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.
22 #define FASTRETRANS_DEBUG 1
24 #include <linux/list.h>
25 #include <linux/tcp.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>
49 extern struct inet_hashinfo tcp_hashinfo
;
51 extern struct percpu_counter tcp_orphan_count
;
52 extern void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
61 #define MAX_TCP_WINDOW 32767U
63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 512
69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
70 #define TCP_FASTRETRANS_THRESH 3
72 /* Maximal reordering. */
73 #define TCP_MAX_REORDERING 127
75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
76 #define TCP_MAX_QUICKACKS 16U
79 #define TCP_URG_VALID 0x0100
80 #define TCP_URG_NOTYET 0x0200
81 #define TCP_URG_READ 0x0400
83 #define TCP_RETR1 3 /*
84 * This is how many retries it does before it
85 * tries to figure out if the gateway is
86 * down. Minimal RFC value is 3; it corresponds
87 * to ~3sec-8min depending on RTO.
90 #define TCP_RETR2 15 /*
91 * This should take at least
92 * 90 minutes to time out.
93 * RFC1122 says that the limit is 100 sec.
94 * 15 is ~13-30min depending on RTO.
97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
98 * connection: ~180sec is RFC minimum */
100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
101 * connection: ~180sec is RFC minimum */
104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
105 * socket. 7 is ~50sec-16min.
109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
110 * state, about 60 seconds */
111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
112 /* BSD style FIN_WAIT2 deadlock breaker.
113 * It used to be 3min, new value is 60sec,
114 * to combine FIN-WAIT-2 timeout with
118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
121 #define TCP_ATO_MIN ((unsigned)(HZ/25))
123 #define TCP_DELACK_MIN 4U
124 #define TCP_ATO_MIN 4U
126 #define TCP_RTO_MAX ((unsigned)(120*HZ))
127 #define TCP_RTO_MIN ((unsigned)(HZ/5))
128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
131 * for local resources.
134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
136 #define TCP_KEEPALIVE_INTVL (75*HZ)
138 #define MAX_TCP_KEEPIDLE 32767
139 #define MAX_TCP_KEEPINTVL 32767
140 #define MAX_TCP_KEEPCNT 127
141 #define MAX_TCP_SYNCNT 127
143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
147 * after this time. It should be equal
148 * (or greater than) TCP_TIMEWAIT_LEN
149 * to provide reliability equal to one
150 * provided by timewait state.
152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
153 * timestamps. It must be less than
154 * minimal timewait lifetime.
160 #define TCPOPT_NOP 1 /* Padding */
161 #define TCPOPT_EOL 0 /* End of options */
162 #define TCPOPT_MSS 2 /* Segment size negotiating */
163 #define TCPOPT_WINDOW 3 /* Window scaling */
164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
165 #define TCPOPT_SACK 5 /* SACK Block */
166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
174 #define TCPOLEN_MSS 4
175 #define TCPOLEN_WINDOW 3
176 #define TCPOLEN_SACK_PERM 2
