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/crypto.h>
31 #include <linux/cryptohash.h>
32 #include <linux/kref.h>
33 #include <linux/ktime.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo;
52 extern struct percpu_counter tcp_orphan_count;
53 void tcp_time_wait(struct sock *sk, int state, int timeo);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 1024
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL 600
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD 8
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
102 * when active opening a connection.
103 * RFC1122 says the minimum retry MUST
104 * be at least 180secs. Nevertheless
105 * this value is corresponding to
106 * 63secs of retransmission with the
107 * current initial RTO.
110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
111 * when passive opening a connection.
112 * This is corresponding to 31secs of
113 * retransmission with the current
117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 * state, about 60 seconds */
119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
120 /* BSD style FIN_WAIT2 deadlock breaker.
121 * It used to be 3min, new value is 60sec,
122 * to combine FIN-WAIT-2 timeout with
126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
129 #define TCP_ATO_MIN ((unsigned)(HZ/25))
131 #define TCP_DELACK_MIN 4U
132 #define TCP_ATO_MIN 4U
134 #define TCP_RTO_MAX ((unsigned)(120*HZ))
135 #define TCP_RTO_MIN ((unsigned)(HZ/5))
136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
138 * used as a fallback RTO for the
139 * initial data transmission if no
140 * valid RTT sample has been acquired,
141 * most likely due to retrans in 3WHS.
144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
145 * for local resources.
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. SOCKOPT1/2 determine how to set max_qlen.
233 #define TFO_SERVER_WO_SOCKOPT1 0x400
234 #define TFO_SERVER_WO_SOCKOPT2 0x800
236 extern struct inet_timewait_death_row tcp_death_row;
238 /* sysctl variables for tcp */
239 extern int sysctl_tcp_timestamps;
240 extern int sysctl_tcp_window_scaling;
241 extern int sysctl_tcp_sack;
242 extern int sysctl_tcp_fin_timeout;
243 extern int sysctl_tcp_fastopen;
244 extern int sysctl_tcp_retrans_collapse;
245 extern int sysctl_tcp_stdurg;
246 extern int sysctl_tcp_rfc1337;
247 extern int sysctl_tcp_abort_on_overflow;
248 extern int sysctl_tcp_max_orphans;
249 extern int sysctl_tcp_fack;
250 extern int sysctl_tcp_reordering;
251 extern int sysctl_tcp_max_reordering;
252 extern int sysctl_tcp_dsack;
253 extern long sysctl_tcp_mem[3];
254 extern int sysctl_tcp_wmem[3];
255 extern int sysctl_tcp_rmem[3];
256 extern int sysctl_tcp_app_win;
257 extern int sysctl_tcp_adv_win_scale;
258 extern int sysctl_tcp_tw_reuse;
259 extern int sysctl_tcp_frto;
260 extern int sysctl_tcp_low_latency;
261 extern int sysctl_tcp_nometrics_save;
262 extern int sysctl_tcp_moderate_rcvbuf;
263 extern int sysctl_tcp_tso_win_divisor;
264 extern int sysctl_tcp_workaround_signed_windows;
265 extern int sysctl_tcp_slow_start_after_idle;
266 extern int sysctl_tcp_thin_linear_timeouts;
267 extern int sysctl_tcp_thin_dupack;
268 extern int sysctl_tcp_early_retrans;
269 extern int sysctl_tcp_limit_output_bytes;
270 extern int sysctl_tcp_challenge_ack_limit;
271 extern unsigned int sysctl_tcp_notsent_lowat;
272 extern int sysctl_tcp_min_tso_segs;
273 extern int sysctl_tcp_min_rtt_wlen;
274 extern int sysctl_tcp_autocorking;
275 extern int sysctl_tcp_invalid_ratelimit;
276 extern int sysctl_tcp_pacing_ss_ratio;
277 extern int sysctl_tcp_pacing_ca_ratio;
279 extern atomic_long_t tcp_memory_allocated;
280 extern struct percpu_counter tcp_sockets_allocated;
281 extern int tcp_memory_pressure;
283 /* optimized version of sk_under_memory_pressure() for TCP sockets */
284 static inline bool tcp_under_memory_pressure(const struct sock *sk)
286 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
287 mem_cgroup_under_socket_pressure(sk->sk_memcg))
290 return tcp_memory_pressure;
293 * The next routines deal with comparing 32 bit unsigned ints
294 * and worry about wraparound (automatic with unsigned arithmetic).
