1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Definitions for the TCP module.
9 * Version: @(#)tcp.h 1.0.5 05/23/93
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
17 #define FASTRETRANS_DEBUG 1
19 #include <linux/list.h>
20 #include <linux/tcp.h>
21 #include <linux/bug.h>
22 #include <linux/slab.h>
23 #include <linux/cache.h>
24 #include <linux/percpu.h>
25 #include <linux/skbuff.h>
26 #include <linux/cryptohash.h>
27 #include <linux/kref.h>
28 #include <linux/ktime.h>
30 #include <net/inet_connection_sock.h>
31 #include <net/inet_timewait_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
35 #include <net/sock_reuseport.h>
39 #include <net/tcp_states.h>
40 #include <net/inet_ecn.h>
43 #include <linux/seq_file.h>
44 #include <linux/memcontrol.h>
45 #include <linux/bpf-cgroup.h>
46 #include <linux/siphash.h>
48 extern struct inet_hashinfo tcp_hashinfo
;
50 extern struct percpu_counter tcp_orphan_count
;
51 void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
53 #define MAX_TCP_HEADER (128 + MAX_HEADER)
54 #define MAX_TCP_OPTION_SPACE 40
55 #define TCP_MIN_SND_MSS 48
56 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
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 */
248 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
250 extern atomic_long_t tcp_memory_allocated
;
251 extern struct percpu_counter tcp_sockets_allocated
;
252 extern unsigned long tcp_memory_pressure
;
254 /* optimized version of sk_under_memory_pressure() for TCP sockets */
255 static inline bool tcp_under_memory_pressure(const struct sock
*sk
)
257 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
258 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
261 return tcp_memory_pressure
;
264 * The next routines deal with comparing 32 bit unsigned ints
265 * and worry about wraparound (automatic with unsigned arithmetic).
268 static inline bool before(__u32 seq1
, __u32 seq2
)
270 return (__s32
)(seq1
-seq2
) < 0;
272 #define after(seq2, seq1) before(seq1, seq2)
274 /* is s2<=s1<=s3 ? */
275 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
277 return seq3
- seq2
>= seq1
- seq2
;
280 static inline bool tcp_out_of_memory(struct sock
*sk
)
282 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
283 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
288 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
290 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
292 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
293 int orphans
= percpu_counter_read_positive(ocp
);
295 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
296 orphans
= percpu_counter_sum_positive(ocp
);
297 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
303 bool tcp_check_oom(struct sock
*sk
, int shift
);
306 extern struct proto tcp_prot
;
308 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
311 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
313 void tcp_tasklet_init(void);
315 int tcp_v4_err(struct sk_buff
*skb
, u32
);
317 void tcp_shutdown(struct sock
*sk
, int how
);
319 int tcp_v4_early_demux(struct sk_buff
*skb
);
320 int tcp_v4_rcv(struct sk_buff
*skb
);
322 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
323 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
324 int tcp_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
325 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
327 int tcp_sendpage_locked(struct sock
*sk
, struct page
*page
, int offset
,
328 size_t size
, int flags
);
329 ssize_t
do_tcp_sendpages(struct sock
*sk
, struct page
*page
, int offset
,
330 size_t size
, int flags
);
331 void tcp_release_cb(struct sock
*sk
);
332 void tcp_wfree(struct sk_buff
*skb
);
333 void tcp_write_timer_handler(struct sock
*sk
);
334 void tcp_delack_timer_handler(struct sock
*sk
);
335 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
336 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
);
337 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
);
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 void tcp_enter_quickack_mode(struct sock
*sk
, unsigned int max_quickacks
);
346 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
347 const unsigned int pkts
)
349 struct inet_connection_sock
*icsk
= inet_csk(sk
);
351 if (icsk
->icsk_ack
.quick
) {
352 if (pkts
>= icsk
->icsk_ack
.quick
) {
353 icsk
->icsk_ack
.quick
= 0;
354 /* Leaving quickack mode we deflate ATO. */
355 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
357 icsk
->icsk_ack
.quick
-= pkts
;
362 #define TCP_ECN_QUEUE_CWR 2
363 #define TCP_ECN_DEMAND_CWR 4
364 #define TCP_ECN_SEEN 8
374 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
376 const struct tcphdr
*th
);
377 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
378 struct request_sock
*req
, bool fastopen
,
380 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
381 struct sk_buff
*skb
);
382 void tcp_enter_loss(struct sock
*sk
);
383 void tcp_cwnd_reduction(struct sock
*sk
, int newly_acked_sacked
, int flag
);
384 void tcp_clear_retrans(struct tcp_sock
*tp
);
385 void tcp_update_metrics(struct sock
*sk
);
386 void tcp_init_metrics(struct sock
*sk
);
387 void tcp_metrics_init(void);
388 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
);
389 void tcp_close(struct sock
*sk
, long timeout
);
390 void tcp_init_sock(struct sock
*sk
);
391 void tcp_init_transfer(struct sock
*sk
, int bpf_op
);
392 __poll_t
tcp_poll(struct file
*file
, struct socket
*sock
,
393 struct poll_table_struct
*wait
);
394 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
395 char __user
*optval
, int __user
*optlen
);
396 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
397 char __user
*optval
, unsigned int optlen
);
398 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
399 char __user
*optval
, int __user
*optlen
);
400 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
401 char __user
*optval
, unsigned int optlen
);
402 void tcp_set_keepalive(struct sock
*sk
, int val
);
403 void tcp_syn_ack_timeout(const struct request_sock
*req
);
404 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
405 int flags
, int *addr_len
);
406 int tcp_set_rcvlowat(struct sock
*sk
, int val
);
407 void tcp_data_ready(struct sock
*sk
);
409 int tcp_mmap(struct file
*file
, struct socket
*sock
,
410 struct vm_area_struct
*vma
);
412 void tcp_parse_options(const struct net
*net
, const struct sk_buff
*skb
,
413 struct tcp_options_received
*opt_rx
,
414 int estab
, struct tcp_fastopen_cookie
*foc
);
415 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
418 * TCP v4 functions exported for the inet6 API
421 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
422 void tcp_v4_mtu_reduced(struct sock
*sk
);
423 void tcp_req_err(struct sock
*sk
, u32 seq
, bool abort
);
424 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
425 struct sock
*tcp_create_openreq_child(const struct sock
*sk
,
426 struct request_sock
*req
,
427 struct sk_buff
*skb
);
428 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
429 struct sock
*tcp_v4_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
430 struct request_sock
*req
,
431 struct dst_entry
*dst
,
432 struct request_sock
*req_unhash
,
434 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
435 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
436 int tcp_connect(struct sock
*sk
);
437 enum tcp_synack_type
{
442 struct sk_buff
*tcp_make_synack(const struct sock
*sk
, struct dst_entry
*dst
,
443 struct request_sock
*req
,
444 struct tcp_fastopen_cookie
*foc
,
445 enum tcp_synack_type synack_type
);
446 int tcp_disconnect(struct sock
*sk
, int flags
);
448 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
449 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
450 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
452 /* From syncookies.c */
453 struct sock
*tcp_get_cookie_sock(struct sock
*sk
, struct sk_buff
*skb
,
454 struct request_sock
*req
,
455 struct dst_entry
*dst
, u32 tsoff
);
456 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
458 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
459 #ifdef CONFIG_SYN_COOKIES
461 /* Syncookies use a monotonic timer which increments every 60 seconds.
462 * This counter is used both as a hash input and partially encoded into
463 * the cookie value. A cookie is only validated further if the delta
464 * between the current counter value and the encoded one is less than this,
465 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
466 * the counter advances immediately after a cookie is generated).