177 #define TCPOLEN_TIMESTAMP 10
178 #define TCPOLEN_MD5SIG 18
179 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
180 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
181 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
182 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
184 /* But this is what stacks really send out. */
185 #define TCPOLEN_TSTAMP_ALIGNED 12
186 #define TCPOLEN_WSCALE_ALIGNED 4
187 #define TCPOLEN_SACKPERM_ALIGNED 4
188 #define TCPOLEN_SACK_BASE 2
189 #define TCPOLEN_SACK_BASE_ALIGNED 4
190 #define TCPOLEN_SACK_PERBLOCK 8
191 #define TCPOLEN_MD5SIG_ALIGNED 20
192 #define TCPOLEN_MSS_ALIGNED 4
194 /* Flags in tp->nonagle */
195 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
196 #define TCP_NAGLE_CORK 2 /* Socket is corked */
197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
199 /* TCP thin-stream limits */
200 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
202 extern struct inet_timewait_death_row tcp_death_row
;
204 /* sysctl variables for tcp */
205 extern int sysctl_tcp_timestamps
;
206 extern int sysctl_tcp_window_scaling
;
207 extern int sysctl_tcp_sack
;
208 extern int sysctl_tcp_fin_timeout
;
209 extern int sysctl_tcp_keepalive_time
;
210 extern int sysctl_tcp_keepalive_probes
;
211 extern int sysctl_tcp_keepalive_intvl
;
212 extern int sysctl_tcp_syn_retries
;
213 extern int sysctl_tcp_synack_retries
;
214 extern int sysctl_tcp_retries1
;
215 extern int sysctl_tcp_retries2
;
216 extern int sysctl_tcp_orphan_retries
;
217 extern int sysctl_tcp_syncookies
;
218 extern int sysctl_tcp_retrans_collapse
;
219 extern int sysctl_tcp_stdurg
;
220 extern int sysctl_tcp_rfc1337
;
221 extern int sysctl_tcp_abort_on_overflow
;
222 extern int sysctl_tcp_max_orphans
;
223 extern int sysctl_tcp_fack
;
224 extern int sysctl_tcp_reordering
;
225 extern int sysctl_tcp_ecn
;
226 extern int sysctl_tcp_dsack
;
227 extern int sysctl_tcp_mem
[3];
228 extern int sysctl_tcp_wmem
[3];
229 extern int sysctl_tcp_rmem
[3];
230 extern int sysctl_tcp_app_win
;
231 extern int sysctl_tcp_adv_win_scale
;
232 extern int sysctl_tcp_tw_reuse
;
233 extern int sysctl_tcp_frto
;
234 extern int sysctl_tcp_frto_response
;
235 extern int sysctl_tcp_low_latency
;
236 extern int sysctl_tcp_dma_copybreak
;
237 extern int sysctl_tcp_nometrics_save
;
238 extern int sysctl_tcp_moderate_rcvbuf
;
239 extern int sysctl_tcp_tso_win_divisor
;
240 extern int sysctl_tcp_abc
;
241 extern int sysctl_tcp_mtu_probing
;
242 extern int sysctl_tcp_base_mss
;
243 extern int sysctl_tcp_workaround_signed_windows
;
244 extern int sysctl_tcp_slow_start_after_idle
;
245 extern int sysctl_tcp_max_ssthresh
;
246 extern int sysctl_tcp_cookie_size
;
247 extern int sysctl_tcp_thin_linear_timeouts
;
248 extern int sysctl_tcp_thin_dupack
;
250 extern atomic_t tcp_memory_allocated
;
251 extern struct percpu_counter tcp_sockets_allocated
;
252 extern int tcp_memory_pressure
;
255 * The next routines deal with comparing 32 bit unsigned ints
256 * and worry about wraparound (automatic with unsigned arithmetic).
