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.
147 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
148 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
149 #define TCP_KEEPALIVE_INTVL (75*HZ)
151 #define MAX_TCP_KEEPIDLE 32767
152 #define MAX_TCP_KEEPINTVL 32767
153 #define MAX_TCP_KEEPCNT 127
154 #define MAX_TCP_SYNCNT 127
156 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
158 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
159 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
160 * after this time. It should be equal
161 * (or greater than) TCP_TIMEWAIT_LEN
162 * to provide reliability equal to one
163 * provided by timewait state.
165 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
166 * timestamps. It must be less than
167 * minimal timewait lifetime.
173 #define TCPOPT_NOP 1 /* Padding */
174 #define TCPOPT_EOL 0 /* End of options */
175 #define TCPOPT_MSS 2 /* Segment size negotiating */
176 #define TCPOPT_WINDOW 3 /* Window scaling */
177 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
178 #define TCPOPT_SACK 5 /* SACK Block */
179 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
180 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
181 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
182 #define TCPOPT_EXP 254 /* Experimental */
183 /* Magic number to be after the option value for sharing TCP
184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
186 #define TCPOPT_FASTOPEN_MAGIC 0xF989
192 #define TCPOLEN_MSS 4
193 #define TCPOLEN_WINDOW 3
194 #define TCPOLEN_SACK_PERM 2
195 #define TCPOLEN_TIMESTAMP 10
196 #define TCPOLEN_MD5SIG 18
197 #define TCPOLEN_FASTOPEN_BASE 2
198 #define TCPOLEN_EXP_FASTOPEN_BASE 4
200 /* But this is what stacks really send out. */
201 #define TCPOLEN_TSTAMP_ALIGNED 12
202 #define TCPOLEN_WSCALE_ALIGNED 4
203 #define TCPOLEN_SACKPERM_ALIGNED 4
204 #define TCPOLEN_SACK_BASE 2
205 #define TCPOLEN_SACK_BASE_ALIGNED 4
206 #define TCPOLEN_SACK_PERBLOCK 8
207 #define TCPOLEN_MD5SIG_ALIGNED 20
208 #define TCPOLEN_MSS_ALIGNED 4
210 /* Flags in tp->nonagle */
211 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
212 #define TCP_NAGLE_CORK 2 /* Socket is corked */
213 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
215 /* TCP thin-stream limits */
216 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
218 /* TCP initial congestion window as per rfc6928 */
219 #define TCP_INIT_CWND 10
221 /* Bit Flags for sysctl_tcp_fastopen */
222 #define TFO_CLIENT_ENABLE 1
223 #define TFO_SERVER_ENABLE 2
224 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
226 /* Accept SYN data w/o any cookie option */
227 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
229 /* Force enable TFO on all listeners, i.e., not requiring the
230 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
232 #define TFO_SERVER_WO_SOCKOPT1 0x400
233 #define TFO_SERVER_WO_SOCKOPT2 0x800
235 extern struct inet_timewait_death_row tcp_death_row;
237 /* sysctl variables for tcp */
238 extern int sysctl_tcp_timestamps;
239 extern int sysctl_tcp_window_scaling;
240 extern int sysctl_tcp_sack;
241 extern int sysctl_tcp_fastopen;
242 extern int sysctl_tcp_retrans_collapse;
243 extern int sysctl_tcp_stdurg;
244 extern int sysctl_tcp_rfc1337;
245 extern int sysctl_tcp_abort_on_overflow;
246 extern int sysctl_tcp_max_orphans;
247 extern int sysctl_tcp_fack;
248 extern int sysctl_tcp_reordering;
249 extern int sysctl_tcp_max_reordering;
250 extern int sysctl_tcp_dsack;
251 extern long sysctl_tcp_mem[3];
252 extern int sysctl_tcp_wmem[3];
253 extern int sysctl_tcp_rmem[3];
254 extern int sysctl_tcp_app_win;
255 extern int sysctl_tcp_adv_win_scale;
256 extern int sysctl_tcp_tw_reuse;
257 extern int sysctl_tcp_frto;
258 extern int sysctl_tcp_low_latency;
259 extern int sysctl_tcp_nometrics_save;
260 extern int sysctl_tcp_moderate_rcvbuf;
261 extern int sysctl_tcp_tso_win_divisor;
262 extern int sysctl_tcp_workaround_signed_windows;
263 extern int sysctl_tcp_slow_start_after_idle;
264 extern int sysctl_tcp_thin_linear_timeouts;
265 extern int sysctl_tcp_thin_dupack;
266 extern int sysctl_tcp_early_retrans;
267 extern int sysctl_tcp_limit_output_bytes;
268 extern int sysctl_tcp_challenge_ack_limit;
269 extern int sysctl_tcp_min_tso_segs;
270 extern int sysctl_tcp_min_rtt_wlen;
271 extern int sysctl_tcp_autocorking;
272 extern int sysctl_tcp_invalid_ratelimit;
273 extern int sysctl_tcp_pacing_ss_ratio;
274 extern int sysctl_tcp_pacing_ca_ratio;
276 extern atomic_long_t tcp_memory_allocated;
277 extern struct percpu_counter tcp_sockets_allocated;
278 extern int tcp_memory_pressure;
280 /* optimized version of sk_under_memory_pressure() for TCP sockets */
281 static inline bool tcp_under_memory_pressure(const struct sock *sk)
283 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
284 mem_cgroup_under_socket_pressure(sk->sk_memcg))
287 return tcp_memory_pressure;
290 * The next routines deal with comparing 32 bit unsigned ints
291 * and worry about wraparound (automatic with unsigned arithmetic).