297 static inline bool before(__u32 seq1, __u32 seq2)
299 return (__s32)(seq1-seq2) < 0;
301 #define after(seq2, seq1) before(seq1, seq2)
303 /* is s2<=s1<=s3 ? */
304 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
306 return seq3 - seq2 >= seq1 - seq2;
309 static inline bool tcp_out_of_memory(struct sock *sk)
311 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
312 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
317 void sk_forced_mem_schedule(struct sock *sk, int size);
319 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
321 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
322 int orphans = percpu_counter_read_positive(ocp);
324 if (orphans << shift > sysctl_tcp_max_orphans) {
325 orphans = percpu_counter_sum_positive(ocp);
326 if (orphans << shift > sysctl_tcp_max_orphans)
332 bool tcp_check_oom(struct sock *sk, int shift);
335 extern struct proto tcp_prot;
337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
338 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
340 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
341 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
343 void tcp_tasklet_init(void);
345 void tcp_v4_err(struct sk_buff *skb, u32);
347 void tcp_shutdown(struct sock *sk, int how);
349 void tcp_v4_early_demux(struct sk_buff *skb);
350 int tcp_v4_rcv(struct sk_buff *skb);
352 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
353 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
354 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
356 void tcp_release_cb(struct sock *sk);
357 void tcp_wfree(struct sk_buff *skb);
358 void tcp_write_timer_handler(struct sock *sk);
359 void tcp_delack_timer_handler(struct sock *sk);
360 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
361 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
362 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
363 const struct tcphdr *th, unsigned int len);
364 void tcp_rcv_space_adjust(struct sock *sk);
365 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
366 void tcp_twsk_destructor(struct sock *sk);
367 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
368 struct pipe_inode_info *pipe, size_t len,
371 static inline void tcp_dec_quickack_mode(struct sock *sk,
372 const unsigned int pkts)
374 struct inet_connection_sock *icsk = inet_csk(sk);
376 if (icsk->icsk_ack.quick) {
377 if (pkts >= icsk->icsk_ack.quick) {
378 icsk->icsk_ack.quick = 0;
379 /* Leaving quickack mode we deflate ATO. */
380 icsk->icsk_ack.ato = TCP_ATO_MIN;
382 icsk->icsk_ack.quick -= pkts;
387 #define TCP_ECN_QUEUE_CWR 2
388 #define TCP_ECN_DEMAND_CWR 4
389 #define TCP_ECN_SEEN 8
399 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
401 const struct tcphdr *th);
402 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
403 struct request_sock *req, bool fastopen);
404 int tcp_child_process(struct sock *parent, struct sock *child,
405 struct sk_buff *skb);
406 void tcp_enter_loss(struct sock *sk);
407 void tcp_clear_retrans(struct tcp_sock *tp);
408 void tcp_update_metrics(struct sock *sk);
409 void tcp_init_metrics(struct sock *sk);
410 void tcp_metrics_init(void);
411 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
412 bool paws_check, bool timestamps);
413 bool tcp_remember_stamp(struct sock *sk);
414 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
415 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
416 void tcp_disable_fack(struct tcp_sock *tp);
417 void tcp_close(struct sock *sk, long timeout);
418 void tcp_init_sock(struct sock *sk);
419 unsigned int tcp_poll(struct file *file, struct socket *sock,
420 struct poll_table_struct *wait);
421 int tcp_getsockopt(struct sock *sk, int level, int optname,
422 char __user *optval, int __user *optlen);
423 int tcp_setsockopt(struct sock *sk, int level, int optname,
424 char __user *optval, unsigned int optlen);
425 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
426 char __user *optval, int __user *optlen);
427 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
428 char __user *optval, unsigned int optlen);
429 void tcp_set_keepalive(struct sock *sk, int val);
430 void tcp_syn_ack_timeout(const struct request_sock *req);
431 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
432 int flags, int *addr_len);
433 void tcp_parse_options(const struct sk_buff *skb,
434 struct tcp_options_received *opt_rx,
435 int estab, struct tcp_fastopen_cookie *foc);
436 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
439 * TCP v4 functions exported for the inet6 API
442 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
443 void tcp_v4_mtu_reduced(struct sock *sk);
444 void tcp_req_err(struct sock *sk, u32 seq);
445 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
446 struct sock *tcp_create_openreq_child(const struct sock *sk,
447 struct request_sock *req,
448 struct sk_buff *skb);
449 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
450 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
451 struct request_sock *req,
452 struct dst_entry *dst,
453 struct request_sock *req_unhash,
455 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
456 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
457 int tcp_connect(struct sock *sk);
458 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
459 struct request_sock *req,
460 struct tcp_fastopen_cookie *foc,
462 int tcp_disconnect(struct sock *sk, int flags);
464 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
465 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
466 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
468 /* From syncookies.c */
469 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
470 struct request_sock *req,
471 struct dst_entry *dst);
472 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
474 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
475 #ifdef CONFIG_SYN_COOKIES
477 /* Syncookies use a monotonic timer which increments every 60 seconds.
478 * This counter is used both as a hash input and partially encoded into
479 * the cookie value. A cookie is only validated further if the delta
480 * between the current counter value and the encoded one is less than this,
481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
482 * the counter advances immediately after a cookie is generated).
484 #define MAX_SYNCOOKIE_AGE 2
485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
488 /* syncookies: remember time of last synqueue overflow
489 * But do not dirty this field too often (once per second is enough)
490 * It is racy as we do not hold a lock, but race is very minor.