259 static inline int before(__u32 seq1
, __u32 seq2
)
261 return (__s32
)(seq1
-seq2
) < 0;
263 #define after(seq2, seq1) before(seq1, seq2)
265 /* is s2<=s1<=s3 ? */
266 static inline int between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
268 return seq3
- seq2
>= seq1
- seq2
;
271 static inline int tcp_too_many_orphans(struct sock
*sk
, int num
)
273 return (num
> sysctl_tcp_max_orphans
) ||
274 (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
275 atomic_read(&tcp_memory_allocated
) > sysctl_tcp_mem
[2]);
278 /* syncookies: remember time of last synqueue overflow */
279 static inline void tcp_synq_overflow(struct sock
*sk
)
281 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= jiffies
;
284 /* syncookies: no recent synqueue overflow on this listening socket? */
285 static inline int tcp_synq_no_recent_overflow(const struct sock
*sk
)
287 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
288 return time_after(jiffies
, last_overflow
+ TCP_TIMEOUT_INIT
);
291 extern struct proto tcp_prot
;
293 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
294 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
295 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
296 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
297 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
299 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
301 extern void tcp_shutdown (struct sock
*sk
, int how
);
303 extern int tcp_v4_rcv(struct sk_buff
*skb
);
305 extern int tcp_v4_remember_stamp(struct sock
*sk
);
306 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
307 extern int tcp_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
308 struct msghdr
*msg
, size_t size
);
309 extern ssize_t
tcp_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
310 size_t size
, int flags
);
311 extern int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
312 extern int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
313 struct tcphdr
*th
, unsigned len
);
314 extern int tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
315 struct tcphdr
*th
, unsigned len
);
316 extern void tcp_rcv_space_adjust(struct sock
*sk
);
317 extern void tcp_cleanup_rbuf(struct sock
*sk
, int copied
);
318 extern int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
319 extern void tcp_twsk_destructor(struct sock
*sk
);
320 extern ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
321 struct pipe_inode_info
*pipe
, size_t len
,
324 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
325 const unsigned int pkts
)
327 struct inet_connection_sock
*icsk
= inet_csk(sk
);
329 if (icsk
->icsk_ack
.quick
) {
330 if (pkts
>= icsk
->icsk_ack
.quick
) {
331 icsk
->icsk_ack
.quick
= 0;
332 /* Leaving quickack mode we deflate ATO. */
333 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
335 icsk
->icsk_ack
.quick
-= pkts
;
339 extern void tcp_enter_quickack_mode(struct sock
*sk
);
342 #define TCP_ECN_QUEUE_CWR 2
343 #define TCP_ECN_DEMAND_CWR 4
345 static __inline__
void
346 TCP_ECN_create_request(struct request_sock
*req
, struct tcphdr
*th
)
348 if (sysctl_tcp_ecn
&& th
->ece
&& th
->cwr
)
349 inet_rsk(req
)->ecn_ok
= 1;
360 extern enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
362 const struct tcphdr
*th
);
363 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
364 struct request_sock
*req
,
365 struct request_sock
**prev
);
366 extern int tcp_child_process(struct sock
*parent
, struct sock
*child
,
367 struct sk_buff
*skb
);
368 extern int tcp_use_frto(struct sock
*sk
);
369 extern void tcp_enter_frto(struct sock
*sk
);
370 extern void tcp_enter_loss(struct sock
*sk
, int how
);
371 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
372 extern void tcp_update_metrics(struct sock
*sk
);
373 extern void tcp_close(struct sock
*sk
, long timeout
);
374 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
,
375 struct poll_table_struct
*wait
);
376 extern int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
377 char __user
*optval
, int __user
*optlen
);
378 extern int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
379 char __user
*optval
, unsigned int optlen
);
380 extern int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
381 char __user
*optval
, int __user
*optlen
);
382 extern int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
383 char __user
*optval
, unsigned int optlen
);
384 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
385 extern void tcp_syn_ack_timeout(struct sock
*sk
, struct request_sock
*req
);
386 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
387 size_t len
, int nonblock
, int flags
, int *addr_len
);
388 extern void tcp_parse_options(struct sk_buff
*skb
,
389 struct tcp_options_received
*opt_rx
, u8
**hvpp
,
391 extern u8
*tcp_parse_md5sig_option(struct tcphdr
*th
);
394 * TCP v4 functions exported for the inet6 API
397 extern void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
398 extern int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
399 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
400 struct request_sock
*req
,
401 struct sk_buff
*skb
);
402 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
403 struct request_sock
*req
,
404 struct dst_entry
*dst
);
405 extern int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
406 extern int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
408 extern int tcp_connect(struct sock
