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 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 512
70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
71 #define TCP_FASTRETRANS_THRESH 3
73 /* Maximal reordering. */
74 #define TCP_MAX_REORDERING 127
76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
77 #define TCP_MAX_QUICKACKS 16U
80 #define TCP_URG_VALID 0x0100
81 #define TCP_URG_NOTYET 0x0200
82 #define TCP_URG_READ 0x0400
84 #define TCP_RETR1 3 /*
85 * This is how many retries it does before it
86 * tries to figure out if the gateway is
87 * down. Minimal RFC value is 3; it corresponds
88 * to ~3sec-8min depending on RTO.
91 #define TCP_RETR2 15 /*
92 * This should take at least
93 * 90 minutes to time out.
94 * RFC1122 says that the limit is 100 sec.
95 * 15 is ~13-30min depending on RTO.
98 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
99 * when active opening a connection.
100 * RFC1122 says the minimum retry MUST
101 * be at least 180secs. Nevertheless
102 * this value is corresponding to
103 * 63secs of retransmission with the
104 * current initial RTO.
107 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
108 * when passive opening a connection.
109 * This is corresponding to 31secs of
110 * retransmission with the current
114 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
115 * state, about 60 seconds */
116 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
117 /* BSD style FIN_WAIT2 deadlock breaker.
118 * It used to be 3min, new value is 60sec,
119 * to combine FIN-WAIT-2 timeout with
123 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
125 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
126 #define TCP_ATO_MIN ((unsigned)(HZ/25))
128 #define TCP_DELACK_MIN 4U
129 #define TCP_ATO_MIN 4U
131 #define TCP_RTO_MAX ((unsigned)(120*HZ))
132 #define TCP_RTO_MIN ((unsigned)(HZ/5))
133 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
134 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
135 * used as a fallback RTO for the
136 * initial data transmission if no
137 * valid RTT sample has been acquired,
138 * most likely due to retrans in 3WHS.
141 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
142 * for local resources.
145 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
146 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
147 #define TCP_KEEPALIVE_INTVL (75*HZ)
149 #define MAX_TCP_KEEPIDLE 32767
150 #define MAX_TCP_KEEPINTVL 32767
151 #define MAX_TCP_KEEPCNT 127
152 #define MAX_TCP_SYNCNT 127
154 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
156 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
157 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
158 * after this time. It should be equal
159 * (or greater than) TCP_TIMEWAIT_LEN
160 * to provide reliability equal to one
161 * provided by timewait state.
163 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
164 * timestamps. It must be less than
165 * minimal timewait lifetime.
171 #define TCPOPT_NOP 1 /* Padding */
172 #define TCPOPT_EOL 0 /* End of options */
173 #define TCPOPT_MSS 2 /* Segment size negotiating */
174 #define TCPOPT_WINDOW 3 /* Window scaling */
175 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
176 #define TCPOPT_SACK 5 /* SACK Block */
177 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
178 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
179 #define TCPOPT_EXP 254 /* Experimental */
180 /* Magic number to be after the option value for sharing TCP
181 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
183 #define TCPOPT_FASTOPEN_MAGIC 0xF989
189 #define TCPOLEN_MSS 4
190 #define TCPOLEN_WINDOW 3
191 #define TCPOLEN_SACK_PERM 2
192 #define TCPOLEN_TIMESTAMP 10
193 #define TCPOLEN_MD5SIG 18
194 #define TCPOLEN_EXP_FASTOPEN_BASE 4
196 /* But this is what stacks really send out. */
197 #define TCPOLEN_TSTAMP_ALIGNED 12
198 #define TCPOLEN_WSCALE_ALIGNED 4
199 #define TCPOLEN_SACKPERM_ALIGNED 4
200 #define TCPOLEN_SACK_BASE 2
201 #define TCPOLEN_SACK_BASE_ALIGNED 4
202 #define TCPOLEN_SACK_PERBLOCK 8
203 #define TCPOLEN_MD5SIG_ALIGNED 20
204 #define TCPOLEN_MSS_ALIGNED 4
206 /* Flags in tp->nonagle */
207 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
208 #define TCP_NAGLE_CORK 2 /* Socket is corked */
209 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
211 /* TCP thin-stream limits */
212 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
214 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
215 #define TCP_INIT_CWND 10
217 /* Bit Flags for sysctl_tcp_fastopen */
218 #define TFO_CLIENT_ENABLE 1
219 #define TFO_SERVER_ENABLE 2
220 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
222 /* Process SYN data but skip cookie validation */
223 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100
224 /* Accept SYN data w/o any cookie option */
225 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
227 /* Force enable TFO on all listeners, i.e., not requiring the
228 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
230 #define TFO_SERVER_WO_SOCKOPT1 0x400
231 #define TFO_SERVER_WO_SOCKOPT2 0x800
232 /* Always create TFO child sockets on a TFO listener even when
233 * cookie/data not present. (For testing purpose!)
