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
57 #define TCP_MIN_SND_MSS 48
58 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
61 * Never offer a window over 32767 without using window scaling. Some
62 * poor stacks do signed 16bit maths!
64 #define MAX_TCP_WINDOW 32767U
66 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
67 #define TCP_MIN_MSS 88U
69 /* The least MTU to use for probing */
70 #define TCP_BASE_MSS 1024
72 /* probing interval, default to 10 minutes as per RFC4821 */
73 #define TCP_PROBE_INTERVAL 600
75 /* Specify interval when tcp mtu probing will stop */
76 #define TCP_PROBE_THRESHOLD 8
78 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
79 #define TCP_FASTRETRANS_THRESH 3
81 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
82 #define TCP_MAX_QUICKACKS 16U
84 /* Maximal number of window scale according to RFC1323 */
85 #define TCP_MAX_WSCALE 14U
88 #define TCP_URG_VALID 0x0100
89 #define TCP_URG_NOTYET 0x0200
90 #define TCP_URG_READ 0x0400
92 #define TCP_RETR1 3 /*
93 * This is how many retries it does before it
94 * tries to figure out if the gateway is
95 * down. Minimal RFC value is 3; it corresponds
96 * to ~3sec-8min depending on RTO.
99 #define TCP_RETR2 15 /*
100 * This should take at least
101 * 90 minutes to time out.
102 * RFC1122 says that the limit is 100 sec.
103 * 15 is ~13-30min depending on RTO.
106 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
107 * when active opening a connection.
108 * RFC1122 says the minimum retry MUST
109 * be at least 180secs. Nevertheless
110 * this value is corresponding to
111 * 63secs of retransmission with the
112 * current initial RTO.
115 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
116 * when passive opening a connection.
117 * This is corresponding to 31secs of
118 * retransmission with the current
122 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
123 * state, about 60 seconds */
124 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
125 /* BSD style FIN_WAIT2 deadlock breaker.
126 * It used to be 3min, new value is 60sec,
127 * to combine FIN-WAIT-2 timeout with
131 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
133 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
134 #define TCP_ATO_MIN ((unsigned)(HZ/25))
136 #define TCP_DELACK_MIN 4U
137 #define TCP_ATO_MIN 4U
139 #define TCP_RTO_MAX ((unsigned)(120*HZ))
140 #define TCP_RTO_MIN ((unsigned)(HZ/5))
141 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
142 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
143 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
144 * used as a fallback RTO for the
145 * initial data transmission if no
146 * valid RTT sample has been acquired,
147 * most likely due to retrans in 3WHS.
150 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
151 * for local resources.
153 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
154 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
155 #define TCP_KEEPALIVE_INTVL (75*HZ)
157 #define MAX_TCP_KEEPIDLE 32767
158 #define MAX_TCP_KEEPINTVL 32767
159 #define MAX_TCP_KEEPCNT 127
160 #define MAX_TCP_SYNCNT 127
162 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
164 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
165 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
166 * after this time. It should be equal
167 * (or greater than) TCP_TIMEWAIT_LEN
168 * to provide reliability equal to one
169 * provided by timewait state.
171 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
172 * timestamps. It must be less than
173 * minimal timewait lifetime.
179 #define TCPOPT_NOP 1 /* Padding */
180 #define TCPOPT_EOL 0 /* End of options */
181 #define TCPOPT_MSS 2 /* Segment size negotiating */
182 #define TCPOPT_WINDOW 3 /* Window scaling */
183 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
184 #define TCPOPT_SACK 5 /* SACK Block */
185 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
186 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
187 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
188 #define TCPOPT_EXP 254 /* Experimental */
189 /* Magic number to be after the option value for sharing TCP
190 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
192 #define TCPOPT_FASTOPEN_MAGIC 0xF989
193 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
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
206 #define TCPOLEN_EXP_SMC_BASE 6
208 /* But this is what stacks really send out. */
209 #define TCPOLEN_TSTAMP_ALIGNED 12
210 #define TCPOLEN_WSCALE_ALIGNED 4
211 #define TCPOLEN_SACKPERM_ALIGNED 4
212 #define TCPOLEN_SACK_BASE 2
213 #define TCPOLEN_SACK_BASE_ALIGNED 4
214 #define TCPOLEN_SACK_PERBLOCK 8
215 #define TCPOLEN_MD5SIG_ALIGNED 20
216 #define TCPOLEN_MSS_ALIGNED 4
217 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
219 /* Flags in tp->nonagle */
220 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
221 #define TCP_NAGLE_CORK 2 /* Socket is corked */
222 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
224 /* TCP thin-stream limits */
225 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
227 /* TCP initial congestion window as per rfc6928 */
228 #define TCP_INIT_CWND 10
230 /* Bit Flags for sysctl_tcp_fastopen */
231 #define TFO_CLIENT_ENABLE 1
232 #define TFO_SERVER_ENABLE 2
233 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
235 /* Accept SYN data w/o any cookie option */
236 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
238 /* Force enable TFO on all listeners, i.e., not requiring the
239 * TCP_FASTOPEN socket option.
241 #define TFO_SERVER_WO_SOCKOPT1 0x400
244 /* sysctl variables for tcp */
245 extern int sysctl_tcp_max_orphans
;
246 extern long sysctl_tcp_mem
[3];
248 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
249 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
251 extern atomic_long_t tcp_memory_allocated
;
252 extern struct percpu_counter tcp_sockets_allocated
;
253 extern unsigned long tcp_memory_pressure
;
255 /* optimized version of sk_under_memory_pressure() for TCP sockets */
256 static inline bool tcp_under_memory_pressure(const struct sock
*sk
)
258 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
259 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
262 return tcp_memory_pressure
;
265 * The next routines deal with comparing 32 bit unsigned ints
266 * and worry about wraparound (automatic with unsigned arithmetic).