*sk
);
409 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
410 struct request_sock
*req
,
411 struct request_values
*rvp
);
412 extern int tcp_disconnect(struct sock
*sk
, int flags
);
415 /* From syncookies.c */
416 extern __u32 syncookie_secret
[2][16-4+SHA_DIGEST_WORDS
];
417 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
418 struct ip_options
*opt
);
419 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
422 extern __u32
cookie_init_timestamp(struct request_sock
*req
);
423 extern bool cookie_check_timestamp(struct tcp_options_received
*opt
, bool *);
425 /* From net/ipv6/syncookies.c */
426 extern struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
427 extern __u32
cookie_v6_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
432 extern void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
434 extern int tcp_may_send_now(struct sock
*sk
);
435 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
436 extern void tcp_retransmit_timer(struct sock
*sk
);
437 extern void tcp_xmit_retransmit_queue(struct sock
*);
438 extern void tcp_simple_retransmit(struct sock
*);
439 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
440 extern int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int);
442 extern void tcp_send_probe0(struct sock
*);
443 extern void tcp_send_partial(struct sock
*);
444 extern int tcp_write_wakeup(struct sock
*);
445 extern void tcp_send_fin(struct sock
*sk
);
446 extern void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
447 extern int tcp_send_synack(struct sock
*);
448 extern void tcp_push_one(struct sock
*, unsigned int mss_now
);
449 extern void tcp_send_ack(struct sock
*sk
);
450 extern void tcp_send_delayed_ack(struct sock
*sk
);
453 extern void tcp_cwnd_application_limited(struct sock
*sk
);
456 extern void tcp_init_xmit_timers(struct sock
*);
457 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
459 inet_csk_clear_xmit_timers(sk
);
462 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
463 extern unsigned int tcp_current_mss(struct sock
*sk
);
465 /* Bound MSS / TSO packet size with the half of the window */
466 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
468 if (tp
->max_window
&& pktsize
> (tp
->max_window
>> 1))
469 return max(tp
->max_window
>> 1, 68U - tp
->tcp_header_len
);
475 extern void tcp_get_info(struct sock
*, struct tcp_info
*);
477 /* Read 'sendfile()'-style from a TCP socket */
478 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
479 unsigned int, size_t);
480 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
481 sk_read_actor_t recv_actor
);
483 extern void tcp_initialize_rcv_mss(struct sock
*sk
);
485 extern int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
486 extern int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
487 extern void tcp_mtup_init(struct sock
*sk
);
489 static inline void tcp_bound_rto(const struct sock
*sk
)
491 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
492 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
495 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
497 return (tp
->srtt
>> 3) + tp
->rttvar
;
500 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
502 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
503 ntohl(TCP_FLAG_ACK
) |
507 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
509 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
512 static inline void tcp_fast_path_check(struct sock
*sk
)
514 struct tcp_sock
*tp
= tcp_sk(sk
);
516 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
518 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
520 tcp_fast_path_on(tp
);
523 /* Compute the actual rto_min value */
524 static inline u32
tcp_rto_min(struct sock
*sk
)
526 struct dst_entry
*dst
= __sk_dst_get(sk
);
527 u32 rto_min
= TCP_RTO_MIN
;
529 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
530 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
534 /* Compute the actual receive window we are currently advertising.
535 * Rcv_nxt can be after the window if our peer push more data
536 * than the offered window.
538 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
540 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
547 /* Choose a new window, without checks for shrinking, and without
548 * scaling applied to the result. The caller does these things
549 * if necessary. This is a "raw" window selection.
551 extern u32
__tcp_select_window(struct sock
*sk
);
553 /* TCP timestamps are only 32-bits, this causes a slight
554 * complication on 64-bit systems since we store a snapshot
555 * of jiffies in the buffer control blocks below. We decided
556 * to use only the low 32-bits of jiffies and hide the ugly
557 * casts with the following macro.
559 #define tcp_time_stamp ((__u32)(jiffies))
561 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
563 #define TCPHDR_FIN 0x01
564 #define TCPHDR_SYN 0x02
565 #define TCPHDR_RST 0x04
566 #define TCPHDR_PSH 0x08
567 #define TCPHDR_ACK 0x10
568 #define TCPHDR_URG 0x20
569 #define TCPHDR_ECE 0x40
570 #define TCPHDR_CWR 0x80
572 /* This is what the send packet queuing engine uses to pass
573 * TCP per-packet control information to the transmission
574 * code. We also store the host-order sequence numbers in
575 * here too. This is 36 bytes on 32-bit architectures,
576 * 40 bytes on 64-bit machines, if this grows please adjust
577 * skbuff.h:skbuff->cb[xxx] size appropriately.
