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>
48 #include <linux/bpf-cgroup.h>
50 extern struct inet_hashinfo tcp_hashinfo
;
52 extern struct percpu_counter tcp_orphan_count
;
53 void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 1024
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL 600
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD 8
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
82 /* Maximal number of window scale according to RFC1323 */
83 #define TCP_MAX_WSCALE 14U
86 #define TCP_URG_VALID 0x0100
87 #define TCP_URG_NOTYET 0x0200
88 #define TCP_URG_READ 0x0400
90 #define TCP_RETR1 3 /*
91 * This is how many retries it does before it
92 * tries to figure out if the gateway is
93 * down. Minimal RFC value is 3; it corresponds
94 * to ~3sec-8min depending on RTO.
97 #define TCP_RETR2 15 /*
98 * This should take at least
99 * 90 minutes to time out.
100 * RFC1122 says that the limit is 100 sec.
101 * 15 is ~13-30min depending on RTO.
104 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
105 * when active opening a connection.
106 * RFC1122 says the minimum retry MUST
107 * be at least 180secs. Nevertheless
108 * this value is corresponding to
109 * 63secs of retransmission with the
110 * current initial RTO.
113 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
114 * when passive opening a connection.
115 * This is corresponding to 31secs of
116 * retransmission with the current
120 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 * state, about 60 seconds */
122 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
123 /* BSD style FIN_WAIT2 deadlock breaker.
124 * It used to be 3min, new value is 60sec,
125 * to combine FIN-WAIT-2 timeout with
129 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
131 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
132 #define TCP_ATO_MIN ((unsigned)(HZ/25))
134 #define TCP_DELACK_MIN 4U
135 #define TCP_ATO_MIN 4U
137 #define TCP_RTO_MAX ((unsigned)(120*HZ))
138 #define TCP_RTO_MIN ((unsigned)(HZ/5))
139 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
140 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
141 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
142 * used as a fallback RTO for the
143 * initial data transmission if no
144 * valid RTT sample has been acquired,
145 * most likely due to retrans in 3WHS.
148 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 * for local resources.
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
191 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
197 #define TCPOLEN_MSS 4
198 #define TCPOLEN_WINDOW 3
199 #define TCPOLEN_SACK_PERM 2
200 #define TCPOLEN_TIMESTAMP 10
201 #define TCPOLEN_MD5SIG 18
202 #define TCPOLEN_FASTOPEN_BASE 2
203 #define TCPOLEN_EXP_FASTOPEN_BASE 4
204 #define TCPOLEN_EXP_SMC_BASE 6
206 /* But this is what stacks really send out. */
207 #define TCPOLEN_TSTAMP_ALIGNED 12
208 #define TCPOLEN_WSCALE_ALIGNED 4
209 #define TCPOLEN_SACKPERM_ALIGNED 4
210 #define TCPOLEN_SACK_BASE 2
211 #define TCPOLEN_SACK_BASE_ALIGNED 4
212 #define TCPOLEN_SACK_PERBLOCK 8
213 #define TCPOLEN_MD5SIG_ALIGNED 20
214 #define TCPOLEN_MSS_ALIGNED 4
215 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
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_max_orphans
;
244 extern long sysctl_tcp_mem
[3];
246 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
247 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
249 extern atomic_long_t tcp_memory_allocated
;
250 extern struct percpu_counter tcp_sockets_allocated
;
251 extern unsigned long tcp_memory_pressure
;
253 /* optimized version of sk_under_memory_pressure() for TCP sockets */
254 static inline bool tcp_under_memory_pressure(const struct sock
*sk
)
256 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
257 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
260 return tcp_memory_pressure
;
263 * The next routines deal with comparing 32 bit unsigned ints
264 * and worry about wraparound (automatic with unsigned arithmetic).
267 static inline bool before(__u32 seq1
, __u32 seq2
)
269 return (__s32
)(seq1
-seq2
) < 0;
271 #define after(seq2, seq1) before(seq1, seq2)
273 /* is s2<=s1<=s3 ? */
274 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
276 return seq3
- seq2
>= seq1
- seq2
;
279 static inline bool tcp_out_of_memory(struct sock
*sk
)
281 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
282 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
287 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
289 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
291 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
292 int orphans
= percpu_counter_read_positive(ocp
);
294 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
295 orphans
= percpu_counter_sum_positive(ocp
);
296 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
302 bool tcp_check_oom(struct sock
*sk
, int shift
);
305 extern struct proto tcp_prot
;
307 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
308 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
312 void tcp_tasklet_init(void);
314 void tcp_v4_err(struct sk_buff
*skb
, u32
);
316 void tcp_shutdown(struct sock
*sk
, int how
);
318 int tcp_v4_early_demux(struct sk_buff
*skb
);
319 int tcp_v4_rcv(struct sk_buff
*skb
);
321 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
322 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
323 int tcp_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
324 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
326 int tcp_sendpage_locked(struct sock
*sk
, struct page
*page
, int offset
,
327 size_t size
, int flags
);
328 ssize_t
do_tcp_sendpages(struct sock
*sk
, struct page
*page
, int offset
,
329 size_t size
, int flags
);
330 void tcp_release_cb(struct sock
*sk
);
331 void tcp_wfree(struct sk_buff
*skb
);
332 void tcp_write_timer_handler(struct sock
*sk
);
333 void tcp_delack_timer_handler(struct sock
*sk
);
334 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
335 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
);
336 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
337 const struct tcphdr
*th
);
338 void tcp_rcv_space_adjust(struct sock
*sk
);
339 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
340 void tcp_twsk_destructor(struct sock
*sk
);
341 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
342 struct pipe_inode_info
*pipe
, size_t len
,
345 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
346 const unsigned int pkts
)
348 struct inet_connection_sock
*icsk
= inet_csk(sk
);
350 if (icsk
->icsk_ack
.quick
) {
351 if (pkts
>= icsk
->icsk_ack
.quick
) {
352 icsk
->icsk_ack
.quick
= 0;
353 /* Leaving quickack mode we deflate ATO. */
354 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
356 icsk
->icsk_ack
.quick
-= pkts
;
361 #define TCP_ECN_QUEUE_CWR 2
362 #define TCP_ECN_DEMAND_CWR 4
363 #define TCP_ECN_SEEN 8
373 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
375 const struct tcphdr
*th
);
376 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
377 struct request_sock
*req
, bool fastopen
);
378 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
379 struct sk_buff
*skb
);
380 void tcp_enter_loss(struct sock
*sk
);
381 void tcp_cwnd_reduction(struct sock
*sk
, int newly_acked_sacked
, int flag
);
382 void tcp_clear_retrans(struct tcp_sock
*tp
);
383 void tcp_update_metrics(struct sock
