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/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.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 extern 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 /* Offer an initial receive window of 10 mss. */
65 #define TCP_DEFAULT_INIT_RCVWND 10
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 512
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
76 /* Maximal reordering. */
77 #define TCP_MAX_REORDERING 127
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
198 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
199 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
200 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
201 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
203 /* But this is what stacks really send out. */
204 #define TCPOLEN_TSTAMP_ALIGNED 12
205 #define TCPOLEN_WSCALE_ALIGNED 4
206 #define TCPOLEN_SACKPERM_ALIGNED 4
207 #define TCPOLEN_SACK_BASE 2
208 #define TCPOLEN_SACK_BASE_ALIGNED 4
209 #define TCPOLEN_SACK_PERBLOCK 8
210 #define TCPOLEN_MD5SIG_ALIGNED 20
211 #define TCPOLEN_MSS_ALIGNED 4
213 /* Flags in tp->nonagle */
214 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
215 #define TCP_NAGLE_CORK 2 /* Socket is corked */
216 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
218 /* TCP thin-stream limits */
219 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
221 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
222 #define TCP_INIT_CWND 10
224 /* Bit Flags for sysctl_tcp_fastopen */
225 #define TFO_CLIENT_ENABLE 1
226 #define TFO_SERVER_ENABLE 2
227 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
229 /* Process SYN data but skip cookie validation */
230 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100
231 /* Accept SYN data w/o any cookie option */
232 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
234 /* Force enable TFO on all listeners, i.e., not requiring the
235 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
237 #define TFO_SERVER_WO_SOCKOPT1 0x400
238 #define TFO_SERVER_WO_SOCKOPT2 0x800
239 /* Always create TFO child sockets on a TFO listener even when
240 * cookie/data not present. (For testing purpose!)
242 #define TFO_SERVER_ALWAYS 0x1000
244 extern struct inet_timewait_death_row tcp_death_row;
246 /* sysctl variables for tcp */
247 extern int sysctl_tcp_timestamps;
248 extern int sysctl_tcp_window_scaling;
249 extern int sysctl_tcp_sack;
250 extern int sysctl_tcp_fin_timeout;
251 extern int sysctl_tcp_keepalive_time;
252 extern int sysctl_tcp_keepalive_probes;
253 extern int sysctl_tcp_keepalive_intvl;
254 extern int sysctl_tcp_syn_retries;
255 extern int sysctl_tcp_synack_retries;
256 extern int sysctl_tcp_retries1;
257 extern int sysctl_tcp_retries2;
258 extern int sysctl_tcp_orphan_retries;
259 extern int sysctl_tcp_syncookies;
260 extern int sysctl_tcp_fastopen;
261 extern int sysctl_tcp_retrans_collapse;
262 extern int sysctl_tcp_stdurg;
263 extern int sysctl_tcp_rfc1337;
264 extern int sysctl_tcp_abort_on_overflow;
265 extern int sysctl_tcp_max_orphans;
266 extern int sysctl_tcp_fack;
267 extern int sysctl_tcp_reordering;
268 extern int sysctl_tcp_dsack;
269 extern int sysctl_tcp_wmem[3];
270 extern int sysctl_tcp_rmem[3];
271 extern int sysctl_tcp_app_win;
272 extern int sysctl_tcp_adv_win_scale;
273 extern int sysctl_tcp_tw_reuse;
274 extern int sysctl_tcp_frto;
275 extern int sysctl_tcp_low_latency;
276 extern int sysctl_tcp_dma_copybreak;
277 extern int sysctl_tcp_nometrics_save;
278 extern int sysctl_tcp_moderate_rcvbuf;
279 extern int sysctl_tcp_tso_win_divisor;
280 extern int sysctl_tcp_mtu_probing;
281 extern int sysctl_tcp_base_mss;
282 extern int sysctl_tcp_workaround_signed_windows;
283 extern int sysctl_tcp_slow_start_after_idle;
284 extern int sysctl_tcp_max_ssthresh;
285 extern int sysctl_tcp_thin_linear_timeouts;
286 extern int sysctl_tcp_thin_dupack;
287 extern int sysctl_tcp_early_retrans;
288 extern int sysctl_tcp_limit_output_bytes;
289 extern int sysctl_tcp_challenge_ack_limit;
290 extern int sysctl_tcp_min_tso_segs;
292 extern atomic_long_t tcp_memory_allocated;
293 extern struct percpu_counter tcp_sockets_allocated;
294 extern int tcp_memory_pressure;
297 * The next routines deal with comparing 32 bit unsigned ints
298 * and worry about wraparound (automatic with unsigned arithmetic).