492 static inline void tcp_synq_overflow(const struct sock *sk)
494 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
495 unsigned long now = jiffies;
497 if (time_after(now, last_overflow + HZ))
498 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
501 /* syncookies: no recent synqueue overflow on this listening socket? */
502 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
504 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
506 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
509 static inline u32 tcp_cookie_time(void)
511 u64 val = get_jiffies_64();
513 do_div(val, TCP_SYNCOOKIE_PERIOD);
517 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
519 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
520 __u32 cookie_init_timestamp(struct request_sock *req);
521 bool cookie_timestamp_decode(struct tcp_options_received *opt);
522 bool cookie_ecn_ok(const struct tcp_options_received *opt,
523 const struct net *net, const struct dst_entry *dst);
525 /* From net/ipv6/syncookies.c */
526 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
528 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
530 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
531 const struct tcphdr *th, u16 *mssp);
532 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
536 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
538 bool tcp_may_send_now(struct sock *sk);
539 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
540 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
541 void tcp_retransmit_timer(struct sock *sk);
542 void tcp_xmit_retransmit_queue(struct sock *);
543 void tcp_simple_retransmit(struct sock *);
544 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
545 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
547 void tcp_send_probe0(struct sock *);
548 void tcp_send_partial(struct sock *);
549 int tcp_write_wakeup(struct sock *, int mib);
550 void tcp_send_fin(struct sock *sk);
551 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
552 int tcp_send_synack(struct sock *);
553 void tcp_push_one(struct sock *, unsigned int mss_now);
554 void tcp_send_ack(struct sock *sk);
555 void tcp_send_delayed_ack(struct sock *sk);
556 void tcp_send_loss_probe(struct sock *sk);
557 bool tcp_schedule_loss_probe(struct sock *sk);
560 void tcp_resume_early_retransmit(struct sock *sk);
561 void tcp_rearm_rto(struct sock *sk);
562 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
563 void tcp_reset(struct sock *sk);
564 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
565 void tcp_fin(struct sock *sk);
568 void tcp_init_xmit_timers(struct sock *);
569 static inline void tcp_clear_xmit_timers(struct sock *sk)
571 inet_csk_clear_xmit_timers(sk);
574 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
575 unsigned int tcp_current_mss(struct sock *sk);
577 /* Bound MSS / TSO packet size with the half of the window */
578 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
582 /* When peer uses tiny windows, there is no use in packetizing
583 * to sub-MSS pieces for the sake of SWS or making sure there
584 * are enough packets in the pipe for fast recovery.
586 * On the other hand, for extremely large MSS devices, handling
587 * smaller than MSS windows in this way does make sense.
589 if (tp->max_window >= 512)
590 cutoff = (tp->max_window >> 1);
592 cutoff = tp->max_window;
594 if (cutoff && pktsize > cutoff)
595 return max_t(int, cutoff, 68U - tp->tcp_header_len);
601 void tcp_get_info(struct sock *, struct tcp_info *);
603 /* Read 'sendfile()'-style from a TCP socket */
604 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
605 unsigned int, size_t);
606 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
607 sk_read_actor_t recv_actor);
609 void tcp_initialize_rcv_mss(struct sock *sk);
611 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
612 int tcp_mss_to_mtu(struct sock *sk, int mss);
613 void tcp_mtup_init(struct sock *sk);
614 void tcp_init_buffer_space(struct sock *sk);
616 static inline void tcp_bound_rto(const struct sock *sk)
618 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
619 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
622 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
624 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
627 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
629 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
630 ntohl(TCP_FLAG_ACK) |
634 static inline void tcp_fast_path_on(struct tcp_sock *tp)
636 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
639 static inline void tcp_fast_path_check(struct sock *sk)
641 struct tcp_sock *tp = tcp_sk(sk);
643 if (skb_queue_empty(&tp->out_of_order_queue) &&
645 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
647 tcp_fast_path_on(tp);
650 /* Compute the actual rto_min value */
651 static inline u32 tcp_rto_min(struct sock *sk)
653 const struct dst_entry *dst = __sk_dst_get(sk);
654 u32 rto_min = TCP_RTO_MIN;
656 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
657 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
661 static inline u32 tcp_rto_min_us(struct sock *sk)
663 return jiffies_to_usecs(tcp_rto_min(sk));
666 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
668 return dst_metric_locked(dst, RTAX_CC_ALGO);
671 /* Minimum RTT in usec. ~0 means not available. */
672 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
674 return tp->rtt_min[0].rtt;
677 /* Compute the actual receive window we are currently advertising.
678 * Rcv_nxt can be after the window if our peer push more data
679 * than the offered window.
681 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
683 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
690 /* Choose a new window, without checks for shrinking, and without
691 * scaling applied to the result. The caller does these things
692 * if necessary. This is a "raw" window selection.
694 u32 __tcp_select_window(struct sock *sk);
696 void tcp_send_window_probe(struct sock *sk);
698 /* TCP timestamps are only 32-bits, this causes a slight
699 * complication on 64-bit systems since we store a snapshot
700 * of jiffies in the buffer control blocks below. We decided
701 * to use only the low 32-bits of jiffies and hide the ugly
702 * casts with the following macro.