235 #define TFO_SERVER_ALWAYS 0x1000
237 extern struct inet_timewait_death_row tcp_death_row;
239 /* sysctl variables for tcp */
240 extern int sysctl_tcp_timestamps;
241 extern int sysctl_tcp_window_scaling;
242 extern int sysctl_tcp_sack;
243 extern int sysctl_tcp_fin_timeout;
244 extern int sysctl_tcp_keepalive_time;
245 extern int sysctl_tcp_keepalive_probes;
246 extern int sysctl_tcp_keepalive_intvl;
247 extern int sysctl_tcp_syn_retries;
248 extern int sysctl_tcp_synack_retries;
249 extern int sysctl_tcp_retries1;
250 extern int sysctl_tcp_retries2;
251 extern int sysctl_tcp_orphan_retries;
252 extern int sysctl_tcp_syncookies;
253 extern int sysctl_tcp_fastopen;
254 extern int sysctl_tcp_retrans_collapse;
255 extern int sysctl_tcp_stdurg;
256 extern int sysctl_tcp_rfc1337;
257 extern int sysctl_tcp_abort_on_overflow;
258 extern int sysctl_tcp_max_orphans;
259 extern int sysctl_tcp_fack;
260 extern int sysctl_tcp_reordering;
261 extern int sysctl_tcp_dsack;
262 extern int sysctl_tcp_wmem[3];
263 extern int sysctl_tcp_rmem[3];
264 extern int sysctl_tcp_app_win;
265 extern int sysctl_tcp_adv_win_scale;
266 extern int sysctl_tcp_tw_reuse;
267 extern int sysctl_tcp_frto;
268 extern int sysctl_tcp_low_latency;
269 extern int sysctl_tcp_dma_copybreak;
270 extern int sysctl_tcp_nometrics_save;
271 extern int sysctl_tcp_moderate_rcvbuf;
272 extern int sysctl_tcp_tso_win_divisor;
273 extern int sysctl_tcp_mtu_probing;
274 extern int sysctl_tcp_base_mss;
275 extern int sysctl_tcp_workaround_signed_windows;
276 extern int sysctl_tcp_slow_start_after_idle;
277 extern int sysctl_tcp_max_ssthresh;
278 extern int sysctl_tcp_thin_linear_timeouts;
279 extern int sysctl_tcp_thin_dupack;
280 extern int sysctl_tcp_early_retrans;
281 extern int sysctl_tcp_limit_output_bytes;
282 extern int sysctl_tcp_challenge_ack_limit;
283 extern unsigned int sysctl_tcp_notsent_lowat;
284 extern int sysctl_tcp_min_tso_segs;
286 extern atomic_long_t tcp_memory_allocated;
287 extern struct percpu_counter tcp_sockets_allocated;
288 extern int tcp_memory_pressure;
291 * The next routines deal with comparing 32 bit unsigned ints
292 * and worry about wraparound (automatic with unsigned arithmetic).
295 static inline bool before(__u32 seq1, __u32 seq2)
297 return (__s32)(seq1-seq2) < 0;
299 #define after(seq2, seq1) before(seq1, seq2)
301 /* is s2<=s1<=s3 ? */
302 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
304 return seq3 - seq2 >= seq1 - seq2;
307 static inline bool tcp_out_of_memory(struct sock *sk)
309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
315 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
317 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
318 int orphans = percpu_counter_read_positive(ocp);
320 if (orphans << shift > sysctl_tcp_max_orphans) {
321 orphans = percpu_counter_sum_positive(ocp);
322 if (orphans << shift > sysctl_tcp_max_orphans)
328 bool tcp_check_oom(struct sock *sk, int shift);
330 /* syncookies: remember time of last synqueue overflow */
331 static inline void tcp_synq_overflow(struct sock *sk)
333 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
336 /* syncookies: no recent synqueue overflow on this listening socket? */
337 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
339 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
340 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
343 extern struct proto tcp_prot;
345 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
346 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
347 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
348 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
349 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
351 void tcp_init_mem(struct net *net);
353 void tcp_tasklet_init(void);
355 void tcp_v4_err(struct sk_buff *skb, u32);
357 void tcp_shutdown(struct sock *sk, int how);
359 void tcp_v4_early_demux(struct sk_buff *skb);
360 int tcp_v4_rcv(struct sk_buff *skb);
362 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
363 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
365 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
367 void tcp_release_cb(struct sock *sk);
368 void tcp_wfree(struct sk_buff *skb);
369 void tcp_write_timer_handler(struct sock *sk);
370 void tcp_delack_timer_handler(struct sock *sk);
371 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
372 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
373 const struct tcphdr *th, unsigned int len);
374 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
375 const struct tcphdr *th, unsigned int len);
376 void tcp_rcv_space_adjust(struct sock *sk);
377 void tcp_cleanup_rbuf(struct sock *sk, int copied);
378 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
379 void tcp_twsk_destructor(struct sock *sk);
380 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
381 struct pipe_inode_info *pipe, size_t len,
384 static inline void tcp_dec_quickack_mode(struct sock *sk,
385 const unsigned int pkts)
387 struct inet_connection_sock *icsk = inet_csk(sk);
389 if (icsk->icsk_ack.quick) {
390 if (pkts >= icsk->icsk_ack.quick) {
391 icsk->icsk_ack.quick = 0;
392 /* Leaving quickack mode we deflate ATO. */
393 icsk->icsk_ack.ato = TCP_ATO_MIN;
395 icsk->icsk_ack.quick -= pkts;
400 #define TCP_ECN_QUEUE_CWR 2
401 #define TCP_ECN_DEMAND_CWR 4
402 #define TCP_ECN_SEEN 8
412 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
414 const struct tcphdr *th);
415 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
416 struct request_sock *req, struct request_sock **prev,
418 int tcp_child_process(struct sock *parent, struct sock *child,
419 struct sk_buff *skb);
420 void tcp_enter_loss(struct sock *sk, int how);
421 void tcp_clear_retrans(struct tcp_sock *tp);