269 static inline bool before(__u32 seq1
, __u32 seq2
)
271 return (__s32
)(seq1
-seq2
) < 0;
273 #define after(seq2, seq1) before(seq1, seq2)
275 /* is s2<=s1<=s3 ? */
276 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
278 return seq3
- seq2
>= seq1
- seq2
;
281 static inline bool tcp_out_of_memory(struct sock
*sk
)
283 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
284 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
289 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
291 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
293 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
294 int orphans
= percpu_counter_read_positive(ocp
);
296 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
297 orphans
= percpu_counter_sum_positive(ocp
);
298 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
304 bool tcp_check_oom(struct sock
*sk
, int shift
);
307 extern struct proto tcp_prot
;
309 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
311 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
312 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
314 void tcp_tasklet_init(void);
316 void tcp_v4_err(struct sk_buff
*skb
, u32
);
318 void tcp_shutdown(struct sock
*sk
, int how
);
320 int tcp_v4_early_demux(struct sk_buff
*skb
);
321 int tcp_v4_rcv(struct sk_buff
*skb
);
323 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
324 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
325 int tcp_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
326 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
328 int tcp_sendpage_locked(struct sock
*sk
, struct page
*page
, int offset
,
329 size_t size
, int flags
);
330 ssize_t
do_tcp_sendpages(struct sock
*sk
, struct page
*page
, int offset
,
331 size_t size
, int flags
);
332 void tcp_release_cb(struct sock
*sk
);
333 void tcp_wfree(struct sk_buff
*skb
);
334 void tcp_write_timer_handler(struct sock
*sk
);
335 void tcp_delack_timer_handler(struct sock
*sk
);
336 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
337 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
);
338 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
339 const struct tcphdr
*th
);
340 void tcp_rcv_space_adjust(struct sock
*sk
);
341 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
342 void tcp_twsk_destructor(struct sock
*sk
);
343 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
344 struct pipe_inode_info
*pipe
, size_t len
,
347 void tcp_enter_quickack_mode(struct sock
*sk
, unsigned int max_quickacks
);
348 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
349 const unsigned int pkts
)
351 struct inet_connection_sock
*icsk
= inet_csk(sk
);
353 if (icsk
->icsk_ack
.quick
) {
354 if (pkts
>= icsk
->icsk_ack
.quick
) {
355 icsk
->icsk_ack
.quick
= 0;
356 /* Leaving quickack mode we deflate ATO. */
357 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
359 icsk
->icsk_ack
.quick
-= pkts
;
364 #define TCP_ECN_QUEUE_CWR 2
365 #define TCP_ECN_DEMAND_CWR 4
366 #define TCP_ECN_SEEN 8
376 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
378 const struct tcphdr
*th
);
379 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
380 struct request_sock
*req
, bool fastopen
);
381 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
382 struct sk_buff
*skb
);
383 void tcp_enter_loss(struct sock
*sk
);
384 void tcp_cwnd_reduction(struct sock
*sk
, int newly_acked_sacked
, int flag
);
385 void tcp_clear_retrans(struct tcp_sock
*tp
);
386 void tcp_update_metrics(struct sock
*sk
);
387 void tcp_init_metrics(struct sock
*sk
);
388 void tcp_metrics_init(void);
389 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
);
390 void tcp_close(struct sock
*sk
, long timeout
);
391 void tcp_init_sock(struct sock
*sk
);
392 void tcp_init_transfer(struct sock
*sk
, int bpf_op
);
393 unsigned int tcp_poll(struct file
*file
, struct socket
*sock
,
394 struct poll_table_struct
*wait
);
395 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
396 char __user
*optval
, int __user
*optlen
);
397 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
398 char __user
*optval
, unsigned int optlen
);
399 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
400 char __user
*optval
, int __user
*optlen
);
401 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
402 char __user
*optval
, unsigned int optlen
);
403 void tcp_set_keepalive(struct sock
*sk
, int val
);
404 void tcp_syn_ack_timeout(const struct request_sock
*req
);
405 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
406 int flags
, int *addr_len
);
407 void tcp_parse_options(const struct net
*net
, const struct sk_buff
*skb
,
408 struct tcp_options_received
*opt_rx
,
409 int estab
, struct tcp_fastopen_cookie
*foc
);
410 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
413 * TCP v4 functions exported for the inet6 API
416 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
417 void tcp_v4_mtu_reduced(struct sock
*sk
);
418 void tcp_req_err(struct sock
*sk
, u32 seq
, bool abort
);
419 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
420 struct sock
*tcp_create_openreq_child(const struct sock
*sk
,
421 struct request_sock
*req
,
422 struct sk_buff
*skb
);
423 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
424 struct sock
*tcp_v4_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
425 struct request_sock
*req
,
426 struct dst_entry
*dst
,
427 struct request_sock
*req_unhash
,
429 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
430 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
431 int tcp_connect(struct sock
*sk
);
432 enum tcp_synack_type
{
437 struct sk_buff
*tcp_make_synack(const struct sock
*sk
, struct dst_entry
*dst
,
438 struct request_sock
*req
,
439 struct tcp_fastopen_cookie
*foc
,
440 enum tcp_synack_type synack_type
);
441 int tcp_disconnect(struct sock
*sk
, int flags
);
443 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
444 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
445 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
447 /* From syncookies.c */
448 struct sock
*tcp_get_cookie_sock(struct sock
*sk
, struct sk_buff
*skb
,
449 struct request_sock
*req
,
450 struct dst_entry
*dst
, u32 tsoff
);
451 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
453 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
454 #ifdef CONFIG_SYN_COOKIES
456 /* Syncookies use a monotonic timer which increments every 60 seconds.
457 * This counter is used both as a hash input and partially encoded into
458 * the cookie value. A cookie is only validated further if the delta
459 * between the current counter value and the encoded one is less than this,
460 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
461 * the counter advances immediately after a cookie is generated).
463 #define MAX_SYNCOOKIE_AGE 2
464 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
465 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
467 /* syncookies: remember time of last synqueue overflow
468 * But do not dirty this field too often (once per second is enough)
469 * It is racy as we do not hold a lock, but race is very minor.
471 static inline void tcp_synq_overflow(const struct sock
*sk
)
473 unsigned long last_overflow
= READ_ONCE(tcp_sk(sk
)->rx_opt
.ts_recent_stamp
);
474 unsigned long now
= jiffies
;
476 if (!time_between32(now
, last_overflow
, last_overflow
+ HZ
))
477 WRITE_ONCE(tcp_sk(sk
)->rx_opt
.ts_recent_stamp
, now
);
480 /* syncookies: no recent synqueue overflow on this listening socket? */
481 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
483 unsigned long last_overflow
= READ_ONCE(tcp_sk(sk
)->rx_opt
.ts_recent_stamp
);
485 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
486 * then we're under synflood. However, we have to use
487 * 'last_overflow - HZ' as lower bound. That's because a concurrent
488 * tcp_synq_overflow() could update .ts_recent_stamp after we read
489 * jiffies but before we store .ts_recent_stamp into last_overflow,
490 * which could lead to rejecting a valid syncookie.