*sk
);
384 void tcp_init_metrics(struct sock
*sk
);
385 void tcp_metrics_init(void);
386 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
);
387 void tcp_close(struct sock
*sk
, long timeout
);
388 void tcp_init_sock(struct sock
*sk
);
389 void tcp_init_transfer(struct sock
*sk
, int bpf_op
);
390 __poll_t
tcp_poll(struct file
*file
, struct socket
*sock
,
391 struct poll_table_struct
*wait
);
392 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
393 char __user
*optval
, int __user
*optlen
);
394 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
395 char __user
*optval
, unsigned int optlen
);
396 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
397 char __user
*optval
, int __user
*optlen
);
398 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
399 char __user
*optval
, unsigned int optlen
);
400 void tcp_set_keepalive(struct sock
*sk
, int val
);
401 void tcp_syn_ack_timeout(const struct request_sock
*req
);
402 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
403 int flags
, int *addr_len
);
404 void tcp_parse_options(const struct net
*net
, const struct sk_buff
*skb
,
405 struct tcp_options_received
*opt_rx
,
406 int estab
, struct tcp_fastopen_cookie
*foc
);
407 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
410 * TCP v4 functions exported for the inet6 API
413 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
414 void tcp_v4_mtu_reduced(struct sock
*sk
);
415 void tcp_req_err(struct sock
*sk
, u32 seq
, bool abort
);
416 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
417 struct sock
*tcp_create_openreq_child(const struct sock
*sk
,
418 struct request_sock
*req
,
419 struct sk_buff
*skb
);
420 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
421 struct sock
*tcp_v4_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
422 struct request_sock
*req
,
423 struct dst_entry
*dst
,
424 struct request_sock
*req_unhash
,
426 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
427 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
428 int tcp_connect(struct sock
*sk
);
429 enum tcp_synack_type
{
434 struct sk_buff
*tcp_make_synack(const struct sock
*sk
, struct dst_entry
*dst
,
435 struct request_sock
*req
,
436 struct tcp_fastopen_cookie
*foc
,
437 enum tcp_synack_type synack_type
);
438 int tcp_disconnect(struct sock
*sk
, int flags
);
440 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
441 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
442 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
444 /* From syncookies.c */
445 struct sock
*tcp_get_cookie_sock(struct sock
*sk
, struct sk_buff
*skb
,
446 struct request_sock
*req
,
447 struct dst_entry
*dst
, u32 tsoff
);
448 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
450 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
451 #ifdef CONFIG_SYN_COOKIES
453 /* Syncookies use a monotonic timer which increments every 60 seconds.
454 * This counter is used both as a hash input and partially encoded into
455 * the cookie value. A cookie is only validated further if the delta
456 * between the current counter value and the encoded one is less than this,
457 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
458 * the counter advances immediately after a cookie is generated).
460 #define MAX_SYNCOOKIE_AGE 2
461 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
462 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
464 /* syncookies: remember time of last synqueue overflow
465 * But do not dirty this field too often (once per second is enough)
466 * It is racy as we do not hold a lock, but race is very minor.
468 static inline void tcp_synq_overflow(const struct sock
*sk
)
470 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
471 unsigned long now
= jiffies
;
473 if (time_after(now
, last_overflow
+ HZ
))
474 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= now
;
477 /* syncookies: no recent synqueue overflow on this listening socket? */
478 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
480 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
482 return time_after(jiffies
, last_overflow
+ TCP_SYNCOOKIE_VALID
);
485 static inline u32
tcp_cookie_time(void)
487 u64 val
= get_jiffies_64();
489 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
493 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
495 __u32
cookie_v4_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
496 u64
cookie_init_timestamp(struct request_sock
*req
);
497 bool cookie_timestamp_decode(const struct net
*net
,
498 struct tcp_options_received
*opt
);
499 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
500 const struct net
*net
, const struct dst_entry
*dst
);
502 /* From net/ipv6/syncookies.c */
503 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
505 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
507 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
508 const struct tcphdr
*th
, u16
*mssp
);
509 __u32
cookie_v6_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
513 u32
tcp_tso_autosize(const struct sock
*sk
, unsigned int mss_now
,
515 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
517 int __tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
518 int tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
519 void tcp_retransmit_timer(struct sock
*sk
);
520 void tcp_xmit_retransmit_queue(struct sock
*);
521 void tcp_simple_retransmit(struct sock
*);
522 void tcp_enter_recovery(struct sock
*sk
, bool ece_ack
);
523 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
525 TCP_FRAG_IN_WRITE_QUEUE
,
526 TCP_FRAG_IN_RTX_QUEUE
,
528 int tcp_fragment(struct sock
*sk
, enum tcp_queue tcp_queue
,
529 struct sk_buff
*skb
, u32 len
,
530 unsigned int mss_now
, gfp_t gfp
);
532 void tcp_send_probe0(struct sock
*);
533 void tcp_send_partial(struct sock
*);
534 int tcp_write_wakeup(struct sock
*, int mib
);
535 void tcp_send_fin(struct sock
*sk
);
536 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
537 int tcp_send_synack(struct sock
*);
538 void tcp_push_one(struct sock
*, unsigned int mss_now
);
539 void tcp_send_ack(struct sock
*sk
);
540 void tcp_send_delayed_ack(struct sock
*sk
);
541 void tcp_send_loss_probe(struct sock
*sk
);
542 bool tcp_schedule_loss_probe(struct sock
*sk
, bool advancing_rto
);
543 void tcp_skb_collapse_tstamp(struct sk_buff
*skb
,
544 const struct sk_buff
*next_skb
);
547 void tcp_rearm_rto(struct sock
*sk
);
548 void tcp_synack_rtt_meas(struct sock
*sk
, struct request_sock
*req
);
549 void tcp_reset(struct sock
*sk
);
550 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock
*tp
, struct sk_buff
*skb
);
551 void tcp_fin(struct sock
*sk
);
554 void tcp_init_xmit_timers(struct sock
*);
555 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
557 hrtimer_cancel(&tcp_sk(sk
)->pacing_timer
);
558 inet_csk_clear_xmit_timers(sk
);
561 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
562 unsigned int tcp_current_mss(struct sock
*sk
);
564 /* Bound MSS / TSO packet size with the half of the window */
565 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
569 /* When peer uses tiny windows, there is no use in packetizing
570 * to sub-MSS pieces for the sake of SWS or making sure there
571 * are enough packets in the pipe for fast recovery.
573 * On the other hand, for extremely large MSS devices, handling
574 * smaller than MSS windows in this way does make sense.
