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>
42 #include <net/mptcp.h>
44 #include <linux/seq_file.h>
45 #include <linux/memcontrol.h>
46 #include <linux/bpf-cgroup.h>
47 #include <linux/siphash.h>
49 extern struct inet_hashinfo tcp_hashinfo
;
51 extern struct percpu_counter tcp_orphan_count
;
52 void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
54 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
56 #define TCP_MIN_SND_MSS 48
57 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
60 * Never offer a window over 32767 without using window scaling. Some
61 * poor stacks do signed 16bit maths!
63 #define MAX_TCP_WINDOW 32767U
65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66 #define TCP_MIN_MSS 88U
68 /* The initial MTU to use for probing */
69 #define TCP_BASE_MSS 1024
71 /* probing interval, default to 10 minutes as per RFC4821 */
72 #define TCP_PROBE_INTERVAL 600
74 /* Specify interval when tcp mtu probing will stop */
75 #define TCP_PROBE_THRESHOLD 8
77 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
78 #define TCP_FASTRETRANS_THRESH 3
80 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
81 #define TCP_MAX_QUICKACKS 16U
83 /* Maximal number of window scale according to RFC1323 */
84 #define TCP_MAX_WSCALE 14U
87 #define TCP_URG_VALID 0x0100
88 #define TCP_URG_NOTYET 0x0200
89 #define TCP_URG_READ 0x0400
91 #define TCP_RETR1 3 /*
92 * This is how many retries it does before it
93 * tries to figure out if the gateway is
94 * down. Minimal RFC value is 3; it corresponds
95 * to ~3sec-8min depending on RTO.
98 #define TCP_RETR2 15 /*
99 * This should take at least
100 * 90 minutes to time out.
101 * RFC1122 says that the limit is 100 sec.
102 * 15 is ~13-30min depending on RTO.
105 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
106 * when active opening a connection.
107 * RFC1122 says the minimum retry MUST
108 * be at least 180secs. Nevertheless
109 * this value is corresponding to
110 * 63secs of retransmission with the
111 * current initial RTO.
114 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
115 * when passive opening a connection.
116 * This is corresponding to 31secs of
117 * retransmission with the current
121 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
122 * state, about 60 seconds */
123 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
124 /* BSD style FIN_WAIT2 deadlock breaker.
125 * It used to be 3min, new value is 60sec,
126 * to combine FIN-WAIT-2 timeout with
130 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
132 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
133 #define TCP_ATO_MIN ((unsigned)(HZ/25))
135 #define TCP_DELACK_MIN 4U
136 #define TCP_ATO_MIN 4U
138 #define TCP_RTO_MAX ((unsigned)(120*HZ))
139 #define TCP_RTO_MIN ((unsigned)(HZ/5))
140 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
141 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
142 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
143 * used as a fallback RTO for the
144 * initial data transmission if no
145 * valid RTT sample has been acquired,
146 * most likely due to retrans in 3WHS.
149 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
150 * for local resources.
152 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
153 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
154 #define TCP_KEEPALIVE_INTVL (75*HZ)
156 #define MAX_TCP_KEEPIDLE 32767
157 #define MAX_TCP_KEEPINTVL 32767
158 #define MAX_TCP_KEEPCNT 127
159 #define MAX_TCP_SYNCNT 127
161 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
163 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
164 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
165 * after this time. It should be equal
166 * (or greater than) TCP_TIMEWAIT_LEN
167 * to provide reliability equal to one
168 * provided by timewait state.
170 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
171 * timestamps. It must be less than
172 * minimal timewait lifetime.
178 #define TCPOPT_NOP 1 /* Padding */
179 #define TCPOPT_EOL 0 /* End of options */
180 #define TCPOPT_MSS 2 /* Segment size negotiating */
181 #define TCPOPT_WINDOW 3 /* Window scaling */
182 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
183 #define TCPOPT_SACK 5 /* SACK Block */
184 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
185 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
186 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
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 */
250 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
252 extern atomic_long_t tcp_memory_allocated
;
253 extern struct percpu_counter tcp_sockets_allocated
;
254 extern unsigned long tcp_memory_pressure
;
256 /* optimized version of sk_under_memory_pressure() for TCP sockets */
257 static inline bool tcp_under_memory_pressure(const struct sock
*sk
)
259 if (mem_cgroup_sockets_enabled
&& sk
->sk_memcg
&&
260 mem_cgroup_under_socket_pressure(sk
->sk_memcg
))
263 return READ_ONCE(tcp_memory_pressure
);
266 * The next routines deal with comparing 32 bit unsigned ints
267 * and worry about wraparound (automatic with unsigned arithmetic).
270 static inline bool before(__u32 seq1
, __u32 seq2
)
272 return (__s32
)(seq1
-seq2
) < 0;
274 #define after(seq2, seq1) before(seq1, seq2)
276 /* is s2<=s1<=s3 ? */
277 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
279 return seq3
- seq2
>= seq1
- seq2
;
282 static inline bool tcp_out_of_memory(struct sock
*sk
)
284 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
285 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
290 void sk_forced_mem_schedule(struct sock
*sk
, int size
);
292 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
294 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
295 int orphans
= percpu_counter_read_positive(ocp
);
297 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
298 orphans
= percpu_counter_sum_positive(ocp
);
299 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
305 bool tcp_check_oom(struct sock
*sk
, int shift
);
308 extern struct proto tcp_prot
;
310 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
311 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
312 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
313 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
315 void tcp_tasklet_init(void);
317 int tcp_v4_err(struct sk_buff
*skb
, u32
);
319 void tcp_shutdown(struct sock
*sk
, int how
);
321 int tcp_v4_early_demux(struct sk_buff
*skb
);
322 int tcp_v4_rcv(struct sk_buff
*skb
);
324 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
325 int tcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
326 int tcp_sendmsg_locked(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
327 int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
, size_t size
,
329 int tcp_sendpage_locked(struct sock
*sk
, struct page
*page
, int offset
,
330 size_t size
, int flags
);
331 ssize_t
do_tcp_sendpages(struct sock
*sk
, struct page
*page
, int offset
,
332 size_t size
, int flags
);
333 int tcp_send_mss(struct sock
*sk
, int *size_goal
, int flags
);
334 void tcp_push(struct sock
*sk
, int flags
, int mss_now
, int nonagle
,
336 void tcp_release_cb(struct sock
*sk
);
337 void tcp_wfree(struct sk_buff
*skb
);
338 void tcp_write_timer_handler(struct sock
*sk
);
339 void tcp_delack_timer_handler(struct sock
*sk
);
340 int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
341 int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
);
342 void tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
);
343 void tcp_rcv_space_adjust(struct sock
*sk
);
344 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
345 void tcp_twsk_destructor(struct sock
*sk
);
346 ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
347 struct pipe_inode_info
*pipe
, size_t len
,
350 void tcp_enter_quickack_mode(struct sock
*sk
, unsigned int max_quickacks
);
351 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
352 const unsigned int pkts
)
354 struct inet_connection_sock
*icsk
= inet_csk(sk
);
356 if (icsk
->icsk_ack
.quick
) {
357 if (pkts
>= icsk
->icsk_ack
.quick
) {
358 icsk
->icsk_ack
.quick
= 0;
359 /* Leaving quickack mode we deflate ATO. */
360 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
362 icsk
->icsk_ack
.quick
-= pkts
;
367 #define TCP_ECN_QUEUE_CWR 2
368 #define TCP_ECN_DEMAND_CWR 4
369 #define TCP_ECN_SEEN 8
379 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
381 const struct tcphdr
*th
);
382 struct sock
*tcp_check_req(struct sock
*sk
, struct sk_buff
*skb
,
383 struct request_sock
*req
, bool fastopen
,
385 int tcp_child_process(struct sock
*parent
, struct sock
*child
,
386 struct sk_buff
*skb
);
387 void tcp_enter_loss(struct sock
*sk
);
388 void tcp_cwnd_reduction(struct sock
*sk
, int newly_acked_sacked
, int flag
);
389 void tcp_clear_retrans(struct tcp_sock
*tp
);
390 void tcp_update_metrics(struct sock
*sk
);
391 void tcp_init_metrics(struct sock
*sk
);
392 void tcp_metrics_init(void);
393 bool tcp_peer_is_proven(struct request_sock
*req
, struct dst_entry
*dst
);
394 void tcp_close(struct sock
*sk
