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 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 (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
61 #define MAX_TCP_WINDOW 32767U
63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 1024
69 /* probing interval, default to 10 minutes as per RFC4821 */
70 #define TCP_PROBE_INTERVAL 600
72 /* Specify interval when tcp mtu probing will stop */
73 #define TCP_PROBE_THRESHOLD 8
75 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
76 #define TCP_FASTRETRANS_THRESH 3
78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
79 #define TCP_MAX_QUICKACKS 16U
82 #define TCP_URG_VALID 0x0100
83 #define TCP_URG_NOTYET 0x0200
84 #define TCP_URG_READ 0x0400
86 #define TCP_RETR1 3 /*
87 * This is how many retries it does before it
88 * tries to figure out if the gateway is
89 * down. Minimal RFC value is 3; it corresponds
90 * to ~3sec-8min depending on RTO.
93 #define TCP_RETR2 15 /*
94 * This should take at least
95 * 90 minutes to time out.
96 * RFC1122 says that the limit is 100 sec.
97 * 15 is ~13-30min depending on RTO.
100 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
101 * when active opening a connection.
102 * RFC1122 says the minimum retry MUST
103 * be at least 180secs. Nevertheless
104 * this value is corresponding to
105 * 63secs of retransmission with the
106 * current initial RTO.
109 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
110 * when passive opening a connection.
111 * This is corresponding to 31secs of
112 * retransmission with the current
116 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
117 * state, about 60 seconds */
118 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
119 /* BSD style FIN_WAIT2 deadlock breaker.
120 * It used to be 3min, new value is 60sec,
121 * to combine FIN-WAIT-2 timeout with
125 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
127 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
128 #define TCP_ATO_MIN ((unsigned)(HZ/25))
130 #define TCP_DELACK_MIN 4U
131 #define TCP_ATO_MIN 4U
133 #define TCP_RTO_MAX ((unsigned)(120*HZ))
134 #define TCP_RTO_MIN ((unsigned)(HZ/5))
135 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
136 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
137 * used as a fallback RTO for the
138 * initial data transmission if no
139 * valid RTT sample has been acquired,
140 * most likely due to retrans in 3WHS.
143 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
144 * for local resources.
146 #define TCP_REO_TIMEOUT_MIN (2000) /* Min RACK reordering timeout in usec */
148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
150 #define TCP_KEEPALIVE_INTVL (75*HZ)
152 #define MAX_TCP_KEEPIDLE 32767
153 #define MAX_TCP_KEEPINTVL 32767
154 #define MAX_TCP_KEEPCNT 127
155 #define MAX_TCP_SYNCNT 127
157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
161 * after this time. It should be equal
162 * (or greater than) TCP_TIMEWAIT_LEN
163 * to provide reliability equal to one
164 * provided by timewait state.
166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
167 * timestamps. It must be less than
168 * minimal timewait lifetime.
174 #define TCPOPT_NOP 1 /* Padding */
175 #define TCPOPT_EOL 0 /* End of options */
176 #define TCPOPT_MSS 2 /* Segment size negotiating */
177 #define TCPOPT_WINDOW 3 /* Window scaling */
178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
179 #define TCPOPT_SACK 5 /* SACK Block */
180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
183 #define TCPOPT_EXP 254 /* Experimental */
184 /* Magic number to be after the option value for sharing TCP
185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
187 #define TCPOPT_FASTOPEN_MAGIC 0xF989
193 #define TCPOLEN_MSS 4
194 #define TCPOLEN_WINDOW 3
195 #define TCPOLEN_SACK_PERM 2
196 #define TCPOLEN_TIMESTAMP 10
197 #define TCPOLEN_MD5SIG 18
198 #define TCPOLEN_FASTOPEN_BASE 2
199 #define TCPOLEN_EXP_FASTOPEN_BASE 4
201 /* But this is what stacks really send out. */
202 #define TCPOLEN_TSTAMP_ALIGNED 12
203 #define TCPOLEN_WSCALE_ALIGNED 4
204 #define TCPOLEN_SACKPERM_ALIGNED 4
205 #define TCPOLEN_SACK_BASE 2
206 #define TCPOLEN_SACK_BASE_ALIGNED 4
207 #define TCPOLEN_SACK_PERBLOCK 8
208 #define TCPOLEN_MD5SIG_ALIGNED 20
209 #define TCPOLEN_MSS_ALIGNED 4
211 /* Flags in tp->nonagle */
212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
213 #define TCP_NAGLE_CORK 2 /* Socket is corked */
214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
216 /* TCP thin-stream limits */
217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
219 /* TCP initial congestion window as per rfc6928 */
220 #define TCP_INIT_CWND 10
222 /* Bit Flags for sysctl_tcp_fastopen */
223 #define TFO_CLIENT_ENABLE 1
224 #define TFO_SERVER_ENABLE 2
225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
227 /* Accept SYN data w/o any cookie option */
228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
230 /* Force enable TFO on all listeners, i.e., not requiring the
231 * TCP_FASTOPEN socket option.
