2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/cryptohash.h>
31 #include <linux/kref.h>
32 #include <linux/ktime.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
49 #include <linux/bpf.h>
50 #include <linux/filter.h>
51 #include <linux/bpf-cgroup.h>
53 extern struct inet_hashinfo tcp_hashinfo;
55 extern struct percpu_counter tcp_orphan_count;
56 void tcp_time_wait(struct sock *sk, int state, int timeo);
58 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
59 #define MAX_TCP_OPTION_SPACE 40
60 #define TCP_MIN_SND_MSS 48
61 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
64 * Never offer a window over 32767 without using window scaling. Some
65 * poor stacks do signed 16bit maths!
67 #define MAX_TCP_WINDOW 32767U
69 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
70 #define TCP_MIN_MSS 88U
72 /* The least MTU to use for probing */
73 #define TCP_BASE_MSS 1024
75 /* probing interval, default to 10 minutes as per RFC4821 */
76 #define TCP_PROBE_INTERVAL 600
78 /* Specify interval when tcp mtu probing will stop */
79 #define TCP_PROBE_THRESHOLD 8
81 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
82 #define TCP_FASTRETRANS_THRESH 3
84 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
85 #define TCP_MAX_QUICKACKS 16U
87 /* Maximal number of window scale according to RFC1323 */
88 #define TCP_MAX_WSCALE 14U
91 #define TCP_URG_VALID 0x0100
92 #define TCP_URG_NOTYET 0x0200
93 #define TCP_URG_READ 0x0400
95 #define TCP_RETR1 3 /*
96 * This is how many retries it does before it
97 * tries to figure out if the gateway is
98 * down. Minimal RFC value is 3; it corresponds
99 * to ~3sec-8min depending on RTO.
102 #define TCP_RETR2 15 /*
103 * This should take at least
104 * 90 minutes to time out.
105 * RFC1122 says that the limit is 100 sec.
106 * 15 is ~13-30min depending on RTO.
109 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
110 * when active opening a connection.
111 * RFC1122 says the minimum retry MUST
112 * be at least 180secs. Nevertheless
113 * this value is corresponding to
114 * 63secs of retransmission with the
115 * current initial RTO.
118 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
119 * when passive opening a connection.
120 * This is corresponding to 31secs of
121 * retransmission with the current
125 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
126 * state, about 60 seconds */
127 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
128 /* BSD style FIN_WAIT2 deadlock breaker.
129 * It used to be 3min, new value is 60sec,
130 * to combine FIN-WAIT-2 timeout with
134 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
136 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
137 #define TCP_ATO_MIN ((unsigned)(HZ/25))
139 #define TCP_DELACK_MIN 4U
140 #define TCP_ATO_MIN 4U
142 #define TCP_RTO_MAX ((unsigned)(120*HZ))
143 #define TCP_RTO_MIN ((unsigned)(HZ/5))
144 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
145 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
146 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
147 * used as a fallback RTO for the
148 * initial data transmission if no
149 * valid RTT sample has been acquired,
150 * most likely due to retrans in 3WHS.
153 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
154 * for local resources.
156 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
157 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
158 #define TCP_KEEPALIVE_INTVL (75*HZ)
160 #define MAX_TCP_KEEPIDLE 32767
161 #define MAX_TCP_KEEPINTVL 32767
162 #define MAX_TCP_KEEPCNT 127
163 #define MAX_TCP_SYNCNT 127
165 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
167 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
168 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
169 * after this time. It should be equal
170 * (or greater than) TCP_TIMEWAIT_LEN
171 * to provide reliability equal to one
172 * provided by timewait state.
174 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
175 * timestamps. It must be less than
176 * minimal timewait lifetime.
182 #define TCPOPT_NOP 1 /* Padding */
183 #define TCPOPT_EOL 0 /* End of options */
184 #define TCPOPT_MSS 2 /* Segment size negotiating */
185 #define TCPOPT_WINDOW 3 /* Window scaling */
186 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
187 #define TCPOPT_SACK 5 /* SACK Block */
188 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
189 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
190 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
191 #define TCPOPT_EXP 254 /* Experimental */
192 /* Magic number to be after the option value for sharing TCP
193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
195 #define TCPOPT_FASTOPEN_MAGIC 0xF989
201 #define TCPOLEN_MSS 4
202 #define TCPOLEN_WINDOW 3
203 #define TCPOLEN_SACK_PERM 2
204 #define TCPOLEN_TIMESTAMP 10
205 #define TCPOLEN_MD5SIG 18
206 #define TCPOLEN_FASTOPEN_BASE 2
207 #define TCPOLEN_EXP_FASTOPEN_BASE 4
209 /* But this is what stacks really send out. */
210 #define TCPOLEN_TSTAMP_ALIGNED 12
211 #define TCPOLEN_WSCALE_ALIGNED 4
212 #define TCPOLEN_SACKPERM_ALIGNED 4
213 #define TCPOLEN_SACK_BASE 2
214 #define TCPOLEN_SACK_BASE_ALIGNED 4
215 #define TCPOLEN_SACK_PERBLOCK 8
216 #define TCPOLEN_MD5SIG_ALIGNED 20
217 #define TCPOLEN_MSS_ALIGNED 4
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_fastopen;
246 extern int sysctl_tcp_retrans_collapse;
247 extern int sysctl_tcp_stdurg;
248 extern int sysctl_tcp_rfc1337;
249 extern int sysctl_tcp_abort_on_overflow;
250 extern int sysctl_tcp_max_orphans;
251 extern int sysctl_tcp_fack;
252 extern int sysctl_tcp_reordering;
253 extern int sysctl_tcp_max_reordering;
254 extern int sysctl_tcp_dsack;
255 extern long sysctl_tcp_mem[3];
256 extern int sysctl_tcp_wmem[3];
257 extern int sysctl_tcp_rmem[3];
258 extern int sysctl_tcp_app_win;
259 extern int sysctl_tcp_adv_win_scale;
260 extern int sysctl_tcp_frto;
261 extern int sysctl_tcp_nometrics_save;
262 extern int sysctl_tcp_moderate_rcvbuf;
263 extern int sysctl_tcp_tso_win_divisor;
264 extern int sysctl_tcp_workaround_signed_windows;
265 extern int sysctl_tcp_slow_start_after_idle;
266 extern int sysctl_tcp_thin_linear_timeouts;
267 extern int sysctl_tcp_thin_dupack;
268 extern int sysctl_tcp_early_retrans;
269 extern int sysctl_tcp_recovery;
270 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
272 extern int sysctl_tcp_limit_output_bytes;
273 extern int sysctl_tcp_challenge_ack_limit;
274 extern int sysctl_tcp_min_tso_segs;
275 extern int sysctl_tcp_min_rtt_wlen;
276 extern int sysctl_tcp_autocorking;
277 extern int sysctl_tcp_invalid_ratelimit;
278 extern int sysctl_tcp_pacing_ss_ratio;
279 extern int sysctl_tcp_pacing_ca_ratio;
281 extern atomic_long_t tcp_memory_allocated;
282 extern struct percpu_counter tcp_sockets_allocated;
283 extern unsigned long tcp_memory_pressure;
285 /* optimized version of sk_under_memory_pressure() for TCP sockets */
286 static inline bool tcp_under_memory_pressure(const struct sock *sk)
288 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
289 mem_cgroup_under_socket_pressure(sk->sk_memcg))
292 return READ_ONCE(tcp_memory_pressure);
295 * The next routines deal with comparing 32 bit unsigned ints
296 * and worry about wraparound (automatic with unsigned arithmetic).
