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 5 /* number of times to retry active opening a
102 * connection: ~180sec is RFC minimum */
104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
105 * connection: ~180sec is RFC minimum */
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
128 * used as a fallback RTO for the
129 * initial data transmission if no
130 * valid RTT sample has been acquired,
131 * most likely due to retrans in 3WHS.
134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
135 * for local resources.
138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
140 #define TCP_KEEPALIVE_INTVL (75*HZ)
142 #define MAX_TCP_KEEPIDLE 32767
143 #define MAX_TCP_KEEPINTVL 32767
144 #define MAX_TCP_KEEPCNT 127
145 #define MAX_TCP_SYNCNT 127
147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
151 * after this time. It should be equal
152 * (or greater than) TCP_TIMEWAIT_LEN
153 * to provide reliability equal to one
154 * provided by timewait state.
156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
157 * timestamps. It must be less than
158 * minimal timewait lifetime.
164 #define TCPOPT_NOP 1 /* Padding */
165 #define TCPOPT_EOL 0 /* End of options */
166 #define TCPOPT_MSS 2 /* Segment size negotiating */
167 #define TCPOPT_WINDOW 3 /* Window scaling */
168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
169 #define TCPOPT_SACK 5 /* SACK Block */
170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
178 #define TCPOLEN_MSS 4
179 #define TCPOLEN_WINDOW 3
180 #define TCPOLEN_SACK_PERM 2
181 #define TCPOLEN_TIMESTAMP 10
182 #define TCPOLEN_MD5SIG 18
183 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
184 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
185 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
186 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
188 /* But this is what stacks really send out. */
189 #define TCPOLEN_TSTAMP_ALIGNED 12
190 #define TCPOLEN_WSCALE_ALIGNED 4
191 #define TCPOLEN_SACKPERM_ALIGNED 4
192 #define TCPOLEN_SACK_BASE 2
193 #define TCPOLEN_SACK_BASE_ALIGNED 4
194 #define TCPOLEN_SACK_PERBLOCK 8
195 #define TCPOLEN_MD5SIG_ALIGNED 20
196 #define TCPOLEN_MSS_ALIGNED 4
198 /* Flags in tp->nonagle */
199 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
200 #define TCP_NAGLE_CORK 2 /* Socket is corked */
201 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
203 /* TCP thin-stream limits */
204 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
206 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
207 #define TCP_INIT_CWND 10
209 extern struct inet_timewait_death_row tcp_death_row;
211 /* sysctl variables for tcp */
212 extern int sysctl_tcp_timestamps;
213 extern int sysctl_tcp_window_scaling;
214 extern int sysctl_tcp_sack;
215 extern int sysctl_tcp_fin_timeout;
216 extern int sysctl_tcp_keepalive_time;
217 extern int sysctl_tcp_keepalive_probes;
218 extern int sysctl_tcp_keepalive_intvl;
219 extern int sysctl_tcp_syn_retries;
220 extern int sysctl_tcp_synack_retries;
221 extern int sysctl_tcp_retries1;
222 extern int sysctl_tcp_retries2;
223 extern int sysctl_tcp_orphan_retries;
224 extern int sysctl_tcp_syncookies;
225 extern int sysctl_tcp_retrans_collapse;
226 extern int sysctl_tcp_stdurg;
227 extern int sysctl_tcp_rfc1337;
228 extern int sysctl_tcp_abort_on_overflow;
229 extern int sysctl_tcp_max_orphans;
230 extern int sysctl_tcp_fack;
231 extern int sysctl_tcp_reordering;
232 extern int sysctl_tcp_ecn;
233 extern int sysctl_tcp_dsack;
234 extern int sysctl_tcp_wmem[3];
235 extern int sysctl_tcp_rmem[3];
236 extern int sysctl_tcp_app_win;
237 extern int sysctl_tcp_adv_win_scale;
238 extern int sysctl_tcp_tw_reuse;
239 extern int sysctl_tcp_frto;
240 extern int sysctl_tcp_frto_response;
241 extern int sysctl_tcp_low_latency;
242 extern int sysctl_tcp_dma_copybreak;
243 extern int sysctl_tcp_nometrics_save;
244 extern int sysctl_tcp_moderate_rcvbuf;
245 extern int sysctl_tcp_tso_win_divisor;
246 extern int sysctl_tcp_abc;
247 extern int sysctl_tcp_mtu_probing;
248 extern int sysctl_tcp_base_mss;
249 extern int sysctl_tcp_workaround_signed_windows;
250 extern int sysctl_tcp_slow_start_after_idle;
251 extern int sysctl_tcp_max_ssthresh;
252 extern int sysctl_tcp_cookie_size;
253 extern int sysctl_tcp_thin_linear_timeouts;
254 extern int sysctl_tcp_thin_dupack;
255 extern int sysctl_tcp_early_retrans;
256 extern int sysctl_tcp_limit_output_bytes;
257 extern int sysctl_tcp_challenge_ack_limit;
259 extern atomic_long_t tcp_memory_allocated;
260 extern struct percpu_counter tcp_sockets_allocated;
261 extern int tcp_memory_pressure;
264 * The next routines deal with comparing 32 bit unsigned ints
265 * and worry about wraparound (automatic with unsigned arithmetic).