301 static inline bool before(__u32 seq1, __u32 seq2)
303 return (__s32)(seq1-seq2) < 0;
305 #define after(seq2, seq1) before(seq1, seq2)
307 /* is s2<=s1<=s3 ? */
308 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
310 return seq3 - seq2 >= seq1 - seq2;
313 static inline bool tcp_out_of_memory(struct sock *sk)
315 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
316 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
321 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
323 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
324 int orphans = percpu_counter_read_positive(ocp);
326 if (orphans << shift > sysctl_tcp_max_orphans) {
327 orphans = percpu_counter_sum_positive(ocp);
328 if (orphans << shift > sysctl_tcp_max_orphans)
334 extern bool tcp_check_oom(struct sock *sk, int shift);
336 /* syncookies: remember time of last synqueue overflow */
337 static inline void tcp_synq_overflow(struct sock *sk)
339 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
342 /* syncookies: no recent synqueue overflow on this listening socket? */
343 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
345 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
346 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
349 extern struct proto tcp_prot;
351 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
352 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
353 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
354 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
355 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
357 extern void tcp_init_mem(struct net *net);
359 extern void tcp_tasklet_init(void);
361 extern void tcp_v4_err(struct sk_buff *skb, u32);
363 extern void tcp_shutdown (struct sock *sk, int how);
365 extern void tcp_v4_early_demux(struct sk_buff *skb);
366 extern int tcp_v4_rcv(struct sk_buff *skb);
368 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
369 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
371 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
372 size_t size, int flags);
373 extern void tcp_release_cb(struct sock *sk);
374 extern void tcp_wfree(struct sk_buff *skb);
375 extern void tcp_write_timer_handler(struct sock *sk);
376 extern void tcp_delack_timer_handler(struct sock *sk);
377 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
378 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
379 const struct tcphdr *th, unsigned int len);
380 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
381 const struct tcphdr *th, unsigned int len);
382 extern void tcp_rcv_space_adjust(struct sock *sk);
383 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
384 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
385 extern void tcp_twsk_destructor(struct sock *sk);
386 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
387 struct pipe_inode_info *pipe, size_t len,
390 static inline void tcp_dec_quickack_mode(struct sock *sk,
391 const unsigned int pkts)
393 struct inet_connection_sock *icsk = inet_csk(sk);
395 if (icsk->icsk_ack.quick) {
396 if (pkts >= icsk->icsk_ack.quick) {
397 icsk->icsk_ack.quick = 0;
398 /* Leaving quickack mode we deflate ATO. */
399 icsk->icsk_ack.ato = TCP_ATO_MIN;
401 icsk->icsk_ack.quick -= pkts;
406 #define TCP_ECN_QUEUE_CWR 2
407 #define TCP_ECN_DEMAND_CWR 4
408 #define TCP_ECN_SEEN 8
418 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
420 const struct tcphdr *th);
421 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
422 struct request_sock *req,
423 struct request_sock **prev,
425 extern int tcp_child_process(struct sock *parent, struct sock *child,
426 struct sk_buff *skb);
427 extern void tcp_enter_loss(struct sock *sk, int how);
428 extern void tcp_clear_retrans(struct tcp_sock *tp);
429 extern void tcp_update_metrics(struct sock *sk);
430 extern void tcp_init_metrics(struct sock *sk);
431 extern void tcp_metrics_init(void);
432 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check);
433 extern bool tcp_remember_stamp(struct sock *sk);
434 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
435 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
436 extern void tcp_disable_fack(struct tcp_sock *tp);
437 extern void tcp_close(struct sock *sk, long timeout);
438 extern void tcp_init_sock(struct sock *sk);
439 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
440 struct poll_table_struct *wait);
441 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
442 char __user *optval, int __user *optlen);
443 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
444 char __user *optval, unsigned int optlen);
445 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
446 char __user *optval, int __user *optlen);
447 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
448 char __user *optval, unsigned int optlen);
449 extern void tcp_set_keepalive(struct sock *sk, int val);
450 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
451 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
452 size_t len, int nonblock, int flags, int *addr_len);
453 extern void tcp_parse_options(const struct sk_buff *skb,
454 struct tcp_options_received *opt_rx,
455 int estab, struct tcp_fastopen_cookie *foc);
456 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
459 * TCP v4 functions exported for the inet6 API
462 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
463 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
464 extern struct sock * tcp_create_openreq_child(struct sock *sk,
465 struct request_sock *req,
466 struct sk_buff *skb);
467 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
468 struct request_sock *req,
469 struct dst_entry *dst);
470 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