468 #define MAX_SYNCOOKIE_AGE 2
469 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
470 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
472 /* syncookies: remember time of last synqueue overflow
473 * But do not dirty this field too often (once per second is enough)
474 * It is racy as we do not hold a lock, but race is very minor.
476 static inline void tcp_synq_overflow(const struct sock
*sk
)
478 unsigned int last_overflow
;
479 unsigned int now
= jiffies
;
481 if (sk
->sk_reuseport
) {
482 struct sock_reuseport
*reuse
;
484 reuse
= rcu_dereference(sk
->sk_reuseport_cb
);
486 last_overflow
= READ_ONCE(reuse
->synq_overflow_ts
);
487 if (time_after32(now
, last_overflow
+ HZ
))
488 WRITE_ONCE(reuse
->synq_overflow_ts
, now
);
493 last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
494 if (time_after32(now
, last_overflow
+ HZ
))
495 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= now
;
498 /* syncookies: no recent synqueue overflow on this listening socket? */
499 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
501 unsigned int last_overflow
;
502 unsigned int now
= jiffies
;
504 if (sk
->sk_reuseport
) {
505 struct sock_reuseport
*reuse
;
507 reuse
= rcu_dereference(sk
->sk_reuseport_cb
);
509 last_overflow
= READ_ONCE(reuse
->synq_overflow_ts
);
510 return time_after32(now
, last_overflow
+
511 TCP_SYNCOOKIE_VALID
);
515 last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
516 return time_after32(now
, last_overflow
+ TCP_SYNCOOKIE_VALID
);
519 static inline u32
tcp_cookie_time(void)
521 u64 val
= get_jiffies_64();
523 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
527 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
529 __u32
cookie_v4_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
530 u64
cookie_init_timestamp(struct request_sock
*req
);
531 bool cookie_timestamp_decode(const struct net
*net
,
532 struct tcp_options_received
*opt
);
533 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
534 const struct net
*net
, const struct dst_entry
*dst
);
536 /* From net/ipv6/syncookies.c */
537 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
539 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
541 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
542 const struct tcphdr
*th
, u16
*mssp
);
543 __u32
cookie_v6_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
547 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
549 int __tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
550 int tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
551 void tcp_retransmit_timer(struct sock
*sk
);
552 void tcp_xmit_retransmit_queue(struct sock
*);
553 void tcp_simple_retransmit(struct sock
*);
554 void tcp_enter_recovery(struct sock
*sk
, bool ece_ack
);
555 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
557 TCP_FRAG_IN_WRITE_QUEUE
,
558 TCP_FRAG_IN_RTX_QUEUE
,
560 int tcp_fragment(struct sock
*sk
, enum tcp_queue tcp_queue
,
561 struct sk_buff
*skb
, u32 len
,
562 unsigned int mss_now
, gfp_t gfp
);
564 void tcp_send_probe0(struct sock
*);
565 void tcp_send_partial(struct sock
*);
566 int tcp_write_wakeup(struct sock
*, int mib
);
567 void tcp_send_fin(struct sock
*sk
);
568 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
569 int tcp_send_synack(struct sock
*);
570 void tcp_push_one(struct sock
*, unsigned int mss_now
);
571 void __tcp_send_ack(struct sock
*sk
, u32 rcv_nxt
);
572 void tcp_send_ack(struct sock
*sk
);
573 void tcp_send_delayed_ack(struct sock
*sk
);
574 void tcp_send_loss_probe(struct sock
*sk
);
575 bool tcp_schedule_loss_probe(struct sock
*sk
, bool advancing_rto
);
576 void tcp_skb_collapse_tstamp(struct sk_buff
*skb
,
577 const struct sk_buff
*next_skb
);
580 void tcp_rearm_rto(struct sock
*sk
);
581 void tcp_synack_rtt_meas(struct sock
*sk
, struct request_sock
*req
);
582 void tcp_reset(struct sock
*sk
);
583 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock
*tp
, struct sk_buff
*skb
);
584 void tcp_fin(struct sock
*sk
);
587 void tcp_init_xmit_timers(struct sock
*);
588 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
590 if (hrtimer_try_to_cancel(&tcp_sk(sk
)->pacing_timer
) == 1)
593 if (hrtimer_try_to_cancel(&tcp_sk(sk
)->compressed_ack_timer
) == 1)
596 inet_csk_clear_xmit_timers(sk
);
599 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
600 unsigned int tcp_current_mss(struct sock
*sk
);
602 /* Bound MSS / TSO packet size with the half of the window */
603 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
607 /* When peer uses tiny windows, there is no use in packetizing
608 * to sub-MSS pieces for the sake of SWS or making sure there
609 * are enough packets in the pipe for fast recovery.
611 * On the other hand, for extremely large MSS devices, handling
612 * smaller than MSS windows in this way does make sense.
614 if (tp
->max_window
> TCP_MSS_DEFAULT
)
615 cutoff
= (tp
->max_window
>> 1);
617 cutoff
= tp
->max_window
;
619 if (cutoff
&& pktsize
> cutoff
)
620 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
626 void tcp_get_info(struct sock
*, struct tcp_info
*);
628 /* Read 'sendfile()'-style from a TCP socket */
629 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
630 sk_read_actor_t recv_actor
);
632 void tcp_initialize_rcv_mss(struct sock
*sk
);
634 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
635 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
636 void tcp_mtup_init(struct sock
*sk
);
637 void tcp_init_buffer_space(struct sock
*sk
);
639 static inline void tcp_bound_rto(const struct sock
*sk
)
641 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
642 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
645 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
647 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
650 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
652 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
653 ntohl(TCP_FLAG_ACK
) |
657 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
659 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
662 static inline void tcp_fast_path_check(struct sock
*sk
)
664 struct tcp_sock
*tp
= tcp_sk(sk
);
666 if (RB_EMPTY_ROOT(&tp
->out_of_order_queue
) &&
668 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
670 tcp_fast_path_on(tp
);
673 /* Compute the actual rto_min value */
674 static inline u32
tcp_rto_min(struct sock
*sk
)
676 const struct dst_entry
*dst
= __sk_dst_get(sk
);
677 u32 rto_min
= TCP_RTO_MIN
;
679 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
680 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
684 static inline u32
tcp_rto_min_us(struct sock
*sk
)
686 return jiffies_to_usecs(tcp_rto_min(sk
));
689 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
691 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
694 /* Minimum RTT in usec. ~0 means not available. */
695 static inline u32
tcp_min_rtt(const struct tcp_sock
*tp
)
697 return minmax_get(&tp
->rtt_min
);
700 /* Compute the actual receive window we are currently advertising.
701 * Rcv_nxt can be after the window if our peer push more data
702 * than the offered window.
704 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
706 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
713 /* Choose a new window, without checks for shrinking, and without
714 * scaling applied to the result. The caller does these things
715 * if necessary. This is a "raw" window selection.
717 u32
__tcp_select_window(struct sock
*sk
);
719 void tcp_send_window_probe(struct sock
*sk
);
721 /* TCP uses 32bit jiffies to save some space.
722 * Note that this is different from tcp_time_stamp, which
723 * historically has been the same until linux-4.13.
725 #define tcp_jiffies32 ((u32)jiffies)
728 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
729 * It is no longer tied to jiffies, but to 1 ms clock.
730 * Note: double check if you want to use tcp_jiffies32 instead of this.