576 if (tp
->max_window
> TCP_MSS_DEFAULT
)
577 cutoff
= (tp
->max_window
>> 1);
579 cutoff
= tp
->max_window
;
581 if (cutoff
&& pktsize
> cutoff
)
582 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
588 void tcp_get_info(struct sock
*, struct tcp_info
*);
590 /* Read 'sendfile()'-style from a TCP socket */
591 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
592 sk_read_actor_t recv_actor
);
594 void tcp_initialize_rcv_mss(struct sock
*sk
);
596 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
597 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
598 void tcp_mtup_init(struct sock
*sk
);
599 void tcp_init_buffer_space(struct sock
*sk
);
601 static inline void tcp_bound_rto(const struct sock
*sk
)
603 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
604 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
607 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
609 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
612 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
614 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
615 ntohl(TCP_FLAG_ACK
) |
619 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
621 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
624 static inline void tcp_fast_path_check(struct sock
*sk
)
626 struct tcp_sock
*tp
= tcp_sk(sk
);
628 if (RB_EMPTY_ROOT(&tp
->out_of_order_queue
) &&
630 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
632 tcp_fast_path_on(tp
);
635 /* Compute the actual rto_min value */
636 static inline u32
tcp_rto_min(struct sock
*sk
)
638 const struct dst_entry
*dst
= __sk_dst_get(sk
);
639 u32 rto_min
= TCP_RTO_MIN
;
641 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
642 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
646 static inline u32
tcp_rto_min_us(struct sock
*sk
)
648 return jiffies_to_usecs(tcp_rto_min(sk
));
651 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
653 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
656 /* Minimum RTT in usec. ~0 means not available. */
657 static inline u32
tcp_min_rtt(const struct tcp_sock
*tp
)
659 return minmax_get(&tp
->rtt_min
);
662 /* Compute the actual receive window we are currently advertising.
663 * Rcv_nxt can be after the window if our peer push more data
664 * than the offered window.
666 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
668 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
675 /* Choose a new window, without checks for shrinking, and without
676 * scaling applied to the result. The caller does these things
677 * if necessary. This is a "raw" window selection.
679 u32
__tcp_select_window(struct sock
*sk
);
681 void tcp_send_window_probe(struct sock
*sk
);
683 /* TCP uses 32bit jiffies to save some space.
684 * Note that this is different from tcp_time_stamp, which
685 * historically has been the same until linux-4.13.
687 #define tcp_jiffies32 ((u32)jiffies)
690 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
691 * It is no longer tied to jiffies, but to 1 ms clock.
692 * Note: double check if you want to use tcp_jiffies32 instead of this.
694 #define TCP_TS_HZ 1000
696 static inline u64
tcp_clock_ns(void)
698 return local_clock();
701 static inline u64
tcp_clock_us(void)
703 return div_u64(tcp_clock_ns(), NSEC_PER_USEC
);
706 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
707 static inline u32
tcp_time_stamp(const struct tcp_sock
*tp
)
709 return div_u64(tp
->tcp_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
712 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
713 static inline u32
tcp_time_stamp_raw(void)
715 return div_u64(tcp_clock_ns(), NSEC_PER_SEC
/ TCP_TS_HZ
);
719 /* Refresh 1us clock of a TCP socket,
720 * ensuring monotically increasing values.
722 static inline void tcp_mstamp_refresh(struct tcp_sock
*tp
)
724 u64 val
= tcp_clock_us();
726 if (val
> tp
->tcp_mstamp
)
727 tp
->tcp_mstamp
= val
;
730 static inline u32
tcp_stamp_us_delta(u64 t1
, u64 t0
)
732 return max_t(s64
, t1
- t0
, 0);
735 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
737 return div_u64(skb
->skb_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
741 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
743 #define TCPHDR_FIN 0x01
744 #define TCPHDR_SYN 0x02
745 #define TCPHDR_RST 0x04
746 #define TCPHDR_PSH 0x08
747 #define TCPHDR_ACK 0x10
748 #define TCPHDR_URG 0x20
749 #define TCPHDR_ECE 0x40
750 #define TCPHDR_CWR 0x80
752 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
754 /* This is what the send packet queuing engine uses to pass
755 * TCP per-packet control information to the transmission code.
756 * We also store the host-order sequence numbers in here too.
757 * This is 44 bytes if IPV6 is enabled.
758 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
761 __u32 seq
; /* Starting sequence number */
762 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
764 /* Note : tcp_tw_isn is used in input path only
765 * (isn chosen by tcp_timewait_state_process())
767 * tcp_gso_segs/size are used in write queue only,
768 * cf tcp_skb_pcount()/tcp_skb_mss()
776 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
778 __u8 sacked
; /* State flags for SACK. */
779 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
780 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
781 #define TCPCB_LOST 0x04 /* SKB is lost */
782 #define TCPCB_TAGBITS 0x07 /* All tag bits */
783 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
784 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
785 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
788 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
789 __u8 txstamp_ack
:1, /* Record TX timestamp for ack? */
790 eor
:1, /* Is skb MSG_EOR marked? */
791 has_rxtstamp
:1, /* SKB has a RX timestamp */
793 __u32 ack_seq
; /* Sequence number ACK'd */
796 /* There is space for up to 24 bytes */
797 __u32 in_flight
:30,/* Bytes in flight at transmit */
798 is_app_limited
:1, /* cwnd not fully used? */
800 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
802 /* start of send pipeline phase */
804 /* when we reached the "delivered" count */
805 u64 delivered_mstamp
;
806 } tx
; /* only used for outgoing skbs */
808 struct inet_skb_parm h4
;
809 #if IS_ENABLED(CONFIG_IPV6)
810 struct inet6_skb_parm h6
;
812 } header
; /* For incoming skbs */
822 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
825 #if IS_ENABLED(CONFIG_IPV6)
826 /* This is the variant of inet6_iif() that must be used by TCP,
827 * as TCP moves IP6CB into a different location in skb->cb[]
829 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
831 bool l3_slave
= ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
);
833 return l3_slave
? skb
->skb_iif
: TCP_SKB_CB(skb
)->header
.h6
.iif
;
836 /* TCP_SKB_CB reference means this can not be used from early demux */
837 static inline int tcp_v6_sdif(const struct sk_buff
*skb
)
839 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
840 if (skb
&& ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
))
841 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
847 static inline bool inet_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
849 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
850 if (!net
->ipv4
.sysctl_tcp_l3mdev_accept
&&
851 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
857 /* TCP_SKB_CB reference means this can not be used from early demux */
858 static inline int tcp_v4_sdif(struct sk_buff
*skb
)
860 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
861 if (skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
862 return TCP_SKB_CB(skb
)->header
.h4
.iif
;
867 /* Due to TSO, an SKB can be composed of multiple actual
868 * packets. To keep these tracked properly, we use this.