, long timeout
);
395 void tcp_init_sock(struct sock
*sk
);
396 void tcp_init_transfer(struct sock
*sk
, int bpf_op
);
397 __poll_t
tcp_poll(struct file
*file
, struct socket
*sock
,
398 struct poll_table_struct
*wait
);
399 int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
400 char __user
*optval
, int __user
*optlen
);
401 int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
402 char __user
*optval
, unsigned int optlen
);
403 int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
404 char __user
*optval
, int __user
*optlen
);
405 int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
406 char __user
*optval
, unsigned int optlen
);
407 void tcp_set_keepalive(struct sock
*sk
, int val
);
408 void tcp_syn_ack_timeout(const struct request_sock
*req
);
409 int tcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
410 int flags
, int *addr_len
);
411 int tcp_set_rcvlowat(struct sock
*sk
, int val
);
412 void tcp_data_ready(struct sock
*sk
);
414 int tcp_mmap(struct file
*file
, struct socket
*sock
,
415 struct vm_area_struct
*vma
);
417 void tcp_parse_options(const struct net
*net
, const struct sk_buff
*skb
,
418 struct tcp_options_received
*opt_rx
,
419 int estab
, struct tcp_fastopen_cookie
*foc
);
420 const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
423 * BPF SKB-less helpers
425 u16
tcp_v4_get_syncookie(struct sock
*sk
, struct iphdr
*iph
,
426 struct tcphdr
*th
, u32
*cookie
);
427 u16
tcp_v6_get_syncookie(struct sock
*sk
, struct ipv6hdr
*iph
,
428 struct tcphdr
*th
, u32
*cookie
);
429 u16
tcp_get_syncookie_mss(struct request_sock_ops
*rsk_ops
,
430 const struct tcp_request_sock_ops
*af_ops
,
431 struct sock
*sk
, struct tcphdr
*th
);
433 * TCP v4 functions exported for the inet6 API
436 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
437 void tcp_v4_mtu_reduced(struct sock
*sk
);
438 void tcp_req_err(struct sock
*sk
, u32 seq
, bool abort
);
439 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
440 struct sock
*tcp_create_openreq_child(const struct sock
*sk
,
441 struct request_sock
*req
,
442 struct sk_buff
*skb
);
443 void tcp_ca_openreq_child(struct sock
*sk
, const struct dst_entry
*dst
);
444 struct sock
*tcp_v4_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
445 struct request_sock
*req
,
446 struct dst_entry
*dst
,
447 struct request_sock
*req_unhash
,
449 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
450 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
);
451 int tcp_connect(struct sock
*sk
);
452 enum tcp_synack_type
{
457 struct sk_buff
*tcp_make_synack(const struct sock
*sk
, struct dst_entry
*dst
,
458 struct request_sock
*req
,
459 struct tcp_fastopen_cookie
*foc
,
460 enum tcp_synack_type synack_type
);
461 int tcp_disconnect(struct sock
*sk
, int flags
);
463 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
464 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
465 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
);
467 /* From syncookies.c */
468 struct sock
*tcp_get_cookie_sock(struct sock
*sk
, struct sk_buff
*skb
,
469 struct request_sock
*req
,
470 struct dst_entry
*dst
, u32 tsoff
);
471 int __cookie_v4_check(const struct iphdr
*iph
, const struct tcphdr
*th
,
473 struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
);
474 #ifdef CONFIG_SYN_COOKIES
476 /* Syncookies use a monotonic timer which increments every 60 seconds.
477 * This counter is used both as a hash input and partially encoded into
478 * the cookie value. A cookie is only validated further if the delta
479 * between the current counter value and the encoded one is less than this,
480 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
481 * the counter advances immediately after a cookie is generated).
483 #define MAX_SYNCOOKIE_AGE 2
484 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
485 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
487 /* syncookies: remember time of last synqueue overflow
488 * But do not dirty this field too often (once per second is enough)
489 * It is racy as we do not hold a lock, but race is very minor.
491 static inline void tcp_synq_overflow(const struct sock
*sk
)
493 unsigned int last_overflow
;
494 unsigned int now
= jiffies
;
496 if (sk
->sk_reuseport
) {
497 struct sock_reuseport
*reuse
;
499 reuse
= rcu_dereference(sk
->sk_reuseport_cb
);
501 last_overflow
= READ_ONCE(reuse
->synq_overflow_ts
);
502 if (!time_between32(now
, last_overflow
,
504 WRITE_ONCE(reuse
->synq_overflow_ts
, now
);
509 last_overflow
= READ_ONCE(tcp_sk(sk
)->rx_opt
.ts_recent_stamp
);
510 if (!time_between32(now
, last_overflow
, last_overflow
+ HZ
))
511 WRITE_ONCE(tcp_sk(sk
)->rx_opt
.ts_recent_stamp
, now
);
514 /* syncookies: no recent synqueue overflow on this listening socket? */
515 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
517 unsigned int last_overflow
;
518 unsigned int now
= jiffies
;
520 if (sk
->sk_reuseport
) {
521 struct sock_reuseport
*reuse
;
523 reuse
= rcu_dereference(sk
->sk_reuseport_cb
);
525 last_overflow
= READ_ONCE(reuse
->synq_overflow_ts
);
526 return !time_between32(now
, last_overflow
- HZ
,
528 TCP_SYNCOOKIE_VALID
);
532 last_overflow
= READ_ONCE(tcp_sk(sk
)->rx_opt
.ts_recent_stamp
);
534 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
535 * then we're under synflood. However, we have to use
536 * 'last_overflow - HZ' as lower bound. That's because a concurrent
537 * tcp_synq_overflow() could update .ts_recent_stamp after we read
538 * jiffies but before we store .ts_recent_stamp into last_overflow,
539 * which could lead to rejecting a valid syncookie.
541 return !time_between32(now
, last_overflow
- HZ
,
542 last_overflow
+ TCP_SYNCOOKIE_VALID
);
545 static inline u32
tcp_cookie_time(void)
547 u64 val
= get_jiffies_64();
549 do_div(val
, TCP_SYNCOOKIE_PERIOD
);
553 u32
__cookie_v4_init_sequence(const struct iphdr
*iph
, const struct tcphdr
*th
,
555 __u32
cookie_v4_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
556 u64
cookie_init_timestamp(struct request_sock
*req
, u64 now
);
557 bool cookie_timestamp_decode(const struct net
*net
,
558 struct tcp_options_received
*opt
);
559 bool cookie_ecn_ok(const struct tcp_options_received
*opt
,
560 const struct net
*net
, const struct dst_entry
*dst
);
562 /* From net/ipv6/syncookies.c */
563 int __cookie_v6_check(const struct ipv6hdr
*iph
, const struct tcphdr
*th
,
565 struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
567 u32
__cookie_v6_init_sequence(const struct ipv6hdr
*iph
,
568 const struct tcphdr
*th
, u16
*mssp
);
569 __u32
cookie_v6_init_sequence(const struct sk_buff
*skb
, __u16
*mss
);
573 void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
575 int __tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
576 int tcp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
, int segs
);
577 void tcp_retransmit_timer(struct sock
*sk
);
578 void tcp_xmit_retransmit_queue(struct sock
*);
579 void tcp_simple_retransmit(struct sock
*);
580 void tcp_enter_recovery(struct sock
*sk
, bool ece_ack
);
581 int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
583 TCP_FRAG_IN_WRITE_QUEUE
,
584 TCP_FRAG_IN_RTX_QUEUE
,
586 int tcp_fragment(struct sock
*sk
, enum tcp_queue tcp_queue
,
587 struct sk_buff
*skb
, u32 len
,
588 unsigned int mss_now
, gfp_t gfp
);
590 void tcp_send_probe0(struct sock
*);
591 void tcp_send_partial(struct sock
*);
592 int tcp_write_wakeup(struct sock
*, int mib
);
593 void tcp_send_fin(struct sock
*sk
);
594 void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
595 int tcp_send_synack(struct sock
*);
596 void tcp_push_one(struct sock
*, unsigned int mss_now
);
597 void __tcp_send_ack(struct sock
*sk
, u32 rcv_nxt
);
598 void tcp_send_ack(struct sock
*sk
);
599 void tcp_send_delayed_ack(struct sock
*sk
);
600 void tcp_send_loss_probe(struct sock
*sk
);
601 bool tcp_schedule_loss_probe(struct sock
*sk
, bool advancing_rto
);
602 void tcp_skb_collapse_tstamp(struct sk_buff
*skb
,
603 const struct sk_buff
*next_skb
);
606 void tcp_rearm_rto(struct sock
*sk
);
607 void tcp_synack_rtt_meas(struct sock
*sk
, struct request_sock
*req
);
608 void tcp_reset(struct sock
*sk
);
609 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock
*tp
, struct sk_buff
*skb
);
610 void tcp_fin(struct sock
*sk
);
613 void tcp_init_xmit_timers(struct sock
*);
614 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
616 if (hrtimer_try_to_cancel(&tcp_sk(sk
)->pacing_timer
) == 1)
619 if (hrtimer_try_to_cancel(&tcp_sk(sk
)->compressed_ack_timer
) == 1)
622 inet_csk_clear_xmit_timers(sk
);
625 unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
626 unsigned int tcp_current_mss(struct sock
*sk
);
628 /* Bound MSS / TSO packet size with the half of the window */
629 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
633 /* When peer uses tiny windows, there is no use in packetizing
634 * to sub-MSS pieces for the sake of SWS or making sure there
635 * are enough packets in the pipe for fast recovery.
637 * On the other hand, for extremely large MSS devices, handling
638 * smaller than MSS windows in this way does make sense.