268 static inline bool before(__u32 seq1, __u32 seq2)
270 return (__s32)(seq1-seq2) < 0;
272 #define after(seq2, seq1) before(seq1, seq2)
274 /* is s2<=s1<=s3 ? */
275 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
277 return seq3 - seq2 >= seq1 - seq2;
280 static inline bool tcp_out_of_memory(struct sock *sk)
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
288 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
290 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
291 int orphans = percpu_counter_read_positive(ocp);
293 if (orphans << shift > sysctl_tcp_max_orphans) {
294 orphans = percpu_counter_sum_positive(ocp);
295 if (orphans << shift > sysctl_tcp_max_orphans)
301 extern bool tcp_check_oom(struct sock *sk, int shift);
303 /* syncookies: remember time of last synqueue overflow */
304 static inline void tcp_synq_overflow(struct sock *sk)
306 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
309 /* syncookies: no recent synqueue overflow on this listening socket? */
310 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
312 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
313 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
316 extern struct proto tcp_prot;
318 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
319 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
320 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
321 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
322 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
324 extern void tcp_init_mem(struct net *net);
326 extern void tcp_tasklet_init(void);
328 extern void tcp_v4_err(struct sk_buff *skb, u32);
330 extern void tcp_shutdown (struct sock *sk, int how);
332 extern void tcp_v4_early_demux(struct sk_buff *skb);
333 extern int tcp_v4_rcv(struct sk_buff *skb);
335 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk);
336 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
337 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
339 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
340 size_t size, int flags);
341 extern void tcp_release_cb(struct sock *sk);
342 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
343 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
344 const struct tcphdr *th, unsigned int len);
345 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
346 const struct tcphdr *th, unsigned int len);
347 extern void tcp_rcv_space_adjust(struct sock *sk);
348 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
349 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
350 extern void tcp_twsk_destructor(struct sock *sk);
351 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
352 struct pipe_inode_info *pipe, size_t len,
355 static inline void tcp_dec_quickack_mode(struct sock *sk,
356 const unsigned int pkts)
358 struct inet_connection_sock *icsk = inet_csk(sk);
360 if (icsk->icsk_ack.quick) {
361 if (pkts >= icsk->icsk_ack.quick) {
362 icsk->icsk_ack.quick = 0;
363 /* Leaving quickack mode we deflate ATO. */
364 icsk->icsk_ack.ato = TCP_ATO_MIN;
366 icsk->icsk_ack.quick -= pkts;
371 #define TCP_ECN_QUEUE_CWR 2
372 #define TCP_ECN_DEMAND_CWR 4
373 #define TCP_ECN_SEEN 8
383 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
385 const struct tcphdr *th);
386 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
387 struct request_sock *req,
388 struct request_sock **prev);
389 extern int tcp_child_process(struct sock *parent, struct sock *child,
390 struct sk_buff *skb);
391 extern bool tcp_use_frto(struct sock *sk);
392 extern void tcp_enter_frto(struct sock *sk);
393 extern void tcp_enter_loss(struct sock *sk, int how);
394 extern void tcp_clear_retrans(struct tcp_sock *tp);
395 extern void tcp_update_metrics(struct sock *sk);
396 extern void tcp_init_metrics(struct sock *sk);
397 extern void tcp_metrics_init(void);
398 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check);
399 extern bool tcp_remember_stamp(struct sock *sk);
400 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
401 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
402 extern void tcp_disable_fack(struct tcp_sock *tp);
403 extern void tcp_close(struct sock *sk, long timeout);
404 extern void tcp_init_sock(struct sock *sk);
405 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
406 struct poll_table_struct *wait);
407 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
408 char __user *optval, int __user *optlen);
409 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
410 char __user *optval, unsigned int optlen);
411 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
412 char __user *optval, int __user *optlen);
413 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
414 char __user *optval, unsigned int optlen);
415 extern void tcp_set_keepalive(struct sock *sk, int val);
416 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
417 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
418 size_t len, int nonblock, int flags, int *addr_len);
419 extern void tcp_parse_options(const struct sk_buff *skb,
420 struct tcp_options_received *opt_rx, const u8 **hvpp,
422 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
425 * TCP v4 functions exported for the inet6 API
428 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
429 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
430 extern struct sock * tcp_create_openreq_child(struct sock *sk,
431 struct request_sock *req,
432 struct sk_buff *skb);
433 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
434 struct request_sock *req,
435 struct dst_entry *dst);
436 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
437 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
