2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo;
52 extern struct percpu_counter tcp_orphan_count;
53 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Offer an initial receive window of 10 mss. */
65 #define TCP_DEFAULT_INIT_RCVWND 10
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 512
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
76 /* Maximal reordering. */
77 #define TCP_MAX_REORDERING 127
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
102 * when active opening a connection.
103 * RFC1122 says the minimum retry MUST
104 * be at least 180secs. Nevertheless
105 * this value is corresponding to
106 * 63secs of retransmission with the
107 * current initial RTO.
110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
111 * when passive opening a connection.
112 * This is corresponding to 31secs of
113 * retransmission with the current
117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 * state, about 60 seconds */
119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
120 /* BSD style FIN_WAIT2 deadlock breaker.
121 * It used to be 3min, new value is 60sec,
122 * to combine FIN-WAIT-2 timeout with
126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
129 #define TCP_ATO_MIN ((unsigned)(HZ/25))
131 #define TCP_DELACK_MIN 4U
132 #define TCP_ATO_MIN 4U
134 #define TCP_RTO_MAX ((unsigned)(120*HZ))
135 #define TCP_RTO_MIN ((unsigned)(HZ/5))
136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
138 * used as a fallback RTO for the
139 * initial data transmission if no
140 * valid RTT sample has been acquired,
141 * most likely due to retrans in 3WHS.
144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
145 * for local resources.
148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
150 #define TCP_KEEPALIVE_INTVL (75*HZ)
152 #define MAX_TCP_KEEPIDLE 32767
153 #define MAX_TCP_KEEPINTVL 32767
154 #define MAX_TCP_KEEPCNT 127
155 #define MAX_TCP_SYNCNT 127
157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
161 * after this time. It should be equal
162 * (or greater than) TCP_TIMEWAIT_LEN
163 * to provide reliability equal to one
164 * provided by timewait state.
166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
167 * timestamps. It must be less than
168 * minimal timewait lifetime.
174 #define TCPOPT_NOP 1 /* Padding */
175 #define TCPOPT_EOL 0 /* End of options */
176 #define TCPOPT_MSS 2 /* Segment size negotiating */
177 #define TCPOPT_WINDOW 3 /* Window scaling */
178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
179 #define TCPOPT_SACK 5 /* SACK Block */
180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
182 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
183 #define TCPOPT_EXP 254 /* Experimental */
184 /* Magic number to be after the option value for sharing TCP
185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
187 #define TCPOPT_FASTOPEN_MAGIC 0xF989
193 #define TCPOLEN_MSS 4
194 #define TCPOLEN_WINDOW 3
195 #define TCPOLEN_SACK_PERM 2
196 #define TCPOLEN_TIMESTAMP 10
197 #define TCPOLEN_MD5SIG 18
198 #define TCPOLEN_EXP_FASTOPEN_BASE 4
199 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
200 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
201 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
202 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
204 /* But this is what stacks really send out. */
205 #define TCPOLEN_TSTAMP_ALIGNED 12
206 #define TCPOLEN_WSCALE_ALIGNED 4
207 #define TCPOLEN_SACKPERM_ALIGNED 4
208 #define TCPOLEN_SACK_BASE 2
209 #define TCPOLEN_SACK_BASE_ALIGNED 4
210 #define TCPOLEN_SACK_PERBLOCK 8
211 #define TCPOLEN_MD5SIG_ALIGNED 20
212 #define TCPOLEN_MSS_ALIGNED 4
214 /* Flags in tp->nonagle */
215 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
216 #define TCP_NAGLE_CORK 2 /* Socket is corked */
217 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
219 /* TCP thin-stream limits */
220 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
222 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
223 #define TCP_INIT_CWND 10
225 /* Bit Flags for sysctl_tcp_fastopen */
226 #define TFO_CLIENT_ENABLE 1
227 #define TFO_SERVER_ENABLE 2
228 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
230 /* Process SYN data but skip cookie validation */
231 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100
232 /* Accept SYN data w/o any cookie option */
233 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
235 /* Force enable TFO on all listeners, i.e., not requiring the
236 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
238 #define TFO_SERVER_WO_SOCKOPT1 0x400
239 #define TFO_SERVER_WO_SOCKOPT2 0x800
240 /* Always create TFO child sockets on a TFO listener even when
241 * cookie/data not present. (For testing purpose!)