640 if (tp
->max_window
> TCP_MSS_DEFAULT
)
641 cutoff
= (tp
->max_window
>> 1);
643 cutoff
= tp
->max_window
;
645 if (cutoff
&& pktsize
> cutoff
)
646 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
652 void tcp_get_info(struct sock
*, struct tcp_info
*);
654 /* Read 'sendfile()'-style from a TCP socket */
655 int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
656 sk_read_actor_t recv_actor
);
658 void tcp_initialize_rcv_mss(struct sock
*sk
);
660 int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
661 int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
662 void tcp_mtup_init(struct sock
*sk
);
663 void tcp_init_buffer_space(struct sock
*sk
);
665 static inline void tcp_bound_rto(const struct sock
*sk
)
667 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
668 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
671 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
673 return usecs_to_jiffies((tp
->srtt_us
>> 3) + tp
->rttvar_us
);
676 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
678 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
679 ntohl(TCP_FLAG_ACK
) |
683 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
685 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
688 static inline void tcp_fast_path_check(struct sock
*sk
)
690 struct tcp_sock
*tp
= tcp_sk(sk
);
692 if (RB_EMPTY_ROOT(&tp
->out_of_order_queue
) &&
694 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
696 tcp_fast_path_on(tp
);
699 /* Compute the actual rto_min value */
700 static inline u32
tcp_rto_min(struct sock
*sk
)
702 const struct dst_entry
*dst
= __sk_dst_get(sk
);
703 u32 rto_min
= TCP_RTO_MIN
;
705 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
706 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
710 static inline u32
tcp_rto_min_us(struct sock
*sk
)
712 return jiffies_to_usecs(tcp_rto_min(sk
));
715 static inline bool tcp_ca_dst_locked(const struct dst_entry
*dst
)
717 return dst_metric_locked(dst
, RTAX_CC_ALGO
);
720 /* Minimum RTT in usec. ~0 means not available. */
721 static inline u32
tcp_min_rtt(const struct tcp_sock
*tp
)
723 return minmax_get(&tp
->rtt_min
);
726 /* Compute the actual receive window we are currently advertising.
727 * Rcv_nxt can be after the window if our peer push more data
728 * than the offered window.
730 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
732 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
739 /* Choose a new window, without checks for shrinking, and without
740 * scaling applied to the result. The caller does these things
741 * if necessary. This is a "raw" window selection.
743 u32
__tcp_select_window(struct sock
*sk
);
745 void tcp_send_window_probe(struct sock
*sk
);
747 /* TCP uses 32bit jiffies to save some space.
748 * Note that this is different from tcp_time_stamp, which
749 * historically has been the same until linux-4.13.
751 #define tcp_jiffies32 ((u32)jiffies)
754 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
755 * It is no longer tied to jiffies, but to 1 ms clock.
756 * Note: double check if you want to use tcp_jiffies32 instead of this.
758 #define TCP_TS_HZ 1000
760 static inline u64
tcp_clock_ns(void)
762 return ktime_get_ns();
765 static inline u64
tcp_clock_us(void)
767 return div_u64(tcp_clock_ns(), NSEC_PER_USEC
);
770 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
771 static inline u32
tcp_time_stamp(const struct tcp_sock
*tp
)
773 return div_u64(tp
->tcp_mstamp
, USEC_PER_SEC
/ TCP_TS_HZ
);
776 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
777 static inline u32
tcp_ns_to_ts(u64 ns
)
779 return div_u64(ns
, NSEC_PER_SEC
/ TCP_TS_HZ
);
782 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
783 static inline u32
tcp_time_stamp_raw(void)
785 return tcp_ns_to_ts(tcp_clock_ns());
788 void tcp_mstamp_refresh(struct tcp_sock
*tp
);
790 static inline u32
tcp_stamp_us_delta(u64 t1
, u64 t0
)
792 return max_t(s64
, t1
- t0
, 0);
795 static inline u32
tcp_skb_timestamp(const struct sk_buff
*skb
)
797 return tcp_ns_to_ts(skb
->skb_mstamp_ns
);
800 /* provide the departure time in us unit */
801 static inline u64
tcp_skb_timestamp_us(const struct sk_buff
*skb
)
803 return div_u64(skb
->skb_mstamp_ns
, NSEC_PER_USEC
);
807 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
809 #define TCPHDR_FIN 0x01
810 #define TCPHDR_SYN 0x02
811 #define TCPHDR_RST 0x04
812 #define TCPHDR_PSH 0x08
813 #define TCPHDR_ACK 0x10
814 #define TCPHDR_URG 0x20
815 #define TCPHDR_ECE 0x40
816 #define TCPHDR_CWR 0x80
818 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
820 /* This is what the send packet queuing engine uses to pass
821 * TCP per-packet control information to the transmission code.
822 * We also store the host-order sequence numbers in here too.
823 * This is 44 bytes if IPV6 is enabled.
824 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
827 __u32 seq
; /* Starting sequence number */
828 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
830 /* Note : tcp_tw_isn is used in input path only
831 * (isn chosen by tcp_timewait_state_process())
833 * tcp_gso_segs/size are used in write queue only,
834 * cf tcp_skb_pcount()/tcp_skb_mss()
842 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
844 __u8 sacked
; /* State flags for SACK. */
845 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
846 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
847 #define TCPCB_LOST 0x04 /* SKB is lost */
848 #define TCPCB_TAGBITS 0x07 /* All tag bits */
849 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
850 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
851 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
854 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
855 __u8 txstamp_ack
:1, /* Record TX timestamp for ack? */
856 eor
:1, /* Is skb MSG_EOR marked? */
857 has_rxtstamp
:1, /* SKB has a RX timestamp */
859 __u32 ack_seq
; /* Sequence number ACK'd */
862 /* There is space for up to 24 bytes */
863 __u32 in_flight
:30,/* Bytes in flight at transmit */
864 is_app_limited
:1, /* cwnd not fully used? */
866 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
868 /* start of send pipeline phase */
870 /* when we reached the "delivered" count */
871 u64 delivered_mstamp
;
872 } tx
; /* only used for outgoing skbs */
874 struct inet_skb_parm h4
;
875 #if IS_ENABLED(CONFIG_IPV6)
876 struct inet6_skb_parm h6
;
878 } header
; /* For incoming skbs */
881 struct sock
*sk_redir
;
887 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
889 static inline void bpf_compute_data_end_sk_skb(struct sk_buff
*skb
)
891 TCP_SKB_CB(skb
)->bpf
.data_end
= skb
->data
+ skb_headlen(skb
);
894 static inline bool tcp_skb_bpf_ingress(const struct sk_buff
*skb
)
896 return TCP_SKB_CB(skb
)->bpf
.flags
& BPF_F_INGRESS
;
899 static inline struct sock
*tcp_skb_bpf_redirect_fetch(struct sk_buff
*skb
)
901 return TCP_SKB_CB(skb
)->bpf
.sk_redir
;
904 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff
*skb
)
906 TCP_SKB_CB(skb
)->bpf
.sk_redir
= NULL
;
909 #if IS_ENABLED(CONFIG_IPV6)
910 /* This is the variant of inet6_iif() that must be used by TCP,
911 * as TCP moves IP6CB into a different location in skb->cb[]
913 static inline int tcp_v6_iif(const struct sk_buff
*skb
)
915 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
918 static inline int tcp_v6_iif_l3_slave(const struct sk_buff
*skb
)
920 bool l3_slave
= ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
);
922 return l3_slave
? skb
->skb_iif
: TCP_SKB_CB(skb
)->header
.h6
.iif
;
925 /* TCP_SKB_CB reference means this can not be used from early demux */
926 static inline int tcp_v6_sdif(const struct sk_buff
*skb
)
928 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
929 if (skb
&& ipv6_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h6
.flags
))
930 return TCP_SKB_CB(skb
)->header
.h6
.iif
;
936 static inline bool inet_exact_dif_match(struct net
*net
, struct sk_buff
*skb
)
938 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
939 if (!net
->ipv4
.sysctl_tcp_l3mdev_accept
&&
940 skb
&& ipv4_l3mdev_skb(IPCB(skb
)->flags
))
946 /* TCP_SKB_CB reference means this can not be used from early demux */
947 static inline int tcp_v4_sdif(struct sk_buff
*skb
)
949 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
950 if (skb
&& ipv4_l3mdev_skb(TCP_SKB_CB(skb
)->header
.h4
.flags
))
951 return TCP_SKB_CB(skb
)->header
.h4
.iif
;
956 /* Due to TSO, an SKB can be composed of multiple actual
957 * packets. To keep these tracked properly, we use this.