439 extern int tcp_connect(struct sock *sk);
440 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
441 struct request_sock *req,
442 struct request_values *rvp);
443 extern int tcp_disconnect(struct sock *sk, int flags);
445 void tcp_connect_init(struct sock *sk);
446 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
447 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
449 /* From syncookies.c */
450 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
451 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
452 struct ip_options *opt);
453 #ifdef CONFIG_SYN_COOKIES
454 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
457 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
465 extern __u32 cookie_init_timestamp(struct request_sock *req);
466 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
468 /* From net/ipv6/syncookies.c */
469 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
470 #ifdef CONFIG_SYN_COOKIES
471 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
474 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
483 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
485 extern bool tcp_may_send_now(struct sock *sk);
486 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
487 extern void tcp_retransmit_timer(struct sock *sk);
488 extern void tcp_xmit_retransmit_queue(struct sock *);
489 extern void tcp_simple_retransmit(struct sock *);
490 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
491 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
493 extern void tcp_send_probe0(struct sock *);
494 extern void tcp_send_partial(struct sock *);
495 extern int tcp_write_wakeup(struct sock *);
496 extern void tcp_send_fin(struct sock *sk);
497 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
498 extern int tcp_send_synack(struct sock *);
499 extern bool tcp_syn_flood_action(struct sock *sk,
500 const struct sk_buff *skb,
502 extern void tcp_push_one(struct sock *, unsigned int mss_now);
503 extern void tcp_send_ack(struct sock *sk);
504 extern void tcp_send_delayed_ack(struct sock *sk);
507 extern void tcp_cwnd_application_limited(struct sock *sk);
508 extern void tcp_resume_early_retransmit(struct sock *sk);
509 extern void tcp_rearm_rto(struct sock *sk);
512 extern void tcp_init_xmit_timers(struct sock *);
513 static inline void tcp_clear_xmit_timers(struct sock *sk)
515 inet_csk_clear_xmit_timers(sk);
518 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
519 extern unsigned int tcp_current_mss(struct sock *sk);
521 /* Bound MSS / TSO packet size with the half of the window */
522 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
526 /* When peer uses tiny windows, there is no use in packetizing
527 * to sub-MSS pieces for the sake of SWS or making sure there
528 * are enough packets in the pipe for fast recovery.
530 * On the other hand, for extremely large MSS devices, handling
531 * smaller than MSS windows in this way does make sense.
533 if (tp->max_window >= 512)
534 cutoff = (tp->max_window >> 1);
536 cutoff = tp->max_window;
538 if (cutoff && pktsize > cutoff)
539 return max_t(int, cutoff, 68U - tp->tcp_header_len);
545 extern void tcp_get_info(const struct sock *, struct tcp_info *);
547 /* Read 'sendfile()'-style from a TCP socket */
548 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
549 unsigned int, size_t);
550 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
551 sk_read_actor_t recv_actor);
553 extern void tcp_initialize_rcv_mss(struct sock *sk);
555 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
556 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
557 extern void tcp_mtup_init(struct sock *sk);
558 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
560 static inline void tcp_bound_rto(const struct sock *sk)
562 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
563 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
566 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
568 return (tp->srtt >> 3) + tp->rttvar;
571 extern void tcp_set_rto(struct sock *sk);
573 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
575 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
576 ntohl(TCP_FLAG_ACK) |
580 static inline void tcp_fast_path_on(struct tcp_sock *tp)
582 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
585 static inline void tcp_fast_path_check(struct sock *sk)
587 struct tcp_sock *tp = tcp_sk(sk);
589 if (skb_queue_empty(&tp->out_of_order_queue) &&
591 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
593 tcp_fast_path_on(tp);
596 /* Compute the actual rto_min value */
597 static inline u32 tcp_rto_min(struct sock *sk)
599 const struct dst_entry *dst = __sk_dst_get(sk);
600 u32 rto_min = TCP_RTO_MIN;
602 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
603 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
607 /* Compute the actual receive window we are currently advertising.
608 * Rcv_nxt can be after the window if our peer push more data
609 * than the offered window.
611 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
613 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
620 /* Choose a new window, without checks for shrinking, and without
621 * scaling applied to the result. The caller does these things
622 * if necessary. This is a "raw" window selection.
624 extern u32 __tcp_select_window(struct sock *sk);
626 void tcp_send_window_probe(struct sock *sk);
628 /* TCP timestamps are only 32-bits, this causes a slight
629 * complication on 64-bit systems since we store a snapshot
630 * of jiffies in the buffer control blocks below. We decided
631 * to use only the low 32-bits of jiffies and hide the ugly
632 * casts with the following macro.