233 #define TFO_SERVER_WO_SOCKOPT1 0x400
236 /* sysctl variables for tcp */
237 extern int sysctl_tcp_timestamps;
238 extern int sysctl_tcp_window_scaling;
239 extern int sysctl_tcp_sack;
240 extern int sysctl_tcp_fastopen;
241 extern int sysctl_tcp_retrans_collapse;
242 extern int sysctl_tcp_stdurg;
243 extern int sysctl_tcp_rfc1337;
244 extern int sysctl_tcp_abort_on_overflow;
245 extern int sysctl_tcp_max_orphans;
246 extern int sysctl_tcp_fack;
247 extern int sysctl_tcp_reordering;
248 extern int sysctl_tcp_max_reordering;
249 extern int sysctl_tcp_dsack;
250 extern long sysctl_tcp_mem[3];
251 extern int sysctl_tcp_wmem[3];
252 extern int sysctl_tcp_rmem[3];
253 extern int sysctl_tcp_app_win;
254 extern int sysctl_tcp_adv_win_scale;
255 extern int sysctl_tcp_frto;
256 extern int sysctl_tcp_low_latency;
257 extern int sysctl_tcp_nometrics_save;
258 extern int sysctl_tcp_moderate_rcvbuf;
259 extern int sysctl_tcp_tso_win_divisor;
260 extern int sysctl_tcp_workaround_signed_windows;
261 extern int sysctl_tcp_slow_start_after_idle;
262 extern int sysctl_tcp_thin_linear_timeouts;
263 extern int sysctl_tcp_thin_dupack;
264 extern int sysctl_tcp_early_retrans;
265 extern int sysctl_tcp_recovery;
266 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
268 extern int sysctl_tcp_limit_output_bytes;
269 extern int sysctl_tcp_challenge_ack_limit;
270 extern int sysctl_tcp_min_tso_segs;
271 extern int sysctl_tcp_min_rtt_wlen;
272 extern int sysctl_tcp_autocorking;
273 extern int sysctl_tcp_invalid_ratelimit;
274 extern int sysctl_tcp_pacing_ss_ratio;
275 extern int sysctl_tcp_pacing_ca_ratio;
277 extern atomic_long_t tcp_memory_allocated;
278 extern struct percpu_counter tcp_sockets_allocated;
279 extern int tcp_memory_pressure;
281 /* optimized version of sk_under_memory_pressure() for TCP sockets */
282 static inline bool tcp_under_memory_pressure(const struct sock *sk)
284 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
285 mem_cgroup_under_socket_pressure(sk->sk_memcg))
288 return tcp_memory_pressure;
291 * The next routines deal with comparing 32 bit unsigned ints
292 * and worry about wraparound (automatic with unsigned arithmetic).
295 static inline bool before(__u32 seq1, __u32 seq2)
297 return (__s32)(seq1-seq2) < 0;
299 #define after(seq2, seq1) before(seq1, seq2)
301 /* is s2<=s1<=s3 ? */
302 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
304 return seq3 - seq2 >= seq1 - seq2;
307 static inline bool tcp_out_of_memory(struct sock *sk)
309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
315 void sk_forced_mem_schedule(struct sock *sk, int size);
317 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
319 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
320 int orphans = percpu_counter_read_positive(ocp);
322 if (orphans << shift > sysctl_tcp_max_orphans) {
323 orphans = percpu_counter_sum_positive(ocp);
324 if (orphans << shift > sysctl_tcp_max_orphans)
330 bool tcp_check_oom(struct sock *sk, int shift);
333 extern struct proto tcp_prot;
335 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
336 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
337 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
338 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
340 void tcp_tasklet_init(void);
342 void tcp_v4_err(struct sk_buff *skb, u32);
344 void tcp_shutdown(struct sock *sk, int how);
346 void tcp_v4_early_demux(struct sk_buff *skb);
347 int tcp_v4_rcv(struct sk_buff *skb);
349 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
350 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
351 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
353 void tcp_release_cb(struct sock *sk);
354 void tcp_wfree(struct sk_buff *skb);
355 void tcp_write_timer_handler(struct sock *sk);
356 void tcp_delack_timer_handler(struct sock *sk);
357 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
358 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
359 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
360 const struct tcphdr *th, unsigned int len);
361 void tcp_rcv_space_adjust(struct sock *sk);
362 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
363 void tcp_twsk_destructor(struct sock *sk);
364 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
365 struct pipe_inode_info *pipe, size_t len,
368 static inline void tcp_dec_quickack_mode(struct sock *sk,
369 const unsigned int pkts)
371 struct inet_connection_sock *icsk = inet_csk(sk);
373 if (icsk->icsk_ack.quick) {
374 if (pkts >= icsk->icsk_ack.quick) {
375 icsk->icsk_ack.quick = 0;
376 /* Leaving quickack mode we deflate ATO. */
377 icsk->icsk_ack.ato = TCP_ATO_MIN;
379 icsk->icsk_ack.quick -= pkts;
384 #define TCP_ECN_QUEUE_CWR 2
385 #define TCP_ECN_DEMAND_CWR 4
386 #define TCP_ECN_SEEN 8
396 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
398 const struct tcphdr *th);
399 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
400 struct request_sock *req, bool fastopen);
401 int tcp_child_process(struct sock *parent, struct sock *child,
402 struct sk_buff *skb);
403 void tcp_enter_loss(struct sock *sk);
404 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
405 void tcp_clear_retrans(struct tcp_sock *tp);
406 void tcp_update_metrics(struct sock *sk);
407 void tcp_init_metrics(struct sock *sk);
408 void tcp_metrics_init(void);
409 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
410 bool paws_check, bool timestamps);
411 bool tcp_remember_stamp(struct sock *sk);
412 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
413 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
414 void tcp_disable_fack(struct tcp_sock *tp);
415 void tcp_close(struct sock *sk, long timeout);
416 void tcp_init_sock(struct sock *sk);
417 unsigned int tcp_poll(struct file *file, struct socket *sock,
418 struct poll_table_struct *wait);
419 int tcp_getsockopt(struct sock *sk, int level, int optname,
420 char __user *optval, int __user *optlen);
421 int tcp_setsockopt(struct sock *sk, int level, int optname,
422 char __user *optval, unsigned int optlen);
423 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
424 char __user *optval, int __user *optlen);
425 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
426 char __user *optval, unsigned int optlen);
427 void tcp_set_keepalive(struct sock *sk, int val);
428 void tcp_syn_ack_timeout(const struct request_sock *req);
429 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
430 int flags, int *addr_len);
431 void tcp_parse_options(const struct sk_buff *skb,
432 struct tcp_options_received *opt_rx,
433 int estab, struct tcp_fastopen_cookie *foc);
434 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
437 * TCP v4 functions exported for the inet6 API
440 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
441 void tcp_v4_mtu_reduced(struct sock *sk);
442 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
443 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
444 struct sock *tcp_create_openreq_child(const struct sock *sk,
445 struct request_sock *req,
446 struct sk_buff *skb);
447 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
448 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
449 struct request_sock *req,
450 struct dst_entry *dst,
451 struct request_sock *req_unhash,
453 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
454 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
455 int tcp_connect(struct sock *sk);
456 enum tcp_synack_type {
461 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
462 struct request_sock *req,
463 struct tcp_fastopen_cookie *foc,
464 enum tcp_synack_type synack_type);
465 int tcp_disconnect(struct sock *sk, int flags);
467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
471 /* From syncookies.c */
472 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
473 struct request_sock *req,
474 struct dst_entry *dst);
475 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
477 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
478 #ifdef CONFIG_SYN_COOKIES
480 /* Syncookies use a monotonic timer which increments every 60 seconds.