492 return !time_between32(jiffies
, last_overflow
- HZ
,
493 last_overflow
+ TCP_SYNCOOKIE_VALID
);
496 static inline u32
tcp_cookie_time(void)
498 u64 val
= get_jiffies_64();
500 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
504 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
506 __u32
cookie_v4_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
507 u64
cookie_init_timestamp(struct request_sock
*req
);
508 bool cookie_timestamp_decode(const struct net
*net
,
509 struct tcp_options_received
*opt
);
510 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
511 const struct net
*net
, const struct dst_entry
*dst
);
513 /* From net/ipv6/syncookies.c */
514 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
516 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
518 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
519 const struct tcphdr
*th
, u16
*mssp
);
520 __u32
cookie_v6_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
524 u32
tcp_tso_autosize(const struct sock
*sk
, unsigned int mss_now
,
526 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
528 int __tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
529 int tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
530 void tcp_retransmit_timer(struct sock
*sk
);
531 void tcp_xmit_retransmit_queue(struct sock
*);
532 void tcp_simple_retransmit(struct sock
*);
533 void tcp_enter_recovery(struct sock
*sk
, bool ece_ack
);
534 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
536 TCP_FRAG_IN_WRITE_QUEUE
,
537 TCP_FRAG_IN_RTX_QUEUE
,
539 int tcp_fragment(struct sock
*sk
, enum tcp_queue tcp_queue
,
540 struct sk_buff
*skb
, u32 len
,
541 unsigned int mss_now
, gfp_t gfp
);
543 void tcp_send_probe0(struct sock
*);
544 void tcp_send_partial(struct sock
*);
545 int tcp_write_wakeup(struct sock
*, int mib
);
546 void tcp_send_fin(struct sock
*sk
);
547 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
548 int tcp_send_synack(struct sock
*);
549 void tcp_push_one(struct sock
*, unsigned int mss_now
);
550 void __tcp_send_ack(struct sock
*sk
, u32 rcv_nxt
);
551 void tcp_send_ack(struct sock
*sk
);
552 void tcp_send_delayed_ack(struct sock
*sk
);
553 void tcp_send_loss_probe(struct sock
*sk
);
554 bool tcp_schedule_loss_probe(struct sock
*sk
, bool advancing_rto
);
555 void tcp_skb_collapse_tstamp(struct sk_buff
*skb
,
556 const struct sk_buff
*next_skb
);
559 void tcp_rearm_rto(struct sock
*sk
);
560 void tcp_synack_rtt_meas(struct sock
*sk
, struct request_sock
*req
);
561 void tcp_reset(struct sock
*sk
);
562 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock
*tp
, struct sk_buff
*skb
);
563 void tcp_fin(struct sock
*sk
);
566 void tcp_init_xmit_timers(struct sock
*);
567 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
569 hrtimer_cancel(&tcp_sk(sk
)->pacing_timer
);
570 inet_csk_clear_xmit_timers(sk
);
573 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
574 unsigned int tcp_current_mss(struct sock
*sk
);
576 /* Bound MSS / TSO packet size with the half of the window */
577 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
581 /* When peer uses tiny windows, there is no use in packetizing
582 * to sub-MSS pieces for the sake of SWS or making sure there
583 * are enough packets in the pipe for fast recovery.
585 * On the other hand, for extremely large MSS devices, handling
586 * smaller than MSS windows in this way does make sense.
588 if (tp
->max_window
> TCP_MSS_DEFAULT
)
589 cutoff
= (tp
->max_window
>> 1);
591 cutoff
= tp
->max_window
;
593 if (cutoff
&& pktsize
> cutoff
)
594 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
600 void tcp_get_info(struct sock
*, struct tcp_info
*);
602 /* Read 'sendfile()'-style from a TCP socket */
603 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
604 sk_read_actor_t recv_actor
);
606 void tcp_initialize_rcv_mss(struct sock
*sk
);
608 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
609 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
610 void tcp_mtup_init(struct sock
*sk
);
611 void tcp_init_buffer_space(struct sock
*sk
);
613 static inline void tcp_bound_rto(const struct sock
*sk
)
615 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
616 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
619 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
621 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
624 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
626 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
627 ntohl(TCP_FLAG_ACK
) |
631 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
633 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
636 static inline void tcp_fast_path_check(struct sock
*sk
)
638 struct tcp_sock
*tp
= tcp_sk(sk
);
640 if (RB_EMPTY_ROOT(&tp
->out_of_order_queue
) &&
642 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
644 tcp_fast_path_on(tp
);
647 /* Compute the actual rto_min value */
648 static inline u32
tcp_rto_min(struct sock
*sk
)
650 const struct dst_entry
*dst
= __sk_dst_get(sk
);
651 u32 rto_min
= TCP_RTO_MIN
;
653 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
654 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
658 static inline u32
tcp_rto_min_us(struct sock
*sk
)
660 return jiffies_to_usecs(tcp_rto_min(sk
));
663 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
665 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
668 /* Minimum RTT in usec. ~0 means not available. */
669 static inline u32
tcp_min_rtt(const struct tcp_sock
*tp
)
671 return minmax_get(&tp
->rtt_min
);
674 /* Compute the actual receive window we are currently advertising.
675 * Rcv_nxt can be after the window if our peer push more data
676 * than the offered window.
678 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
680 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
687 /* Choose a new window, without checks for shrinking, and without
688 * scaling applied to the result. The caller does these things
689 * if necessary. This is a "raw" window selection.
691 u32
__tcp_select_window(struct sock
*sk
);
693 void tcp_send_window_probe(struct sock
*sk
);
695 /* TCP uses 32bit jiffies to save some space.
696 * Note that this is different from tcp_time_stamp, which
697 * historically has been the same until linux-4.13.
699 #define tcp_jiffies32 ((u32)jiffies)
702 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
703 * It is no longer tied to jiffies, but to 1 ms clock.
704 * Note: double check if you want to use tcp_jiffies32 instead of this.
706 #define TCP_TS_HZ 1000
708 static inline u64
tcp_clock_ns(void)
710 return local_clock();
713 static inline u64
tcp_clock_us(void)
715 return div_u64(tcp_clock_ns(), NSEC_PER_USEC
);
718 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
719 static inline u32
tcp_time_stamp(const struct tcp_sock
*tp
)
721 return div_u64(tp
->tcp_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
724 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
725 static inline u32
tcp_time_stamp_raw(void)
727 return div_u64(tcp_clock_ns(), NSEC_PER_SEC
/ TCP_TS_HZ
);
731 /* Refresh 1us clock of a TCP socket,
732 * ensuring monotically increasing values.
734 static inline void tcp_mstamp_refresh(struct tcp_sock
*tp
)
736 u64 val
= tcp_clock_us();
738 if (val
> tp
->tcp_mstamp
)
739 tp
->tcp_mstamp
= val
;
742 static inline u32
tcp_stamp_us_delta(u64 t1
, u64 t0
)
744 return max_t(s64
, t1
- t0
, 0);
747 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
749 return div_u64(skb
->skb_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
753 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
755 #define TCPHDR_FIN 0x01
756 #define TCPHDR_SYN 0x02
757 #define TCPHDR_RST 0x04
758 #define TCPHDR_PSH 0x08
759 #define TCPHDR_ACK 0x10
760 #define TCPHDR_URG 0x20
761 #define TCPHDR_ECE 0x40
762 #define TCPHDR_CWR 0x80
764 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
766 /* This is what the send packet queuing engine uses to pass
767 * TCP per-packet control information to the transmission code.
768 * We also store the host-order sequence numbers in here too.
769 * This is 44 bytes if IPV6 is enabled.