299 static inline bool before(__u32 seq1, __u32 seq2)
301 return (__s32)(seq1-seq2) < 0;
303 #define after(seq2, seq1) before(seq1, seq2)
305 /* is s2<=s1<=s3 ? */
306 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
308 return seq3 - seq2 >= seq1 - seq2;
311 static inline bool tcp_out_of_memory(struct sock *sk)
313 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
314 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
319 void sk_forced_mem_schedule(struct sock *sk, int size);
321 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
323 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
324 int orphans = percpu_counter_read_positive(ocp);
326 if (orphans << shift > sysctl_tcp_max_orphans) {
327 orphans = percpu_counter_sum_positive(ocp);
328 if (orphans << shift > sysctl_tcp_max_orphans)
334 bool tcp_check_oom(struct sock *sk, int shift);
337 extern struct proto tcp_prot;
339 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
340 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
341 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
342 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
344 void tcp_tasklet_init(void);
346 void tcp_v4_err(struct sk_buff *skb, u32);
348 void tcp_shutdown(struct sock *sk, int how);
350 int tcp_v4_early_demux(struct sk_buff *skb);
351 int tcp_v4_rcv(struct sk_buff *skb);
353 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
354 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
355 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
356 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
358 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
359 size_t size, int flags);
360 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
361 size_t size, int flags);
362 void tcp_release_cb(struct sock *sk);
363 void tcp_wfree(struct sk_buff *skb);
364 void tcp_write_timer_handler(struct sock *sk);
365 void tcp_delack_timer_handler(struct sock *sk);
366 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
367 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
368 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
369 const struct tcphdr *th);
370 void tcp_rcv_space_adjust(struct sock *sk);
371 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
372 void tcp_twsk_destructor(struct sock *sk);
373 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
374 struct pipe_inode_info *pipe, size_t len,
377 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
378 static inline void tcp_dec_quickack_mode(struct sock *sk,
379 const unsigned int pkts)
381 struct inet_connection_sock *icsk = inet_csk(sk);
383 if (icsk->icsk_ack.quick) {
384 if (pkts >= icsk->icsk_ack.quick) {
385 icsk->icsk_ack.quick = 0;
386 /* Leaving quickack mode we deflate ATO. */
387 icsk->icsk_ack.ato = TCP_ATO_MIN;
389 icsk->icsk_ack.quick -= pkts;
394 #define TCP_ECN_QUEUE_CWR 2
395 #define TCP_ECN_DEMAND_CWR 4
396 #define TCP_ECN_SEEN 8
406 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
408 const struct tcphdr *th);
409 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
410 struct request_sock *req, bool fastopen);
411 int tcp_child_process(struct sock *parent, struct sock *child,
412 struct sk_buff *skb);
413 void tcp_enter_loss(struct sock *sk);
414 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
415 void tcp_clear_retrans(struct tcp_sock *tp);
416 void tcp_update_metrics(struct sock *sk);
417 void tcp_init_metrics(struct sock *sk);
418 void tcp_metrics_init(void);
419 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
420 void tcp_disable_fack(struct tcp_sock *tp);
421 void tcp_close(struct sock *sk, long timeout);
422 void tcp_init_sock(struct sock *sk);
423 unsigned int tcp_poll(struct file *file, struct socket *sock,
424 struct poll_table_struct *wait);
425 int tcp_getsockopt(struct sock *sk, int level, int optname,
426 char __user *optval, int __user *optlen);
427 int tcp_setsockopt(struct sock *sk, int level, int optname,
428 char __user *optval, unsigned int optlen);
429 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, int __user *optlen);
431 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
432 char __user *optval, unsigned int optlen);
433 void tcp_set_keepalive(struct sock *sk, int val);
434 void tcp_syn_ack_timeout(const struct request_sock *req);
435 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
436 int flags, int *addr_len);
437 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
438 struct tcp_options_received *opt_rx,
439 int estab, struct tcp_fastopen_cookie *foc);
440 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
443 * TCP v4 functions exported for the inet6 API
446 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
447 void tcp_v4_mtu_reduced(struct sock *sk);
448 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
449 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
450 struct sock *tcp_create_openreq_child(const struct sock *sk,
451 struct request_sock *req,
452 struct sk_buff *skb);
453 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
454 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
455 struct request_sock *req,
456 struct dst_entry *dst,
457 struct request_sock *req_unhash,
459 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
460 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
461 int tcp_connect(struct sock *sk);
462 enum tcp_synack_type {
467 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
468 struct request_sock *req,
469 struct tcp_fastopen_cookie *foc,
470 enum tcp_synack_type synack_type);
471 int tcp_disconnect(struct sock *sk, int flags);
473 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
474 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
475 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
477 /* From syncookies.c */
478 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
479 struct request_sock *req,
480 struct dst_entry *dst, u32 tsoff);
481 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
483 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
484 #ifdef CONFIG_SYN_COOKIES
486 /* Syncookies use a monotonic timer which increments every 60 seconds.
487 * This counter is used both as a hash input and partially encoded into
488 * the cookie value. A cookie is only validated further if the delta
489 * between the current counter value and the encoded one is less than this,
490 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
491 * the counter advances immediately after a cookie is generated).