634 #define tcp_time_stamp ((__u32)(jiffies))
636 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
638 #define TCPHDR_FIN 0x01
639 #define TCPHDR_SYN 0x02
640 #define TCPHDR_RST 0x04
641 #define TCPHDR_PSH 0x08
642 #define TCPHDR_ACK 0x10
643 #define TCPHDR_URG 0x20
644 #define TCPHDR_ECE 0x40
645 #define TCPHDR_CWR 0x80
647 /* This is what the send packet queuing engine uses to pass
648 * TCP per-packet control information to the transmission code.
649 * We also store the host-order sequence numbers in here too.
650 * This is 44 bytes if IPV6 is enabled.
651 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
655 struct inet_skb_parm h4;
656 #if IS_ENABLED(CONFIG_IPV6)
657 struct inet6_skb_parm h6;
659 } header; /* For incoming frames */
660 __u32 seq; /* Starting sequence number */
661 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
662 __u32 when; /* used to compute rtt's */
663 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
665 __u8 sacked; /* State flags for SACK/FACK. */
666 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
667 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
668 #define TCPCB_LOST 0x04 /* SKB is lost */
669 #define TCPCB_TAGBITS 0x07 /* All tag bits */
670 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
671 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
673 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
675 __u32 ack_seq; /* Sequence number ACK'd */
678 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
680 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
682 * If we receive a SYN packet with these bits set, it means a network is
683 * playing bad games with TOS bits. In order to avoid possible false congestion
684 * notifications, we disable TCP ECN negociation.
687 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
689 const struct tcphdr *th = tcp_hdr(skb);
691 if (sysctl_tcp_ecn && th->ece && th->cwr &&
692 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
693 inet_rsk(req)->ecn_ok = 1;
696 /* Due to TSO, an SKB can be composed of multiple actual
697 * packets. To keep these tracked properly, we use this.
699 static inline int tcp_skb_pcount(const struct sk_buff *skb)
701 return skb_shinfo(skb)->gso_segs;
704 /* This is valid iff tcp_skb_pcount() > 1. */
705 static inline int tcp_skb_mss(const struct sk_buff *skb)
707 return skb_shinfo(skb)->gso_size;
710 /* Events passed to congestion control interface */
712 CA_EVENT_TX_START, /* first transmit when no packets in flight */
713 CA_EVENT_CWND_RESTART, /* congestion window restart */
714 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
715 CA_EVENT_FRTO, /* fast recovery timeout */
716 CA_EVENT_LOSS, /* loss timeout */
717 CA_EVENT_FAST_ACK, /* in sequence ack */
718 CA_EVENT_SLOW_ACK, /* other ack */
722 * Interface for adding new TCP congestion control handlers
724 #define TCP_CA_NAME_MAX 16
725 #define TCP_CA_MAX 128
726 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
728 #define TCP_CONG_NON_RESTRICTED 0x1
729 #define TCP_CONG_RTT_STAMP 0x2
731 struct tcp_congestion_ops {
732 struct list_head list;
735 /* initialize private data (optional) */
736 void (*init)(struct sock *sk);
737 /* cleanup private data (optional) */
738 void (*release)(struct sock *sk);
740 /* return slow start threshold (required) */
741 u32 (*ssthresh)(struct sock *sk);
742 /* lower bound for congestion window (optional) */
743 u32 (*min_cwnd)(const struct sock *sk);
744 /* do new cwnd calculation (required) */
745 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
746 /* call before changing ca_state (optional) */
747 void (*set_state)(struct sock *sk, u8 new_state);
748 /* call when cwnd event occurs (optional) */
749 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
750 /* new value of cwnd after loss (optional) */
751 u32 (*undo_cwnd)(struct sock *sk);
752 /* hook for packet ack accounting (optional) */
753 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
754 /* get info for inet_diag (optional) */
755 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
757 char name[TCP_CA_NAME_MAX];
758 struct module *owner;
761 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
762 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
764 extern void tcp_init_congestion_control(struct sock *sk);
765 extern void tcp_cleanup_congestion_control(struct sock *sk);
766 extern int tcp_set_default_congestion_control(const char *name);
767 extern void tcp_get_default_congestion_control(char *name);
768 extern void tcp_get_available_congestion_control(char *buf, size_t len);
769 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
770 extern int tcp_set_allowed_congestion_control(char *allowed);
771 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
772 extern void tcp_slow_start(struct tcp_sock *tp);
773 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
775 extern struct tcp_congestion_ops tcp_init_congestion_ops;
776 extern u32 tcp_reno_ssthresh(struct sock *sk);
777 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
778 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
779 extern struct tcp_congestion_ops tcp_reno;
781 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
783 struct inet_connection_sock *icsk = inet_csk(sk);
785 if (icsk->icsk_ca_ops->set_state)
786 icsk->icsk_ca_ops->set_state(sk, ca_state);
787 icsk->icsk_ca_state = ca_state;
790 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
792 const struct inet_connection_sock *icsk = inet_csk(sk);
794 if (icsk->icsk_ca_ops->cwnd_event)
795 icsk->icsk_ca_ops->cwnd_event(sk, event);
798 /* These functions determine how the current flow behaves in respect of SACK
799 * handling. SACK is negotiated with the peer, and therefore it can vary
800 * between different flows.
