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_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_EXP_FASTOPEN_BASE 4
199 /* But this is what stacks really send out. */
200 #define TCPOLEN_TSTAMP_ALIGNED 12
201 #define TCPOLEN_WSCALE_ALIGNED 4
202 #define TCPOLEN_SACKPERM_ALIGNED 4
203 #define TCPOLEN_SACK_BASE 2
204 #define TCPOLEN_SACK_BASE_ALIGNED 4
205 #define TCPOLEN_SACK_PERBLOCK 8
206 #define TCPOLEN_MD5SIG_ALIGNED 20
207 #define TCPOLEN_MSS_ALIGNED 4
209 /* Flags in tp->nonagle */
210 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
211 #define TCP_NAGLE_CORK 2 /* Socket is corked */
212 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
214 /* TCP thin-stream limits */
215 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
217 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
218 #define TCP_INIT_CWND 10
220 /* Bit Flags for sysctl_tcp_fastopen */
221 #define TFO_CLIENT_ENABLE 1
222 #define TFO_SERVER_ENABLE 2
223 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
225 /* Accept SYN data w/o any cookie option */
226 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
228 /* Force enable TFO on all listeners, i.e., not requiring the
229 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
231 #define TFO_SERVER_WO_SOCKOPT1 0x400
232 #define TFO_SERVER_WO_SOCKOPT2 0x800
234 extern struct inet_timewait_death_row tcp_death_row;
236 /* sysctl variables for tcp */
237 extern int sysctl_tcp_timestamps;
238 extern int sysctl_tcp_window_scaling;
239 extern int sysctl_tcp_sack;
240 extern int sysctl_tcp_fin_timeout;
241 extern int sysctl_tcp_keepalive_time;
242 extern int sysctl_tcp_keepalive_probes;
243 extern int sysctl_tcp_keepalive_intvl;
244 extern int sysctl_tcp_syn_retries;
245 extern int sysctl_tcp_synack_retries;
246 extern int sysctl_tcp_retries1;
247 extern int sysctl_tcp_retries2;
248 extern int sysctl_tcp_orphan_retries;
249 extern int sysctl_tcp_syncookies;
250 extern int sysctl_tcp_fastopen;
251 extern int sysctl_tcp_retrans_collapse;
252 extern int sysctl_tcp_stdurg;
253 extern int sysctl_tcp_rfc1337;
254 extern int sysctl_tcp_abort_on_overflow;
255 extern int sysctl_tcp_max_orphans;
256 extern int sysctl_tcp_fack;
257 extern int sysctl_tcp_reordering;
258 extern int sysctl_tcp_max_reordering;
259 extern int sysctl_tcp_dsack;
260 extern long sysctl_tcp_mem[3];
261 extern int sysctl_tcp_wmem[3];
262 extern int sysctl_tcp_rmem[3];
263 extern int sysctl_tcp_app_win;
264 extern int sysctl_tcp_adv_win_scale;
265 extern int sysctl_tcp_tw_reuse;
266 extern int sysctl_tcp_frto;
267 extern int sysctl_tcp_low_latency;
268 extern int sysctl_tcp_nometrics_save;
269 extern int sysctl_tcp_moderate_rcvbuf;
270 extern int sysctl_tcp_tso_win_divisor;
271 extern int sysctl_tcp_workaround_signed_windows;
272 extern int sysctl_tcp_slow_start_after_idle;
273 extern int sysctl_tcp_thin_linear_timeouts;
274 extern int sysctl_tcp_thin_dupack;
275 extern int sysctl_tcp_early_retrans;
276 extern int sysctl_tcp_limit_output_bytes;
277 extern int sysctl_tcp_challenge_ack_limit;
278 extern unsigned int sysctl_tcp_notsent_lowat;
279 extern int sysctl_tcp_min_tso_segs;
280 extern int sysctl_tcp_autocorking;
281 extern int sysctl_tcp_invalid_ratelimit;
283 extern atomic_long_t tcp_memory_allocated;
284 extern struct percpu_counter tcp_sockets_allocated;
285 extern int tcp_memory_pressure;
288 * The next routines deal with comparing 32 bit unsigned ints
289 * and worry about wraparound (automatic with unsigned arithmetic).