294 static inline bool before(__u32 seq1, __u32 seq2)
296 return (__s32)(seq1-seq2) < 0;
298 #define after(seq2, seq1) before(seq1, seq2)
300 /* is s2<=s1<=s3 ? */
301 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
303 return seq3 - seq2 >= seq1 - seq2;
306 static inline bool tcp_out_of_memory(struct sock *sk)
308 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
309 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
314 void sk_forced_mem_schedule(struct sock *sk, int size);
316 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
318 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
319 int orphans = percpu_counter_read_positive(ocp);
321 if (orphans << shift > sysctl_tcp_max_orphans) {
322 orphans = percpu_counter_sum_positive(ocp);
323 if (orphans << shift > sysctl_tcp_max_orphans)
329 bool tcp_check_oom(struct sock *sk, int shift);
332 extern struct proto tcp_prot;
334 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
335 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
336 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
337 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
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_clear_retrans(struct tcp_sock *tp);
405 void tcp_update_metrics(struct sock *sk);
406 void tcp_init_metrics(struct sock *sk);
407 void tcp_metrics_init(void);
408 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
409 bool paws_check, bool timestamps);
410 bool tcp_remember_stamp(struct sock *sk);
411 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
412 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
413 void tcp_disable_fack(struct tcp_sock *tp);
414 void tcp_close(struct sock *sk, long timeout);
415 void tcp_init_sock(struct sock *sk);
416 unsigned int tcp_poll(struct file *file, struct socket *sock,
417 struct poll_table_struct *wait);
418 int tcp_getsockopt(struct sock *sk, int level, int optname,
419 char __user *optval, int __user *optlen);
420 int tcp_setsockopt(struct sock *sk, int level, int optname,
421 char __user *optval, unsigned int optlen);
422 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
423 char __user *optval, int __user *optlen);
424 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
425 char __user *optval, unsigned int optlen);
426 void tcp_set_keepalive(struct sock *sk, int val);
427 void tcp_syn_ack_timeout(const struct request_sock *req);
428 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
429 int flags, int *addr_len);
430 void tcp_parse_options(const struct sk_buff *skb,
431 struct tcp_options_received *opt_rx,
432 int estab, struct tcp_fastopen_cookie *foc);
433 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
436 * TCP v4 functions exported for the inet6 API
439 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
440 void tcp_v4_mtu_reduced(struct sock *sk);
441 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
442 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
443 struct sock *tcp_create_openreq_child(const struct sock *sk,
444 struct request_sock *req,
445 struct sk_buff *skb);
446 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
447 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
448 struct request_sock *req,
449 struct dst_entry *dst,
450 struct request_sock *req_unhash,
452 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
453 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
454 int tcp_connect(struct sock *sk);
455 enum tcp_synack_type {
460 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
461 struct request_sock *req,
462 struct tcp_fastopen_cookie *foc,
463 enum tcp_synack_type synack_type);
464 int tcp_disconnect(struct sock *sk, int flags);
466 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
467 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
468 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
470 /* From syncookies.c */
471 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
472 struct request_sock *req,
473 struct dst_entry *dst);
474 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
476 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
477 #ifdef CONFIG_SYN_COOKIES
479 /* Syncookies use a monotonic timer which increments every 60 seconds.
480 * This counter is used both as a hash input and partially encoded into
481 * the cookie value. A cookie is only validated further if the delta
482 * between the current counter value and the encoded one is less than this,
483 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
484 * the counter advances immediately after a cookie is generated).
486 #define MAX_SYNCOOKIE_AGE 2
487 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
488 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
490 /* syncookies: remember time of last synqueue overflow
491 * But do not dirty this field too often (once per second is enough)
492 * It is racy as we do not hold a lock, but race is very minor.