870 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
872 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
875 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
877 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
880 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
882 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
885 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
886 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
888 return TCP_SKB_CB(skb
)->tcp_gso_size
;
891 static inline bool tcp_skb_can_collapse_to(const struct sk_buff
*skb
)
893 return likely(!TCP_SKB_CB(skb
)->eor
);
896 /* Events passed to congestion control interface */
898 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
899 CA_EVENT_CWND_RESTART
, /* congestion window restart */
900 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
901 CA_EVENT_LOSS
, /* loss timeout */
902 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
903 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
904 CA_EVENT_DELAYED_ACK
, /* Delayed ack is sent */
905 CA_EVENT_NON_DELAYED_ACK
,
908 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
909 enum tcp_ca_ack_event_flags
{
910 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
911 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
912 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
916 * Interface for adding new TCP congestion control handlers
918 #define TCP_CA_NAME_MAX 16
919 #define TCP_CA_MAX 128
920 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
922 #define TCP_CA_UNSPEC 0
924 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
925 #define TCP_CONG_NON_RESTRICTED 0x1
926 /* Requires ECN/ECT set on all packets */
927 #define TCP_CONG_NEEDS_ECN 0x2
937 /* A rate sample measures the number of (original/retransmitted) data
938 * packets delivered "delivered" over an interval of time "interval_us".
939 * The tcp_rate.c code fills in the rate sample, and congestion
940 * control modules that define a cong_control function to run at the end
941 * of ACK processing can optionally chose to consult this sample when
942 * setting cwnd and pacing rate.
943 * A sample is invalid if "delivered" or "interval_us" is negative.
946 u64 prior_mstamp
; /* starting timestamp for interval */
947 u32 prior_delivered
; /* tp->delivered at "prior_mstamp" */
948 s32 delivered
; /* number of packets delivered over interval */
949 long interval_us
; /* time for tp->delivered to incr "delivered" */
950 long rtt_us
; /* RTT of last (S)ACKed packet (or -1) */
951 int losses
; /* number of packets marked lost upon ACK */
952 u32 acked_sacked
; /* number of packets newly (S)ACKed upon ACK */
953 u32 prior_in_flight
; /* in flight before this ACK */
954 bool is_app_limited
; /* is sample from packet with bubble in pipe? */
955 bool is_retrans
; /* is sample from retransmission? */
956 bool is_ack_delayed
; /* is this (likely) a delayed ACK? */
959 struct tcp_congestion_ops
{
960 struct list_head list
;
964 /* initialize private data (optional) */
965 void (*init
)(struct sock
*sk
);
966 /* cleanup private data (optional) */
967 void (*release
)(struct sock
*sk
);
969 /* return slow start threshold (required) */
970 u32 (*ssthresh
)(struct sock
*sk
);
971 /* do new cwnd calculation (required) */
972 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
973 /* call before changing ca_state (optional) */
974 void (*set_state
)(struct sock
*sk
, u8 new_state
);
975 /* call when cwnd event occurs (optional) */
976 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
977 /* call when ack arrives (optional) */
978 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
979 /* new value of cwnd after loss (required) */
980 u32 (*undo_cwnd
)(struct sock
*sk
);
981 /* hook for packet ack accounting (optional) */
982 void (*pkts_acked
)(struct sock
*sk
, const struct ack_sample
*sample
);
983 /* suggest number of segments for each skb to transmit (optional) */
984 u32 (*tso_segs_goal
)(struct sock
*sk
);
985 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
986 u32 (*sndbuf_expand
)(struct sock
*sk
);
987 /* call when packets are delivered to update cwnd and pacing rate,
988 * after all the ca_state processing. (optional)
990 void (*cong_control
)(struct sock
*sk
, const struct rate_sample
*rs
);
991 /* get info for inet_diag (optional) */
992 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
993 union tcp_cc_info
*info
);
995 char name
[TCP_CA_NAME_MAX
];
996 struct module
*owner
;
999 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1000 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1002 void tcp_assign_congestion_control(struct sock
*sk
);
1003 void tcp_init_congestion_control(struct sock
*sk
);
1004 void tcp_cleanup_congestion_control(struct sock
*sk
);
1005 int tcp_set_default_congestion_control(struct net
*net
, const char *name
);
1006 void tcp_get_default_congestion_control(struct net
*net
, char *name
);
1007 void tcp_get_available_congestion_control(char *buf
, size_t len
);
1008 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
1009 int tcp_set_allowed_congestion_control(char *allowed
);
1010 int tcp_set_congestion_control(struct sock
*sk
, const char *name
, bool load
, bool reinit
);
1011 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
1012 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
1014 u32
tcp_reno_ssthresh(struct sock
*sk
);
1015 u32
tcp_reno_undo_cwnd(struct sock
*sk
);
1016 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
1017 extern struct tcp_congestion_ops tcp_reno
;
1019 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
1020 u32
tcp_ca_get_key_by_name(struct net
*net
, const char *name
, bool *ecn_ca
);
1022 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
1024 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
1030 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
1032 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1034 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
1037 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
1039 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1041 if (icsk
->icsk_ca_ops
->set_state
)
1042 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
1043 icsk
->icsk_ca_state
= ca_state
;
1046 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
1048 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1050 if (icsk
->icsk_ca_ops
->cwnd_event
)
1051 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
1054 /* From tcp_rate.c */
1055 void tcp_rate_skb_sent(struct sock
*sk
, struct sk_buff
*skb
);
1056 void tcp_rate_skb_delivered(struct sock
*sk
, struct sk_buff
*skb
,
1057 struct rate_sample
*rs
);
1058 void tcp_rate_gen(struct sock
*sk
, u32 delivered
, u32 lost
,
1059 bool is_sack_reneg
, struct rate_sample
*rs
);
1060 void tcp_rate_check_app_limited(struct sock
*sk
);
1062 /* These functions determine how the current flow behaves in respect of SACK
1063 * handling. SACK is negotiated with the peer, and therefore it can vary
1064 * between different flows.
1066 * tcp_is_sack - SACK enabled
1067 * tcp_is_reno - No SACK
1069 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
1071 return tp
->rx_opt
.sack_ok
;
1074 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
1076 return !tcp_is_sack(tp
);
1079 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
1081 return tp
->sacked_out
+ tp
->lost_out
;
1084 /* This determines how many packets are "in the network" to the best
1085 * of our knowledge. In many cases it is conservative, but where
1086 * detailed information is available from the receiver (via SACK
1087 * blocks etc.) we can make more aggressive calculations.
1089 * Use this for decisions involving congestion control, use just
1090 * tp->packets_out to determine if the send queue is empty or not.