292 static inline bool before(__u32 seq1, __u32 seq2)
294 return (__s32)(seq1-seq2) < 0;
296 #define after(seq2, seq1) before(seq1, seq2)
298 /* is s2<=s1<=s3 ? */
299 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
301 return seq3 - seq2 >= seq1 - seq2;
304 static inline bool tcp_out_of_memory(struct sock *sk)
306 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
307 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
312 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
314 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
315 int orphans = percpu_counter_read_positive(ocp);
317 if (orphans << shift > sysctl_tcp_max_orphans) {
318 orphans = percpu_counter_sum_positive(ocp);
319 if (orphans << shift > sysctl_tcp_max_orphans)
325 bool tcp_check_oom(struct sock *sk, int shift);
327 /* syncookies: remember time of last synqueue overflow */
328 static inline void tcp_synq_overflow(struct sock *sk)
330 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
333 /* syncookies: no recent synqueue overflow on this listening socket? */
334 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
336 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
337 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
340 extern struct proto tcp_prot;
342 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
343 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
344 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
345 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
346 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
348 void tcp_tasklet_init(void);
350 void tcp_v4_err(struct sk_buff *skb, u32);
352 void tcp_shutdown(struct sock *sk, int how);
354 void tcp_v4_early_demux(struct sk_buff *skb);
355 int tcp_v4_rcv(struct sk_buff *skb);
357 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
358 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
359 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
361 void tcp_release_cb(struct sock *sk);
362 void tcp_wfree(struct sk_buff *skb);
363 void tcp_write_timer_handler(struct sock *sk);
364 void tcp_delack_timer_handler(struct sock *sk);
365 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
366 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
367 const struct tcphdr *th, unsigned int len);
368 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
369 const struct tcphdr *th, unsigned int len);
370 void tcp_rcv_space_adjust(struct sock *sk);
371 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
372 void tcp_twsk_destructor(struct sock *sk);
373 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
374 struct pipe_inode_info *pipe, size_t len,
377 static inline void tcp_dec_quickack_mode(struct sock *sk,
378 const unsigned int pkts)
380 struct inet_connection_sock *icsk = inet_csk(sk);
382 if (icsk->icsk_ack.quick) {
383 if (pkts >= icsk->icsk_ack.quick) {
384 icsk->icsk_ack.quick = 0;
385 /* Leaving quickack mode we deflate ATO. */
386 icsk->icsk_ack.ato = TCP_ATO_MIN;
388 icsk->icsk_ack.quick -= pkts;
393 #define TCP_ECN_QUEUE_CWR 2
394 #define TCP_ECN_DEMAND_CWR 4
395 #define TCP_ECN_SEEN 8
405 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
407 const struct tcphdr *th);
408 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
409 struct request_sock *req, bool fastopen);
410 int tcp_child_process(struct sock *parent, struct sock *child,
411 struct sk_buff *skb);
412 void tcp_enter_loss(struct sock *sk);
413 void tcp_clear_retrans(struct tcp_sock *tp);
414 void tcp_update_metrics(struct sock *sk);
415 void tcp_init_metrics(struct sock *sk);
416 void tcp_metrics_init(void);
417 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
418 bool paws_check, bool timestamps);
419 bool tcp_remember_stamp(struct sock *sk);
420 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
421 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
422 void tcp_disable_fack(struct tcp_sock *tp);
423 void tcp_close(struct sock *sk, long timeout);
424 void tcp_init_sock(struct sock *sk);
425 unsigned int tcp_poll(struct file *file, struct socket *sock,
426 struct poll_table_struct *wait);
427 int tcp_getsockopt(struct sock *sk, int level, int optname,
428 char __user *optval, int __user *optlen);
429 int tcp_setsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, unsigned int optlen);
431 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
432 char __user *optval, int __user *optlen);
433 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
434 char __user *optval, unsigned int optlen);
435 void tcp_set_keepalive(struct sock *sk, int val);
436 void tcp_syn_ack_timeout(const struct request_sock *req);
437 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
438 int flags, int *addr_len);
439 void tcp_parse_options(const struct sk_buff *skb,
440 struct tcp_options_received *opt_rx,
441 int estab, struct tcp_fastopen_cookie *foc);
442 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
445 * TCP v4 functions exported for the inet6 API
448 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
449 void tcp_v4_mtu_reduced(struct sock *sk);
450 void tcp_req_err(struct sock *sk, u32 seq);
451 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
452 struct sock *tcp_create_openreq_child(struct sock *sk,
453 struct request_sock *req,
454 struct sk_buff *skb);
455 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
456 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
457 struct request_sock *req,
458 struct dst_entry *dst);
459 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
460 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
461 int tcp_connect(struct sock *sk);
462 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
463 struct request_sock *req,
464 struct tcp_fastopen_cookie *foc);
465 int tcp_disconnect(struct sock *sk, int flags);
467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
471 /* From syncookies.c */
472 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
486 static inline u32 tcp_cookie_time(void)