959 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
961 return TCP_SKB_CB(skb
)->tcp_gso_segs
;
964 static inline void tcp_skb_pcount_set(struct sk_buff
*skb
, int segs
)
966 TCP_SKB_CB(skb
)->tcp_gso_segs
= segs
;
969 static inline void tcp_skb_pcount_add(struct sk_buff
*skb
, int segs
)
971 TCP_SKB_CB(skb
)->tcp_gso_segs
+= segs
;
974 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
975 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
977 return TCP_SKB_CB(skb
)->tcp_gso_size
;
980 static inline bool tcp_skb_can_collapse_to(const struct sk_buff
*skb
)
982 return likely(!TCP_SKB_CB(skb
)->eor
);
985 static inline bool tcp_skb_can_collapse(const struct sk_buff
*to
,
986 const struct sk_buff
*from
)
988 return likely(tcp_skb_can_collapse_to(to
) &&
989 mptcp_skb_can_collapse(to
, from
));
992 /* Events passed to congestion control interface */
994 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
995 CA_EVENT_CWND_RESTART
, /* congestion window restart */
996 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
997 CA_EVENT_LOSS
, /* loss timeout */
998 CA_EVENT_ECN_NO_CE
, /* ECT set, but not CE marked */
999 CA_EVENT_ECN_IS_CE
, /* received CE marked IP packet */
1002 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1003 enum tcp_ca_ack_event_flags
{
1004 CA_ACK_SLOWPATH
= (1 << 0), /* In slow path processing */
1005 CA_ACK_WIN_UPDATE
= (1 << 1), /* ACK updated window */
1006 CA_ACK_ECE
= (1 << 2), /* ECE bit is set on ack */
1010 * Interface for adding new TCP congestion control handlers
1012 #define TCP_CA_NAME_MAX 16
1013 #define TCP_CA_MAX 128
1014 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1016 #define TCP_CA_UNSPEC 0
1018 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1019 #define TCP_CONG_NON_RESTRICTED 0x1
1020 /* Requires ECN/ECT set on all packets */
1021 #define TCP_CONG_NEEDS_ECN 0x2
1022 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1032 /* A rate sample measures the number of (original/retransmitted) data
1033 * packets delivered "delivered" over an interval of time "interval_us".
1034 * The tcp_rate.c code fills in the rate sample, and congestion
1035 * control modules that define a cong_control function to run at the end
1036 * of ACK processing can optionally chose to consult this sample when
1037 * setting cwnd and pacing rate.
1038 * A sample is invalid if "delivered" or "interval_us" is negative.
1040 struct rate_sample
{
1041 u64 prior_mstamp
; /* starting timestamp for interval */
1042 u32 prior_delivered
; /* tp->delivered at "prior_mstamp" */
1043 s32 delivered
; /* number of packets delivered over interval */
1044 long interval_us
; /* time for tp->delivered to incr "delivered" */
1045 u32 snd_interval_us
; /* snd interval for delivered packets */
1046 u32 rcv_interval_us
; /* rcv interval for delivered packets */
1047 long rtt_us
; /* RTT of last (S)ACKed packet (or -1) */
1048 int losses
; /* number of packets marked lost upon ACK */
1049 u32 acked_sacked
; /* number of packets newly (S)ACKed upon ACK */
1050 u32 prior_in_flight
; /* in flight before this ACK */
1051 bool is_app_limited
; /* is sample from packet with bubble in pipe? */
1052 bool is_retrans
; /* is sample from retransmission? */
1053 bool is_ack_delayed
; /* is this (likely) a delayed ACK? */
1056 struct tcp_congestion_ops
{
1057 struct list_head list
;
1061 /* initialize private data (optional) */
1062 void (*init
)(struct sock
*sk
);
1063 /* cleanup private data (optional) */
1064 void (*release
)(struct sock
*sk
);
1066 /* return slow start threshold (required) */
1067 u32 (*ssthresh
)(struct sock
*sk
);
1068 /* do new cwnd calculation (required) */
1069 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 acked
);
1070 /* call before changing ca_state (optional) */
1071 void (*set_state
)(struct sock
*sk
, u8 new_state
);
1072 /* call when cwnd event occurs (optional) */
1073 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
1074 /* call when ack arrives (optional) */
1075 void (*in_ack_event
)(struct sock
*sk
, u32 flags
);
1076 /* new value of cwnd after loss (required) */
1077 u32 (*undo_cwnd
)(struct sock
*sk
);
1078 /* hook for packet ack accounting (optional) */
1079 void (*pkts_acked
)(struct sock
*sk
, const struct ack_sample
*sample
);
1080 /* override sysctl_tcp_min_tso_segs */
1081 u32 (*min_tso_segs
)(struct sock
*sk
);
1082 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1083 u32 (*sndbuf_expand
)(struct sock
*sk
);
1084 /* call when packets are delivered to update cwnd and pacing rate,
1085 * after all the ca_state processing. (optional)
1087 void (*cong_control
)(struct sock
*sk
, const struct rate_sample
*rs
);
1088 /* get info for inet_diag (optional) */
1089 size_t (*get_info
)(struct sock
*sk
, u32 ext
, int *attr
,
1090 union tcp_cc_info
*info
);
1092 char name
[TCP_CA_NAME_MAX
];
1093 struct module
*owner
;
1096 int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1097 void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1099 void tcp_assign_congestion_control(struct sock
*sk
);
1100 void tcp_init_congestion_control(struct sock
*sk
);
1101 void tcp_cleanup_congestion_control(struct sock
*sk
);
1102 int tcp_set_default_congestion_control(struct net
*net
, const char *name
);
1103 void tcp_get_default_congestion_control(struct net
*net
, char *name
);
1104 void tcp_get_available_congestion_control(char *buf
, size_t len
);
1105 void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
1106 int tcp_set_allowed_congestion_control(char *allowed
);
1107 int tcp_set_congestion_control(struct sock
*sk
, const char *name
, bool load
,
1108 bool reinit
, bool cap_net_admin
);
1109 u32
tcp_slow_start(struct tcp_sock
*tp
, u32 acked
);
1110 void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
, u32 acked
);
1112 u32
tcp_reno_ssthresh(struct sock
*sk
);
1113 u32
tcp_reno_undo_cwnd(struct sock
*sk
);
1114 void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 acked
);
1115 extern struct tcp_congestion_ops tcp_reno
;
1117 struct tcp_congestion_ops
*tcp_ca_find(const char *name
);
1118 struct tcp_congestion_ops
*tcp_ca_find_key(u32 key
);
1119 u32
tcp_ca_get_key_by_name(struct net
*net
, const char *name
, bool *ecn_ca
);
1121 char *tcp_ca_get_name_by_key(u32 key
, char *buffer
);
1123 static inline char *tcp_ca_get_name_by_key(u32 key
, char *buffer
)
1129 static inline bool tcp_ca_needs_ecn(const struct sock
*sk
)
1131 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1133 return icsk
->icsk_ca_ops
->flags
& TCP_CONG_NEEDS_ECN
;
1136 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
1138 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1140 if (icsk
->icsk_ca_ops
->set_state
)
1141 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
1142 icsk
->icsk_ca_state
= ca_state
;
1145 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
1147 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
1149 if (icsk
->icsk_ca_ops
->cwnd_event
)
1150 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
1153 /* From tcp_rate.c */
1154 void tcp_rate_skb_sent(struct sock
*sk
, struct sk_buff
*skb
);
1155 void tcp_rate_skb_delivered(struct sock
*sk
, struct sk_buff
*skb
,
1156 struct rate_sample
*rs
);
1157 void tcp_rate_gen(struct sock
*sk
, u32 delivered
, u32 lost
,
1158 bool is_sack_reneg
, struct rate_sample
*rs
);
1159 void tcp_rate_check_app_limited(struct sock
*sk
);
1161 /* These functions determine how the current flow behaves in respect of SACK
1162 * handling. SACK is negotiated with the peer, and therefore it can vary
1163 * between different flows.
1165 * tcp_is_sack - SACK enabled
1166 * tcp_is_reno - No SACK
1168 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
1170 return likely(tp
->rx_opt
.sack_ok
);
1173 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
1175 return !tcp_is_sack(tp
);
1178 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
1180 return tp
->sacked_out
+ tp
->lost_out
;
1183 /* This determines how many packets are "in the network" to the best
1184 * of our knowledge. In many cases it is conservative, but where
1185 * detailed information is available from the receiver (via SACK
1186 * blocks etc.) we can make more aggressive calculations.
1188 * Use this for decisions involving congestion control, use just
1189 * tp->packets_out to determine if the send queue is empty or not.
1191 * Read this equation as:
1193 * "Packets sent once on transmission queue" MINUS
1194 * "Packets left network, but not honestly ACKed yet" PLUS
1195 * "Packets fast retransmitted"
1197 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1199 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
1202 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1204 static inline bool tcp_in_slow_start(const struct tcp_sock
*tp
)
1206 return tp
->snd_cwnd
< tp
->snd_ssthresh
;
1209 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
1211 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
1214 static inline bool tcp_in_cwnd_reduction(const struct sock
*sk
)
1216 return (TCPF_CA_CWR
| TCPF_CA_Recovery
) &
1217 (1 << inet_csk(sk
)->icsk_ca_state
);
1220 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1221 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1224 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
1226 const struct tcp_sock
*tp
= tcp_sk(sk
);
1228 if (tcp_in_cwnd_reduction(sk
))
1229 return tp
->snd_ssthresh
;
1231 return max(tp
->snd_ssthresh
,
1232 ((tp
->snd_cwnd
>> 1) +
1233 (tp
->snd_cwnd
>> 2)));
1236 /* Use define here intentionally to get WARN_ON location shown at the caller */
1237 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1239 void tcp_enter_cwr(struct sock
*sk
);
1240 __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
1242 /* The maximum number of MSS of available cwnd for which TSO defers
1243 * sending if not using sysctl_tcp_tso_win_divisor.
