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
;
278 extern int sysctl_tcp_default_init_rwnd
;
280 extern atomic_long_t tcp_memory_allocated
;
281 extern struct percpu_counter tcp_sockets_allocated
;
282 extern unsigned long tcp_memory_pressure
;
284 /* optimized version of sk_under_memory_pressure() for TCP sockets */
285 static inline bool tcp_under_memory_pressure(const struct sock
*sk
)
287 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
288 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
291 return tcp_memory_pressure
;
294 * The next routines deal with comparing 32 bit unsigned ints
295 * and worry about wraparound (automatic with unsigned arithmetic).
298 static inline bool before(__u32 seq1
, __u32 seq2
)
300 return (__s32
)(seq1
-seq2
) < 0;
302 #define after(seq2, seq1) before(seq1, seq2)
304 /* is s2<=s1<=s3 ? */
305 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
307 return seq3
- seq2
>= seq1
- seq2
;
310 static inline bool tcp_out_of_memory(struct sock
*sk
)
312 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
313 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
318 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
320 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
322 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
323 int orphans
= percpu_counter_read_positive(ocp
);
325 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
326 orphans
= percpu_counter_sum_positive(ocp
);
327 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
333 bool tcp_check_oom(struct sock
*sk
, int shift
);
336 extern struct proto tcp_prot
;
338 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
339 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
340 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
341 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
343 void tcp_tasklet_init(void);
345 void tcp_v4_err(struct sk_buff
*skb
, u32
);
347 void tcp_shutdown(struct sock
*sk
, int how
);
349 int tcp_v4_early_demux(struct sk_buff
*skb
);
350 int tcp_v4_rcv(struct sk_buff
*skb
);
352 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
353 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
354 int tcp_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
355 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
357 int tcp_sendpage_locked(struct sock
*sk
, struct page
*page
, int offset
,
358 size_t size
, int flags
);
359 ssize_t
do_tcp_sendpages(struct sock
*sk
, struct page
*page
, int offset
,
360 size_t size
, int flags
);
361 void tcp_release_cb(struct sock
*sk
);
362 void tcp_wfree(struct sk_buff
*skb
);
363 void tcp_write_timer_handler(struct sock
*sk
);
364 void tcp_delack_timer_handler(struct sock
*sk
);
365 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
366 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
);
367 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
368 const struct tcphdr
*th
);
369 void tcp_rcv_space_adjust(struct sock
*sk
);
370 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
371 void tcp_twsk_destructor(struct sock
*sk
);
372 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
373 struct pipe_inode_info
*pipe
, size_t len
,
376 void tcp_enter_quickack_mode(struct sock
*sk
, unsigned int max_quickacks
);
377 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
378 const unsigned int pkts
)
380 struct inet_connection_sock
*icsk
= inet_csk(sk
);
382 if (icsk
->icsk_ack
.quick
) {
383 if (pkts
>= icsk
->icsk_ack
.quick
) {
384 icsk
->icsk_ack
.quick
= 0;
385 /* Leaving quickack mode we deflate ATO. */
386 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
388 icsk
->icsk_ack
.quick
-= pkts
;
393 #define TCP_ECN_QUEUE_CWR 2
394 #define TCP_ECN_DEMAND_CWR 4
395 #define TCP_ECN_SEEN 8
405 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
407 const struct tcphdr
*th
);
408 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
409 struct request_sock
*req
, bool fastopen
);
410 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
411 struct sk_buff
*skb
);
412 void tcp_enter_loss(struct sock
*sk
);
413 void tcp_cwnd_reduction(struct sock
*sk
, int newly_acked_sacked
, int flag
);
414 void tcp_clear_retrans(struct tcp_sock
*tp
);
415 void tcp_update_metrics(struct sock
*sk
);
416 void tcp_init_metrics(struct sock
*sk
);
417 void tcp_metrics_init(void);
418 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
);
419 void tcp_disable_fack(struct tcp_sock
*tp
);
420 void tcp_close(struct sock
*sk
, long timeout
);
421 void tcp_init_sock(struct sock
*sk
);
422 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
,
423 struct poll_table_struct
*wait
);
424 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
425 char __user
*optval
, int __user
*optlen
);
426 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
427 char __user
*optval
, unsigned int optlen
);
428 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
429 char __user
*optval
, int __user
*optlen
);
430 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
431 char __user
*optval
, unsigned int optlen
);
432 void tcp_set_keepalive(struct sock
*sk
, int val
);
433 void tcp_syn_ack_timeout(const struct request_sock
*req
);
434 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
435 int flags
, int *addr_len
);
436 void tcp_parse_options(const struct net
*net
, const struct sk_buff
*skb
,
437 struct tcp_options_received
*opt_rx
,
438 int estab
, struct tcp_fastopen_cookie
*foc
);
439 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
442 * TCP v4 functions exported for the inet6 API
445 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
446 void tcp_v4_mtu_reduced(struct sock
*sk
);
447 void tcp_req_err(struct sock
*sk
, u32 seq
, bool abort
);
448 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
449 struct sock
*tcp_create_openreq_child(const struct sock
*sk
,
450 struct request_sock
*req
,
451 struct sk_buff
*skb
);
452 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
453 struct sock
*tcp_v4_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
454 struct request_sock
*req
,
455 struct dst_entry
*dst
,
456 struct request_sock
*req_unhash
,
458 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
459 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
460 int tcp_connect(struct sock
*sk
);
461 enum tcp_synack_type
{
466 struct sk_buff
*tcp_make_synack(const struct sock
*sk
, struct dst_entry
*dst
,
467 struct request_sock
*req
,
468 struct tcp_fastopen_cookie
*foc
,
469 enum tcp_synack_type synack_type
);
470 int tcp_disconnect(struct sock
*sk
, int flags
);
472 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
473 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
474 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
476 /* From syncookies.c */
477 struct sock
*tcp_get_cookie_sock(struct sock
*sk
, struct sk_buff
*skb
,
478 struct request_sock
*req
,
479 struct dst_entry
*dst
, u32 tsoff
);
480 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
482 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
483 #ifdef CONFIG_SYN_COOKIES
485 /* Syncookies use a monotonic timer which increments every 60 seconds.