732 #define TCP_TS_HZ 1000
734 static inline u64
tcp_clock_ns(void)
736 return ktime_get_ns();
739 static inline u64
tcp_clock_us(void)
741 return div_u64(tcp_clock_ns(), NSEC_PER_USEC
);
744 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
745 static inline u32
tcp_time_stamp(const struct tcp_sock
*tp
)
747 return div_u64(tp
->tcp_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
750 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
751 static inline u32
tcp_time_stamp_raw(void)
753 return div_u64(tcp_clock_ns(), NSEC_PER_SEC
/ TCP_TS_HZ
);
756 void tcp_mstamp_refresh(struct tcp_sock
*tp
);
758 static inline u32
tcp_stamp_us_delta(u64 t1
, u64 t0
)
760 return max_t(s64
, t1
- t0
, 0);
763 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
765 return div_u64(skb
->skb_mstamp_ns
, NSEC_PER_SEC
/ TCP_TS_HZ
);
768 /* provide the departure time in us unit */
769 static inline u64
tcp_skb_timestamp_us(const struct sk_buff
*skb
)
771 return div_u64(skb
->skb_mstamp_ns
, NSEC_PER_USEC
);
775 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
777 #define TCPHDR_FIN 0x01
778 #define TCPHDR_SYN 0x02
779 #define TCPHDR_RST 0x04
780 #define TCPHDR_PSH 0x08
781 #define TCPHDR_ACK 0x10
782 #define TCPHDR_URG 0x20
783 #define TCPHDR_ECE 0x40
784 #define TCPHDR_CWR 0x80
786 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
788 /* This is what the send packet queuing engine uses to pass
789 * TCP per-packet control information to the transmission code.
790 * We also store the host-order sequence numbers in here too.
791 * This is 44 bytes if IPV6 is enabled.
792 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
795 __u32 seq
; /* Starting sequence number */
796 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
798 /* Note : tcp_tw_isn is used in input path only
799 * (isn chosen by tcp_timewait_state_process())
801 * tcp_gso_segs/size are used in write queue only,
802 * cf tcp_skb_pcount()/tcp_skb_mss()
810 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
812 __u8 sacked
; /* State flags for SACK. */
813 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
814 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
815 #define TCPCB_LOST 0x04 /* SKB is lost */
816 #define TCPCB_TAGBITS 0x07 /* All tag bits */
817 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
818 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
819 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
822 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
823 __u8 txstamp_ack
:1, /* Record TX timestamp for ack? */
824 eor
:1, /* Is skb MSG_EOR marked? */
825 has_rxtstamp
:1, /* SKB has a RX timestamp */
827 __u32 ack_seq
; /* Sequence number ACK'd */
830 /* There is space for up to 24 bytes */
831 __u32 in_flight
:30,/* Bytes in flight at transmit */
832 is_app_limited
:1, /* cwnd not fully used? */
834 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
836 /* start of send pipeline phase */
838 /* when we reached the "delivered" count */
839 u64 delivered_mstamp
;
840 } tx
; /* only used for outgoing skbs */
842 struct inet_skb_parm h4
;
843 #if IS_ENABLED(CONFIG_IPV6)
844 struct inet6_skb_parm h6
;
846 } header
; /* For incoming skbs */
849 struct sock
*sk_redir
;
855 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
857 static inline void bpf_compute_data_end_sk_skb(struct sk_buff
*skb
)
859 TCP_SKB_CB(skb
)->bpf
.data_end
= skb
->data
+ skb_headlen(skb
);
862 static inline bool tcp_skb_bpf_ingress(const struct sk_buff
*skb
)
864 return TCP_SKB_CB(skb
)->bpf
.flags
& BPF_F_INGRESS
;
867 static inline struct sock
*tcp_skb_bpf_redirect_fetch(struct sk_buff
*skb
)
869 return TCP_SKB_CB(skb
)->bpf
.sk_redir
;
872 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff
*skb
)
874 TCP_SKB_CB(skb
)->bpf
.sk_redir
= NULL
;
877 #if IS_ENABLED(CONFIG_IPV6)
878 /* This is the variant of inet6_iif() that must be used by TCP,
879 * as TCP moves IP6CB into a different location in skb->cb[]
881 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
883 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
886 static inline int tcp_v6_iif_l3_slave(const struct sk_buff
*skb
)
888 bool l3_slave
= ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
);
890 return l3_slave
? skb
->skb_iif
: TCP_SKB_CB(skb
)->header
.h6
.iif
;
893 /* TCP_SKB_CB reference means this can not be used from early demux */
894 static inline int tcp_v6_sdif(const struct sk_buff
*skb
)
896 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
897 if (skb
&& ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
))
898 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
904 static inline bool inet_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
906 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
907 if (!net
->ipv4
.sysctl_tcp_l3mdev_accept
&&
908 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
914 /* TCP_SKB_CB reference means this can not be used from early demux */
915 static inline int tcp_v4_sdif(struct sk_buff
*skb
)
917 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
918 if (skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
919 return TCP_SKB_CB(skb
)->header
.h4
.iif
;
924 /* Due to TSO, an SKB can be composed of multiple actual
925 * packets. To keep these tracked properly, we use this.
927 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
929 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
932 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
934 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
937 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
939 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
942 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
943 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
945 return TCP_SKB_CB(skb
)->tcp_gso_size
;
948 static inline bool tcp_skb_can_collapse_to(const struct sk_buff
*skb
)
950 return likely(!TCP_SKB_CB(skb
)->eor
);
953 /* Events passed to congestion control interface */
955 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
956 CA_EVENT_CWND_RESTART
, /* congestion window restart */
957 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
958 CA_EVENT_LOSS
, /* loss timeout */
959 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
960 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
963 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
964 enum tcp_ca_ack_event_flags
{
965 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
966 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
967 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
971 * Interface for adding new TCP congestion control handlers
973 #define TCP_CA_NAME_MAX 16
974 #define TCP_CA_MAX 128
975 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
977 #define TCP_CA_UNSPEC 0
979 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
980 #define TCP_CONG_NON_RESTRICTED 0x1
981 /* Requires ECN/ECT set on all packets */
982 #define TCP_CONG_NEEDS_ECN 0x2
992 /* A rate sample measures the number of (original/retransmitted) data
993 * packets delivered "delivered" over an interval of time "interval_us".
994 * The tcp_rate.c code fills in the rate sample, and congestion
995 * control modules that define a cong_control function to run at the end
996 * of ACK processing can optionally chose to consult this sample when
997 * setting cwnd and pacing rate.
998 * A sample is invalid if "delivered" or "interval_us" is negative.