1245 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
1250 /* Returns end sequence number of the receiver's advertised window */
1251 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
1253 return tp
->snd_una
+ tp
->snd_wnd
;
1256 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1257 * flexible approach. The RFC suggests cwnd should not be raised unless
1258 * it was fully used previously. And that's exactly what we do in
1259 * congestion avoidance mode. But in slow start we allow cwnd to grow
1260 * as long as the application has used half the cwnd.
1262 * cwnd is 10 (IW10), but application sends 9 frames.
1263 * We allow cwnd to reach 18 when all frames are ACKed.
1264 * This check is safe because it's as aggressive as slow start which already
1265 * risks 100% overshoot. The advantage is that we discourage application to
1266 * either send more filler packets or data to artificially blow up the cwnd
1267 * usage, and allow application-limited process to probe bw more aggressively.
1269 static inline bool tcp_is_cwnd_limited(const struct sock
*sk
)
1271 const struct tcp_sock
*tp
= tcp_sk(sk
);
1273 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1274 if (tcp_in_slow_start(tp
))
1275 return tp
->snd_cwnd
< 2 * tp
->max_packets_out
;
1277 return tp
->is_cwnd_limited
;
1280 /* BBR congestion control needs pacing.
1281 * Same remark for SO_MAX_PACING_RATE.
1282 * sch_fq packet scheduler is efficiently handling pacing,
1283 * but is not always installed/used.
1284 * Return true if TCP stack should pace packets itself.
1286 static inline bool tcp_needs_internal_pacing(const struct sock
*sk
)
1288 return smp_load_acquire(&sk
->sk_pacing_status
) == SK_PACING_NEEDED
;
1291 /* Return in jiffies the delay before one skb is sent.
1292 * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1294 static inline unsigned long tcp_pacing_delay(const struct sock
*sk
,
1295 const struct sk_buff
*skb
)
1297 s64 pacing_delay
= skb
? skb
->tstamp
: tcp_sk(sk
)->tcp_wstamp_ns
;
1299 pacing_delay
-= tcp_sk(sk
)->tcp_clock_cache
;
1301 return pacing_delay
> 0 ? nsecs_to_jiffies(pacing_delay
) : 0;
1304 static inline void tcp_reset_xmit_timer(struct sock
*sk
,
1307 const unsigned long max_when
,
1308 const struct sk_buff
*skb
)
1310 inet_csk_reset_xmit_timer(sk
, what
, when
+ tcp_pacing_delay(sk
, skb
),
1314 /* Something is really bad, we could not queue an additional packet,
1315 * because qdisc is full or receiver sent a 0 window, or we are paced.
1316 * We do not want to add fuel to the fire, or abort too early,
1317 * so make sure the timer we arm now is at least 200ms in the future,
1318 * regardless of current icsk_rto value (as it could be ~2ms)
1320 static inline unsigned long tcp_probe0_base(const struct sock
*sk
)
1322 return max_t(unsigned long, inet_csk(sk
)->icsk_rto
, TCP_RTO_MIN
);
1325 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1326 static inline unsigned long tcp_probe0_when(const struct sock
*sk
,
1327 unsigned long max_when
)
1329 u64 when
= (u64
)tcp_probe0_base(sk
) << inet_csk(sk
)->icsk_backoff
;
1331 return (unsigned long)min_t(u64
, when
, max_when
);
1334 static inline void tcp_check_probe_timer(struct sock
*sk
)
1336 if (!tcp_sk(sk
)->packets_out
&& !inet_csk(sk
)->icsk_pending
)
1337 tcp_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
1338 tcp_probe0_base(sk
), TCP_RTO_MAX
,
1342 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
1347 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
1353 * Calculate(/check) TCP checksum
1355 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
1356 __be32 daddr
, __wsum base
)
1358 return csum_tcpudp_magic(saddr
, daddr
, len
, IPPROTO_TCP
, base
);
1361 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
1363 return !skb_csum_unnecessary(skb
) &&
1364 __skb_checksum_complete(skb
);
1367 bool tcp_add_backlog(struct sock
*sk
, struct sk_buff
*skb
);
1368 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
);
1369 void tcp_set_state(struct sock
*sk
, int state
);
1370 void tcp_done(struct sock
*sk
);
1371 int tcp_abort(struct sock
*sk
, int err
);
1373 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1376 rx_opt
->num_sacks
= 0;
1379 u32
tcp_default_init_rwnd(u32 mss
);
1380 void tcp_cwnd_restart(struct sock
*sk
, s32 delta
);
1382 static inline void tcp_slow_start_after_idle_check(struct sock
*sk
)
1384 const struct tcp_congestion_ops
*ca_ops
= inet_csk(sk
)->icsk_ca_ops
;
1385 struct tcp_sock
*tp
= tcp_sk(sk
);
1388 if (!sock_net(sk
)->ipv4
.sysctl_tcp_slow_start_after_idle
|| tp
->packets_out
||
1389 ca_ops
->cong_control
)
1391 delta
= tcp_jiffies32
- tp
->lsndtime
;
1392 if (delta
> inet_csk(sk
)->icsk_rto
)
1393 tcp_cwnd_restart(sk
, delta
);
1396 /* Determine a window scaling and initial window to offer. */
1397 void tcp_select_initial_window(const struct sock
*sk
, int __space
,
1398 __u32 mss
, __u32
*rcv_wnd
,
1399 __u32
*window_clamp
, int wscale_ok
,
1400 __u8
*rcv_wscale
, __u32 init_rcv_wnd
);
1402 static inline int tcp_win_from_space(const struct sock
*sk
, int space
)
1404 int tcp_adv_win_scale
= sock_net(sk
)->ipv4
.sysctl_tcp_adv_win_scale
;
1406 return tcp_adv_win_scale
<= 0 ?
1407 (space
>>(-tcp_adv_win_scale
)) :
1408 space
- (space
>>tcp_adv_win_scale
);
1411 /* Note: caller must be prepared to deal with negative returns */
1412 static inline int tcp_space(const struct sock
*sk
)
1414 return tcp_win_from_space(sk
, READ_ONCE(sk
->sk_rcvbuf
) -
1415 READ_ONCE(sk
->sk_backlog
.len
) -
1416 atomic_read(&sk
->sk_rmem_alloc
));
1419 static inline int tcp_full_space(const struct sock
*sk
)
1421 return tcp_win_from_space(sk
, READ_ONCE(sk
->sk_rcvbuf
));
1424 extern void tcp_openreq_init_rwin(struct request_sock
*req
,
1425 const struct sock
*sk_listener
,
1426 const struct dst_entry
*dst
);
1428 void tcp_enter_memory_pressure(struct sock
*sk
);
1429 void tcp_leave_memory_pressure(struct sock
*sk
);
1431 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1433 struct net
*net
= sock_net((struct sock
*)tp
);
1435 return tp
->keepalive_intvl
? : net
->ipv4
.sysctl_tcp_keepalive_intvl
;
1438 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1440 struct net
*net
= sock_net((struct sock
*)tp
);
1442 return tp
->keepalive_time
? : net
->ipv4
.sysctl_tcp_keepalive_time
;
1445 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1447 struct net
*net
= sock_net((struct sock
*)tp
);
1449 return tp
->keepalive_probes
? : net
->ipv4
.sysctl_tcp_keepalive_probes
;
1452 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1454 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1456 return min_t(u32
, tcp_jiffies32
- icsk
->icsk_ack
.lrcvtime
,
1457 tcp_jiffies32
- tp
->rcv_tstamp
);
1460 static inline int tcp_fin_time(const struct sock
*sk
)
1462 int fin_timeout
= tcp_sk(sk
)->linger2
? : sock_net(sk
)->ipv4
.sysctl_tcp_fin_timeout
;
1463 const int rto
= inet_csk(sk
)->icsk_rto
;
1465 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1466 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1471 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1474 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1476 if (unlikely(!time_before32(ktime_get_seconds(),
1477 rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
)))
1480 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1481 * then following tcp messages have valid values. Ignore 0 value,
1482 * or else 'negative' tsval might forbid us to accept their packets.
1484 if (!rx_opt
->ts_recent
)
1489 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1492 if (tcp_paws_check(rx_opt
, 0))
1495 /* RST segments are not recommended to carry timestamp,
1496 and, if they do, it is recommended to ignore PAWS because
1497 "their cleanup function should take precedence over timestamps."
1498 Certainly, it is mistake. It is necessary to understand the reasons
1499 of this constraint to relax it: if peer reboots, clock may go
1500 out-of-sync and half-open connections will not be reset.
1501 Actually, the problem would be not existing if all
1502 the implementations followed draft about maintaining clock
1503 via reboots. Linux-2.2 DOES NOT!