486 * This counter is used both as a hash input and partially encoded into
487 * the cookie value. A cookie is only validated further if the delta
488 * between the current counter value and the encoded one is less than this,
489 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
490 * the counter advances immediately after a cookie is generated).
492 #define MAX_SYNCOOKIE_AGE 2
493 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
494 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
496 /* syncookies: remember time of last synqueue overflow
497 * But do not dirty this field too often (once per second is enough)
498 * It is racy as we do not hold a lock, but race is very minor.
500 static inline void tcp_synq_overflow(const struct sock
*sk
)
502 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
503 unsigned long now
= jiffies
;
505 if (time_after(now
, last_overflow
+ HZ
))
506 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= now
;
509 /* syncookies: no recent synqueue overflow on this listening socket? */
510 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
512 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
514 return time_after(jiffies
, last_overflow
+ TCP_SYNCOOKIE_VALID
);
517 static inline u32
tcp_cookie_time(void)
519 u64 val
= get_jiffies_64();
521 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
525 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
527 __u32
cookie_v4_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
528 u64
cookie_init_timestamp(struct request_sock
*req
);
529 bool cookie_timestamp_decode(const struct net
*net
,
530 struct tcp_options_received
*opt
);
531 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
532 const struct net
*net
, const struct dst_entry
*dst
);
534 /* From net/ipv6/syncookies.c */
535 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
537 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
539 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
540 const struct tcphdr
*th
, u16
*mssp
);
541 __u32
cookie_v6_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
545 u32
tcp_tso_autosize(const struct sock
*sk
, unsigned int mss_now
,
547 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
549 int __tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
550 int tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
551 void tcp_retransmit_timer(struct sock
*sk
);
552 void tcp_xmit_retransmit_queue(struct sock
*);
553 void tcp_simple_retransmit(struct sock
*);
554 void tcp_enter_recovery(struct sock
*sk
, bool ece_ack
);
555 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
556 int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int, gfp_t
);
558 void tcp_send_probe0(struct sock
*);
559 void tcp_send_partial(struct sock
*);
560 int tcp_write_wakeup(struct sock
*, int mib
);
561 void tcp_send_fin(struct sock
*sk
);
562 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
563 int tcp_send_synack(struct sock
*);
564 void tcp_push_one(struct sock
*, unsigned int mss_now
);
565 void __tcp_send_ack(struct sock
*sk
, u32 rcv_nxt
);
566 void tcp_send_ack(struct sock
*sk
);
567 void tcp_send_delayed_ack(struct sock
*sk
);
568 void tcp_send_loss_probe(struct sock
*sk
);
569 bool tcp_schedule_loss_probe(struct sock
*sk
, bool advancing_rto
);
570 void tcp_skb_collapse_tstamp(struct sk_buff
*skb
,
571 const struct sk_buff
*next_skb
);
574 void tcp_rearm_rto(struct sock
*sk
);
575 void tcp_synack_rtt_meas(struct sock
*sk
, struct request_sock
*req
);
576 void tcp_reset(struct sock
*sk
);
577 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock
*tp
, struct sk_buff
*skb
);
578 void tcp_fin(struct sock
*sk
);
581 void tcp_init_xmit_timers(struct sock
*);
582 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
584 hrtimer_cancel(&tcp_sk(sk
)->pacing_timer
);
585 inet_csk_clear_xmit_timers(sk
);
588 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
589 unsigned int tcp_current_mss(struct sock
*sk
);
591 /* Bound MSS / TSO packet size with the half of the window */
592 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
596 /* When peer uses tiny windows, there is no use in packetizing
597 * to sub-MSS pieces for the sake of SWS or making sure there
598 * are enough packets in the pipe for fast recovery.
600 * On the other hand, for extremely large MSS devices, handling
601 * smaller than MSS windows in this way does make sense.
603 if (tp
->max_window
> TCP_MSS_DEFAULT
)
604 cutoff
= (tp
->max_window
>> 1);
606 cutoff
= tp
->max_window
;
608 if (cutoff
&& pktsize
> cutoff
)
609 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
615 void tcp_get_info(struct sock
*, struct tcp_info
*);
617 /* Read 'sendfile()'-style from a TCP socket */
618 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
619 sk_read_actor_t recv_actor
);
621 void tcp_initialize_rcv_mss(struct sock
*sk
);
623 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
624 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
625 void tcp_mtup_init(struct sock
*sk
);
626 void tcp_init_buffer_space(struct sock
*sk
);
628 static inline void tcp_bound_rto(const struct sock
*sk
)
630 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
631 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
634 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
636 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
639 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
641 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
642 ntohl(TCP_FLAG_ACK
) |
646 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
648 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
651 static inline void tcp_fast_path_check(struct sock
*sk
)
653 struct tcp_sock
*tp
= tcp_sk(sk
);
655 if (RB_EMPTY_ROOT(&tp
->out_of_order_queue
) &&
657 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
659 tcp_fast_path_on(tp
);
662 /* Compute the actual rto_min value */
663 static inline u32
tcp_rto_min(struct sock
*sk
)
665 const struct dst_entry
*dst
= __sk_dst_get(sk
);
666 u32 rto_min
= TCP_RTO_MIN
;
668 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
669 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
673 static inline u32
tcp_rto_min_us(struct sock
*sk
)
675 return jiffies_to_usecs(tcp_rto_min(sk
));
678 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
680 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
683 /* Minimum RTT in usec. ~0 means not available. */
684 static inline u32
tcp_min_rtt(const struct tcp_sock
*tp
)
686 return minmax_get(&tp
->rtt_min
);
689 /* Compute the actual receive window we are currently advertising.
690 * Rcv_nxt can be after the window if our peer push more data
691 * than the offered window.
693 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
695 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
702 /* Choose a new window, without checks for shrinking, and without
703 * scaling applied to the result. The caller does these things
704 * if necessary. This is a "raw" window selection.
706 u32
__tcp_select_window(struct sock
*sk
);
708 void tcp_send_window_probe(struct sock
*sk
);
710 /* TCP uses 32bit jiffies to save some space.
711 * Note that this is different from tcp_time_stamp, which
712 * historically has been the same until linux-4.13.