488 u64 val = get_jiffies_64();
490 do_div(val, 60 * HZ);
494 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
496 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
498 __u32 cookie_init_timestamp(struct request_sock *req);
499 bool cookie_timestamp_decode(struct tcp_options_received *opt);
500 bool cookie_ecn_ok(const struct tcp_options_received *opt,
501 const struct net *net, const struct dst_entry *dst);
503 /* From net/ipv6/syncookies.c */
504 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
506 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
508 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
509 const struct tcphdr *th, u16 *mssp);
510 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
515 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
517 bool tcp_may_send_now(struct sock *sk);
518 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
519 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
520 void tcp_retransmit_timer(struct sock *sk);
521 void tcp_xmit_retransmit_queue(struct sock *);
522 void tcp_simple_retransmit(struct sock *);
523 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
524 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
526 void tcp_send_probe0(struct sock *);
527 void tcp_send_partial(struct sock *);
528 int tcp_write_wakeup(struct sock *);
529 void tcp_send_fin(struct sock *sk);
530 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
531 int tcp_send_synack(struct sock *);
532 void tcp_push_one(struct sock *, unsigned int mss_now);
533 void tcp_send_ack(struct sock *sk);
534 void tcp_send_delayed_ack(struct sock *sk);
535 void tcp_send_loss_probe(struct sock *sk);
536 bool tcp_schedule_loss_probe(struct sock *sk);
539 void tcp_resume_early_retransmit(struct sock *sk);
540 void tcp_rearm_rto(struct sock *sk);
541 void tcp_reset(struct sock *sk);
544 void tcp_init_xmit_timers(struct sock *);
545 static inline void tcp_clear_xmit_timers(struct sock *sk)
547 inet_csk_clear_xmit_timers(sk);
550 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
551 unsigned int tcp_current_mss(struct sock *sk);
553 /* Bound MSS / TSO packet size with the half of the window */
554 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
558 /* When peer uses tiny windows, there is no use in packetizing
559 * to sub-MSS pieces for the sake of SWS or making sure there
560 * are enough packets in the pipe for fast recovery.
562 * On the other hand, for extremely large MSS devices, handling
563 * smaller than MSS windows in this way does make sense.
565 if (tp->max_window >= 512)
566 cutoff = (tp->max_window >> 1);
568 cutoff = tp->max_window;
570 if (cutoff && pktsize > cutoff)
571 return max_t(int, cutoff, 68U - tp->tcp_header_len);
577 void tcp_get_info(const struct sock *, struct tcp_info *);
579 /* Read 'sendfile()'-style from a TCP socket */
580 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
581 unsigned int, size_t);
582 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
583 sk_read_actor_t recv_actor);
585 void tcp_initialize_rcv_mss(struct sock *sk);
587 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
588 int tcp_mss_to_mtu(struct sock *sk, int mss);
589 void tcp_mtup_init(struct sock *sk);
590 void tcp_init_buffer_space(struct sock *sk);
592 static inline void tcp_bound_rto(const struct sock *sk)
594 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
595 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
598 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
600 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
603 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
605 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
606 ntohl(TCP_FLAG_ACK) |
610 static inline void tcp_fast_path_on(struct tcp_sock *tp)
612 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
615 static inline void tcp_fast_path_check(struct sock *sk)
617 struct tcp_sock *tp = tcp_sk(sk);
619 if (skb_queue_empty(&tp->out_of_order_queue) &&
621 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
623 tcp_fast_path_on(tp);
626 /* Compute the actual rto_min value */
627 static inline u32 tcp_rto_min(struct sock *sk)
629 const struct dst_entry *dst = __sk_dst_get(sk);
630 u32 rto_min = TCP_RTO_MIN;
632 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
633 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
637 static inline u32 tcp_rto_min_us(struct sock *sk)
639 return jiffies_to_usecs(tcp_rto_min(sk));
642 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
644 return dst_metric_locked(dst, RTAX_CC_ALGO);
647 /* Compute the actual receive window we are currently advertising.
648 * Rcv_nxt can be after the window if our peer push more data
649 * than the offered window.
651 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
653 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
660 /* Choose a new window, without checks for shrinking, and without
661 * scaling applied to the result. The caller does these things
662 * if necessary. This is a "raw" window selection.
664 u32 __tcp_select_window(struct sock *sk);
666 void tcp_send_window_probe(struct sock *sk);
668 /* TCP timestamps are only 32-bits, this causes a slight
669 * complication on 64-bit systems since we store a snapshot
670 * of jiffies in the buffer control blocks below. We decided
671 * to use only the low 32-bits of jiffies and hide the ugly
672 * casts with the following macro.
674 #define tcp_time_stamp ((__u32)(jiffies))
676 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
678 return skb->skb_mstamp.stamp_jiffies;
682 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
684 #define TCPHDR_FIN 0x01
685 #define TCPHDR_SYN 0x02
686 #define TCPHDR_RST 0x04
687 #define TCPHDR_PSH 0x08
688 #define TCPHDR_ACK 0x10
689 #define TCPHDR_URG 0x20
690 #define TCPHDR_ECE 0x40
691 #define TCPHDR_CWR 0x80
693 /* This is what the send packet queuing engine uses to pass
694 * TCP per-packet control information to the transmission code.