1000 struct rate_sample
{
1001 u64 prior_mstamp
; /* starting timestamp for interval */
1002 u32 prior_delivered
; /* tp->delivered at "prior_mstamp" */
1003 s32 delivered
; /* number of packets delivered over interval */
1004 long interval_us
; /* time for tp->delivered to incr "delivered" */
1005 u32 snd_interval_us
; /* snd interval for delivered packets */
1006 u32 rcv_interval_us
; /* rcv interval for delivered packets */
1007 long rtt_us
; /* RTT of last (S)ACKed packet (or -1) */
1008 int losses
; /* number of packets marked lost upon ACK */
1009 u32 acked_sacked
; /* number of packets newly (S)ACKed upon ACK */
1010 u32 prior_in_flight
; /* in flight before this ACK */
1011 bool is_app_limited
; /* is sample from packet with bubble in pipe? */
1012 bool is_retrans
; /* is sample from retransmission? */
1013 bool is_ack_delayed
; /* is this (likely) a delayed ACK? */
1016 struct tcp_congestion_ops
{
1017 struct list_head list
;
1021 /* initialize private data (optional) */
1022 void (*init
)(struct sock
*sk
);
1023 /* cleanup private data (optional) */
1024 void (*release
)(struct sock
*sk
);
1026 /* return slow start threshold (required) */
1027 u32 (*ssthresh
)(struct sock
*sk
);
1028 /* do new cwnd calculation (required) */
1029 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
1030 /* call before changing ca_state (optional) */
1031 void (*set_state
)(struct sock
*sk
, u8 new_state
);
1032 /* call when cwnd event occurs (optional) */
1033 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
1034 /* call when ack arrives (optional) */
1035 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
1036 /* new value of cwnd after loss (required) */
1037 u32 (*undo_cwnd
)(struct sock
*sk
);
1038 /* hook for packet ack accounting (optional) */
1039 void (*pkts_acked
)(struct sock
*sk
, const struct ack_sample
*sample
);
1040 /* override sysctl_tcp_min_tso_segs */
1041 u32 (*min_tso_segs
)(struct sock
*sk
);
1042 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1043 u32 (*sndbuf_expand
)(struct sock
*sk
);
1044 /* call when packets are delivered to update cwnd and pacing rate,
1045 * after all the ca_state processing. (optional)
1047 void (*cong_control
)(struct sock
*sk
, const struct rate_sample
*rs
);
1048 /* get info for inet_diag (optional) */
1049 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
1050 union tcp_cc_info
*info
);
1052 char name
[TCP_CA_NAME_MAX
];
1053 struct module
*owner
;
1056 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1057 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1059 void tcp_assign_congestion_control(struct sock
*sk
);
1060 void tcp_init_congestion_control(struct sock
*sk
);
1061 void tcp_cleanup_congestion_control(struct sock
*sk
);
1062 int tcp_set_default_congestion_control(struct net
*net
, const char *name
);
1063 void tcp_get_default_congestion_control(struct net
*net
, char *name
);
1064 void tcp_get_available_congestion_control(char *buf
, size_t len
);
1065 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
1066 int tcp_set_allowed_congestion_control(char *allowed
);
1067 int tcp_set_congestion_control(struct sock
*sk
, const char *name
, bool load
,
1068 bool reinit
, bool cap_net_admin
);
1069 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
1070 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
1072 u32
tcp_reno_ssthresh(struct sock
*sk
);
1073 u32
tcp_reno_undo_cwnd(struct sock
*sk
);
1074 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
1075 extern struct tcp_congestion_ops tcp_reno
;
1077 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
1078 u32
tcp_ca_get_key_by_name(struct net
*net
, const char *name
, bool *ecn_ca
);
1080 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
1082 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
1088 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
1090 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1092 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
1095 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
1097 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1099 if (icsk
->icsk_ca_ops
->set_state
)
1100 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
1101 icsk
->icsk_ca_state
= ca_state
;
1104 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
1106 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1108 if (icsk
->icsk_ca_ops
->cwnd_event
)
1109 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
1112 /* From tcp_rate.c */
1113 void tcp_rate_skb_sent(struct sock
*sk
, struct sk_buff
*skb
);
1114 void tcp_rate_skb_delivered(struct sock
*sk
, struct sk_buff
*skb
,
1115 struct rate_sample
*rs
);
1116 void tcp_rate_gen(struct sock
*sk
, u32 delivered
, u32 lost
,
1117 bool is_sack_reneg
, struct rate_sample
*rs
);
1118 void tcp_rate_check_app_limited(struct sock
*sk
);
1120 /* These functions determine how the current flow behaves in respect of SACK
1121 * handling. SACK is negotiated with the peer, and therefore it can vary
1122 * between different flows.
1124 * tcp_is_sack - SACK enabled
1125 * tcp_is_reno - No SACK
1127 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
1129 return likely(tp
->rx_opt
.sack_ok
);
1132 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
1134 return !tcp_is_sack(tp
);
1137 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
1139 return tp
->sacked_out
+ tp
->lost_out
;
1142 /* This determines how many packets are "in the network" to the best
1143 * of our knowledge. In many cases it is conservative, but where
1144 * detailed information is available from the receiver (via SACK
1145 * blocks etc.) we can make more aggressive calculations.
1147 * Use this for decisions involving congestion control, use just
1148 * tp->packets_out to determine if the send queue is empty or not.
1150 * Read this equation as:
1152 * "Packets sent once on transmission queue" MINUS
1153 * "Packets left network, but not honestly ACKed yet" PLUS
1154 * "Packets fast retransmitted"
1156 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1158 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
1161 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1163 static inline bool tcp_in_slow_start(const struct tcp_sock
*tp
)
1165 return tp
->snd_cwnd
< tp
->snd_ssthresh
;
1168 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
1170 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
1173 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
1175 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
1176 (1 << inet_csk(sk
)->icsk_ca_state
);
1179 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1180 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1183 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
1185 const struct tcp_sock
*tp
= tcp_sk(sk
);
1187 if (tcp_in_cwnd_reduction(sk
))
1188 return tp
->snd_ssthresh
;
1190 return max(tp
->snd_ssthresh
,
1191 ((tp
->snd_cwnd
>> 1) +
1192 (tp
->snd_cwnd
>> 2)));
1195 /* Use define here intentionally to get WARN_ON location shown at the caller */
1196 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1198 void tcp_enter_cwr(struct sock
*sk
);
1199 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1201 /* The maximum number of MSS of available cwnd for which TSO defers
1202 * sending if not using sysctl_tcp_tso_win_divisor.
1204 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1209 /* Returns end sequence number of the receiver's advertised window */
1210 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1212 return tp
->snd_una
+ tp
->snd_wnd
;
1215 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1216 * flexible approach. The RFC suggests cwnd should not be raised unless
1217 * it was fully used previously. And that's exactly what we do in
1218 * congestion avoidance mode. But in slow start we allow cwnd to grow
1219 * as long as the application has used half the cwnd.
1221 * cwnd is 10 (IW10), but application sends 9 frames.
1222 * We allow cwnd to reach 18 when all frames are ACKed.
1223 * This check is safe because it's as aggressive as slow start which already
1224 * risks 100% overshoot. The advantage is that we discourage application to
1225 * either send more filler packets or data to artificially blow up the cwnd
1226 * usage, and allow application-limited process to probe bw more aggressively.
1228 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1230 const struct tcp_sock
*tp
= tcp_sk(sk
);
1232 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1233 if (tcp_in_slow_start(tp
))
1234 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1236 return tp
->is_cwnd_limited
;
1239 /* BBR congestion control needs pacing.
1240 * Same remark for SO_MAX_PACING_RATE.
1241 * sch_fq packet scheduler is efficiently handling pacing,
1242 * but is not always installed/used.
1243 * Return true if TCP stack should pace packets itself.
1245 static inline bool tcp_needs_internal_pacing(const struct sock
*sk
)
1247 return smp_load_acquire(&sk
->sk_pacing_status
) == SK_PACING_NEEDED
;
1250 /* Return in jiffies the delay before one skb is sent.
1251 * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1253 static inline unsigned long tcp_pacing_delay(const struct sock
*sk
,
1254 const struct sk_buff
*skb
)
1256 s64 pacing_delay
= skb
? skb
->tstamp
: tcp_sk(sk
)->tcp_wstamp_ns
;
1258 pacing_delay
-= tcp_sk(sk
)->tcp_clock_cache
;
1260 return pacing_delay
> 0 ? nsecs_to_jiffies(pacing_delay
) : 0;
1263 static inline void tcp_reset_xmit_timer(struct sock
*sk
,
1266 const unsigned long max_when
,
1267 const struct sk_buff
*skb
)
1269 inet_csk_reset_xmit_timer(sk
, what
, when
+ tcp_pacing_delay(sk
, skb
),
1273 /* Something is really bad, we could not queue an additional packet,
1274 * because qdisc is full or receiver sent a 0 window, or we are paced.