494 static inline void tcp_synq_overflow(const struct sock *sk)
496 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
497 unsigned long now = jiffies;
499 if (time_after(now, last_overflow + HZ))
500 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
503 /* syncookies: no recent synqueue overflow on this listening socket? */
504 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
506 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
508 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
511 static inline u32 tcp_cookie_time(void)
513 u64 val = get_jiffies_64();
515 do_div(val, TCP_SYNCOOKIE_PERIOD);
519 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
521 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
522 __u32 cookie_init_timestamp(struct request_sock *req);
523 bool cookie_timestamp_decode(struct tcp_options_received *opt);
524 bool cookie_ecn_ok(const struct tcp_options_received *opt,
525 const struct net *net, const struct dst_entry *dst);
527 /* From net/ipv6/syncookies.c */
528 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
530 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
532 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
533 const struct tcphdr *th, u16 *mssp);
534 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
538 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
540 bool tcp_may_send_now(struct sock *sk);
541 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
542 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
543 void tcp_retransmit_timer(struct sock *sk);
544 void tcp_xmit_retransmit_queue(struct sock *);
545 void tcp_simple_retransmit(struct sock *);
546 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
547 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
549 void tcp_send_probe0(struct sock *);
550 void tcp_send_partial(struct sock *);
551 int tcp_write_wakeup(struct sock *, int mib);
552 void tcp_send_fin(struct sock *sk);
553 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
554 int tcp_send_synack(struct sock *);
555 void tcp_push_one(struct sock *, unsigned int mss_now);
556 void tcp_send_ack(struct sock *sk);
557 void tcp_send_delayed_ack(struct sock *sk);
558 void tcp_send_loss_probe(struct sock *sk);
559 bool tcp_schedule_loss_probe(struct sock *sk);
560 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
561 const struct sk_buff *next_skb);
564 void tcp_resume_early_retransmit(struct sock *sk);
565 void tcp_rearm_rto(struct sock *sk);
566 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
567 void tcp_reset(struct sock *sk);
568 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
569 void tcp_fin(struct sock *sk);
572 void tcp_init_xmit_timers(struct sock *);
573 static inline void tcp_clear_xmit_timers(struct sock *sk)
575 inet_csk_clear_xmit_timers(sk);
578 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
579 unsigned int tcp_current_mss(struct sock *sk);
581 /* Bound MSS / TSO packet size with the half of the window */
582 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
586 /* When peer uses tiny windows, there is no use in packetizing
587 * to sub-MSS pieces for the sake of SWS or making sure there
588 * are enough packets in the pipe for fast recovery.
590 * On the other hand, for extremely large MSS devices, handling
591 * smaller than MSS windows in this way does make sense.
593 if (tp->max_window >= 512)
594 cutoff = (tp->max_window >> 1);
596 cutoff = tp->max_window;
598 if (cutoff && pktsize > cutoff)
599 return max_t(int, cutoff, 68U - tp->tcp_header_len);
605 void tcp_get_info(struct sock *, struct tcp_info *);
607 /* Read 'sendfile()'-style from a TCP socket */
608 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
609 unsigned int, size_t);
610 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
611 sk_read_actor_t recv_actor);
613 void tcp_initialize_rcv_mss(struct sock *sk);
615 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
616 int tcp_mss_to_mtu(struct sock *sk, int mss);
617 void tcp_mtup_init(struct sock *sk);
618 void tcp_init_buffer_space(struct sock *sk);
620 static inline void tcp_bound_rto(const struct sock *sk)
622 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
623 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
626 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
628 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
631 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
633 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
634 ntohl(TCP_FLAG_ACK) |
638 static inline void tcp_fast_path_on(struct tcp_sock *tp)
640 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
643 static inline void tcp_fast_path_check(struct sock *sk)
645 struct tcp_sock *tp = tcp_sk(sk);
647 if (skb_queue_empty(&tp->out_of_order_queue) &&
649 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
651 tcp_fast_path_on(tp);
654 /* Compute the actual rto_min value */
655 static inline u32 tcp_rto_min(struct sock *sk)
657 const struct dst_entry *dst = __sk_dst_get(sk);
658 u32 rto_min = TCP_RTO_MIN;
660 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
661 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
665 static inline u32 tcp_rto_min_us(struct sock *sk)
667 return jiffies_to_usecs(tcp_rto_min(sk));
670 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
672 return dst_metric_locked(dst, RTAX_CC_ALGO);
675 /* Minimum RTT in usec. ~0 means not available. */
676 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
678 return tp->rtt_min[0].rtt;
681 /* Compute the actual receive window we are currently advertising.
682 * Rcv_nxt can be after the window if our peer push more data
683 * than the offered window.
685 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
687 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
694 /* Choose a new window, without checks for shrinking, and without
695 * scaling applied to the result. The caller does these things
696 * if necessary. This is a "raw" window selection.
698 u32 __tcp_select_window(struct sock *sk);
700 void tcp_send_window_probe(struct sock *sk);
702 /* TCP timestamps are only 32-bits, this causes a slight
703 * complication on 64-bit systems since we store a snapshot
704 * of jiffies in the buffer control blocks below. We decided
705 * to use only the low 32-bits of jiffies and hide the ugly
706 * casts with the following macro.
708 #define tcp_time_stamp ((__u32)(jiffies))
710 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
712 return skb->skb_mstamp.stamp_jiffies;
716 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
718 #define TCPHDR_FIN 0x01
719 #define TCPHDR_SYN 0x02
720 #define TCPHDR_RST 0x04
721 #define TCPHDR_PSH 0x08
722 #define TCPHDR_ACK 0x10
723 #define TCPHDR_URG 0x20
724 #define TCPHDR_ECE 0x40
725 #define TCPHDR_CWR 0x80
727 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
729 /* This is what the send packet queuing engine uses to pass
730 * TCP per-packet control information to the transmission code.
731 * We also store the host-order sequence numbers in here too.
732 * This is 44 bytes if IPV6 is enabled.