802 * tcp_is_sack - SACK enabled
803 * tcp_is_reno - No SACK
804 * tcp_is_fack - FACK enabled, implies SACK enabled
806 static inline int tcp_is_sack(const struct tcp_sock *tp)
808 return tp->rx_opt.sack_ok;
811 static inline bool tcp_is_reno(const struct tcp_sock *tp)
813 return !tcp_is_sack(tp);
816 static inline bool tcp_is_fack(const struct tcp_sock *tp)
818 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
821 static inline void tcp_enable_fack(struct tcp_sock *tp)
823 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
826 /* TCP early-retransmit (ER) is similar to but more conservative than
827 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
829 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
831 tp->do_early_retrans = sysctl_tcp_early_retrans &&
832 !sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3;
833 tp->early_retrans_delayed = 0;
836 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
838 tp->do_early_retrans = 0;
841 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
843 return tp->sacked_out + tp->lost_out;
846 /* This determines how many packets are "in the network" to the best
847 * of our knowledge. In many cases it is conservative, but where
848 * detailed information is available from the receiver (via SACK
849 * blocks etc.) we can make more aggressive calculations.
851 * Use this for decisions involving congestion control, use just
852 * tp->packets_out to determine if the send queue is empty or not.
854 * Read this equation as:
856 * "Packets sent once on transmission queue" MINUS
857 * "Packets left network, but not honestly ACKed yet" PLUS
858 * "Packets fast retransmitted"
860 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
862 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
865 #define TCP_INFINITE_SSTHRESH 0x7fffffff
867 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
869 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
872 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
873 * The exception is rate halving phase, when cwnd is decreasing towards
876 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
878 const struct tcp_sock *tp = tcp_sk(sk);
880 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
881 return tp->snd_ssthresh;
883 return max(tp->snd_ssthresh,
884 ((tp->snd_cwnd >> 1) +
885 (tp->snd_cwnd >> 2)));
888 /* Use define here intentionally to get WARN_ON location shown at the caller */
889 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
891 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
892 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
894 /* The maximum number of MSS of available cwnd for which TSO defers
895 * sending if not using sysctl_tcp_tso_win_divisor.
897 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
902 /* Slow start with delack produces 3 packets of burst, so that
903 * it is safe "de facto". This will be the default - same as
904 * the default reordering threshold - but if reordering increases,
905 * we must be able to allow cwnd to burst at least this much in order
906 * to not pull it back when holes are filled.
908 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
910 return tp->reordering;
913 /* Returns end sequence number of the receiver's advertised window */
914 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
916 return tp->snd_una + tp->snd_wnd;
918 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
920 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
921 const struct sk_buff *skb)
924 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
927 static inline void tcp_check_probe_timer(struct sock *sk)
929 const struct tcp_sock *tp = tcp_sk(sk);
930 const struct inet_connection_sock *icsk = inet_csk(sk);
932 if (!tp->packets_out && !icsk->icsk_pending)
933 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
934 icsk->icsk_rto, TCP_RTO_MAX);
937 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
942 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
948 * Calculate(/check) TCP checksum
950 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
951 __be32 daddr, __wsum base)
953 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
956 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
958 return __skb_checksum_complete(skb);
961 static inline bool tcp_checksum_complete(struct sk_buff *skb)
963 return !skb_csum_unnecessary(skb) &&
964 __tcp_checksum_complete(skb);
967 /* Prequeue for VJ style copy to user, combined with checksumming. */
969 static inline void tcp_prequeue_init(struct tcp_sock *tp)
971 tp->ucopy.task = NULL;
973 tp->ucopy.memory = 0;
974 skb_queue_head_init(&tp->ucopy.prequeue);
975 #ifdef CONFIG_NET_DMA
976 tp->ucopy.dma_chan = NULL;
977 tp->ucopy.wakeup = 0;
978 tp->ucopy.pinned_list = NULL;
979 tp->ucopy.dma_cookie = 0;
983 /* Packet is added to VJ-style prequeue for processing in process
984 * context, if a reader task is waiting. Apparently, this exciting
985 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
986 * failed somewhere. Latency? Burstiness? Well, at least now we will
987 * see, why it failed. 8)8) --ANK
989 * NOTE: is this not too big to inline?