422 void tcp_update_metrics(struct sock *sk);
423 void tcp_init_metrics(struct sock *sk);
424 void tcp_metrics_init(void);
425 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
427 bool tcp_remember_stamp(struct sock *sk);
428 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
429 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
430 void tcp_disable_fack(struct tcp_sock *tp);
431 void tcp_close(struct sock *sk, long timeout);
432 void tcp_init_sock(struct sock *sk);
433 unsigned int tcp_poll(struct file *file, struct socket *sock,
434 struct poll_table_struct *wait);
435 int tcp_getsockopt(struct sock *sk, int level, int optname,
436 char __user *optval, int __user *optlen);
437 int tcp_setsockopt(struct sock *sk, int level, int optname,
438 char __user *optval, unsigned int optlen);
439 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
440 char __user *optval, int __user *optlen);
441 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
442 char __user *optval, unsigned int optlen);
443 void tcp_set_keepalive(struct sock *sk, int val);
444 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
445 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
446 size_t len, int nonblock, int flags, int *addr_len);
447 void tcp_parse_options(const struct sk_buff *skb,
448 struct tcp_options_received *opt_rx,
449 int estab, struct tcp_fastopen_cookie *foc);
450 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
453 * TCP v4 functions exported for the inet6 API
456 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
457 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
458 struct sock *tcp_create_openreq_child(struct sock *sk,
459 struct request_sock *req,
460 struct sk_buff *skb);
461 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
462 struct request_sock *req,
463 struct dst_entry *dst);
464 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
465 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
466 int tcp_connect(struct sock *sk);
467 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
468 struct request_sock *req,
469 struct tcp_fastopen_cookie *foc);
470 int tcp_disconnect(struct sock *sk, int flags);
472 void tcp_connect_init(struct sock *sk);
473 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
474 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
475 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
477 /* From syncookies.c */
478 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
480 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
481 struct ip_options *opt);
482 #ifdef CONFIG_SYN_COOKIES
483 #include <linux/ktime.h>
485 /* Syncookies use a monotonic timer which increments every 64 seconds.
486 * This counter is used both as a hash input and partially encoded into
487 * the cookie value. A cookie is only validated further if the delta
488 * between the current counter value and the encoded one is less than this,
489 * i.e. a sent cookie is valid only at most for 128 seconds (or less if
490 * the counter advances immediately after a cookie is generated).
492 #define MAX_SYNCOOKIE_AGE 2
494 static inline u32 tcp_cookie_time(void)
497 getnstimeofday(&now);
498 return now.tv_sec >> 6; /* 64 seconds granularity */
501 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
503 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mss);
505 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
513 __u32 cookie_init_timestamp(struct request_sock *req);
514 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net,
517 /* From net/ipv6/syncookies.c */
518 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
520 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
521 #ifdef CONFIG_SYN_COOKIES
522 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
523 const struct tcphdr *th, u16 *mssp);
524 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
527 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
536 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
538 bool tcp_may_send_now(struct sock *sk);
539 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
540 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
541 void tcp_retransmit_timer(struct sock *sk);
542 void tcp_xmit_retransmit_queue(struct sock *);
543 void tcp_simple_retransmit(struct sock *);
544 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
545 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
547 void tcp_send_probe0(struct sock *);
548 void tcp_send_partial(struct sock *);
549 int tcp_write_wakeup(struct sock *);
550 void tcp_send_fin(struct sock *sk);
551 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
552 int tcp_send_synack(struct sock *);
553 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
555 void tcp_push_one(struct sock *, unsigned int mss_now);
556 void tcp_send_ack(struct sock *sk);
557 void tcp_send_delayed_ack(struct sock *sk);
558 void tcp_send_loss_probe(struct sock *sk);
559 bool tcp_schedule_loss_probe(struct sock *sk);
562 void tcp_cwnd_application_limited(struct sock *sk);
563 void tcp_resume_early_retransmit(struct sock *sk);
564 void tcp_rearm_rto(struct sock *sk);
565 void tcp_reset(struct sock *sk);
568 void tcp_init_xmit_timers(struct sock *);
569 static inline void tcp_clear_xmit_timers(struct sock *sk)
571 inet_csk_clear_xmit_timers(sk);
574 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
575 unsigned int tcp_current_mss(struct sock *sk);
577 /* Bound MSS / TSO packet size with the half of the window */
578 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
582 /* When peer uses tiny windows, there is no use in packetizing
583 * to sub-MSS pieces for the sake of SWS or making sure there
584 * are enough packets in the pipe for fast recovery.