243 #define TFO_SERVER_ALWAYS 0x1000
245 extern struct inet_timewait_death_row tcp_death_row;
247 /* sysctl variables for tcp */
248 extern int sysctl_tcp_timestamps;
249 extern int sysctl_tcp_window_scaling;
250 extern int sysctl_tcp_sack;
251 extern int sysctl_tcp_fin_timeout;
252 extern int sysctl_tcp_keepalive_time;
253 extern int sysctl_tcp_keepalive_probes;
254 extern int sysctl_tcp_keepalive_intvl;
255 extern int sysctl_tcp_syn_retries;
256 extern int sysctl_tcp_synack_retries;
257 extern int sysctl_tcp_retries1;
258 extern int sysctl_tcp_retries2;
259 extern int sysctl_tcp_orphan_retries;
260 extern int sysctl_tcp_syncookies;
261 extern int sysctl_tcp_fastopen;
262 extern int sysctl_tcp_retrans_collapse;
263 extern int sysctl_tcp_stdurg;
264 extern int sysctl_tcp_rfc1337;
265 extern int sysctl_tcp_abort_on_overflow;
266 extern int sysctl_tcp_max_orphans;
267 extern int sysctl_tcp_fack;
268 extern int sysctl_tcp_reordering;
269 extern int sysctl_tcp_ecn;
270 extern int sysctl_tcp_dsack;
271 extern int sysctl_tcp_wmem[3];
272 extern int sysctl_tcp_rmem[3];
273 extern int sysctl_tcp_app_win;
274 extern int sysctl_tcp_adv_win_scale;
275 extern int sysctl_tcp_tw_reuse;
276 extern int sysctl_tcp_frto;
277 extern int sysctl_tcp_frto_response;
278 extern int sysctl_tcp_low_latency;
279 extern int sysctl_tcp_dma_copybreak;
280 extern int sysctl_tcp_nometrics_save;
281 extern int sysctl_tcp_moderate_rcvbuf;
282 extern int sysctl_tcp_tso_win_divisor;
283 extern int sysctl_tcp_abc;
284 extern int sysctl_tcp_mtu_probing;
285 extern int sysctl_tcp_base_mss;
286 extern int sysctl_tcp_workaround_signed_windows;
287 extern int sysctl_tcp_slow_start_after_idle;
288 extern int sysctl_tcp_max_ssthresh;
289 extern int sysctl_tcp_cookie_size;
290 extern int sysctl_tcp_thin_linear_timeouts;
291 extern int sysctl_tcp_thin_dupack;
292 extern int sysctl_tcp_early_retrans;
293 extern int sysctl_tcp_limit_output_bytes;
294 extern int sysctl_tcp_challenge_ack_limit;
296 extern atomic_long_t tcp_memory_allocated;
297 extern struct percpu_counter tcp_sockets_allocated;
298 extern int tcp_memory_pressure;
301 * The next routines deal with comparing 32 bit unsigned ints
302 * and worry about wraparound (automatic with unsigned arithmetic).
305 static inline bool before(__u32 seq1, __u32 seq2)
307 return (__s32)(seq1-seq2) < 0;
309 #define after(seq2, seq1) before(seq1, seq2)
311 /* is s2<=s1<=s3 ? */
312 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
314 return seq3 - seq2 >= seq1 - seq2;
317 static inline bool tcp_out_of_memory(struct sock *sk)
319 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
320 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
325 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
327 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
328 int orphans = percpu_counter_read_positive(ocp);
330 if (orphans << shift > sysctl_tcp_max_orphans) {
331 orphans = percpu_counter_sum_positive(ocp);
332 if (orphans << shift > sysctl_tcp_max_orphans)
338 extern bool tcp_check_oom(struct sock *sk, int shift);
340 /* syncookies: remember time of last synqueue overflow */
341 static inline void tcp_synq_overflow(struct sock *sk)
343 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
346 /* syncookies: no recent synqueue overflow on this listening socket? */
347 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
349 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
350 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
353 extern struct proto tcp_prot;
355 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
356 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
357 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
358 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
359 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
361 extern void tcp_init_mem(struct net *net);
363 extern void tcp_tasklet_init(void);
365 extern void tcp_v4_err(struct sk_buff *skb, u32);
367 extern void tcp_shutdown (struct sock *sk, int how);
369 extern void tcp_v4_early_demux(struct sk_buff *skb);
370 extern int tcp_v4_rcv(struct sk_buff *skb);
372 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk);
373 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
374 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
376 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
377 size_t size, int flags);
378 extern void tcp_release_cb(struct sock *sk);
379 extern void tcp_write_timer_handler(struct sock *sk);
380 extern void tcp_delack_timer_handler(struct sock *sk);
381 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
382 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
383 const struct tcphdr *th, unsigned int len);
384 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
385 const struct tcphdr *th, unsigned int len);
386 extern void tcp_rcv_space_adjust(struct sock *sk);
387 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
388 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
389 extern void tcp_twsk_destructor(struct sock *sk);
390 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
391 struct pipe_inode_info *pipe, size_t len,
394 static inline void tcp_dec_quickack_mode(struct sock *sk,
395 const unsigned int pkts)
397 struct inet_connection_sock *icsk = inet_csk(sk);
399 if (icsk->icsk_ack.quick) {
400 if (pkts >= icsk->icsk_ack.quick) {
401 icsk->icsk_ack.quick = 0;
402 /* Leaving quickack mode we deflate ATO. */
403 icsk->icsk_ack.ato = TCP_ATO_MIN;
405 icsk->icsk_ack.quick -= pkts;
410 #define TCP_ECN_QUEUE_CWR 2
411 #define TCP_ECN_DEMAND_CWR 4
412 #define TCP_ECN_SEEN 8
422 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
424 const struct tcphdr *th);
425 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
426 struct request_sock *req,
427 struct request_sock **prev,
429 extern int tcp_child_process(struct sock *parent, struct sock *child,
430 struct sk_buff *skb);
431 extern bool tcp_use_frto(struct sock *sk);
432 extern void tcp_enter_frto(struct sock *sk);
433 extern void tcp_enter_loss(struct sock *sk, int how);
434 extern void tcp_clear_retrans(struct tcp_sock *tp);
435 extern void tcp_update_metrics(struct sock *sk);
436 extern void tcp_init_metrics(struct sock *sk);
437 extern void tcp_metrics_init(void);
438 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check);
439 extern bool tcp_remember_stamp(struct sock *sk);
440 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
441 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
442 extern void tcp_disable_fack(struct tcp_sock *tp);
443 extern void tcp_close(struct sock *sk, long timeout);
444 extern void tcp_init_sock(struct sock *sk);