695 * We also store the host-order sequence numbers in here too.
696 * This is 44 bytes if IPV6 is enabled.
697 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
700 __u32 seq; /* Starting sequence number */
701 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
703 /* Note : tcp_tw_isn is used in input path only
704 * (isn chosen by tcp_timewait_state_process())
706 * tcp_gso_segs is used in write queue only,
707 * cf tcp_skb_pcount()
712 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
714 __u8 sacked; /* State flags for SACK/FACK. */
715 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
716 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
717 #define TCPCB_LOST 0x04 /* SKB is lost */
718 #define TCPCB_TAGBITS 0x07 /* All tag bits */
719 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
720 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
721 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
724 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
726 __u32 ack_seq; /* Sequence number ACK'd */
728 struct inet_skb_parm h4;
729 #if IS_ENABLED(CONFIG_IPV6)
730 struct inet6_skb_parm h6;
732 } header; /* For incoming frames */
735 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
738 #if IS_ENABLED(CONFIG_IPV6)
739 /* This is the variant of inet6_iif() that must be used by TCP,
740 * as TCP moves IP6CB into a different location in skb->cb[]
742 static inline int tcp_v6_iif(const struct sk_buff *skb)
744 return TCP_SKB_CB(skb)->header.h6.iif;
748 /* Due to TSO, an SKB can be composed of multiple actual
749 * packets. To keep these tracked properly, we use this.
751 static inline int tcp_skb_pcount(const struct sk_buff *skb)
753 return TCP_SKB_CB(skb)->tcp_gso_segs;
756 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
758 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
761 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
763 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
766 /* This is valid iff tcp_skb_pcount() > 1. */
767 static inline int tcp_skb_mss(const struct sk_buff *skb)
769 return skb_shinfo(skb)->gso_size;
772 /* Events passed to congestion control interface */
774 CA_EVENT_TX_START, /* first transmit when no packets in flight */
775 CA_EVENT_CWND_RESTART, /* congestion window restart */
776 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
777 CA_EVENT_LOSS, /* loss timeout */
778 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
779 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
780 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
781 CA_EVENT_NON_DELAYED_ACK,
784 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
785 enum tcp_ca_ack_event_flags {
786 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
787 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
788 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
792 * Interface for adding new TCP congestion control handlers
794 #define TCP_CA_NAME_MAX 16
795 #define TCP_CA_MAX 128
796 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
798 #define TCP_CA_UNSPEC 0
800 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
801 #define TCP_CONG_NON_RESTRICTED 0x1
802 /* Requires ECN/ECT set on all packets */
803 #define TCP_CONG_NEEDS_ECN 0x2
805 struct tcp_congestion_ops {
806 struct list_head list;
810 /* initialize private data (optional) */
811 void (*init)(struct sock *sk);
812 /* cleanup private data (optional) */
813 void (*release)(struct sock *sk);
815 /* return slow start threshold (required) */
816 u32 (*ssthresh)(struct sock *sk);
817 /* do new cwnd calculation (required) */
818 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
819 /* call before changing ca_state (optional) */
820 void (*set_state)(struct sock *sk, u8 new_state);
821 /* call when cwnd event occurs (optional) */
822 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
823 /* call when ack arrives (optional) */
824 void (*in_ack_event)(struct sock *sk, u32 flags);
825 /* new value of cwnd after loss (optional) */
826 u32 (*undo_cwnd)(struct sock *sk);
827 /* hook for packet ack accounting (optional) */
828 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
829 /* get info for inet_diag (optional) */
830 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
832 char name[TCP_CA_NAME_MAX];
833 struct module *owner;
836 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
837 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
839 void tcp_assign_congestion_control(struct sock *sk);
840 void tcp_init_congestion_control(struct sock *sk);
841 void tcp_cleanup_congestion_control(struct sock *sk);
842 int tcp_set_default_congestion_control(const char *name);
843 void tcp_get_default_congestion_control(char *name);
844 void tcp_get_available_congestion_control(char *buf, size_t len);
845 void tcp_get_allowed_congestion_control(char *buf, size_t len);
846 int tcp_set_allowed_congestion_control(char *allowed);
847 int tcp_set_congestion_control(struct sock *sk, const char *name);
848 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
849 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
851 u32 tcp_reno_ssthresh(struct sock *sk);
852 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
853 extern struct tcp_congestion_ops tcp_reno;
855 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
856 u32 tcp_ca_get_key_by_name(const char *name);
858 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
860 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
866 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
868 const struct inet_connection_sock *icsk = inet_csk(sk);
870 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
873 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
875 struct inet_connection_sock *icsk = inet_csk(sk);
877 if (icsk->icsk_ca_ops->set_state)
878 icsk->icsk_ca_ops->set_state(sk, ca_state);
879 icsk->icsk_ca_state = ca_state;
882 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
884 const struct inet_connection_sock *icsk = inet_csk(sk);
886 if (icsk->icsk_ca_ops->cwnd_event)
887 icsk->icsk_ca_ops->cwnd_event(sk, event);
890 /* These functions determine how the current flow behaves in respect of SACK
891 * handling. SACK is negotiated with the peer, and therefore it can vary
892 * between different flows.