481 * This counter is used both as a hash input and partially encoded into
482 * the cookie value. A cookie is only validated further if the delta
483 * between the current counter value and the encoded one is less than this,
484 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
485 * the counter advances immediately after a cookie is generated).
487 #define MAX_SYNCOOKIE_AGE 2
488 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
489 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
491 /* syncookies: remember time of last synqueue overflow
492 * But do not dirty this field too often (once per second is enough)
493 * It is racy as we do not hold a lock, but race is very minor.
495 static inline void tcp_synq_overflow(const struct sock *sk)
497 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
498 unsigned long now = jiffies;
500 if (time_after(now, last_overflow + HZ))
501 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
504 /* syncookies: no recent synqueue overflow on this listening socket? */
505 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
507 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
509 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
512 static inline u32 tcp_cookie_time(void)
514 u64 val = get_jiffies_64();
516 do_div(val, TCP_SYNCOOKIE_PERIOD);
520 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
522 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
523 __u32 cookie_init_timestamp(struct request_sock *req);
524 bool cookie_timestamp_decode(struct tcp_options_received *opt);
525 bool cookie_ecn_ok(const struct tcp_options_received *opt,
526 const struct net *net, const struct dst_entry *dst);
528 /* From net/ipv6/syncookies.c */
529 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
531 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
533 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
534 const struct tcphdr *th, u16 *mssp);
535 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
539 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
541 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
543 bool tcp_may_send_now(struct sock *sk);
544 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
545 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
546 void tcp_retransmit_timer(struct sock *sk);
547 void tcp_xmit_retransmit_queue(struct sock *);
548 void tcp_simple_retransmit(struct sock *);
549 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
550 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
551 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
553 void tcp_send_probe0(struct sock *);
554 void tcp_send_partial(struct sock *);
555 int tcp_write_wakeup(struct sock *, int mib);
556 void tcp_send_fin(struct sock *sk);
557 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
558 int tcp_send_synack(struct sock *);
559 void tcp_push_one(struct sock *, unsigned int mss_now);
560 void tcp_send_ack(struct sock *sk);
561 void tcp_send_delayed_ack(struct sock *sk);
562 void tcp_send_loss_probe(struct sock *sk);
563 bool tcp_schedule_loss_probe(struct sock *sk);
564 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
565 const struct sk_buff *next_skb);
568 void tcp_rearm_rto(struct sock *sk);
569 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
570 void tcp_reset(struct sock *sk);
571 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
572 void tcp_fin(struct sock *sk);
575 void tcp_init_xmit_timers(struct sock *);
576 static inline void tcp_clear_xmit_timers(struct sock *sk)
578 inet_csk_clear_xmit_timers(sk);
581 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
582 unsigned int tcp_current_mss(struct sock *sk);
584 /* Bound MSS / TSO packet size with the half of the window */
585 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
589 /* When peer uses tiny windows, there is no use in packetizing
590 * to sub-MSS pieces for the sake of SWS or making sure there
591 * are enough packets in the pipe for fast recovery.
593 * On the other hand, for extremely large MSS devices, handling
594 * smaller than MSS windows in this way does make sense.
596 if (tp->max_window > TCP_MSS_DEFAULT)
597 cutoff = (tp->max_window >> 1);
599 cutoff = tp->max_window;
601 if (cutoff && pktsize > cutoff)
602 return max_t(int, cutoff, 68U - tp->tcp_header_len);
608 void tcp_get_info(struct sock *, struct tcp_info *);
610 /* Read 'sendfile()'-style from a TCP socket */
611 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
612 sk_read_actor_t recv_actor);
614 void tcp_initialize_rcv_mss(struct sock *sk);
616 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
617 int tcp_mss_to_mtu(struct sock *sk, int mss);
618 void tcp_mtup_init(struct sock *sk);
619 void tcp_init_buffer_space(struct sock *sk);
621 static inline void tcp_bound_rto(const struct sock *sk)
623 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
624 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
627 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
629 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
632 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
634 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
635 ntohl(TCP_FLAG_ACK) |
639 static inline void tcp_fast_path_on(struct tcp_sock *tp)
641 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
644 static inline void tcp_fast_path_check(struct sock *sk)
646 struct tcp_sock *tp = tcp_sk(sk);
648 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
650 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
652 tcp_fast_path_on(tp);
655 /* Compute the actual rto_min value */
656 static inline u32 tcp_rto_min(struct sock *sk)
658 const struct dst_entry *dst = __sk_dst_get(sk);
659 u32 rto_min = TCP_RTO_MIN;
661 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
662 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
666 static inline u32 tcp_rto_min_us(struct sock *sk)
668 return jiffies_to_usecs(tcp_rto_min(sk));
671 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
673 return dst_metric_locked(dst, RTAX_CC_ALGO);
676 /* Minimum RTT in usec. ~0 means not available. */
677 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
679 return minmax_get(&tp->rtt_min);
682 /* Compute the actual receive window we are currently advertising.