704 #define tcp_time_stamp ((__u32)(jiffies))
706 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
708 return skb->skb_mstamp.stamp_jiffies;
712 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
714 #define TCPHDR_FIN 0x01
715 #define TCPHDR_SYN 0x02
716 #define TCPHDR_RST 0x04
717 #define TCPHDR_PSH 0x08
718 #define TCPHDR_ACK 0x10
719 #define TCPHDR_URG 0x20
720 #define TCPHDR_ECE 0x40
721 #define TCPHDR_CWR 0x80
723 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
725 /* This is what the send packet queuing engine uses to pass
726 * TCP per-packet control information to the transmission code.
727 * We also store the host-order sequence numbers in here too.
728 * This is 44 bytes if IPV6 is enabled.
729 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
732 __u32 seq; /* Starting sequence number */
733 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
735 /* Note : tcp_tw_isn is used in input path only
736 * (isn chosen by tcp_timewait_state_process())
738 * tcp_gso_segs/size are used in write queue only,
739 * cf tcp_skb_pcount()/tcp_skb_mss()
747 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
749 __u8 sacked; /* State flags for SACK/FACK. */
750 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
751 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
752 #define TCPCB_LOST 0x04 /* SKB is lost */
753 #define TCPCB_TAGBITS 0x07 /* All tag bits */
754 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
755 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
756 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
759 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
761 __u32 ack_seq; /* Sequence number ACK'd */
763 struct inet_skb_parm h4;
764 #if IS_ENABLED(CONFIG_IPV6)
765 struct inet6_skb_parm h6;
767 } header; /* For incoming frames */
770 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
773 #if IS_ENABLED(CONFIG_IPV6)
774 /* This is the variant of inet6_iif() that must be used by TCP,
775 * as TCP moves IP6CB into a different location in skb->cb[]
777 static inline int tcp_v6_iif(const struct sk_buff *skb)
779 return TCP_SKB_CB(skb)->header.h6.iif;
783 /* Due to TSO, an SKB can be composed of multiple actual
784 * packets. To keep these tracked properly, we use this.
786 static inline int tcp_skb_pcount(const struct sk_buff *skb)
788 return TCP_SKB_CB(skb)->tcp_gso_segs;
791 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
793 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
796 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
798 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
801 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
802 static inline int tcp_skb_mss(const struct sk_buff *skb)
804 return TCP_SKB_CB(skb)->tcp_gso_size;
807 /* Events passed to congestion control interface */
809 CA_EVENT_TX_START, /* first transmit when no packets in flight */
810 CA_EVENT_CWND_RESTART, /* congestion window restart */
811 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
812 CA_EVENT_LOSS, /* loss timeout */
813 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
814 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
815 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
816 CA_EVENT_NON_DELAYED_ACK,
819 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
820 enum tcp_ca_ack_event_flags {
821 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
822 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
823 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
827 * Interface for adding new TCP congestion control handlers
829 #define TCP_CA_NAME_MAX 16
830 #define TCP_CA_MAX 128
831 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
833 #define TCP_CA_UNSPEC 0
835 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
836 #define TCP_CONG_NON_RESTRICTED 0x1
837 /* Requires ECN/ECT set on all packets */
838 #define TCP_CONG_NEEDS_ECN 0x2
842 struct tcp_congestion_ops {
843 struct list_head list;
847 /* initialize private data (optional) */
848 void (*init)(struct sock *sk);
849 /* cleanup private data (optional) */
850 void (*release)(struct sock *sk);
852 /* return slow start threshold (required) */
853 u32 (*ssthresh)(struct sock *sk);
854 /* do new cwnd calculation (required) */
855 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
856 /* call before changing ca_state (optional) */
857 void (*set_state)(struct sock *sk, u8 new_state);
858 /* call when cwnd event occurs (optional) */
859 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
860 /* call when ack arrives (optional) */
861 void (*in_ack_event)(struct sock *sk, u32 flags);
862 /* new value of cwnd after loss (optional) */
863 u32 (*undo_cwnd)(struct sock *sk);
864 /* hook for packet ack accounting (optional) */
865 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
866 /* get info for inet_diag (optional) */
867 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
868 union tcp_cc_info *info);
870 char name[TCP_CA_NAME_MAX];
871 struct module *owner;
874 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
875 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
877 void tcp_assign_congestion_control(struct sock *sk);
878 void tcp_init_congestion_control(struct sock *sk);
879 void tcp_cleanup_congestion_control(struct sock *sk);
880 int tcp_set_default_congestion_control(const char *name);
881 void tcp_get_default_congestion_control(char *name);
882 void tcp_get_available_congestion_control(char *buf, size_t len);
883 void tcp_get_allowed_congestion_control(char *buf, size_t len);
884 int tcp_set_allowed_congestion_control(char *allowed);
885 int tcp_set_congestion_control(struct sock *sk, const char *name);
886 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
887 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
889 u32 tcp_reno_ssthresh(struct sock *sk);
890 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
891 extern struct tcp_congestion_ops tcp_reno;
893 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
894 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
896 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
898 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
904 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
906 const struct inet_connection_sock *icsk = inet_csk(sk);
908 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
911 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
913 struct inet_connection_sock *icsk = inet_csk(sk);
915 if (icsk->icsk_ca_ops->set_state)
916 icsk->icsk_ca_ops->set_state(sk, ca_state);
917 icsk->icsk_ca_state = ca_state;
920 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
922 const struct inet_connection_sock *icsk = inet_csk(sk);
924 if (icsk->icsk_ca_ops->cwnd_event)
925 icsk->icsk_ca_ops->cwnd_event(sk, event);
928 /* These functions determine how the current flow behaves in respect of SACK
929 * handling. SACK is negotiated with the peer, and therefore it can vary
930 * between different flows.
