2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/crypto.h>
31 #include <linux/cryptohash.h>
32 #include <linux/kref.h>
33 #include <linux/ktime.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo;
52 extern struct percpu_counter tcp_orphan_count;
53 void tcp_time_wait(struct sock *sk, int state, int timeo);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 512
70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
71 #define TCP_FASTRETRANS_THRESH 3
73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
74 #define TCP_MAX_QUICKACKS 16U
77 #define TCP_URG_VALID 0x0100
78 #define TCP_URG_NOTYET 0x0200
79 #define TCP_URG_READ 0x0400
81 #define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
88 #define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
95 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
96 * when active opening a connection.
97 * RFC1122 says the minimum retry MUST
98 * be at least 180secs. Nevertheless
99 * this value is corresponding to
100 * 63secs of retransmission with the
101 * current initial RTO.
104 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
105 * when passive opening a connection.
106 * This is corresponding to 31secs of
107 * retransmission with the current
111 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
112 * state, about 60 seconds */
113 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
114 /* BSD style FIN_WAIT2 deadlock breaker.
115 * It used to be 3min, new value is 60sec,
116 * to combine FIN-WAIT-2 timeout with
120 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
122 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
123 #define TCP_ATO_MIN ((unsigned)(HZ/25))
125 #define TCP_DELACK_MIN 4U
126 #define TCP_ATO_MIN 4U
128 #define TCP_RTO_MAX ((unsigned)(120*HZ))
129 #define TCP_RTO_MIN ((unsigned)(HZ/5))
130 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
131 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
132 * used as a fallback RTO for the
133 * initial data transmission if no
134 * valid RTT sample has been acquired,
135 * most likely due to retrans in 3WHS.
138 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
139 * for local resources.
142 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
143 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
144 #define TCP_KEEPALIVE_INTVL (75*HZ)
146 #define MAX_TCP_KEEPIDLE 32767
147 #define MAX_TCP_KEEPINTVL 32767
148 #define MAX_TCP_KEEPCNT 127
149 #define MAX_TCP_SYNCNT 127
151 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
153 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
154 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
155 * after this time. It should be equal
156 * (or greater than) TCP_TIMEWAIT_LEN
157 * to provide reliability equal to one
158 * provided by timewait state.
160 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
161 * timestamps. It must be less than
162 * minimal timewait lifetime.
168 #define TCPOPT_NOP 1 /* Padding */
169 #define TCPOPT_EOL 0 /* End of options */
170 #define TCPOPT_MSS 2 /* Segment size negotiating */
171 #define TCPOPT_WINDOW 3 /* Window scaling */
172 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
173 #define TCPOPT_SACK 5 /* SACK Block */
174 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
175 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
176 #define TCPOPT_EXP 254 /* Experimental */
177 /* Magic number to be after the option value for sharing TCP
178 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
180 #define TCPOPT_FASTOPEN_MAGIC 0xF989
186 #define TCPOLEN_MSS 4
187 #define TCPOLEN_WINDOW 3
188 #define TCPOLEN_SACK_PERM 2
189 #define TCPOLEN_TIMESTAMP 10
190 #define TCPOLEN_MD5SIG 18
191 #define TCPOLEN_EXP_FASTOPEN_BASE 4
193 /* But this is what stacks really send out. */
194 #define TCPOLEN_TSTAMP_ALIGNED 12
195 #define TCPOLEN_WSCALE_ALIGNED 4
196 #define TCPOLEN_SACKPERM_ALIGNED 4
197 #define TCPOLEN_SACK_BASE 2
198 #define TCPOLEN_SACK_BASE_ALIGNED 4
199 #define TCPOLEN_SACK_PERBLOCK 8
200 #define TCPOLEN_MD5SIG_ALIGNED 20
201 #define TCPOLEN_MSS_ALIGNED 4
203 /* Flags in tp->nonagle */
204 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
205 #define TCP_NAGLE_CORK 2 /* Socket is corked */
206 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
208 /* TCP thin-stream limits */
209 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
211 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
212 #define TCP_INIT_CWND 10
214 /* Bit Flags for sysctl_tcp_fastopen */
215 #define TFO_CLIENT_ENABLE 1
216 #define TFO_SERVER_ENABLE 2
217 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
219 /* Accept SYN data w/o any cookie option */
220 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
222 /* Force enable TFO on all listeners, i.e., not requiring the
223 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
225 #define TFO_SERVER_WO_SOCKOPT1 0x400
226 #define TFO_SERVER_WO_SOCKOPT2 0x800
228 extern struct inet_timewait_death_row tcp_death_row;
230 /* sysctl variables for tcp */
231 extern int sysctl_tcp_timestamps;
232 extern int sysctl_tcp_window_scaling;
233 extern int sysctl_tcp_sack;
234 extern int sysctl_tcp_fin_timeout;
235 extern int sysctl_tcp_keepalive_time;
236 extern int sysctl_tcp_keepalive_probes;
237 extern int sysctl_tcp_keepalive_intvl;
238 extern int sysctl_tcp_syn_retries;
239 extern int sysctl_tcp_synack_retries;
240 extern int sysctl_tcp_retries1;
241 extern int sysctl_tcp_retries2;