586 * On the other hand, for extremely large MSS devices, handling
587 * smaller than MSS windows in this way does make sense.
589 if (tp->max_window >= 512)
590 cutoff = (tp->max_window >> 1);
592 cutoff = tp->max_window;
594 if (cutoff && pktsize > cutoff)
595 return max_t(int, cutoff, 68U - tp->tcp_header_len);
601 void tcp_get_info(const struct sock *, struct tcp_info *);
603 /* Read 'sendfile()'-style from a TCP socket */
604 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
605 unsigned int, size_t);
606 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
607 sk_read_actor_t recv_actor);
609 void tcp_initialize_rcv_mss(struct sock *sk);
611 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
612 int tcp_mss_to_mtu(struct sock *sk, int mss);
613 void tcp_mtup_init(struct sock *sk);
614 void tcp_init_buffer_space(struct sock *sk);
616 static inline void tcp_bound_rto(const struct sock *sk)
618 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
619 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
622 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
624 return (tp->srtt >> 3) + tp->rttvar;
627 void tcp_set_rto(struct sock *sk);
629 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
631 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
632 ntohl(TCP_FLAG_ACK) |
636 static inline void tcp_fast_path_on(struct tcp_sock *tp)
638 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
641 static inline void tcp_fast_path_check(struct sock *sk)
643 struct tcp_sock *tp = tcp_sk(sk);
645 if (skb_queue_empty(&tp->out_of_order_queue) &&
647 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
649 tcp_fast_path_on(tp);
652 /* Compute the actual rto_min value */
653 static inline u32 tcp_rto_min(struct sock *sk)
655 const struct dst_entry *dst = __sk_dst_get(sk);
656 u32 rto_min = TCP_RTO_MIN;
658 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
659 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
663 /* Compute the actual receive window we are currently advertising.
664 * Rcv_nxt can be after the window if our peer push more data
665 * than the offered window.
667 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
669 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
676 /* Choose a new window, without checks for shrinking, and without
677 * scaling applied to the result. The caller does these things
678 * if necessary. This is a "raw" window selection.
680 u32 __tcp_select_window(struct sock *sk);
682 void tcp_send_window_probe(struct sock *sk);
684 /* TCP timestamps are only 32-bits, this causes a slight
685 * complication on 64-bit systems since we store a snapshot
686 * of jiffies in the buffer control blocks below. We decided
687 * to use only the low 32-bits of jiffies and hide the ugly
688 * casts with the following macro.
690 #define tcp_time_stamp ((__u32)(jiffies))
692 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
694 #define TCPHDR_FIN 0x01
695 #define TCPHDR_SYN 0x02
696 #define TCPHDR_RST 0x04
697 #define TCPHDR_PSH 0x08
698 #define TCPHDR_ACK 0x10
699 #define TCPHDR_URG 0x20
700 #define TCPHDR_ECE 0x40
701 #define TCPHDR_CWR 0x80
703 /* This is what the send packet queuing engine uses to pass
704 * TCP per-packet control information to the transmission code.
705 * We also store the host-order sequence numbers in here too.
706 * This is 44 bytes if IPV6 is enabled.
707 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
711 struct inet_skb_parm h4;
712 #if IS_ENABLED(CONFIG_IPV6)
713 struct inet6_skb_parm h6;
715 } header; /* For incoming frames */
716 __u32 seq; /* Starting sequence number */
717 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
718 __u32 when; /* used to compute rtt's */
719 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
721 __u8 sacked; /* State flags for SACK/FACK. */
722 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
723 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
724 #define TCPCB_LOST 0x04 /* SKB is lost */
725 #define TCPCB_TAGBITS 0x07 /* All tag bits */
726 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
727 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
729 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
731 __u32 ack_seq; /* Sequence number ACK'd */
734 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
736 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
738 * If we receive a SYN packet with these bits set, it means a network is
739 * playing bad games with TOS bits. In order to avoid possible false congestion
740 * notifications, we disable TCP ECN negociation.
743 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
746 const struct tcphdr *th = tcp_hdr(skb);
748 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr &&
749 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
750 inet_rsk(req)->ecn_ok = 1;
753 /* Due to TSO, an SKB can be composed of multiple actual
754 * packets. To keep these tracked properly, we use this.