932 * tcp_is_sack - SACK enabled
933 * tcp_is_reno - No SACK
934 * tcp_is_fack - FACK enabled, implies SACK enabled
936 static inline int tcp_is_sack(const struct tcp_sock *tp)
938 return tp->rx_opt.sack_ok;
941 static inline bool tcp_is_reno(const struct tcp_sock *tp)
943 return !tcp_is_sack(tp);
946 static inline bool tcp_is_fack(const struct tcp_sock *tp)
948 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
951 static inline void tcp_enable_fack(struct tcp_sock *tp)
953 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
956 /* TCP early-retransmit (ER) is similar to but more conservative than
957 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
959 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
961 struct net *net = sock_net((struct sock *)tp);
963 tp->do_early_retrans = sysctl_tcp_early_retrans &&
964 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
965 net->ipv4.sysctl_tcp_reordering == 3;
968 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
970 tp->do_early_retrans = 0;
973 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
975 return tp->sacked_out + tp->lost_out;
978 /* This determines how many packets are "in the network" to the best
979 * of our knowledge. In many cases it is conservative, but where
980 * detailed information is available from the receiver (via SACK
981 * blocks etc.) we can make more aggressive calculations.
983 * Use this for decisions involving congestion control, use just
984 * tp->packets_out to determine if the send queue is empty or not.
986 * Read this equation as:
988 * "Packets sent once on transmission queue" MINUS
989 * "Packets left network, but not honestly ACKed yet" PLUS
990 * "Packets fast retransmitted"
992 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
994 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
997 #define TCP_INFINITE_SSTHRESH 0x7fffffff
999 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1001 return tp->snd_cwnd < tp->snd_ssthresh;
1004 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1006 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1009 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1011 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1012 (1 << inet_csk(sk)->icsk_ca_state);
1015 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1016 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1019 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1021 const struct tcp_sock *tp = tcp_sk(sk);
1023 if (tcp_in_cwnd_reduction(sk))
1024 return tp->snd_ssthresh;
1026 return max(tp->snd_ssthresh,
1027 ((tp->snd_cwnd >> 1) +
1028 (tp->snd_cwnd >> 2)));
1031 /* Use define here intentionally to get WARN_ON location shown at the caller */
1032 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1034 void tcp_enter_cwr(struct sock *sk);
1035 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1037 /* The maximum number of MSS of available cwnd for which TSO defers
1038 * sending if not using sysctl_tcp_tso_win_divisor.
1040 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1045 /* Slow start with delack produces 3 packets of burst, so that
1046 * it is safe "de facto". This will be the default - same as
1047 * the default reordering threshold - but if reordering increases,
1048 * we must be able to allow cwnd to burst at least this much in order
1049 * to not pull it back when holes are filled.
1051 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1053 return tp->reordering;
1056 /* Returns end sequence number of the receiver's advertised window */
1057 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1059 return tp->snd_una + tp->snd_wnd;
1062 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1063 * flexible approach. The RFC suggests cwnd should not be raised unless
1064 * it was fully used previously. And that's exactly what we do in
1065 * congestion avoidance mode. But in slow start we allow cwnd to grow
1066 * as long as the application has used half the cwnd.
1068 * cwnd is 10 (IW10), but application sends 9 frames.
1069 * We allow cwnd to reach 18 when all frames are ACKed.
1070 * This check is safe because it's as aggressive as slow start which already
1071 * risks 100% overshoot. The advantage is that we discourage application to
1072 * either send more filler packets or data to artificially blow up the cwnd
1073 * usage, and allow application-limited process to probe bw more aggressively.
1075 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1077 const struct tcp_sock *tp = tcp_sk(sk);
1079 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1080 if (tcp_in_slow_start(tp))
1081 return tp->snd_cwnd < 2 * tp->max_packets_out;
1083 return tp->is_cwnd_limited;
1086 /* Something is really bad, we could not queue an additional packet,
1087 * because qdisc is full or receiver sent a 0 window.