991 static inline bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
993 struct tcp_sock *tp = tcp_sk(sk);
995 if (sysctl_tcp_low_latency || !tp->ucopy.task)
998 __skb_queue_tail(&tp->ucopy.prequeue, skb);
999 tp->ucopy.memory += skb->truesize;
1000 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1001 struct sk_buff *skb1;
1003 BUG_ON(sock_owned_by_user(sk));
1005 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1006 sk_backlog_rcv(sk, skb1);
1007 NET_INC_STATS_BH(sock_net(sk),
1008 LINUX_MIB_TCPPREQUEUEDROPPED);
1011 tp->ucopy.memory = 0;
1012 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1013 wake_up_interruptible_sync_poll(sk_sleep(sk),
1014 POLLIN | POLLRDNORM | POLLRDBAND);
1015 if (!inet_csk_ack_scheduled(sk))
1016 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1017 (3 * tcp_rto_min(sk)) / 4,
1027 static const char *statename[]={
1028 "Unused","Established","Syn Sent","Syn Recv",
1029 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1030 "Close Wait","Last ACK","Listen","Closing"
1033 extern void tcp_set_state(struct sock *sk, int state);
1035 extern void tcp_done(struct sock *sk);
1037 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1040 rx_opt->num_sacks = 0;
1043 /* Determine a window scaling and initial window to offer. */
1044 extern void tcp_select_initial_window(int __space, __u32 mss,
1045 __u32 *rcv_wnd, __u32 *window_clamp,
1046 int wscale_ok, __u8 *rcv_wscale,
1047 __u32 init_rcv_wnd);
1049 static inline int tcp_win_from_space(int space)
1051 return sysctl_tcp_adv_win_scale<=0 ?
1052 (space>>(-sysctl_tcp_adv_win_scale)) :
1053 space - (space>>sysctl_tcp_adv_win_scale);
1056 /* Note: caller must be prepared to deal with negative returns */
1057 static inline int tcp_space(const struct sock *sk)
1059 return tcp_win_from_space(sk->sk_rcvbuf -
1060 atomic_read(&sk->sk_rmem_alloc));
1063 static inline int tcp_full_space(const struct sock *sk)
1065 return tcp_win_from_space(sk->sk_rcvbuf);
1068 static inline void tcp_openreq_init(struct request_sock *req,
1069 struct tcp_options_received *rx_opt,
1070 struct sk_buff *skb)
1072 struct inet_request_sock *ireq = inet_rsk(req);
1074 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1076 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1077 req->mss = rx_opt->mss_clamp;
1078 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1079 ireq->tstamp_ok = rx_opt->tstamp_ok;
1080 ireq->sack_ok = rx_opt->sack_ok;
1081 ireq->snd_wscale = rx_opt->snd_wscale;
1082 ireq->wscale_ok = rx_opt->wscale_ok;
1085 ireq->rmt_port = tcp_hdr(skb)->source;
1086 ireq->loc_port = tcp_hdr(skb)->dest;
1089 extern void tcp_enter_memory_pressure(struct sock *sk);
1091 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1093 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1096 static inline int keepalive_time_when(const struct tcp_sock *tp)
1098 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1101 static inline int keepalive_probes(const struct tcp_sock *tp)
1103 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1106 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1108 const struct inet_connection_sock *icsk = &tp->inet_conn;
1110 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1111 tcp_time_stamp - tp->rcv_tstamp);
1114 static inline int tcp_fin_time(const struct sock *sk)
1116 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1117 const int rto = inet_csk(sk)->icsk_rto;
1119 if (fin_timeout < (rto << 2) - (rto >> 1))
1120 fin_timeout = (rto << 2) - (rto >> 1);
1125 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1128 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1130 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1133 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1134 * then following tcp messages have valid values. Ignore 0 value,
1135 * or else 'negative' tsval might forbid us to accept their packets.
1137 if (!rx_opt->ts_recent)
1142 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1145 if (tcp_paws_check(rx_opt, 0))
1148 /* RST segments are not recommended to carry timestamp,
1149 and, if they do, it is recommended to ignore PAWS because
1150 "their cleanup function should take precedence over timestamps."
1151 Certainly, it is mistake. It is necessary to understand the reasons
1152 of this constraint to relax it: if peer reboots, clock may go
1153 out-of-sync and half-open connections will not be reset.
1154 Actually, the problem would be not existing if all
1155 the implementations followed draft about maintaining clock
1156 via reboots. Linux-2.2 DOES NOT!