445 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
446 struct poll_table_struct *wait);
447 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
448 char __user *optval, int __user *optlen);
449 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
450 char __user *optval, unsigned int optlen);
451 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
452 char __user *optval, int __user *optlen);
453 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
454 char __user *optval, unsigned int optlen);
455 extern void tcp_set_keepalive(struct sock *sk, int val);
456 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
457 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
458 size_t len, int nonblock, int flags, int *addr_len);
459 extern void tcp_parse_options(const struct sk_buff *skb,
460 struct tcp_options_received *opt_rx, const u8 **hvpp,
461 int estab, struct tcp_fastopen_cookie *foc);
462 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
465 * TCP v4 functions exported for the inet6 API
468 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
469 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
470 extern struct sock * tcp_create_openreq_child(struct sock *sk,
471 struct request_sock *req,
472 struct sk_buff *skb);
473 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
474 struct request_sock *req,
475 struct dst_entry *dst);
476 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
477 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
479 extern int tcp_connect(struct sock *sk);
480 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
481 struct request_sock *req,
482 struct request_values *rvp,
483 struct tcp_fastopen_cookie *foc);
484 extern int tcp_disconnect(struct sock *sk, int flags);
486 void tcp_connect_init(struct sock *sk);
487 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
488 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
489 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
491 /* From syncookies.c */
492 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
493 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
494 struct ip_options *opt);
495 #ifdef CONFIG_SYN_COOKIES
496 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
499 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
507 extern __u32 cookie_init_timestamp(struct request_sock *req);
508 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
510 /* From net/ipv6/syncookies.c */
511 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
512 #ifdef CONFIG_SYN_COOKIES
513 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
516 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
525 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
527 extern bool tcp_may_send_now(struct sock *sk);
528 extern int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
529 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
530 extern void tcp_retransmit_timer(struct sock *sk);
531 extern void tcp_xmit_retransmit_queue(struct sock *);
532 extern void tcp_simple_retransmit(struct sock *);
533 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
534 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
536 extern void tcp_send_probe0(struct sock *);
537 extern void tcp_send_partial(struct sock *);
538 extern int tcp_write_wakeup(struct sock *);
539 extern void tcp_send_fin(struct sock *sk);
540 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
541 extern int tcp_send_synack(struct sock *);
542 extern bool tcp_syn_flood_action(struct sock *sk,
543 const struct sk_buff *skb,
545 extern void tcp_push_one(struct sock *, unsigned int mss_now);
546 extern void tcp_send_ack(struct sock *sk);
547 extern void tcp_send_delayed_ack(struct sock *sk);
550 extern void tcp_cwnd_application_limited(struct sock *sk);
551 extern void tcp_resume_early_retransmit(struct sock *sk);
552 extern void tcp_rearm_rto(struct sock *sk);
553 extern void tcp_reset(struct sock *sk);
556 extern void tcp_init_xmit_timers(struct sock *);
557 static inline void tcp_clear_xmit_timers(struct sock *sk)
559 inet_csk_clear_xmit_timers(sk);
562 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
563 extern unsigned int tcp_current_mss(struct sock *sk);
565 /* Bound MSS / TSO packet size with the half of the window */
566 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
570 /* When peer uses tiny windows, there is no use in packetizing
571 * to sub-MSS pieces for the sake of SWS or making sure there
572 * are enough packets in the pipe for fast recovery.
574 * On the other hand, for extremely large MSS devices, handling
575 * smaller than MSS windows in this way does make sense.
577 if (tp->max_window >= 512)
578 cutoff = (tp->max_window >> 1);
580 cutoff = tp->max_window;
582 if (cutoff && pktsize > cutoff)
583 return max_t(int, cutoff, 68U - tp->tcp_header_len);
589 extern void tcp_get_info(const struct sock *, struct tcp_info *);
591 /* Read 'sendfile()'-style from a TCP socket */
592 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
593 unsigned int, size_t);
594 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
595 sk_read_actor_t recv_actor);
597 extern void tcp_initialize_rcv_mss(struct sock *sk);
599 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
600 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
601 extern void tcp_mtup_init(struct sock *sk);
602 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
603 extern void tcp_init_buffer_space(struct sock *sk);
605 static inline void tcp_bound_rto(const struct sock *sk)
607 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
608 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
611 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
613 return (tp->srtt >> 3) + tp->rttvar;
616 extern void tcp_set_rto(struct sock *sk);
618 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
620 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
621 ntohl(TCP_FLAG_ACK) |
625 static inline void tcp_fast_path_on(struct tcp_sock *tp)
627 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
630 static inline void tcp_fast_path_check(struct sock *sk)
632 struct tcp_sock *tp = tcp_sk(sk);
634 if (skb_queue_empty(&tp->out_of_order_queue) &&
636 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
638 tcp_fast_path_on(tp);
641 /* Compute the actual rto_min value */
642 static inline u32 tcp_rto_min(struct sock *sk)
644 const struct dst_entry *dst = __sk_dst_get(sk);
645 u32 rto_min = TCP_RTO_MIN;
647 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
648 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
652 /* Compute the actual receive window we are currently advertising.
