_in(struct tcp_sock
*tp
, const struct sk_buff
*skb
)
1973 segs_in
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
1974 tp
->segs_in
+= segs_in
;
1975 if (skb
->len
> tcp_hdrlen(skb
))
1976 tp
->data_segs_in
+= segs_in
;
1980 * TCP listen path runs lockless.
1981 * We forced "struct sock" to be const qualified to make sure
1982 * we don't modify one of its field by mistake.
1983 * Here, we increment sk_drops which is an atomic_t, so we can safely
1984 * make sock writable again.
1986 static inline void tcp_listendrop(const struct sock
*sk
)
1988 atomic_inc(&((struct sock
*)sk
)->sk_drops
);
1989 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1992 enum hrtimer_restart
tcp_pace_kick(struct hrtimer
*timer
);