770 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
773 __u32 seq
; /* Starting sequence number */
774 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
776 /* Note : tcp_tw_isn is used in input path only
777 * (isn chosen by tcp_timewait_state_process())
779 * tcp_gso_segs/size are used in write queue only,
780 * cf tcp_skb_pcount()/tcp_skb_mss()
788 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
790 __u8 sacked
; /* State flags for SACK. */
791 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
792 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
793 #define TCPCB_LOST 0x04 /* SKB is lost */
794 #define TCPCB_TAGBITS 0x07 /* All tag bits */
795 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
796 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
797 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
800 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
801 __u8 txstamp_ack
:1, /* Record TX timestamp for ack? */
802 eor
:1, /* Is skb MSG_EOR marked? */
803 has_rxtstamp
:1, /* SKB has a RX timestamp */
805 __u32 ack_seq
; /* Sequence number ACK'd */
808 /* There is space for up to 24 bytes */
809 __u32 in_flight
:30,/* Bytes in flight at transmit */
810 is_app_limited
:1, /* cwnd not fully used? */
812 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
814 /* start of send pipeline phase */
816 /* when we reached the "delivered" count */
817 u64 delivered_mstamp
;
818 } tx
; /* only used for outgoing skbs */
820 struct inet_skb_parm h4
;
821 #if IS_ENABLED(CONFIG_IPV6)
822 struct inet6_skb_parm h6
;
824 } header
; /* For incoming skbs */
834 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
837 #if IS_ENABLED(CONFIG_IPV6)
838 /* This is the variant of inet6_iif() that must be used by TCP,
839 * as TCP moves IP6CB into a different location in skb->cb[]
841 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
843 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
846 static inline int tcp_v6_iif_l3_slave(const struct sk_buff
*skb
)
848 bool l3_slave
= ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
);
850 return l3_slave
? skb
->skb_iif
: TCP_SKB_CB(skb
)->header
.h6
.iif
;
853 /* TCP_SKB_CB reference means this can not be used from early demux */
854 static inline int tcp_v6_sdif(const struct sk_buff
*skb
)
856 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
857 if (skb
&& ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
))
858 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
864 static inline bool inet_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
866 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
867 if (!net
->ipv4
.sysctl_tcp_l3mdev_accept
&&
868 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
874 /* TCP_SKB_CB reference means this can not be used from early demux */
875 static inline int tcp_v4_sdif(struct sk_buff
*skb
)
877 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
878 if (skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
879 return TCP_SKB_CB(skb
)->header
.h4
.iif
;
884 /* Due to TSO, an SKB can be composed of multiple actual
885 * packets. To keep these tracked properly, we use this.
887 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
889 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
892 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
894 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
897 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
899 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
902 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
903 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
905 return TCP_SKB_CB(skb
)->tcp_gso_size
;
908 static inline bool tcp_skb_can_collapse_to(const struct sk_buff
*skb
)
910 return likely(!TCP_SKB_CB(skb
)->eor
);
913 /* Events passed to congestion control interface */
915 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
916 CA_EVENT_CWND_RESTART
, /* congestion window restart */
917 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
918 CA_EVENT_LOSS
, /* loss timeout */
919 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
920 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
923 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
924 enum tcp_ca_ack_event_flags
{
925 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
926 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
927 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
931 * Interface for adding new TCP congestion control handlers
933 #define TCP_CA_NAME_MAX 16
934 #define TCP_CA_MAX 128
935 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
937 #define TCP_CA_UNSPEC 0
939 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
940 #define TCP_CONG_NON_RESTRICTED 0x1
941 /* Requires ECN/ECT set on all packets */
942 #define TCP_CONG_NEEDS_ECN 0x2
952 /* A rate sample measures the number of (original/retransmitted) data
953 * packets delivered "delivered" over an interval of time "interval_us".
954 * The tcp_rate.c code fills in the rate sample, and congestion
955 * control modules that define a cong_control function to run at the end
956 * of ACK processing can optionally chose to consult this sample when
957 * setting cwnd and pacing rate.
958 * A sample is invalid if "delivered" or "interval_us" is negative.
961 u64 prior_mstamp
; /* starting timestamp for interval */
962 u32 prior_delivered
; /* tp->delivered at "prior_mstamp" */
963 s32 delivered
; /* number of packets delivered over interval */
964 long interval_us
; /* time for tp->delivered to incr "delivered" */
965 long rtt_us
; /* RTT of last (S)ACKed packet (or -1) */
966 int losses
; /* number of packets marked lost upon ACK */
967 u32 acked_sacked
; /* number of packets newly (S)ACKed upon ACK */
968 u32 prior_in_flight
; /* in flight before this ACK */
969 bool is_app_limited
; /* is sample from packet with bubble in pipe? */
970 bool is_retrans
; /* is sample from retransmission? */
973 struct tcp_congestion_ops
{
974 struct list_head list
;
978 /* initialize private data (optional) */
979 void (*init
)(struct sock
*sk
);
980 /* cleanup private data (optional) */
981 void (*release
)(struct sock
*sk
);
983 /* return slow start threshold (required) */
984 u32 (*ssthresh
)(struct sock
*sk
);
985 /* do new cwnd calculation (required) */
986 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
987 /* call before changing ca_state (optional) */
988 void (*set_state
)(struct sock
*sk
, u8 new_state
);
989 /* call when cwnd event occurs (optional) */
990 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
991 /* call when ack arrives (optional) */
992 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
993 /* new value of cwnd after loss (required) */
994 u32 (*undo_cwnd
)(struct sock
*sk
);
995 /* hook for packet ack accounting (optional) */
996 void (*pkts_acked
)(struct sock
*sk
, const struct ack_sample
*sample
);
997 /* suggest number of segments for each skb to transmit (optional) */
998 u32 (*tso_segs_goal
)(struct sock
*sk
);
999 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1000 u32 (*sndbuf_expand
)(struct sock
*sk
);
1001 /* call when packets are delivered to update cwnd and pacing rate,
1002 * after all the ca_state processing. (optional)
1004 void (*cong_control
)(struct sock
*sk
, const struct rate_sample
*rs
);
1005 /* get info for inet_diag (optional) */
1006 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
1007 union tcp_cc_info
*info
);
1009 char name
[TCP_CA_NAME_MAX
];
1010 struct module
*owner
;
1013 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1014 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1016 void tcp_assign_congestion_control(struct sock
*sk
);
1017 void tcp_init_congestion_control(struct sock
*sk
);
1018 void tcp_cleanup_congestion_control(struct sock
*sk
);
1019 int tcp_set_default_congestion_control(struct net
*net
, const char *name
);
1020 void tcp_get_default_congestion_control(struct net
*net
, char *name
);
1021 void tcp_get_available_congestion_control(char *buf
, size_t len
);
1022 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
1023 int tcp_set_allowed_congestion_control(char *allowed
);
1024 int tcp_set_congestion_control(struct sock
*sk
, const char *name
, bool load
,
1025 bool reinit
, bool cap_net_admin
);
1026 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
1027 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
1029 u32
tcp_reno_ssthresh(struct sock
*sk
);
1030 u32
tcp_reno_undo_cwnd(struct sock
*sk
);
1031 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
1032 extern struct tcp_congestion_ops tcp_reno
;
1034 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
1035 u32
tcp_ca_get_key_by_name(struct net
*net
, const char *name
, bool *ecn_ca
);
1037 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
1039 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
1045 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
1047 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1049 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
1052 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
1054 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1056 if (icsk
->icsk_ca_ops
->set_state
)
1057 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
1058 icsk
->icsk_ca_state
= ca_state
;
1061 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
1063 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1065 if (icsk
->icsk_ca_ops
->cwnd_event
)
1066 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
1069 /* From tcp_rate.c */
1070 void tcp_rate_skb_sent(struct sock
*sk
, struct sk_buff
*skb
);
1071 void tcp_rate_skb_delivered(struct sock
*sk
, struct sk_buff
*skb
,
1072 struct rate_sample
*rs
);
1073 void tcp_rate_gen(struct sock
*sk
, u32 delivered
, u32 lost
,
1074 bool is_sack_reneg
, struct rate_sample
*rs
);
1075 void tcp_rate_check_app_limited(struct sock
*sk
);
1077 /* These functions determine how the current flow behaves in respect of SACK
1078 * handling. SACK is negotiated with the peer, and therefore it can vary
1079 * between different flows.