683 * Rcv_nxt can be after the window if our peer push more data
684 * than the offered window.
686 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
688 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
695 /* Choose a new window, without checks for shrinking, and without
696 * scaling applied to the result. The caller does these things
697 * if necessary. This is a "raw" window selection.
699 u32 __tcp_select_window(struct sock *sk);
701 void tcp_send_window_probe(struct sock *sk);
703 /* TCP timestamps are only 32-bits, this causes a slight
704 * complication on 64-bit systems since we store a snapshot
705 * of jiffies in the buffer control blocks below. We decided
706 * to use only the low 32-bits of jiffies and hide the ugly
707 * casts with the following macro.
709 #define tcp_time_stamp ((__u32)(jiffies))
711 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
713 return skb->skb_mstamp.stamp_jiffies;
717 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
719 #define TCPHDR_FIN 0x01
720 #define TCPHDR_SYN 0x02
721 #define TCPHDR_RST 0x04
722 #define TCPHDR_PSH 0x08
723 #define TCPHDR_ACK 0x10
724 #define TCPHDR_URG 0x20
725 #define TCPHDR_ECE 0x40
726 #define TCPHDR_CWR 0x80
728 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
730 /* This is what the send packet queuing engine uses to pass
731 * TCP per-packet control information to the transmission code.
732 * We also store the host-order sequence numbers in here too.
733 * This is 44 bytes if IPV6 is enabled.
734 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
737 __u32 seq; /* Starting sequence number */
738 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
740 /* Note : tcp_tw_isn is used in input path only
741 * (isn chosen by tcp_timewait_state_process())
743 * tcp_gso_segs/size are used in write queue only,
744 * cf tcp_skb_pcount()/tcp_skb_mss()
752 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
754 __u8 sacked; /* State flags for SACK/FACK. */
755 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
756 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
757 #define TCPCB_LOST 0x04 /* SKB is lost */
758 #define TCPCB_TAGBITS 0x07 /* All tag bits */
759 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
760 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
761 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
764 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
765 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
766 eor:1, /* Is skb MSG_EOR marked? */
768 __u32 ack_seq; /* Sequence number ACK'd */
771 /* There is space for up to 24 bytes */
772 __u32 in_flight:30,/* Bytes in flight at transmit */
773 is_app_limited:1, /* cwnd not fully used? */
775 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
777 /* start of send pipeline phase */
778 struct skb_mstamp first_tx_mstamp;
779 /* when we reached the "delivered" count */
780 struct skb_mstamp delivered_mstamp;
781 } tx; /* only used for outgoing skbs */
783 struct inet_skb_parm h4;
784 #if IS_ENABLED(CONFIG_IPV6)
785 struct inet6_skb_parm h6;
787 } header; /* For incoming skbs */
791 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
794 #if IS_ENABLED(CONFIG_IPV6)
795 /* This is the variant of inet6_iif() that must be used by TCP,
796 * as TCP moves IP6CB into a different location in skb->cb[]
798 static inline int tcp_v6_iif(const struct sk_buff *skb)
800 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
802 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
806 /* TCP_SKB_CB reference means this can not be used from early demux */
807 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
809 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
810 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
811 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
817 /* Due to TSO, an SKB can be composed of multiple actual
818 * packets. To keep these tracked properly, we use this.
820 static inline int tcp_skb_pcount(const struct sk_buff *skb)
822 return TCP_SKB_CB(skb)->tcp_gso_segs;
825 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
827 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
830 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
832 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
835 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
836 static inline int tcp_skb_mss(const struct sk_buff *skb)
838 return TCP_SKB_CB(skb)->tcp_gso_size;
841 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
843 return likely(!TCP_SKB_CB(skb)->eor);
846 /* Events passed to congestion control interface */
848 CA_EVENT_TX_START, /* first transmit when no packets in flight */
849 CA_EVENT_CWND_RESTART, /* congestion window restart */
850 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
851 CA_EVENT_LOSS, /* loss timeout */
852 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
853 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
854 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
855 CA_EVENT_NON_DELAYED_ACK,
858 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
859 enum tcp_ca_ack_event_flags {
860 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
861 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
862 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
866 * Interface for adding new TCP congestion control handlers
868 #define TCP_CA_NAME_MAX 16
869 #define TCP_CA_MAX 128
870 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
872 #define TCP_CA_UNSPEC 0
874 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
875 #define TCP_CONG_NON_RESTRICTED 0x1
876 /* Requires ECN/ECT set on all packets */
877 #define TCP_CONG_NEEDS_ECN 0x2
887 /* A rate sample measures the number of (original/retransmitted) data
888 * packets delivered "delivered" over an interval of time "interval_us".
889 * The tcp_rate.c code fills in the rate sample, and congestion
890 * control modules that define a cong_control function to run at the end
891 * of ACK processing can optionally chose to consult this sample when
892 * setting cwnd and pacing rate.
893 * A sample is invalid if "delivered" or "interval_us" is negative.