714 #define tcp_jiffies32 ((u32)jiffies)
717 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
718 * It is no longer tied to jiffies, but to 1 ms clock.
719 * Note: double check if you want to use tcp_jiffies32 instead of this.
721 #define TCP_TS_HZ 1000
723 static inline u64
tcp_clock_ns(void)
725 return local_clock();
728 static inline u64
tcp_clock_us(void)
730 return div_u64(tcp_clock_ns(), NSEC_PER_USEC
);
733 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
734 static inline u32
tcp_time_stamp(const struct tcp_sock
*tp
)
736 return div_u64(tp
->tcp_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
739 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
740 static inline u32
tcp_time_stamp_raw(void)
742 return div_u64(tcp_clock_ns(), NSEC_PER_SEC
/ TCP_TS_HZ
);
746 /* Refresh 1us clock of a TCP socket,
747 * ensuring monotically increasing values.
749 static inline void tcp_mstamp_refresh(struct tcp_sock
*tp
)
751 u64 val
= tcp_clock_us();
753 if (val
> tp
->tcp_mstamp
)
754 tp
->tcp_mstamp
= val
;
757 static inline u32
tcp_stamp_us_delta(u64 t1
, u64 t0
)
759 return max_t(s64
, t1
- t0
, 0);
762 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
764 return div_u64(skb
->skb_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
768 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
770 #define TCPHDR_FIN 0x01
771 #define TCPHDR_SYN 0x02
772 #define TCPHDR_RST 0x04
773 #define TCPHDR_PSH 0x08
774 #define TCPHDR_ACK 0x10
775 #define TCPHDR_URG 0x20
776 #define TCPHDR_ECE 0x40
777 #define TCPHDR_CWR 0x80
779 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
781 /* This is what the send packet queuing engine uses to pass
782 * TCP per-packet control information to the transmission code.
783 * We also store the host-order sequence numbers in here too.
784 * This is 44 bytes if IPV6 is enabled.
785 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
788 __u32 seq
; /* Starting sequence number */
789 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
791 /* Note : tcp_tw_isn is used in input path only
792 * (isn chosen by tcp_timewait_state_process())
794 * tcp_gso_segs/size are used in write queue only,
795 * cf tcp_skb_pcount()/tcp_skb_mss()
803 /* Used to stash the receive timestamp while this skb is in the
804 * out of order queue, as skb->tstamp is overwritten by the
809 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
811 __u8 sacked
; /* State flags for SACK/FACK. */
812 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
813 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
814 #define TCPCB_LOST 0x04 /* SKB is lost */
815 #define TCPCB_TAGBITS 0x07 /* All tag bits */
816 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
817 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
818 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
821 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
822 __u8 txstamp_ack
:1, /* Record TX timestamp for ack? */
823 eor
:1, /* Is skb MSG_EOR marked? */
824 has_rxtstamp
:1, /* SKB has a RX timestamp */
826 __u32 ack_seq
; /* Sequence number ACK'd */
829 /* There is space for up to 24 bytes */
830 __u32 in_flight
:30,/* Bytes in flight at transmit */
831 is_app_limited
:1, /* cwnd not fully used? */
833 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
835 /* start of send pipeline phase */
837 /* when we reached the "delivered" count */
838 u64 delivered_mstamp
;
839 } tx
; /* only used for outgoing skbs */
841 struct inet_skb_parm h4
;
842 #if IS_ENABLED(CONFIG_IPV6)
843 struct inet6_skb_parm h6
;
845 } header
; /* For incoming skbs */
855 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
858 #if IS_ENABLED(CONFIG_IPV6)
859 /* This is the variant of inet6_iif() that must be used by TCP,
860 * as TCP moves IP6CB into a different location in skb->cb[]
862 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
864 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
867 static inline int tcp_v6_iif_l3_slave(const struct sk_buff
*skb
)
869 bool l3_slave
= ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
);
871 return l3_slave
? skb
->skb_iif
: TCP_SKB_CB(skb
)->header
.h6
.iif
;
874 /* TCP_SKB_CB reference means this can not be used from early demux */
875 static inline int tcp_v6_sdif(const struct sk_buff
*skb
)
877 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
878 if (skb
&& ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
))
879 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
885 static inline bool inet_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
887 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
888 if (!net
->ipv4
.sysctl_tcp_l3mdev_accept
&&
889 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
895 /* TCP_SKB_CB reference means this can not be used from early demux */
896 static inline int tcp_v4_sdif(struct sk_buff
*skb
)
898 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
899 if (skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
900 return TCP_SKB_CB(skb
)->header
.h4
.iif
;
905 /* Due to TSO, an SKB can be composed of multiple actual
906 * packets. To keep these tracked properly, we use this.
908 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
910 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
913 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
915 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
918 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
920 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
923 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
924 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
926 return TCP_SKB_CB(skb
)->tcp_gso_size
;
929 static inline bool tcp_skb_can_collapse_to(const struct sk_buff
*skb
)
931 return likely(!TCP_SKB_CB(skb
)->eor
);
934 /* Events passed to congestion control interface */
936 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
937 CA_EVENT_CWND_RESTART
, /* congestion window restart */
938 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
939 CA_EVENT_LOSS
, /* loss timeout */
940 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
941 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
944 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
945 enum tcp_ca_ack_event_flags
{
946 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
947 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
948 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
952 * Interface for adding new TCP congestion control handlers
954 #define TCP_CA_NAME_MAX 16
955 #define TCP_CA_MAX 128
956 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
958 #define TCP_CA_UNSPEC 0
960 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
961 #define TCP_CONG_NON_RESTRICTED 0x1
962 /* Requires ECN/ECT set on all packets */
963 #define TCP_CONG_NEEDS_ECN 0x2
973 /* A rate sample measures the number of (original/retransmitted) data
974 * packets delivered "delivered" over an interval of time "interval_us".
975 * The tcp_rate.c code fills in the rate sample, and congestion
976 * control modules that define a cong_control function to run at the end
977 * of ACK processing can optionally chose to consult this sample when
978 * setting cwnd and pacing rate.
979 * A sample is invalid if "delivered" or "interval_us" is negative.