242 extern int sysctl_tcp_orphan_retries;
243 extern int sysctl_tcp_syncookies;
244 extern int sysctl_tcp_fastopen;
245 extern int sysctl_tcp_retrans_collapse;
246 extern int sysctl_tcp_stdurg;
247 extern int sysctl_tcp_rfc1337;
248 extern int sysctl_tcp_abort_on_overflow;
249 extern int sysctl_tcp_max_orphans;
250 extern int sysctl_tcp_fack;
251 extern int sysctl_tcp_reordering;
252 extern int sysctl_tcp_max_reordering;
253 extern int sysctl_tcp_dsack;
254 extern long sysctl_tcp_mem[3];
255 extern int sysctl_tcp_wmem[3];
256 extern int sysctl_tcp_rmem[3];
257 extern int sysctl_tcp_app_win;
258 extern int sysctl_tcp_adv_win_scale;
259 extern int sysctl_tcp_tw_reuse;
260 extern int sysctl_tcp_frto;
261 extern int sysctl_tcp_low_latency;
262 extern int sysctl_tcp_nometrics_save;
263 extern int sysctl_tcp_moderate_rcvbuf;
264 extern int sysctl_tcp_tso_win_divisor;
265 extern int sysctl_tcp_mtu_probing;
266 extern int sysctl_tcp_base_mss;
267 extern int sysctl_tcp_workaround_signed_windows;
268 extern int sysctl_tcp_slow_start_after_idle;
269 extern int sysctl_tcp_thin_linear_timeouts;
270 extern int sysctl_tcp_thin_dupack;
271 extern int sysctl_tcp_early_retrans;
272 extern int sysctl_tcp_limit_output_bytes;
273 extern int sysctl_tcp_challenge_ack_limit;
274 extern unsigned int sysctl_tcp_notsent_lowat;
275 extern int sysctl_tcp_min_tso_segs;
276 extern int sysctl_tcp_autocorking;
278 extern atomic_long_t tcp_memory_allocated;
279 extern struct percpu_counter tcp_sockets_allocated;
280 extern int tcp_memory_pressure;
283 * The next routines deal with comparing 32 bit unsigned ints
284 * and worry about wraparound (automatic with unsigned arithmetic).
287 static inline bool before(__u32 seq1, __u32 seq2)
289 return (__s32)(seq1-seq2) < 0;
291 #define after(seq2, seq1) before(seq1, seq2)
293 /* is s2<=s1<=s3 ? */
294 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
296 return seq3 - seq2 >= seq1 - seq2;
299 static inline bool tcp_out_of_memory(struct sock *sk)
301 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
302 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
307 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
309 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
310 int orphans = percpu_counter_read_positive(ocp);
312 if (orphans << shift > sysctl_tcp_max_orphans) {
313 orphans = percpu_counter_sum_positive(ocp);
314 if (orphans << shift > sysctl_tcp_max_orphans)
320 bool tcp_check_oom(struct sock *sk, int shift);
322 /* syncookies: remember time of last synqueue overflow */
323 static inline void tcp_synq_overflow(struct sock *sk)
325 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
328 /* syncookies: no recent synqueue overflow on this listening socket? */
329 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
331 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
332 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
335 extern struct proto tcp_prot;
337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
338 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
340 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
341 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
343 void tcp_tasklet_init(void);
345 void tcp_v4_err(struct sk_buff *skb, u32);
347 void tcp_shutdown(struct sock *sk, int how);
349 void tcp_v4_early_demux(struct sk_buff *skb);
350 int tcp_v4_rcv(struct sk_buff *skb);
352 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
353 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
355 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
357 void tcp_release_cb(struct sock *sk);
358 void tcp_wfree(struct sk_buff *skb);
359 void tcp_write_timer_handler(struct sock *sk);
360 void tcp_delack_timer_handler(struct sock *sk);
361 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
362 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
363 const struct tcphdr *th, unsigned int len);
364 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
365 const struct tcphdr *th, unsigned int len);
366 void tcp_rcv_space_adjust(struct sock *sk);
367 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
368 void tcp_twsk_destructor(struct sock *sk);
369 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
370 struct pipe_inode_info *pipe, size_t len,
373 static inline void tcp_dec_quickack_mode(struct sock *sk,
374 const unsigned int pkts)
376 struct inet_connection_sock *icsk = inet_csk(sk);
378 if (icsk->icsk_ack.quick) {
379 if (pkts >= icsk->icsk_ack.quick) {
380 icsk->icsk_ack.quick = 0;
381 /* Leaving quickack mode we deflate ATO. */
382 icsk->icsk_ack.ato = TCP_ATO_MIN;
384 icsk->icsk_ack.quick -= pkts;
389 #define TCP_ECN_QUEUE_CWR 2
390 #define TCP_ECN_DEMAND_CWR 4
391 #define TCP_ECN_SEEN 8
401 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
403 const struct tcphdr *th);
404 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
405 struct request_sock *req, struct request_sock **prev,
407 int tcp_child_process(struct sock *parent, struct sock *child,
408 struct sk_buff *skb);
409 void tcp_enter_loss(struct sock *sk);
410 void tcp_clear_retrans(struct tcp_sock *tp);
411 void tcp_update_metrics(struct sock *sk);
412 void tcp_init_metrics(struct sock *sk);
413 void tcp_metrics_init(void);
414 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
415 bool paws_check, bool timestamps);
416 bool tcp_remember_stamp(struct sock *sk);
417 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