756 static inline int tcp_skb_pcount(const struct sk_buff *skb)
758 return skb_shinfo(skb)->gso_segs;
761 /* This is valid iff tcp_skb_pcount() > 1. */
762 static inline int tcp_skb_mss(const struct sk_buff *skb)
764 return skb_shinfo(skb)->gso_size;
767 /* Events passed to congestion control interface */
769 CA_EVENT_TX_START, /* first transmit when no packets in flight */
770 CA_EVENT_CWND_RESTART, /* congestion window restart */
771 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
772 CA_EVENT_LOSS, /* loss timeout */
773 CA_EVENT_FAST_ACK, /* in sequence ack */
774 CA_EVENT_SLOW_ACK, /* other ack */
778 * Interface for adding new TCP congestion control handlers
780 #define TCP_CA_NAME_MAX 16
781 #define TCP_CA_MAX 128
782 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
784 #define TCP_CONG_NON_RESTRICTED 0x1
785 #define TCP_CONG_RTT_STAMP 0x2
787 struct tcp_congestion_ops {
788 struct list_head list;
791 /* initialize private data (optional) */
792 void (*init)(struct sock *sk);
793 /* cleanup private data (optional) */
794 void (*release)(struct sock *sk);
796 /* return slow start threshold (required) */
797 u32 (*ssthresh)(struct sock *sk);
798 /* lower bound for congestion window (optional) */
799 u32 (*min_cwnd)(const struct sock *sk);
800 /* do new cwnd calculation (required) */
801 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
802 /* call before changing ca_state (optional) */
803 void (*set_state)(struct sock *sk, u8 new_state);
804 /* call when cwnd event occurs (optional) */
805 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
806 /* new value of cwnd after loss (optional) */
807 u32 (*undo_cwnd)(struct sock *sk);
808 /* hook for packet ack accounting (optional) */
809 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
810 /* get info for inet_diag (optional) */
811 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
813 char name[TCP_CA_NAME_MAX];
814 struct module *owner;
817 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
818 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
820 void tcp_init_congestion_control(struct sock *sk);
821 void tcp_cleanup_congestion_control(struct sock *sk);
822 int tcp_set_default_congestion_control(const char *name);
823 void tcp_get_default_congestion_control(char *name);
824 void tcp_get_available_congestion_control(char *buf, size_t len);
825 void tcp_get_allowed_congestion_control(char *buf, size_t len);
826 int tcp_set_allowed_congestion_control(char *allowed);
827 int tcp_set_congestion_control(struct sock *sk, const char *name);
828 void tcp_slow_start(struct tcp_sock *tp);
829 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
831 extern struct tcp_congestion_ops tcp_init_congestion_ops;
832 u32 tcp_reno_ssthresh(struct sock *sk);
833 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
834 u32 tcp_reno_min_cwnd(const struct sock *sk);
835 extern struct tcp_congestion_ops tcp_reno;
837 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
839 struct inet_connection_sock *icsk = inet_csk(sk);
841 if (icsk->icsk_ca_ops->set_state)
842 icsk->icsk_ca_ops->set_state(sk, ca_state);
843 icsk->icsk_ca_state = ca_state;
846 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
848 const struct inet_connection_sock *icsk = inet_csk(sk);
850 if (icsk->icsk_ca_ops->cwnd_event)
851 icsk->icsk_ca_ops->cwnd_event(sk, event);
854 /* These functions determine how the current flow behaves in respect of SACK
855 * handling. SACK is negotiated with the peer, and therefore it can vary
856 * between different flows.
858 * tcp_is_sack - SACK enabled
859 * tcp_is_reno - No SACK
860 * tcp_is_fack - FACK enabled, implies SACK enabled
862 static inline int tcp_is_sack(const struct tcp_sock *tp)
864 return tp->rx_opt.sack_ok;
867 static inline bool tcp_is_reno(const struct tcp_sock *tp)
869 return !tcp_is_sack(tp);
872 static inline bool tcp_is_fack(const struct tcp_sock *tp)
874 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
877 static inline void tcp_enable_fack(struct tcp_sock *tp)
879 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
882 /* TCP early-retransmit (ER) is similar to but more conservative than
883 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
885 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
887 tp->do_early_retrans = sysctl_tcp_early_retrans &&
888 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
889 sysctl_tcp_reordering == 3;
892 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
894 tp->do_early_retrans = 0;
897 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
899 return tp->sacked_out + tp->lost_out;
902 /* This determines how many packets are "in the network" to the best
903 * of our knowledge. In many cases it is conservative, but where
904 * detailed information is available from the receiver (via SACK
905 * blocks etc.) we can make more aggressive calculations.
907 * Use this for decisions involving congestion control, use just
908 * tp->packets_out to determine if the send queue is empty or not.