653 * Rcv_nxt can be after the window if our peer push more data
654 * than the offered window.
656 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
658 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
665 /* Choose a new window, without checks for shrinking, and without
666 * scaling applied to the result. The caller does these things
667 * if necessary. This is a "raw" window selection.
669 extern u32 __tcp_select_window(struct sock *sk);
671 void tcp_send_window_probe(struct sock *sk);
673 /* TCP timestamps are only 32-bits, this causes a slight
674 * complication on 64-bit systems since we store a snapshot
675 * of jiffies in the buffer control blocks below. We decided
676 * to use only the low 32-bits of jiffies and hide the ugly
677 * casts with the following macro.
679 #define tcp_time_stamp ((__u32)(jiffies))
681 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
683 #define TCPHDR_FIN 0x01
684 #define TCPHDR_SYN 0x02
685 #define TCPHDR_RST 0x04
686 #define TCPHDR_PSH 0x08
687 #define TCPHDR_ACK 0x10
688 #define TCPHDR_URG 0x20
689 #define TCPHDR_ECE 0x40
690 #define TCPHDR_CWR 0x80
692 /* This is what the send packet queuing engine uses to pass
693 * TCP per-packet control information to the transmission code.
694 * We also store the host-order sequence numbers in here too.
695 * This is 44 bytes if IPV6 is enabled.
696 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
700 struct inet_skb_parm h4;
701 #if IS_ENABLED(CONFIG_IPV6)
702 struct inet6_skb_parm h6;
704 } header; /* For incoming frames */
705 __u32 seq; /* Starting sequence number */
706 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
707 __u32 when; /* used to compute rtt's */
708 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
710 __u8 sacked; /* State flags for SACK/FACK. */
711 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
712 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
713 #define TCPCB_LOST 0x04 /* SKB is lost */
714 #define TCPCB_TAGBITS 0x07 /* All tag bits */
715 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
716 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
718 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
720 __u32 ack_seq; /* Sequence number ACK'd */
723 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
725 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
727 * If we receive a SYN packet with these bits set, it means a network is
728 * playing bad games with TOS bits. In order to avoid possible false congestion
729 * notifications, we disable TCP ECN negociation.
732 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
734 const struct tcphdr *th = tcp_hdr(skb);
736 if (sysctl_tcp_ecn && th->ece && th->cwr &&
737 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
738 inet_rsk(req)->ecn_ok = 1;
741 /* Due to TSO, an SKB can be composed of multiple actual
742 * packets. To keep these tracked properly, we use this.
744 static inline int tcp_skb_pcount(const struct sk_buff *skb)
746 return skb_shinfo(skb)->gso_segs;
749 /* This is valid iff tcp_skb_pcount() > 1. */
750 static inline int tcp_skb_mss(const struct sk_buff *skb)
752 return skb_shinfo(skb)->gso_size;
755 /* Events passed to congestion control interface */
757 CA_EVENT_TX_START, /* first transmit when no packets in flight */
758 CA_EVENT_CWND_RESTART, /* congestion window restart */
759 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
760 CA_EVENT_FRTO, /* fast recovery timeout */
761 CA_EVENT_LOSS, /* loss timeout */
762 CA_EVENT_FAST_ACK, /* in sequence ack */
763 CA_EVENT_SLOW_ACK, /* other ack */
767 * Interface for adding new TCP congestion control handlers
769 #define TCP_CA_NAME_MAX 16
770 #define TCP_CA_MAX 128
771 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
773 #define TCP_CONG_NON_RESTRICTED 0x1
774 #define TCP_CONG_RTT_STAMP 0x2
776 struct tcp_congestion_ops {
777 struct list_head list;
780 /* initialize private data (optional) */
781 void (*init)(struct sock *sk);
782 /* cleanup private data (optional) */
783 void (*release)(struct sock *sk);
785 /* return slow start threshold (required) */
786 u32 (*ssthresh)(struct sock *sk);
787 /* lower bound for congestion window (optional) */
788 u32 (*min_cwnd)(const struct sock *sk);
789 /* do new cwnd calculation (required) */
790 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
791 /* call before changing ca_state (optional) */
792 void (*set_state)(struct sock *sk, u8 new_state);
793 /* call when cwnd event occurs (optional) */
794 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
795 /* new value of cwnd after loss (optional) */
796 u32 (*undo_cwnd)(struct sock *sk);
797 /* hook for packet ack accounting (optional) */
798 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
799 /* get info for inet_diag (optional) */
800 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
802 char name[TCP_CA_NAME_MAX];
803 struct module *owner;
806 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
807 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
809 extern void tcp_init_congestion_control(struct sock *sk);
810 extern void tcp_cleanup_congestion_control(struct sock *sk);
811 extern int tcp_set_default_congestion_control(const char *name);
812 extern void tcp_get_default_congestion_control(char *name);
813 extern void tcp_get_available_congestion_control(char *buf, size_t len);
814 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
815 extern int tcp_set_allowed_congestion_control(char *allowed);
816 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
817 extern void tcp_slow_start(struct tcp_sock *tp);
818 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
820 extern struct tcp_congestion_ops tcp_init_congestion_ops;
821 extern u32 tcp_reno_ssthresh(struct sock *sk);
822 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
823 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
824 extern struct tcp_congestion_ops tcp_reno;
826 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
828 struct inet_connection_sock *icsk = inet_csk(sk);
830 if (icsk->icsk_ca_ops->set_state)
831 icsk->icsk_ca_ops->set_state(sk, ca_state);
832 icsk->icsk_ca_state = ca_state;
835 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
837 const struct inet_connection_sock *icsk = inet_csk(sk);
839 if (icsk->icsk_ca_ops->cwnd_event)
840 icsk->icsk_ca_ops->cwnd_event(sk, event);
843 /* These functions determine how the current flow behaves in respect of SACK
844 * handling. SACK is negotiated with the peer, and therefore it can vary
845 * between different flows.