493 #define MAX_SYNCOOKIE_AGE 2
494 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
495 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
497 /* syncookies: remember time of last synqueue overflow
498 * But do not dirty this field too often (once per second is enough)
499 * It is racy as we do not hold a lock, but race is very minor.
501 static inline void tcp_synq_overflow(const struct sock *sk)
503 unsigned long last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
504 unsigned long now = jiffies;
506 if (!time_between32(now, last_overflow, last_overflow + HZ))
507 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
510 /* syncookies: no recent synqueue overflow on this listening socket? */
511 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
513 unsigned long last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
515 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
516 * then we're under synflood. However, we have to use
517 * 'last_overflow - HZ' as lower bound. That's because a concurrent
518 * tcp_synq_overflow() could update .ts_recent_stamp after we read
519 * jiffies but before we store .ts_recent_stamp into last_overflow,
520 * which could lead to rejecting a valid syncookie.
522 return !time_between32(jiffies, last_overflow - HZ,
523 last_overflow + TCP_SYNCOOKIE_VALID);
526 static inline u32 tcp_cookie_time(void)
528 u64 val = get_jiffies_64();
530 do_div(val, TCP_SYNCOOKIE_PERIOD);
534 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
536 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
537 u64 cookie_init_timestamp(struct request_sock *req);
538 bool cookie_timestamp_decode(const struct net *net,
539 struct tcp_options_received *opt);
540 bool cookie_ecn_ok(const struct tcp_options_received *opt,
541 const struct net *net, const struct dst_entry *dst);
543 /* From net/ipv6/syncookies.c */
544 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
546 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
548 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
549 const struct tcphdr *th, u16 *mssp);
550 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
554 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
556 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
558 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
559 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
560 void tcp_retransmit_timer(struct sock *sk);
561 void tcp_xmit_retransmit_queue(struct sock *);
562 void tcp_simple_retransmit(struct sock *);
563 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
564 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
565 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
567 void tcp_send_probe0(struct sock *);
568 void tcp_send_partial(struct sock *);
569 int tcp_write_wakeup(struct sock *, int mib);
570 void tcp_send_fin(struct sock *sk);
571 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
572 int tcp_send_synack(struct sock *);
573 void tcp_push_one(struct sock *, unsigned int mss_now);
574 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
575 void tcp_send_ack(struct sock *sk);
576 void tcp_send_delayed_ack(struct sock *sk);
577 void tcp_send_loss_probe(struct sock *sk);
578 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
579 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
580 const struct sk_buff *next_skb);
583 void tcp_rearm_rto(struct sock *sk);
584 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
585 void tcp_reset(struct sock *sk);
586 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
587 void tcp_fin(struct sock *sk);
588 void tcp_check_space(struct sock *sk);
591 void tcp_init_xmit_timers(struct sock *);
592 static inline void tcp_clear_xmit_timers(struct sock *sk)
594 hrtimer_cancel(&tcp_sk(sk)->pacing_timer);
595 inet_csk_clear_xmit_timers(sk);
598 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
599 unsigned int tcp_current_mss(struct sock *sk);
601 /* Bound MSS / TSO packet size with the half of the window */
602 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
606 /* When peer uses tiny windows, there is no use in packetizing
607 * to sub-MSS pieces for the sake of SWS or making sure there
608 * are enough packets in the pipe for fast recovery.
610 * On the other hand, for extremely large MSS devices, handling
611 * smaller than MSS windows in this way does make sense.
613 if (tp->max_window > TCP_MSS_DEFAULT)
614 cutoff = (tp->max_window >> 1);
616 cutoff = tp->max_window;
618 if (cutoff && pktsize > cutoff)
619 return max_t(int, cutoff, 68U - tp->tcp_header_len);
625 void tcp_get_info(struct sock *, struct tcp_info *);
627 /* Read 'sendfile()'-style from a TCP socket */
628 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
629 sk_read_actor_t recv_actor);
631 void tcp_initialize_rcv_mss(struct sock *sk);
633 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
634 int tcp_mss_to_mtu(struct sock *sk, int mss);
635 void tcp_mtup_init(struct sock *sk);
636 void tcp_init_buffer_space(struct sock *sk);
638 static inline void tcp_bound_rto(const struct sock *sk)
640 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
641 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
644 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
646 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
649 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
651 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
652 ntohl(TCP_FLAG_ACK) |
656 static inline void tcp_fast_path_on(struct tcp_sock *tp)
658 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
661 static inline void tcp_fast_path_check(struct sock *sk)
663 struct tcp_sock *tp = tcp_sk(sk);
665 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
667 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
669 tcp_fast_path_on(tp);
672 /* Compute the actual rto_min value */
673 static inline u32 tcp_rto_min(struct sock *sk)
675 const struct dst_entry *dst = __sk_dst_get(sk);
676 u32 rto_min = TCP_RTO_MIN;
678 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
679 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
683 static inline u32 tcp_rto_min_us(struct sock *sk)
685 return jiffies_to_usecs(tcp_rto_min(sk));
688 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
690 return dst_metric_locked(dst, RTAX_CC_ALGO);
693 /* Minimum RTT in usec. ~0 means not available. */
694 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
696 return minmax_get(&tp->rtt_min);
699 /* Compute the actual receive window we are currently advertising.
700 * Rcv_nxt can be after the window if our peer push more data
701 * than the offered window.
703 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
705 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
712 /* Choose a new window, without checks for shrinking, and without
713 * scaling applied to the result. The caller does these things
714 * if necessary. This is a "raw" window selection.
716 u32 __tcp_select_window(struct sock *sk);
718 void tcp_send_window_probe(struct sock *sk);
720 /* TCP uses 32bit jiffies to save some space.
721 * Note that this is different from tcp_time_stamp, which
722 * historically has been the same until linux-4.13.
724 #define tcp_jiffies32 ((u32)jiffies)
727 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
728 * It is no longer tied to jiffies, but to 1 ms clock.
729 * Note: double check if you want to use tcp_jiffies32 instead of this.
731 #define TCP_TS_HZ 1000
733 static inline u64 tcp_clock_ns(void)
735 return local_clock();
738 static inline u64 tcp_clock_us(void)
740 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
743 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
744 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
746 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
749 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
750 static inline u32 tcp_time_stamp_raw(void)
752 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
756 /* Refresh 1us clock of a TCP socket,
757 * ensuring monotically increasing values.