418 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
419 void tcp_disable_fack(struct tcp_sock *tp);
420 void tcp_close(struct sock *sk, long timeout);
421 void tcp_init_sock(struct sock *sk);
422 unsigned int tcp_poll(struct file *file, struct socket *sock,
423 struct poll_table_struct *wait);
424 int tcp_getsockopt(struct sock *sk, int level, int optname,
425 char __user *optval, int __user *optlen);
426 int tcp_setsockopt(struct sock *sk, int level, int optname,
427 char __user *optval, unsigned int optlen);
428 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
429 char __user *optval, int __user *optlen);
430 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
431 char __user *optval, unsigned int optlen);
432 void tcp_set_keepalive(struct sock *sk, int val);
433 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
434 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
435 size_t len, int nonblock, int flags, int *addr_len);
436 void tcp_parse_options(const struct sk_buff *skb,
437 struct tcp_options_received *opt_rx,
438 int estab, struct tcp_fastopen_cookie *foc);
439 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
442 * TCP v4 functions exported for the inet6 API
445 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
446 void tcp_v4_mtu_reduced(struct sock *sk);
447 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
448 struct sock *tcp_create_openreq_child(struct sock *sk,
449 struct request_sock *req,
450 struct sk_buff *skb);
451 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
452 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
453 struct request_sock *req,
454 struct dst_entry *dst);
455 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
456 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
457 int tcp_connect(struct sock *sk);
458 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
459 struct request_sock *req,
460 struct tcp_fastopen_cookie *foc);
461 int tcp_disconnect(struct sock *sk, int flags);
463 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
464 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
465 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
467 /* From syncookies.c */
468 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
470 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
471 #ifdef CONFIG_SYN_COOKIES
473 /* Syncookies use a monotonic timer which increments every 60 seconds.
474 * This counter is used both as a hash input and partially encoded into
475 * the cookie value. A cookie is only validated further if the delta
476 * between the current counter value and the encoded one is less than this,
477 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
478 * the counter advances immediately after a cookie is generated).
480 #define MAX_SYNCOOKIE_AGE 2
482 static inline u32 tcp_cookie_time(void)
484 u64 val = get_jiffies_64();
486 do_div(val, 60 * HZ);
490 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
492 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
494 __u32 cookie_init_timestamp(struct request_sock *req);
495 bool cookie_timestamp_decode(struct tcp_options_received *opt);
496 bool cookie_ecn_ok(const struct tcp_options_received *opt,
497 const struct net *net, const struct dst_entry *dst);
499 /* From net/ipv6/syncookies.c */
500 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
502 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
504 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
505 const struct tcphdr *th, u16 *mssp);
506 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
511 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
513 bool tcp_may_send_now(struct sock *sk);
514 int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
515 int tcp_retransmit_skb(struct sock *, struct sk_buff *);
516 void tcp_retransmit_timer(struct sock *sk);
517 void tcp_xmit_retransmit_queue(struct sock *);
518 void tcp_simple_retransmit(struct sock *);
519 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
520 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
522 void tcp_send_probe0(struct sock *);
523 void tcp_send_partial(struct sock *);
524 int tcp_write_wakeup(struct sock *);
525 void tcp_send_fin(struct sock *sk);
526 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
527 int tcp_send_synack(struct sock *);
528 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
530 void tcp_push_one(struct sock *, unsigned int mss_now);
531 void tcp_send_ack(struct sock *sk);
532 void tcp_send_delayed_ack(struct sock *sk);
533 void tcp_send_loss_probe(struct sock *sk);
534 bool tcp_schedule_loss_probe(struct sock *sk);
537 void tcp_resume_early_retransmit(struct sock *sk);
538 void tcp_rearm_rto(struct sock *sk);
539 void tcp_reset(struct sock *sk);
542 void tcp_init_xmit_timers(struct sock *);
543 static inline void tcp_clear_xmit_timers(struct sock *sk)
545 inet_csk_clear_xmit_timers(sk);
548 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
549 unsigned int tcp_current_mss(struct sock *sk);
551 /* Bound MSS / TSO packet size with the half of the window */
552 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
556 /* When peer uses tiny windows, there is no use in packetizing
557 * to sub-MSS pieces for the sake of SWS or making sure there
558 * are enough packets in the pipe for fast recovery.
560 * On the other hand, for extremely large MSS devices, handling
561 * smaller than MSS windows in this way does make sense.