982 u64 prior_mstamp
; /* starting timestamp for interval */
983 u32 prior_delivered
; /* tp->delivered at "prior_mstamp" */
984 s32 delivered
; /* number of packets delivered over interval */
985 long interval_us
; /* time for tp->delivered to incr "delivered" */
986 long rtt_us
; /* RTT of last (S)ACKed packet (or -1) */
987 int losses
; /* number of packets marked lost upon ACK */
988 u32 acked_sacked
; /* number of packets newly (S)ACKed upon ACK */
989 u32 prior_in_flight
; /* in flight before this ACK */
990 bool is_app_limited
; /* is sample from packet with bubble in pipe? */
991 bool is_retrans
; /* is sample from retransmission? */
994 struct tcp_congestion_ops
{
995 struct list_head list
;
999 /* initialize private data (optional) */
1000 void (*init
)(struct sock
*sk
);
1001 /* cleanup private data (optional) */
1002 void (*release
)(struct sock
*sk
);
1004 /* return slow start threshold (required) */
1005 u32 (*ssthresh
)(struct sock
*sk
);
1006 /* do new cwnd calculation (required) */
1007 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
1008 /* call before changing ca_state (optional) */
1009 void (*set_state
)(struct sock
*sk
, u8 new_state
);
1010 /* call when cwnd event occurs (optional) */
1011 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
1012 /* call when ack arrives (optional) */
1013 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
1014 /* new value of cwnd after loss (required) */
1015 u32 (*undo_cwnd
)(struct sock
*sk
);
1016 /* hook for packet ack accounting (optional) */
1017 void (*pkts_acked
)(struct sock
*sk
, const struct ack_sample
*sample
);
1018 /* suggest number of segments for each skb to transmit (optional) */
1019 u32 (*tso_segs_goal
)(struct sock
*sk
);
1020 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1021 u32 (*sndbuf_expand
)(struct sock
*sk
);
1022 /* call when packets are delivered to update cwnd and pacing rate,
1023 * after all the ca_state processing. (optional)
1025 void (*cong_control
)(struct sock
*sk
, const struct rate_sample
*rs
);
1026 /* get info for inet_diag (optional) */
1027 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
1028 union tcp_cc_info
*info
);
1030 char name
[TCP_CA_NAME_MAX
];
1031 struct module
*owner
;
1034 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1035 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1037 void tcp_assign_congestion_control(struct sock
*sk
);
1038 void tcp_init_congestion_control(struct sock
*sk
);
1039 void tcp_cleanup_congestion_control(struct sock
*sk
);
1040 int tcp_set_default_congestion_control(const char *name
);
1041 void tcp_get_default_congestion_control(char *name
);
1042 void tcp_get_available_congestion_control(char *buf
, size_t len
);
1043 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
1044 int tcp_set_allowed_congestion_control(char *allowed
);
1045 int tcp_set_congestion_control(struct sock
*sk
, const char *name
, bool load
, bool reinit
);
1046 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
1047 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
1049 u32
tcp_reno_ssthresh(struct sock
*sk
);
1050 u32
tcp_reno_undo_cwnd(struct sock
*sk
);
1051 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
1052 extern struct tcp_congestion_ops tcp_reno
;
1054 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
1055 u32
tcp_ca_get_key_by_name(const char *name
, bool *ecn_ca
);
1057 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
1059 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
1065 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
1067 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1069 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
1072 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
1074 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1076 if (icsk
->icsk_ca_ops
->set_state
)
1077 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
1078 icsk
->icsk_ca_state
= ca_state
;
1081 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
1083 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1085 if (icsk
->icsk_ca_ops
->cwnd_event
)
1086 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
1089 /* From tcp_rate.c */
1090 void tcp_rate_skb_sent(struct sock
*sk
, struct sk_buff
*skb
);
1091 void tcp_rate_skb_delivered(struct sock
*sk
, struct sk_buff
*skb
,
1092 struct rate_sample
*rs
);
1093 void tcp_rate_gen(struct sock
*sk
, u32 delivered
, u32 lost
,
1094 bool is_sack_reneg
, struct rate_sample
*rs
);
1095 void tcp_rate_check_app_limited(struct sock
*sk
);
1097 /* These functions determine how the current flow behaves in respect of SACK
1098 * handling. SACK is negotiated with the peer, and therefore it can vary
1099 * between different flows.
1101 * tcp_is_sack - SACK enabled
1102 * tcp_is_reno - No SACK
1103 * tcp_is_fack - FACK enabled, implies SACK enabled
1105 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
1107 return tp
->rx_opt
.sack_ok
;
1110 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
1112 return !tcp_is_sack(tp
);
1115 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
1117 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
1120 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
1122 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
1125 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
1127 return tp
->sacked_out
+ tp
->lost_out
;
1130 /* This determines how many packets are "in the network" to the best
1131 * of our knowledge. In many cases it is conservative, but where
1132 * detailed information is available from the receiver (via SACK
1133 * blocks etc.) we can make more aggressive calculations.
1135 * Use this for decisions involving congestion control, use just
1136 * tp->packets_out to determine if the send queue is empty or not.
1138 * Read this equation as:
1140 * "Packets sent once on transmission queue" MINUS
1141 * "Packets left network, but not honestly ACKed yet" PLUS
1142 * "Packets fast retransmitted"
1144 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1146 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
1149 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1151 static inline bool tcp_in_slow_start(const struct tcp_sock
*tp
)
1153 return tp
->snd_cwnd
< tp
->snd_ssthresh
;
1156 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
1158 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
1161 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
1163 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
1164 (1 << inet_csk(sk
)->icsk_ca_state
);
1167 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1168 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1171 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
1173 const struct tcp_sock
*tp
= tcp_sk(sk
);
1175 if (tcp_in_cwnd_reduction(sk
))
1176 return tp
->snd_ssthresh
;
1178 return max(tp
->snd_ssthresh
,
1179 ((tp
->snd_cwnd
>> 1) +
1180 (tp
->snd_cwnd
>> 2)));
1183 /* Use define here intentionally to get WARN_ON location shown at the caller */
1184 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1186 void tcp_enter_cwr(struct sock
*sk
);
1187 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1189 /* The maximum number of MSS of available cwnd for which TSO defers
1190 * sending if not using sysctl_tcp_tso_win_divisor.