1081 * tcp_is_sack - SACK enabled
1082 * tcp_is_reno - No SACK
1084 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
1086 return tp
->rx_opt
.sack_ok
;
1089 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
1091 return !tcp_is_sack(tp
);
1094 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
1096 return tp
->sacked_out
+ tp
->lost_out
;
1099 /* This determines how many packets are "in the network" to the best
1100 * of our knowledge. In many cases it is conservative, but where
1101 * detailed information is available from the receiver (via SACK
1102 * blocks etc.) we can make more aggressive calculations.
1104 * Use this for decisions involving congestion control, use just
1105 * tp->packets_out to determine if the send queue is empty or not.
1107 * Read this equation as:
1109 * "Packets sent once on transmission queue" MINUS
1110 * "Packets left network, but not honestly ACKed yet" PLUS
1111 * "Packets fast retransmitted"
1113 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1115 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
1118 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1120 static inline bool tcp_in_slow_start(const struct tcp_sock
*tp
)
1122 return tp
->snd_cwnd
< tp
->snd_ssthresh
;
1125 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
1127 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
1130 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
1132 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
1133 (1 << inet_csk(sk
)->icsk_ca_state
);
1136 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1137 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1140 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
1142 const struct tcp_sock
*tp
= tcp_sk(sk
);
1144 if (tcp_in_cwnd_reduction(sk
))
1145 return tp
->snd_ssthresh
;
1147 return max(tp
->snd_ssthresh
,
1148 ((tp
->snd_cwnd
>> 1) +
1149 (tp
->snd_cwnd
>> 2)));
1152 /* Use define here intentionally to get WARN_ON location shown at the caller */
1153 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1155 void tcp_enter_cwr(struct sock
*sk
);
1156 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1158 /* The maximum number of MSS of available cwnd for which TSO defers
1159 * sending if not using sysctl_tcp_tso_win_divisor.
1161 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1166 /* Returns end sequence number of the receiver's advertised window */
1167 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1169 return tp
->snd_una
+ tp
->snd_wnd
;
1172 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1173 * flexible approach. The RFC suggests cwnd should not be raised unless
1174 * it was fully used previously. And that's exactly what we do in
1175 * congestion avoidance mode. But in slow start we allow cwnd to grow
1176 * as long as the application has used half the cwnd.
1178 * cwnd is 10 (IW10), but application sends 9 frames.
1179 * We allow cwnd to reach 18 when all frames are ACKed.
1180 * This check is safe because it's as aggressive as slow start which already
1181 * risks 100% overshoot. The advantage is that we discourage application to
1182 * either send more filler packets or data to artificially blow up the cwnd
1183 * usage, and allow application-limited process to probe bw more aggressively.
1185 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1187 const struct tcp_sock
*tp
= tcp_sk(sk
);
1189 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1190 if (tcp_in_slow_start(tp
))
1191 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1193 return tp
->is_cwnd_limited
;
1196 /* Something is really bad, we could not queue an additional packet,
1197 * because qdisc is full or receiver sent a 0 window.
1198 * We do not want to add fuel to the fire, or abort too early,
1199 * so make sure the timer we arm now is at least 200ms in the future,
1200 * regardless of current icsk_rto value (as it could be ~2ms)
1202 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1204 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1207 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1208 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1209 unsigned long max_when
)
1211 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1213 return (unsigned long)min_t(u64
, when
, max_when
);
1216 static inline void tcp_check_probe_timer(struct sock
*sk
)
1218 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1219 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1220 tcp_probe0_base(sk
), TCP_RTO_MAX
);
1223 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1228 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1234 * Calculate(/check) TCP checksum
1236 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1237 __be32 daddr
, __wsum base
)
1239 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1242 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
1244 return __skb_checksum_complete(skb
);
1247 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1249 return !skb_csum_unnecessary(skb
) &&
1250 __tcp_checksum_complete(skb
);
1253 bool tcp_add_backlog(struct sock
*sk
, struct sk_buff
*skb
);
1254 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
);
1259 static const char *statename
[]={
1260 "Unused","Established","Syn Sent","Syn Recv",
1261 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1262 "Close Wait","Last ACK","Listen","Closing"
1265 void tcp_set_state(struct sock
*sk
, int state
);
1267 void tcp_done(struct sock
*sk
);
1269 int tcp_abort(struct sock
*sk
, int err
);
1271 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1274 rx_opt
->num_sacks
= 0;
1277 u32
tcp_default_init_rwnd(u32 mss
);
1278 void tcp_cwnd_restart(struct sock
*sk
, s32 delta
);
1280 static inline void tcp_slow_start_after_idle_check(struct sock
*sk
)
1282 const struct tcp_congestion_ops
*ca_ops
= inet_csk(sk
)->icsk_ca_ops
;
1283 struct tcp_sock
*tp
= tcp_sk(sk
);
1286 if (!sock_net(sk
)->ipv4
.sysctl_tcp_slow_start_after_idle
|| tp
->packets_out
||
1287 ca_ops
->cong_control
)
1289 delta
= tcp_jiffies32
- tp
->lsndtime
;
1290 if (delta
> inet_csk(sk
)->icsk_rto
)
1291 tcp_cwnd_restart(sk
, delta
);
1294 /* Determine a window scaling and initial window to offer. */
1295 void tcp_select_initial_window(const struct sock
*sk
, int __space
,
1296 __u32 mss
, __u32
*rcv_wnd
,
1297 __u32
*window_clamp
, int wscale_ok
,
1298 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1300 static inline int tcp_win_from_space(const struct sock
*sk
, int space
)
1302 int tcp_adv_win_scale
= sock_net(sk
)->ipv4
.sysctl_tcp_adv_win_scale
;
1304 return tcp_adv_win_scale
<= 0 ?