563 if (tp->max_window >= 512)
564 cutoff = (tp->max_window >> 1);
566 cutoff = tp->max_window;
568 if (cutoff && pktsize > cutoff)
569 return max_t(int, cutoff, 68U - tp->tcp_header_len);
575 void tcp_get_info(const struct sock *, struct tcp_info *);
577 /* Read 'sendfile()'-style from a TCP socket */
578 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
579 unsigned int, size_t);
580 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
581 sk_read_actor_t recv_actor);
583 void tcp_initialize_rcv_mss(struct sock *sk);
585 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
586 int tcp_mss_to_mtu(struct sock *sk, int mss);
587 void tcp_mtup_init(struct sock *sk);
588 void tcp_init_buffer_space(struct sock *sk);
590 static inline void tcp_bound_rto(const struct sock *sk)
592 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
593 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
596 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
598 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
601 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
603 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
604 ntohl(TCP_FLAG_ACK) |
608 static inline void tcp_fast_path_on(struct tcp_sock *tp)
610 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
613 static inline void tcp_fast_path_check(struct sock *sk)
615 struct tcp_sock *tp = tcp_sk(sk);
617 if (skb_queue_empty(&tp->out_of_order_queue) &&
619 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
621 tcp_fast_path_on(tp);
624 /* Compute the actual rto_min value */
625 static inline u32 tcp_rto_min(struct sock *sk)
627 const struct dst_entry *dst = __sk_dst_get(sk);
628 u32 rto_min = TCP_RTO_MIN;
630 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
631 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
635 static inline u32 tcp_rto_min_us(struct sock *sk)
637 return jiffies_to_usecs(tcp_rto_min(sk));
640 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
642 return dst_metric_locked(dst, RTAX_CC_ALGO);
645 /* Compute the actual receive window we are currently advertising.
646 * Rcv_nxt can be after the window if our peer push more data
647 * than the offered window.
649 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
651 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
658 /* Choose a new window, without checks for shrinking, and without
659 * scaling applied to the result. The caller does these things
660 * if necessary. This is a "raw" window selection.
662 u32 __tcp_select_window(struct sock *sk);
664 void tcp_send_window_probe(struct sock *sk);
666 /* TCP timestamps are only 32-bits, this causes a slight
667 * complication on 64-bit systems since we store a snapshot
668 * of jiffies in the buffer control blocks below. We decided
669 * to use only the low 32-bits of jiffies and hide the ugly
670 * casts with the following macro.
672 #define tcp_time_stamp ((__u32)(jiffies))
674 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
676 return skb->skb_mstamp.stamp_jiffies;
680 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
682 #define TCPHDR_FIN 0x01
683 #define TCPHDR_SYN 0x02
684 #define TCPHDR_RST 0x04
685 #define TCPHDR_PSH 0x08
686 #define TCPHDR_ACK 0x10
687 #define TCPHDR_URG 0x20
688 #define TCPHDR_ECE 0x40
689 #define TCPHDR_CWR 0x80
691 /* This is what the send packet queuing engine uses to pass
692 * TCP per-packet control information to the transmission code.
693 * We also store the host-order sequence numbers in here too.
694 * This is 44 bytes if IPV6 is enabled.
695 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
698 __u32 seq; /* Starting sequence number */
699 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
701 /* Note : tcp_tw_isn is used in input path only
702 * (isn chosen by tcp_timewait_state_process())
704 * tcp_gso_segs is used in write queue only,
705 * cf tcp_skb_pcount()
710 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
712 __u8 sacked; /* State flags for SACK/FACK. */
713 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
714 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
715 #define TCPCB_LOST 0x04 /* SKB is lost */
716 #define TCPCB_TAGBITS 0x07 /* All tag bits */
717 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
718 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
719 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
722 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
724 __u32 ack_seq; /* Sequence number ACK'd */
726 struct inet_skb_parm h4;
727 #if IS_ENABLED(CONFIG_IPV6)
728 struct inet6_skb_parm h6;
730 } header; /* For incoming frames */
733 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
736 #if IS_ENABLED(CONFIG_IPV6)
737 /* This is the variant of inet6_iif() that must be used by TCP,
738 * as TCP moves IP6CB into a different location in skb->cb[]
740 static inline int tcp_v6_iif(const struct sk_buff *skb)
742 return TCP_SKB_CB(skb)->header.h6.iif;
746 /* Due to TSO, an SKB can be composed of multiple actual
747 * packets. To keep these tracked properly, we use this.