733 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
736 __u32 seq; /* Starting sequence number */
737 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
739 /* Note : tcp_tw_isn is used in input path only
740 * (isn chosen by tcp_timewait_state_process())
742 * tcp_gso_segs/size are used in write queue only,
743 * cf tcp_skb_pcount()/tcp_skb_mss()
751 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
753 __u8 sacked; /* State flags for SACK/FACK. */
754 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
755 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
756 #define TCPCB_LOST 0x04 /* SKB is lost */
757 #define TCPCB_TAGBITS 0x07 /* All tag bits */
758 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
759 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
760 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
763 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
764 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
766 __u32 ack_seq; /* Sequence number ACK'd */
768 struct inet_skb_parm h4;
769 #if IS_ENABLED(CONFIG_IPV6)
770 struct inet6_skb_parm h6;
772 } header; /* For incoming frames */
775 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
778 #if IS_ENABLED(CONFIG_IPV6)
779 /* This is the variant of inet6_iif() that must be used by TCP,
780 * as TCP moves IP6CB into a different location in skb->cb[]
782 static inline int tcp_v6_iif(const struct sk_buff *skb)
784 return TCP_SKB_CB(skb)->header.h6.iif;
788 /* Due to TSO, an SKB can be composed of multiple actual
789 * packets. To keep these tracked properly, we use this.
791 static inline int tcp_skb_pcount(const struct sk_buff *skb)
793 return TCP_SKB_CB(skb)->tcp_gso_segs;
796 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
798 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
801 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
803 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
806 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
807 static inline int tcp_skb_mss(const struct sk_buff *skb)
809 return TCP_SKB_CB(skb)->tcp_gso_size;
812 /* Events passed to congestion control interface */
814 CA_EVENT_TX_START, /* first transmit when no packets in flight */
815 CA_EVENT_CWND_RESTART, /* congestion window restart */
816 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
817 CA_EVENT_LOSS, /* loss timeout */
818 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
819 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
820 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
821 CA_EVENT_NON_DELAYED_ACK,
824 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
825 enum tcp_ca_ack_event_flags {
826 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
827 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
828 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
832 * Interface for adding new TCP congestion control handlers
834 #define TCP_CA_NAME_MAX 16
835 #define TCP_CA_MAX 128
836 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
838 #define TCP_CA_UNSPEC 0
840 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
841 #define TCP_CONG_NON_RESTRICTED 0x1
842 /* Requires ECN/ECT set on all packets */
843 #define TCP_CONG_NEEDS_ECN 0x2
847 struct tcp_congestion_ops {
848 struct list_head list;
852 /* initialize private data (optional) */
853 void (*init)(struct sock *sk);
854 /* cleanup private data (optional) */
855 void (*release)(struct sock *sk);
857 /* return slow start threshold (required) */
858 u32 (*ssthresh)(struct sock *sk);
859 /* do new cwnd calculation (required) */
860 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
861 /* call before changing ca_state (optional) */
862 void (*set_state)(struct sock *sk, u8 new_state);
863 /* call when cwnd event occurs (optional) */
864 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
865 /* call when ack arrives (optional) */
866 void (*in_ack_event)(struct sock *sk, u32 flags);
867 /* new value of cwnd after loss (optional) */
868 u32 (*undo_cwnd)(struct sock *sk);
869 /* hook for packet ack accounting (optional) */
870 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
871 /* get info for inet_diag (optional) */
872 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
873 union tcp_cc_info *info);
875 char name[TCP_CA_NAME_MAX];
876 struct module *owner;
879 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
880 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
882 void tcp_assign_congestion_control(struct sock *sk);
883 void tcp_init_congestion_control(struct sock *sk);
884 void tcp_cleanup_congestion_control(struct sock *sk);
885 int tcp_set_default_congestion_control(const char *name);
886 void tcp_get_default_congestion_control(char *name);
887 void tcp_get_available_congestion_control(char *buf, size_t len);
888 void tcp_get_allowed_congestion_control(char *buf, size_t len);
889 int tcp_set_allowed_congestion_control(char *allowed);
890 int tcp_set_congestion_control(struct sock *sk, const char *name);
891 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
892 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
894 u32 tcp_reno_ssthresh(struct sock *sk);
895 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
896 extern struct tcp_congestion_ops tcp_reno;
898 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
899 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
901 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
903 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
909 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
911 const struct inet_connection_sock *icsk = inet_csk(sk);
913 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
916 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
918 struct inet_connection_sock *icsk = inet_csk(sk);
920 if (icsk->icsk_ca_ops->set_state)
921 icsk->icsk_ca_ops->set_state(sk, ca_state);
922 icsk->icsk_ca_state = ca_state;
925 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
927 const struct inet_connection_sock *icsk = inet_csk(sk);
929 if (icsk->icsk_ca_ops->cwnd_event)
930 icsk->icsk_ca_ops->cwnd_event(sk, event);
933 /* These functions determine how the current flow behaves in respect of SACK
934 * handling. SACK is negotiated with the peer, and therefore it can vary
935 * between different flows.