1158 However, we can relax time bounds for RST segments to MSL.
1160 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1165 static inline void tcp_mib_init(struct net *net)
1168 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1169 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1170 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1171 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1175 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1177 tp->lost_skb_hint = NULL;
1178 tp->scoreboard_skb_hint = NULL;
1181 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1183 tcp_clear_retrans_hints_partial(tp);
1184 tp->retransmit_skb_hint = NULL;
1190 union tcp_md5_addr {
1192 #if IS_ENABLED(CONFIG_IPV6)
1197 /* - key database */
1198 struct tcp_md5sig_key {
1199 struct hlist_node node;
1201 u8 family; /* AF_INET or AF_INET6 */
1202 union tcp_md5_addr addr;
1203 u8 key[TCP_MD5SIG_MAXKEYLEN];
1204 struct rcu_head rcu;
1208 struct tcp_md5sig_info {
1209 struct hlist_head head;
1210 struct rcu_head rcu;
1213 /* - pseudo header */
1214 struct tcp4_pseudohdr {
1222 struct tcp6_pseudohdr {
1223 struct in6_addr saddr;
1224 struct in6_addr daddr;
1226 __be32 protocol; /* including padding */
1229 union tcp_md5sum_block {
1230 struct tcp4_pseudohdr ip4;
1231 #if IS_ENABLED(CONFIG_IPV6)
1232 struct tcp6_pseudohdr ip6;
1236 /* - pool: digest algorithm, hash description and scratch buffer */
1237 struct tcp_md5sig_pool {
1238 struct hash_desc md5_desc;
1239 union tcp_md5sum_block md5_blk;
1243 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1244 const struct sock *sk,
1245 const struct request_sock *req,
1246 const struct sk_buff *skb);
1247 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1248 int family, const u8 *newkey,
1249 u8 newkeylen, gfp_t gfp);
1250 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1252 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1253 struct sock *addr_sk);
1255 #ifdef CONFIG_TCP_MD5SIG
1256 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1257 const union tcp_md5_addr *addr, int family);
1258 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1260 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1261 const union tcp_md5_addr *addr,
1266 #define tcp_twsk_md5_key(twsk) NULL
1269 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1270 extern void tcp_free_md5sig_pool(void);
1272 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1273 extern void tcp_put_md5sig_pool(void);
1275 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1276 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1277 unsigned int header_len);
1278 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1279 const struct tcp_md5sig_key *key);
1281 /* write queue abstraction */
1282 static inline void tcp_write_queue_purge(struct sock *sk)
1284 struct sk_buff *skb;
1286 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1287 sk_wmem_free_skb(sk, skb);
1289 tcp_clear_all_retrans_hints(tcp_sk(sk));
1292 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1294 return skb_peek(&sk->sk_write_queue);
1297 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1299 return skb_peek_tail(&sk->sk_write_queue);
1302 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1303 const struct sk_buff *skb)
1305 return skb_queue_next(&sk->sk_write_queue, skb);
1308 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1309 const struct sk_buff *skb)
1311 return skb_queue_prev(&sk->sk_write_queue, skb);
1314 #define tcp_for_write_queue(skb, sk) \
1315 skb_queue_walk(&(sk)->sk_write_queue, skb)
1317 #define tcp_for_write_queue_from(skb, sk) \
1318 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1320 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1321 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1323 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1325 return sk->sk_send_head;
1328 static inline bool tcp_skb_is_last(const struct sock *sk,
1329 const struct sk_buff *skb)
1331 return skb_queue_is_last(&sk->sk_write_queue, skb);
1334 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1336 if (tcp_skb_is_last(sk, skb))
1337 sk->sk_send_head = NULL;
1339 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1342 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1344 if (sk->sk_send_head == skb_unlinked)
1345 sk->sk_send_head = NULL;
1348 static inline void tcp_init_send_head(struct sock *sk)
1350 sk->sk_send_head = NULL;
1353 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1355 __skb_queue_tail(&sk->sk_write_queue, skb);
1358 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1360 __tcp_add_write_queue_tail(sk, skb);
1362 /* Queue it, remembering where we must start sending. */
1363 if (sk->sk_send_head == NULL) {
1364 sk->sk_send_head = skb;
1366 if (tcp_sk(sk)->highest_sack == NULL)
1367 tcp_sk(sk)->highest_sack = skb;
1371 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1373 __skb_queue_head(&sk->sk_write_queue, skb);
1376 /* Insert buff after skb on the write queue of sk. */
1377 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1378 struct sk_buff *buff,
1381 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1384 /* Insert new before skb on the write queue of sk. */
1385 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1386 struct sk_buff *skb,
1389 __skb_queue_before(&sk->sk_write_queue, skb, new);
1391 if (sk->sk_send_head == skb)
1392 sk->sk_send_head = new;
1395 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1397 __skb_unlink(skb, &sk->sk_write_queue);
1400 static inline bool tcp_write_queue_empty(struct sock *sk)
1402 return skb_queue_empty(&sk->sk_write_queue);
1405 static inline void tcp_push_pending_frames(struct sock *sk)
1407 if (tcp_send_head(sk)) {
1408 struct tcp_sock *tp = tcp_sk(sk);
1410 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1414 /* Start sequence of the skb just after the highest skb with SACKed
1415 * bit, valid only if sacked_out > 0 or when the caller has ensured
1416 * validity by itself.