749 static inline int tcp_skb_pcount(const struct sk_buff *skb)
751 return TCP_SKB_CB(skb)->tcp_gso_segs;
754 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
756 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
759 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
761 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
764 /* This is valid iff tcp_skb_pcount() > 1. */
765 static inline int tcp_skb_mss(const struct sk_buff *skb)
767 return skb_shinfo(skb)->gso_size;
770 /* Events passed to congestion control interface */
772 CA_EVENT_TX_START, /* first transmit when no packets in flight */
773 CA_EVENT_CWND_RESTART, /* congestion window restart */
774 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
775 CA_EVENT_LOSS, /* loss timeout */
776 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
777 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
778 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
779 CA_EVENT_NON_DELAYED_ACK,
782 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
783 enum tcp_ca_ack_event_flags {
784 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
785 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
786 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
790 * Interface for adding new TCP congestion control handlers
792 #define TCP_CA_NAME_MAX 16
793 #define TCP_CA_MAX 128
794 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
796 #define TCP_CA_UNSPEC 0
798 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
799 #define TCP_CONG_NON_RESTRICTED 0x1
800 /* Requires ECN/ECT set on all packets */
801 #define TCP_CONG_NEEDS_ECN 0x2
803 struct tcp_congestion_ops {
804 struct list_head list;
808 /* initialize private data (optional) */
809 void (*init)(struct sock *sk);
810 /* cleanup private data (optional) */
811 void (*release)(struct sock *sk);
813 /* return slow start threshold (required) */
814 u32 (*ssthresh)(struct sock *sk);
815 /* do new cwnd calculation (required) */
816 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
817 /* call before changing ca_state (optional) */
818 void (*set_state)(struct sock *sk, u8 new_state);
819 /* call when cwnd event occurs (optional) */
820 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
821 /* call when ack arrives (optional) */
822 void (*in_ack_event)(struct sock *sk, u32 flags);
823 /* new value of cwnd after loss (optional) */
824 u32 (*undo_cwnd)(struct sock *sk);
825 /* hook for packet ack accounting (optional) */
826 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
827 /* get info for inet_diag (optional) */
828 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
830 char name[TCP_CA_NAME_MAX];
831 struct module *owner;
834 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
835 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
837 void tcp_assign_congestion_control(struct sock *sk);
838 void tcp_init_congestion_control(struct sock *sk);
839 void tcp_cleanup_congestion_control(struct sock *sk);
840 int tcp_set_default_congestion_control(const char *name);
841 void tcp_get_default_congestion_control(char *name);
842 void tcp_get_available_congestion_control(char *buf, size_t len);
843 void tcp_get_allowed_congestion_control(char *buf, size_t len);
844 int tcp_set_allowed_congestion_control(char *allowed);
845 int tcp_set_congestion_control(struct sock *sk, const char *name);
846 void tcp_slow_start(struct tcp_sock *tp, u32 acked);
847 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
849 u32 tcp_reno_ssthresh(struct sock *sk);
850 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
851 extern struct tcp_congestion_ops tcp_reno;
853 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
854 u32 tcp_ca_get_key_by_name(const char *name);
856 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
858 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
864 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
866 const struct inet_connection_sock *icsk = inet_csk(sk);
868 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
871 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
873 struct inet_connection_sock *icsk = inet_csk(sk);
875 if (icsk->icsk_ca_ops->set_state)
876 icsk->icsk_ca_ops->set_state(sk, ca_state);
877 icsk->icsk_ca_state = ca_state;
880 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
882 const struct inet_connection_sock *icsk = inet_csk(sk);
884 if (icsk->icsk_ca_ops->cwnd_event)
885 icsk->icsk_ca_ops->cwnd_event(sk, event);
888 /* These functions determine how the current flow behaves in respect of SACK
889 * handling. SACK is negotiated with the peer, and therefore it can vary
890 * between different flows.
892 * tcp_is_sack - SACK enabled
893 * tcp_is_reno - No SACK
894 * tcp_is_fack - FACK enabled, implies SACK enabled
896 static inline int tcp_is_sack(const struct tcp_sock *tp)
898 return tp->rx_opt.sack_ok;
901 static inline bool tcp_is_reno(const struct tcp_sock *tp)
903 return !tcp_is_sack(tp);
906 static inline bool tcp_is_fack(const struct tcp_sock *tp)
908 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
911 static inline void tcp_enable_fack(struct tcp_sock *tp)
913 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
916 /* TCP early-retransmit (ER) is similar to but more conservative than
917 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
919 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
921 tp->do_early_retrans = sysctl_tcp_early_retrans &&
922 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
923 sysctl_tcp_reordering == 3;
926 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
928 tp->do_early_retrans = 0;
931 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
933 return tp->sacked_out + tp->lost_out;
936 /* This determines how many packets are "in the network" to the best
937 * of our knowledge. In many cases it is conservative, but where
938 * detailed information is available from the receiver (via SACK
939 * blocks etc.) we can make more aggressive calculations.
941 * Use this for decisions involving congestion control, use just
942 * tp->packets_out to determine if the send queue is empty or not.
944 * Read this equation as:
946 * "Packets sent once on transmission queue" MINUS
947 * "Packets left network, but not honestly ACKed yet" PLUS
948 * "Packets fast retransmitted"
950 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
952 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
955 #define TCP_INFINITE_SSTHRESH 0x7fffffff
957 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
959 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
962 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
964 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
965 (1 << inet_csk(sk)->icsk_ca_state);
968 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
969 * The exception is cwnd reduction phase, when cwnd is decreasing towards
972 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
974 const struct tcp_sock *tp = tcp_sk(sk);
976 if (tcp_in_cwnd_reduction(sk))
977 return tp->snd_ssthresh;
979 return max(tp->snd_ssthresh,
980 ((tp->snd_cwnd >> 1) +
981 (tp->snd_cwnd >> 2)));
984 /* Use define here intentionally to get WARN_ON location shown at the caller */
985 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
987 void tcp_enter_cwr(struct sock *sk);
988 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
990 /* The maximum number of MSS of available cwnd for which TSO defers
991 * sending if not using sysctl_tcp_tso_win_divisor.
993 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
998 /* Slow start with delack produces 3 packets of burst, so that
999 * it is safe "de facto". This will be the default - same as
1000 * the default reordering threshold - but if reordering increases,
1001 * we must be able to allow cwnd to burst at least this much in order
1002 * to not pull it back when holes are filled.
1004 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1006 return tp->reordering;
1009 /* Returns end sequence number of the receiver's advertised window */
1010 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1012 return tp->snd_una + tp->snd_wnd;
1015 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1016 * flexible approach. The RFC suggests cwnd should not be raised unless
1017 * it was fully used previously. And that's exactly what we do in
1018 * congestion avoidance mode. But in slow start we allow cwnd to grow
1019 * as long as the application has used half the cwnd.