937 * tcp_is_sack - SACK enabled
938 * tcp_is_reno - No SACK
939 * tcp_is_fack - FACK enabled, implies SACK enabled
941 static inline int tcp_is_sack(const struct tcp_sock *tp)
943 return tp->rx_opt.sack_ok;
946 static inline bool tcp_is_reno(const struct tcp_sock *tp)
948 return !tcp_is_sack(tp);
951 static inline bool tcp_is_fack(const struct tcp_sock *tp)
953 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
956 static inline void tcp_enable_fack(struct tcp_sock *tp)
958 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
961 /* TCP early-retransmit (ER) is similar to but more conservative than
962 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
964 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
966 struct net *net = sock_net((struct sock *)tp);
968 tp->do_early_retrans = sysctl_tcp_early_retrans &&
969 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
970 net->ipv4.sysctl_tcp_reordering == 3;
973 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
975 tp->do_early_retrans = 0;
978 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
980 return tp->sacked_out + tp->lost_out;
983 /* This determines how many packets are "in the network" to the best
984 * of our knowledge. In many cases it is conservative, but where
985 * detailed information is available from the receiver (via SACK
986 * blocks etc.) we can make more aggressive calculations.
988 * Use this for decisions involving congestion control, use just
989 * tp->packets_out to determine if the send queue is empty or not.
991 * Read this equation as:
993 * "Packets sent once on transmission queue" MINUS
994 * "Packets left network, but not honestly ACKed yet" PLUS
995 * "Packets fast retransmitted"
997 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
999 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1002 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1004 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1006 return tp->snd_cwnd < tp->snd_ssthresh;
1009 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1011 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1014 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1016 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1017 (1 << inet_csk(sk)->icsk_ca_state);
1020 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1021 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1024 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1026 const struct tcp_sock *tp = tcp_sk(sk);
1028 if (tcp_in_cwnd_reduction(sk))
1029 return tp->snd_ssthresh;
1031 return max(tp->snd_ssthresh,
1032 ((tp->snd_cwnd >> 1) +
1033 (tp->snd_cwnd >> 2)));
1036 /* Use define here intentionally to get WARN_ON location shown at the caller */
1037 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1039 void tcp_enter_cwr(struct sock *sk);
1040 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1042 /* The maximum number of MSS of available cwnd for which TSO defers
1043 * sending if not using sysctl_tcp_tso_win_divisor.
1045 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1050 /* Returns end sequence number of the receiver's advertised window */
1051 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1053 return tp->snd_una + tp->snd_wnd;
1056 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1057 * flexible approach. The RFC suggests cwnd should not be raised unless
1058 * it was fully used previously. And that's exactly what we do in
1059 * congestion avoidance mode. But in slow start we allow cwnd to grow
1060 * as long as the application has used half the cwnd.
1062 * cwnd is 10 (IW10), but application sends 9 frames.
1063 * We allow cwnd to reach 18 when all frames are ACKed.
1064 * This check is safe because it's as aggressive as slow start which already
1065 * risks 100% overshoot. The advantage is that we discourage application to
1066 * either send more filler packets or data to artificially blow up the cwnd
1067 * usage, and allow application-limited process to probe bw more aggressively.
1069 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1071 const struct tcp_sock *tp = tcp_sk(sk);
1073 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1074 if (tcp_in_slow_start(tp))
1075 return tp->snd_cwnd < 2 * tp->max_packets_out;
1077 return tp->is_cwnd_limited;
1080 /* Something is really bad, we could not queue an additional packet,
1081 * because qdisc is full or receiver sent a 0 window.
1082 * We do not want to add fuel to the fire, or abort too early,
1083 * so make sure the timer we arm now is at least 200ms in the future,
1084 * regardless of current icsk_rto value (as it could be ~2ms)
1086 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1088 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1091 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1092 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1093 unsigned long max_when)
1095 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1097 return (unsigned long)min_t(u64, when, max_when);
1100 static inline void tcp_check_probe_timer(struct sock *sk)
1102 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1103 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1104 tcp_probe0_base(sk), TCP_RTO_MAX);
1107 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1112 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1118 * Calculate(/check) TCP checksum
1120 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1121 __be32 daddr, __wsum base)
1123 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1126 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1128 return __skb_checksum_complete(skb);
1131 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1133 return !skb_csum_unnecessary(skb) &&
1134 __tcp_checksum_complete(skb);
1137 /* Prequeue for VJ style copy to user, combined with checksumming. */
1139 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1141 tp->ucopy.task = NULL;
1143 tp->ucopy.memory = 0;
1144 skb_queue_head_init(&tp->ucopy.prequeue);
1147 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1152 static const char *statename[]={
1153 "Unused","Established","Syn Sent","Syn Recv",
1154 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1155 "Close Wait","Last ACK","Listen","Closing"
1158 void tcp_set_state(struct sock *sk, int state);
1160 void tcp_done(struct sock *sk);
1162 int tcp_abort(struct sock *sk, int err);
1164 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1167 rx_opt->num_sacks = 0;
1170 u32 tcp_default_init_rwnd(u32 mss);
1171 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1173 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1175 struct tcp_sock *tp = tcp_sk(sk);
1178 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1180 delta = tcp_time_stamp - tp->lsndtime;
1181 if (delta > inet_csk(sk)->icsk_rto)
1182 tcp_cwnd_restart(sk, delta);
1185 /* Determine a window scaling and initial window to offer. */
1186 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1187 __u32 *window_clamp, int wscale_ok,
1188 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1190 static inline int tcp_win_from_space(int space)
1192 return sysctl_tcp_adv_win_scale<=0 ?