1192 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1197 /* Returns end sequence number of the receiver's advertised window */
1198 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1200 return tp
->snd_una
+ tp
->snd_wnd
;
1203 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1204 * flexible approach. The RFC suggests cwnd should not be raised unless
1205 * it was fully used previously. And that's exactly what we do in
1206 * congestion avoidance mode. But in slow start we allow cwnd to grow
1207 * as long as the application has used half the cwnd.
1209 * cwnd is 10 (IW10), but application sends 9 frames.
1210 * We allow cwnd to reach 18 when all frames are ACKed.
1211 * This check is safe because it's as aggressive as slow start which already
1212 * risks 100% overshoot. The advantage is that we discourage application to
1213 * either send more filler packets or data to artificially blow up the cwnd
1214 * usage, and allow application-limited process to probe bw more aggressively.
1216 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1218 const struct tcp_sock
*tp
= tcp_sk(sk
);
1220 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1221 if (tcp_in_slow_start(tp
))
1222 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1224 return tp
->is_cwnd_limited
;
1227 /* Something is really bad, we could not queue an additional packet,
1228 * because qdisc is full or receiver sent a 0 window.
1229 * We do not want to add fuel to the fire, or abort too early,
1230 * so make sure the timer we arm now is at least 200ms in the future,
1231 * regardless of current icsk_rto value (as it could be ~2ms)
1233 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1235 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1238 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1239 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1240 unsigned long max_when
)
1242 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1244 return (unsigned long)min_t(u64
, when
, max_when
);
1247 static inline void tcp_check_probe_timer(struct sock
*sk
)
1249 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1250 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1251 tcp_probe0_base(sk
), TCP_RTO_MAX
);
1254 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1259 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1265 * Calculate(/check) TCP checksum
1267 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1268 __be32 daddr
, __wsum base
)
1270 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1273 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1275 return __skb_checksum_complete(skb
);
1278 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1280 return !skb_csum_unnecessary(skb
) &&
1281 __tcp_checksum_complete(skb
);
1284 bool tcp_add_backlog(struct sock
*sk
, struct sk_buff
*skb
);
1285 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
);
1290 static const char *statename
[]={
1291 "Unused","Established","Syn Sent","Syn Recv",
1292 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1293 "Close Wait","Last ACK","Listen","Closing"
1296 void tcp_set_state(struct sock
*sk
, int state
);
1298 void tcp_done(struct sock
*sk
);
1300 int tcp_abort(struct sock
*sk
, int err
);
1302 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1305 rx_opt
->num_sacks
= 0;
1308 u32
tcp_default_init_rwnd(u32 mss
);
1309 void tcp_cwnd_restart(struct sock
*sk
, s32 delta
);
1311 static inline void tcp_slow_start_after_idle_check(struct sock
*sk
)
1313 const struct tcp_congestion_ops
*ca_ops
= inet_csk(sk
)->icsk_ca_ops
;
1314 struct tcp_sock
*tp
= tcp_sk(sk
);
1317 if (!sysctl_tcp_slow_start_after_idle
|| tp
->packets_out
||
1318 ca_ops
->cong_control
)
1320 delta
= tcp_jiffies32
- tp
->lsndtime
;
1321 if (delta
> inet_csk(sk
)->icsk_rto
)
1322 tcp_cwnd_restart(sk
, delta
);
1325 /* Determine a window scaling and initial window to offer. */
1326 void tcp_select_initial_window(int __space
, __u32 mss
, __u32
*rcv_wnd
,
1327 __u32
*window_clamp
, int wscale_ok
,
1328 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1330 static inline int tcp_win_from_space(int space
)
1332 int tcp_adv_win_scale
= sysctl_tcp_adv_win_scale
;
1334 return tcp_adv_win_scale
<= 0 ?
1335 (space
>>(-tcp_adv_win_scale
)) :
1336 space
- (space
>>tcp_adv_win_scale
);
1339 /* Note: caller must be prepared to deal with negative returns */
1340 static inline int tcp_space(const struct sock
*sk
)
1342 return tcp_win_from_space(sk
->sk_rcvbuf
-
1343 atomic_read(&sk
->sk_rmem_alloc
));
1346 static inline int tcp_full_space(const struct sock
*sk
)
1348 return tcp_win_from_space(sk
->sk_rcvbuf
);
1351 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1352 const struct sock
*sk_listener
,
1353 const struct dst_entry
*dst
);
1355 void tcp_enter_memory_pressure(struct sock
*sk
);
1356 void tcp_leave_memory_pressure(struct sock
*sk
);
1358 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1360 struct net
*net
= sock_net((struct sock
*)tp
);
1362 return tp
->keepalive_intvl
? : net
->ipv4
.sysctl_tcp_keepalive_intvl
;
1365 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1367 struct net
*net
= sock_net((struct sock
*)tp
);
1369 return tp
->keepalive_time
? : net
->ipv4
.sysctl_tcp_keepalive_time
;
1372 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1374 struct net
*net
= sock_net((struct sock
*)tp
);
1376 return tp
->keepalive_probes
? : net
->ipv4
.sysctl_tcp_keepalive_probes
;
1379 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1381 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1383 return min_t(u32
, tcp_jiffies32
- icsk
->icsk_ack
.lrcvtime
,
1384 tcp_jiffies32
- tp
->rcv_tstamp
);
1387 static inline int tcp_fin_time(const struct sock
*sk
)
1389 int fin_timeout
= tcp_sk(sk
)->linger2
? : sock_net(sk
)->ipv4
.sysctl_tcp_fin_timeout
;
1390 const int rto
= inet_csk(sk
)->icsk_rto
;
1392 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1393 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1398 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1401 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1403 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1406 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1407 * then following tcp messages have valid values. Ignore 0 value,
1408 * or else 'negative' tsval might forbid us to accept their packets.
1410 if (!rx_opt
->ts_recent
)
1415 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1418 if (tcp_paws_check(rx_opt
, 0))
1421 /* RST segments are not recommended to carry timestamp,
1422 and, if they do, it is recommended to ignore PAWS because
1423 "their cleanup function should take precedence over timestamps."
1424 Certainly, it is mistake. It is necessary to understand the reasons
1425 of this constraint to relax it: if peer reboots, clock may go
1426 out-of-sync and half-open connections will not be reset.