894 * tcp_is_sack - SACK enabled
895 * tcp_is_reno - No SACK
896 * tcp_is_fack - FACK enabled, implies SACK enabled
898 static inline int tcp_is_sack(const struct tcp_sock *tp)
900 return tp->rx_opt.sack_ok;
903 static inline bool tcp_is_reno(const struct tcp_sock *tp)
905 return !tcp_is_sack(tp);
908 static inline bool tcp_is_fack(const struct tcp_sock *tp)
910 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
913 static inline void tcp_enable_fack(struct tcp_sock *tp)
915 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
918 /* TCP early-retransmit (ER) is similar to but more conservative than
919 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
921 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
923 tp->do_early_retrans = sysctl_tcp_early_retrans &&
924 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
925 sysctl_tcp_reordering == 3;
928 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
930 tp->do_early_retrans = 0;
933 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
935 return tp->sacked_out + tp->lost_out;
938 /* This determines how many packets are "in the network" to the best
939 * of our knowledge. In many cases it is conservative, but where
940 * detailed information is available from the receiver (via SACK
941 * blocks etc.) we can make more aggressive calculations.
943 * Use this for decisions involving congestion control, use just
944 * tp->packets_out to determine if the send queue is empty or not.
946 * Read this equation as:
948 * "Packets sent once on transmission queue" MINUS
949 * "Packets left network, but not honestly ACKed yet" PLUS
950 * "Packets fast retransmitted"
952 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
954 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
957 #define TCP_INFINITE_SSTHRESH 0x7fffffff
959 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
961 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
964 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
966 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
967 (1 << inet_csk(sk)->icsk_ca_state);
970 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
971 * The exception is cwnd reduction phase, when cwnd is decreasing towards
974 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
976 const struct tcp_sock *tp = tcp_sk(sk);
978 if (tcp_in_cwnd_reduction(sk))
979 return tp->snd_ssthresh;
981 return max(tp->snd_ssthresh,
982 ((tp->snd_cwnd >> 1) +
983 (tp->snd_cwnd >> 2)));
986 /* Use define here intentionally to get WARN_ON location shown at the caller */
987 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
989 void tcp_enter_cwr(struct sock *sk);
990 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
992 /* The maximum number of MSS of available cwnd for which TSO defers
993 * sending if not using sysctl_tcp_tso_win_divisor.
995 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1000 /* Slow start with delack produces 3 packets of burst, so that
1001 * it is safe "de facto". This will be the default - same as
1002 * the default reordering threshold - but if reordering increases,
1003 * we must be able to allow cwnd to burst at least this much in order
1004 * to not pull it back when holes are filled.
1006 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1008 return tp->reordering;
1011 /* Returns end sequence number of the receiver's advertised window */
1012 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1014 return tp->snd_una + tp->snd_wnd;
1017 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1018 * flexible approach. The RFC suggests cwnd should not be raised unless
1019 * it was fully used previously. And that's exactly what we do in
1020 * congestion avoidance mode. But in slow start we allow cwnd to grow
1021 * as long as the application has used half the cwnd.
1023 * cwnd is 10 (IW10), but application sends 9 frames.
1024 * We allow cwnd to reach 18 when all frames are ACKed.
1025 * This check is safe because it's as aggressive as slow start which already
1026 * risks 100% overshoot. The advantage is that we discourage application to
1027 * either send more filler packets or data to artificially blow up the cwnd
1028 * usage, and allow application-limited process to probe bw more aggressively.