1193 (space>>(-sysctl_tcp_adv_win_scale)) :
1194 space - (space>>sysctl_tcp_adv_win_scale);
1197 /* Note: caller must be prepared to deal with negative returns */
1198 static inline int tcp_space(const struct sock *sk)
1200 return tcp_win_from_space(sk->sk_rcvbuf -
1201 atomic_read(&sk->sk_rmem_alloc));
1204 static inline int tcp_full_space(const struct sock *sk)
1206 return tcp_win_from_space(sk->sk_rcvbuf);
1209 extern void tcp_openreq_init_rwin(struct request_sock *req,
1210 const struct sock *sk_listener,
1211 const struct dst_entry *dst);
1213 void tcp_enter_memory_pressure(struct sock *sk);
1215 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1217 struct net *net = sock_net((struct sock *)tp);
1219 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1222 static inline int keepalive_time_when(const struct tcp_sock *tp)
1224 struct net *net = sock_net((struct sock *)tp);
1226 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1229 static inline int keepalive_probes(const struct tcp_sock *tp)
1231 struct net *net = sock_net((struct sock *)tp);
1233 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1236 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1238 const struct inet_connection_sock *icsk = &tp->inet_conn;
1240 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1241 tcp_time_stamp - tp->rcv_tstamp);
1244 static inline int tcp_fin_time(const struct sock *sk)
1246 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1247 const int rto = inet_csk(sk)->icsk_rto;
1249 if (fin_timeout < (rto << 2) - (rto >> 1))
1250 fin_timeout = (rto << 2) - (rto >> 1);
1255 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1258 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1260 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1263 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1264 * then following tcp messages have valid values. Ignore 0 value,
1265 * or else 'negative' tsval might forbid us to accept their packets.
1267 if (!rx_opt->ts_recent)
1272 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1275 if (tcp_paws_check(rx_opt, 0))
1278 /* RST segments are not recommended to carry timestamp,
1279 and, if they do, it is recommended to ignore PAWS because
1280 "their cleanup function should take precedence over timestamps."
1281 Certainly, it is mistake. It is necessary to understand the reasons
1282 of this constraint to relax it: if peer reboots, clock may go
1283 out-of-sync and half-open connections will not be reset.
1284 Actually, the problem would be not existing if all
1285 the implementations followed draft about maintaining clock
1286 via reboots. Linux-2.2 DOES NOT!
1288 However, we can relax time bounds for RST segments to MSL.
1290 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1295 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1296 int mib_idx, u32 *last_oow_ack_time);
1298 static inline void tcp_mib_init(struct net *net)
1301 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1302 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1303 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1304 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1308 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1310 tp->lost_skb_hint = NULL;
1313 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1315 tcp_clear_retrans_hints_partial(tp);
1316 tp->retransmit_skb_hint = NULL;
1319 union tcp_md5_addr {
1321 #if IS_ENABLED(CONFIG_IPV6)
1326 /* - key database */
1327 struct tcp_md5sig_key {
1328 struct hlist_node node;
1330 u8 family; /* AF_INET or AF_INET6 */
1331 union tcp_md5_addr addr;
1332 u8 key[TCP_MD5SIG_MAXKEYLEN];
1333 struct rcu_head rcu;
1337 struct tcp_md5sig_info {
1338 struct hlist_head head;
1339 struct rcu_head rcu;
1342 /* - pseudo header */
1343 struct tcp4_pseudohdr {
1351 struct tcp6_pseudohdr {
1352 struct in6_addr saddr;
1353 struct in6_addr daddr;
1355 __be32 protocol; /* including padding */
1358 union tcp_md5sum_block {
1359 struct tcp4_pseudohdr ip4;
1360 #if IS_ENABLED(CONFIG_IPV6)
1361 struct tcp6_pseudohdr ip6;
1365 /* - pool: digest algorithm, hash description and scratch buffer */
1366 struct tcp_md5sig_pool {
1367 struct ahash_request *md5_req;
1368 union tcp_md5sum_block md5_blk;
1372 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1373 const struct sock *sk, const struct sk_buff *skb);
1374 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1375 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1376 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1378 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1379 const struct sock *addr_sk);
1381 #ifdef CONFIG_TCP_MD5SIG
1382 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1383 const union tcp_md5_addr *addr,
1385 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1387 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1388 const union tcp_md5_addr *addr,
1393 #define tcp_twsk_md5_key(twsk) NULL
1396 bool tcp_alloc_md5sig_pool(void);
1398 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1399 static inline void tcp_put_md5sig_pool(void)
1404 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1405 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1406 unsigned int header_len);
1407 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1408 const struct tcp_md5sig_key *key);
1410 /* From tcp_fastopen.c */
1411 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1412 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1413 unsigned long *last_syn_loss);