910 * Read this equation as:
912 * "Packets sent once on transmission queue" MINUS
913 * "Packets left network, but not honestly ACKed yet" PLUS
914 * "Packets fast retransmitted"
916 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
918 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
921 #define TCP_INFINITE_SSTHRESH 0x7fffffff
923 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
925 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
928 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
930 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
931 (1 << inet_csk(sk)->icsk_ca_state);
934 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
935 * The exception is cwnd reduction phase, when cwnd is decreasing towards
938 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
940 const struct tcp_sock *tp = tcp_sk(sk);
942 if (tcp_in_cwnd_reduction(sk))
943 return tp->snd_ssthresh;
945 return max(tp->snd_ssthresh,
946 ((tp->snd_cwnd >> 1) +
947 (tp->snd_cwnd >> 2)));
950 /* Use define here intentionally to get WARN_ON location shown at the caller */
951 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
953 void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
954 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
956 /* The maximum number of MSS of available cwnd for which TSO defers
957 * sending if not using sysctl_tcp_tso_win_divisor.
959 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
964 /* Slow start with delack produces 3 packets of burst, so that
965 * it is safe "de facto". This will be the default - same as
966 * the default reordering threshold - but if reordering increases,
967 * we must be able to allow cwnd to burst at least this much in order
968 * to not pull it back when holes are filled.
970 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
972 return tp->reordering;
975 /* Returns end sequence number of the receiver's advertised window */
976 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
978 return tp->snd_una + tp->snd_wnd;
980 bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
982 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
983 const struct sk_buff *skb)
986 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
989 static inline void tcp_check_probe_timer(struct sock *sk)
991 const struct tcp_sock *tp = tcp_sk(sk);
992 const struct inet_connection_sock *icsk = inet_csk(sk);
994 if (!tp->packets_out && !icsk->icsk_pending)
995 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
996 icsk->icsk_rto, TCP_RTO_MAX);
999 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1004 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1010 * Calculate(/check) TCP checksum
1012 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1013 __be32 daddr, __wsum base)
1015 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1018 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1020 return __skb_checksum_complete(skb);
1023 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1025 return !skb_csum_unnecessary(skb) &&
1026 __tcp_checksum_complete(skb);
1029 /* Prequeue for VJ style copy to user, combined with checksumming. */
1031 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1033 tp->ucopy.task = NULL;
1035 tp->ucopy.memory = 0;
1036 skb_queue_head_init(&tp->ucopy.prequeue);
1037 #ifdef CONFIG_NET_DMA
1038 tp->ucopy.dma_chan = NULL;
1039 tp->ucopy.wakeup = 0;
1040 tp->ucopy.pinned_list = NULL;
1041 tp->ucopy.dma_cookie = 0;
1045 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1050 static const char *statename[]={
1051 "Unused","Established","Syn Sent","Syn Recv",
1052 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1053 "Close Wait","Last ACK","Listen","Closing"
1056 void tcp_set_state(struct sock *sk, int state);
1058 void tcp_done(struct sock *sk);
1060 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1063 rx_opt->num_sacks = 0;
1066 u32 tcp_default_init_rwnd(u32 mss);
1068 /* Determine a window scaling and initial window to offer. */
1069 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1070 __u32 *window_clamp, int wscale_ok,
1071 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1073 static inline int tcp_win_from_space(int space)
1075 return sysctl_tcp_adv_win_scale<=0 ?
1076 (space>>(-sysctl_tcp_adv_win_scale)) :
1077 space - (space>>sysctl_tcp_adv_win_scale);
1080 /* Note: caller must be prepared to deal with negative returns */
1081 static inline int tcp_space(const struct sock *sk)
1083 return tcp_win_from_space(sk->sk_rcvbuf -
1084 atomic_read(&sk->sk_rmem_alloc));
1087 static inline int tcp_full_space(const struct sock *sk)
1089 return tcp_win_from_space(sk->sk_rcvbuf);
1092 static inline void tcp_openreq_init(struct request_sock *req,
1093 struct tcp_options_received *rx_opt,
1094 struct sk_buff *skb)
1096 struct inet_request_sock *ireq = inet_rsk(req);
1098 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1100 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1101 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1102 tcp_rsk(req)->snt_synack = 0;
1103 req->mss = rx_opt->mss_clamp;
1104 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1105 ireq->tstamp_ok = rx_opt->tstamp_ok;
1106 ireq->sack_ok = rx_opt->sack_ok;
1107 ireq->snd_wscale = rx_opt->snd_wscale;
1108 ireq->wscale_ok = rx_opt->wscale_ok;
1111 ireq->ir_rmt_port = tcp_hdr(skb)->source;
1112 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
1115 void tcp_enter_memory_pressure(struct sock *sk);
1117 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1119 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1122 static inline int keepalive_time_when(const struct tcp_sock *tp)
1124 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1127 static inline int keepalive_probes(const struct tcp_sock *tp)
1129 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1132 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1134 const struct inet_connection_sock *icsk = &tp->inet_conn;
1136 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1137 tcp_time_stamp - tp->rcv_tstamp);
1140 static inline int tcp_fin_time(const struct sock *sk)
1142 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1143 const int rto = inet_csk(sk)->icsk_rto;
1145 if (fin_timeout < (rto << 2) - (rto >> 1))
1146 fin_timeout = (rto << 2) - (rto >> 1);
1151 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1154 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1156 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1159 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1160 * then following tcp messages have valid values. Ignore 0 value,
1161 * or else 'negative' tsval might forbid us to accept their packets.