471 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
473 extern int tcp_connect(struct sock *sk);
474 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
475 struct request_sock *req,
476 struct tcp_fastopen_cookie *foc);
477 extern int tcp_disconnect(struct sock *sk, int flags);
479 void tcp_connect_init(struct sock *sk);
480 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
481 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
482 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
484 /* From syncookies.c */
485 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
486 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
487 struct ip_options *opt);
488 #ifdef CONFIG_SYN_COOKIES
489 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
492 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
500 extern __u32 cookie_init_timestamp(struct request_sock *req);
501 extern bool cookie_check_timestamp(struct tcp_options_received *opt,
502 struct net *net, bool *ecn_ok);
504 /* From net/ipv6/syncookies.c */
505 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
506 #ifdef CONFIG_SYN_COOKIES
507 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
510 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
519 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
521 extern bool tcp_may_send_now(struct sock *sk);
522 extern int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
523 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
524 extern void tcp_retransmit_timer(struct sock *sk);
525 extern void tcp_xmit_retransmit_queue(struct sock *);
526 extern void tcp_simple_retransmit(struct sock *);
527 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
528 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
530 extern void tcp_send_probe0(struct sock *);
531 extern void tcp_send_partial(struct sock *);
532 extern int tcp_write_wakeup(struct sock *);
533 extern void tcp_send_fin(struct sock *sk);
534 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
535 extern int tcp_send_synack(struct sock *);
536 extern bool tcp_syn_flood_action(struct sock *sk,
537 const struct sk_buff *skb,
539 extern void tcp_push_one(struct sock *, unsigned int mss_now);
540 extern void tcp_send_ack(struct sock *sk);
541 extern void tcp_send_delayed_ack(struct sock *sk);
542 extern void tcp_send_loss_probe(struct sock *sk);
543 extern bool tcp_schedule_loss_probe(struct sock *sk);
546 extern void tcp_cwnd_application_limited(struct sock *sk);
547 extern void tcp_resume_early_retransmit(struct sock *sk);
548 extern void tcp_rearm_rto(struct sock *sk);
549 extern void tcp_reset(struct sock *sk);
552 extern void tcp_init_xmit_timers(struct sock *);
553 static inline void tcp_clear_xmit_timers(struct sock *sk)
555 inet_csk_clear_xmit_timers(sk);
558 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
559 extern unsigned int tcp_current_mss(struct sock *sk);
561 /* Bound MSS / TSO packet size with the half of the window */
562 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
566 /* When peer uses tiny windows, there is no use in packetizing
567 * to sub-MSS pieces for the sake of SWS or making sure there
568 * are enough packets in the pipe for fast recovery.
570 * On the other hand, for extremely large MSS devices, handling
571 * smaller than MSS windows in this way does make sense.
573 if (tp->max_window >= 512)
574 cutoff = (tp->max_window >> 1);
576 cutoff = tp->max_window;
578 if (cutoff && pktsize > cutoff)
579 return max_t(int, cutoff, 68U - tp->tcp_header_len);
585 extern void tcp_get_info(const struct sock *, struct tcp_info *);
587 /* Read 'sendfile()'-style from a TCP socket */
588 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
589 unsigned int, size_t);
590 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
591 sk_read_actor_t recv_actor);
593 extern void tcp_initialize_rcv_mss(struct sock *sk);
595 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
596 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
597 extern void tcp_mtup_init(struct sock *sk);
598 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
599 extern void tcp_init_buffer_space(struct sock *sk);
601 static inline void tcp_bound_rto(const struct sock *sk)
603 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
604 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
607 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
609 return (tp->srtt >> 3) + tp->rttvar;
612 extern void tcp_set_rto(struct sock *sk);
614 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
616 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
617 ntohl(TCP_FLAG_ACK) |
621 static inline void tcp_fast_path_on(struct tcp_sock *tp)
623 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
626 static inline void tcp_fast_path_check(struct sock *sk)
628 struct tcp_sock *tp = tcp_sk(sk);
630 if (skb_queue_empty(&tp->out_of_order_queue) &&
632 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
634 tcp_fast_path_on(tp);
637 /* Compute the actual rto_min value */
638 static inline u32 tcp_rto_min(struct sock *sk)
640 const struct dst_entry *dst = __sk_dst_get(sk);
641 u32 rto_min = TCP_RTO_MIN;
643 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
644 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
648 /* Compute the actual receive window we are currently advertising.
649 * Rcv_nxt can be after the window if our peer push more data
650 * than the offered window.
652 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
654 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
661 /* Choose a new window, without checks for shrinking, and without
662 * scaling applied to the result. The caller does these things
663 * if necessary. This is a "raw" window selection.