1414 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1415 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1417 struct tcp_fastopen_request {
1418 /* Fast Open cookie. Size 0 means a cookie request */
1419 struct tcp_fastopen_cookie cookie;
1420 struct msghdr *data; /* data in MSG_FASTOPEN */
1422 int copied; /* queued in tcp_connect() */
1424 void tcp_free_fastopen_req(struct tcp_sock *tp);
1426 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1427 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1428 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1429 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1430 struct request_sock *req,
1431 struct tcp_fastopen_cookie *foc,
1432 struct dst_entry *dst);
1433 void tcp_fastopen_init_key_once(bool publish);
1434 #define TCP_FASTOPEN_KEY_LENGTH 16
1436 /* Fastopen key context */
1437 struct tcp_fastopen_context {
1438 struct crypto_cipher *tfm;
1439 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1440 struct rcu_head rcu;
1443 /* write queue abstraction */
1444 static inline void tcp_write_queue_purge(struct sock *sk)
1446 struct sk_buff *skb;
1448 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1449 sk_wmem_free_skb(sk, skb);
1451 tcp_clear_all_retrans_hints(tcp_sk(sk));
1454 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1456 return skb_peek(&sk->sk_write_queue);
1459 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1461 return skb_peek_tail(&sk->sk_write_queue);
1464 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1465 const struct sk_buff *skb)
1467 return skb_queue_next(&sk->sk_write_queue, skb);
1470 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1471 const struct sk_buff *skb)
1473 return skb_queue_prev(&sk->sk_write_queue, skb);
1476 #define tcp_for_write_queue(skb, sk) \
1477 skb_queue_walk(&(sk)->sk_write_queue, skb)
1479 #define tcp_for_write_queue_from(skb, sk) \
1480 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1482 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1483 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1485 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1487 return sk->sk_send_head;
1490 static inline bool tcp_skb_is_last(const struct sock *sk,
1491 const struct sk_buff *skb)
1493 return skb_queue_is_last(&sk->sk_write_queue, skb);
1496 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1498 if (tcp_skb_is_last(sk, skb))
1499 sk->sk_send_head = NULL;
1501 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1504 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1506 if (sk->sk_send_head == skb_unlinked)
1507 sk->sk_send_head = NULL;
1510 static inline void tcp_init_send_head(struct sock *sk)
1512 sk->sk_send_head = NULL;
1515 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1517 __skb_queue_tail(&sk->sk_write_queue, skb);
1520 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1522 __tcp_add_write_queue_tail(sk, skb);
1524 /* Queue it, remembering where we must start sending. */
1525 if (sk->sk_send_head == NULL) {
1526 sk->sk_send_head = skb;
1528 if (tcp_sk(sk)->highest_sack == NULL)
1529 tcp_sk(sk)->highest_sack = skb;
1533 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1535 __skb_queue_head(&sk->sk_write_queue, skb);
1538 /* Insert buff after skb on the write queue of sk. */
1539 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1540 struct sk_buff *buff,
1543 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1546 /* Insert new before skb on the write queue of sk. */
1547 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1548 struct sk_buff *skb,
1551 __skb_queue_before(&sk->sk_write_queue, skb, new);
1553 if (sk->sk_send_head == skb)
1554 sk->sk_send_head = new;
1557 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1559 __skb_unlink(skb, &sk->sk_write_queue);
1562 static inline bool tcp_write_queue_empty(struct sock *sk)
1564 return skb_queue_empty(&sk->sk_write_queue);
1567 static inline void tcp_push_pending_frames(struct sock *sk)
1569 if (tcp_send_head(sk)) {
1570 struct tcp_sock *tp = tcp_sk(sk);
1572 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1576 /* Start sequence of the skb just after the highest skb with SACKed
1577 * bit, valid only if sacked_out > 0 or when the caller has ensured
1578 * validity by itself.
1580 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1582 if (!tp->sacked_out)
1585 if (tp->highest_sack == NULL)
1588 return TCP_SKB_CB(tp->highest_sack)->seq;
1591 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1593 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1594 tcp_write_queue_next(sk, skb);
1597 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1599 return tcp_sk(sk)->highest_sack;
1602 static inline void tcp_highest_sack_reset(struct sock *sk)
1604 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1607 /* Called when old skb is about to be deleted (to be combined with new skb) */
1608 static inline void tcp_highest_sack_combine(struct sock *sk,
1609 struct sk_buff *old,
1610 struct sk_buff *new)
1612 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1613 tcp_sk(sk)->highest_sack = new;
1616 /* This helper checks if socket has IP_TRANSPARENT set */
1617 static inline bool inet_sk_transparent(const struct sock *sk)
1619 switch (sk->sk_state) {
1621 return inet_twsk(sk)->tw_transparent;
1622 case TCP_NEW_SYN_RECV:
1623 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1625 return inet_sk(sk)->transparent;
1628 /* Determines whether this is a thin stream (which may suffer from
1629 * increased latency). Used to trigger latency-reducing mechanisms.