896 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
897 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
898 s32 delivered; /* number of packets delivered over interval */
899 long interval_us; /* time for tp->delivered to incr "delivered" */
900 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
901 int losses; /* number of packets marked lost upon ACK */
902 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
903 u32 prior_in_flight; /* in flight before this ACK */
904 bool is_app_limited; /* is sample from packet with bubble in pipe? */
905 bool is_retrans; /* is sample from retransmission? */
908 struct tcp_congestion_ops {
909 struct list_head list;
913 /* initialize private data (optional) */
914 void (*init)(struct sock *sk);
915 /* cleanup private data (optional) */
916 void (*release)(struct sock *sk);
918 /* return slow start threshold (required) */
919 u32 (*ssthresh)(struct sock *sk);
920 /* do new cwnd calculation (required) */
921 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
922 /* call before changing ca_state (optional) */
923 void (*set_state)(struct sock *sk, u8 new_state);
924 /* call when cwnd event occurs (optional) */
925 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
926 /* call when ack arrives (optional) */
927 void (*in_ack_event)(struct sock *sk, u32 flags);
928 /* new value of cwnd after loss (optional) */
929 u32 (*undo_cwnd)(struct sock *sk);
930 /* hook for packet ack accounting (optional) */
931 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
932 /* suggest number of segments for each skb to transmit (optional) */
933 u32 (*tso_segs_goal)(struct sock *sk);
934 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
935 u32 (*sndbuf_expand)(struct sock *sk);
936 /* call when packets are delivered to update cwnd and pacing rate,
937 * after all the ca_state processing. (optional)
939 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
940 /* get info for inet_diag (optional) */
941 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
942 union tcp_cc_info *info);
944 char name[TCP_CA_NAME_MAX];
945 struct module *owner;
948 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
949 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
951 void tcp_assign_congestion_control(struct sock *sk);
952 void tcp_init_congestion_control(struct sock *sk);
953 void tcp_cleanup_congestion_control(struct sock *sk);
954 int tcp_set_default_congestion_control(const char *name);
955 void tcp_get_default_congestion_control(char *name);
956 void tcp_get_available_congestion_control(char *buf, size_t len);
957 void tcp_get_allowed_congestion_control(char *buf, size_t len);
958 int tcp_set_allowed_congestion_control(char *allowed);
959 int tcp_set_congestion_control(struct sock *sk, const char *name);
960 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
961 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
963 u32 tcp_reno_ssthresh(struct sock *sk);
964 u32 tcp_reno_undo_cwnd(struct sock *sk);
965 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
966 extern struct tcp_congestion_ops tcp_reno;
968 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
969 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
971 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
973 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
979 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
981 const struct inet_connection_sock *icsk = inet_csk(sk);
983 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
986 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
988 struct inet_connection_sock *icsk = inet_csk(sk);
990 if (icsk->icsk_ca_ops->set_state)
991 icsk->icsk_ca_ops->set_state(sk, ca_state);
992 icsk->icsk_ca_state = ca_state;
995 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
997 const struct inet_connection_sock *icsk = inet_csk(sk);
999 if (icsk->icsk_ca_ops->cwnd_event)
1000 icsk->icsk_ca_ops->cwnd_event(sk, event);
1003 /* From tcp_rate.c */
1004 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1005 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1006 struct rate_sample *rs);
1007 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1008 struct skb_mstamp *now, struct rate_sample *rs);
1009 void tcp_rate_check_app_limited(struct sock *sk);
1011 /* These functions determine how the current flow behaves in respect of SACK
1012 * handling. SACK is negotiated with the peer, and therefore it can vary
1013 * between different flows.
1015 * tcp_is_sack - SACK enabled
1016 * tcp_is_reno - No SACK
1017 * tcp_is_fack - FACK enabled, implies SACK enabled
1019 static inline int tcp_is_sack(const struct tcp_sock *tp)
1021 return tp->rx_opt.sack_ok;
1024 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1026 return !tcp_is_sack(tp);
1029 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1031 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1034 static inline void tcp_enable_fack(struct tcp_sock *tp)
1036 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1039 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1041 return tp->sacked_out + tp->lost_out;
1044 /* This determines how many packets are "in the network" to the best
1045 * of our knowledge. In many cases it is conservative, but where
1046 * detailed information is available from the receiver (via SACK
1047 * blocks etc.) we can make more aggressive calculations.
1049 * Use this for decisions involving congestion control, use just
1050 * tp->packets_out to determine if the send queue is empty or not.
1052 * Read this equation as:
1054 * "Packets sent once on transmission queue" MINUS
1055 * "Packets left network, but not honestly ACKed yet" PLUS
1056 * "Packets fast retransmitted"
1058 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1060 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1063 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1065 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1067 return tp->snd_cwnd < tp->snd_ssthresh;
1070 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1072 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1075 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1077 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1078 (1 << inet_csk(sk)->icsk_ca_state);
1081 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1082 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1085 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1087 const struct tcp_sock *tp = tcp_sk(sk);
1089 if (tcp_in_cwnd_reduction(sk))
1090 return tp->snd_ssthresh;
1092 return max(tp->snd_ssthresh,
1093 ((tp->snd_cwnd >> 1) +
1094 (tp->snd_cwnd >> 2)));
1097 /* Use define here intentionally to get WARN_ON location shown at the caller */
1098 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1100 void tcp_enter_cwr(struct sock *sk);
1101 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1103 /* The maximum number of MSS of available cwnd for which TSO defers
1104 * sending if not using sysctl_tcp_tso_win_divisor.
1106 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1111 /* Returns end sequence number of the receiver's advertised window */
1112 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1114 return tp->snd_una + tp->snd_wnd;
1117 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1118 * flexible approach. The RFC suggests cwnd should not be raised unless
1119 * it was fully used previously. And that's exactly what we do in
1120 * congestion avoidance mode. But in slow start we allow cwnd to grow
1121 * as long as the application has used half the cwnd.
1123 * cwnd is 10 (IW10), but application sends 9 frames.
1124 * We allow cwnd to reach 18 when all frames are ACKed.
1125 * This check is safe because it's as aggressive as slow start which already
1126 * risks 100% overshoot. The advantage is that we discourage application to
1127 * either send more filler packets or data to artificially blow up the cwnd
1128 * usage, and allow application-limited process to probe bw more aggressively.
1130 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1132 const struct tcp_sock *tp = tcp_sk(sk);
1134 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1135 if (tcp_in_slow_start(tp))
1136 return tp->snd_cwnd < 2 * tp->max_packets_out;
1138 return tp->is_cwnd_limited;
1141 /* Something is really bad, we could not queue an additional packet,
1142 * because qdisc is full or receiver sent a 0 window.