1092 * Read this equation as:
1094 * "Packets sent once on transmission queue" MINUS
1095 * "Packets left network, but not honestly ACKed yet" PLUS
1096 * "Packets fast retransmitted"
1098 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1100 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
1103 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1105 static inline bool tcp_in_slow_start(const struct tcp_sock
*tp
)
1107 return tp
->snd_cwnd
< tp
->snd_ssthresh
;
1110 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
1112 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
1115 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
1117 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
1118 (1 << inet_csk(sk
)->icsk_ca_state
);
1121 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1122 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1125 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
1127 const struct tcp_sock
*tp
= tcp_sk(sk
);
1129 if (tcp_in_cwnd_reduction(sk
))
1130 return tp
->snd_ssthresh
;
1132 return max(tp
->snd_ssthresh
,
1133 ((tp
->snd_cwnd
>> 1) +
1134 (tp
->snd_cwnd
>> 2)));
1137 /* Use define here intentionally to get WARN_ON location shown at the caller */
1138 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1140 void tcp_enter_cwr(struct sock
*sk
);
1141 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1143 /* The maximum number of MSS of available cwnd for which TSO defers
1144 * sending if not using sysctl_tcp_tso_win_divisor.
1146 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1151 /* Returns end sequence number of the receiver's advertised window */
1152 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1154 return tp
->snd_una
+ tp
->snd_wnd
;
1157 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1158 * flexible approach. The RFC suggests cwnd should not be raised unless
1159 * it was fully used previously. And that's exactly what we do in
1160 * congestion avoidance mode. But in slow start we allow cwnd to grow
1161 * as long as the application has used half the cwnd.
1163 * cwnd is 10 (IW10), but application sends 9 frames.
1164 * We allow cwnd to reach 18 when all frames are ACKed.
1165 * This check is safe because it's as aggressive as slow start which already
1166 * risks 100% overshoot. The advantage is that we discourage application to
1167 * either send more filler packets or data to artificially blow up the cwnd
1168 * usage, and allow application-limited process to probe bw more aggressively.
1170 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1172 const struct tcp_sock
*tp
= tcp_sk(sk
);
1174 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1175 if (tcp_in_slow_start(tp
))
1176 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1178 return tp
->is_cwnd_limited
;
1181 /* Something is really bad, we could not queue an additional packet,
1182 * because qdisc is full or receiver sent a 0 window.
1183 * We do not want to add fuel to the fire, or abort too early,
1184 * so make sure the timer we arm now is at least 200ms in the future,
1185 * regardless of current icsk_rto value (as it could be ~2ms)
1187 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1189 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1192 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1193 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1194 unsigned long max_when
)
1196 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1198 return (unsigned long)min_t(u64
, when
, max_when
);
1201 static inline void tcp_check_probe_timer(struct sock
*sk
)
1203 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1204 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1205 tcp_probe0_base(sk
), TCP_RTO_MAX
);
1208 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1213 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1219 * Calculate(/check) TCP checksum
1221 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1222 __be32 daddr
, __wsum base
)
1224 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1227 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1229 return __skb_checksum_complete(skb
);
1232 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1234 return !skb_csum_unnecessary(skb
) &&
1235 __tcp_checksum_complete(skb
);
1238 bool tcp_add_backlog(struct sock
*sk
, struct sk_buff
*skb
);
1239 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
);
1244 static const char *statename
[]={
1245 "Unused","Established","Syn Sent","Syn Recv",
1246 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1247 "Close Wait","Last ACK","Listen","Closing"
1250 void tcp_set_state(struct sock
*sk
, int state
);
1252 void tcp_done(struct sock
*sk
);
1254 int tcp_abort(struct sock
*sk
, int err
);
1256 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1259 rx_opt
->num_sacks
= 0;
1262 u32
tcp_default_init_rwnd(u32 mss
);
1263 void tcp_cwnd_restart(struct sock
*sk
, s32 delta
);
1265 static inline void tcp_slow_start_after_idle_check(struct sock
*sk
)
1267 const struct tcp_congestion_ops
*ca_ops
= inet_csk(sk
)->icsk_ca_ops
;
1268 struct tcp_sock
*tp
= tcp_sk(sk
);
1271 if (!sock_net(sk
)->ipv4
.sysctl_tcp_slow_start_after_idle
|| tp
->packets_out
||
1272 ca_ops
->cong_control
)
1274 delta
= tcp_jiffies32
- tp
->lsndtime
;
1275 if (delta
> inet_csk(sk
)->icsk_rto
)
1276 tcp_cwnd_restart(sk
, delta
);
1279 /* Determine a window scaling and initial window to offer. */
1280 void tcp_select_initial_window(const struct sock
*sk
, int __space
,
1281 __u32 mss
, __u32
*rcv_wnd
,
1282 __u32
*window_clamp
, int wscale_ok
,
1283 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1285 static inline int tcp_win_from_space(const struct sock
*sk
, int space
)
1287 int tcp_adv_win_scale
= sock_net(sk
)->ipv4
.sysctl_tcp_adv_win_scale
;
1289 return tcp_adv_win_scale
<= 0 ?
1290 (space
>>(-tcp_adv_win_scale
)) :
1291 space
- (space
>>tcp_adv_win_scale
);
1294 /* Note: caller must be prepared to deal with negative returns */
1295 static inline int tcp_space(const struct sock
*sk
)
1297 return tcp_win_from_space(sk
, sk
->sk_rcvbuf
-
1298 atomic_read(&sk
->sk_rmem_alloc
));
1301 static inline int tcp_full_space(const struct sock
*sk
)
1303 return tcp_win_from_space(sk
, sk
->sk_rcvbuf
);
1306 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1307 const struct sock
*sk_listener
,
1308 const struct dst_entry
*dst
);
1310 void tcp_enter_memory_pressure(struct sock
*sk
);
1311 void tcp_leave_memory_pressure(struct sock
*sk
);
1313 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1315 struct net
*net
= sock_net((struct sock
*)tp
);
1317 return tp
->keepalive_intvl
? : net
->ipv4
.sysctl_tcp_keepalive_intvl
;
1320 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1322 struct net
*net
= sock_net((struct sock
*)tp
);
1324 return tp
->keepalive_time
? : net
->ipv4
.sysctl_tcp_keepalive_time
;
1327 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1329 struct net
*net
= sock_net((struct sock
*)tp
);
1331 return tp
->keepalive_probes
? : net
->ipv4
.sysctl_tcp_keepalive_probes
;
1334 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1336 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1338 return min_t(u32
, tcp_jiffies32
- icsk
->icsk_ack
.lrcvtime
,
1339 tcp_jiffies32
- tp
->rcv_tstamp
);
1342 static inline int tcp_fin_time(const struct sock
*sk
)
1344 int fin_timeout
= tcp_sk(sk
)->linger2
? : sock_net(sk
)->ipv4
.sysctl_tcp_fin_timeout
;
1345 const int rto
= inet_csk(sk
)->icsk_rto
;
1347 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1348 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1353 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1356 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1358 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1361 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1362 * then following tcp messages have valid values. Ignore 0 value,
1363 * or else 'negative' tsval might forbid us to accept their packets.