1030 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1032 const struct tcp_sock *tp = tcp_sk(sk);
1034 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1035 if (tp->snd_cwnd <= tp->snd_ssthresh)
1036 return tp->snd_cwnd < 2 * tp->max_packets_out;
1038 return tp->is_cwnd_limited;
1041 static inline void tcp_check_probe_timer(struct sock *sk)
1043 const struct tcp_sock *tp = tcp_sk(sk);
1044 const struct inet_connection_sock *icsk = inet_csk(sk);
1046 if (!tp->packets_out && !icsk->icsk_pending)
1047 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1048 icsk->icsk_rto, TCP_RTO_MAX);
1051 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1056 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1062 * Calculate(/check) TCP checksum
1064 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1065 __be32 daddr, __wsum base)
1067 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1070 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1072 return __skb_checksum_complete(skb);
1075 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1077 return !skb_csum_unnecessary(skb) &&
1078 __tcp_checksum_complete(skb);
1081 /* Prequeue for VJ style copy to user, combined with checksumming. */
1083 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1085 tp->ucopy.task = NULL;
1087 tp->ucopy.memory = 0;
1088 skb_queue_head_init(&tp->ucopy.prequeue);
1091 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1096 static const char *statename[]={
1097 "Unused","Established","Syn Sent","Syn Recv",
1098 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1099 "Close Wait","Last ACK","Listen","Closing"
1102 void tcp_set_state(struct sock *sk, int state);
1104 void tcp_done(struct sock *sk);
1106 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1109 rx_opt->num_sacks = 0;
1112 u32 tcp_default_init_rwnd(u32 mss);
1114 /* Determine a window scaling and initial window to offer. */
1115 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1116 __u32 *window_clamp, int wscale_ok,
1117 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1119 static inline int tcp_win_from_space(int space)
1121 return sysctl_tcp_adv_win_scale<=0 ?
1122 (space>>(-sysctl_tcp_adv_win_scale)) :
1123 space - (space>>sysctl_tcp_adv_win_scale);
1126 /* Note: caller must be prepared to deal with negative returns */
1127 static inline int tcp_space(const struct sock *sk)
1129 return tcp_win_from_space(sk->sk_rcvbuf -
1130 atomic_read(&sk->sk_rmem_alloc));
1133 static inline int tcp_full_space(const struct sock *sk)
1135 return tcp_win_from_space(sk->sk_rcvbuf);
1138 extern void tcp_openreq_init_rwin(struct request_sock *req,
1139 struct sock *sk, struct dst_entry *dst);
1141 void tcp_enter_memory_pressure(struct sock *sk);
1143 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1145 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1148 static inline int keepalive_time_when(const struct tcp_sock *tp)
1150 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1153 static inline int keepalive_probes(const struct tcp_sock *tp)
1155 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1158 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1160 const struct inet_connection_sock *icsk = &tp->inet_conn;
1162 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1163 tcp_time_stamp - tp->rcv_tstamp);
1166 static inline int tcp_fin_time(const struct sock *sk)
1168 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1169 const int rto = inet_csk(sk)->icsk_rto;
1171 if (fin_timeout < (rto << 2) - (rto >> 1))
1172 fin_timeout = (rto << 2) - (rto >> 1);
1177 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1180 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1182 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1185 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1186 * then following tcp messages have valid values. Ignore 0 value,
1187 * or else 'negative' tsval might forbid us to accept their packets.
1189 if (!rx_opt->ts_recent)
1194 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1197 if (tcp_paws_check(rx_opt, 0))
1200 /* RST segments are not recommended to carry timestamp,
1201 and, if they do, it is recommended to ignore PAWS because
1202 "their cleanup function should take precedence over timestamps."
1203 Certainly, it is mistake. It is necessary to understand the reasons
1204 of this constraint to relax it: if peer reboots, clock may go
1205 out-of-sync and half-open connections will not be reset.
1206 Actually, the problem would be not existing if all
1207 the implementations followed draft about maintaining clock
1208 via reboots. Linux-2.2 DOES NOT!