1143 * We do not want to add fuel to the fire, or abort too early,
1144 * so make sure the timer we arm now is at least 200ms in the future,
1145 * regardless of current icsk_rto value (as it could be ~2ms)
1147 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1149 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1152 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1153 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1154 unsigned long max_when)
1156 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1158 return (unsigned long)min_t(u64, when, max_when);
1161 static inline void tcp_check_probe_timer(struct sock *sk)
1163 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1164 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1165 tcp_probe0_base(sk), TCP_RTO_MAX);
1168 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1173 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1179 * Calculate(/check) TCP checksum
1181 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1182 __be32 daddr, __wsum base)
1184 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1187 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1189 return __skb_checksum_complete(skb);
1192 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1194 return !skb_csum_unnecessary(skb) &&
1195 __tcp_checksum_complete(skb);
1198 /* Prequeue for VJ style copy to user, combined with checksumming. */
1200 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1202 tp->ucopy.task = NULL;
1204 tp->ucopy.memory = 0;
1205 skb_queue_head_init(&tp->ucopy.prequeue);
1208 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1209 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1210 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1215 static const char *statename[]={
1216 "Unused","Established","Syn Sent","Syn Recv",
1217 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1218 "Close Wait","Last ACK","Listen","Closing"
1221 void tcp_set_state(struct sock *sk, int state);
1223 void tcp_done(struct sock *sk);
1225 int tcp_abort(struct sock *sk, int err);
1227 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1230 rx_opt->num_sacks = 0;
1233 u32 tcp_default_init_rwnd(u32 mss);
1234 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1236 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1238 struct tcp_sock *tp = tcp_sk(sk);
1241 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1243 delta = tcp_time_stamp - tp->lsndtime;
1244 if (delta > inet_csk(sk)->icsk_rto)
1245 tcp_cwnd_restart(sk, delta);
1248 /* Determine a window scaling and initial window to offer. */
1249 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1250 __u32 *window_clamp, int wscale_ok,
1251 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1253 static inline int tcp_win_from_space(int space)
1255 return sysctl_tcp_adv_win_scale<=0 ?
1256 (space>>(-sysctl_tcp_adv_win_scale)) :
1257 space - (space>>sysctl_tcp_adv_win_scale);
1260 /* Note: caller must be prepared to deal with negative returns */
1261 static inline int tcp_space(const struct sock *sk)
1263 return tcp_win_from_space(sk->sk_rcvbuf -
1264 atomic_read(&sk->sk_rmem_alloc));
1267 static inline int tcp_full_space(const struct sock *sk)
1269 return tcp_win_from_space(sk->sk_rcvbuf);
1272 extern void tcp_openreq_init_rwin(struct request_sock *req,
1273 const struct sock *sk_listener,
1274 const struct dst_entry *dst);
1276 void tcp_enter_memory_pressure(struct sock *sk);
1278 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1280 struct net *net = sock_net((struct sock *)tp);
1282 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1285 static inline int keepalive_time_when(const struct tcp_sock *tp)
1287 struct net *net = sock_net((struct sock *)tp);
1289 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1292 static inline int keepalive_probes(const struct tcp_sock *tp)
1294 struct net *net = sock_net((struct sock *)tp);
1296 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1299 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1301 const struct inet_connection_sock *icsk = &tp->inet_conn;
1303 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1304 tcp_time_stamp - tp->rcv_tstamp);
1307 static inline int tcp_fin_time(const struct sock *sk)
1309 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1310 const int rto = inet_csk(sk)->icsk_rto;
1312 if (fin_timeout < (rto << 2) - (rto >> 1))
1313 fin_timeout = (rto << 2) - (rto >> 1);
1318 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1321 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1323 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1326 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1327 * then following tcp messages have valid values. Ignore 0 value,
1328 * or else 'negative' tsval might forbid us to accept their packets.
1330 if (!rx_opt->ts_recent)
1335 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1338 if (tcp_paws_check(rx_opt, 0))
1341 /* RST segments are not recommended to carry timestamp,
1342 and, if they do, it is recommended to ignore PAWS because
1343 "their cleanup function should take precedence over timestamps."
1344 Certainly, it is mistake. It is necessary to understand the reasons
1345 of this constraint to relax it: if peer reboots, clock may go
1346 out-of-sync and half-open connections will not be reset.
1347 Actually, the problem would be not existing if all
1348 the implementations followed draft about maintaining clock
1349 via reboots. Linux-2.2 DOES NOT!