665 extern u32 __tcp_select_window(struct sock *sk);
667 void tcp_send_window_probe(struct sock *sk);
669 /* TCP timestamps are only 32-bits, this causes a slight
670 * complication on 64-bit systems since we store a snapshot
671 * of jiffies in the buffer control blocks below. We decided
672 * to use only the low 32-bits of jiffies and hide the ugly
673 * casts with the following macro.
675 #define tcp_time_stamp ((__u32)(jiffies))
677 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
679 #define TCPHDR_FIN 0x01
680 #define TCPHDR_SYN 0x02
681 #define TCPHDR_RST 0x04
682 #define TCPHDR_PSH 0x08
683 #define TCPHDR_ACK 0x10
684 #define TCPHDR_URG 0x20
685 #define TCPHDR_ECE 0x40
686 #define TCPHDR_CWR 0x80
688 /* This is what the send packet queuing engine uses to pass
689 * TCP per-packet control information to the transmission code.
690 * We also store the host-order sequence numbers in here too.
691 * This is 44 bytes if IPV6 is enabled.
692 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
696 struct inet_skb_parm h4;
697 #if IS_ENABLED(CONFIG_IPV6)
698 struct inet6_skb_parm h6;
700 } header; /* For incoming frames */
701 __u32 seq; /* Starting sequence number */
702 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
703 __u32 when; /* used to compute rtt's */
704 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
706 __u8 sacked; /* State flags for SACK/FACK. */
707 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
708 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
709 #define TCPCB_LOST 0x04 /* SKB is lost */
710 #define TCPCB_TAGBITS 0x07 /* All tag bits */
711 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
712 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
714 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
716 __u32 ack_seq; /* Sequence number ACK'd */
719 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
721 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
723 * If we receive a SYN packet with these bits set, it means a network is
724 * playing bad games with TOS bits. In order to avoid possible false congestion
725 * notifications, we disable TCP ECN negociation.
728 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
731 const struct tcphdr *th = tcp_hdr(skb);
733 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr &&
734 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
735 inet_rsk(req)->ecn_ok = 1;
738 /* Due to TSO, an SKB can be composed of multiple actual
739 * packets. To keep these tracked properly, we use this.
741 static inline int tcp_skb_pcount(const struct sk_buff *skb)
743 return skb_shinfo(skb)->gso_segs;
746 /* This is valid iff tcp_skb_pcount() > 1. */
747 static inline int tcp_skb_mss(const struct sk_buff *skb)
749 return skb_shinfo(skb)->gso_size;
752 /* Events passed to congestion control interface */
754 CA_EVENT_TX_START, /* first transmit when no packets in flight */
755 CA_EVENT_CWND_RESTART, /* congestion window restart */
756 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
757 CA_EVENT_LOSS, /* loss timeout */
758 CA_EVENT_FAST_ACK, /* in sequence ack */
759 CA_EVENT_SLOW_ACK, /* other ack */
763 * Interface for adding new TCP congestion control handlers
765 #define TCP_CA_NAME_MAX 16
766 #define TCP_CA_MAX 128
767 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
769 #define TCP_CONG_NON_RESTRICTED 0x1
770 #define TCP_CONG_RTT_STAMP 0x2
772 struct tcp_congestion_ops {
773 struct list_head list;
776 /* initialize private data (optional) */
777 void (*init)(struct sock *sk);
778 /* cleanup private data (optional) */
779 void (*release)(struct sock *sk);
781 /* return slow start threshold (required) */
782 u32 (*ssthresh)(struct sock *sk);
783 /* lower bound for congestion window (optional) */
784 u32 (*min_cwnd)(const struct sock *sk);
785 /* do new cwnd calculation (required) */
786 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
787 /* call before changing ca_state (optional) */
788 void (*set_state)(struct sock *sk, u8 new_state);
789 /* call when cwnd event occurs (optional) */
790 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
791 /* new value of cwnd after loss (optional) */
792 u32 (*undo_cwnd)(struct sock *sk);
793 /* hook for packet ack accounting (optional) */
794 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
795 /* get info for inet_diag (optional) */
796 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
798 char name[TCP_CA_NAME_MAX];
799 struct module *owner;
802 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
803 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
805 extern void tcp_init_congestion_control(struct sock *sk);
806 extern void tcp_cleanup_congestion_control(struct sock *sk);
807 extern int tcp_set_default_congestion_control(const char *name);
808 extern void tcp_get_default_congestion_control(char *name);
809 extern void tcp_get_available_congestion_control(char *buf, size_t len);
810 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
811 extern int tcp_set_allowed_congestion_control(char *allowed);
812 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
813 extern void tcp_slow_start(struct tcp_sock *tp);
814 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
816 extern struct tcp_congestion_ops tcp_init_congestion_ops;
817 extern u32 tcp_reno_ssthresh(struct sock *sk);
818 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
819 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
820 extern struct tcp_congestion_ops tcp_reno;
822 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
824 struct inet_connection_sock *icsk = inet_csk(sk);
826 if (icsk->icsk_ca_ops->set_state)
827 icsk->icsk_ca_ops->set_state(sk, ca_state);
828 icsk->icsk_ca_state = ca_state;
831 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
833 const struct inet_connection_sock *icsk = inet_csk(sk);
835 if (icsk->icsk_ca_ops->cwnd_event)
836 icsk->icsk_ca_ops->cwnd_event(sk, event);
839 /* These functions determine how the current flow behaves in respect of SACK
840 * handling. SACK is negotiated with the peer, and therefore it can vary
841 * between different flows.