1305 (space
>>(-tcp_adv_win_scale
)) :
1306 space
- (space
>>tcp_adv_win_scale
);
1309 /* Note: caller must be prepared to deal with negative returns */
1310 static inline int tcp_space(const struct sock
*sk
)
1312 return tcp_win_from_space(sk
, sk
->sk_rcvbuf
- sk
->sk_backlog
.len
-
1313 atomic_read(&sk
->sk_rmem_alloc
));
1316 static inline int tcp_full_space(const struct sock
*sk
)
1318 return tcp_win_from_space(sk
, sk
->sk_rcvbuf
);
1321 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1322 const struct sock
*sk_listener
,
1323 const struct dst_entry
*dst
);
1325 void tcp_enter_memory_pressure(struct sock
*sk
);
1326 void tcp_leave_memory_pressure(struct sock
*sk
);
1328 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1330 struct net
*net
= sock_net((struct sock
*)tp
);
1332 return tp
->keepalive_intvl
? : net
->ipv4
.sysctl_tcp_keepalive_intvl
;
1335 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1337 struct net
*net
= sock_net((struct sock
*)tp
);
1339 return tp
->keepalive_time
? : net
->ipv4
.sysctl_tcp_keepalive_time
;
1342 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1344 struct net
*net
= sock_net((struct sock
*)tp
);
1346 return tp
->keepalive_probes
? : net
->ipv4
.sysctl_tcp_keepalive_probes
;
1349 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1351 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1353 return min_t(u32
, tcp_jiffies32
- icsk
->icsk_ack
.lrcvtime
,
1354 tcp_jiffies32
- tp
->rcv_tstamp
);
1357 static inline int tcp_fin_time(const struct sock
*sk
)
1359 int fin_timeout
= tcp_sk(sk
)->linger2
? : sock_net(sk
)->ipv4
.sysctl_tcp_fin_timeout
;
1360 const int rto
= inet_csk(sk
)->icsk_rto
;
1362 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1363 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1368 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1371 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1373 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1376 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1377 * then following tcp messages have valid values. Ignore 0 value,
1378 * or else 'negative' tsval might forbid us to accept their packets.
1380 if (!rx_opt
->ts_recent
)
1385 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1388 if (tcp_paws_check(rx_opt
, 0))
1391 /* RST segments are not recommended to carry timestamp,
1392 and, if they do, it is recommended to ignore PAWS because
1393 "their cleanup function should take precedence over timestamps."
1394 Certainly, it is mistake. It is necessary to understand the reasons
1395 of this constraint to relax it: if peer reboots, clock may go
1396 out-of-sync and half-open connections will not be reset.
1397 Actually, the problem would be not existing if all
1398 the implementations followed draft about maintaining clock
1399 via reboots. Linux-2.2 DOES NOT!
1401 However, we can relax time bounds for RST segments to MSL.
1403 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1408 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1409 int mib_idx
, u32
*last_oow_ack_time
);
1411 static inline void tcp_mib_init(struct net
*net
)
1414 TCP_ADD_STATS(net
, TCP_MIB_RTOALGORITHM
, 1);
1415 TCP_ADD_STATS(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1416 TCP_ADD_STATS(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1417 TCP_ADD_STATS(net
, TCP_MIB_MAXCONN
, -1);
1421 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1423 tp
->lost_skb_hint
= NULL
;
1426 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1428 tcp_clear_retrans_hints_partial(tp
);
1429 tp
->retransmit_skb_hint
= NULL
;
1432 union tcp_md5_addr
{
1434 #if IS_ENABLED(CONFIG_IPV6)
1439 /* - key database */
1440 struct tcp_md5sig_key
{
1441 struct hlist_node node
;
1443 u8 family
; /* AF_INET or AF_INET6 */
1444 union tcp_md5_addr addr
;
1446 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1447 struct rcu_head rcu
;
1451 struct tcp_md5sig_info
{
1452 struct hlist_head head
;
1453 struct rcu_head rcu
;
1456 /* - pseudo header */
1457 struct tcp4_pseudohdr
{
1465 struct tcp6_pseudohdr
{
1466 struct in6_addr saddr
;
1467 struct in6_addr daddr
;
1469 __be32 protocol
; /* including padding */
1472 union tcp_md5sum_block
{
1473 struct tcp4_pseudohdr ip4
;
1474 #if IS_ENABLED(CONFIG_IPV6)
1475 struct tcp6_pseudohdr ip6
;
1479 /* - pool: digest algorithm, hash description and scratch buffer */
1480 struct tcp_md5sig_pool
{
1481 struct ahash_request
*md5_req
;
1486 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1487 const struct sock
*sk
, const struct sk_buff
*skb
);
1488 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1489 int family
, u8 prefixlen
, const u8
*newkey
, u8 newkeylen
,
1491 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1492 int family
, u8 prefixlen
);
1493 struct tcp_md5sig_key
*tcp_v4_md5_lookup(const struct sock
*sk
,
1494 const struct sock
*addr_sk
);
1496 #ifdef CONFIG_TCP_MD5SIG
1497 struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1498 const union tcp_md5_addr
*addr
,
1500 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1502 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1503 const union tcp_md5_addr
*addr
,
1508 #define tcp_twsk_md5_key(twsk) NULL
1511 bool tcp_alloc_md5sig_pool(void);
1513 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1514 static inline void tcp_put_md5sig_pool(void)
1519 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1520 unsigned int header_len
);
1521 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1522 const struct tcp_md5sig_key
*key
);
1524 /* From tcp_fastopen.c */
1525 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1526 struct tcp_fastopen_cookie
*cookie
, int *syn_loss
,
1527 unsigned long *last_syn_loss
);
1528 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1529 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1531 struct tcp_fastopen_request
{
1532 /* Fast Open cookie. Size 0 means a cookie request */
1533 struct tcp_fastopen_cookie cookie
;
1534 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1536 int copied
; /* queued in tcp_connect() */
1538 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1539 void tcp_fastopen_destroy_cipher(struct sock
*sk
);
1540 void tcp_fastopen_ctx_destroy(struct net
*net
);
1541 int tcp_fastopen_reset_cipher(struct net
*net
, struct sock
*sk
,
1542 void *key
, unsigned int len
);
1543 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
);
1544 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1545 struct request_sock
*req
,
1546 struct tcp_fastopen_cookie
*foc
,
1547 const struct dst_entry
*dst
);
1548 void tcp_fastopen_init_key_once(struct net
*net
);
1549 bool tcp_fastopen_cookie_check(struct sock
*sk
, u16
*mss
,
1550 struct tcp_fastopen_cookie
*cookie
);
1551 bool tcp_fastopen_defer_connect(struct sock
*sk
, int *err
);
1552 #define TCP_FASTOPEN_KEY_LENGTH 16
1554 /* Fastopen key context */
1555 struct tcp_fastopen_context
{
1556 struct crypto_cipher
*tfm
;
1557 __u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
1558 struct rcu_head rcu
;
1561 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout
;
1562 void tcp_fastopen_active_disable(struct sock
*sk
);
1563 bool tcp_fastopen_active_should_disable(struct sock
*sk
);
1564 void tcp_fastopen_active_disable_ofo_check(struct sock
*sk
);
1565 void tcp_fastopen_active_timeout_reset(void);
1567 /* Latencies incurred by various limits for a sender. They are
1568 * chronograph-like stats that are mutually exclusive.