1163 if (!rx_opt->ts_recent)
1168 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1171 if (tcp_paws_check(rx_opt, 0))
1174 /* RST segments are not recommended to carry timestamp,
1175 and, if they do, it is recommended to ignore PAWS because
1176 "their cleanup function should take precedence over timestamps."
1177 Certainly, it is mistake. It is necessary to understand the reasons
1178 of this constraint to relax it: if peer reboots, clock may go
1179 out-of-sync and half-open connections will not be reset.
1180 Actually, the problem would be not existing if all
1181 the implementations followed draft about maintaining clock
1182 via reboots. Linux-2.2 DOES NOT!
1184 However, we can relax time bounds for RST segments to MSL.
1186 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1191 static inline void tcp_mib_init(struct net *net)
1194 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1195 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1196 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1197 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1201 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1203 tp->lost_skb_hint = NULL;
1206 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1208 tcp_clear_retrans_hints_partial(tp);
1209 tp->retransmit_skb_hint = NULL;
1215 union tcp_md5_addr {
1217 #if IS_ENABLED(CONFIG_IPV6)
1222 /* - key database */
1223 struct tcp_md5sig_key {
1224 struct hlist_node node;
1226 u8 family; /* AF_INET or AF_INET6 */
1227 union tcp_md5_addr addr;
1228 u8 key[TCP_MD5SIG_MAXKEYLEN];
1229 struct rcu_head rcu;
1233 struct tcp_md5sig_info {
1234 struct hlist_head head;
1235 struct rcu_head rcu;
1238 /* - pseudo header */
1239 struct tcp4_pseudohdr {
1247 struct tcp6_pseudohdr {
1248 struct in6_addr saddr;
1249 struct in6_addr daddr;
1251 __be32 protocol; /* including padding */
1254 union tcp_md5sum_block {
1255 struct tcp4_pseudohdr ip4;
1256 #if IS_ENABLED(CONFIG_IPV6)
1257 struct tcp6_pseudohdr ip6;
1261 /* - pool: digest algorithm, hash description and scratch buffer */
1262 struct tcp_md5sig_pool {
1263 struct hash_desc md5_desc;
1264 union tcp_md5sum_block md5_blk;
1268 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1269 const struct sock *sk, const struct request_sock *req,
1270 const struct sk_buff *skb);
1271 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1272 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1273 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1275 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1276 struct sock *addr_sk);
1278 #ifdef CONFIG_TCP_MD5SIG
1279 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1280 const union tcp_md5_addr *addr,
1282 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1284 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1285 const union tcp_md5_addr *addr,
1290 #define tcp_twsk_md5_key(twsk) NULL
1293 bool tcp_alloc_md5sig_pool(void);
1295 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1296 static inline void tcp_put_md5sig_pool(void)
1301 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1302 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1303 unsigned int header_len);
1304 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1305 const struct tcp_md5sig_key *key);
1307 /* From tcp_fastopen.c */
1308 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1309 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1310 unsigned long *last_syn_loss);
1311 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1312 struct tcp_fastopen_cookie *cookie, bool syn_lost);
1313 struct tcp_fastopen_request {
1314 /* Fast Open cookie. Size 0 means a cookie request */
1315 struct tcp_fastopen_cookie cookie;
1316 struct msghdr *data; /* data in MSG_FASTOPEN */
1317 u16 copied; /* queued in tcp_connect() */
1319 void tcp_free_fastopen_req(struct tcp_sock *tp);
1321 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1322 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1323 void tcp_fastopen_cookie_gen(__be32 src, __be32 dst,
1324 struct tcp_fastopen_cookie *foc);
1325 void tcp_fastopen_init_key_once(bool publish);
1326 #define TCP_FASTOPEN_KEY_LENGTH 16
1328 /* Fastopen key context */
1329 struct tcp_fastopen_context {
1330 struct crypto_cipher *tfm;
1331 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1332 struct rcu_head rcu;
1335 /* write queue abstraction */
1336 static inline void tcp_write_queue_purge(struct sock *sk)
1338 struct sk_buff *skb;
1340 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1341 sk_wmem_free_skb(sk, skb);
1343 tcp_clear_all_retrans_hints(tcp_sk(sk));
1346 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1348 return skb_peek(&sk->sk_write_queue);
1351 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1353 return skb_peek_tail(&sk->sk_write_queue);
1356 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1357 const struct sk_buff *skb)
1359 return skb_queue_next(&sk->sk_write_queue, skb);
1362 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1363 const struct sk_buff *skb)
1365 return skb_queue_prev(&sk->sk_write_queue, skb);
1368 #define tcp_for_write_queue(skb, sk) \
1369 skb_queue_walk(&(sk)->sk_write_queue, skb)
1371 #define tcp_for_write_queue_from(skb, sk) \
1372 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1374 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1375 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1377 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1379 return sk->sk_send_head;