1418 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1420 if (!tp->sacked_out)
1423 if (tp->highest_sack == NULL)
1426 return TCP_SKB_CB(tp->highest_sack)->seq;
1429 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1431 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1432 tcp_write_queue_next(sk, skb);
1435 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1437 return tcp_sk(sk)->highest_sack;
1440 static inline void tcp_highest_sack_reset(struct sock *sk)
1442 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1445 /* Called when old skb is about to be deleted (to be combined with new skb) */
1446 static inline void tcp_highest_sack_combine(struct sock *sk,
1447 struct sk_buff *old,
1448 struct sk_buff *new)
1450 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1451 tcp_sk(sk)->highest_sack = new;
1454 /* Determines whether this is a thin stream (which may suffer from
1455 * increased latency). Used to trigger latency-reducing mechanisms.
1457 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1459 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1463 enum tcp_seq_states {
1464 TCP_SEQ_STATE_LISTENING,
1465 TCP_SEQ_STATE_OPENREQ,
1466 TCP_SEQ_STATE_ESTABLISHED,
1467 TCP_SEQ_STATE_TIME_WAIT,
1470 int tcp_seq_open(struct inode *inode, struct file *file);
1472 struct tcp_seq_afinfo {
1475 const struct file_operations *seq_fops;
1476 struct seq_operations seq_ops;
1479 struct tcp_iter_state {
1480 struct seq_net_private p;
1482 enum tcp_seq_states state;
1483 struct sock *syn_wait_sk;
1484 int bucket, offset, sbucket, num, uid;
1488 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1489 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1491 extern struct request_sock_ops tcp_request_sock_ops;
1492 extern struct request_sock_ops tcp6_request_sock_ops;
1494 extern void tcp_v4_destroy_sock(struct sock *sk);
1496 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1497 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1498 netdev_features_t features);
1499 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1500 struct sk_buff *skb);
1501 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1502 struct sk_buff *skb);
1503 extern int tcp_gro_complete(struct sk_buff *skb);
1504 extern int tcp4_gro_complete(struct sk_buff *skb);
1506 #ifdef CONFIG_PROC_FS
1507 extern int tcp4_proc_init(void);
1508 extern void tcp4_proc_exit(void);
1511 /* TCP af-specific functions */
1512 struct tcp_sock_af_ops {
1513 #ifdef CONFIG_TCP_MD5SIG
1514 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1515 struct sock *addr_sk);
1516 int (*calc_md5_hash) (char *location,
1517 struct tcp_md5sig_key *md5,
1518 const struct sock *sk,
1519 const struct request_sock *req,
1520 const struct sk_buff *skb);
1521 int (*md5_parse) (struct sock *sk,
1522 char __user *optval,
1527 struct tcp_request_sock_ops {
1528 #ifdef CONFIG_TCP_MD5SIG
1529 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1530 struct request_sock *req);
1531 int (*calc_md5_hash) (char *location,
1532 struct tcp_md5sig_key *md5,
1533 const struct sock *sk,
1534 const struct request_sock *req,
1535 const struct sk_buff *skb);
1539 /* Using SHA1 for now, define some constants.
1541 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1542 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1543 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1545 extern int tcp_cookie_generator(u32 *bakery);
1548 * struct tcp_cookie_values - each socket needs extra space for the
1549 * cookies, together with (optional) space for any SYN data.
1551 * A tcp_sock contains a pointer to the current value, and this is
1552 * cloned to the tcp_timewait_sock.
1554 * @cookie_pair: variable data from the option exchange.
1556 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1557 * indicates default (sysctl_tcp_cookie_size).
1558 * After cookie sent, remembers size of cookie.
1559 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1561 * @s_data_desired: user specified tcpct_s_data_desired. When the
1562 * constant payload is specified (@s_data_constant),
1563 * holds its length instead.
1564 * Range 0 to TCP_MSS_DESIRED.
1566 * @s_data_payload: constant data that is to be included in the
1567 * payload of SYN or SYNACK segments when the
1568 * cookie option is present.
1570 struct tcp_cookie_values {
1572 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1573 u8 cookie_pair_size;
1575 u16 s_data_desired:11,
1580 u8 s_data_payload[0];
1583 static inline void tcp_cookie_values_release(struct kref *kref)
1585 kfree(container_of(kref, struct tcp_cookie_values, kref));
1588 /* The length of constant payload data. Note that s_data_desired is
1589 * overloaded, depending on s_data_constant: either the length of constant
1590 * data (returned here) or the limit on variable data.
1592 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1594 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1595 ? tp->cookie_values->s_data_desired
1600 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1602 * As tcp_request_sock has already been extended in other places, the
1603 * only remaining method is to pass stack values along as function
1604 * parameters. These parameters are not needed after sending SYNACK.
1606 * @cookie_bakery: cryptographic secret and message workspace.
1608 * @cookie_plus: bytes in authenticator/cookie option, copied from
1609 * struct tcp_options_received (above).
1611 struct tcp_extend_values {
1612 struct request_values rv;
1613 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1619 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1621 return (struct tcp_extend_values *)rvp;
1624 extern void tcp_v4_init(void);
1625 extern void tcp_init(void);