1505 However, we can relax time bounds for RST segments to MSL.
1507 if (rst
&& !time_before32(ktime_get_seconds(),
1508 rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
))
1513 bool tcp_oow_rate_limited(struct net
*net
, const struct sk_buff
*skb
,
1514 int mib_idx
, u32
*last_oow_ack_time
);
1516 static inline void tcp_mib_init(struct net
*net
)
1519 TCP_ADD_STATS(net
, TCP_MIB_RTOALGORITHM
, 1);
1520 TCP_ADD_STATS(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1521 TCP_ADD_STATS(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1522 TCP_ADD_STATS(net
, TCP_MIB_MAXCONN
, -1);
1526 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1528 tp
->lost_skb_hint
= NULL
;
1531 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1533 tcp_clear_retrans_hints_partial(tp
);
1534 tp
->retransmit_skb_hint
= NULL
;
1537 union tcp_md5_addr
{
1539 #if IS_ENABLED(CONFIG_IPV6)
1544 /* - key database */
1545 struct tcp_md5sig_key
{
1546 struct hlist_node node
;
1548 u8 family
; /* AF_INET or AF_INET6 */
1550 union tcp_md5_addr addr
;
1551 int l3index
; /* set if key added with L3 scope */
1552 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1553 struct rcu_head rcu
;
1557 struct tcp_md5sig_info
{
1558 struct hlist_head head
;
1559 struct rcu_head rcu
;
1562 /* - pseudo header */
1563 struct tcp4_pseudohdr
{
1571 struct tcp6_pseudohdr
{
1572 struct in6_addr saddr
;
1573 struct in6_addr daddr
;
1575 __be32 protocol
; /* including padding */
1578 union tcp_md5sum_block
{
1579 struct tcp4_pseudohdr ip4
;
1580 #if IS_ENABLED(CONFIG_IPV6)
1581 struct tcp6_pseudohdr ip6
;
1585 /* - pool: digest algorithm, hash description and scratch buffer */
1586 struct tcp_md5sig_pool
{
1587 struct ahash_request
*md5_req
;
1592 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1593 const struct sock
*sk
, const struct sk_buff
*skb
);
1594 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1595 int family
, u8 prefixlen
, int l3index
,
1596 const u8
*newkey
, u8 newkeylen
, gfp_t gfp
);
1597 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1598 int family
, u8 prefixlen
, int l3index
);
1599 struct tcp_md5sig_key
*tcp_v4_md5_lookup(const struct sock
*sk
,
1600 const struct sock
*addr_sk
);
1602 #ifdef CONFIG_TCP_MD5SIG
1603 #include <linux/jump_label.h>
1604 extern struct static_key_false tcp_md5_needed
;
1605 struct tcp_md5sig_key
*__tcp_md5_do_lookup(const struct sock
*sk
, int l3index
,
1606 const union tcp_md5_addr
*addr
,
1608 static inline struct tcp_md5sig_key
*
1609 tcp_md5_do_lookup(const struct sock
*sk
, int l3index
,
1610 const union tcp_md5_addr
*addr
, int family
)
1612 if (!static_branch_unlikely(&tcp_md5_needed
))
1614 return __tcp_md5_do_lookup(sk
, l3index
, addr
, family
);
1617 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1619 static inline struct tcp_md5sig_key
*
1620 tcp_md5_do_lookup(const struct sock
*sk
, int l3index
,
1621 const union tcp_md5_addr
*addr
, int family
)
1625 #define tcp_twsk_md5_key(twsk) NULL
1628 bool tcp_alloc_md5sig_pool(void);
1630 struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1631 static inline void tcp_put_md5sig_pool(void)
1636 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1637 unsigned int header_len
);
1638 int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1639 const struct tcp_md5sig_key
*key
);
1641 /* From tcp_fastopen.c */
1642 void tcp_fastopen_cache_get(struct sock
*sk
, u16
*mss
,
1643 struct tcp_fastopen_cookie
*cookie
);
1644 void tcp_fastopen_cache_set(struct sock
*sk
, u16 mss
,
1645 struct tcp_fastopen_cookie
*cookie
, bool syn_lost
,
1647 struct tcp_fastopen_request
{
1648 /* Fast Open cookie. Size 0 means a cookie request */
1649 struct tcp_fastopen_cookie cookie
;
1650 struct msghdr
*data
; /* data in MSG_FASTOPEN */
1652 int copied
; /* queued in tcp_connect() */
1653 struct ubuf_info
*uarg
;
1655 void tcp_free_fastopen_req(struct tcp_sock
*tp
);
1656 void tcp_fastopen_destroy_cipher(struct sock
*sk
);
1657 void tcp_fastopen_ctx_destroy(struct net
*net
);
1658 int tcp_fastopen_reset_cipher(struct net
*net
, struct sock
*sk
,
1659 void *primary_key
, void *backup_key
);
1660 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
);
1661 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
1662 struct request_sock
*req
,
1663 struct tcp_fastopen_cookie
*foc
,
1664 const struct dst_entry
*dst
);
1665 void tcp_fastopen_init_key_once(struct net
*net
);
1666 bool tcp_fastopen_cookie_check(struct sock
*sk
, u16
*mss
,
1667 struct tcp_fastopen_cookie
*cookie
);
1668 bool tcp_fastopen_defer_connect(struct sock
*sk
, int *err
);
1669 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1670 #define TCP_FASTOPEN_KEY_MAX 2
1671 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1672 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1674 /* Fastopen key context */
1675 struct tcp_fastopen_context
{
1676 siphash_key_t key
[TCP_FASTOPEN_KEY_MAX
];
1678 struct rcu_head rcu
;
1681 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout
;
1682 void tcp_fastopen_active_disable(struct sock
*sk
);
1683 bool tcp_fastopen_active_should_disable(struct sock
*sk
);
1684 void tcp_fastopen_active_disable_ofo_check(struct sock
*sk
);
1685 void tcp_fastopen_active_detect_blackhole(struct sock
*sk
, bool expired
);
1687 /* Caller needs to wrap with rcu_read_(un)lock() */
1689 struct tcp_fastopen_context
*tcp_fastopen_get_ctx(const struct sock
*sk
)
1691 struct tcp_fastopen_context
*ctx
;
1693 ctx
= rcu_dereference(inet_csk(sk
)->icsk_accept_queue
.fastopenq
.ctx
);
1695 ctx
= rcu_dereference(sock_net(sk
)->ipv4
.tcp_fastopen_ctx
);
1700 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie
*foc
,
1701 const struct tcp_fastopen_cookie
*orig
)
1703 if (orig
->len
== TCP_FASTOPEN_COOKIE_SIZE
&&
1704 orig
->len
== foc
->len
&&
1705 !memcmp(orig
->val
, foc
->val
, foc
->len
))
1711 int tcp_fastopen_context_len(const struct tcp_fastopen_context
*ctx
)
1716 /* Latencies incurred by various limits for a sender. They are
1717 * chronograph-like stats that are mutually exclusive.
1721 TCP_CHRONO_BUSY
, /* Actively sending data (non-empty write queue) */
1722 TCP_CHRONO_RWND_LIMITED
, /* Stalled by insufficient receive window */
1723 TCP_CHRONO_SNDBUF_LIMITED
, /* Stalled by insufficient send buffer */
1727 void tcp_chrono_start(struct sock
*sk
, const enum tcp_chrono type
);
1728 void tcp_chrono_stop(struct sock
*sk
, const enum tcp_chrono type
);
1730 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1731 * the same memory storage than skb->destructor/_skb_refdst
1733 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff
*skb
)
1735 skb
->destructor
= NULL
;
1736 skb
->_skb_refdst
= 0UL;
1739 #define tcp_skb_tsorted_save(skb) { \
1740 unsigned long _save = skb->_skb_refdst; \
1741 skb->_skb_refdst = 0UL;
1743 #define tcp_skb_tsorted_restore(skb) \
1744 skb->_skb_refdst = _save; \
1747 void tcp_write_queue_purge(struct sock
*sk
);
1749 static inline struct sk_buff
*tcp_rtx_queue_head(const struct sock
*sk
)
1751 return skb_rb_first(&sk
->tcp_rtx_queue
);
1754 static inline struct sk_buff
*tcp_rtx_queue_tail(const struct sock
*sk
)
1756 return skb_rb_last(&sk
->tcp_rtx_queue
);
1759 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1761 return skb_peek(&sk
->sk_write_queue
);
1764 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1766 return skb_peek_tail(&sk
->sk_write_queue
);
1769 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1770 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1772 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1774 return skb_peek(&sk
->sk_write_queue
);
1777 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1778 const struct sk_buff
*skb
)
1780 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1784 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1787 * Since the write queue can have a temporary empty skb in it,
1788 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1790 static inline bool tcp_write_queue_empty(const struct sock
*sk
)
1792 const struct tcp_sock
*tp
= tcp_sk(sk
);
1794 return tp
->write_seq
== tp
->snd_nxt
;
1797 static inline bool tcp_rtx_queue_empty(const struct sock
*sk
)
1799 return RB_EMPTY_ROOT(&sk
->tcp_rtx_queue
);
1802 static inline bool tcp_rtx_and_write_queues_empty(const struct sock
*sk
)
1804 return tcp_rtx_queue_empty(sk
) && tcp_write_queue_empty(sk
);
1807 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1809 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1811 /* Queue it, remembering where we must start sending. */
1812 if (sk
->sk_write_queue
.next
== skb
)
1813 tcp_chrono_start(sk
, TCP_CHRONO_BUSY
);
1816 /* Insert new before skb on the write queue of sk. */
1817 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1818 struct sk_buff
*skb
,
1821 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1824 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1826 tcp_skb_tsorted_anchor_cleanup(skb
);
1827 __skb_unlink(skb
, &sk
->sk_write_queue
);
1830 void tcp_rbtree_insert(struct rb_root
*root
, struct sk_buff
*skb
);
1832 static inline void tcp_rtx_queue_unlink(struct sk_buff
*skb
, struct sock
*sk
)
1834 tcp_skb_tsorted_anchor_cleanup(skb
);
1835 rb_erase(&skb
->rbnode
, &sk
->tcp_rtx_queue
);
1838 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff
*skb
, struct sock
*sk
)
1840 list_del(&skb
->tcp_tsorted_anchor
);
1841 tcp_rtx_queue_unlink(skb
, sk
);
1842 sk_wmem_free_skb(sk
, skb
);
1845 static inline void tcp_push_pending_frames(struct sock
*sk
)
1847 if (tcp_send_head(sk
)) {
1848 struct tcp_sock
*tp
= tcp_sk(sk
);
1850 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1854 /* Start sequence of the skb just after the highest skb with SACKed
1855 * bit, valid only if sacked_out > 0 or when the caller has ensured
1856 * validity by itself.