1382 static inline bool tcp_skb_is_last(const struct sock *sk,
1383 const struct sk_buff *skb)
1385 return skb_queue_is_last(&sk->sk_write_queue, skb);
1388 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1390 if (tcp_skb_is_last(sk, skb))
1391 sk->sk_send_head = NULL;
1393 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1396 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1398 if (sk->sk_send_head == skb_unlinked)
1399 sk->sk_send_head = NULL;
1402 static inline void tcp_init_send_head(struct sock *sk)
1404 sk->sk_send_head = NULL;
1407 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1409 __skb_queue_tail(&sk->sk_write_queue, skb);
1412 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1414 __tcp_add_write_queue_tail(sk, skb);
1416 /* Queue it, remembering where we must start sending. */
1417 if (sk->sk_send_head == NULL) {
1418 sk->sk_send_head = skb;
1420 if (tcp_sk(sk)->highest_sack == NULL)
1421 tcp_sk(sk)->highest_sack = skb;
1425 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1427 __skb_queue_head(&sk->sk_write_queue, skb);
1430 /* Insert buff after skb on the write queue of sk. */
1431 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1432 struct sk_buff *buff,
1435 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1438 /* Insert new before skb on the write queue of sk. */
1439 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1440 struct sk_buff *skb,
1443 __skb_queue_before(&sk->sk_write_queue, skb, new);
1445 if (sk->sk_send_head == skb)
1446 sk->sk_send_head = new;
1449 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1451 __skb_unlink(skb, &sk->sk_write_queue);
1454 static inline bool tcp_write_queue_empty(struct sock *sk)
1456 return skb_queue_empty(&sk->sk_write_queue);
1459 static inline void tcp_push_pending_frames(struct sock *sk)
1461 if (tcp_send_head(sk)) {
1462 struct tcp_sock *tp = tcp_sk(sk);
1464 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1468 /* Start sequence of the skb just after the highest skb with SACKed
1469 * bit, valid only if sacked_out > 0 or when the caller has ensured
1470 * validity by itself.
1472 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1474 if (!tp->sacked_out)
1477 if (tp->highest_sack == NULL)
1480 return TCP_SKB_CB(tp->highest_sack)->seq;
1483 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1485 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1486 tcp_write_queue_next(sk, skb);
1489 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1491 return tcp_sk(sk)->highest_sack;
1494 static inline void tcp_highest_sack_reset(struct sock *sk)
1496 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1499 /* Called when old skb is about to be deleted (to be combined with new skb) */
1500 static inline void tcp_highest_sack_combine(struct sock *sk,
1501 struct sk_buff *old,
1502 struct sk_buff *new)
1504 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1505 tcp_sk(sk)->highest_sack = new;
1508 /* Determines whether this is a thin stream (which may suffer from
1509 * increased latency). Used to trigger latency-reducing mechanisms.
1511 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1513 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1517 enum tcp_seq_states {
1518 TCP_SEQ_STATE_LISTENING,
1519 TCP_SEQ_STATE_OPENREQ,
1520 TCP_SEQ_STATE_ESTABLISHED,
1523 int tcp_seq_open(struct inode *inode, struct file *file);
1525 struct tcp_seq_afinfo {
1528 const struct file_operations *seq_fops;
1529 struct seq_operations seq_ops;
1532 struct tcp_iter_state {
1533 struct seq_net_private p;
1535 enum tcp_seq_states state;
1536 struct sock *syn_wait_sk;
1537 int bucket, offset, sbucket, num;
1542 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1543 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1545 extern struct request_sock_ops tcp_request_sock_ops;
1546 extern struct request_sock_ops tcp6_request_sock_ops;
1548 void tcp_v4_destroy_sock(struct sock *sk);
1550 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1551 netdev_features_t features);
1552 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1553 int tcp_gro_complete(struct sk_buff *skb);
1555 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1557 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1559 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1562 static inline bool tcp_stream_memory_free(const struct sock *sk)
1564 const struct tcp_sock *tp = tcp_sk(sk);
1565 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1567 return notsent_bytes < tcp_notsent_lowat(tp);
1570 #ifdef CONFIG_PROC_FS
1571 int tcp4_proc_init(void);
1572 void tcp4_proc_exit(void);
1575 /* TCP af-specific functions */
1576 struct tcp_sock_af_ops {
1577 #ifdef CONFIG_TCP_MD5SIG
1578 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1579 struct sock *addr_sk);
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);
1585 int (*md5_parse) (struct sock *sk,
1586 char __user *optval,
1591 struct tcp_request_sock_ops {
1592 #ifdef CONFIG_TCP_MD5SIG
1593 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1594 struct request_sock *req);
1595 int (*calc_md5_hash) (char *location,
1596 struct tcp_md5sig_key *md5,
1597 const struct sock *sk,
1598 const struct request_sock *req,
1599 const struct sk_buff *skb);
1603 int tcpv4_offload_init(void);
1605 void tcp_v4_init(void);
1606 void tcp_init(void);