1572 TCP_CHRONO_BUSY
, /* Actively sending data (non-empty write queue) */
1573 TCP_CHRONO_RWND_LIMITED
, /* Stalled by insufficient receive window */
1574 TCP_CHRONO_SNDBUF_LIMITED
, /* Stalled by insufficient send buffer */
1578 void tcp_chrono_start(struct sock
*sk
, const enum tcp_chrono type
);
1579 void tcp_chrono_stop(struct sock
*sk
, const enum tcp_chrono type
);
1581 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1582 * the same memory storage than skb->destructor/_skb_refdst
1584 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff
*skb
)
1586 skb
->destructor
= NULL
;
1587 skb
->_skb_refdst
= 0UL;
1590 #define tcp_skb_tsorted_save(skb) { \
1591 unsigned long _save = skb->_skb_refdst; \
1592 skb->_skb_refdst = 0UL;
1594 #define tcp_skb_tsorted_restore(skb) \
1595 skb->_skb_refdst = _save; \
1598 void tcp_write_queue_purge(struct sock
*sk
);
1600 static inline struct sk_buff
*tcp_rtx_queue_head(const struct sock
*sk
)
1602 return skb_rb_first(&sk
->tcp_rtx_queue
);
1605 static inline struct sk_buff
*tcp_rtx_queue_tail(const struct sock
*sk
)
1607 return skb_rb_last(&sk
->tcp_rtx_queue
);
1610 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1612 return skb_peek(&sk
->sk_write_queue
);
1615 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1617 return skb_peek_tail(&sk
->sk_write_queue
);
1620 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1621 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1623 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1625 return skb_peek(&sk
->sk_write_queue
);
1628 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1629 const struct sk_buff
*skb
)
1631 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1634 static inline bool tcp_write_queue_empty(const struct sock
*sk
)
1636 return skb_queue_empty(&sk
->sk_write_queue
);
1639 static inline bool tcp_rtx_queue_empty(const struct sock
*sk
)
1641 return RB_EMPTY_ROOT(&sk
->tcp_rtx_queue
);
1644 static inline bool tcp_rtx_and_write_queues_empty(const struct sock
*sk
)
1646 return tcp_rtx_queue_empty(sk
) && tcp_write_queue_empty(sk
);
1649 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1651 if (tcp_write_queue_empty(sk
))
1652 tcp_chrono_stop(sk
, TCP_CHRONO_BUSY
);
1655 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1657 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1660 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1662 __tcp_add_write_queue_tail(sk
, skb
);
1664 /* Queue it, remembering where we must start sending. */
1665 if (sk
->sk_write_queue
.next
== skb
)
1666 tcp_chrono_start(sk
, TCP_CHRONO_BUSY
);
1669 /* Insert new before skb on the write queue of sk. */
1670 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1671 struct sk_buff
*skb
,
1674 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1677 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1679 tcp_skb_tsorted_anchor_cleanup(skb
);
1680 __skb_unlink(skb
, &sk
->sk_write_queue
);
1683 void tcp_rbtree_insert(struct rb_root
*root
, struct sk_buff
*skb
);
1685 static inline void tcp_rtx_queue_unlink(struct sk_buff
*skb
, struct sock
*sk
)
1687 tcp_skb_tsorted_anchor_cleanup(skb
);
1688 rb_erase(&skb
->rbnode
, &sk
->tcp_rtx_queue
);
1691 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff
*skb
, struct sock
*sk
)
1693 list_del(&skb
->tcp_tsorted_anchor
);
1694 tcp_rtx_queue_unlink(skb
, sk
);
1695 sk_wmem_free_skb(sk
, skb
);
1698 static inline void tcp_push_pending_frames(struct sock
*sk
)
1700 if (tcp_send_head(sk
)) {
1701 struct tcp_sock
*tp
= tcp_sk(sk
);
1703 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1707 /* Start sequence of the skb just after the highest skb with SACKed
1708 * bit, valid only if sacked_out > 0 or when the caller has ensured
1709 * validity by itself.
1711 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1713 if (!tp
->sacked_out
)
1716 if (tp
->highest_sack
== NULL
)
1719 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1722 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1724 tcp_sk(sk
)->highest_sack
= skb_rb_next(skb
);
1727 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1729 return tcp_sk(sk
)->highest_sack
;
1732 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1734 tcp_sk(sk
)->highest_sack
= tcp_rtx_queue_head(sk
);
1737 /* Called when old skb is about to be deleted and replaced by new skb */
1738 static inline void tcp_highest_sack_replace(struct sock
*sk
,
1739 struct sk_buff
*old
,
1740 struct sk_buff
*new)
1742 if (old
== tcp_highest_sack(sk
))
1743 tcp_sk(sk
)->highest_sack
= new;
1746 /* This helper checks if socket has IP_TRANSPARENT set */
1747 static inline bool inet_sk_transparent(const struct sock
*sk
)
1749 switch (sk
->sk_state
) {
1751 return inet_twsk(sk
)->tw_transparent
;
1752 case TCP_NEW_SYN_RECV
:
1753 return inet_rsk(inet_reqsk(sk
))->no_srccheck
;
1755 return inet_sk(sk
)->transparent
;
1758 /* Determines whether this is a thin stream (which may suffer from
1759 * increased latency). Used to trigger latency-reducing mechanisms.
1761 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1763 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1767 enum tcp_seq_states
{
1768 TCP_SEQ_STATE_LISTENING
,
1769 TCP_SEQ_STATE_ESTABLISHED
,
1772 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1774 struct tcp_seq_afinfo
{
1777 const struct file_operations
*seq_fops
;
1778 struct seq_operations seq_ops
;
1781 struct tcp_iter_state
{
1782 struct seq_net_private p
;
1784 enum tcp_seq_states state
;
1785 struct sock
*syn_wait_sk
;
1786 int bucket
, offset
, sbucket
, num
;
1790 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1791 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1793 extern struct request_sock_ops tcp_request_sock_ops
;
1794 extern struct request_sock_ops tcp6_request_sock_ops
;
1796 void tcp_v4_destroy_sock(struct sock
*sk
);
1798 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1799 netdev_features_t features
);
1800 struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
);
1801 int tcp_gro_complete(struct sk_buff
*skb
);
1803 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1805 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1807 struct net
*net
= sock_net((struct sock
*)tp
);
1808 return tp
->notsent_lowat
?: net
->ipv4
.sysctl_tcp_notsent_lowat
;
1811 static inline bool tcp_stream_memory_free(const struct sock
*sk
)
1813 const struct tcp_sock
*tp
= tcp_sk(sk
);
1814 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1816 return notsent_bytes
< tcp_notsent_lowat(tp
);
1819 #ifdef CONFIG_PROC_FS
1820 int tcp4_proc_init(void);
1821 void tcp4_proc_exit(void);
1824 int tcp_rtx_synack(const struct sock
*sk
, struct request_sock
*req
);