1275 * We do not want to add fuel to the fire, or abort too early,
1276 * so make sure the timer we arm now is at least 200ms in the future,
1277 * regardless of current icsk_rto value (as it could be ~2ms)
1279 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1281 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1284 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1285 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1286 unsigned long max_when
)
1288 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1290 return (unsigned long)min_t(u64
, when
, max_when
);
1293 static inline void tcp_check_probe_timer(struct sock
*sk
)
1295 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1296 tcp_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1297 tcp_probe0_base(sk
), TCP_RTO_MAX
,
1301 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1306 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1312 * Calculate(/check) TCP checksum
1314 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1315 __be32 daddr
, __wsum base
)
1317 return csum_tcpudp_magic(saddr
, daddr
, len
, IPPROTO_TCP
, base
);
1320 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1322 return !skb_csum_unnecessary(skb
) &&
1323 __skb_checksum_complete(skb
);
1326 bool tcp_add_backlog(struct sock
*sk
, struct sk_buff
*skb
);
1327 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
);
1328 void tcp_set_state(struct sock
*sk
, int state
);
1329 void tcp_done(struct sock
*sk
);
1330 int tcp_abort(struct sock
*sk
, int err
);
1332 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1335 rx_opt
->num_sacks
= 0;
1338 u32
tcp_default_init_rwnd(u32 mss
);
1339 void tcp_cwnd_restart(struct sock
*sk
, s32 delta
);
1341 static inline void tcp_slow_start_after_idle_check(struct sock
*sk
)
1343 const struct tcp_congestion_ops
*ca_ops
= inet_csk(sk
)->icsk_ca_ops
;
1344 struct tcp_sock
*tp
= tcp_sk(sk
);
1347 if (!sock_net(sk
)->ipv4
.sysctl_tcp_slow_start_after_idle
|| tp
->packets_out
||
1348 ca_ops
->cong_control
)
1350 delta
= tcp_jiffies32
- tp
->lsndtime
;
1351 if (delta
> inet_csk(sk
)->icsk_rto
)
1352 tcp_cwnd_restart(sk
, delta
);
1355 /* Determine a window scaling and initial window to offer. */
1356 void tcp_select_initial_window(const struct sock
*sk
, int __space
,
1357 __u32 mss
, __u32
*rcv_wnd
,
1358 __u32
*window_clamp
, int wscale_ok
,
1359 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1361 static inline int tcp_win_from_space(const struct sock
*sk
, int space
)
1363 int tcp_adv_win_scale
= sock_net(sk
)->ipv4
.sysctl_tcp_adv_win_scale
;
1365 return tcp_adv_win_scale
<= 0 ?
1366 (space
>>(-tcp_adv_win_scale
)) :
1367 space
- (space
>>tcp_adv_win_scale
);
1370 /* Note: caller must be prepared to deal with negative returns */
1371 static inline int tcp_space(const struct sock
*sk
)
1373 return tcp_win_from_space(sk
, sk
->sk_rcvbuf
- sk
->sk_backlog
.len
-
1374 atomic_read(&sk
->sk_rmem_alloc
));
1377 static inline int tcp_full_space(const struct sock
*sk
)
1379 return tcp_win_from_space(sk
, sk
->sk_rcvbuf
);
1382 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1383 const struct sock
*sk_listener
,
1384 const struct dst_entry
*dst
);
1386 void tcp_enter_memory_pressure(struct sock
*sk
);
1387 void tcp_leave_memory_pressure(struct sock
*sk
);
1389 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1391 struct net
*net
= sock_net((struct sock
*)tp
);
1393 return tp
->keepalive_intvl
? : net
->ipv4
.sysctl_tcp_keepalive_intvl
;
1396 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1398 struct net
*net
= sock_net((struct sock
*)tp
);
1400 return tp
->keepalive_time
? : net
->ipv4
.sysctl_tcp_keepalive_time
;
1403 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1405 struct net
*net
= sock_net((struct sock
*)tp
);
1407 return tp
->keepalive_probes
? : net
->ipv4
.sysctl_tcp_keepalive_probes
;
1410 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1412 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1414 return min_t(u32
, tcp_jiffies32
- icsk
->icsk_ack
.lrcvtime
,
1415 tcp_jiffies32
- tp
->rcv_tstamp
);
1418 static inline int tcp_fin_time(const struct sock
*sk
)
1420 int fin_timeout
= tcp_sk(sk
)->linger2
? : sock_net(sk
)->ipv4
.sysctl_tcp_fin_timeout
;
1421 const int rto
= inet_csk(sk
)->icsk_rto
;
1423 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1424 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1429 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1432 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1434 if (unlikely(!time_before32(ktime_get_seconds(),
1435 rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
)))
1438 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1439 * then following tcp messages have valid values. Ignore 0 value,
1440 * or else 'negative' tsval might forbid us to accept their packets.
1442 if (!rx_opt
->ts_recent
)
1447 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1450 if (tcp_paws_check(rx_opt
, 0))
1453 /* RST segments are not recommended to carry timestamp,
1454 and, if they do, it is recommended to ignore PAWS because
1455 "their cleanup function should take precedence over timestamps."
1456 Certainly, it is mistake. It is necessary to understand the reasons
1457 of this constraint to relax it: if peer reboots, clock may go
1458 out-of-sync and half-open connections will not be reset.
1459 Actually, the problem would be not existing if all
1460 the implementations followed draft about maintaining clock
1461 via reboots. Linux-2.2 DOES NOT!