759 static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
761 u64 val = tcp_clock_us();
763 if (val > tp->tcp_mstamp)
764 tp->tcp_mstamp = val;
767 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
769 return max_t(s64, t1 - t0, 0);
772 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
774 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
778 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
780 #define TCPHDR_FIN 0x01
781 #define TCPHDR_SYN 0x02
782 #define TCPHDR_RST 0x04
783 #define TCPHDR_PSH 0x08
784 #define TCPHDR_ACK 0x10
785 #define TCPHDR_URG 0x20
786 #define TCPHDR_ECE 0x40
787 #define TCPHDR_CWR 0x80
789 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
791 /* This is what the send packet queuing engine uses to pass
792 * TCP per-packet control information to the transmission code.
793 * We also store the host-order sequence numbers in here too.
794 * This is 44 bytes if IPV6 is enabled.
795 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
798 __u32 seq; /* Starting sequence number */
799 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
801 /* Note : tcp_tw_isn is used in input path only
802 * (isn chosen by tcp_timewait_state_process())
804 * tcp_gso_segs/size are used in write queue only,
805 * cf tcp_skb_pcount()/tcp_skb_mss()
813 /* Used to stash the receive timestamp while this skb is in the
814 * out of order queue, as skb->tstamp is overwritten by the
819 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
821 __u8 sacked; /* State flags for SACK/FACK. */
822 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
823 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
824 #define TCPCB_LOST 0x04 /* SKB is lost */
825 #define TCPCB_TAGBITS 0x07 /* All tag bits */
826 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
827 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
828 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
831 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
832 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
833 eor:1, /* Is skb MSG_EOR marked? */
834 has_rxtstamp:1, /* SKB has a RX timestamp */
836 __u32 ack_seq; /* Sequence number ACK'd */
839 /* There is space for up to 24 bytes */
840 __u32 in_flight:30,/* Bytes in flight at transmit */
841 is_app_limited:1, /* cwnd not fully used? */
843 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
845 /* start of send pipeline phase */
847 /* when we reached the "delivered" count */
848 u64 delivered_mstamp;
849 } tx; /* only used for outgoing skbs */
851 struct inet_skb_parm h4;
852 #if IS_ENABLED(CONFIG_IPV6)
853 struct inet6_skb_parm h6;
855 } header; /* For incoming skbs */
865 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
868 #if IS_ENABLED(CONFIG_IPV6)
869 /* This is the variant of inet6_iif() that must be used by TCP,
870 * as TCP moves IP6CB into a different location in skb->cb[]
872 static inline int tcp_v6_iif(const struct sk_buff *skb)
874 return TCP_SKB_CB(skb)->header.h6.iif;
877 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
879 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
881 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
884 /* TCP_SKB_CB reference means this can not be used from early demux */
885 static inline int tcp_v6_sdif(const struct sk_buff *skb)
887 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
888 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
889 return TCP_SKB_CB(skb)->header.h6.iif;
894 void tcp_v6_early_demux(struct sk_buff *skb);
897 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
899 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
900 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
901 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
907 /* TCP_SKB_CB reference means this can not be used from early demux */
908 static inline int tcp_v4_sdif(struct sk_buff *skb)
910 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
911 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
912 return TCP_SKB_CB(skb)->header.h4.iif;
917 /* Due to TSO, an SKB can be composed of multiple actual
918 * packets. To keep these tracked properly, we use this.
920 static inline int tcp_skb_pcount(const struct sk_buff *skb)
922 return TCP_SKB_CB(skb)->tcp_gso_segs;
925 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
927 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
930 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
932 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
935 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
936 static inline int tcp_skb_mss(const struct sk_buff *skb)
938 return TCP_SKB_CB(skb)->tcp_gso_size;
941 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
943 return likely(!TCP_SKB_CB(skb)->eor);
946 /* Events passed to congestion control interface */
948 CA_EVENT_TX_START, /* first transmit when no packets in flight */
949 CA_EVENT_CWND_RESTART, /* congestion window restart */
950 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
951 CA_EVENT_LOSS, /* loss timeout */
952 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
953 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
956 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
957 enum tcp_ca_ack_event_flags {
958 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
959 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
960 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
964 * Interface for adding new TCP congestion control handlers
966 #define TCP_CA_NAME_MAX 16
967 #define TCP_CA_MAX 128
968 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
970 #define TCP_CA_UNSPEC 0
972 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
973 #define TCP_CONG_NON_RESTRICTED 0x1
974 /* Requires ECN/ECT set on all packets */
975 #define TCP_CONG_NEEDS_ECN 0x2
985 /* A rate sample measures the number of (original/retransmitted) data
986 * packets delivered "delivered" over an interval of time "interval_us".
987 * The tcp_rate.c code fills in the rate sample, and congestion
988 * control modules that define a cong_control function to run at the end
989 * of ACK processing can optionally chose to consult this sample when
990 * setting cwnd and pacing rate.
991 * A sample is invalid if "delivered" or "interval_us" is negative.