1021 * cwnd is 10 (IW10), but application sends 9 frames.
1022 * We allow cwnd to reach 18 when all frames are ACKed.
1023 * This check is safe because it's as aggressive as slow start which already
1024 * risks 100% overshoot. The advantage is that we discourage application to
1025 * either send more filler packets or data to artificially blow up the cwnd
1026 * usage, and allow application-limited process to probe bw more aggressively.
1028 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1030 const struct tcp_sock *tp = tcp_sk(sk);
1032 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1033 if (tp->snd_cwnd <= tp->snd_ssthresh)
1034 return tp->snd_cwnd < 2 * tp->max_packets_out;
1036 return tp->is_cwnd_limited;
1039 static inline void tcp_check_probe_timer(struct sock *sk)
1041 const struct tcp_sock *tp = tcp_sk(sk);
1042 const struct inet_connection_sock *icsk = inet_csk(sk);
1044 if (!tp->packets_out && !icsk->icsk_pending)
1045 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1046 icsk->icsk_rto, TCP_RTO_MAX);
1049 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1054 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1060 * Calculate(/check) TCP checksum
1062 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1063 __be32 daddr, __wsum base)
1065 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1068 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1070 return __skb_checksum_complete(skb);
1073 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1075 return !skb_csum_unnecessary(skb) &&
1076 __tcp_checksum_complete(skb);
1079 /* Prequeue for VJ style copy to user, combined with checksumming. */
1081 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1083 tp->ucopy.task = NULL;
1085 tp->ucopy.memory = 0;
1086 skb_queue_head_init(&tp->ucopy.prequeue);
1089 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1094 static const char *statename[]={
1095 "Unused","Established","Syn Sent","Syn Recv",
1096 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1097 "Close Wait","Last ACK","Listen","Closing"
1100 void tcp_set_state(struct sock *sk, int state);
1102 void tcp_done(struct sock *sk);
1104 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1107 rx_opt->num_sacks = 0;
1110 u32 tcp_default_init_rwnd(u32 mss);
1112 /* Determine a window scaling and initial window to offer. */
1113 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1114 __u32 *window_clamp, int wscale_ok,
1115 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1117 static inline int tcp_win_from_space(int space)
1119 return sysctl_tcp_adv_win_scale<=0 ?
1120 (space>>(-sysctl_tcp_adv_win_scale)) :
1121 space - (space>>sysctl_tcp_adv_win_scale);
1124 /* Note: caller must be prepared to deal with negative returns */
1125 static inline int tcp_space(const struct sock *sk)
1127 return tcp_win_from_space(sk->sk_rcvbuf -
1128 atomic_read(&sk->sk_rmem_alloc));
1131 static inline int tcp_full_space(const struct sock *sk)
1133 return tcp_win_from_space(sk->sk_rcvbuf);
1136 static inline void tcp_openreq_init(struct request_sock *req,
1137 struct tcp_options_received *rx_opt,
1138 struct sk_buff *skb, struct sock *sk)
1140 struct inet_request_sock *ireq = inet_rsk(req);
1142 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1144 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1145 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1146 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1147 req->mss = rx_opt->mss_clamp;
1148 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1149 ireq->tstamp_ok = rx_opt->tstamp_ok;
1150 ireq->sack_ok = rx_opt->sack_ok;
1151 ireq->snd_wscale = rx_opt->snd_wscale;
1152 ireq->wscale_ok = rx_opt->wscale_ok;
1155 ireq->ir_rmt_port = tcp_hdr(skb)->source;
1156 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
1157 ireq->ir_mark = inet_request_mark(sk, skb);
1160 extern void tcp_openreq_init_rwin(struct request_sock *req,
1161 struct sock *sk, struct dst_entry *dst);
1163 void tcp_enter_memory_pressure(struct sock *sk);
1165 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1167 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1170 static inline int keepalive_time_when(const struct tcp_sock *tp)
1172 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1175 static inline int keepalive_probes(const struct tcp_sock *tp)
1177 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1180 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1182 const struct inet_connection_sock *icsk = &tp->inet_conn;
1184 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1185 tcp_time_stamp - tp->rcv_tstamp);
1188 static inline int tcp_fin_time(const struct sock *sk)
1190 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1191 const int rto = inet_csk(sk)->icsk_rto;
1193 if (fin_timeout < (rto << 2) - (rto >> 1))
1194 fin_timeout = (rto << 2) - (rto >> 1);
1199 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1202 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1204 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1207 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1208 * then following tcp messages have valid values. Ignore 0 value,
1209 * or else 'negative' tsval might forbid us to accept their packets.
1211 if (!rx_opt->ts_recent)
1216 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1219 if (tcp_paws_check(rx_opt, 0))
1222 /* RST segments are not recommended to carry timestamp,
1223 and, if they do, it is recommended to ignore PAWS because
1224 "their cleanup function should take precedence over timestamps."
1225 Certainly, it is mistake. It is necessary to understand the reasons
1226 of this constraint to relax it: if peer reboots, clock may go
1227 out-of-sync and half-open connections will not be reset.
1228 Actually, the problem would be not existing if all
1229 the implementations followed draft about maintaining clock
1230 via reboots. Linux-2.2 DOES NOT!