1463 However, we can relax time bounds for RST segments to MSL.
1465 if (rst
&& !time_before32(ktime_get_seconds(),
1466 rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
))
1471 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1472 int mib_idx
, u32
*last_oow_ack_time
);
1474 static inline void tcp_mib_init(struct net
*net
)
1477 TCP_ADD_STATS(net
, TCP_MIB_RTOALGORITHM
, 1);
1478 TCP_ADD_STATS(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1479 TCP_ADD_STATS(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1480 TCP_ADD_STATS(net
, TCP_MIB_MAXCONN
, -1);
1484 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1486 tp
->lost_skb_hint
= NULL
;
1489 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1491 tcp_clear_retrans_hints_partial(tp
);
1492 tp
->retransmit_skb_hint
= NULL
;
1495 union tcp_md5_addr
{
1497 #if IS_ENABLED(CONFIG_IPV6)
1502 /* - key database */
1503 struct tcp_md5sig_key
{
1504 struct hlist_node node
;
1506 u8 family
; /* AF_INET or AF_INET6 */
1507 union tcp_md5_addr addr
;
1509 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1510 struct rcu_head rcu
;
1514 struct tcp_md5sig_info
{
1515 struct hlist_head head
;
1516 struct rcu_head rcu
;
1519 /* - pseudo header */
1520 struct tcp4_pseudohdr
{
1528 struct tcp6_pseudohdr
{
1529 struct in6_addr saddr
;
1530 struct in6_addr daddr
;
1532 __be32 protocol
; /* including padding */
1535 union tcp_md5sum_block
{
1536 struct tcp4_pseudohdr ip4
;
1537 #if IS_ENABLED(CONFIG_IPV6)
1538 struct tcp6_pseudohdr ip6
;
1542 /* - pool: digest algorithm, hash description and scratch buffer */
1543 struct tcp_md5sig_pool
{
1544 struct ahash_request
*md5_req
;
1549 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1550 const struct sock
*sk
, const struct sk_buff
*skb
);
1551 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1552 int family
, u8 prefixlen
, const u8
*newkey
, u8 newkeylen
,
1554 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1555 int family
, u8 prefixlen
);
1556 struct tcp_md5sig_key
*tcp_v4_md5_lookup(const struct sock
*sk
,
1557 const struct sock
*addr_sk
);
1559 #ifdef CONFIG_TCP_MD5SIG
1560 #include <linux/jump_label.h>
1561 extern struct static_key_false tcp_md5_needed
;
1562 struct tcp_md5sig_key
*__tcp_md5_do_lookup(const struct sock
*sk
,
1563 const union tcp_md5_addr
*addr
,
1565 static inline struct tcp_md5sig_key
*
1566 tcp_md5_do_lookup(const struct sock
*sk
,
1567 const union tcp_md5_addr
*addr
,
1570 if (!static_branch_unlikely(&tcp_md5_needed
))
1572 return __tcp_md5_do_lookup(sk
, addr
, family
);
1575 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1577 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(const struct sock
*sk
,
1578 const union tcp_md5_addr
*addr
,
1583 #define tcp_twsk_md5_key(twsk) NULL
1586 bool tcp_alloc_md5sig_pool(void);
1588 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1589 static inline void tcp_put_md5sig_pool(void)
1594 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1595 unsigned int header_len
);
1596 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1597 const struct tcp_md5sig_key
*key
);
1599 /* From tcp_fastopen.c */
1600 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1601 struct tcp_fastopen_cookie
*cookie
);
1602 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1603 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1605 struct tcp_fastopen_request
{
1606 /* Fast Open cookie. Size 0 means a cookie request */
1607 struct tcp_fastopen_cookie cookie
;
1608 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1610 int copied
; /* queued in tcp_connect() */
1611 struct ubuf_info
*uarg
;
1613 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1614 void tcp_fastopen_destroy_cipher(struct sock
*sk
);
1615 void tcp_fastopen_ctx_destroy(struct net
*net
);
1616 int tcp_fastopen_reset_cipher(struct net
*net
, struct sock
*sk
,
1617 void *primary_key
, void *backup_key
);
1618 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
);
1619 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1620 struct request_sock
*req
,
1621 struct tcp_fastopen_cookie
*foc
,
1622 const struct dst_entry
*dst
);
1623 void tcp_fastopen_init_key_once(struct net
*net
);
1624 bool tcp_fastopen_cookie_check(struct sock
*sk
, u16
*mss
,
1625 struct tcp_fastopen_cookie
*cookie
);
1626 bool tcp_fastopen_defer_connect(struct sock
*sk
, int *err
);
1627 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1628 #define TCP_FASTOPEN_KEY_MAX 2
1629 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1630 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1632 /* Fastopen key context */
1633 struct tcp_fastopen_context
{
1634 siphash_key_t key
[TCP_FASTOPEN_KEY_MAX
];
1636 struct rcu_head rcu
;
1639 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout
;
1640 void tcp_fastopen_active_disable(struct sock
*sk
);
1641 bool tcp_fastopen_active_should_disable(struct sock
*sk
);
1642 void tcp_fastopen_active_disable_ofo_check(struct sock
*sk
);
1643 void tcp_fastopen_active_detect_blackhole(struct sock
*sk
, bool expired
);
1645 /* Caller needs to wrap with rcu_read_(un)lock() */
1647 struct tcp_fastopen_context
*tcp_fastopen_get_ctx(const struct sock
*sk
)
1649 struct tcp_fastopen_context
*ctx
;
1651 ctx
= rcu_dereference(inet_csk(sk
)->icsk_accept_queue
.fastopenq
.ctx
);
1653 ctx
= rcu_dereference(sock_net(sk
)->ipv4
.tcp_fastopen_ctx
);
1658 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie
*foc
,
1659 const struct tcp_fastopen_cookie
*orig
)
1661 if (orig
->len
== TCP_FASTOPEN_COOKIE_SIZE
&&
1662 orig
->len
== foc
->len
&&
1663 !memcmp(orig
->val
, foc
->val
, foc
->len
))
1669 int tcp_fastopen_context_len(const struct tcp_fastopen_context
*ctx
)
1674 /* Latencies incurred by various limits for a sender. They are
1675 * chronograph-like stats that are mutually exclusive.
1679 TCP_CHRONO_BUSY
, /* Actively sending data (non-empty write queue) */
1680 TCP_CHRONO_RWND_LIMITED
, /* Stalled by insufficient receive window */
1681 TCP_CHRONO_SNDBUF_LIMITED
, /* Stalled by insufficient send buffer */
1685 void tcp_chrono_start(struct sock
*sk
, const enum tcp_chrono type
);
1686 void tcp_chrono_stop(struct sock
*sk
, const enum tcp_chrono type
);
1688 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1689 * the same memory storage than skb->destructor/_skb_refdst
1691 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff
*skb
)
1693 skb
->destructor
= NULL
;
1694 skb
->_skb_refdst
= 0UL;
1697 #define tcp_skb_tsorted_save(skb) { \
1698 unsigned long _save = skb->_skb_refdst; \
1699 skb->_skb_refdst = 0UL;
1701 #define tcp_skb_tsorted_restore(skb) \
1702 skb->_skb_refdst = _save; \
1705 void tcp_write_queue_purge(struct sock
*sk
);
1707 static inline struct sk_buff
*tcp_rtx_queue_head(const struct sock
*sk
)
1709 return skb_rb_first(&sk
->tcp_rtx_queue
);
1712 static inline struct sk_buff
*tcp_rtx_queue_tail(const struct sock
*sk
)
1714 return skb_rb_last(&sk
->tcp_rtx_queue
);
1717 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1719 return skb_peek(&sk
->sk_write_queue
);
1722 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1724 return skb_peek_tail(&sk
->sk_write_queue
);
1727 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1728 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1730 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1732 return skb_peek(&sk
->sk_write_queue
);
1735 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1736 const struct sk_buff
*skb
)
1738 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1741 static inline bool tcp_write_queue_empty(const struct sock
*sk
)
1743 return skb_queue_empty(&sk
->sk_write_queue
);
1746 static inline bool tcp_rtx_queue_empty(const struct sock
*sk
)
1748 return RB_EMPTY_ROOT(&sk
->tcp_rtx_queue
);
1751 static inline bool tcp_rtx_and_write_queues_empty(const struct sock
*sk
)
1753 return tcp_rtx_queue_empty(sk
) && tcp_write_queue_empty(sk
);
1756 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1758 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1760 /* Queue it, remembering where we must start sending. */
1761 if (sk
->sk_write_queue
.next
== skb
)
1762 tcp_chrono_start(sk
, TCP_CHRONO_BUSY
);
1765 /* Insert new before skb on the write queue of sk. */
1766 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1767 struct sk_buff
*skb
,
1770 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1773 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1775 tcp_skb_tsorted_anchor_cleanup(skb
);
1776 __skb_unlink(skb
, &sk
->sk_write_queue
);
1779 void tcp_rbtree_insert(struct rb_root
*root
, struct sk_buff
*skb
);
1781 static inline void tcp_rtx_queue_unlink(struct sk_buff
*skb
, struct sock
*sk
)
1783 tcp_skb_tsorted_anchor_cleanup(skb
);
1784 rb_erase(&skb
->rbnode
, &sk
->tcp_rtx_queue
);
1787 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff
*skb
, struct sock
*sk
)
1789 list_del(&skb
->tcp_tsorted_anchor
);
1790 tcp_rtx_queue_unlink(skb
, sk
);
1791 sk_wmem_free_skb(sk
, skb
);
1794 static inline void tcp_push_pending_frames(struct sock
*sk
)
1796 if (tcp_send_head(sk
)) {
1797 struct tcp_sock
*tp
= tcp_sk(sk
);
1799 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1803 /* Start sequence of the skb just after the highest skb with SACKed
1804 * bit, valid only if sacked_out > 0 or when the caller has ensured
1805 * validity by itself.