581 struct inet_skb_parm h4
;
582 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
583 struct inet6_skb_parm h6
;
585 } header
; /* For incoming frames */
586 __u32 seq
; /* Starting sequence number */
587 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
588 __u32 when
; /* used to compute rtt's */
589 __u8 flags
; /* TCP header flags. */
590 __u8 sacked
; /* State flags for SACK/FACK. */
591 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
592 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
593 #define TCPCB_LOST 0x04 /* SKB is lost */
594 #define TCPCB_TAGBITS 0x07 /* All tag bits */
596 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
597 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
599 __u32 ack_seq
; /* Sequence number ACK'd */
602 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
604 /* Due to TSO, an SKB can be composed of multiple actual
605 * packets. To keep these tracked properly, we use this.
607 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
609 return skb_shinfo(skb
)->gso_segs
;
612 /* This is valid iff tcp_skb_pcount() > 1. */
613 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
615 return skb_shinfo(skb
)->gso_size
;
618 /* Events passed to congestion control interface */
620 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
621 CA_EVENT_CWND_RESTART
, /* congestion window restart */
622 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
623 CA_EVENT_FRTO
, /* fast recovery timeout */
624 CA_EVENT_LOSS
, /* loss timeout */
625 CA_EVENT_FAST_ACK
, /* in sequence ack */
626 CA_EVENT_SLOW_ACK
, /* other ack */
630 * Interface for adding new TCP congestion control handlers
632 #define TCP_CA_NAME_MAX 16
633 #define TCP_CA_MAX 128
634 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
636 #define TCP_CONG_NON_RESTRICTED 0x1
637 #define TCP_CONG_RTT_STAMP 0x2
639 struct tcp_congestion_ops
{
640 struct list_head list
;
643 /* initialize private data (optional) */
644 void (*init
)(struct sock
*sk
);
645 /* cleanup private data (optional) */
646 void (*release
)(struct sock
*sk
);
648 /* return slow start threshold (required) */
649 u32 (*ssthresh
)(struct sock
*sk
);
650 /* lower bound for congestion window (optional) */
651 u32 (*min_cwnd
)(const struct sock
*sk
);
652 /* do new cwnd calculation (required) */
653 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 in_flight
);
654 /* call before changing ca_state (optional) */
655 void (*set_state
)(struct sock
*sk
, u8 new_state
);
656 /* call when cwnd event occurs (optional) */
657 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
658 /* new value of cwnd after loss (optional) */
659 u32 (*undo_cwnd
)(struct sock
*sk
);
660 /* hook for packet ack accounting (optional) */
661 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
662 /* get info for inet_diag (optional) */
663 void (*get_info
)(struct sock
*sk
, u32 ext
, struct sk_buff
*skb
);
665 char name
[TCP_CA_NAME_MAX
];
666 struct module
*owner
;
669 extern int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
670 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
672 extern void tcp_init_congestion_control(struct sock
*sk
);
673 extern void tcp_cleanup_congestion_control(struct sock
*sk
);
674 extern int tcp_set_default_congestion_control(const char *name
);
675 extern void tcp_get_default_congestion_control(char *name
);
676 extern void tcp_get_available_congestion_control(char *buf
, size_t len
);
677 extern void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
678 extern int tcp_set_allowed_congestion_control(char *allowed
);
679 extern int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
680 extern void tcp_slow_start(struct tcp_sock
*tp
);
681 extern void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
);
683 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
684 extern u32
tcp_reno_ssthresh(struct sock
*sk
);
685 extern void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 in_flight
);
686 extern u32
tcp_reno_min_cwnd(const struct sock
*sk
);
687 extern struct tcp_congestion_ops tcp_reno
;
689 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
691 struct inet_connection_sock
*icsk
= inet_csk(sk
);
693 if (icsk
->icsk_ca_ops
->set_state
)
694 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
695 icsk
->icsk_ca_state
= ca_state
;
698 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
700 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
702 if (icsk
->icsk_ca_ops
->cwnd_event
)
703 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
706 /* These functions determine how the current flow behaves in respect of SACK
707 * handling. SACK is negotiated with the peer, and therefore it can vary
708 * between different flows.