1427 Actually, the problem would be not existing if all
1428 the implementations followed draft about maintaining clock
1429 via reboots. Linux-2.2 DOES NOT!
1431 However, we can relax time bounds for RST segments to MSL.
1433 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1438 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1439 int mib_idx
, u32
*last_oow_ack_time
);
1441 static inline void tcp_mib_init(struct net
*net
)
1444 TCP_ADD_STATS(net
, TCP_MIB_RTOALGORITHM
, 1);
1445 TCP_ADD_STATS(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1446 TCP_ADD_STATS(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1447 TCP_ADD_STATS(net
, TCP_MIB_MAXCONN
, -1);
1451 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1453 tp
->lost_skb_hint
= NULL
;
1456 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1458 tcp_clear_retrans_hints_partial(tp
);
1459 tp
->retransmit_skb_hint
= NULL
;
1462 union tcp_md5_addr
{
1464 #if IS_ENABLED(CONFIG_IPV6)
1469 /* - key database */
1470 struct tcp_md5sig_key
{
1471 struct hlist_node node
;
1473 u8 family
; /* AF_INET or AF_INET6 */
1474 union tcp_md5_addr addr
;
1476 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1477 struct rcu_head rcu
;
1481 struct tcp_md5sig_info
{
1482 struct hlist_head head
;
1483 struct rcu_head rcu
;
1486 /* - pseudo header */
1487 struct tcp4_pseudohdr
{
1495 struct tcp6_pseudohdr
{
1496 struct in6_addr saddr
;
1497 struct in6_addr daddr
;
1499 __be32 protocol
; /* including padding */
1502 union tcp_md5sum_block
{
1503 struct tcp4_pseudohdr ip4
;
1504 #if IS_ENABLED(CONFIG_IPV6)
1505 struct tcp6_pseudohdr ip6
;
1509 /* - pool: digest algorithm, hash description and scratch buffer */
1510 struct tcp_md5sig_pool
{
1511 struct ahash_request
*md5_req
;
1516 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1517 const struct sock
*sk
, const struct sk_buff
*skb
);
1518 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1519 int family
, u8 prefixlen
, const u8
*newkey
, u8 newkeylen
,
1521 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1522 int family
, u8 prefixlen
);
1523 struct tcp_md5sig_key
*tcp_v4_md5_lookup(const struct sock
*sk
,
1524 const struct sock
*addr_sk
);
1526 #ifdef CONFIG_TCP_MD5SIG
1527 struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1528 const union tcp_md5_addr
*addr
,
1530 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1532 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1533 const union tcp_md5_addr
*addr
,
1538 #define tcp_twsk_md5_key(twsk) NULL
1541 bool tcp_alloc_md5sig_pool(void);
1543 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1544 static inline void tcp_put_md5sig_pool(void)
1549 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1550 unsigned int header_len
);
1551 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1552 const struct tcp_md5sig_key
*key
);
1554 /* From tcp_fastopen.c */
1555 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1556 struct tcp_fastopen_cookie
*cookie
, int *syn_loss
,
1557 unsigned long *last_syn_loss
);
1558 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1559 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1561 struct tcp_fastopen_request
{
1562 /* Fast Open cookie. Size 0 means a cookie request */
1563 struct tcp_fastopen_cookie cookie
;
1564 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1566 int copied
; /* queued in tcp_connect() */
1568 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1570 extern struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
1571 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
);
1572 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
);
1573 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1574 struct request_sock
*req
,
1575 struct tcp_fastopen_cookie
*foc
);
1576 void tcp_fastopen_init_key_once(bool publish
);
1577 bool tcp_fastopen_cookie_check(struct sock
*sk
, u16
*mss
,
1578 struct tcp_fastopen_cookie
*cookie
);
1579 bool tcp_fastopen_defer_connect(struct sock
*sk
, int *err
);
1580 #define TCP_FASTOPEN_KEY_LENGTH 16
1582 /* Fastopen key context */
1583 struct tcp_fastopen_context
{
1584 struct crypto_cipher
*tfm
;
1585 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1586 struct rcu_head rcu
;
1589 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout
;
1590 void tcp_fastopen_active_disable(struct sock
*sk
);
1591 bool tcp_fastopen_active_should_disable(struct sock
*sk
);
1592 void tcp_fastopen_active_disable_ofo_check(struct sock
*sk
);
1593 void tcp_fastopen_active_timeout_reset(void);
1595 /* Latencies incurred by various limits for a sender. They are
1596 * chronograph-like stats that are mutually exclusive.