1232 However, we can relax time bounds for RST segments to MSL.
1234 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1239 static inline void tcp_mib_init(struct net *net)
1242 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1243 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1244 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1245 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1249 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1251 tp->lost_skb_hint = NULL;
1254 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1256 tcp_clear_retrans_hints_partial(tp);
1257 tp->retransmit_skb_hint = NULL;
1263 union tcp_md5_addr {
1265 #if IS_ENABLED(CONFIG_IPV6)
1270 /* - key database */
1271 struct tcp_md5sig_key {
1272 struct hlist_node node;
1274 u8 family; /* AF_INET or AF_INET6 */
1275 union tcp_md5_addr addr;
1276 u8 key[TCP_MD5SIG_MAXKEYLEN];
1277 struct rcu_head rcu;
1281 struct tcp_md5sig_info {
1282 struct hlist_head head;
1283 struct rcu_head rcu;
1286 /* - pseudo header */
1287 struct tcp4_pseudohdr {
1295 struct tcp6_pseudohdr {
1296 struct in6_addr saddr;
1297 struct in6_addr daddr;
1299 __be32 protocol; /* including padding */
1302 union tcp_md5sum_block {
1303 struct tcp4_pseudohdr ip4;
1304 #if IS_ENABLED(CONFIG_IPV6)
1305 struct tcp6_pseudohdr ip6;
1309 /* - pool: digest algorithm, hash description and scratch buffer */
1310 struct tcp_md5sig_pool {
1311 struct hash_desc md5_desc;
1312 union tcp_md5sum_block md5_blk;
1316 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1317 const struct sock *sk, const struct request_sock *req,
1318 const struct sk_buff *skb);
1319 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1320 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1321 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1323 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1324 struct sock *addr_sk);
1326 #ifdef CONFIG_TCP_MD5SIG
1327 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1328 const union tcp_md5_addr *addr,
1330 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1332 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1333 const union tcp_md5_addr *addr,
1338 #define tcp_twsk_md5_key(twsk) NULL
1341 bool tcp_alloc_md5sig_pool(void);
1343 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1344 static inline void tcp_put_md5sig_pool(void)
1349 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1350 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1351 unsigned int header_len);
1352 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1353 const struct tcp_md5sig_key *key);
1355 /* From tcp_fastopen.c */
1356 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1357 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1358 unsigned long *last_syn_loss);
1359 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1360 struct tcp_fastopen_cookie *cookie, bool syn_lost);
1361 struct tcp_fastopen_request {
1362 /* Fast Open cookie. Size 0 means a cookie request */
1363 struct tcp_fastopen_cookie cookie;
1364 struct msghdr *data; /* data in MSG_FASTOPEN */
1366 int copied; /* queued in tcp_connect() */
1368 void tcp_free_fastopen_req(struct tcp_sock *tp);
1370 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1371 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1372 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1373 struct request_sock *req,
1374 struct tcp_fastopen_cookie *foc,
1375 struct dst_entry *dst);
1376 void tcp_fastopen_init_key_once(bool publish);
1377 #define TCP_FASTOPEN_KEY_LENGTH 16
1379 /* Fastopen key context */
1380 struct tcp_fastopen_context {
1381 struct crypto_cipher *tfm;
1382 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1383 struct rcu_head rcu;
1386 /* write queue abstraction */
1387 static inline void tcp_write_queue_purge(struct sock *sk)
1389 struct sk_buff *skb;
1391 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1392 sk_wmem_free_skb(sk, skb);
1394 tcp_clear_all_retrans_hints(tcp_sk(sk));
1397 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1399 return skb_peek(&sk->sk_write_queue);
1402 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1404 return skb_peek_tail(&sk->sk_write_queue);
1407 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1408 const struct sk_buff *skb)
1410 return skb_queue_next(&sk->sk_write_queue, skb);
1413 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1414 const struct sk_buff *skb)
1416 return skb_queue_prev(&sk->sk_write_queue, skb);
1419 #define tcp_for_write_queue(skb, sk) \
1420 skb_queue_walk(&(sk)->sk_write_queue, skb)
1422 #define tcp_for_write_queue_from(skb, sk) \
1423 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1425 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1426 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1428 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1430 return sk->sk_send_head;
1433 static inline bool tcp_skb_is_last(const struct sock *sk,
1434 const struct sk_buff *skb)
1436 return skb_queue_is_last(&sk->sk_write_queue, skb);
1439 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1441 if (tcp_skb_is_last(sk, skb))
1442 sk->sk_send_head = NULL;
1444 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1447 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1449 if (sk->sk_send_head == skb_unlinked)
1450 sk->sk_send_head = NULL;
1453 static inline void tcp_init_send_head(struct sock *sk)
1455 sk->sk_send_head = NULL;
1458 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1460 __skb_queue_tail(&sk->sk_write_queue, skb);
1463 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1465 __tcp_add_write_queue_tail(sk, skb);
1467 /* Queue it, remembering where we must start sending. */
1468 if (sk->sk_send_head == NULL) {
1469 sk->sk_send_head = skb;
1471 if (tcp_sk(sk)->highest_sack == NULL)
1472 tcp_sk(sk)->highest_sack = skb;
1476 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1478 __skb_queue_head(&sk->sk_write_queue, skb);
1481 /* Insert buff after skb on the write queue of sk. */
1482 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1483 struct sk_buff *buff,
1486 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1489 /* Insert new before skb on the write queue of sk. */
1490 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1491 struct sk_buff *skb,
1494 __skb_queue_before(&sk->sk_write_queue, skb, new);
1496 if (sk->sk_send_head == skb)
1497 sk->sk_send_head = new;
1500 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1502 __skb_unlink(skb, &sk->sk_write_queue);
1505 static inline bool tcp_write_queue_empty(struct sock *sk)
1507 return skb_queue_empty(&sk->sk_write_queue);
1510 static inline void tcp_push_pending_frames(struct sock *sk)
1512 if (tcp_send_head(sk)) {
1513 struct tcp_sock *tp = tcp_sk(sk);
1515 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1519 /* Start sequence of the skb just after the highest skb with SACKed
1520 * bit, valid only if sacked_out > 0 or when the caller has ensured
1521 * validity by itself.