847 * tcp_is_sack - SACK enabled
848 * tcp_is_reno - No SACK
849 * tcp_is_fack - FACK enabled, implies SACK enabled
851 static inline int tcp_is_sack(const struct tcp_sock *tp)
853 return tp->rx_opt.sack_ok;
856 static inline bool tcp_is_reno(const struct tcp_sock *tp)
858 return !tcp_is_sack(tp);
861 static inline bool tcp_is_fack(const struct tcp_sock *tp)
863 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
866 static inline void tcp_enable_fack(struct tcp_sock *tp)
868 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
871 /* TCP early-retransmit (ER) is similar to but more conservative than
872 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
874 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
876 tp->do_early_retrans = sysctl_tcp_early_retrans &&
877 !sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3;
878 tp->early_retrans_delayed = 0;
881 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
883 tp->do_early_retrans = 0;
886 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
888 return tp->sacked_out + tp->lost_out;
891 /* This determines how many packets are "in the network" to the best
892 * of our knowledge. In many cases it is conservative, but where
893 * detailed information is available from the receiver (via SACK
894 * blocks etc.) we can make more aggressive calculations.
896 * Use this for decisions involving congestion control, use just
897 * tp->packets_out to determine if the send queue is empty or not.
899 * Read this equation as:
901 * "Packets sent once on transmission queue" MINUS
902 * "Packets left network, but not honestly ACKed yet" PLUS
903 * "Packets fast retransmitted"
905 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
907 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
910 #define TCP_INFINITE_SSTHRESH 0x7fffffff
912 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
914 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
917 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
919 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
920 (1 << inet_csk(sk)->icsk_ca_state);
923 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
924 * The exception is cwnd reduction phase, when cwnd is decreasing towards
927 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
929 const struct tcp_sock *tp = tcp_sk(sk);
931 if (tcp_in_cwnd_reduction(sk))
932 return tp->snd_ssthresh;
934 return max(tp->snd_ssthresh,
935 ((tp->snd_cwnd >> 1) +
936 (tp->snd_cwnd >> 2)));
939 /* Use define here intentionally to get WARN_ON location shown at the caller */
940 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
942 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
943 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
945 /* The maximum number of MSS of available cwnd for which TSO defers
946 * sending if not using sysctl_tcp_tso_win_divisor.
948 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
953 /* Slow start with delack produces 3 packets of burst, so that
954 * it is safe "de facto". This will be the default - same as
955 * the default reordering threshold - but if reordering increases,
956 * we must be able to allow cwnd to burst at least this much in order
957 * to not pull it back when holes are filled.
959 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
961 return tp->reordering;
964 /* Returns end sequence number of the receiver's advertised window */
965 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
967 return tp->snd_una + tp->snd_wnd;
969 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
971 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
972 const struct sk_buff *skb)
975 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
978 static inline void tcp_check_probe_timer(struct sock *sk)
980 const struct tcp_sock *tp = tcp_sk(sk);
981 const struct inet_connection_sock *icsk = inet_csk(sk);
983 if (!tp->packets_out && !icsk->icsk_pending)
984 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
985 icsk->icsk_rto, TCP_RTO_MAX);
988 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
993 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
999 * Calculate(/check) TCP checksum
1001 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1002 __be32 daddr, __wsum base)
1004 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1007 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1009 return __skb_checksum_complete(skb);
1012 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1014 return !skb_csum_unnecessary(skb) &&
1015 __tcp_checksum_complete(skb);
1018 /* Prequeue for VJ style copy to user, combined with checksumming. */
1020 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1022 tp->ucopy.task = NULL;
1024 tp->ucopy.memory = 0;
1025 skb_queue_head_init(&tp->ucopy.prequeue);
1026 #ifdef CONFIG_NET_DMA
1027 tp->ucopy.dma_chan = NULL;
1028 tp->ucopy.wakeup = 0;
1029 tp->ucopy.pinned_list = NULL;
1030 tp->ucopy.dma_cookie = 0;
1034 /* Packet is added to VJ-style prequeue for processing in process
1035 * context, if a reader task is waiting. Apparently, this exciting
1036 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1037 * failed somewhere. Latency? Burstiness? Well, at least now we will
1038 * see, why it failed. 8)8) --ANK
1040 * NOTE: is this not too big to inline?
1042 static inline bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1044 struct tcp_sock *tp = tcp_sk(sk);
1046 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1049 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1050 tp->ucopy.memory += skb->truesize;
1051 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1052 struct sk_buff *skb1;
1054 BUG_ON(sock_owned_by_user(sk));
1056 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1057 sk_backlog_rcv(sk, skb1);
1058 NET_INC_STATS_BH(sock_net(sk),
1059 LINUX_MIB_TCPPREQUEUEDROPPED);
1062 tp->ucopy.memory = 0;
1063 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1064 wake_up_interruptible_sync_poll(sk_sleep(sk),
1065 POLLIN | POLLRDNORM | POLLRDBAND);
1066 if (!inet_csk_ack_scheduled(sk))
1067 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1068 (3 * tcp_rto_min(sk)) / 4,
1078 static const char *statename[]={
1079 "Unused","Established","Syn Sent","Syn Recv",
1080 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1081 "Close Wait","Last ACK","Listen","Closing"
1084 extern void tcp_set_state(struct sock *sk, int state);
1086 extern void tcp_done(struct sock *sk);
1088 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1091 rx_opt->num_sacks = 0;
1094 /* Determine a window scaling and initial window to offer. */
1095 extern void tcp_select_initial_window(int __space, __u32 mss,
1096 __u32 *rcv_wnd, __u32 *window_clamp,
1097 int wscale_ok, __u8 *rcv_wscale,
1098 __u32 init_rcv_wnd);
1100 static inline int tcp_win_from_space(int space)
1102 return sysctl_tcp_adv_win_scale<=0 ?
1103 (space>>(-sysctl_tcp_adv_win_scale)) :
1104 space - (space>>sysctl_tcp_adv_win_scale);
1107 /* Note: caller must be prepared to deal with negative returns */
1108 static inline int tcp_space(const struct sock *sk)
1110 return tcp_win_from_space(sk->sk_rcvbuf -
1111 atomic_read(&sk->sk_rmem_alloc));
1114 static inline int tcp_full_space(const struct sock *sk)
1116 return tcp_win_from_space(sk->sk_rcvbuf);
1119 static inline void tcp_openreq_init(struct request_sock *req,
1120 struct tcp_options_received *rx_opt,
1121 struct sk_buff *skb)
1123 struct inet_request_sock *ireq = inet_rsk(req);
1125 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1127 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1128 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1129 tcp_rsk(req)->snt_synack = 0;
1130 req->mss = rx_opt->mss_clamp;
1131 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1132 ireq->tstamp_ok = rx_opt->tstamp_ok;
1133 ireq->sack_ok = rx_opt->sack_ok;
1134 ireq->snd_wscale = rx_opt->snd_wscale;
1135 ireq->wscale_ok = rx_opt->wscale_ok;
1138 ireq->rmt_port = tcp_hdr(skb)->source;
1139 ireq->loc_port = tcp_hdr(skb)->dest;
1142 /* Compute time elapsed between SYNACK and the ACK completing 3WHS */
1143 static inline void tcp_synack_rtt_meas(struct sock *sk,
1144 struct request_sock *req)
1146 if (tcp_rsk(req)->snt_synack)
1147 tcp_valid_rtt_meas(sk,
1148 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1151 extern void tcp_enter_memory_pressure(struct sock *sk);
1153 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1155 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1158 static inline int keepalive_time_when(const struct tcp_sock *tp)
1160 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1163 static inline int keepalive_probes(const struct tcp_sock *tp)
1165 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1168 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1170 const struct inet_connection_sock *icsk = &tp->inet_conn;
1172 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1173 tcp_time_stamp - tp->rcv_tstamp);
1176 static inline int tcp_fin_time(const struct sock *sk)
1178 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1179 const int rto = inet_csk(sk)->icsk_rto;
1181 if (fin_timeout < (rto << 2) - (rto >> 1))
1182 fin_timeout = (rto << 2) - (rto >> 1);
1187 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1190 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1192 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1195 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1196 * then following tcp messages have valid values. Ignore 0 value,
1197 * or else 'negative' tsval might forbid us to accept their packets.
1199 if (!rx_opt->ts_recent)
1204 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1207 if (tcp_paws_check(rx_opt, 0))
1210 /* RST segments are not recommended to carry timestamp,
1211 and, if they do, it is recommended to ignore PAWS because
1212 "their cleanup function should take precedence over timestamps."
1213 Certainly, it is mistake. It is necessary to understand the reasons
1214 of this constraint to relax it: if peer reboots, clock may go
1215 out-of-sync and half-open connections will not be reset.
1216 Actually, the problem would be not existing if all
1217 the implementations followed draft about maintaining clock
1218 via reboots. Linux-2.2 DOES NOT!
1220 However, we can relax time bounds for RST segments to MSL.
1222 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1227 static inline void tcp_mib_init(struct net *net)
1230 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1231 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1232 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1233 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1237 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1239 tp->lost_skb_hint = NULL;
1240 tp->scoreboard_skb_hint = NULL;
1243 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1245 tcp_clear_retrans_hints_partial(tp);
1246 tp->retransmit_skb_hint = NULL;
1252 union tcp_md5_addr {
1254 #if IS_ENABLED(CONFIG_IPV6)
1259 /* - key database */
1260 struct tcp_md5sig_key {
1261 struct hlist_node node;
1263 u8 family; /* AF_INET or AF_INET6 */
1264 union tcp_md5_addr addr;
1265 u8 key[TCP_MD5SIG_MAXKEYLEN];
1266 struct rcu_head rcu;
1270 struct tcp_md5sig_info {
1271 struct hlist_head head;
1272 struct rcu_head rcu;
1275 /* - pseudo header */
1276 struct tcp4_pseudohdr {
1284 struct tcp6_pseudohdr {
1285 struct in6_addr saddr;
1286 struct in6_addr daddr;
1288 __be32 protocol; /* including padding */
1291 union tcp_md5sum_block {
1292 struct tcp4_pseudohdr ip4;
1293 #if IS_ENABLED(CONFIG_IPV6)
1294 struct tcp6_pseudohdr ip6;
1298 /* - pool: digest algorithm, hash description and scratch buffer */
1299 struct tcp_md5sig_pool {
1300 struct hash_desc md5_desc;
1301 union tcp_md5sum_block md5_blk;
1305 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1306 const struct sock *sk,
1307 const struct request_sock *req,
1308 const struct sk_buff *skb);
1309 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1310 int family, const u8 *newkey,
1311 u8 newkeylen, gfp_t gfp);
1312 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1314 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1315 struct sock *addr_sk);
1317 #ifdef CONFIG_TCP_MD5SIG
1318 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1319 const union tcp_md5_addr *addr, int family);
1320 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1322 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1323 const union tcp_md5_addr *addr,
1328 #define tcp_twsk_md5_key(twsk) NULL
1331 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1332 extern void tcp_free_md5sig_pool(void);
1334 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1335 extern void tcp_put_md5sig_pool(void);
1337 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1338 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1339 unsigned int header_len);
1340 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1341 const struct tcp_md5sig_key *key);
1343 /* From tcp_fastopen.c */
1344 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1345 struct tcp_fastopen_cookie *cookie,
1346 int *syn_loss, unsigned long *last_syn_loss);
1347 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1348 struct tcp_fastopen_cookie *cookie,
1350 struct tcp_fastopen_request {
1351 /* Fast Open cookie. Size 0 means a cookie request */
1352 struct tcp_fastopen_cookie cookie;
1353 struct msghdr *data; /* data in MSG_FASTOPEN */
1354 u16 copied; /* queued in tcp_connect() */
1356 void tcp_free_fastopen_req(struct tcp_sock *tp);
1358 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1359 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1360 void tcp_fastopen_cookie_gen(__be32 addr, struct tcp_fastopen_cookie *foc);
1362 #define TCP_FASTOPEN_KEY_LENGTH 16
1364 /* Fastopen key context */
1365 struct tcp_fastopen_context {
1366 struct crypto_cipher __rcu *tfm;
1367 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1368 struct rcu_head rcu;
1371 /* write queue abstraction */
1372 static inline void tcp_write_queue_purge(struct sock *sk)
1374 struct sk_buff *skb;
1376 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1377 sk_wmem_free_skb(sk, skb);
1379 tcp_clear_all_retrans_hints(tcp_sk(sk));
1382 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1384 return skb_peek(&sk->sk_write_queue);
1387 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1389 return skb_peek_tail(&sk->sk_write_queue);
1392 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1393 const struct sk_buff *skb)
1395 return skb_queue_next(&sk->sk_write_queue, skb);
1398 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1399 const struct sk_buff *skb)
1401 return skb_queue_prev(&sk->sk_write_queue, skb);
1404 #define tcp_for_write_queue(skb, sk) \
1405 skb_queue_walk(&(sk)->sk_write_queue, skb)
1407 #define tcp_for_write_queue_from(skb, sk) \
1408 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1410 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1411 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1413 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1415 return sk->sk_send_head;
1418 static inline bool tcp_skb_is_last(const struct sock *sk,
1419 const struct sk_buff *skb)
1421 return skb_queue_is_last(&sk->sk_write_queue, skb);
1424 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1426 if (tcp_skb_is_last(sk, skb))
1427 sk->sk_send_head = NULL;
1429 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1432 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1434 if (sk->sk_send_head == skb_unlinked)
1435 sk->sk_send_head = NULL;
1438 static inline void tcp_init_send_head(struct sock *sk)
1440 sk->sk_send_head = NULL;
1443 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1445 __skb_queue_tail(&sk->sk_write_queue, skb);
1448 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1450 __tcp_add_write_queue_tail(sk, skb);
1452 /* Queue it, remembering where we must start sending. */
1453 if (sk->sk_send_head == NULL) {
1454 sk->sk_send_head = skb;
1456 if (tcp_sk(sk)->highest_sack == NULL)
1457 tcp_sk(sk)->highest_sack = skb;
1461 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1463 __skb_queue_head(&sk->sk_write_queue, skb);
1466 /* Insert buff after skb on the write queue of sk. */
1467 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1468 struct sk_buff *buff,
1471 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1474 /* Insert new before skb on the write queue of sk. */
1475 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1476 struct sk_buff *skb,
1479 __skb_queue_before(&sk->sk_write_queue, skb, new);
1481 if (sk->sk_send_head == skb)
1482 sk->sk_send_head = new;
1485 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1487 __skb_unlink(skb, &sk->sk_write_queue);
1490 static inline bool tcp_write_queue_empty(struct sock *sk)
1492 return skb_queue_empty(&sk->sk_write_queue);
1495 static inline void tcp_push_pending_frames(struct sock *sk)
1497 if (tcp_send_head(sk)) {
1498 struct tcp_sock *tp = tcp_sk(sk);
1500 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1504 /* Start sequence of the skb just after the highest skb with SACKed
1505 * bit, valid only if sacked_out > 0 or when the caller has ensured
1506 * validity by itself.
1508 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1510 if (!tp->sacked_out)
1513 if (tp->highest_sack == NULL)
1516 return TCP_SKB_CB(tp->highest_sack)->seq;
1519 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1521 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1522 tcp_write_queue_next(sk, skb);
1525 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1527 return tcp_sk(sk)->highest_sack;
1530 static inline void tcp_highest_sack_reset(struct sock *sk)
1532 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1535 /* Called when old skb is about to be deleted (to be combined with new skb) */
1536 static inline void tcp_highest_sack_combine(struct sock *sk,
1537 struct sk_buff *old,
1538 struct sk_buff *new)
1540 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1541 tcp_sk(sk)->highest_sack = new;
1544 /* Determines whether this is a thin stream (which may suffer from
1545 * increased latency). Used to trigger latency-reducing mechanisms.
1547 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1549 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1553 enum tcp_seq_states {
1554 TCP_SEQ_STATE_LISTENING,
1555 TCP_SEQ_STATE_OPENREQ,
1556 TCP_SEQ_STATE_ESTABLISHED,
1557 TCP_SEQ_STATE_TIME_WAIT,
1560 int tcp_seq_open(struct inode *inode, struct file *file);
1562 struct tcp_seq_afinfo {
1565 const struct file_operations *seq_fops;
1566 struct seq_operations seq_ops;
1569 struct tcp_iter_state {
1570 struct seq_net_private p;
1572 enum tcp_seq_states state;
1573 struct sock *syn_wait_sk;
1574 int bucket, offset, sbucket, num;
1579 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1580 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1582 extern struct request_sock_ops tcp_request_sock_ops;
1583 extern struct request_sock_ops tcp6_request_sock_ops;
1585 extern void tcp_v4_destroy_sock(struct sock *sk);
1587 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1588 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1589 netdev_features_t features);
1590 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1591 struct sk_buff *skb);
1592 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1593 struct sk_buff *skb);
1594 extern int tcp_gro_complete(struct sk_buff *skb);
1595 extern int tcp4_gro_complete(struct sk_buff *skb);
1597 #ifdef CONFIG_PROC_FS
1598 extern int tcp4_proc_init(void);
1599 extern void tcp4_proc_exit(void);
1602 /* TCP af-specific functions */
1603 struct tcp_sock_af_ops {
1604 #ifdef CONFIG_TCP_MD5SIG
1605 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1606 struct sock *addr_sk);
1607 int (*calc_md5_hash) (char *location,
1608 struct tcp_md5sig_key *md5,
1609 const struct sock *sk,
1610 const struct request_sock *req,
1611 const struct sk_buff *skb);
1612 int (*md5_parse) (struct sock *sk,
1613 char __user *optval,
1618 struct tcp_request_sock_ops {
1619 #ifdef CONFIG_TCP_MD5SIG
1620 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1621 struct request_sock *req);
1622 int (*calc_md5_hash) (char *location,
1623 struct tcp_md5sig_key *md5,
1624 const struct sock *sk,
1625 const struct request_sock *req,
1626 const struct sk_buff *skb);
1630 /* Using SHA1 for now, define some constants.
1632 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1633 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1634 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1636 extern int tcp_cookie_generator(u32 *bakery);
1639 * struct tcp_cookie_values - each socket needs extra space for the
1640 * cookies, together with (optional) space for any SYN data.
1642 * A tcp_sock contains a pointer to the current value, and this is
1643 * cloned to the tcp_timewait_sock.
1645 * @cookie_pair: variable data from the option exchange.
1647 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1648 * indicates default (sysctl_tcp_cookie_size).
1649 * After cookie sent, remembers size of cookie.
1650 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1652 * @s_data_desired: user specified tcpct_s_data_desired. When the
1653 * constant payload is specified (@s_data_constant),
1654 * holds its length instead.
1655 * Range 0 to TCP_MSS_DESIRED.
1657 * @s_data_payload: constant data that is to be included in the
1658 * payload of SYN or SYNACK segments when the
1659 * cookie option is present.
1661 struct tcp_cookie_values {
1663 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1664 u8 cookie_pair_size;
1666 u16 s_data_desired:11,
1671 u8 s_data_payload[0];
1674 static inline void tcp_cookie_values_release(struct kref *kref)
1676 kfree(container_of(kref, struct tcp_cookie_values, kref));
1679 /* The length of constant payload data. Note that s_data_desired is
1680 * overloaded, depending on s_data_constant: either the length of constant
1681 * data (returned here) or the limit on variable data.
1683 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1685 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1686 ? tp->cookie_values->s_data_desired
1691 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1693 * As tcp_request_sock has already been extended in other places, the
1694 * only remaining method is to pass stack values along as function
1695 * parameters. These parameters are not needed after sending SYNACK.
1697 * @cookie_bakery: cryptographic secret and message workspace.
1699 * @cookie_plus: bytes in authenticator/cookie option, copied from
1700 * struct tcp_options_received (above).
1702 struct tcp_extend_values {
1703 struct request_values rv;
1704 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1710 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1712 return (struct tcp_extend_values *)rvp;
1715 extern void tcp_v4_init(void);
1716 extern void tcp_init(void);