710 * tcp_is_sack - SACK enabled
711 * tcp_is_reno - No SACK
712 * tcp_is_fack - FACK enabled, implies SACK enabled
714 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
716 return tp
->rx_opt
.sack_ok
;
719 static inline int tcp_is_reno(const struct tcp_sock
*tp
)
721 return !tcp_is_sack(tp
);
724 static inline int tcp_is_fack(const struct tcp_sock
*tp
)
726 return tp
->rx_opt
.sack_ok
& 2;
729 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
731 tp
->rx_opt
.sack_ok
|= 2;
734 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
736 return tp
->sacked_out
+ tp
->lost_out
;
739 /* This determines how many packets are "in the network" to the best
740 * of our knowledge. In many cases it is conservative, but where
741 * detailed information is available from the receiver (via SACK
742 * blocks etc.) we can make more aggressive calculations.
744 * Use this for decisions involving congestion control, use just
745 * tp->packets_out to determine if the send queue is empty or not.
747 * Read this equation as:
749 * "Packets sent once on transmission queue" MINUS
750 * "Packets left network, but not honestly ACKed yet" PLUS
751 * "Packets fast retransmitted"
753 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
755 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
758 #define TCP_INFINITE_SSTHRESH 0x7fffffff
760 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
762 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
765 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
766 * The exception is rate halving phase, when cwnd is decreasing towards
769 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
771 const struct tcp_sock
*tp
= tcp_sk(sk
);
772 if ((1 << inet_csk(sk
)->icsk_ca_state
) & (TCPF_CA_CWR
| TCPF_CA_Recovery
))
773 return tp
->snd_ssthresh
;
775 return max(tp
->snd_ssthresh
,
776 ((tp
->snd_cwnd
>> 1) +
777 (tp
->snd_cwnd
>> 2)));
780 /* Use define here intentionally to get WARN_ON location shown at the caller */
781 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
783 extern void tcp_enter_cwr(struct sock
*sk
, const int set_ssthresh
);
784 extern __u32
tcp_init_cwnd(struct tcp_sock
*tp
, struct dst_entry
*dst
);
786 /* Slow start with delack produces 3 packets of burst, so that
787 * it is safe "de facto". This will be the default - same as
788 * the default reordering threshold - but if reordering increases,
789 * we must be able to allow cwnd to burst at least this much in order
790 * to not pull it back when holes are filled.
792 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
794 return tp
->reordering
;
797 /* Returns end sequence number of the receiver's advertised window */
798 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
800 return tp
->snd_una
+ tp
->snd_wnd
;
802 extern int tcp_is_cwnd_limited(const struct sock
*sk
, u32 in_flight
);
804 static inline void tcp_minshall_update(struct tcp_sock
*tp
, unsigned int mss
,
805 const struct sk_buff
*skb
)
808 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
811 static inline void tcp_check_probe_timer(struct sock
*sk
)
813 struct tcp_sock
*tp
= tcp_sk(sk
);
814 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
816 if (!tp
->packets_out
&& !icsk
->icsk_pending
)
817 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
818 icsk
->icsk_rto
, TCP_RTO_MAX
);
821 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
826 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
832 * Calculate(/check) TCP checksum
834 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
835 __be32 daddr
, __wsum base
)
837 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
840 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
842 return __skb_checksum_complete(skb
);
845 static inline int tcp_checksum_complete(struct sk_buff
*skb
)
847 return !skb_csum_unnecessary(skb
) &&
848 __tcp_checksum_complete(skb
);
851 /* Prequeue for VJ style copy to user, combined with checksumming. */
853 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
855 tp
->ucopy
.task
= NULL
;
857 tp
->ucopy
.memory
= 0;
858 skb_queue_head_init(&tp
->ucopy
.prequeue
);
859 #ifdef CONFIG_NET_DMA
860 tp
->ucopy
.dma_chan
= NULL
;
861 tp
->ucopy
.wakeup
= 0;
862 tp
->ucopy
.pinned_list
= NULL
;
863 tp
->ucopy
.dma_cookie
= 0;
867 /* Packet is added to VJ-style prequeue for processing in process
868 * context, if a reader task is waiting. Apparently, this exciting
869 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
870 * failed somewhere. Latency? Burstiness? Well, at least now we will
871 * see, why it failed. 8)8) --ANK
873 * NOTE: is this not too big to inline?