1825 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1826 const struct tcp_request_sock_ops
*af_ops
,
1827 struct sock
*sk
, struct sk_buff
*skb
);
1829 /* TCP af-specific functions */
1830 struct tcp_sock_af_ops
{
1831 #ifdef CONFIG_TCP_MD5SIG
1832 struct tcp_md5sig_key
*(*md5_lookup
) (const struct sock
*sk
,
1833 const struct sock
*addr_sk
);
1834 int (*calc_md5_hash
)(char *location
,
1835 const struct tcp_md5sig_key
*md5
,
1836 const struct sock
*sk
,
1837 const struct sk_buff
*skb
);
1838 int (*md5_parse
)(struct sock
*sk
,
1840 char __user
*optval
,
1845 struct tcp_request_sock_ops
{
1847 #ifdef CONFIG_TCP_MD5SIG
1848 struct tcp_md5sig_key
*(*req_md5_lookup
)(const struct sock
*sk
,
1849 const struct sock
*addr_sk
);
1850 int (*calc_md5_hash
) (char *location
,
1851 const struct tcp_md5sig_key
*md5
,
1852 const struct sock
*sk
,
1853 const struct sk_buff
*skb
);
1855 void (*init_req
)(struct request_sock
*req
,
1856 const struct sock
*sk_listener
,
1857 struct sk_buff
*skb
);
1858 #ifdef CONFIG_SYN_COOKIES
1859 __u32 (*cookie_init_seq
)(const struct sk_buff
*skb
,
1862 struct dst_entry
*(*route_req
)(const struct sock
*sk
, struct flowi
*fl
,
1863 const struct request_sock
*req
);
1864 u32 (*init_seq
)(const struct sk_buff
*skb
);
1865 u32 (*init_ts_off
)(const struct net
*net
, const struct sk_buff
*skb
);
1866 int (*send_synack
)(const struct sock
*sk
, struct dst_entry
*dst
,
1867 struct flowi
*fl
, struct request_sock
*req
,
1868 struct tcp_fastopen_cookie
*foc
,
1869 enum tcp_synack_type synack_type
);
1872 #ifdef CONFIG_SYN_COOKIES
1873 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1874 const struct sock
*sk
, struct sk_buff
*skb
,
1877 tcp_synq_overflow(sk
);
1878 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_SYNCOOKIESSENT
);
1879 return ops
->cookie_init_seq(skb
, mss
);
1882 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1883 const struct sock
*sk
, struct sk_buff
*skb
,
1890 int tcpv4_offload_init(void);
1892 void tcp_v4_init(void);
1893 void tcp_init(void);
1895 /* tcp_recovery.c */
1896 extern void tcp_rack_mark_lost(struct sock
*sk
);
1897 extern void tcp_rack_advance(struct tcp_sock
*tp
, u8 sacked
, u32 end_seq
,
1899 extern void tcp_rack_reo_timeout(struct sock
*sk
);
1900 extern void tcp_rack_update_reo_wnd(struct sock
*sk
, struct rate_sample
*rs
);
1902 /* At how many usecs into the future should the RTO fire? */
1903 static inline s64
tcp_rto_delta_us(const struct sock
*sk
)
1905 const struct sk_buff
*skb
= tcp_rtx_queue_head(sk
);
1906 u32 rto
= inet_csk(sk
)->icsk_rto
;
1907 u64 rto_time_stamp_us
= skb
->skb_mstamp
+ jiffies_to_usecs(rto
);
1909 return rto_time_stamp_us
- tcp_sk(sk
)->tcp_mstamp
;
1913 * Save and compile IPv4 options, return a pointer to it
1915 static inline struct ip_options_rcu
*tcp_v4_save_options(struct net
*net
,
1916 struct sk_buff
*skb
)
1918 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
1919 struct ip_options_rcu
*dopt
= NULL
;
1922 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
1924 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
1925 if (dopt
&& __ip_options_echo(net
, &dopt
->opt
, skb
, opt
)) {
1933 /* locally generated TCP pure ACKs have skb->truesize == 2
1934 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1935 * This is much faster than dissecting the packet to find out.
1936 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1938 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
1940 return skb
->truesize
== 2;
1943 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)
1948 static inline int tcp_inq(struct sock
*sk
)
1950 struct tcp_sock
*tp
= tcp_sk(sk
);
1953 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
1955 } else if (sock_flag(sk
, SOCK_URGINLINE
) ||
1957 before(tp
->urg_seq
, tp
->copied_seq
) ||
1958 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
1960 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
1962 /* Subtract 1, if FIN was received */
1963 if (answ
&& sock_flag(sk
, SOCK_DONE
))
1966 answ
= tp
->urg_seq
- tp
->copied_seq
;
1972 int tcp_peek_len(struct socket
*sock
);
1974 static inline void tcp_segs_in(struct tcp_sock
*tp
, const struct sk_buff
*skb
)
1978 segs_in
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
1979 tp
->segs_in
+= segs_in
;
1980 if (skb
->len
> tcp_hdrlen(skb
))
1981 tp
->data_segs_in
+= segs_in
;
1985 * TCP listen path runs lockless.
1986 * We forced "struct sock" to be const qualified to make sure
1987 * we don't modify one of its field by mistake.
1988 * Here, we increment sk_drops which is an atomic_t, so we can safely
1989 * make sock writable again.
1991 static inline void tcp_listendrop(const struct sock
*sk
)
1993 atomic_inc(&((struct sock
*)sk
)->sk_drops
);
1994 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1997 enum hrtimer_restart
tcp_pace_kick(struct hrtimer
*timer
);
2000 * Interface for adding Upper Level Protocols over TCP
2003 #define TCP_ULP_NAME_MAX 16
2004 #define TCP_ULP_MAX 128
2005 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2007 struct tcp_ulp_ops
{
2008 struct list_head list
;
2010 /* initialize ulp */
2011 int (*init
)(struct sock
*sk
);
2013 void (*release
)(struct sock
*sk
);
2015 char name
[TCP_ULP_NAME_MAX
];
2016 struct module
*owner
;
2018 int tcp_register_ulp(struct tcp_ulp_ops
*type
);
2019 void tcp_unregister_ulp(struct tcp_ulp_ops
*type
);
2020 int tcp_set_ulp(struct sock
*sk
, const char *name
);
2021 void tcp_get_available_ulp(char *buf
, size_t len
);
2022 void tcp_cleanup_ulp(struct sock
*sk
);
2024 #define MODULE_ALIAS_TCP_ULP(name) \
2025 __MODULE_INFO(alias, alias_userspace, name); \
2026 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2028 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2029 * is < 0, then the BPF op failed (for example if the loaded BPF
2030 * program does not support the chosen operation or there is no BPF
2034 static inline int tcp_call_bpf(struct sock
*sk
, int op
)
2036 struct bpf_sock_ops_kern sock_ops
;
2039 if (sk_fullsock(sk
))
2040 sock_owned_by_me(sk
);
2042 memset(&sock_ops
, 0, sizeof(sock_ops
));
2046 ret
= BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops
);
2048 ret
= sock_ops
.reply
;
2054 static inline int tcp_call_bpf(struct sock
*sk
, int op
)
2060 static inline u32
tcp_timeout_init(struct sock
*sk
)
2064 timeout
= tcp_call_bpf(sk
, BPF_SOCK_OPS_TIMEOUT_INIT
);
2067 timeout
= TCP_TIMEOUT_INIT
;
2071 static inline u32
tcp_rwnd_init_bpf(struct sock
*sk
)
2075 rwnd
= tcp_call_bpf(sk
, BPF_SOCK_OPS_RWND_INIT
);
2082 static inline bool tcp_bpf_ca_needs_ecn(struct sock
*sk
)
2084 return (tcp_call_bpf(sk
, BPF_SOCK_OPS_NEEDS_ECN
) == 1);
2087 #if IS_ENABLED(CONFIG_SMC)
2088 extern struct static_key_false tcp_have_smc
;