1088 * We do not want to add fuel to the fire, or abort too early,
1089 * so make sure the timer we arm now is at least 200ms in the future,
1090 * regardless of current icsk_rto value (as it could be ~2ms)
1092 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1094 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1097 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1098 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1099 unsigned long max_when)
1101 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1103 return (unsigned long)min_t(u64, when, max_when);
1106 static inline void tcp_check_probe_timer(struct sock *sk)
1108 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1109 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1110 tcp_probe0_base(sk), TCP_RTO_MAX);
1113 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1118 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1124 * Calculate(/check) TCP checksum
1126 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1127 __be32 daddr, __wsum base)
1129 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1132 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1134 return __skb_checksum_complete(skb);
1137 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1139 return !skb_csum_unnecessary(skb) &&
1140 __tcp_checksum_complete(skb);
1143 /* Prequeue for VJ style copy to user, combined with checksumming. */
1145 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1147 tp->ucopy.task = NULL;
1149 tp->ucopy.memory = 0;
1150 skb_queue_head_init(&tp->ucopy.prequeue);
1153 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1158 static const char *statename[]={
1159 "Unused","Established","Syn Sent","Syn Recv",
1160 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1161 "Close Wait","Last ACK","Listen","Closing"
1164 void tcp_set_state(struct sock *sk, int state);
1166 void tcp_done(struct sock *sk);
1168 int tcp_abort(struct sock *sk, int err);
1170 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1173 rx_opt->num_sacks = 0;
1176 u32 tcp_default_init_rwnd(u32 mss);
1177 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1179 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1181 struct tcp_sock *tp = tcp_sk(sk);
1184 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1186 delta = tcp_time_stamp - tp->lsndtime;
1187 if (delta > inet_csk(sk)->icsk_rto)
1188 tcp_cwnd_restart(sk, delta);
1191 /* Determine a window scaling and initial window to offer. */
1192 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1193 __u32 *window_clamp, int wscale_ok,
1194 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1196 static inline int tcp_win_from_space(int space)
1198 return sysctl_tcp_adv_win_scale<=0 ?
1199 (space>>(-sysctl_tcp_adv_win_scale)) :
1200 space - (space>>sysctl_tcp_adv_win_scale);
1203 /* Note: caller must be prepared to deal with negative returns */
1204 static inline int tcp_space(const struct sock *sk)
1206 return tcp_win_from_space(sk->sk_rcvbuf -
1207 atomic_read(&sk->sk_rmem_alloc));
1210 static inline int tcp_full_space(const struct sock *sk)
1212 return tcp_win_from_space(sk->sk_rcvbuf);
1215 extern void tcp_openreq_init_rwin(struct request_sock *req,
1216 const struct sock *sk_listener,
1217 const struct dst_entry *dst);
1219 void tcp_enter_memory_pressure(struct sock *sk);
1221 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1223 struct net *net = sock_net((struct sock *)tp);
1225 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1228 static inline int keepalive_time_when(const struct tcp_sock *tp)
1230 struct net *net = sock_net((struct sock *)tp);
1232 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1235 static inline int keepalive_probes(const struct tcp_sock *tp)
1237 struct net *net = sock_net((struct sock *)tp);
1239 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1242 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1244 const struct inet_connection_sock *icsk = &tp->inet_conn;
1246 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1247 tcp_time_stamp - tp->rcv_tstamp);
1250 static inline int tcp_fin_time(const struct sock *sk)
1252 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1253 const int rto = inet_csk(sk)->icsk_rto;
1255 if (fin_timeout < (rto << 2) - (rto >> 1))
1256 fin_timeout = (rto << 2) - (rto >> 1);
1261 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1264 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1266 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1269 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1270 * then following tcp messages have valid values. Ignore 0 value,
1271 * or else 'negative' tsval might forbid us to accept their packets.
1273 if (!rx_opt->ts_recent)
1278 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1281 if (tcp_paws_check(rx_opt, 0))
1284 /* RST segments are not recommended to carry timestamp,
1285 and, if they do, it is recommended to ignore PAWS because
1286 "their cleanup function should take precedence over timestamps."
1287 Certainly, it is mistake. It is necessary to understand the reasons
1288 of this constraint to relax it: if peer reboots, clock may go
1289 out-of-sync and half-open connections will not be reset.
1290 Actually, the problem would be not existing if all
1291 the implementations followed draft about maintaining clock
1292 via reboots. Linux-2.2 DOES NOT!