994 u64 prior_mstamp; /* starting timestamp for interval */
995 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
996 s32 delivered; /* number of packets delivered over interval */
997 long interval_us; /* time for tp->delivered to incr "delivered" */
998 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
999 int losses; /* number of packets marked lost upon ACK */
1000 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1001 u32 prior_in_flight; /* in flight before this ACK */
1002 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1003 bool is_retrans; /* is sample from retransmission? */
1006 struct tcp_congestion_ops {
1007 struct list_head list;
1011 /* initialize private data (optional) */
1012 void (*init)(struct sock *sk);
1013 /* cleanup private data (optional) */
1014 void (*release)(struct sock *sk);
1016 /* return slow start threshold (required) */
1017 u32 (*ssthresh)(struct sock *sk);
1018 /* do new cwnd calculation (required) */
1019 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1020 /* call before changing ca_state (optional) */
1021 void (*set_state)(struct sock *sk, u8 new_state);
1022 /* call when cwnd event occurs (optional) */
1023 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1024 /* call when ack arrives (optional) */
1025 void (*in_ack_event)(struct sock *sk, u32 flags);
1026 /* new value of cwnd after loss (required) */
1027 u32 (*undo_cwnd)(struct sock *sk);
1028 /* hook for packet ack accounting (optional) */
1029 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1030 /* suggest number of segments for each skb to transmit (optional) */
1031 u32 (*tso_segs_goal)(struct sock *sk);
1032 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1033 u32 (*sndbuf_expand)(struct sock *sk);
1034 /* call when packets are delivered to update cwnd and pacing rate,
1035 * after all the ca_state processing. (optional)
1037 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1038 /* get info for inet_diag (optional) */
1039 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1040 union tcp_cc_info *info);
1042 char name[TCP_CA_NAME_MAX];
1043 struct module *owner;
1046 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1047 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1049 void tcp_assign_congestion_control(struct sock *sk);
1050 void tcp_init_congestion_control(struct sock *sk);
1051 void tcp_cleanup_congestion_control(struct sock *sk);
1052 int tcp_set_default_congestion_control(const char *name);
1053 void tcp_get_default_congestion_control(char *name);
1054 void tcp_get_available_congestion_control(char *buf, size_t len);
1055 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1056 int tcp_set_allowed_congestion_control(char *allowed);
1057 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1058 bool reinit, bool cap_net_admin);
1059 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1060 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1062 u32 tcp_reno_ssthresh(struct sock *sk);
1063 u32 tcp_reno_undo_cwnd(struct sock *sk);
1064 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1065 extern struct tcp_congestion_ops tcp_reno;
1067 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1068 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
1070 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1072 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1078 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1080 const struct inet_connection_sock *icsk = inet_csk(sk);
1082 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1085 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1087 struct inet_connection_sock *icsk = inet_csk(sk);
1089 if (icsk->icsk_ca_ops->set_state)
1090 icsk->icsk_ca_ops->set_state(sk, ca_state);
1091 icsk->icsk_ca_state = ca_state;
1094 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1096 const struct inet_connection_sock *icsk = inet_csk(sk);
1098 if (icsk->icsk_ca_ops->cwnd_event)
1099 icsk->icsk_ca_ops->cwnd_event(sk, event);
1102 /* From tcp_rate.c */
1103 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1104 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1105 struct rate_sample *rs);
1106 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1107 bool is_sack_reneg, struct rate_sample *rs);
1108 void tcp_rate_check_app_limited(struct sock *sk);
1110 /* These functions determine how the current flow behaves in respect of SACK
1111 * handling. SACK is negotiated with the peer, and therefore it can vary
1112 * between different flows.
1114 * tcp_is_sack - SACK enabled
1115 * tcp_is_reno - No SACK
1116 * tcp_is_fack - FACK enabled, implies SACK enabled
1118 static inline int tcp_is_sack(const struct tcp_sock *tp)
1120 return tp->rx_opt.sack_ok;
1123 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1125 return !tcp_is_sack(tp);
1128 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1130 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1133 static inline void tcp_enable_fack(struct tcp_sock *tp)
1135 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1138 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1140 return tp->sacked_out + tp->lost_out;
1143 /* This determines how many packets are "in the network" to the best
1144 * of our knowledge. In many cases it is conservative, but where
1145 * detailed information is available from the receiver (via SACK
1146 * blocks etc.) we can make more aggressive calculations.
1148 * Use this for decisions involving congestion control, use just
1149 * tp->packets_out to determine if the send queue is empty or not.
1151 * Read this equation as:
1153 * "Packets sent once on transmission queue" MINUS
1154 * "Packets left network, but not honestly ACKed yet" PLUS
1155 * "Packets fast retransmitted"
1157 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1159 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1162 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1164 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1166 return tp->snd_cwnd < tp->snd_ssthresh;
1169 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1171 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1174 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1176 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1177 (1 << inet_csk(sk)->icsk_ca_state);
1180 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1181 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1184 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1186 const struct tcp_sock *tp = tcp_sk(sk);
1188 if (tcp_in_cwnd_reduction(sk))
1189 return tp->snd_ssthresh;
1191 return max(tp->snd_ssthresh,
1192 ((tp->snd_cwnd >> 1) +
1193 (tp->snd_cwnd >> 2)));
1196 /* Use define here intentionally to get WARN_ON location shown at the caller */
1197 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1199 void tcp_enter_cwr(struct sock *sk);
1200 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1202 /* The maximum number of MSS of available cwnd for which TSO defers
1203 * sending if not using sysctl_tcp_tso_win_divisor.
1205 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1210 /* Returns end sequence number of the receiver's advertised window */
1211 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1213 return tp->snd_una + tp->snd_wnd;
1216 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1217 * flexible approach. The RFC suggests cwnd should not be raised unless
1218 * it was fully used previously. And that's exactly what we do in
1219 * congestion avoidance mode. But in slow start we allow cwnd to grow
1220 * as long as the application has used half the cwnd.
1222 * cwnd is 10 (IW10), but application sends 9 frames.
1223 * We allow cwnd to reach 18 when all frames are ACKed.
1224 * This check is safe because it's as aggressive as slow start which already
1225 * risks 100% overshoot. The advantage is that we discourage application to
1226 * either send more filler packets or data to artificially blow up the cwnd
1227 * usage, and allow application-limited process to probe bw more aggressively.
1229 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1231 const struct tcp_sock *tp = tcp_sk(sk);
1233 if (tp->is_cwnd_limited)
1236 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1237 if (tcp_in_slow_start(tp))
1238 return tp->snd_cwnd < 2 * tp->max_packets_out;
1243 /* Something is really bad, we could not queue an additional packet,
1244 * because qdisc is full or receiver sent a 0 window.