1210 However, we can relax time bounds for RST segments to MSL.
1212 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1217 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1218 int mib_idx, u32 *last_oow_ack_time);
1220 static inline void tcp_mib_init(struct net *net)
1223 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1224 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1225 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1226 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1230 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1232 tp->lost_skb_hint = NULL;
1235 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1237 tcp_clear_retrans_hints_partial(tp);
1238 tp->retransmit_skb_hint = NULL;
1244 union tcp_md5_addr {
1246 #if IS_ENABLED(CONFIG_IPV6)
1251 /* - key database */
1252 struct tcp_md5sig_key {
1253 struct hlist_node node;
1255 u8 family; /* AF_INET or AF_INET6 */
1256 union tcp_md5_addr addr;
1257 u8 key[TCP_MD5SIG_MAXKEYLEN];
1258 struct rcu_head rcu;
1262 struct tcp_md5sig_info {
1263 struct hlist_head head;
1264 struct rcu_head rcu;
1267 /* - pseudo header */
1268 struct tcp4_pseudohdr {
1276 struct tcp6_pseudohdr {
1277 struct in6_addr saddr;
1278 struct in6_addr daddr;
1280 __be32 protocol; /* including padding */
1283 union tcp_md5sum_block {
1284 struct tcp4_pseudohdr ip4;
1285 #if IS_ENABLED(CONFIG_IPV6)
1286 struct tcp6_pseudohdr ip6;
1290 /* - pool: digest algorithm, hash description and scratch buffer */
1291 struct tcp_md5sig_pool {
1292 struct hash_desc md5_desc;
1293 union tcp_md5sum_block md5_blk;
1297 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1298 const struct sock *sk, const struct sk_buff *skb);
1299 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1300 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1301 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1303 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1304 const struct sock *addr_sk);
1306 #ifdef CONFIG_TCP_MD5SIG
1307 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1308 const union tcp_md5_addr *addr,
1310 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1312 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1313 const union tcp_md5_addr *addr,
1318 #define tcp_twsk_md5_key(twsk) NULL
1321 bool tcp_alloc_md5sig_pool(void);
1323 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1324 static inline void tcp_put_md5sig_pool(void)
1329 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1330 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1331 unsigned int header_len);
1332 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1333 const struct tcp_md5sig_key *key);
1335 /* From tcp_fastopen.c */
1336 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1337 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1338 unsigned long *last_syn_loss);
1339 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1340 struct tcp_fastopen_cookie *cookie, bool syn_lost);
1341 struct tcp_fastopen_request {
1342 /* Fast Open cookie. Size 0 means a cookie request */
1343 struct tcp_fastopen_cookie cookie;
1344 struct msghdr *data; /* data in MSG_FASTOPEN */
1346 int copied; /* queued in tcp_connect() */
1348 void tcp_free_fastopen_req(struct tcp_sock *tp);
1350 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1351 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1352 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1353 struct request_sock *req,
1354 struct tcp_fastopen_cookie *foc,
1355 struct dst_entry *dst);
1356 void tcp_fastopen_init_key_once(bool publish);
1357 #define TCP_FASTOPEN_KEY_LENGTH 16
1359 /* Fastopen key context */
1360 struct tcp_fastopen_context {
1361 struct crypto_cipher *tfm;
1362 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1363 struct rcu_head rcu;
1366 /* write queue abstraction */
1367 static inline void tcp_write_queue_purge(struct sock *sk)
1369 struct sk_buff *skb;
1371 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1372 sk_wmem_free_skb(sk, skb);
1374 tcp_clear_all_retrans_hints(tcp_sk(sk));
1377 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1379 return skb_peek(&sk->sk_write_queue);
1382 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1384 return skb_peek_tail(&sk->sk_write_queue);
1387 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1388 const struct sk_buff *skb)
1390 return skb_queue_next(&sk->sk_write_queue, skb);
1393 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1394 const struct sk_buff *skb)
1396 return skb_queue_prev(&sk->sk_write_queue, skb);
1399 #define tcp_for_write_queue(skb, sk) \
1400 skb_queue_walk(&(sk)->sk_write_queue, skb)
1402 #define tcp_for_write_queue_from(skb, sk) \
1403 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1405 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1406 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1408 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1410 return sk->sk_send_head;
1413 static inline bool tcp_skb_is_last(const struct sock *sk,
1414 const struct sk_buff *skb)
1416 return skb_queue_is_last(&sk->sk_write_queue, skb);
1419 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1421 if (tcp_skb_is_last(sk, skb))
1422 sk->sk_send_head = NULL;
1424 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1427 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1429 if (sk->sk_send_head == skb_unlinked)
1430 sk->sk_send_head = NULL;
1433 static inline void tcp_init_send_head(struct sock *sk)
1435 sk->sk_send_head = NULL;
1438 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1440 __skb_queue_tail(&sk->sk_write_queue, skb);
1443 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1445 __tcp_add_write_queue_tail(sk, skb);
1447 /* Queue it, remembering where we must start sending. */
1448 if (sk->sk_send_head == NULL) {
1449 sk->sk_send_head = skb;
1451 if (tcp_sk(sk)->highest_sack == NULL)
1452 tcp_sk(sk)->highest_sack = skb;
1456 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1458 __skb_queue_head(&sk->sk_write_queue, skb);
1461 /* Insert buff after skb on the write queue of sk. */
1462 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1463 struct sk_buff *buff,
1466 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1469 /* Insert new before skb on the write queue of sk. */
1470 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1471 struct sk_buff *skb,
1474 __skb_queue_before(&sk->sk_write_queue, skb, new);
1476 if (sk->sk_send_head == skb)
1477 sk->sk_send_head = new;
1480 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1482 __skb_unlink(skb, &sk->sk_write_queue);
1485 static inline bool tcp_write_queue_empty(struct sock *sk)
1487 return skb_queue_empty(&sk->sk_write_queue);
1490 static inline void tcp_push_pending_frames(struct sock *sk)
1492 if (tcp_send_head(sk)) {
1493 struct tcp_sock *tp = tcp_sk(sk);
1495 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1499 /* Start sequence of the skb just after the highest skb with SACKed
1500 * bit, valid only if sacked_out > 0 or when the caller has ensured
1501 * validity by itself.