1600 TCP_CHRONO_BUSY
, /* Actively sending data (non-empty write queue) */
1601 TCP_CHRONO_RWND_LIMITED
, /* Stalled by insufficient receive window */
1602 TCP_CHRONO_SNDBUF_LIMITED
, /* Stalled by insufficient send buffer */
1606 void tcp_chrono_start(struct sock
*sk
, const enum tcp_chrono type
);
1607 void tcp_chrono_stop(struct sock
*sk
, const enum tcp_chrono type
);
1609 static inline void tcp_init_send_head(struct sock
*sk
)
1611 sk
->sk_send_head
= NULL
;
1614 static inline void tcp_init_send_head(struct sock
*sk
);
1616 /* write queue abstraction */
1617 static inline void tcp_write_queue_purge(struct sock
*sk
)
1619 struct sk_buff
*skb
;
1621 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1622 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1623 sk_wmem_free_skb(sk
, skb
);
1624 tcp_init_send_head(sk
);
1626 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1627 tcp_init_send_head(sk
);
1628 tcp_sk(sk
)->packets_out
= 0;
1629 inet_csk(sk
)->icsk_backoff
= 0;
1632 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1634 return skb_peek(&sk
->sk_write_queue
);
1637 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1639 return skb_peek_tail(&sk
->sk_write_queue
);
1642 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1643 const struct sk_buff
*skb
)
1645 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1648 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1649 const struct sk_buff
*skb
)
1651 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1654 #define tcp_for_write_queue(skb, sk) \
1655 skb_queue_walk(&(sk)->sk_write_queue, skb)
1657 #define tcp_for_write_queue_from(skb, sk) \
1658 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1660 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1661 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1663 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1665 return sk
->sk_send_head
;
1668 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1669 const struct sk_buff
*skb
)
1671 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1674 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1676 if (tcp_skb_is_last(sk
, skb
))
1677 sk
->sk_send_head
= NULL
;
1679 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1682 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1684 if (sk
->sk_send_head
== skb_unlinked
) {
1685 sk
->sk_send_head
= NULL
;
1686 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1688 if (tcp_sk(sk
)->highest_sack
== skb_unlinked
)
1689 tcp_sk(sk
)->highest_sack
= NULL
;
1692 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1694 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1697 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1699 __tcp_add_write_queue_tail(sk
, skb
);
1701 /* Queue it, remembering where we must start sending. */
1702 if (sk
->sk_send_head
== NULL
) {
1703 sk
->sk_send_head
= skb
;
1704 tcp_chrono_start(sk
, TCP_CHRONO_BUSY
);
1706 if (tcp_sk(sk
)->highest_sack
== NULL
)
1707 tcp_sk(sk
)->highest_sack
= skb
;
1711 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1713 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1716 /* Insert buff after skb on the write queue of sk. */
1717 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1718 struct sk_buff
*buff
,
1721 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1724 /* Insert new before skb on the write queue of sk. */
1725 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1726 struct sk_buff
*skb
,
1729 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1731 if (sk
->sk_send_head
== skb
)
1732 sk
->sk_send_head
= new;
1735 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1737 __skb_unlink(skb
, &sk
->sk_write_queue
);
1740 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1742 return skb_queue_empty(&sk
->sk_write_queue
);
1745 static inline void tcp_push_pending_frames(struct sock
*sk
)
1747 if (tcp_send_head(sk
)) {
1748 struct tcp_sock
*tp
= tcp_sk(sk
);
1750 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1754 /* Start sequence of the skb just after the highest skb with SACKed
1755 * bit, valid only if sacked_out > 0 or when the caller has ensured
1756 * validity by itself.
1758 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1760 if (!tp
->sacked_out
)
1763 if (tp
->highest_sack
== NULL
)
1766 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1769 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1771 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1772 tcp_write_queue_next(sk
, skb
);
1775 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1777 return tcp_sk(sk
)->highest_sack
;
1780 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1782 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1785 /* Called when old skb is about to be deleted and replaced by new skb */
1786 static inline void tcp_highest_sack_replace(struct sock
*sk
,
1787 struct sk_buff
*old
,
1788 struct sk_buff
*new)
1790 if (old
== tcp_highest_sack(sk
))
1791 tcp_sk(sk
)->highest_sack
= new;
1794 /* This helper checks if socket has IP_TRANSPARENT set */
1795 static inline bool inet_sk_transparent(const struct sock
*sk
)
1797 switch (sk
->sk_state
) {
1799 return inet_twsk(sk
)->tw_transparent
;
1800 case TCP_NEW_SYN_RECV
:
1801 return inet_rsk(inet_reqsk(sk
))->no_srccheck
;
1803 return inet_sk(sk
)->transparent
;
1806 /* Determines whether this is a thin stream (which may suffer from
1807 * increased latency). Used to trigger latency-reducing mechanisms.
1809 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1811 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1815 enum tcp_seq_states
{
1816 TCP_SEQ_STATE_LISTENING
,
1817 TCP_SEQ_STATE_ESTABLISHED
,
1820 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1822 struct tcp_seq_afinfo
{
1825 const struct file_operations
*seq_fops
;
1826 struct seq_operations seq_ops
;
1829 struct tcp_iter_state
{
1830 struct seq_net_private p
;
1832 enum tcp_seq_states state
;
1833 struct sock
*syn_wait_sk
;
1834 int bucket
, offset
, sbucket
, num
;
1838 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1839 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1841 extern struct request_sock_ops tcp_request_sock_ops
;
1842 extern struct request_sock_ops tcp6_request_sock_ops
;
1844 void tcp_v4_destroy_sock(struct sock
*sk
);
1846 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1847 netdev_features_t features
);
1848 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1849 int tcp_gro_complete(struct sk_buff
*skb
);
1851 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1853 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1855 struct net
*net
= sock_net((struct sock
*)tp
);
1856 return tp
->notsent_lowat
?: net
->ipv4
.sysctl_tcp_notsent_lowat
;
1859 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1861 const struct tcp_sock
*tp
= tcp_sk(sk