1858 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1860 if (!tp
->sacked_out
)
1863 if (tp
->highest_sack
== NULL
)
1866 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1869 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1871 tcp_sk(sk
)->highest_sack
= skb_rb_next(skb
);
1874 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1876 return tcp_sk(sk
)->highest_sack
;
1879 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1881 tcp_sk(sk
)->highest_sack
= tcp_rtx_queue_head(sk
);
1884 /* Called when old skb is about to be deleted and replaced by new skb */
1885 static inline void tcp_highest_sack_replace(struct sock
*sk
,
1886 struct sk_buff
*old
,
1887 struct sk_buff
*new)
1889 if (old
== tcp_highest_sack(sk
))
1890 tcp_sk(sk
)->highest_sack
= new;
1893 /* This helper checks if socket has IP_TRANSPARENT set */
1894 static inline bool inet_sk_transparent(const struct sock
*sk
)
1896 switch (sk
->sk_state
) {
1898 return inet_twsk(sk
)->tw_transparent
;
1899 case TCP_NEW_SYN_RECV
:
1900 return inet_rsk(inet_reqsk(sk
))->no_srccheck
;
1902 return inet_sk(sk
)->transparent
;
1905 /* Determines whether this is a thin stream (which may suffer from
1906 * increased latency). Used to trigger latency-reducing mechanisms.
1908 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1910 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1914 enum tcp_seq_states
{
1915 TCP_SEQ_STATE_LISTENING
,
1916 TCP_SEQ_STATE_ESTABLISHED
,
1919 void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
);
1920 void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
);
1921 void tcp_seq_stop(struct seq_file
*seq
, void *v
);
1923 struct tcp_seq_afinfo
{
1927 struct tcp_iter_state
{
1928 struct seq_net_private p
;
1929 enum tcp_seq_states state
;
1930 struct sock
*syn_wait_sk
;
1931 int bucket
, offset
, sbucket
, num
;
1935 extern struct request_sock_ops tcp_request_sock_ops
;
1936 extern struct request_sock_ops tcp6_request_sock_ops
;
1938 void tcp_v4_destroy_sock(struct sock
*sk
);
1940 struct sk_buff
*tcp_gso_segment(struct sk_buff
*skb
,
1941 netdev_features_t features
);
1942 struct sk_buff
*tcp_gro_receive(struct list_head
*head
, struct sk_buff
*skb
);
1943 int tcp_gro_complete(struct sk_buff
*skb
);
1945 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
);
1947 static inline u32
tcp_notsent_lowat(const struct tcp_sock
*tp
)
1949 struct net
*net
= sock_net((struct sock
*)tp
);
1950 return tp
->notsent_lowat
?: net
->ipv4
.sysctl_tcp_notsent_lowat
;
1953 /* @wake is one when sk_stream_write_space() calls us.
1954 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1955 * This mimics the strategy used in sock_def_write_space().
1957 static inline bool tcp_stream_memory_free(const struct sock
*sk
, int wake
)
1959 const struct tcp_sock
*tp
= tcp_sk(sk
);
1960 u32 notsent_bytes
= READ_ONCE(tp
->write_seq
) -
1961 READ_ONCE(tp
->snd_nxt
);
1963 return (notsent_bytes
<< wake
) < tcp_notsent_lowat(tp
);
1966 #ifdef CONFIG_PROC_FS
1967 int tcp4_proc_init(void);
1968 void tcp4_proc_exit(void);
1971 int tcp_rtx_synack(const struct sock
*sk
, struct request_sock
*req
);
1972 int tcp_conn_request(struct request_sock_ops
*rsk_ops
,
1973 const struct tcp_request_sock_ops
*af_ops
,
1974 struct sock
*sk
, struct sk_buff
*skb
);
1976 /* TCP af-specific functions */
1977 struct tcp_sock_af_ops
{
1978 #ifdef CONFIG_TCP_MD5SIG
1979 struct tcp_md5sig_key
*(*md5_lookup
) (const struct sock
*sk
,
1980 const struct sock
*addr_sk
);
1981 int (*calc_md5_hash
)(char *location
,
1982 const struct tcp_md5sig_key
*md5
,
1983 const struct sock
*sk
,
1984 const struct sk_buff
*skb
);
1985 int (*md5_parse
)(struct sock
*sk
,
1987 char __user
*optval
,
1992 struct tcp_request_sock_ops
{
1994 #ifdef CONFIG_TCP_MD5SIG
1995 struct tcp_md5sig_key
*(*req_md5_lookup
)(const struct sock
*sk
,
1996 const struct sock
*addr_sk
);
1997 int (*calc_md5_hash
) (char *location
,
1998 const struct tcp_md5sig_key
*md5
,
1999 const struct sock
*sk
,
2000 const struct sk_buff
*skb
);
2002 void (*init_req
)(struct request_sock
*req
,
2003 const struct sock
*sk_listener
,
2004 struct sk_buff
*skb
);
2005 #ifdef CONFIG_SYN_COOKIES
2006 __u32 (*cookie_init_seq
)(const struct sk_buff
*skb
,
2009 struct dst_entry
*(*route_req
)(const struct sock
*sk
, struct flowi
*fl
,
2010 const struct request_sock
*req
);
2011 u32 (*init_seq
)(const struct sk_buff
*skb
);
2012 u32 (*init_ts_off
)(const struct net
*net
, const struct sk_buff
*skb
);
2013 int (*send_synack
)(const struct sock
*sk
, struct dst_entry
*dst
,
2014 struct flowi
*fl
, struct request_sock
*req
,
2015 struct tcp_fastopen_cookie
*foc
,
2016 enum tcp_synack_type synack_type
);
2019 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
;
2020 #if IS_ENABLED(CONFIG_IPV6)
2021 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops
;
2024 #ifdef CONFIG_SYN_COOKIES
2025 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
2026 const struct sock
*sk
, struct sk_buff
*skb
,
2029 tcp_synq_overflow(sk
);
2030 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_SYNCOOKIESSENT
);
2031 return ops
->cookie_init_seq(skb
, mss
);
2034 static inline __u32
cookie_init_sequence(const struct tcp_request_sock_ops
*ops
,
2035 const struct sock
*sk
, struct sk_buff
*skb
,
2042 int tcpv4_offload_init(void);
2044 void tcp_v4_init(void);
2045 void tcp_init(void);
2047 /* tcp_recovery.c */
2048 void tcp_mark_skb_lost(struct sock
*sk
, struct sk_buff
*skb
);
2049 void tcp_newreno_mark_lost(struct sock
*sk
, bool snd_una_advanced
);
2050 extern s32
tcp_rack_skb_timeout(struct tcp_sock
*tp
, struct sk_buff
*skb
,
2052 extern void tcp_rack_mark_lost(struct sock
*sk
);
2053 extern void tcp_rack_advance(struct tcp_sock
*tp
, u8 sacked
, u32 end_seq
,
2055 extern void tcp_rack_reo_timeout(struct sock
*sk
);
2056 extern void tcp_rack_update_reo_wnd(struct sock
*sk
, struct rate_sample
*rs
);
2058 /* At how many usecs into the future should the RTO fire? */
2059 static inline s64
tcp_rto_delta_us(const struct sock
*sk
)
2061 const struct sk_buff
*skb
= tcp_rtx_queue_head(sk
);
2062 u32 rto
= inet_csk(sk
)->icsk_rto
;
2063 u64 rto_time_stamp_us
= tcp_skb_timestamp_us(skb
) + jiffies_to_usecs(rto
);
2065 return rto_time_stamp_us
- tcp_sk(sk
)->tcp_mstamp
;
2069 * Save and compile IPv4 options, return a pointer to it
2071 static inline struct ip_options_rcu
*tcp_v4_save_options(struct net
*net
,
2072 struct sk_buff
*skb
)
2074 const struct ip_options
*opt
= &TCP_SKB_CB(skb
)->header
.h4
.opt
;
2075 struct ip_options_rcu
*dopt
= NULL
;
2078 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
2080 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
2081 if (dopt
&& __ip_options_echo(net
, &dopt
->opt
, skb
, opt
)) {
2089 /* locally generated TCP pure ACKs have skb->truesize == 2
2090 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2091 * This is much faster than dissecting the packet to find out.