1351 However, we can relax time bounds for RST segments to MSL.
1353 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1358 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1359 int mib_idx, u32 *last_oow_ack_time);
1361 static inline void tcp_mib_init(struct net *net)
1364 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1365 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1366 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1367 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1371 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1373 tp->lost_skb_hint = NULL;
1376 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1378 tcp_clear_retrans_hints_partial(tp);
1379 tp->retransmit_skb_hint = NULL;
1382 union tcp_md5_addr {
1384 #if IS_ENABLED(CONFIG_IPV6)
1389 /* - key database */
1390 struct tcp_md5sig_key {
1391 struct hlist_node node;
1393 u8 family; /* AF_INET or AF_INET6 */
1394 union tcp_md5_addr addr;
1395 u8 key[TCP_MD5SIG_MAXKEYLEN];
1396 struct rcu_head rcu;
1400 struct tcp_md5sig_info {
1401 struct hlist_head head;
1402 struct rcu_head rcu;
1405 /* - pseudo header */
1406 struct tcp4_pseudohdr {
1414 struct tcp6_pseudohdr {
1415 struct in6_addr saddr;
1416 struct in6_addr daddr;
1418 __be32 protocol; /* including padding */
1421 union tcp_md5sum_block {
1422 struct tcp4_pseudohdr ip4;
1423 #if IS_ENABLED(CONFIG_IPV6)
1424 struct tcp6_pseudohdr ip6;
1428 /* - pool: digest algorithm, hash description and scratch buffer */
1429 struct tcp_md5sig_pool {
1430 struct ahash_request *md5_req;
1435 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1436 const struct sock *sk, const struct sk_buff *skb);
1437 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1438 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1439 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1441 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1442 const struct sock *addr_sk);
1444 #ifdef CONFIG_TCP_MD5SIG
1445 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1446 const union tcp_md5_addr *addr,
1448 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1450 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1451 const union tcp_md5_addr *addr,
1456 #define tcp_twsk_md5_key(twsk) NULL
1459 bool tcp_alloc_md5sig_pool(void);
1461 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1462 static inline void tcp_put_md5sig_pool(void)
1467 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1468 unsigned int header_len);
1469 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1470 const struct tcp_md5sig_key *key);
1472 /* From tcp_fastopen.c */
1473 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1474 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1475 unsigned long *last_syn_loss);
1476 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1477 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1479 struct tcp_fastopen_request {
1480 /* Fast Open cookie. Size 0 means a cookie request */
1481 struct tcp_fastopen_cookie cookie;
1482 struct msghdr *data; /* data in MSG_FASTOPEN */
1484 int copied; /* queued in tcp_connect() */
1486 void tcp_free_fastopen_req(struct tcp_sock *tp);
1488 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1489 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1490 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1491 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1492 struct request_sock *req,
1493 struct tcp_fastopen_cookie *foc,
1494 struct dst_entry *dst);
1495 void tcp_fastopen_init_key_once(bool publish);
1496 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1497 struct tcp_fastopen_cookie *cookie);
1498 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1499 #define TCP_FASTOPEN_KEY_LENGTH 16
1501 /* Fastopen key context */
1502 struct tcp_fastopen_context {
1503 struct crypto_cipher *tfm;
1504 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1505 struct rcu_head rcu;
1508 /* Latencies incurred by various limits for a sender. They are
1509 * chronograph-like stats that are mutually exclusive.
1513 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1514 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1515 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1519 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1520 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1522 /* write queue abstraction */
1523 static inline void tcp_write_queue_purge(struct sock *sk)
1525 struct sk_buff *skb;
1527 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1528 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1529 sk_wmem_free_skb(sk, skb);
1531 tcp_clear_all_retrans_hints(tcp_sk(sk));
1534 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1536 return skb_peek(&sk->sk_write_queue);
1539 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1541 return skb_peek_tail(&sk->sk_write_queue);
1544 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1545 const struct sk_buff *skb)
1547 return skb_queue_next(&sk->sk_write_queue, skb);
1550 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1551 const struct sk_buff *skb)
1553 return skb_queue_prev(&sk->sk_write_queue, skb);
1556 #define tcp_for_write_queue(skb, sk) \
1557 skb_queue_walk(&(sk)->sk_write_queue, skb)
1559 #define tcp_for_write_queue_from(skb, sk) \
1560 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1562 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1563 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1565 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1567 return sk->sk_send_head;
1570 static inline bool tcp_skb_is_last(const struct sock *sk,
1571 const struct sk_buff *skb)
1573 return skb_queue_is_last(&sk->sk_write_queue, skb);
1576 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1578 if (tcp_skb_is_last(sk, skb))
1579 sk->sk_send_head = NULL;
1581 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1584 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1586 if (sk->sk_send_head == skb_unlinked) {
1587 sk->sk_send_head = NULL;
1588 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1590 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1591 tcp_sk(sk)->highest_sack = NULL;
1594 static inline void tcp_init_send_head(struct sock *sk)
1596 sk->sk_send_head = NULL;
1599 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1601 __skb_queue_tail(&sk->sk_write_queue, skb);
1604 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1606 __tcp_add_write_queue_tail(sk, skb);
1608 /* Queue it, remembering where we must start sending. */
1609 if (sk->sk_send_head == NULL) {
1610 sk->sk_send_head = skb;
1611 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1613 if (tcp_sk(sk)->highest_sack == NULL)
1614 tcp_sk(sk)->highest_sack = skb;
1618 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1620 __skb_queue_head(&sk->sk_write_queue, skb);
1623 /* Insert buff after skb on the write queue of sk. */
1624 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1625 struct sk_buff *buff,
1628 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1631 /* Insert new before skb on the write queue of sk. */
1632 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1633 struct sk_buff *skb,
1636 __skb_queue_before(&sk->sk_write_queue, skb, new);
1638 if (sk->sk_send_head == skb)
1639 sk->sk_send_head = new;
1642 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1644 __skb_unlink(skb, &sk->sk_write_queue);
1647 static inline bool tcp_write_queue_empty(struct sock *sk)
1649 return skb_queue_empty(&sk->sk_write_queue);
1652 static inline void tcp_push_pending_frames(struct sock *sk)
1654 if (tcp_send_head(sk)) {
1655 struct tcp_sock *tp = tcp_sk(sk);
1657 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1661 /* Start sequence of the skb just after the highest skb with SACKed
1662 * bit, valid only if sacked_out > 0 or when the caller has ensured
1663 * validity by itself.