1245 * We do not want to add fuel to the fire, or abort too early,
1246 * so make sure the timer we arm now is at least 200ms in the future,
1247 * regardless of current icsk_rto value (as it could be ~2ms)
1249 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1251 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1254 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1255 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1256 unsigned long max_when)
1258 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1260 return (unsigned long)min_t(u64, when, max_when);
1263 static inline void tcp_check_probe_timer(struct sock *sk)
1265 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1266 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1267 tcp_probe0_base(sk), TCP_RTO_MAX);
1270 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1275 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1281 * Calculate(/check) TCP checksum
1283 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1284 __be32 daddr, __wsum base)
1286 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1289 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1291 return __skb_checksum_complete(skb);
1294 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1296 return !skb_csum_unnecessary(skb) &&
1297 __tcp_checksum_complete(skb);
1300 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1301 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1306 static const char *statename[]={
1307 "Unused","Established","Syn Sent","Syn Recv",
1308 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1309 "Close Wait","Last ACK","Listen","Closing"
1312 void tcp_set_state(struct sock *sk, int state);
1314 void tcp_done(struct sock *sk);
1316 int tcp_abort(struct sock *sk, int err);
1318 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1321 rx_opt->num_sacks = 0;
1324 u32 tcp_default_init_rwnd(u32 mss);
1325 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1327 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1329 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1330 struct tcp_sock *tp = tcp_sk(sk);
1333 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1334 ca_ops->cong_control)
1336 delta = tcp_jiffies32 - tp->lsndtime;
1337 if (delta > inet_csk(sk)->icsk_rto)
1338 tcp_cwnd_restart(sk, delta);
1341 /* Determine a window scaling and initial window to offer. */
1342 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1343 __u32 *window_clamp, int wscale_ok,
1344 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1346 static inline int tcp_win_from_space(int space)
1348 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1350 return tcp_adv_win_scale <= 0 ?
1351 (space>>(-tcp_adv_win_scale)) :
1352 space - (space>>tcp_adv_win_scale);
1355 /* Note: caller must be prepared to deal with negative returns */
1356 static inline int tcp_space(const struct sock *sk)
1358 return tcp_win_from_space(sk->sk_rcvbuf -
1359 atomic_read(&sk->sk_rmem_alloc));
1362 static inline int tcp_full_space(const struct sock *sk)
1364 return tcp_win_from_space(sk->sk_rcvbuf);
1367 extern void tcp_openreq_init_rwin(struct request_sock *req,
1368 const struct sock *sk_listener,
1369 const struct dst_entry *dst);
1371 void tcp_enter_memory_pressure(struct sock *sk);
1372 void tcp_leave_memory_pressure(struct sock *sk);
1374 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1376 struct net *net = sock_net((struct sock *)tp);
1378 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1381 static inline int keepalive_time_when(const struct tcp_sock *tp)
1383 struct net *net = sock_net((struct sock *)tp);
1385 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1388 static inline int keepalive_probes(const struct tcp_sock *tp)
1390 struct net *net = sock_net((struct sock *)tp);
1392 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1395 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1397 const struct inet_connection_sock *icsk = &tp->inet_conn;
1399 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1400 tcp_jiffies32 - tp->rcv_tstamp);
1403 static inline int tcp_fin_time(const struct sock *sk)
1405 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1406 const int rto = inet_csk(sk)->icsk_rto;
1408 if (fin_timeout < (rto << 2) - (rto >> 1))
1409 fin_timeout = (rto << 2) - (rto >> 1);
1414 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1417 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1419 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1422 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1423 * then following tcp messages have valid values. Ignore 0 value,
1424 * or else 'negative' tsval might forbid us to accept their packets.
1426 if (!rx_opt->ts_recent)
1431 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1434 if (tcp_paws_check(rx_opt, 0))
1437 /* RST segments are not recommended to carry timestamp,
1438 and, if they do, it is recommended to ignore PAWS because
1439 "their cleanup function should take precedence over timestamps."
1440 Certainly, it is mistake. It is necessary to understand the reasons
1441 of this constraint to relax it: if peer reboots, clock may go
1442 out-of-sync and half-open connections will not be reset.
1443 Actually, the problem would be not existing if all
1444 the implementations followed draft about maintaining clock
1445 via reboots. Linux-2.2 DOES NOT!
1447 However, we can relax time bounds for RST segments to MSL.
1449 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1454 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1455 int mib_idx, u32 *last_oow_ack_time);
1457 static inline void tcp_mib_init(struct net *net)
1460 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1461 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1462 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1463 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1467 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1469 tp->lost_skb_hint = NULL;
1472 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1474 tcp_clear_retrans_hints_partial(tp);
1475 tp->retransmit_skb_hint = NULL;
1478 union tcp_md5_addr {
1480 #if IS_ENABLED(CONFIG_IPV6)
1485 /* - key database */
1486 struct tcp_md5sig_key {
1487 struct hlist_node node;
1489 u8 family; /* AF_INET or AF_INET6 */
1490 union tcp_md5_addr addr;
1492 u8 key[TCP_MD5SIG_MAXKEYLEN];
1493 struct rcu_head rcu;
1497 struct tcp_md5sig_info {
1498 struct hlist_head head;
1499 struct rcu_head rcu;
1502 /* - pseudo header */
1503 struct tcp4_pseudohdr {
1511 struct tcp6_pseudohdr {
1512 struct in6_addr saddr;
1513 struct in6_addr daddr;
1515 __be32 protocol; /* including padding */
1518 union tcp_md5sum_block {
1519 struct tcp4_pseudohdr ip4;
1520 #if IS_ENABLED(CONFIG_IPV6)
1521 struct tcp6_pseudohdr ip6;
1525 /* - pool: digest algorithm, hash description and scratch buffer */
1526 struct tcp_md5sig_pool {
1527 struct ahash_request *md5_req;
1532 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1533 const struct sock *sk, const struct sk_buff *skb);
1534 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1535 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1537 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1538 int family, u8 prefixlen);
1539 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1540 const struct sock *addr_sk);
1542 #ifdef CONFIG_TCP_MD5SIG
1543 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1544 const union tcp_md5_addr *addr,
1546 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1548 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1549 const union tcp_md5_addr *addr,
1554 #define tcp_twsk_md5_key(twsk) NULL
1557 bool tcp_alloc_md5sig_pool(void);
1559 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1560 static inline void tcp_put_md5sig_pool(void)
1565 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1566 unsigned int header_len);
1567 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1568 const struct tcp_md5sig_key *key);
1570 /* From tcp_fastopen.c */
1571 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1572 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1573 unsigned long *last_syn_loss);
1574 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1575 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1577 struct tcp_fastopen_request {
1578 /* Fast Open cookie. Size 0 means a cookie request */
1579 struct tcp_fastopen_cookie cookie;
1580 struct msghdr *data; /* data in MSG_FASTOPEN */
1582 int copied; /* queued in tcp_connect() */
1584 void tcp_free_fastopen_req(struct tcp_sock *tp);
1586 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1587 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1588 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1589 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1590 struct request_sock *req,
1591 struct tcp_fastopen_cookie *foc);
1592 void tcp_fastopen_init_key_once(bool publish);
1593 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1594 struct tcp_fastopen_cookie *cookie);
1595 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1596 #define TCP_FASTOPEN_KEY_LENGTH 16
1598 /* Fastopen key context */
1599 struct tcp_fastopen_context {
1600 struct crypto_cipher *tfm;
1601 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1602 struct rcu_head rcu;
1605 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1606 void tcp_fastopen_active_disable(struct sock *sk);
1607 bool tcp_fastopen_active_should_disable(struct sock *sk);
1608 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1609 void tcp_fastopen_active_timeout_reset(void);
1611 /* Latencies incurred by various limits for a sender. They are
1612 * chronograph-like stats that are mutually exclusive.