843 * tcp_is_sack - SACK enabled
844 * tcp_is_reno - No SACK
845 * tcp_is_fack - FACK enabled, implies SACK enabled
847 static inline int tcp_is_sack(const struct tcp_sock *tp)
849 return tp->rx_opt.sack_ok;
852 static inline bool tcp_is_reno(const struct tcp_sock *tp)
854 return !tcp_is_sack(tp);
857 static inline bool tcp_is_fack(const struct tcp_sock *tp)
859 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
862 static inline void tcp_enable_fack(struct tcp_sock *tp)
864 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
867 /* TCP early-retransmit (ER) is similar to but more conservative than
868 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
870 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
872 tp->do_early_retrans = sysctl_tcp_early_retrans &&
873 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
874 sysctl_tcp_reordering == 3;
877 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
879 tp->do_early_retrans = 0;
882 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
884 return tp->sacked_out + tp->lost_out;
887 /* This determines how many packets are "in the network" to the best
888 * of our knowledge. In many cases it is conservative, but where
889 * detailed information is available from the receiver (via SACK
890 * blocks etc.) we can make more aggressive calculations.
892 * Use this for decisions involving congestion control, use just
893 * tp->packets_out to determine if the send queue is empty or not.
895 * Read this equation as:
897 * "Packets sent once on transmission queue" MINUS
898 * "Packets left network, but not honestly ACKed yet" PLUS
899 * "Packets fast retransmitted"
901 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
903 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
906 #define TCP_INFINITE_SSTHRESH 0x7fffffff
908 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
910 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
913 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
915 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
916 (1 << inet_csk(sk)->icsk_ca_state);
919 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
920 * The exception is cwnd reduction phase, when cwnd is decreasing towards
923 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
925 const struct tcp_sock *tp = tcp_sk(sk);
927 if (tcp_in_cwnd_reduction(sk))
928 return tp->snd_ssthresh;
930 return max(tp->snd_ssthresh,
931 ((tp->snd_cwnd >> 1) +
932 (tp->snd_cwnd >> 2)));
935 /* Use define here intentionally to get WARN_ON location shown at the caller */
936 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
938 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
939 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
941 /* The maximum number of MSS of available cwnd for which TSO defers
942 * sending if not using sysctl_tcp_tso_win_divisor.
944 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
949 /* Slow start with delack produces 3 packets of burst, so that
950 * it is safe "de facto". This will be the default - same as
951 * the default reordering threshold - but if reordering increases,
952 * we must be able to allow cwnd to burst at least this much in order
953 * to not pull it back when holes are filled.