1294 However, we can relax time bounds for RST segments to MSL.
1296 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1301 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1302 int mib_idx, u32 *last_oow_ack_time);
1304 static inline void tcp_mib_init(struct net *net)
1307 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1308 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1309 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1310 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1314 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1316 tp->lost_skb_hint = NULL;
1319 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1321 tcp_clear_retrans_hints_partial(tp);
1322 tp->retransmit_skb_hint = NULL;
1328 union tcp_md5_addr {
1330 #if IS_ENABLED(CONFIG_IPV6)
1335 /* - key database */
1336 struct tcp_md5sig_key {
1337 struct hlist_node node;
1339 u8 family; /* AF_INET or AF_INET6 */
1340 union tcp_md5_addr addr;
1341 u8 key[TCP_MD5SIG_MAXKEYLEN];
1342 struct rcu_head rcu;
1346 struct tcp_md5sig_info {
1347 struct hlist_head head;
1348 struct rcu_head rcu;
1351 /* - pseudo header */
1352 struct tcp4_pseudohdr {
1360 struct tcp6_pseudohdr {
1361 struct in6_addr saddr;
1362 struct in6_addr daddr;
1364 __be32 protocol; /* including padding */
1367 union tcp_md5sum_block {
1368 struct tcp4_pseudohdr ip4;
1369 #if IS_ENABLED(CONFIG_IPV6)
1370 struct tcp6_pseudohdr ip6;
1374 /* - pool: digest algorithm, hash description and scratch buffer */
1375 struct tcp_md5sig_pool {
1376 struct hash_desc md5_desc;
1377 union tcp_md5sum_block md5_blk;
1381 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1382 const struct sock *sk, const struct sk_buff *skb);
1383 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1384 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1385 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1387 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1388 const struct sock *addr_sk);
1390 #ifdef CONFIG_TCP_MD5SIG
1391 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1392 const union tcp_md5_addr *addr,
1394 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1396 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1397 const union tcp_md5_addr *addr,
1402 #define tcp_twsk_md5_key(twsk) NULL
1405 bool tcp_alloc_md5sig_pool(void);
1407 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1408 static inline void tcp_put_md5sig_pool(void)
1413 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1414 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1415 unsigned int header_len);
1416 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1417 const struct tcp_md5sig_key *key);
1419 /* From tcp_fastopen.c */
1420 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1421 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1422 unsigned long *last_syn_loss);
1423 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1424 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1426 struct tcp_fastopen_request {
1427 /* Fast Open cookie. Size 0 means a cookie request */
1428 struct tcp_fastopen_cookie cookie;
1429 struct msghdr *data; /* data in MSG_FASTOPEN */
1431 int copied; /* queued in tcp_connect() */
1433 void tcp_free_fastopen_req(struct tcp_sock *tp);
1435 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1436 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1437 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1438 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1439 struct request_sock *req,
1440 struct tcp_fastopen_cookie *foc,
1441 struct dst_entry *dst);
1442 void tcp_fastopen_init_key_once(bool publish);
1443 #define TCP_FASTOPEN_KEY_LENGTH 16
1445 /* Fastopen key context */
1446 struct tcp_fastopen_context {
1447 struct crypto_cipher *tfm;
1448 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1449 struct rcu_head rcu;
1452 /* write queue abstraction */
1453 static inline void tcp_write_queue_purge(struct sock *sk)
1455 struct sk_buff *skb;
1457 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1458 sk_wmem_free_skb(sk, skb);
1460 tcp_clear_all_retrans_hints(tcp_sk(sk));
1463 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1465 return skb_peek(&sk->sk_write_queue);
1468 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1470 return skb_peek_tail(&sk->sk_write_queue);
1473 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1474 const struct sk_buff *skb)
1476 return skb_queue_next(&sk->sk_write_queue, skb);
1479 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1480 const struct sk_buff *skb)
1482 return skb_queue_prev(&sk->sk_write_queue, skb);
1485 #define tcp_for_write_queue(skb, sk) \
1486 skb_queue_walk(&(sk)->sk_write_queue, skb)
1488 #define tcp_for_write_queue_from(skb, sk) \
1489 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1491 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1492 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1494 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1496 return sk->sk_send_head;
1499 static inline bool tcp_skb_is_last(const struct sock *sk,
1500 const struct sk_buff *skb)
1502 return skb_queue_is_last(&sk->sk_write_queue, skb);
1505 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1507 if (tcp_skb_is_last(sk, skb))
1508 sk->sk_send_head = NULL;
1510 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1513 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1515 if (sk->sk_send_head == skb_unlinked)
1516 sk->sk_send_head = NULL;
1519 static inline void tcp_init_send_head(struct sock *sk)
1521 sk->sk_send_head = NULL;
1524 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1526 __skb_queue_tail(&sk->sk_write_queue, skb);
1529 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1531 __tcp_add_write_queue_tail(sk, skb);
1533 /* Queue it, remembering where we must start sending. */
1534 if (sk->sk_send_head == NULL) {
1535 sk->sk_send_head = skb;
1537 if (tcp_sk(sk)->highest_sack == NULL)
1538 tcp_sk(sk)->highest_sack = skb;
1542 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1544 __skb_queue_head(&sk->sk_write_queue, skb);
1547 /* Insert buff after skb on the write queue of sk. */
1548 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1549 struct sk_buff *buff,
1552 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1555 /* Insert new before skb on the write queue of sk. */
1556 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1557 struct sk_buff *skb,
1560 __skb_queue_before(&sk->sk_write_queue, skb, new);
1562 if (sk->sk_send_head == skb)
1563 sk->sk_send_head = new;
1566 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1568 __skb_unlink(skb, &sk->sk_write_queue);
1571 static inline bool tcp_write_queue_empty(struct sock *sk)
1573 return skb_queue_empty(&sk->sk_write_queue);
1576 static inline void tcp_push_pending_frames(struct sock *sk)
1578 if (tcp_send_head(sk)) {
1579 struct tcp_sock *tp = tcp_sk(sk);
1581 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1585 /* Start sequence of the skb just after the highest skb with SACKed
1586 * bit, valid only if sacked_out > 0 or when the caller has ensured
1587 * validity by itself.