1365 if (!rx_opt
->ts_recent
)
1370 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1373 if (tcp_paws_check(rx_opt
, 0))
1376 /* RST segments are not recommended to carry timestamp,
1377 and, if they do, it is recommended to ignore PAWS because
1378 "their cleanup function should take precedence over timestamps."
1379 Certainly, it is mistake. It is necessary to understand the reasons
1380 of this constraint to relax it: if peer reboots, clock may go
1381 out-of-sync and half-open connections will not be reset.
1382 Actually, the problem would be not existing if all
1383 the implementations followed draft about maintaining clock
1384 via reboots. Linux-2.2 DOES NOT!
1386 However, we can relax time bounds for RST segments to MSL.
1388 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1393 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1394 int mib_idx
, u32
*last_oow_ack_time
);
1396 static inline void tcp_mib_init(struct net
*net
)
1399 TCP_ADD_STATS(net
, TCP_MIB_RTOALGORITHM
, 1);
1400 TCP_ADD_STATS(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1401 TCP_ADD_STATS(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1402 TCP_ADD_STATS(net
, TCP_MIB_MAXCONN
, -1);
1406 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1408 tp
->lost_skb_hint
= NULL
;
1411 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1413 tcp_clear_retrans_hints_partial(tp
);
1414 tp
->retransmit_skb_hint
= NULL
;
1417 union tcp_md5_addr
{
1419 #if IS_ENABLED(CONFIG_IPV6)
1424 /* - key database */
1425 struct tcp_md5sig_key
{
1426 struct hlist_node node
;
1428 u8 family
; /* AF_INET or AF_INET6 */
1429 union tcp_md5_addr addr
;
1431 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1432 struct rcu_head rcu
;
1436 struct tcp_md5sig_info
{
1437 struct hlist_head head
;
1438 struct rcu_head rcu
;
1441 /* - pseudo header */
1442 struct tcp4_pseudohdr
{
1450 struct tcp6_pseudohdr
{
1451 struct in6_addr saddr
;
1452 struct in6_addr daddr
;
1454 __be32 protocol
; /* including padding */
1457 union tcp_md5sum_block
{
1458 struct tcp4_pseudohdr ip4
;
1459 #if IS_ENABLED(CONFIG_IPV6)
1460 struct tcp6_pseudohdr ip6
;
1464 /* - pool: digest algorithm, hash description and scratch buffer */
1465 struct tcp_md5sig_pool
{
1466 struct ahash_request
*md5_req
;
1471 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1472 const struct sock
*sk
, const struct sk_buff
*skb
);
1473 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1474 int family
, u8 prefixlen
, const u8
*newkey
, u8 newkeylen
,
1476 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1477 int family
, u8 prefixlen
);
1478 struct tcp_md5sig_key
*tcp_v4_md5_lookup(const struct sock
*sk
,
1479 const struct sock
*addr_sk
);
1481 #ifdef CONFIG_TCP_MD5SIG
1482 struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1483 const union tcp_md5_addr
*addr
,
1485 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1487 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1488 const union tcp_md5_addr
*addr
,
1493 #define tcp_twsk_md5_key(twsk) NULL
1496 bool tcp_alloc_md5sig_pool(void);
1498 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1499 static inline void tcp_put_md5sig_pool(void)
1504 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1505 unsigned int header_len
);
1506 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1507 const struct tcp_md5sig_key
*key
);
1509 /* From tcp_fastopen.c */
1510 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1511 struct tcp_fastopen_cookie
*cookie
);
1512 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1513 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1515 struct tcp_fastopen_request
{
1516 /* Fast Open cookie. Size 0 means a cookie request */
1517 struct tcp_fastopen_cookie cookie
;
1518 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1520 int copied
; /* queued in tcp_connect() */
1522 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1523 void tcp_fastopen_destroy_cipher(struct sock
*sk
);
1524 void tcp_fastopen_ctx_destroy(struct net
*net
);
1525 int tcp_fastopen_reset_cipher(struct net
*net
, struct sock
*sk
,
1526 void *key
, unsigned int len
);
1527 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
);
1528 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1529 struct request_sock
*req
,
1530 struct tcp_fastopen_cookie
*foc
,
1531 const struct dst_entry
*dst
);
1532 void tcp_fastopen_init_key_once(struct net
*net
);
1533 bool tcp_fastopen_cookie_check(struct sock
*sk
, u16
*mss
,
1534 struct tcp_fastopen_cookie
*cookie
);
1535 bool tcp_fastopen_defer_connect(struct sock
*sk
, int *err
);
1536 #define TCP_FASTOPEN_KEY_LENGTH 16
1538 /* Fastopen key context */
1539 struct tcp_fastopen_context
{
1540 struct crypto_cipher
*tfm
;
1541 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1542 struct rcu_head rcu
;
1545 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout
;
1546 void tcp_fastopen_active_disable(struct sock
*sk
);
1547 bool tcp_fastopen_active_should_disable(struct sock
*sk
);
1548 void tcp_fastopen_active_disable_ofo_check(struct sock
*sk
);
1549 void tcp_fastopen_active_detect_blackhole(struct sock
*sk
, bool expired
);
1551 /* Latencies incurred by various limits for a sender. They are
1552 * chronograph-like stats that are mutually exclusive.