1523 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1525 if (!tp->sacked_out)
1528 if (tp->highest_sack == NULL)
1531 return TCP_SKB_CB(tp->highest_sack)->seq;
1534 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1536 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1537 tcp_write_queue_next(sk, skb);
1540 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1542 return tcp_sk(sk)->highest_sack;
1545 static inline void tcp_highest_sack_reset(struct sock *sk)
1547 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1550 /* Called when old skb is about to be deleted (to be combined with new skb) */
1551 static inline void tcp_highest_sack_combine(struct sock *sk,
1552 struct sk_buff *old,
1553 struct sk_buff *new)
1555 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1556 tcp_sk(sk)->highest_sack = new;
1559 /* Determines whether this is a thin stream (which may suffer from
1560 * increased latency). Used to trigger latency-reducing mechanisms.
1562 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1564 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1568 enum tcp_seq_states {
1569 TCP_SEQ_STATE_LISTENING,
1570 TCP_SEQ_STATE_OPENREQ,
1571 TCP_SEQ_STATE_ESTABLISHED,
1574 int tcp_seq_open(struct inode *inode, struct file *file);
1576 struct tcp_seq_afinfo {
1579 const struct file_operations *seq_fops;
1580 struct seq_operations seq_ops;
1583 struct tcp_iter_state {
1584 struct seq_net_private p;
1586 enum tcp_seq_states state;
1587 struct sock *syn_wait_sk;
1588 int bucket, offset, sbucket, num;
1593 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1594 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1596 extern struct request_sock_ops tcp_request_sock_ops;
1597 extern struct request_sock_ops tcp6_request_sock_ops;
1599 void tcp_v4_destroy_sock(struct sock *sk);
1601 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1602 netdev_features_t features);
1603 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1604 int tcp_gro_complete(struct sk_buff *skb);
1606 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1608 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1610 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1613 static inline bool tcp_stream_memory_free(const struct sock *sk)
1615 const struct tcp_sock *tp = tcp_sk(sk);
1616 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1618 return notsent_bytes < tcp_notsent_lowat(tp);
1621 #ifdef CONFIG_PROC_FS
1622 int tcp4_proc_init(void);
1623 void tcp4_proc_exit(void);
1626 int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
1627 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1628 const struct tcp_request_sock_ops *af_ops,
1629 struct sock *sk, struct sk_buff *skb);
1631 /* TCP af-specific functions */
1632 struct tcp_sock_af_ops {
1633 #ifdef CONFIG_TCP_MD5SIG
1634 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1635 struct sock *addr_sk);
1636 int (*calc_md5_hash) (char *location,
1637 struct tcp_md5sig_key *md5,
1638 const struct sock *sk,
1639 const struct request_sock *req,
1640 const struct sk_buff *skb);
1641 int (*md5_parse) (struct sock *sk,
1642 char __user *optval,
1647 struct tcp_request_sock_ops {
1649 #ifdef CONFIG_TCP_MD5SIG
1650 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1651 struct request_sock *req);
1652 int (*calc_md5_hash) (char *location,
1653 struct tcp_md5sig_key *md5,
1654 const struct sock *sk,
1655 const struct request_sock *req,
1656 const struct sk_buff *skb);
1658 void (*init_req)(struct request_sock *req, struct sock *sk,
1659 struct sk_buff *skb);
1660 #ifdef CONFIG_SYN_COOKIES
1661 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1664 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1665 const struct request_sock *req,
1667 __u32 (*init_seq)(const struct sk_buff *skb);
1668 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1669 struct flowi *fl, struct request_sock *req,
1670 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
1671 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1672 const unsigned long timeout);
1675 #ifdef CONFIG_SYN_COOKIES
1676 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1677 struct sock *sk, struct sk_buff *skb,
1680 return ops->cookie_init_seq(sk, skb, mss);
1683 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1684 struct sock *sk, struct sk_buff *skb,
1691 int tcpv4_offload_init(void);
1693 void tcp_v4_init(void);
1694 void tcp_init(void);
1697 * Save and compile IPv4 options, return a pointer to it
1699 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1701 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1702 struct ip_options_rcu *dopt = NULL;
1705 int opt_size = sizeof(*dopt) + opt->optlen;
1707 dopt = kmalloc(opt_size, GFP_ATOMIC);
1708 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {