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 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
373 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
375 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
376 size_t size, int flags);
377 extern void tcp_release_cb(struct sock *sk);
378 extern void tcp_write_timer_handler(struct sock *sk);
379 extern void tcp_delack_timer_handler(struct sock *sk);
380 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
381 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
382 const struct tcphdr *th, unsigned int len);
383 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
384 const struct tcphdr *th, unsigned int len);
385 extern void tcp_rcv_space_adjust(struct sock *sk);
386 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
387 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
388 extern void tcp_twsk_destructor(struct sock *sk);
389 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
390 struct pipe_inode_info *pipe, size_t len,
393 static inline void tcp_dec_quickack_mode(struct sock *sk,
394 const unsigned int pkts)
396 struct inet_connection_sock *icsk = inet_csk(sk);
398 if (icsk->icsk_ack.quick) {
399 if (pkts >= icsk->icsk_ack.quick) {
400 icsk->icsk_ack.quick = 0;
401 /* Leaving quickack mode we deflate ATO. */
402 icsk->icsk_ack.ato = TCP_ATO_MIN;
404 icsk->icsk_ack.quick -= pkts;
409 #define TCP_ECN_QUEUE_CWR 2
410 #define TCP_ECN_DEMAND_CWR 4
411 #define TCP_ECN_SEEN 8
421 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
423 const struct tcphdr *th);
424 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
425 struct request_sock *req,
426 struct request_sock **prev,
428 extern int tcp_child_process(struct sock *parent, struct sock *child,
429 struct sk_buff *skb);
430 extern bool tcp_use_frto(struct sock *sk);
431 extern void tcp_enter_frto(struct sock *sk);
432 extern void tcp_enter_loss(struct sock *sk, int how);
433 extern void tcp_clear_retrans(struct tcp_sock *tp);
434 extern void tcp_update_metrics(struct sock *sk);
435 extern void tcp_init_metrics(struct sock *sk);
436 extern void tcp_metrics_init(void);
437 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check);
438 extern bool tcp_remember_stamp(struct sock *sk);
439 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
440 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
441 extern void tcp_disable_fack(struct tcp_sock *tp);
442 extern void tcp_close(struct sock *sk, long timeout);
443 extern void tcp_init_sock(struct sock *sk);
444 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
445 struct poll_table_struct *wait);
446 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
447 char __user *optval, int __user *optlen);
448 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
449 char __user *optval, unsigned int optlen);
450 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
451 char __user *optval, int __user *optlen);
452 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
453 char __user *optval, unsigned int optlen);
454 extern void tcp_set_keepalive(struct sock *sk, int val);
455 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
456 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
457 size_t len, int nonblock, int flags, int *addr_len);
458 extern void tcp_parse_options(const struct sk_buff *skb,
459 struct tcp_options_received *opt_rx, const u8 **hvpp,
460 int estab, struct tcp_fastopen_cookie *foc);
461 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
464 * TCP v4 functions exported for the inet6 API
467 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
468 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
469 extern struct sock * tcp_create_openreq_child(struct sock *sk,
470 struct request_sock *req,
471 struct sk_buff *skb);
472 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
473 struct request_sock *req,
474 struct dst_entry *dst);
475 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
476 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
478 extern int tcp_connect(struct sock *sk);
479 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
480 struct request_sock *req,
481 struct request_values *rvp,
482 struct tcp_fastopen_cookie *foc);
483 extern int tcp_disconnect(struct sock *sk, int flags);
485 void tcp_connect_init(struct sock *sk);
486 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
487 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
488 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
490 /* From syncookies.c */
491 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
492 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
493 struct ip_options *opt);
494 #ifdef CONFIG_SYN_COOKIES
495 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
498 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
506 extern __u32 cookie_init_timestamp(struct request_sock *req);
507 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
509 /* From net/ipv6/syncookies.c */
510 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
511 #ifdef CONFIG_SYN_COOKIES
512 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
515 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
524 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
526 extern bool tcp_may_send_now(struct sock *sk);
527 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
528 extern void tcp_retransmit_timer(struct sock *sk);
529 extern void tcp_xmit_retransmit_queue(struct sock *);
530 extern void tcp_simple_retransmit(struct sock *);
531 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
532 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
534 extern void tcp_send_probe0(struct sock *);
535 extern void tcp_send_partial(struct sock *);
536 extern int tcp_write_wakeup(struct sock *);
537 extern void tcp_send_fin(struct sock *sk);
538 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
539 extern int tcp_send_synack(struct sock *);
540 extern bool tcp_syn_flood_action(struct sock *sk,
541 const struct sk_buff *skb,
543 extern void tcp_push_one(struct sock *, unsigned int mss_now);
544 extern void tcp_send_ack(struct sock *sk);
545 extern void tcp_send_delayed_ack(struct sock *sk);
548 extern void tcp_cwnd_application_limited(struct sock *sk);
549 extern void tcp_resume_early_retransmit(struct sock *sk);
550 extern void tcp_rearm_rto(struct sock *sk);
551 extern void tcp_reset(struct sock *sk);
554 extern void tcp_init_xmit_timers(struct sock *);
555 static inline void tcp_clear_xmit_timers(struct sock *sk)
557 inet_csk_clear_xmit_timers(sk);
560 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
561 extern unsigned int tcp_current_mss(struct sock *sk);
563 /* Bound MSS / TSO packet size with the half of the window */
564 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
568 /* When peer uses tiny windows, there is no use in packetizing
569 * to sub-MSS pieces for the sake of SWS or making sure there
570 * are enough packets in the pipe for fast recovery.
572 * On the other hand, for extremely large MSS devices, handling
573 * smaller than MSS windows in this way does make sense.
575 if (tp->max_window >= 512)
576 cutoff = (tp->max_window >> 1);
578 cutoff = tp->max_window;
580 if (cutoff && pktsize > cutoff)
581 return max_t(int, cutoff, 68U - tp->tcp_header_len);
587 extern void tcp_get_info(const struct sock *, struct tcp_info *);
589 /* Read 'sendfile()'-style from a TCP socket */
590 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
591 unsigned int, size_t);
592 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
593 sk_read_actor_t recv_actor);
595 extern void tcp_initialize_rcv_mss(struct sock *sk);
597 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
598 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
599 extern void tcp_mtup_init(struct sock *sk);
600 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
601 extern void tcp_init_buffer_space(struct sock *sk);
603 static inline void tcp_bound_rto(const struct sock *sk)
605 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
606 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
609 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
611 return (tp->srtt >> 3) + tp->rttvar;
614 extern void tcp_set_rto(struct sock *sk);
616 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
618 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
619 ntohl(TCP_FLAG_ACK) |
623 static inline void tcp_fast_path_on(struct tcp_sock *tp)
625 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
628 static inline void tcp_fast_path_check(struct sock *sk)
630 struct tcp_sock *tp = tcp_sk(sk);
632 if (skb_queue_empty(&tp->out_of_order_queue) &&
634 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
636 tcp_fast_path_on(tp);
639 /* Compute the actual rto_min value */
640 static inline u32 tcp_rto_min(struct sock *sk)
642 const struct dst_entry *dst = __sk_dst_get(sk);
643 u32 rto_min = TCP_RTO_MIN;
645 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
646 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
650 /* Compute the actual receive window we are currently advertising.
651 * Rcv_nxt can be after the window if our peer push more data
652 * than the offered window.
654 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
656 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
663 /* Choose a new window, without checks for shrinking, and without
664 * scaling applied to the result. The caller does these things
665 * if necessary. This is a "raw" window selection.
667 extern u32 __tcp_select_window(struct sock *sk);
669 void tcp_send_window_probe(struct sock *sk);
671 /* TCP timestamps are only 32-bits, this causes a slight
672 * complication on 64-bit systems since we store a snapshot
673 * of jiffies in the buffer control blocks below. We decided
674 * to use only the low 32-bits of jiffies and hide the ugly
675 * casts with the following macro.
677 #define tcp_time_stamp ((__u32)(jiffies))
679 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
681 #define TCPHDR_FIN 0x01
682 #define TCPHDR_SYN 0x02
683 #define TCPHDR_RST 0x04
684 #define TCPHDR_PSH 0x08
685 #define TCPHDR_ACK 0x10
686 #define TCPHDR_URG 0x20
687 #define TCPHDR_ECE 0x40
688 #define TCPHDR_CWR 0x80
690 /* This is what the send packet queuing engine uses to pass
691 * TCP per-packet control information to the transmission code.
692 * We also store the host-order sequence numbers in here too.
693 * This is 44 bytes if IPV6 is enabled.
694 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
698 struct inet_skb_parm h4;
699 #if IS_ENABLED(CONFIG_IPV6)
700 struct inet6_skb_parm h6;
702 } header; /* For incoming frames */
703 __u32 seq; /* Starting sequence number */
704 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
705 __u32 when; /* used to compute rtt's */
706 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
708 __u8 sacked; /* State flags for SACK/FACK. */
709 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
710 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
711 #define TCPCB_LOST 0x04 /* SKB is lost */
712 #define TCPCB_TAGBITS 0x07 /* All tag bits */
713 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
714 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
716 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
718 __u32 ack_seq; /* Sequence number ACK'd */
721 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
723 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
725 * If we receive a SYN packet with these bits set, it means a network is
726 * playing bad games with TOS bits. In order to avoid possible false congestion
727 * notifications, we disable TCP ECN negociation.
730 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
732 const struct tcphdr *th = tcp_hdr(skb);
734 if (sysctl_tcp_ecn && th->ece && th->cwr &&
735 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
736 inet_rsk(req)->ecn_ok = 1;
739 /* Due to TSO, an SKB can be composed of multiple actual
740 * packets. To keep these tracked properly, we use this.
742 static inline int tcp_skb_pcount(const struct sk_buff *skb)
744 return skb_shinfo(skb)->gso_segs;
747 /* This is valid iff tcp_skb_pcount() > 1. */
748 static inline int tcp_skb_mss(const struct sk_buff *skb)
750 return skb_shinfo(skb)->gso_size;
753 /* Events passed to congestion control interface */
755 CA_EVENT_TX_START, /* first transmit when no packets in flight */
756 CA_EVENT_CWND_RESTART, /* congestion window restart */
757 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
758 CA_EVENT_FRTO, /* fast recovery timeout */
759 CA_EVENT_LOSS, /* loss timeout */
760 CA_EVENT_FAST_ACK, /* in sequence ack */
761 CA_EVENT_SLOW_ACK, /* other ack */
765 * Interface for adding new TCP congestion control handlers
767 #define TCP_CA_NAME_MAX 16
768 #define TCP_CA_MAX 128
769 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
771 #define TCP_CONG_NON_RESTRICTED 0x1
772 #define TCP_CONG_RTT_STAMP 0x2
774 struct tcp_congestion_ops {
775 struct list_head list;
778 /* initialize private data (optional) */
779 void (*init)(struct sock *sk);
780 /* cleanup private data (optional) */
781 void (*release)(struct sock *sk);
783 /* return slow start threshold (required) */
784 u32 (*ssthresh)(struct sock *sk);
785 /* lower bound for congestion window (optional) */
786 u32 (*min_cwnd)(const struct sock *sk);
787 /* do new cwnd calculation (required) */
788 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
789 /* call before changing ca_state (optional) */
790 void (*set_state)(struct sock *sk, u8 new_state);
791 /* call when cwnd event occurs (optional) */
792 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
793 /* new value of cwnd after loss (optional) */
794 u32 (*undo_cwnd)(struct sock *sk);
795 /* hook for packet ack accounting (optional) */
796 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
797 /* get info for inet_diag (optional) */
798 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
800 char name[TCP_CA_NAME_MAX];
801 struct module *owner;
804 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
805 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
807 extern void tcp_init_congestion_control(struct sock *sk);
808 extern void tcp_cleanup_congestion_control(struct sock *sk);
809 extern int tcp_set_default_congestion_control(const char *name);
810 extern void tcp_get_default_congestion_control(char *name);
811 extern void tcp_get_available_congestion_control(char *buf, size_t len);
812 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
813 extern int tcp_set_allowed_congestion_control(char *allowed);
814 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
815 extern void tcp_slow_start(struct tcp_sock *tp);
816 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
818 extern struct tcp_congestion_ops tcp_init_congestion_ops;
819 extern u32 tcp_reno_ssthresh(struct sock *sk);
820 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
821 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
822 extern struct tcp_congestion_ops tcp_reno;
824 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
826 struct inet_connection_sock *icsk = inet_csk(sk);
828 if (icsk->icsk_ca_ops->set_state)
829 icsk->icsk_ca_ops->set_state(sk, ca_state);
830 icsk->icsk_ca_state = ca_state;
833 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
835 const struct inet_connection_sock *icsk = inet_csk(sk);
837 if (icsk->icsk_ca_ops->cwnd_event)
838 icsk->icsk_ca_ops->cwnd_event(sk, event);
841 /* These functions determine how the current flow behaves in respect of SACK
842 * handling. SACK is negotiated with the peer, and therefore it can vary
843 * between different flows.
845 * tcp_is_sack - SACK enabled
846 * tcp_is_reno - No SACK
847 * tcp_is_fack - FACK enabled, implies SACK enabled
849 static inline int tcp_is_sack(const struct tcp_sock *tp)
851 return tp->rx_opt.sack_ok;
854 static inline bool tcp_is_reno(const struct tcp_sock *tp)
856 return !tcp_is_sack(tp);
859 static inline bool tcp_is_fack(const struct tcp_sock *tp)
861 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
864 static inline void tcp_enable_fack(struct tcp_sock *tp)
866 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
869 /* TCP early-retransmit (ER) is similar to but more conservative than
870 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
872 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
874 tp->do_early_retrans = sysctl_tcp_early_retrans &&
875 !sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3;
876 tp->early_retrans_delayed = 0;
879 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
881 tp->do_early_retrans = 0;
884 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
886 return tp->sacked_out + tp->lost_out;
889 /* This determines how many packets are "in the network" to the best
890 * of our knowledge. In many cases it is conservative, but where
891 * detailed information is available from the receiver (via SACK
892 * blocks etc.) we can make more aggressive calculations.
894 * Use this for decisions involving congestion control, use just
895 * tp->packets_out to determine if the send queue is empty or not.
897 * Read this equation as:
899 * "Packets sent once on transmission queue" MINUS
900 * "Packets left network, but not honestly ACKed yet" PLUS
901 * "Packets fast retransmitted"
903 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
905 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
908 #define TCP_INFINITE_SSTHRESH 0x7fffffff
910 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
912 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
915 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
917 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
918 (1 << inet_csk(sk)->icsk_ca_state);
921 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
922 * The exception is cwnd reduction phase, when cwnd is decreasing towards
925 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
927 const struct tcp_sock *tp = tcp_sk(sk);
929 if (tcp_in_cwnd_reduction(sk))
930 return tp->snd_ssthresh;
932 return max(tp->snd_ssthresh,
933 ((tp->snd_cwnd >> 1) +
934 (tp->snd_cwnd >> 2)));
937 /* Use define here intentionally to get WARN_ON location shown at the caller */
938 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
940 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
941 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
943 /* The maximum number of MSS of available cwnd for which TSO defers
944 * sending if not using sysctl_tcp_tso_win_divisor.
946 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
951 /* Slow start with delack produces 3 packets of burst, so that
952 * it is safe "de facto". This will be the default - same as
953 * the default reordering threshold - but if reordering increases,
954 * we must be able to allow cwnd to burst at least this much in order
955 * to not pull it back when holes are filled.
957 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
959 return tp->reordering;
962 /* Returns end sequence number of the receiver's advertised window */
963 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
965 return tp->snd_una + tp->snd_wnd;
967 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
969 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
970 const struct sk_buff *skb)
973 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
976 static inline void tcp_check_probe_timer(struct sock *sk)
978 const struct tcp_sock *tp = tcp_sk(sk);
979 const struct inet_connection_sock *icsk = inet_csk(sk);
981 if (!tp->packets_out && !icsk->icsk_pending)
982 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
983 icsk->icsk_rto, TCP_RTO_MAX);
986 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
991 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
997 * Calculate(/check) TCP checksum
999 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1000 __be32 daddr, __wsum base)
1002 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1005 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1007 return __skb_checksum_complete(skb);
1010 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1012 return !skb_csum_unnecessary(skb) &&
1013 __tcp_checksum_complete(skb);
1016 /* Prequeue for VJ style copy to user, combined with checksumming. */
1018 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1020 tp->ucopy.task = NULL;
1022 tp->ucopy.memory = 0;
1023 skb_queue_head_init(&tp->ucopy.prequeue);
1024 #ifdef CONFIG_NET_DMA
1025 tp->ucopy.dma_chan = NULL;
1026 tp->ucopy.wakeup = 0;
1027 tp->ucopy.pinned_list = NULL;
1028 tp->ucopy.dma_cookie = 0;
1032 /* Packet is added to VJ-style prequeue for processing in process
1033 * context, if a reader task is waiting. Apparently, this exciting
1034 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1035 * failed somewhere. Latency? Burstiness? Well, at least now we will
1036 * see, why it failed. 8)8) --ANK
1038 * NOTE: is this not too big to inline?
1040 static inline bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1042 struct tcp_sock *tp = tcp_sk(sk);
1044 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1047 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1048 tp->ucopy.memory += skb->truesize;
1049 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1050 struct sk_buff *skb1;
1052 BUG_ON(sock_owned_by_user(sk));
1054 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1055 sk_backlog_rcv(sk, skb1);
1056 NET_INC_STATS_BH(sock_net(sk),
1057 LINUX_MIB_TCPPREQUEUEDROPPED);
1060 tp->ucopy.memory = 0;
1061 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1062 wake_up_interruptible_sync_poll(sk_sleep(sk),
1063 POLLIN | POLLRDNORM | POLLRDBAND);
1064 if (!inet_csk_ack_scheduled(sk))
1065 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1066 (3 * tcp_rto_min(sk)) / 4,
1076 static const char *statename[]={
1077 "Unused","Established","Syn Sent","Syn Recv",
1078 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1079 "Close Wait","Last ACK","Listen","Closing"
1082 extern void tcp_set_state(struct sock *sk, int state);
1084 extern void tcp_done(struct sock *sk);
1086 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1089 rx_opt->num_sacks = 0;
1092 /* Determine a window scaling and initial window to offer. */
1093 extern void tcp_select_initial_window(int __space, __u32 mss,
1094 __u32 *rcv_wnd, __u32 *window_clamp,
1095 int wscale_ok, __u8 *rcv_wscale,
1096 __u32 init_rcv_wnd);
1098 static inline int tcp_win_from_space(int space)
1100 return sysctl_tcp_adv_win_scale<=0 ?
1101 (space>>(-sysctl_tcp_adv_win_scale)) :
1102 space - (space>>sysctl_tcp_adv_win_scale);
1105 /* Note: caller must be prepared to deal with negative returns */
1106 static inline int tcp_space(const struct sock *sk)
1108 return tcp_win_from_space(sk->sk_rcvbuf -
1109 atomic_read(&sk->sk_rmem_alloc));
1112 static inline int tcp_full_space(const struct sock *sk)
1114 return tcp_win_from_space(sk->sk_rcvbuf);
1117 static inline void tcp_openreq_init(struct request_sock *req,
1118 struct tcp_options_received *rx_opt,
1119 struct sk_buff *skb)
1121 struct inet_request_sock *ireq = inet_rsk(req);
1123 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1125 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1126 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1127 tcp_rsk(req)->snt_synack = 0;
1128 req->mss = rx_opt->mss_clamp;
1129 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1130 ireq->tstamp_ok = rx_opt->tstamp_ok;
1131 ireq->sack_ok = rx_opt->sack_ok;
1132 ireq->snd_wscale = rx_opt->snd_wscale;
1133 ireq->wscale_ok = rx_opt->wscale_ok;
1136 ireq->rmt_port = tcp_hdr(skb)->source;
1137 ireq->loc_port = tcp_hdr(skb)->dest;
1140 /* Compute time elapsed between SYNACK and the ACK completing 3WHS */
1141 static inline void tcp_synack_rtt_meas(struct sock *sk,
1142 struct request_sock *req)
1144 if (tcp_rsk(req)->snt_synack)
1145 tcp_valid_rtt_meas(sk,
1146 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1149 extern void tcp_enter_memory_pressure(struct sock *sk);
1151 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1153 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1156 static inline int keepalive_time_when(const struct tcp_sock *tp)
1158 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1161 static inline int keepalive_probes(const struct tcp_sock *tp)
1163 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1166 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1168 const struct inet_connection_sock *icsk = &tp->inet_conn;
1170 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1171 tcp_time_stamp - tp->rcv_tstamp);
1174 static inline int tcp_fin_time(const struct sock *sk)
1176 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1177 const int rto = inet_csk(sk)->icsk_rto;
1179 if (fin_timeout < (rto << 2) - (rto >> 1))
1180 fin_timeout = (rto << 2) - (rto >> 1);
1185 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1188 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1190 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1193 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1194 * then following tcp messages have valid values. Ignore 0 value,
1195 * or else 'negative' tsval might forbid us to accept their packets.
1197 if (!rx_opt->ts_recent)
1202 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1205 if (tcp_paws_check(rx_opt, 0))
1208 /* RST segments are not recommended to carry timestamp,
1209 and, if they do, it is recommended to ignore PAWS because
1210 "their cleanup function should take precedence over timestamps."
1211 Certainly, it is mistake. It is necessary to understand the reasons
1212 of this constraint to relax it: if peer reboots, clock may go
1213 out-of-sync and half-open connections will not be reset.
1214 Actually, the problem would be not existing if all
1215 the implementations followed draft about maintaining clock
1216 via reboots. Linux-2.2 DOES NOT!
1218 However, we can relax time bounds for RST segments to MSL.
1220 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1225 static inline void tcp_mib_init(struct net *net)
1228 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1229 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1230 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1231 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1235 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1237 tp->lost_skb_hint = NULL;
1238 tp->scoreboard_skb_hint = NULL;
1241 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1243 tcp_clear_retrans_hints_partial(tp);
1244 tp->retransmit_skb_hint = NULL;
1250 union tcp_md5_addr {
1252 #if IS_ENABLED(CONFIG_IPV6)
1257 /* - key database */
1258 struct tcp_md5sig_key {
1259 struct hlist_node node;
1261 u8 family; /* AF_INET or AF_INET6 */
1262 union tcp_md5_addr addr;
1263 u8 key[TCP_MD5SIG_MAXKEYLEN];
1264 struct rcu_head rcu;
1268 struct tcp_md5sig_info {
1269 struct hlist_head head;
1270 struct rcu_head rcu;
1273 /* - pseudo header */
1274 struct tcp4_pseudohdr {
1282 struct tcp6_pseudohdr {
1283 struct in6_addr saddr;
1284 struct in6_addr daddr;
1286 __be32 protocol; /* including padding */
1289 union tcp_md5sum_block {
1290 struct tcp4_pseudohdr ip4;
1291 #if IS_ENABLED(CONFIG_IPV6)
1292 struct tcp6_pseudohdr ip6;
1296 /* - pool: digest algorithm, hash description and scratch buffer */
1297 struct tcp_md5sig_pool {
1298 struct hash_desc md5_desc;
1299 union tcp_md5sum_block md5_blk;
1303 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1304 const struct sock *sk,
1305 const struct request_sock *req,
1306 const struct sk_buff *skb);
1307 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1308 int family, const u8 *newkey,
1309 u8 newkeylen, gfp_t gfp);
1310 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1312 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1313 struct sock *addr_sk);
1315 #ifdef CONFIG_TCP_MD5SIG
1316 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1317 const union tcp_md5_addr *addr, int family);
1318 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1320 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1321 const union tcp_md5_addr *addr,
1326 #define tcp_twsk_md5_key(twsk) NULL
1329 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1330 extern void tcp_free_md5sig_pool(void);
1332 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1333 extern void tcp_put_md5sig_pool(void);
1335 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1336 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1337 unsigned int header_len);
1338 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1339 const struct tcp_md5sig_key *key);
1341 /* From tcp_fastopen.c */
1342 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1343 struct tcp_fastopen_cookie *cookie,
1344 int *syn_loss, unsigned long *last_syn_loss);
1345 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1346 struct tcp_fastopen_cookie *cookie,
1348 struct tcp_fastopen_request {
1349 /* Fast Open cookie. Size 0 means a cookie request */
1350 struct tcp_fastopen_cookie cookie;
1351 struct msghdr *data; /* data in MSG_FASTOPEN */
1352 u16 copied; /* queued in tcp_connect() */
1354 void tcp_free_fastopen_req(struct tcp_sock *tp);
1356 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1357 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1358 void tcp_fastopen_cookie_gen(__be32 addr, struct tcp_fastopen_cookie *foc);
1360 #define TCP_FASTOPEN_KEY_LENGTH 16
1362 /* Fastopen key context */
1363 struct tcp_fastopen_context {
1364 struct crypto_cipher __rcu *tfm;
1365 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1366 struct rcu_head rcu;
1369 /* write queue abstraction */
1370 static inline void tcp_write_queue_purge(struct sock *sk)
1372 struct sk_buff *skb;
1374 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1375 sk_wmem_free_skb(sk, skb);
1377 tcp_clear_all_retrans_hints(tcp_sk(sk));
1380 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1382 return skb_peek(&sk->sk_write_queue);
1385 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1387 return skb_peek_tail(&sk->sk_write_queue);
1390 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1391 const struct sk_buff *skb)
1393 return skb_queue_next(&sk->sk_write_queue, skb);
1396 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1397 const struct sk_buff *skb)
1399 return skb_queue_prev(&sk->sk_write_queue, skb);
1402 #define tcp_for_write_queue(skb, sk) \
1403 skb_queue_walk(&(sk)->sk_write_queue, skb)
1405 #define tcp_for_write_queue_from(skb, sk) \
1406 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1408 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1409 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1411 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1413 return sk->sk_send_head;
1416 static inline bool tcp_skb_is_last(const struct sock *sk,
1417 const struct sk_buff *skb)
1419 return skb_queue_is_last(&sk->sk_write_queue, skb);
1422 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1424 if (tcp_skb_is_last(sk, skb))
1425 sk->sk_send_head = NULL;
1427 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1430 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1432 if (sk->sk_send_head == skb_unlinked)
1433 sk->sk_send_head = NULL;
1436 static inline void tcp_init_send_head(struct sock *sk)
1438 sk->sk_send_head = NULL;
1441 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1443 __skb_queue_tail(&sk->sk_write_queue, skb);
1446 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1448 __tcp_add_write_queue_tail(sk, skb);
1450 /* Queue it, remembering where we must start sending. */
1451 if (sk->sk_send_head == NULL) {
1452 sk->sk_send_head = skb;
1454 if (tcp_sk(sk)->highest_sack == NULL)
1455 tcp_sk(sk)->highest_sack = skb;
1459 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1461 __skb_queue_head(&sk->sk_write_queue, skb);
1464 /* Insert buff after skb on the write queue of sk. */
1465 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1466 struct sk_buff *buff,
1469 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1472 /* Insert new before skb on the write queue of sk. */
1473 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1474 struct sk_buff *skb,
1477 __skb_queue_before(&sk->sk_write_queue, skb, new);
1479 if (sk->sk_send_head == skb)
1480 sk->sk_send_head = new;
1483 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1485 __skb_unlink(skb, &sk->sk_write_queue);
1488 static inline bool tcp_write_queue_empty(struct sock *sk)
1490 return skb_queue_empty(&sk->sk_write_queue);
1493 static inline void tcp_push_pending_frames(struct sock *sk)
1495 if (tcp_send_head(sk)) {
1496 struct tcp_sock *tp = tcp_sk(sk);
1498 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1502 /* Start sequence of the skb just after the highest skb with SACKed
1503 * bit, valid only if sacked_out > 0 or when the caller has ensured
1504 * validity by itself.
1506 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1508 if (!tp->sacked_out)
1511 if (tp->highest_sack == NULL)
1514 return TCP_SKB_CB(tp->highest_sack)->seq;
1517 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1519 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1520 tcp_write_queue_next(sk, skb);
1523 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1525 return tcp_sk(sk)->highest_sack;
1528 static inline void tcp_highest_sack_reset(struct sock *sk)
1530 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1533 /* Called when old skb is about to be deleted (to be combined with new skb) */
1534 static inline void tcp_highest_sack_combine(struct sock *sk,
1535 struct sk_buff *old,
1536 struct sk_buff *new)
1538 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1539 tcp_sk(sk)->highest_sack = new;
1542 /* Determines whether this is a thin stream (which may suffer from
1543 * increased latency). Used to trigger latency-reducing mechanisms.
1545 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1547 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1551 enum tcp_seq_states {
1552 TCP_SEQ_STATE_LISTENING,
1553 TCP_SEQ_STATE_OPENREQ,
1554 TCP_SEQ_STATE_ESTABLISHED,
1555 TCP_SEQ_STATE_TIME_WAIT,
1558 int tcp_seq_open(struct inode *inode, struct file *file);
1560 struct tcp_seq_afinfo {
1563 const struct file_operations *seq_fops;
1564 struct seq_operations seq_ops;
1567 struct tcp_iter_state {
1568 struct seq_net_private p;
1570 enum tcp_seq_states state;
1571 struct sock *syn_wait_sk;
1572 int bucket, offset, sbucket, num;
1577 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1578 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1580 extern struct request_sock_ops tcp_request_sock_ops;
1581 extern struct request_sock_ops tcp6_request_sock_ops;
1583 extern void tcp_v4_destroy_sock(struct sock *sk);
1585 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1586 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1587 netdev_features_t features);
1588 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1589 struct sk_buff *skb);
1590 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1591 struct sk_buff *skb);
1592 extern int tcp_gro_complete(struct sk_buff *skb);
1593 extern int tcp4_gro_complete(struct sk_buff *skb);
1595 #ifdef CONFIG_PROC_FS
1596 extern int tcp4_proc_init(void);
1597 extern void tcp4_proc_exit(void);
1600 /* TCP af-specific functions */
1601 struct tcp_sock_af_ops {
1602 #ifdef CONFIG_TCP_MD5SIG
1603 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1604 struct sock *addr_sk);
1605 int (*calc_md5_hash) (char *location,
1606 struct tcp_md5sig_key *md5,
1607 const struct sock *sk,
1608 const struct request_sock *req,
1609 const struct sk_buff *skb);
1610 int (*md5_parse) (struct sock *sk,
1611 char __user *optval,
1616 struct tcp_request_sock_ops {
1617 #ifdef CONFIG_TCP_MD5SIG
1618 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1619 struct request_sock *req);
1620 int (*calc_md5_hash) (char *location,
1621 struct tcp_md5sig_key *md5,
1622 const struct sock *sk,
1623 const struct request_sock *req,
1624 const struct sk_buff *skb);
1628 /* Using SHA1 for now, define some constants.
1630 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1631 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1632 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1634 extern int tcp_cookie_generator(u32 *bakery);
1637 * struct tcp_cookie_values - each socket needs extra space for the
1638 * cookies, together with (optional) space for any SYN data.
1640 * A tcp_sock contains a pointer to the current value, and this is
1641 * cloned to the tcp_timewait_sock.
1643 * @cookie_pair: variable data from the option exchange.
1645 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1646 * indicates default (sysctl_tcp_cookie_size).
1647 * After cookie sent, remembers size of cookie.
1648 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1650 * @s_data_desired: user specified tcpct_s_data_desired. When the
1651 * constant payload is specified (@s_data_constant),
1652 * holds its length instead.
1653 * Range 0 to TCP_MSS_DESIRED.
1655 * @s_data_payload: constant data that is to be included in the
1656 * payload of SYN or SYNACK segments when the
1657 * cookie option is present.
1659 struct tcp_cookie_values {
1661 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1662 u8 cookie_pair_size;
1664 u16 s_data_desired:11,
1669 u8 s_data_payload[0];
1672 static inline void tcp_cookie_values_release(struct kref *kref)
1674 kfree(container_of(kref, struct tcp_cookie_values, kref));
1677 /* The length of constant payload data. Note that s_data_desired is
1678 * overloaded, depending on s_data_constant: either the length of constant
1679 * data (returned here) or the limit on variable data.
1681 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1683 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1684 ? tp->cookie_values->s_data_desired
1689 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1691 * As tcp_request_sock has already been extended in other places, the
1692 * only remaining method is to pass stack values along as function
1693 * parameters. These parameters are not needed after sending SYNACK.
1695 * @cookie_bakery: cryptographic secret and message workspace.
1697 * @cookie_plus: bytes in authenticator/cookie option, copied from
1698 * struct tcp_options_received (above).
1700 struct tcp_extend_values {
1701 struct request_values rv;
1702 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1708 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1710 return (struct tcp_extend_values *)rvp;
1713 extern void tcp_v4_init(void);
1714 extern void tcp_init(void);