1665 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1667 if (!tp->sacked_out)
1670 if (tp->highest_sack == NULL)
1673 return TCP_SKB_CB(tp->highest_sack)->seq;
1676 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1678 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1679 tcp_write_queue_next(sk, skb);
1682 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1684 return tcp_sk(sk)->highest_sack;
1687 static inline void tcp_highest_sack_reset(struct sock *sk)
1689 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1692 /* Called when old skb is about to be deleted (to be combined with new skb) */
1693 static inline void tcp_highest_sack_combine(struct sock *sk,
1694 struct sk_buff *old,
1695 struct sk_buff *new)
1697 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1698 tcp_sk(sk)->highest_sack = new;
1701 /* This helper checks if socket has IP_TRANSPARENT set */
1702 static inline bool inet_sk_transparent(const struct sock *sk)
1704 switch (sk->sk_state) {
1706 return inet_twsk(sk)->tw_transparent;
1707 case TCP_NEW_SYN_RECV:
1708 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1710 return inet_sk(sk)->transparent;
1713 /* Determines whether this is a thin stream (which may suffer from
1714 * increased latency). Used to trigger latency-reducing mechanisms.
1716 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1718 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1722 enum tcp_seq_states {
1723 TCP_SEQ_STATE_LISTENING,
1724 TCP_SEQ_STATE_ESTABLISHED,
1727 int tcp_seq_open(struct inode *inode, struct file *file);
1729 struct tcp_seq_afinfo {
1732 const struct file_operations *seq_fops;
1733 struct seq_operations seq_ops;
1736 struct tcp_iter_state {
1737 struct seq_net_private p;
1739 enum tcp_seq_states state;
1740 struct sock *syn_wait_sk;
1741 int bucket, offset, sbucket, num;
1745 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1746 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1748 extern struct request_sock_ops tcp_request_sock_ops;
1749 extern struct request_sock_ops tcp6_request_sock_ops;
1751 void tcp_v4_destroy_sock(struct sock *sk);
1753 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1754 netdev_features_t features);
1755 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1756 int tcp_gro_complete(struct sk_buff *skb);
1758 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1760 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1762 struct net *net = sock_net((struct sock *)tp);
1763 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1766 static inline bool tcp_stream_memory_free(const struct sock *sk)
1768 const struct tcp_sock *tp = tcp_sk(sk);
1769 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1771 return notsent_bytes < tcp_notsent_lowat(tp);
1774 #ifdef CONFIG_PROC_FS
1775 int tcp4_proc_init(void);
1776 void tcp4_proc_exit(void);
1779 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1780 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1781 const struct tcp_request_sock_ops *af_ops,
1782 struct sock *sk, struct sk_buff *skb);
1784 /* TCP af-specific functions */
1785 struct tcp_sock_af_ops {
1786 #ifdef CONFIG_TCP_MD5SIG
1787 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1788 const struct sock *addr_sk);
1789 int (*calc_md5_hash)(char *location,
1790 const struct tcp_md5sig_key *md5,
1791 const struct sock *sk,
1792 const struct sk_buff *skb);
1793 int (*md5_parse)(struct sock *sk,
1794 char __user *optval,
1799 struct tcp_request_sock_ops {
1801 #ifdef CONFIG_TCP_MD5SIG
1802 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1803 const struct sock *addr_sk);
1804 int (*calc_md5_hash) (char *location,
1805 const struct tcp_md5sig_key *md5,
1806 const struct sock *sk,
1807 const struct sk_buff *skb);
1809 void (*init_req)(struct request_sock *req,
1810 const struct sock *sk_listener,
1811 struct sk_buff *skb);
1812 #ifdef CONFIG_SYN_COOKIES
1813 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1816 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1817 const struct request_sock *req,
1819 __u32 (*init_seq)(const struct sk_buff *skb, u32 *tsoff);
1820 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1821 struct flowi *fl, struct request_sock *req,
1822 struct tcp_fastopen_cookie *foc,
1823 enum tcp_synack_type synack_type);
1826 #ifdef CONFIG_SYN_COOKIES
1827 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1828 const struct sock *sk, struct sk_buff *skb,
1831 tcp_synq_overflow(sk);
1832 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1833 return ops->cookie_init_seq(skb, mss);
1836 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1837 const struct sock *sk, struct sk_buff *skb,
1844 int tcpv4_offload_init(void);
1846 void tcp_v4_init(void);
1847 void tcp_init(void);
1849 /* tcp_recovery.c */
1850 extern void tcp_rack_mark_lost(struct sock *sk, const struct skb_mstamp *now);
1851 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1852 const struct skb_mstamp *xmit_time,
1853 const struct skb_mstamp *ack_time);
1854 extern void tcp_rack_reo_timeout(struct sock *sk);
1857 * Save and compile IPv4 options, return a pointer to it
1859 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1861 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1862 struct ip_options_rcu *dopt = NULL;
1865 int opt_size = sizeof(*dopt) + opt->optlen;
1867 dopt = kmalloc(opt_size, GFP_ATOMIC);
1868 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1876 /* locally generated TCP pure ACKs have skb->truesize == 2
1877 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1878 * This is much faster than dissecting the packet to find out.
1879 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1881 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1883 return skb->truesize == 2;
1886 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1891 static inline int tcp_inq(struct sock *sk)
1893 struct tcp_sock *tp = tcp_sk(sk);
1896 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1898 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1900 before(tp->urg_seq, tp->copied_seq) ||
1901 !before(tp->urg_seq, tp->rcv_nxt)) {
1903 answ = tp->rcv_nxt - tp->copied_seq;
1905 /* Subtract 1, if FIN was received */
1906 if (answ && sock_flag(sk, SOCK_DONE))
1909 answ = tp->urg_seq - tp->copied_seq;
1915 int tcp_peek_len(struct socket *sock);
1917 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1921 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1922 tp->segs_in += segs_in;
1923 if (skb->len > tcp_hdrlen(skb))
1924 tp->data_segs_in += segs_in;
1928 * TCP listen path runs lockless.
1929 * We forced "struct sock" to be const qualified to make sure
1930 * we don't modify one of its field by mistake.
1931 * Here, we increment sk_drops which is an atomic_t, so we can safely
1932 * make sock writable again.
1934 static inline void tcp_listendrop(const struct sock *sk)
1936 atomic_inc(&((struct sock *)sk)->sk_drops);
1937 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);