955 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
957 return tp->reordering;
960 /* Returns end sequence number of the receiver's advertised window */
961 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
963 return tp->snd_una + tp->snd_wnd;
965 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
967 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
968 const struct sk_buff *skb)
971 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
974 static inline void tcp_check_probe_timer(struct sock *sk)
976 const struct tcp_sock *tp = tcp_sk(sk);
977 const struct inet_connection_sock *icsk = inet_csk(sk);
979 if (!tp->packets_out && !icsk->icsk_pending)
980 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
981 icsk->icsk_rto, TCP_RTO_MAX);
984 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
989 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
995 * Calculate(/check) TCP checksum
997 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
998 __be32 daddr, __wsum base)
1000 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1003 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1005 return __skb_checksum_complete(skb);
1008 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1010 return !skb_csum_unnecessary(skb) &&
1011 __tcp_checksum_complete(skb);
1014 /* Prequeue for VJ style copy to user, combined with checksumming. */
1016 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1018 tp->ucopy.task = NULL;
1020 tp->ucopy.memory = 0;
1021 skb_queue_head_init(&tp->ucopy.prequeue);
1022 #ifdef CONFIG_NET_DMA
1023 tp->ucopy.dma_chan = NULL;
1024 tp->ucopy.wakeup = 0;
1025 tp->ucopy.pinned_list = NULL;
1026 tp->ucopy.dma_cookie = 0;
1030 extern bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1035 static const char *statename[]={
1036 "Unused","Established","Syn Sent","Syn Recv",
1037 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1038 "Close Wait","Last ACK","Listen","Closing"
1041 extern void tcp_set_state(struct sock *sk, int state);
1043 extern void tcp_done(struct sock *sk);
1045 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1048 rx_opt->num_sacks = 0;
1051 /* Determine a window scaling and initial window to offer. */
1052 extern void tcp_select_initial_window(int __space, __u32 mss,
1053 __u32 *rcv_wnd, __u32 *window_clamp,
1054 int wscale_ok, __u8 *rcv_wscale,
1055 __u32 init_rcv_wnd);
1057 static inline int tcp_win_from_space(int space)
1059 return sysctl_tcp_adv_win_scale<=0 ?
1060 (space>>(-sysctl_tcp_adv_win_scale)) :
1061 space - (space>>sysctl_tcp_adv_win_scale);
1064 /* Note: caller must be prepared to deal with negative returns */
1065 static inline int tcp_space(const struct sock *sk)
1067 return tcp_win_from_space(sk->sk_rcvbuf -
1068 atomic_read(&sk->sk_rmem_alloc));
1071 static inline int tcp_full_space(const struct sock *sk)
1073 return tcp_win_from_space(sk->sk_rcvbuf);
1076 static inline void tcp_openreq_init(struct request_sock *req,
1077 struct tcp_options_received *rx_opt,
1078 struct sk_buff *skb)
1080 struct inet_request_sock *ireq = inet_rsk(req);
1082 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1084 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1085 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1086 tcp_rsk(req)->snt_synack = 0;
1087 req->mss = rx_opt->mss_clamp;
1088 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1089 ireq->tstamp_ok = rx_opt->tstamp_ok;
1090 ireq->sack_ok = rx_opt->sack_ok;
1091 ireq->snd_wscale = rx_opt->snd_wscale;
1092 ireq->wscale_ok = rx_opt->wscale_ok;
1095 ireq->rmt_port = tcp_hdr(skb)->source;
1096 ireq->loc_port = tcp_hdr(skb)->dest;
1099 /* Compute time elapsed between SYNACK and the ACK completing 3WHS */
1100 static inline void tcp_synack_rtt_meas(struct sock *sk,
1101 struct request_sock *req)
1103 if (tcp_rsk(req)->snt_synack)
1104 tcp_valid_rtt_meas(sk,
1105 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1108 extern void tcp_enter_memory_pressure(struct sock *sk);
1110 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1112 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1115 static inline int keepalive_time_when(const struct tcp_sock *tp)
1117 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1120 static inline int keepalive_probes(const struct tcp_sock *tp)
1122 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1125 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1127 const struct inet_connection_sock *icsk = &tp->inet_conn;
1129 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1130 tcp_time_stamp - tp->rcv_tstamp);
1133 static inline int tcp_fin_time(const struct sock *sk)
1135 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1136 const int rto = inet_csk(sk)->icsk_rto;
1138 if (fin_timeout < (rto << 2) - (rto >> 1))
1139 fin_timeout = (rto << 2) - (rto >> 1);
1144 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1147 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1149 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1152 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1153 * then following tcp messages have valid values. Ignore 0 value,
1154 * or else 'negative' tsval might forbid us to accept their packets.
1156 if (!rx_opt->ts_recent)
1161 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1164 if (tcp_paws_check(rx_opt, 0))
1167 /* RST segments are not recommended to carry timestamp,
1168 and, if they do, it is recommended to ignore PAWS because
1169 "their cleanup function should take precedence over timestamps."
1170 Certainly, it is mistake. It is necessary to understand the reasons
1171 of this constraint to relax it: if peer reboots, clock may go
1172 out-of-sync and half-open connections will not be reset.
1173 Actually, the problem would be not existing if all
1174 the implementations followed draft about maintaining clock
1175 via reboots. Linux-2.2 DOES NOT!