1503 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1505 if (!tp->sacked_out)
1508 if (tp->highest_sack == NULL)
1511 return TCP_SKB_CB(tp->highest_sack)->seq;
1514 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1516 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1517 tcp_write_queue_next(sk, skb);
1520 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1522 return tcp_sk(sk)->highest_sack;
1525 static inline void tcp_highest_sack_reset(struct sock *sk)
1527 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1530 /* Called when old skb is about to be deleted (to be combined with new skb) */
1531 static inline void tcp_highest_sack_combine(struct sock *sk,
1532 struct sk_buff *old,
1533 struct sk_buff *new)
1535 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1536 tcp_sk(sk)->highest_sack = new;
1539 /* Determines whether this is a thin stream (which may suffer from
1540 * increased latency). Used to trigger latency-reducing mechanisms.
1542 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1544 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1548 enum tcp_seq_states {
1549 TCP_SEQ_STATE_LISTENING,
1550 TCP_SEQ_STATE_OPENREQ,
1551 TCP_SEQ_STATE_ESTABLISHED,
1554 int tcp_seq_open(struct inode *inode, struct file *file);
1556 struct tcp_seq_afinfo {
1559 const struct file_operations *seq_fops;
1560 struct seq_operations seq_ops;
1563 struct tcp_iter_state {
1564 struct seq_net_private p;
1566 enum tcp_seq_states state;
1567 struct sock *syn_wait_sk;
1568 int bucket, offset, sbucket, num;
1573 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1574 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1576 extern struct request_sock_ops tcp_request_sock_ops;
1577 extern struct request_sock_ops tcp6_request_sock_ops;
1579 void tcp_v4_destroy_sock(struct sock *sk);
1581 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1582 netdev_features_t features);
1583 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1584 int tcp_gro_complete(struct sk_buff *skb);
1586 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1588 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1590 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1593 static inline bool tcp_stream_memory_free(const struct sock *sk)
1595 const struct tcp_sock *tp = tcp_sk(sk);
1596 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1598 return notsent_bytes < tcp_notsent_lowat(tp);
1601 #ifdef CONFIG_PROC_FS
1602 int tcp4_proc_init(void);
1603 void tcp4_proc_exit(void);
1606 int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
1607 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1608 const struct tcp_request_sock_ops *af_ops,
1609 struct sock *sk, struct sk_buff *skb);
1611 /* TCP af-specific functions */
1612 struct tcp_sock_af_ops {
1613 #ifdef CONFIG_TCP_MD5SIG
1614 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1615 const struct sock *addr_sk);
1616 int (*calc_md5_hash)(char *location,
1617 const struct tcp_md5sig_key *md5,
1618 const struct sock *sk,
1619 const struct sk_buff *skb);
1620 int (*md5_parse)(struct sock *sk,
1621 char __user *optval,
1626 struct tcp_request_sock_ops {
1628 #ifdef CONFIG_TCP_MD5SIG
1629 struct tcp_md5sig_key *(*req_md5_lookup)(struct sock *sk,
1630 const struct sock *addr_sk);
1631 int (*calc_md5_hash) (char *location,
1632 const struct tcp_md5sig_key *md5,
1633 const struct sock *sk,
1634 const struct sk_buff *skb);
1636 void (*init_req)(struct request_sock *req, struct sock *sk,
1637 struct sk_buff *skb);
1638 #ifdef CONFIG_SYN_COOKIES
1639 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1642 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1643 const struct request_sock *req,
1645 __u32 (*init_seq)(const struct sk_buff *skb);
1646 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1647 struct flowi *fl, struct request_sock *req,
1648 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
1649 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1650 const unsigned long timeout);
1653 #ifdef CONFIG_SYN_COOKIES
1654 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1655 struct sock *sk, struct sk_buff *skb,
1658 return ops->cookie_init_seq(sk, skb, mss);
1661 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1662 struct sock *sk, struct sk_buff *skb,
1669 int tcpv4_offload_init(void);
1671 void tcp_v4_init(void);
1672 void tcp_init(void);
1675 * Save and compile IPv4 options, return a pointer to it
1677 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1679 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1680 struct ip_options_rcu *dopt = NULL;
1683 int opt_size = sizeof(*dopt) + opt->optlen;
1685 dopt = kmalloc(opt_size, GFP_ATOMIC);
1686 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1694 /* locally generated TCP pure ACKs have skb->truesize == 2
1695 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1696 * This is much faster than dissecting the packet to find out.
1697 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1699 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1701 return skb->truesize == 2;
1704 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)