1556 TCP_CHRONO_BUSY
, /* Actively sending data (non-empty write queue) */
1557 TCP_CHRONO_RWND_LIMITED
, /* Stalled by insufficient receive window */
1558 TCP_CHRONO_SNDBUF_LIMITED
, /* Stalled by insufficient send buffer */
1562 void tcp_chrono_start(struct sock
*sk
, const enum tcp_chrono type
);
1563 void tcp_chrono_stop(struct sock
*sk
, const enum tcp_chrono type
);
1565 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1566 * the same memory storage than skb->destructor/_skb_refdst
1568 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff
*skb
)
1570 skb
->destructor
= NULL
;
1571 skb
->_skb_refdst
= 0UL;
1574 #define tcp_skb_tsorted_save(skb) { \
1575 unsigned long _save = skb->_skb_refdst; \
1576 skb->_skb_refdst = 0UL;
1578 #define tcp_skb_tsorted_restore(skb) \
1579 skb->_skb_refdst = _save; \
1582 void tcp_write_queue_purge(struct sock
*sk
);
1584 static inline struct sk_buff
*tcp_rtx_queue_head(const struct sock
*sk
)
1586 return skb_rb_first(&sk
->tcp_rtx_queue
);
1589 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1591 return skb_peek(&sk
->sk_write_queue
);
1594 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1596 return skb_peek_tail(&sk
->sk_write_queue
);
1599 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1600 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1602 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1604 return skb_peek(&sk
->sk_write_queue
);
1607 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1608 const struct sk_buff
*skb
)
1610 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1613 static inline bool tcp_write_queue_empty(const struct sock
*sk
)
1615 return skb_queue_empty(&sk
->sk_write_queue
);
1618 static inline bool tcp_rtx_queue_empty(const struct sock
*sk
)
1620 return RB_EMPTY_ROOT(&sk
->tcp_rtx_queue
);
1623 static inline bool tcp_rtx_and_write_queues_empty(const struct sock
*sk
)
1625 return tcp_rtx_queue_empty(sk
) && tcp_write_queue_empty(sk
);
1628 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1630 if (tcp_write_queue_empty(sk
))
1631 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1634 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1636 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1639 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1641 __tcp_add_write_queue_tail(sk
, skb
);
1643 /* Queue it, remembering where we must start sending. */
1644 if (sk
->sk_write_queue
.next
== skb
)
1645 tcp_chrono_start(sk
, TCP_CHRONO_BUSY
);
1648 /* Insert new before skb on the write queue of sk. */
1649 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1650 struct sk_buff
*skb
,
1653 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1656 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1658 tcp_skb_tsorted_anchor_cleanup(skb
);
1659 __skb_unlink(skb
, &sk
->sk_write_queue
);
1662 void tcp_rbtree_insert(struct rb_root
*root
, struct sk_buff
*skb
);
1664 static inline void tcp_rtx_queue_unlink(struct sk_buff
*skb
, struct sock
*sk
)
1666 tcp_skb_tsorted_anchor_cleanup(skb
);
1667 rb_erase(&skb
->rbnode
, &sk
->tcp_rtx_queue
);
1670 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff
*skb
, struct sock
*sk
)
1672 list_del(&skb
->tcp_tsorted_anchor
);
1673 tcp_rtx_queue_unlink(skb
, sk
);
1674 sk_wmem_free_skb(sk
, skb
);
1677 static inline void tcp_push_pending_frames(struct sock
*sk
)
1679 if (tcp_send_head(sk
)) {
1680 struct tcp_sock
*tp
= tcp_sk(sk
);
1682 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1686 /* Start sequence of the skb just after the highest skb with SACKed
1687 * bit, valid only if sacked_out > 0 or when the caller has ensured
1688 * validity by itself.
1690 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1692 if (!tp
->sacked_out
)
1695 if (tp
->highest_sack
== NULL
)
1698 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1701 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1703 tcp_sk(sk
)->highest_sack
= skb_rb_next(skb
);
1706 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1708 return tcp_sk(sk
)->highest_sack
;
1711 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1713 tcp_sk(sk
)->highest_sack
= tcp_rtx_queue_head(sk
);
1716 /* Called when old skb is about to be deleted and replaced by new skb */
1717 static inline void tcp_highest_sack_replace(struct sock
*sk
,
1718 struct sk_buff
*old
,
1719 struct sk_buff
*new)
1721 if (old
== tcp_highest_sack(sk
))
1722 tcp_sk(sk
)->highest_sack
= new;
1725 /* This helper checks if socket has IP_TRANSPARENT set */
1726 static inline bool inet_sk_transparent(const struct sock
*sk
)
1728 switch (sk
->sk_state
) {
1730 return inet_twsk(sk
)->tw_transparent
;
1731 case TCP_NEW_SYN_RECV
:
1732 return inet_rsk(inet_reqsk(sk
))->no_srccheck
;
1734 return inet_sk(sk
)->transparent
;
1737 /* Determines whether this is a thin stream (which may suffer from
1738 * increased latency). Used to trigger latency-reducing mechanisms.
1740 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1742 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1746 enum tcp_seq_states
{
1747 TCP_SEQ_STATE_LISTENING
,
1748 TCP_SEQ_STATE_ESTABLISHED
,
1751 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1753 struct tcp_seq_afinfo
{
1756 const struct file_operations
*seq_fops
;
1757 struct seq_operations seq_ops
;
1760 struct tcp_iter_state
{
1761 struct seq_net_private p
;
1763 enum tcp_seq_states state
;
1764 struct sock
*syn_wait_sk
;
1765 int bucket
, offset
, sbucket
, num
;
1769 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1770 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1772 extern struct request_sock_ops tcp_request_sock_ops
;
1773 extern struct request_sock_ops tcp6_request_sock_ops
;
1775 void tcp_v4_destroy_sock(struct sock
*sk
);
1777 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1778 netdev_features_t features
);
1779 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1780 int tcp_gro_complete(struct sk_buff
*skb
);
1782 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1784 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1786 struct net
*net
= sock_net((struct sock
*)tp
);
1787 return tp
->notsent_lowat
?: net
->ipv4
.sysctl_tcp_notsent_lowat
;
1790 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1792 const struct tcp_sock
*tp
= tcp_sk(sk
);
1793 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1795 return notsent_bytes
< tcp_notsent_lowat(tp
);
1798 #ifdef CONFIG_PROC_FS
1799 int tcp4_proc_init(void);
1800 void tcp4_proc_exit(void);
1803 int tcp_rtx_synack(const struct sock
*sk
, struct request_sock
*req
);
1804 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1805 const struct tcp_request_sock_ops
*af_ops
,
1806 struct sock
*sk
, struct sk_buff
*skb
);
1808 /* TCP af-specific functions */
1809 struct tcp_sock_af_ops
{
1810 #ifdef CONFIG_TCP_MD5SIG
1811 struct tcp_md5sig_key
*(*md5_lookup
) (const struct sock
*sk
,
1812 const struct sock
*addr_sk
);
1813 int (*calc_md5_hash
)(char *location
,