1177 However, we can relax time bounds for RST segments to MSL.
1179 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1184 static inline void tcp_mib_init(struct net *net)
1187 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1188 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1189 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1190 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1194 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1196 tp->lost_skb_hint = NULL;
1197 tp->scoreboard_skb_hint = NULL;
1200 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1202 tcp_clear_retrans_hints_partial(tp);
1203 tp->retransmit_skb_hint = NULL;
1209 union tcp_md5_addr {
1211 #if IS_ENABLED(CONFIG_IPV6)
1216 /* - key database */
1217 struct tcp_md5sig_key {
1218 struct hlist_node node;
1220 u8 family; /* AF_INET or AF_INET6 */
1221 union tcp_md5_addr addr;
1222 u8 key[TCP_MD5SIG_MAXKEYLEN];
1223 struct rcu_head rcu;
1227 struct tcp_md5sig_info {
1228 struct hlist_head head;
1229 struct rcu_head rcu;
1232 /* - pseudo header */
1233 struct tcp4_pseudohdr {
1241 struct tcp6_pseudohdr {
1242 struct in6_addr saddr;
1243 struct in6_addr daddr;
1245 __be32 protocol; /* including padding */
1248 union tcp_md5sum_block {
1249 struct tcp4_pseudohdr ip4;
1250 #if IS_ENABLED(CONFIG_IPV6)
1251 struct tcp6_pseudohdr ip6;
1255 /* - pool: digest algorithm, hash description and scratch buffer */
1256 struct tcp_md5sig_pool {
1257 struct hash_desc md5_desc;
1258 union tcp_md5sum_block md5_blk;
1262 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1263 const struct sock *sk,
1264 const struct request_sock *req,
1265 const struct sk_buff *skb);
1266 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1267 int family, const u8 *newkey,
1268 u8 newkeylen, gfp_t gfp);
1269 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1271 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1272 struct sock *addr_sk);
1274 #ifdef CONFIG_TCP_MD5SIG
1275 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1276 const union tcp_md5_addr *addr, int family);
1277 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1279 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1280 const union tcp_md5_addr *addr,
1285 #define tcp_twsk_md5_key(twsk) NULL
1288 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1289 extern void tcp_free_md5sig_pool(void);
1291 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1292 extern void tcp_put_md5sig_pool(void);
1294 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1295 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1296 unsigned int header_len);
1297 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1298 const struct tcp_md5sig_key *key);
1300 /* From tcp_fastopen.c */
1301 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1302 struct tcp_fastopen_cookie *cookie,
1303 int *syn_loss, unsigned long *last_syn_loss);
1304 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1305 struct tcp_fastopen_cookie *cookie,
1307 struct tcp_fastopen_request {
1308 /* Fast Open cookie. Size 0 means a cookie request */
1309 struct tcp_fastopen_cookie cookie;
1310 struct msghdr *data; /* data in MSG_FASTOPEN */
1312 int copied; /* queued in tcp_connect() */
1314 void tcp_free_fastopen_req(struct tcp_sock *tp);
1316 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1317 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1318 void tcp_fastopen_cookie_gen(__be32 addr, struct tcp_fastopen_cookie *foc);
1320 #define TCP_FASTOPEN_KEY_LENGTH 16
1322 /* Fastopen key context */
1323 struct tcp_fastopen_context {
1324 struct crypto_cipher __rcu *tfm;
1325 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1326 struct rcu_head rcu;
1329 /* write queue abstraction */
1330 static inline void tcp_write_queue_purge(struct sock *sk)
1332 struct sk_buff *skb;
1334 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1335 sk_wmem_free_skb(sk, skb);
1337 tcp_clear_all_retrans_hints(tcp_sk(sk));
1340 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1342 return skb_peek(&sk->sk_write_queue);
1345 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1347 return skb_peek_tail(&sk->sk_write_queue);
1350 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1351 const struct sk_buff *skb)
1353 return skb_queue_next(&sk->sk_write_queue, skb);
1356 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1357 const struct sk_buff *skb)
1359 return skb_queue_prev(&sk->sk_write_queue, skb);
1362 #define tcp_for_write_queue(skb, sk) \
1363 skb_queue_walk(&(sk)->sk_write_queue, skb)
1365 #define tcp_for_write_queue_from(skb, sk) \
1366 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1368 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1369 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1371 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1373 return sk->sk_send_head;
1376 static inline bool tcp_skb_is_last(const struct sock *sk,
1377 const struct sk_buff *skb)
1379 return skb_queue_is_last(&sk->sk_write_queue, skb);
1382 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1384 if (tcp_skb_is_last(sk, skb))