1807 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1809 if (!tp
->sacked_out
)
1812 if (tp
->highest_sack
== NULL
)
1815 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1818 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1820 tcp_sk(sk
)->highest_sack
= skb_rb_next(skb
);
1823 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1825 return tcp_sk(sk
)->highest_sack
;
1828 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1830 tcp_sk(sk
)->highest_sack
= tcp_rtx_queue_head(sk
);
1833 /* Called when old skb is about to be deleted and replaced by new skb */
1834 static inline void tcp_highest_sack_replace(struct sock
*sk
,
1835 struct sk_buff
*old
,
1836 struct sk_buff
*new)
1838 if (old
== tcp_highest_sack(sk
))
1839 tcp_sk(sk
)->highest_sack
= new;
1842 /* This helper checks if socket has IP_TRANSPARENT set */
1843 static inline bool inet_sk_transparent(const struct sock
*sk
)
1845 switch (sk
->sk_state
) {
1847 return inet_twsk(sk
)->tw_transparent
;
1848 case TCP_NEW_SYN_RECV
:
1849 return inet_rsk(inet_reqsk(sk
))->no_srccheck
;
1851 return inet_sk(sk
)->transparent
;
1854 /* Determines whether this is a thin stream (which may suffer from
1855 * increased latency). Used to trigger latency-reducing mechanisms.
1857 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1859 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1863 enum tcp_seq_states
{
1864 TCP_SEQ_STATE_LISTENING
,
1865 TCP_SEQ_STATE_ESTABLISHED
,
1868 void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
);
1869 void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
);
1870 void tcp_seq_stop(struct seq_file
*seq
, void *v
);
1872 struct tcp_seq_afinfo
{
1876 struct tcp_iter_state
{
1877 struct seq_net_private p
;
1878 enum tcp_seq_states state
;
1879 struct sock
*syn_wait_sk
;
1880 int bucket
, offset
, sbucket
, num
;
1884 extern struct request_sock_ops tcp_request_sock_ops
;
1885 extern struct request_sock_ops tcp6_request_sock_ops
;
1887 void tcp_v4_destroy_sock(struct sock
*sk
);
1889 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1890 netdev_features_t features
);
1891 struct sk_buff
*tcp_gro_receive(struct list_head
*head
, struct sk_buff
*skb
);
1892 int tcp_gro_complete(struct sk_buff
*skb
);
1894 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1896 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1898 struct net
*net
= sock_net((struct sock
*)tp
);
1899 return tp
->notsent_lowat
?: net
->ipv4
.sysctl_tcp_notsent_lowat
;
1902 /* @wake is one when sk_stream_write_space() calls us.
1903 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1904 * This mimics the strategy used in sock_def_write_space().
1906 static inline bool tcp_stream_memory_free(const struct sock
*sk
, int wake
)
1908 const struct tcp_sock
*tp
= tcp_sk(sk
);
1909 u32 notsent_bytes
= tp
->write_seq
- tp
->snd_nxt
;
1911 return (notsent_bytes
<< wake
) < tcp_notsent_lowat(tp
);
1914 #ifdef CONFIG_PROC_FS
1915 int tcp4_proc_init(void);
1916 void tcp4_proc_exit(void);
1919 int tcp_rtx_synack(const struct sock
*sk
, struct request_sock
*req
);
1920 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1921 const struct tcp_request_sock_ops
*af_ops
,
1922 struct sock
*sk
, struct sk_buff
*skb
);
1924 /* TCP af-specific functions */
1925 struct tcp_sock_af_ops
{
1926 #ifdef CONFIG_TCP_MD5SIG
1927 struct tcp_md5sig_key
*(*md5_lookup
) (const struct sock
*sk
,
1928 const struct sock
*addr_sk
);
1929 int (*calc_md5_hash
)(char *location
,
1930 const struct tcp_md5sig_key
*md5
,
1931 const struct sock
*sk
,
1932 const struct sk_buff
*skb
);
1933 int (*md5_parse
)(struct sock
*sk
,
1935 char __user
*optval
,
1940 struct tcp_request_sock_ops
{
1942 #ifdef CONFIG_TCP_MD5SIG
1943 struct tcp_md5sig_key
*(*req_md5_lookup
)(const struct sock
*sk
,
1944 const struct sock
*addr_sk
);
1945 int (*calc_md5_hash
) (char *location
,
1946 const struct tcp_md5sig_key
*md5
,
1947 const struct sock
*sk
,
1948 const struct sk_buff
*skb
);
1950 void (*init_req
)(struct request_sock
*req
,
1951 const struct sock
*sk_listener
,
1952 struct sk_buff
*skb
);
1953 #ifdef CONFIG_SYN_COOKIES
1954 __u32 (*cookie_init_seq
)(const struct sk_buff
*skb
,
1957 struct dst_entry
*(*route_req
)(const struct sock
*sk
, struct flowi
*fl
,
1958 const struct request_sock
*req
);
1959 u32 (*init_seq
)(const struct sk_buff
*skb
);
1960 u32 (*init_ts_off
)(const struct net
*net
, const struct sk_buff
*skb
);
1961 int (*send_synack
)(const struct sock
*sk
, struct dst_entry
*dst
,
1962 struct flowi
*fl
, struct request_sock
*req
,
1963 struct tcp_fastopen_cookie
*foc
,
1964 enum tcp_synack_type synack_type
);
1967 #ifdef CONFIG_SYN_COOKIES
1968 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1969 const struct sock
*sk
, struct sk_buff
*skb
,
1972 tcp_synq_overflow(sk
);
1973 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_SYNCOOKIESSENT
);
1974 return ops
->cookie_init_seq(skb
, mss
);
1977 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
1978 const struct sock
*sk
, struct sk_buff
*skb
,
1985 int tcpv4_offload_init(void);
1987 void tcp_v4_init(void);
1988 void tcp_init(void);
1990 /* tcp_recovery.c */
1991 void tcp_mark_skb_lost(struct sock
*sk
, struct sk_buff
*skb
);
1992 void tcp_newreno_mark_lost(struct sock
*sk
, bool snd_una_advanced
);
1993 extern s32
tcp_rack_skb_timeout(struct tcp_sock
*tp
, struct sk_buff
*skb
,
1995 extern void tcp_rack_mark_lost(struct sock
*sk
);
1996 extern void tcp_rack_advance(struct tcp_sock
*tp
, u8 sacked
, u32 end_seq
,
1998 extern void tcp_rack_reo_timeout(struct sock
*sk
);
1999 extern void tcp_rack_update_reo_wnd(struct sock
*sk
, struct rate_sample
*rs
);
2001 /* At how many usecs into the future should the RTO fire? */
2002 static inline s64
tcp_rto_delta_us(const struct sock
*sk
)
2004 const struct sk_buff
*skb
= tcp_rtx_queue_head(sk
);
2005 u32 rto
= inet_csk(sk
)->icsk_rto
;
2006 u64 rto_time_stamp_us
= tcp_skb_timestamp_us(skb
) + jiffies_to_usecs(rto
);
2008 return rto_time_stamp_us
- tcp_sk(sk
)->tcp_mstamp
;
2012 * Save and compile IPv4 options, return a pointer to it
2014 static inline struct ip_options_rcu
*tcp_v4_save_options(struct net
*net
,
2015 struct sk_buff
*skb
)
2017 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
2018 struct ip_options_rcu
*dopt
= NULL
;
2021 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
2023 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
2024 if (dopt
&& __ip_options_echo(net
, &dopt
->opt
, skb
, opt
)) {
2032 /* locally generated TCP pure ACKs have skb->truesize == 2
2033 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2034 * This is much faster than dissecting the packet to find out.