1631 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1633 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1637 enum tcp_seq_states {
1638 TCP_SEQ_STATE_LISTENING,
1639 TCP_SEQ_STATE_ESTABLISHED,
1642 int tcp_seq_open(struct inode *inode, struct file *file);
1644 struct tcp_seq_afinfo {
1647 const struct file_operations *seq_fops;
1648 struct seq_operations seq_ops;
1651 struct tcp_iter_state {
1652 struct seq_net_private p;
1654 enum tcp_seq_states state;
1655 struct sock *syn_wait_sk;
1656 int bucket, offset, sbucket, num;
1660 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1661 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1663 extern struct request_sock_ops tcp_request_sock_ops;
1664 extern struct request_sock_ops tcp6_request_sock_ops;
1666 void tcp_v4_destroy_sock(struct sock *sk);
1668 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1669 netdev_features_t features);
1670 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1671 int tcp_gro_complete(struct sk_buff *skb);
1673 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1675 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1677 struct net *net = sock_net((struct sock *)tp);
1678 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1681 static inline bool tcp_stream_memory_free(const struct sock *sk)
1683 const struct tcp_sock *tp = tcp_sk(sk);
1684 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1686 return notsent_bytes < tcp_notsent_lowat(tp);
1689 #ifdef CONFIG_PROC_FS
1690 int tcp4_proc_init(void);
1691 void tcp4_proc_exit(void);
1694 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1695 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1696 const struct tcp_request_sock_ops *af_ops,
1697 struct sock *sk, struct sk_buff *skb);
1699 /* TCP af-specific functions */
1700 struct tcp_sock_af_ops {
1701 #ifdef CONFIG_TCP_MD5SIG
1702 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1703 const struct sock *addr_sk);
1704 int (*calc_md5_hash)(char *location,
1705 const struct tcp_md5sig_key *md5,
1706 const struct sock *sk,
1707 const struct sk_buff *skb);
1708 int (*md5_parse)(struct sock *sk,
1709 char __user *optval,
1714 struct tcp_request_sock_ops {
1716 #ifdef CONFIG_TCP_MD5SIG
1717 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1718 const struct sock *addr_sk);
1719 int (*calc_md5_hash) (char *location,
1720 const struct tcp_md5sig_key *md5,
1721 const struct sock *sk,
1722 const struct sk_buff *skb);
1724 void (*init_req)(struct request_sock *req,
1725 const struct sock *sk_listener,
1726 struct sk_buff *skb);
1727 #ifdef CONFIG_SYN_COOKIES
1728 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1731 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1732 const struct request_sock *req,
1734 __u32 (*init_seq)(const struct sk_buff *skb);
1735 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1736 struct flowi *fl, struct request_sock *req,
1737 struct tcp_fastopen_cookie *foc,
1738 enum tcp_synack_type synack_type);
1741 #ifdef CONFIG_SYN_COOKIES
1742 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1743 const struct sock *sk, struct sk_buff *skb,
1746 tcp_synq_overflow(sk);
1747 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1748 return ops->cookie_init_seq(skb, mss);
1751 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1752 const struct sock *sk, struct sk_buff *skb,
1759 int tcpv4_offload_init(void);
1761 void tcp_v4_init(void);
1762 void tcp_init(void);
1764 /* tcp_recovery.c */
1766 /* Flags to enable various loss recovery features. See below */
1767 extern int sysctl_tcp_recovery;
1769 /* Use TCP RACK to detect (some) tail and retransmit losses */
1770 #define TCP_RACK_LOST_RETRANS 0x1
1772 extern int tcp_rack_mark_lost(struct sock *sk);
1774 extern void tcp_rack_advance(struct tcp_sock *tp,
1775 const struct skb_mstamp *xmit_time, u8 sacked);
1778 * Save and compile IPv4 options, return a pointer to it
1780 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1782 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1783 struct ip_options_rcu *dopt = NULL;
1786 int opt_size = sizeof(*dopt) + opt->optlen;
1788 dopt = kmalloc(opt_size, GFP_ATOMIC);
1789 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1797 /* locally generated TCP pure ACKs have skb->truesize == 2
1798 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1799 * This is much faster than dissecting the packet to find out.
1800 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1802 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1804 return skb->truesize == 2;
1807 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1812 static inline int tcp_inq(struct sock *sk)
1814 struct tcp_sock *tp = tcp_sk(sk);
1817 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1819 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1821 before(tp->urg_seq, tp->copied_seq) ||
1822 !before(tp->urg_seq, tp->rcv_nxt)) {
1824 answ = tp->rcv_nxt - tp->copied_seq;
1826 /* Subtract 1, if FIN was received */
1827 if (answ && sock_flag(sk, SOCK_DONE))
1830 answ = tp->urg_seq - tp->copied_seq;
1836 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1840 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1841 tp->segs_in += segs_in;
1842 if (skb->len > tcp_hdrlen(skb))
1843 tp->data_segs_in += segs_in;
1847 * TCP listen path runs lockless.
1848 * We forced "struct sock" to be const qualified to make sure
1849 * we don't modify one of its field by mistake.
1850 * Here, we increment sk_drops which is an atomic_t, so we can safely
1851 * make sock writable again.
1853 static inline void tcp_listendrop(const struct sock *sk)
1855 atomic_inc(&((struct sock *)sk)->sk_drops);
1856 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);