1589 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1591 if (!tp->sacked_out)
1594 if (tp->highest_sack == NULL)
1597 return TCP_SKB_CB(tp->highest_sack)->seq;
1600 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1602 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1603 tcp_write_queue_next(sk, skb);
1606 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1608 return tcp_sk(sk)->highest_sack;
1611 static inline void tcp_highest_sack_reset(struct sock *sk)
1613 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1616 /* Called when old skb is about to be deleted (to be combined with new skb) */
1617 static inline void tcp_highest_sack_combine(struct sock *sk,
1618 struct sk_buff *old,
1619 struct sk_buff *new)
1621 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1622 tcp_sk(sk)->highest_sack = new;
1625 /* This helper checks if socket has IP_TRANSPARENT set */
1626 static inline bool inet_sk_transparent(const struct sock *sk)
1628 switch (sk->sk_state) {
1630 return inet_twsk(sk)->tw_transparent;
1631 case TCP_NEW_SYN_RECV:
1632 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1634 return inet_sk(sk)->transparent;
1637 /* Determines whether this is a thin stream (which may suffer from
1638 * increased latency). Used to trigger latency-reducing mechanisms.
1640 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1642 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1646 enum tcp_seq_states {
1647 TCP_SEQ_STATE_LISTENING,
1648 TCP_SEQ_STATE_ESTABLISHED,
1651 int tcp_seq_open(struct inode *inode, struct file *file);
1653 struct tcp_seq_afinfo {
1656 const struct file_operations *seq_fops;
1657 struct seq_operations seq_ops;
1660 struct tcp_iter_state {
1661 struct seq_net_private p;
1663 enum tcp_seq_states state;
1664 struct sock *syn_wait_sk;
1665 int bucket, offset, sbucket, num;
1669 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1670 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1672 extern struct request_sock_ops tcp_request_sock_ops;
1673 extern struct request_sock_ops tcp6_request_sock_ops;
1675 void tcp_v4_destroy_sock(struct sock *sk);
1677 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1678 netdev_features_t features);
1679 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1680 int tcp_gro_complete(struct sk_buff *skb);
1682 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1684 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1686 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1689 static inline bool tcp_stream_memory_free(const struct sock *sk)
1691 const struct tcp_sock *tp = tcp_sk(sk);
1692 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1694 return notsent_bytes < tcp_notsent_lowat(tp);
1697 #ifdef CONFIG_PROC_FS
1698 int tcp4_proc_init(void);
1699 void tcp4_proc_exit(void);
1702 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1703 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1704 const struct tcp_request_sock_ops *af_ops,
1705 struct sock *sk, struct sk_buff *skb);
1707 /* TCP af-specific functions */
1708 struct tcp_sock_af_ops {
1709 #ifdef CONFIG_TCP_MD5SIG
1710 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1711 const struct sock *addr_sk);
1712 int (*calc_md5_hash)(char *location,
1713 const struct tcp_md5sig_key *md5,
1714 const struct sock *sk,
1715 const struct sk_buff *skb);
1716 int (*md5_parse)(struct sock *sk,
1717 char __user *optval,
1722 struct tcp_request_sock_ops {
1724 #ifdef CONFIG_TCP_MD5SIG
1725 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1726 const struct sock *addr_sk);
1727 int (*calc_md5_hash) (char *location,
1728 const struct tcp_md5sig_key *md5,
1729 const struct sock *sk,
1730 const struct sk_buff *skb);
1732 void (*init_req)(struct request_sock *req,
1733 const struct sock *sk_listener,
1734 struct sk_buff *skb);
1735 #ifdef CONFIG_SYN_COOKIES
1736 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1739 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1740 const struct request_sock *req,
1742 __u32 (*init_seq)(const struct sk_buff *skb);
1743 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1744 struct flowi *fl, struct request_sock *req,
1745 struct tcp_fastopen_cookie *foc,
1749 #ifdef CONFIG_SYN_COOKIES
1750 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1751 const struct sock *sk, struct sk_buff *skb,
1754 tcp_synq_overflow(sk);
1755 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1756 return ops->cookie_init_seq(skb, mss);
1759 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1760 const struct sock *sk, struct sk_buff *skb,
1767 int tcpv4_offload_init(void);
1769 void tcp_v4_init(void);
1770 void tcp_init(void);
1772 /* tcp_recovery.c */
1774 /* Flags to enable various loss recovery features. See below */
1775 extern int sysctl_tcp_recovery;
1777 /* Use TCP RACK to detect (some) tail and retransmit losses */
1778 #define TCP_RACK_LOST_RETRANS 0x1
1780 extern int tcp_rack_mark_lost(struct sock *sk);
1782 extern void tcp_rack_advance(struct tcp_sock *tp,
1783 const struct skb_mstamp *xmit_time, u8 sacked);
1786 * Save and compile IPv4 options, return a pointer to it
1788 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1790 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1791 struct ip_options_rcu *dopt = NULL;
1794 int opt_size = sizeof(*dopt) + opt->optlen;
1796 dopt = kmalloc(opt_size, GFP_ATOMIC);
1797 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1805 /* locally generated TCP pure ACKs have skb->truesize == 2
1806 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1807 * This is much faster than dissecting the packet to find out.
1808 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1810 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1812 return skb->truesize == 2;
1815 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)