2035 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2037 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
2039 return skb
->truesize
== 2;
2042 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)
2047 static inline int tcp_inq(struct sock
*sk
)
2049 struct tcp_sock
*tp
= tcp_sk(sk
);
2052 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
2054 } else if (sock_flag(sk
, SOCK_URGINLINE
) ||
2056 before(tp
->urg_seq
, tp
->copied_seq
) ||
2057 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
2059 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
2061 /* Subtract 1, if FIN was received */
2062 if (answ
&& sock_flag(sk
, SOCK_DONE
))
2065 answ
= tp
->urg_seq
- tp
->copied_seq
;
2071 int tcp_peek_len(struct socket
*sock
);
2073 static inline void tcp_segs_in(struct tcp_sock
*tp
, const struct sk_buff
*skb
)
2077 segs_in
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
2078 tp
->segs_in
+= segs_in
;
2079 if (skb
->len
> tcp_hdrlen(skb
))
2080 tp
->data_segs_in
+= segs_in
;
2084 * TCP listen path runs lockless.
2085 * We forced "struct sock" to be const qualified to make sure
2086 * we don't modify one of its field by mistake.
2087 * Here, we increment sk_drops which is an atomic_t, so we can safely
2088 * make sock writable again.
2090 static inline void tcp_listendrop(const struct sock
*sk
)
2092 atomic_inc(&((struct sock
*)sk
)->sk_drops
);
2093 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
2096 enum hrtimer_restart
tcp_pace_kick(struct hrtimer
*timer
);
2099 * Interface for adding Upper Level Protocols over TCP
2102 #define TCP_ULP_NAME_MAX 16
2103 #define TCP_ULP_MAX 128
2104 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2106 struct tcp_ulp_ops
{
2107 struct list_head list
;
2109 /* initialize ulp */
2110 int (*init
)(struct sock
*sk
);
2112 void (*release
)(struct sock
*sk
);
2114 char name
[TCP_ULP_NAME_MAX
];
2115 struct module
*owner
;
2117 int tcp_register_ulp(struct tcp_ulp_ops
*type
);
2118 void tcp_unregister_ulp(struct tcp_ulp_ops
*type
);
2119 int tcp_set_ulp(struct sock
*sk
, const char *name
);
2120 void tcp_get_available_ulp(char *buf
, size_t len
);
2121 void tcp_cleanup_ulp(struct sock
*sk
);
2123 #define MODULE_ALIAS_TCP_ULP(name) \
2124 __MODULE_INFO(alias, alias_userspace, name); \
2125 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2130 int tcp_bpf_init(struct sock
*sk
);
2131 void tcp_bpf_reinit(struct sock
*sk
);
2132 int tcp_bpf_sendmsg_redir(struct sock
*sk
, struct sk_msg
*msg
, u32 bytes
,
2134 int tcp_bpf_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
,
2135 int nonblock
, int flags
, int *addr_len
);
2136 int __tcp_bpf_recvmsg(struct sock
*sk
, struct sk_psock
*psock
,
2137 struct msghdr
*msg
, int len
, int flags
);
2139 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2140 * is < 0, then the BPF op failed (for example if the loaded BPF
2141 * program does not support the chosen operation or there is no BPF
2145 static inline int tcp_call_bpf(struct sock
*sk
, int op
, u32 nargs
, u32
*args
)
2147 struct bpf_sock_ops_kern sock_ops
;
2150 memset(&sock_ops
, 0, offsetof(struct bpf_sock_ops_kern
, temp
));
2151 if (sk_fullsock(sk
)) {
2152 sock_ops
.is_fullsock
= 1;
2153 sock_owned_by_me(sk
);
2159 memcpy(sock_ops
.args
, args
, nargs
* sizeof(*args
));
2161 ret
= BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops
);
2163 ret
= sock_ops
.reply
;
2169 static inline int tcp_call_bpf_2arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
)
2171 u32 args
[2] = {arg1
, arg2
};
2173 return tcp_call_bpf(sk
, op
, 2, args
);
2176 static inline int tcp_call_bpf_3arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
,
2179 u32 args
[3] = {arg1
, arg2
, arg3
};
2181 return tcp_call_bpf(sk
, op
, 3, args
);
2185 static inline int tcp_call_bpf(struct sock
*sk
, int op
, u32 nargs
, u32
*args
)
2190 static inline int tcp_call_bpf_2arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
)
2195 static inline int tcp_call_bpf_3arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
,
2203 static inline u32
tcp_timeout_init(struct sock
*sk
)
2207 timeout
= tcp_call_bpf(sk
, BPF_SOCK_OPS_TIMEOUT_INIT
, 0, NULL
);
2210 timeout
= TCP_TIMEOUT_INIT
;
2214 static inline u32
tcp_rwnd_init_bpf(struct sock
*sk
)
2218 rwnd
= tcp_call_bpf(sk
, BPF_SOCK_OPS_RWND_INIT
, 0, NULL
);
2225 static inline bool tcp_bpf_ca_needs_ecn(struct sock
*sk
)
2227 return (tcp_call_bpf(sk
, BPF_SOCK_OPS_NEEDS_ECN
, 0, NULL
) == 1);
2230 static inline void tcp_bpf_rtt(struct sock
*sk
)
2232 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk
), BPF_SOCK_OPS_RTT_CB_FLAG
))
2233 tcp_call_bpf(sk
, BPF_SOCK_OPS_RTT_CB
, 0, NULL
);
2236 #if IS_ENABLED(CONFIG_SMC)
2237 extern struct static_key_false tcp_have_smc
;
2240 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2241 void clean_acked_data_enable(struct inet_connection_sock
*icsk
,
2242 void (*cad
)(struct sock
*sk
, u32 ack_seq
));
2243 void clean_acked_data_disable(struct inet_connection_sock
*icsk
);
2244 void clean_acked_data_flush(void);
2247 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled
);
2248 static inline void tcp_add_tx_delay(struct sk_buff
*skb
,
2249 const struct tcp_sock
*tp
)
2251 if (static_branch_unlikely(&tcp_tx_delay_enabled
))
2252 skb
->skb_mstamp_ns
+= (u64
)tp
->tcp_tx_delay
* NSEC_PER_USEC
;
2255 /* Compute Earliest Departure Time for some control packets
2256 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2258 static inline u64
tcp_transmit_time(const struct sock
*sk
)
2260 if (static_branch_unlikely(&tcp_tx_delay_enabled
)) {
2261 u32 delay
= (sk
->sk_state
== TCP_TIME_WAIT
) ?
2262 tcp_twsk(sk
)->tw_tx_delay
: tcp_sk(sk
)->tcp_tx_delay
;
2264 return tcp_clock_ns() + (u64
)delay
* NSEC_PER_USEC
;