2092 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2094 static inline bool skb_is_tcp_pure_ack(const struct sk_buff
*skb
)
2096 return skb
->truesize
== 2;
2099 static inline void skb_set_tcp_pure_ack(struct sk_buff
*skb
)
2104 static inline int tcp_inq(struct sock
*sk
)
2106 struct tcp_sock
*tp
= tcp_sk(sk
);
2109 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
2111 } else if (sock_flag(sk
, SOCK_URGINLINE
) ||
2113 before(tp
->urg_seq
, tp
->copied_seq
) ||
2114 !before(tp
->urg_seq
, tp
->rcv_nxt
)) {
2116 answ
= tp
->rcv_nxt
- tp
->copied_seq
;
2118 /* Subtract 1, if FIN was received */
2119 if (answ
&& sock_flag(sk
, SOCK_DONE
))
2122 answ
= tp
->urg_seq
- tp
->copied_seq
;
2128 int tcp_peek_len(struct socket
*sock
);
2130 static inline void tcp_segs_in(struct tcp_sock
*tp
, const struct sk_buff
*skb
)
2134 segs_in
= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
2135 tp
->segs_in
+= segs_in
;
2136 if (skb
->len
> tcp_hdrlen(skb
))
2137 tp
->data_segs_in
+= segs_in
;
2141 * TCP listen path runs lockless.
2142 * We forced "struct sock" to be const qualified to make sure
2143 * we don't modify one of its field by mistake.
2144 * Here, we increment sk_drops which is an atomic_t, so we can safely
2145 * make sock writable again.
2147 static inline void tcp_listendrop(const struct sock
*sk
)
2149 atomic_inc(&((struct sock
*)sk
)->sk_drops
);
2150 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
2153 enum hrtimer_restart
tcp_pace_kick(struct hrtimer
*timer
);
2156 * Interface for adding Upper Level Protocols over TCP
2159 #define TCP_ULP_NAME_MAX 16
2160 #define TCP_ULP_MAX 128
2161 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2163 struct tcp_ulp_ops
{
2164 struct list_head list
;
2166 /* initialize ulp */
2167 int (*init
)(struct sock
*sk
);
2169 void (*update
)(struct sock
*sk
, struct proto
*p
,
2170 void (*write_space
)(struct sock
*sk
));
2172 void (*release
)(struct sock
*sk
);
2174 int (*get_info
)(const struct sock
*sk
, struct sk_buff
*skb
);
2175 size_t (*get_info_size
)(const struct sock
*sk
);
2177 void (*clone
)(const struct request_sock
*req
, struct sock
*newsk
,
2178 const gfp_t priority
);
2180 char name
[TCP_ULP_NAME_MAX
];
2181 struct module
*owner
;
2183 int tcp_register_ulp(struct tcp_ulp_ops
*type
);
2184 void tcp_unregister_ulp(struct tcp_ulp_ops
*type
);
2185 int tcp_set_ulp(struct sock
*sk
, const char *name
);
2186 void tcp_get_available_ulp(char *buf
, size_t len
);
2187 void tcp_cleanup_ulp(struct sock
*sk
);
2188 void tcp_update_ulp(struct sock
*sk
, struct proto
*p
,
2189 void (*write_space
)(struct sock
*sk
));
2191 #define MODULE_ALIAS_TCP_ULP(name) \
2192 __MODULE_INFO(alias, alias_userspace, name); \
2193 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2198 #ifdef CONFIG_BPF_STREAM_PARSER
2199 struct proto
*tcp_bpf_get_proto(struct sock
*sk
, struct sk_psock
*psock
);
2200 void tcp_bpf_clone(const struct sock
*sk
, struct sock
*newsk
);
2202 static inline void tcp_bpf_clone(const struct sock
*sk
, struct sock
*newsk
)
2205 #endif /* CONFIG_BPF_STREAM_PARSER */
2207 #ifdef CONFIG_NET_SOCK_MSG
2208 int tcp_bpf_sendmsg_redir(struct sock
*sk
, struct sk_msg
*msg
, u32 bytes
,
2210 int __tcp_bpf_recvmsg(struct sock
*sk
, struct sk_psock
*psock
,
2211 struct msghdr
*msg
, int len
, int flags
);
2212 #endif /* CONFIG_NET_SOCK_MSG */
2214 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2215 * is < 0, then the BPF op failed (for example if the loaded BPF
2216 * program does not support the chosen operation or there is no BPF
2220 static inline int tcp_call_bpf(struct sock
*sk
, int op
, u32 nargs
, u32
*args
)
2222 struct bpf_sock_ops_kern sock_ops
;
2225 memset(&sock_ops
, 0, offsetof(struct bpf_sock_ops_kern
, temp
));
2226 if (sk_fullsock(sk
)) {
2227 sock_ops
.is_fullsock
= 1;
2228 sock_owned_by_me(sk
);
2234 memcpy(sock_ops
.args
, args
, nargs
* sizeof(*args
));
2236 ret
= BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops
);
2238 ret
= sock_ops
.reply
;
2244 static inline int tcp_call_bpf_2arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
)
2246 u32 args
[2] = {arg1
, arg2
};
2248 return tcp_call_bpf(sk
, op
, 2, args
);
2251 static inline int tcp_call_bpf_3arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
,
2254 u32 args
[3] = {arg1
, arg2
, arg3
};
2256 return tcp_call_bpf(sk
, op
, 3, args
);
2260 static inline int tcp_call_bpf(struct sock
*sk
, int op
, u32 nargs
, u32
*args
)
2265 static inline int tcp_call_bpf_2arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
)
2270 static inline int tcp_call_bpf_3arg(struct sock
*sk
, int op
, u32 arg1
, u32 arg2
,
2278 static inline u32
tcp_timeout_init(struct sock
*sk
)
2282 timeout
= tcp_call_bpf(sk
, BPF_SOCK_OPS_TIMEOUT_INIT
, 0, NULL
);
2285 timeout
= TCP_TIMEOUT_INIT
;
2289 static inline u32
tcp_rwnd_init_bpf(struct sock
*sk
)
2293 rwnd
= tcp_call_bpf(sk
, BPF_SOCK_OPS_RWND_INIT
, 0, NULL
);
2300 static inline bool tcp_bpf_ca_needs_ecn(struct sock
*sk
)
2302 return (tcp_call_bpf(sk
, BPF_SOCK_OPS_NEEDS_ECN
, 0, NULL
) == 1);
2305 static inline void tcp_bpf_rtt(struct sock
*sk
)
2307 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk
), BPF_SOCK_OPS_RTT_CB_FLAG
))
2308 tcp_call_bpf(sk
, BPF_SOCK_OPS_RTT_CB
, 0, NULL
);
2311 #if IS_ENABLED(CONFIG_SMC)
2312 extern struct static_key_false tcp_have_smc
;
2315 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2316 void clean_acked_data_enable(struct inet_connection_sock
*icsk
,
2317 void (*cad
)(struct sock
*sk
, u32 ack_seq
));
2318 void clean_acked_data_disable(struct inet_connection_sock
*icsk
);
2319 void clean_acked_data_flush(void);
2322 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled
);
2323 static inline void tcp_add_tx_delay(struct sk_buff
*skb
,
2324 const struct tcp_sock
*tp
)
2326 if (static_branch_unlikely(&tcp_tx_delay_enabled
))
2327 skb
->skb_mstamp_ns
+= (u64
)tp
->tcp_tx_delay
* NSEC_PER_USEC
;
2330 /* Compute Earliest Departure Time for some control packets
2331 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2333 static inline u64
tcp_transmit_time(const struct sock
*sk
)
2335 if (static_branch_unlikely(&tcp_tx_delay_enabled
)) {
2336 u32 delay
= (sk
->sk_state
== TCP_TIME_WAIT
) ?
2337 tcp_twsk(sk
)->tw_tx_delay
: tcp_sk(sk
)->tcp_tx_delay
;
2339 return tcp_clock_ns() + (u64
)delay
* NSEC_PER_USEC
;