);
1862 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1864 return notsent_bytes
< tcp_notsent_lowat(tp
);
1867 #ifdef CONFIG_PROC_FS
1868 int tcp4_proc_init(void);
1869 void tcp4_proc_exit(void);
1872 int tcp_rtx_synack(const struct sock
*sk
, struct request_sock
*req
);
1873 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1874 const struct tcp_request_sock_ops
*af_ops
,
1875 struct sock
*sk
, struct sk_buff
*skb
);
1877 /* TCP af-specific functions */
1878 struct tcp_sock_af_ops
{
1879 #ifdef CONFIG_TCP_MD5SIG
1880 struct tcp_md5sig_key
*(*md5_lookup
) (const struct sock
*sk
,
1881 const struct sock
*addr_sk
);
1882 int (*calc_md5_hash
)(char *location
,
1883 const struct tcp_md5sig_key
*md5
,
1884 const struct sock
*sk
,
1885 const struct sk_buff
*skb
);
1886 int (*md5_parse
)(struct sock
*sk
,
1888 char __user
*optval
,
1893 struct tcp_request_sock_ops
{
1895 #ifdef CONFIG_TCP_MD5SIG
1896 struct tcp_md5sig_key
*(*req_md5_lookup
)(const struct sock
*sk
,
1897 const struct sock
*addr_sk
);
1898 int (*calc_md5_hash
) (char *location
,
1899 const struct tcp_md5sig_key
*md5
,
1900 const struct sock
*sk
,
1901 const struct sk_buff
*skb
);
1903 void (*init_req
)(struct request_sock
*req
,
1904 const struct sock
*sk_listener
,
1905 struct sk_buff
*skb
);
1906 #ifdef CONFIG_SYN_COOKIES
1907 __u32 (*cookie_init_seq
)(const struct sk_buff
*skb
,
1910 struct dst_entry
*(*route_req
)(const struct sock
*sk
, struct flowi
*fl
,
1911 const struct request_sock
*req
);
1912 u32 (*init_seq
)(const struct sk_buff
*skb
);
1913 u32 (*init_ts_off
)(const struct net
*net
, const struct sk_buff
*skb
);
1914 int (*send_synack
)(const struct sock
*sk
, struct dst_entry
*dst
,
1915 struct flowi
*fl
, struct request_sock
*req
,
1916 struct tcp_fastopen_cookie
*foc
,
1917 enum tcp_synack_type synack_type
);
1920 #ifdef CONFIG_SYN_COOKIES
1921 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1922 const struct sock
*sk
, struct sk_buff
*skb
,
1925 tcp_synq_overflow(sk
);
1926 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_SYNCOOKIESSENT
);
1927 return ops
->cookie_init_seq(skb
, mss
);
1930 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1931 const struct sock
*sk
, struct sk_buff
*skb
,
1938 int tcpv4_offload_init(void);
1940 void tcp_v4_init(void);
1941 void tcp_init(void);
1943 /* tcp_recovery.c */
1944 extern void tcp_rack_mark_lost(struct sock
*sk
);
1945 extern void tcp_rack_advance(struct tcp_sock
*tp
, u8 sacked
, u32 end_seq
,
1947 extern void tcp_rack_reo_timeout(struct sock
*sk
);
1949 /* At how many usecs into the future should the RTO fire? */
1950 static inline s64
tcp_rto_delta_us(const struct sock
*sk
)
1952 const struct sk_buff
*skb
= tcp_write_queue_head(sk
);
1953 u32 rto
= inet_csk(sk
)->icsk_rto
;
1954 u64 rto_time_stamp_us
= skb
->skb_mstamp
+ jiffies_to_usecs(rto
);
1956 return rto_time_stamp_us
- tcp_sk(sk
)->tcp_mstamp
;
1960 * Save and compile IPv4 options, return a pointer to it
1962 static inline struct ip_options_rcu
*tcp_v4_save_options(struct net
*net
,
1963 struct sk_buff
*skb
)
1965 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
1966 struct ip_options_rcu
*dopt
= NULL
;
1969 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
1971 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
1972 if (dopt
&& __ip_options_echo(net
, &dopt
->opt
, skb
, opt
)) {
1980 /* locally generated TCP pure ACKs have skb->truesize == 2
1981 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1982 * This is much faster than dissecting the packet to find out.
1983 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1985 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
1987 return skb
->truesize
== 2;
1990 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)
1995 static inline int tcp_inq(struct sock
*sk
)
1997 struct tcp_sock
*tp
= tcp_sk(sk
);
2000 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
2002 } else if (sock_flag(sk
, SOCK_URGINLINE
) ||
2004 before(tp
->urg_seq
, tp
->copied_seq
) ||
2005 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
2007 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
2009 /* Subtract 1, if FIN was received */
2010 if (answ
&& sock_flag(sk
, SOCK_DONE
))
2013 answ
= tp
->urg_seq
- tp
->copied_seq
;
2019 int tcp_peek_len(struct socket
*sock
);
2021 static inline void tcp_segs_in(struct tcp_sock
*tp
, const struct sk_buff
*skb
)
2025 segs_in
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
2026 tp
->segs_in
+= segs_in
;
2027 if (skb
->len
> tcp_hdrlen(skb
))
2028 tp
->data_segs_in
+= segs_in
;
2032 * TCP listen path runs lockless.
2033 * We forced "struct sock" to be const qualified to make sure
2034 * we don't modify one of its field by mistake.
2035 * Here, we increment sk_drops which is an atomic_t, so we can safely
2036 * make sock writable again.
2038 static inline void tcp_listendrop(const struct sock
*sk
)
2040 atomic_inc(&((struct sock
*)sk
)->sk_drops
);
2041 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
2044 enum hrtimer_restart
tcp_pace_kick(struct hrtimer
*timer
);
2047 * Interface for adding Upper Level Protocols over TCP
2050 #define TCP_ULP_NAME_MAX 16
2051 #define TCP_ULP_MAX 128
2052 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2054 struct tcp_ulp_ops
{
2055 struct list_head list
;
2057 /* initialize ulp */
2058 int (*init
)(struct sock
*sk
);
2060 void (*release
)(struct sock
*sk
);
2062 char name
[TCP_ULP_NAME_MAX
];
2063 struct module
*owner
;
2065 int tcp_register_ulp(struct tcp_ulp_ops
*type
);
2066 void tcp_unregister_ulp(struct tcp_ulp_ops
*type
);
2067 int tcp_set_ulp(struct sock
*sk
, const char *name
);
2068 void tcp_get_available_ulp(char *buf
, size_t len
);
2069 void tcp_cleanup_ulp(struct sock
*sk
);
2071 #define MODULE_ALIAS_TCP_ULP(name) \
2072 __MODULE_INFO(alias, alias_userspace, name); \
2073 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2075 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2076 * is < 0, then the BPF op failed (for example if the loaded BPF
2077 * program does not support the chosen operation or there is no BPF
2081 static inline int tcp_call_bpf(struct sock
*sk
, int op
)
2083 struct bpf_sock_ops_kern sock_ops
;
2086 if (sk_fullsock(sk
))
2087 sock_owned_by_me(sk
);
2089 memset(&sock_ops
, 0, sizeof(sock_ops
));
2093 ret
= BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops
);
2095 ret
= sock_ops
.reply
;
2101 static inline int tcp_call_bpf(struct sock
*sk
, int op
)
2107 static inline u32
tcp_timeout_init(struct sock
*sk
)
2111 timeout
= tcp_call_bpf(sk
, BPF_SOCK_OPS_TIMEOUT_INIT
);
2114 timeout
= TCP_TIMEOUT_INIT
;
2118 static inline u32
tcp_rwnd_init_bpf(struct sock
*sk
)
2122 rwnd
= tcp_call_bpf(sk
, BPF_SOCK_OPS_RWND_INIT
);
2129 static inline bool tcp_bpf_ca_needs_ecn(struct sock
*sk
)
2131 return (tcp_call_bpf(sk
, BPF_SOCK_OPS_NEEDS_ECN
) == 1);