1616 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1617 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1618 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1622 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1623 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1625 static inline void tcp_init_send_head(struct sock *sk)
1627 sk->sk_send_head = NULL;
1630 /* write queue abstraction */
1631 static inline void tcp_write_queue_purge(struct sock *sk)
1633 struct sk_buff *skb;
1635 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1636 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1637 sk_wmem_free_skb(sk, skb);
1639 tcp_clear_all_retrans_hints(tcp_sk(sk));
1640 tcp_init_send_head(sk);
1641 tcp_sk(sk)->packets_out = 0;
1642 inet_csk(sk)->icsk_backoff = 0;
1645 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1647 return skb_peek(&sk->sk_write_queue);
1650 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1652 return skb_peek_tail(&sk->sk_write_queue);
1655 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1656 const struct sk_buff *skb)
1658 return skb_queue_next(&sk->sk_write_queue, skb);
1661 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1662 const struct sk_buff *skb)
1664 return skb_queue_prev(&sk->sk_write_queue, skb);
1667 #define tcp_for_write_queue(skb, sk) \
1668 skb_queue_walk(&(sk)->sk_write_queue, skb)
1670 #define tcp_for_write_queue_from(skb, sk) \
1671 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1673 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1674 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1676 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1678 return sk->sk_send_head;
1681 static inline bool tcp_skb_is_last(const struct sock *sk,
1682 const struct sk_buff *skb)
1684 return skb_queue_is_last(&sk->sk_write_queue, skb);
1687 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1689 if (tcp_skb_is_last(sk, skb))
1690 sk->sk_send_head = NULL;
1692 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1695 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1697 if (sk->sk_send_head == skb_unlinked) {
1698 sk->sk_send_head = NULL;
1699 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1701 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1702 tcp_sk(sk)->highest_sack = NULL;
1705 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1707 struct sk_buff *skb = tcp_write_queue_head(sk);
1709 if (skb == tcp_send_head(sk))
1715 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1717 struct sk_buff *skb = tcp_send_head(sk);
1719 /* empty retransmit queue, for example due to zero window */
1720 if (skb == tcp_write_queue_head(sk))
1723 return skb ? tcp_write_queue_prev(sk, skb) : tcp_write_queue_tail(sk);
1726 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1728 __skb_queue_tail(&sk->sk_write_queue, skb);
1731 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1733 __tcp_add_write_queue_tail(sk, skb);
1735 /* Queue it, remembering where we must start sending. */
1736 if (sk->sk_send_head == NULL) {
1737 sk->sk_send_head = skb;
1738 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1740 if (tcp_sk(sk)->highest_sack == NULL)
1741 tcp_sk(sk)->highest_sack = skb;
1745 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1747 __skb_queue_head(&sk->sk_write_queue, skb);
1750 /* Insert buff after skb on the write queue of sk. */
1751 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1752 struct sk_buff *buff,
1755 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1758 /* Insert new before skb on the write queue of sk. */
1759 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1760 struct sk_buff *skb,
1763 __skb_queue_before(&sk->sk_write_queue, skb, new);
1765 if (sk->sk_send_head == skb)
1766 sk->sk_send_head = new;
1769 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1771 __skb_unlink(skb, &sk->sk_write_queue);
1774 static inline bool tcp_write_queue_empty(struct sock *sk)
1776 return skb_queue_empty(&sk->sk_write_queue);
1779 static inline void tcp_push_pending_frames(struct sock *sk)
1781 if (tcp_send_head(sk)) {
1782 struct tcp_sock *tp = tcp_sk(sk);
1784 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1788 /* Start sequence of the skb just after the highest skb with SACKed
1789 * bit, valid only if sacked_out > 0 or when the caller has ensured
1790 * validity by itself.
1792 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1794 if (!tp->sacked_out)
1797 if (tp->highest_sack == NULL)
1800 return TCP_SKB_CB(tp->highest_sack)->seq;
1803 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1805 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1806 tcp_write_queue_next(sk, skb);
1809 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1811 return tcp_sk(sk)->highest_sack;
1814 static inline void tcp_highest_sack_reset(struct sock *sk)
1816 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1819 /* Called when old skb is about to be deleted and replaced by new skb */
1820 static inline void tcp_highest_sack_replace(struct sock *sk,
1821 struct sk_buff *old,
1822 struct sk_buff *new)
1824 if (old == tcp_highest_sack(sk))
1825 tcp_sk(sk)->highest_sack = new;
1828 /* This helper checks if socket has IP_TRANSPARENT set */
1829 static inline bool inet_sk_transparent(const struct sock *sk)
1831 switch (sk->sk_state) {
1833 return inet_twsk(sk)->tw_transparent;
1834 case TCP_NEW_SYN_RECV:
1835 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1837 return inet_sk(sk)->transparent;
1840 /* Determines whether this is a thin stream (which may suffer from
1841 * increased latency). Used to trigger latency-reducing mechanisms.