1385 sk->sk_send_head = NULL;
1387 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1390 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1392 if (sk->sk_send_head == skb_unlinked)
1393 sk->sk_send_head = NULL;
1396 static inline void tcp_init_send_head(struct sock *sk)
1398 sk->sk_send_head = NULL;
1401 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1403 __skb_queue_tail(&sk->sk_write_queue, skb);
1406 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1408 __tcp_add_write_queue_tail(sk, skb);
1410 /* Queue it, remembering where we must start sending. */
1411 if (sk->sk_send_head == NULL) {
1412 sk->sk_send_head = skb;
1414 if (tcp_sk(sk)->highest_sack == NULL)
1415 tcp_sk(sk)->highest_sack = skb;
1419 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1421 __skb_queue_head(&sk->sk_write_queue, skb);
1424 /* Insert buff after skb on the write queue of sk. */
1425 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1426 struct sk_buff *buff,
1429 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1432 /* Insert new before skb on the write queue of sk. */
1433 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1434 struct sk_buff *skb,
1437 __skb_queue_before(&sk->sk_write_queue, skb, new);
1439 if (sk->sk_send_head == skb)
1440 sk->sk_send_head = new;
1443 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1445 __skb_unlink(skb, &sk->sk_write_queue);
1448 static inline bool tcp_write_queue_empty(struct sock *sk)
1450 return skb_queue_empty(&sk->sk_write_queue);
1453 static inline void tcp_push_pending_frames(struct sock *sk)
1455 if (tcp_send_head(sk)) {
1456 struct tcp_sock *tp = tcp_sk(sk);
1458 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1462 /* Start sequence of the skb just after the highest skb with SACKed
1463 * bit, valid only if sacked_out > 0 or when the caller has ensured
1464 * validity by itself.
1466 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1468 if (!tp->sacked_out)
1471 if (tp->highest_sack == NULL)
1474 return TCP_SKB_CB(tp->highest_sack)->seq;
1477 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1479 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1480 tcp_write_queue_next(sk, skb);
1483 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1485 return tcp_sk(sk)->highest_sack;
1488 static inline void tcp_highest_sack_reset(struct sock *sk)
1490 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1493 /* Called when old skb is about to be deleted (to be combined with new skb) */
1494 static inline void tcp_highest_sack_combine(struct sock *sk,
1495 struct sk_buff *old,
1496 struct sk_buff *new)
1498 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1499 tcp_sk(sk)->highest_sack = new;
1502 /* Determines whether this is a thin stream (which may suffer from
1503 * increased latency). Used to trigger latency-reducing mechanisms.
1505 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1507 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1511 enum tcp_seq_states {
1512 TCP_SEQ_STATE_LISTENING,
1513 TCP_SEQ_STATE_OPENREQ,
1514 TCP_SEQ_STATE_ESTABLISHED,
1515 TCP_SEQ_STATE_TIME_WAIT,
1518 int tcp_seq_open(struct inode *inode, struct file *file);
1520 struct tcp_seq_afinfo {
1523 const struct file_operations *seq_fops;
1524 struct seq_operations seq_ops;
1527 struct tcp_iter_state {
1528 struct seq_net_private p;
1530 enum tcp_seq_states state;
1531 struct sock *syn_wait_sk;
1532 int bucket, offset, sbucket, num;
1537 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1538 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1540 extern struct request_sock_ops tcp_request_sock_ops;
1541 extern struct request_sock_ops tcp6_request_sock_ops;
1543 extern void tcp_v4_destroy_sock(struct sock *sk);
1545 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1546 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1547 netdev_features_t features);
1548 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1549 struct sk_buff *skb);
1550 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1551 struct sk_buff *skb);
1552 extern int tcp_gro_complete(struct sk_buff *skb);
1553 extern int tcp4_gro_complete(struct sk_buff *skb);
1555 #ifdef CONFIG_PROC_FS
1556 extern int tcp4_proc_init(void);
1557 extern void tcp4_proc_exit(void);
1560 /* TCP af-specific functions */
1561 struct tcp_sock_af_ops {
1562 #ifdef CONFIG_TCP_MD5SIG
1563 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1564 struct sock *addr_sk);
1565 int (*calc_md5_hash) (char *location,
1566 struct tcp_md5sig_key *md5,
1567 const struct sock *sk,
1568 const struct request_sock *req,
1569 const struct sk_buff *skb);
1570 int (*md5_parse) (struct sock *sk,
1571 char __user *optval,
1576 struct tcp_request_sock_ops {
1577 #ifdef CONFIG_TCP_MD5SIG
1578 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1579 struct request_sock *req);
1580 int (*calc_md5_hash) (char *location,
1581 struct tcp_md5sig_key *md5,
1582 const struct sock *sk,
1583 const struct request_sock *req,
1584 const struct sk_buff *skb);
1588 extern void tcp_v4_init(void);
1589 extern void tcp_init(void);