1814 const struct tcp_md5sig_key
*md5
,
1815 const struct sock
*sk
,
1816 const struct sk_buff
*skb
);
1817 int (*md5_parse
)(struct sock
*sk
,
1819 char __user
*optval
,
1824 struct tcp_request_sock_ops
{
1826 #ifdef CONFIG_TCP_MD5SIG
1827 struct tcp_md5sig_key
*(*req_md5_lookup
)(const struct sock
*sk
,
1828 const struct sock
*addr_sk
);
1829 int (*calc_md5_hash
) (char *location
,
1830 const struct tcp_md5sig_key
*md5
,
1831 const struct sock
*sk
,
1832 const struct sk_buff
*skb
);
1834 void (*init_req
)(struct request_sock
*req
,
1835 const struct sock
*sk_listener
,
1836 struct sk_buff
*skb
);
1837 #ifdef CONFIG_SYN_COOKIES
1838 __u32 (*cookie_init_seq
)(const struct sk_buff
*skb
,
1841 struct dst_entry
*(*route_req
)(const struct sock
*sk
, struct flowi
*fl
,
1842 const struct request_sock
*req
);
1843 u32 (*init_seq
)(const struct sk_buff
*skb
);
1844 u32 (*init_ts_off
)(const struct net
*net
, const struct sk_buff
*skb
);
1845 int (*send_synack
)(const struct sock
*sk
, struct dst_entry
*dst
,
1846 struct flowi
*fl
, struct request_sock
*req
,
1847 struct tcp_fastopen_cookie
*foc
,
1848 enum tcp_synack_type synack_type
);
1851 #ifdef CONFIG_SYN_COOKIES
1852 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1853 const struct sock
*sk
, struct sk_buff
*skb
,
1856 tcp_synq_overflow(sk
);
1857 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_SYNCOOKIESSENT
);
1858 return ops
->cookie_init_seq(skb
, mss
);
1861 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1862 const struct sock
*sk
, struct sk_buff
*skb
,
1869 int tcpv4_offload_init(void);
1871 void tcp_v4_init(void);
1872 void tcp_init(void);
1874 /* tcp_recovery.c */
1875 extern void tcp_rack_mark_lost(struct sock
*sk
);
1876 extern void tcp_rack_advance(struct tcp_sock
*tp
, u8 sacked
, u32 end_seq
,
1878 extern void tcp_rack_reo_timeout(struct sock
*sk
);
1879 extern void tcp_rack_update_reo_wnd(struct sock
*sk
, struct rate_sample
*rs
);
1881 /* At how many usecs into the future should the RTO fire? */
1882 static inline s64
tcp_rto_delta_us(const struct sock
*sk
)
1884 const struct sk_buff
*skb
= tcp_rtx_queue_head(sk
);
1885 u32 rto
= inet_csk(sk
)->icsk_rto
;
1886 u64 rto_time_stamp_us
= skb
->skb_mstamp
+ jiffies_to_usecs(rto
);
1888 return rto_time_stamp_us
- tcp_sk(sk
)->tcp_mstamp
;
1892 * Save and compile IPv4 options, return a pointer to it
1894 static inline struct ip_options_rcu
*tcp_v4_save_options(struct net
*net
,
1895 struct sk_buff
*skb
)
1897 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
1898 struct ip_options_rcu
*dopt
= NULL
;
1901 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
1903 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
1904 if (dopt
&& __ip_options_echo(net
, &dopt
->opt
, skb
, opt
)) {
1912 /* locally generated TCP pure ACKs have skb->truesize == 2
1913 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1914 * This is much faster than dissecting the packet to find out.
1915 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1917 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
1919 return skb
->truesize
== 2;
1922 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)
1927 static inline int tcp_inq(struct sock
*sk
)
1929 struct tcp_sock
*tp
= tcp_sk(sk
);
1932 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
1934 } else if (sock_flag(sk
, SOCK_URGINLINE
) ||
1936 before(tp
->urg_seq
, tp
->copied_seq
) ||
1937 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
1939 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
1941 /* Subtract 1, if FIN was received */
1942 if (answ
&& sock_flag(sk
, SOCK_DONE
))
1945 answ
= tp
->urg_seq
- tp
->copied_seq
;
1951 int tcp_peek_len(struct socket
*sock
);
1953 static inline void tcp_segs_in(struct tcp_sock
*tp
, const struct sk_buff
*skb
)
1957 segs_in
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
1958 tp
->segs_in
+= segs_in
;
1959 if (skb
->len
> tcp_hdrlen(skb
))
1960 tp
->data_segs_in
+= segs_in
;
1964 * TCP listen path runs lockless.
1965 * We forced "struct sock" to be const qualified to make sure
1966 * we don't modify one of its field by mistake.
1967 * Here, we increment sk_drops which is an atomic_t, so we can safely
1968 * make sock writable again.
1970 static inline void tcp_listendrop(const struct sock
*sk
)
1972 atomic_inc(&((struct sock
*)sk
)->sk_drops
);
1973 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1976 enum hrtimer_restart
tcp_pace_kick(struct hrtimer
*timer
);
1979 * Interface for adding Upper Level Protocols over TCP
1982 #define TCP_ULP_NAME_MAX 16
1983 #define TCP_ULP_MAX 128
1984 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
1991 struct tcp_ulp_ops
{
1992 struct list_head list
;
1994 /* initialize ulp */
1995 int (*init
)(struct sock
*sk
);
1997 void (*release
)(struct sock
*sk
);
2000 char name
[TCP_ULP_NAME_MAX
];
2002 struct module
*owner
;
2004 int tcp_register_ulp(struct tcp_ulp_ops
*type
);
2005 void tcp_unregister_ulp(struct tcp_ulp_ops
*type
);
2006 int tcp_set_ulp(struct sock
*sk
, const char *name
);
2007 int tcp_set_ulp_id(struct sock
*sk
, const int ulp
);
2008 void tcp_get_available_ulp(char *buf
, size_t len
);
2009 void tcp_cleanup_ulp(struct sock
*sk
);
2011 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2012 * is < 0, then the BPF op failed (for example if the loaded BPF
2013 * program does not support the chosen operation or there is no BPF
2017 static inline int tcp_call_bpf(struct sock
*sk
, int op
, u32 nargs
, u32
*args
)
2019 struct bpf_sock_ops_kern sock_ops
;
2022 memset(&sock_ops
, 0, offsetof(struct bpf_sock_ops_kern
, temp
));
2023 if (sk_fullsock(sk
)) {
2024 sock_ops
.is_fullsock
= 1;
2025 sock_owned_by_me(sk
);
2031 memcpy(sock_ops
.args
, args
, nargs
* sizeof(*args
));
2033 ret
= BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops
);
2035 ret
= sock_ops
.reply
;
2041 static inline int tcp_call_bpf_2arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
)
2043 u32 args
[2] = {arg1
, arg2
};
2045 return tcp_call_bpf(sk
, op
, 2, args
);
2048 static inline int tcp_call_bpf_3arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
,
2051 u32 args
[3] = {arg1
, arg2
, arg3
};
2053 return tcp_call_bpf(sk
, op
, 3, args
);
2057 static inline int tcp_call_bpf(struct sock
*sk
, int op
, u32 nargs
, u32
*args
)
2062 static inline int tcp_call_bpf_2arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
)
2067 static inline int tcp_call_bpf_3arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
,
2075 static inline u32
tcp_timeout_init(struct sock
*sk
)
2079 timeout
= tcp_call_bpf(sk
, BPF_SOCK_OPS_TIMEOUT_INIT
, 0, NULL
);
2082 timeout
= TCP_TIMEOUT_INIT
;
2086 static inline u32
tcp_rwnd_init_bpf(struct sock
*sk
)
2090 rwnd
= tcp_call_bpf(sk
, BPF_SOCK_OPS_RWND_INIT
, 0, NULL
);
2097 static inline bool tcp_bpf_ca_needs_ecn(struct sock
*sk
)
2099 return (tcp_call_bpf(sk
, BPF_SOCK_OPS_NEEDS_ECN
, 0, NULL
) == 1);
2102 #if IS_ENABLED(CONFIG_SMC)
2103 extern struct static_key_false tcp_have_smc
;