1843 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1845 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1849 enum tcp_seq_states {
1850 TCP_SEQ_STATE_LISTENING,
1851 TCP_SEQ_STATE_ESTABLISHED,
1854 int tcp_seq_open(struct inode *inode, struct file *file);
1856 struct tcp_seq_afinfo {
1859 const struct file_operations *seq_fops;
1860 struct seq_operations seq_ops;
1863 struct tcp_iter_state {
1864 struct seq_net_private p;
1866 enum tcp_seq_states state;
1867 struct sock *syn_wait_sk;
1868 int bucket, offset, sbucket, num;
1872 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1873 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1875 extern struct request_sock_ops tcp_request_sock_ops;
1876 extern struct request_sock_ops tcp6_request_sock_ops;
1878 void tcp_v4_destroy_sock(struct sock *sk);
1880 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1881 netdev_features_t features);
1882 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1883 int tcp_gro_complete(struct sk_buff *skb);
1885 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1887 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1889 struct net *net = sock_net((struct sock *)tp);
1892 val = READ_ONCE(tp->notsent_lowat);
1894 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
1897 static inline bool tcp_stream_memory_free(const struct sock *sk)
1899 const struct tcp_sock *tp = tcp_sk(sk);
1900 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1902 return notsent_bytes < tcp_notsent_lowat(tp);
1905 #ifdef CONFIG_PROC_FS
1906 int tcp4_proc_init(void);
1907 void tcp4_proc_exit(void);
1910 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1911 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1912 const struct tcp_request_sock_ops *af_ops,
1913 struct sock *sk, struct sk_buff *skb);
1915 /* TCP af-specific functions */
1916 struct tcp_sock_af_ops {
1917 #ifdef CONFIG_TCP_MD5SIG
1918 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1919 const struct sock *addr_sk);
1920 int (*calc_md5_hash)(char *location,
1921 const struct tcp_md5sig_key *md5,
1922 const struct sock *sk,
1923 const struct sk_buff *skb);
1924 int (*md5_parse)(struct sock *sk,
1926 char __user *optval,
1931 struct tcp_request_sock_ops {
1933 #ifdef CONFIG_TCP_MD5SIG
1934 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1935 const struct sock *addr_sk);
1936 int (*calc_md5_hash) (char *location,
1937 const struct tcp_md5sig_key *md5,
1938 const struct sock *sk,
1939 const struct sk_buff *skb);
1941 void (*init_req)(struct request_sock *req,
1942 const struct sock *sk_listener,
1943 struct sk_buff *skb);
1944 #ifdef CONFIG_SYN_COOKIES
1945 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1948 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1949 const struct request_sock *req);
1950 u32 (*init_seq)(const struct sk_buff *skb);
1951 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1952 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1953 struct flowi *fl, struct request_sock *req,
1954 struct tcp_fastopen_cookie *foc,
1955 enum tcp_synack_type synack_type);
1958 #ifdef CONFIG_SYN_COOKIES
1959 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1960 const struct sock *sk, struct sk_buff *skb,
1963 tcp_synq_overflow(sk);
1964 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1965 return ops->cookie_init_seq(skb, mss);
1968 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1969 const struct sock *sk, struct sk_buff *skb,
1976 int tcpv4_offload_init(void);
1978 void tcp_v4_init(void);
1979 void tcp_init(void);
1981 /* tcp_recovery.c */
1982 extern bool tcp_rack_mark_lost(struct sock *sk);
1983 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1985 extern void tcp_rack_reo_timeout(struct sock *sk);
1987 /* At how many usecs into the future should the RTO fire? */
1988 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1990 const struct sk_buff *skb = tcp_write_queue_head(sk);
1991 u32 rto = inet_csk(sk)->icsk_rto;
1992 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1994 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1998 * Save and compile IPv4 options, return a pointer to it
2000 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2001 struct sk_buff *skb)
2003 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2004 struct ip_options_rcu *dopt = NULL;
2007 int opt_size = sizeof(*dopt) + opt->optlen;
2009 dopt = kmalloc(opt_size, GFP_ATOMIC);
2010 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2018 /* locally generated TCP pure ACKs have skb->truesize == 2
2019 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2020 * This is much faster than dissecting the packet to find out.
2021 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2023 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2025 return skb->truesize == 2;
2028 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2033 static inline int tcp_inq(struct sock *sk)
2035 struct tcp_sock *tp = tcp_sk(sk);
2038 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2040 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2042 before(tp->urg_seq, tp->copied_seq) ||
2043 !before(tp->urg_seq, tp->rcv_nxt)) {
2045 answ = tp->rcv_nxt - tp->copied_seq;
2047 /* Subtract 1, if FIN was received */
2048 if (answ && sock_flag(sk, SOCK_DONE))
2051 answ = tp->urg_seq - tp->copied_seq;
2057 int tcp_peek_len(struct socket *sock);
2059 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2063 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2064 tp->segs_in += segs_in;
2065 if (skb->len > tcp_hdrlen(skb))
2066 tp->data_segs_in += segs_in;
2070 * TCP listen path runs lockless.
2071 * We forced "struct sock" to be const qualified to make sure
2072 * we don't modify one of its field by mistake.
2073 * Here, we increment sk_drops which is an atomic_t, so we can safely
2074 * make sock writable again.
2076 static inline void tcp_listendrop(const struct sock *sk)
2078 atomic_inc(&((struct sock *)sk)->sk_drops);
2079 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2082 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2085 * Interface for adding Upper Level Protocols over TCP
2088 #define TCP_ULP_NAME_MAX 16
2089 #define TCP_ULP_MAX 128
2090 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2092 struct tcp_ulp_ops {
2093 struct list_head list;
2095 /* initialize ulp */
2096 int (*init)(struct sock *sk);
2098 void (*release)(struct sock *sk);
2100 char name[TCP_ULP_NAME_MAX];
2101 struct module *owner;
2103 int tcp_register_ulp(struct tcp_ulp_ops *type);
2104 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2105 int tcp_set_ulp(struct sock *sk, const char *name);
2106 void tcp_get_available_ulp(char *buf, size_t len);
2107 void tcp_cleanup_ulp(struct sock *sk);
2109 #define MODULE_ALIAS_TCP_ULP(name) \
2110 __MODULE_INFO(alias, alias_userspace, name); \
2111 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2113 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2114 * is < 0, then the BPF op failed (for example if the loaded BPF
2115 * program does not support the chosen operation or there is no BPF
2119 static inline int tcp_call_bpf(struct sock *sk, int op)
2121 struct bpf_sock_ops_kern sock_ops;
2124 if (sk_fullsock(sk))
2125 sock_owned_by_me(sk);
2127 memset(&sock_ops, 0, sizeof(sock_ops));
2131 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2133 ret = sock_ops.reply;
2139 static inline int tcp_call_bpf(struct sock *sk, int op)
2145 static inline u32 tcp_timeout_init(struct sock *sk)
2149 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT);
2152 timeout = TCP_TIMEOUT_INIT;
2156 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2160 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT);
2167 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2169 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1);