875 static inline int tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
877 struct tcp_sock
*tp
= tcp_sk(sk
);
879 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
882 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
883 tp
->ucopy
.memory
+= skb
->truesize
;
884 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
885 struct sk_buff
*skb1
;
887 BUG_ON(sock_owned_by_user(sk
));
889 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
890 sk_backlog_rcv(sk
, skb1
);
891 NET_INC_STATS_BH(sock_net(sk
),
892 LINUX_MIB_TCPPREQUEUEDROPPED
);
895 tp
->ucopy
.memory
= 0;
896 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
897 wake_up_interruptible_sync_poll(sk_sleep(sk
),
898 POLLIN
| POLLRDNORM
| POLLRDBAND
);
899 if (!inet_csk_ack_scheduled(sk
))
900 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
901 (3 * tcp_rto_min(sk
)) / 4,
911 static const char *statename
[]={
912 "Unused","Established","Syn Sent","Syn Recv",
913 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
914 "Close Wait","Last ACK","Listen","Closing"
917 extern void tcp_set_state(struct sock
*sk
, int state
);
919 extern void tcp_done(struct sock
*sk
);
921 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
924 rx_opt
->num_sacks
= 0;
927 /* Determine a window scaling and initial window to offer. */
928 extern void tcp_select_initial_window(int __space
, __u32 mss
,
929 __u32
*rcv_wnd
, __u32
*window_clamp
,
930 int wscale_ok
, __u8
*rcv_wscale
,
933 static inline int tcp_win_from_space(int space
)
935 return sysctl_tcp_adv_win_scale
<=0 ?
936 (space
>>(-sysctl_tcp_adv_win_scale
)) :
937 space
- (space
>>sysctl_tcp_adv_win_scale
);
940 /* Note: caller must be prepared to deal with negative returns */
941 static inline int tcp_space(const struct sock
*sk
)
943 return tcp_win_from_space(sk
->sk_rcvbuf
-
944 atomic_read(&sk
->sk_rmem_alloc
));
947 static inline int tcp_full_space(const struct sock
*sk
)
949 return tcp_win_from_space(sk
->sk_rcvbuf
);
952 static inline void tcp_openreq_init(struct request_sock
*req
,
953 struct tcp_options_received
*rx_opt
,
956 struct inet_request_sock
*ireq
= inet_rsk(req
);
958 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
960 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
961 req
->mss
= rx_opt
->mss_clamp
;
962 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
963 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
964 ireq
->sack_ok
= rx_opt
->sack_ok
;
965 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
966 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
969 ireq
->rmt_port
= tcp_hdr(skb
)->source
;
970 ireq
->loc_port
= tcp_hdr(skb
)->dest
;
973 extern void tcp_enter_memory_pressure(struct sock
*sk
);
975 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
977 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
980 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
982 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
985 static inline int keepalive_probes(const struct tcp_sock
*tp
)
987 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
990 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
992 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
994 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
995 tcp_time_stamp
- tp
->rcv_tstamp
);
998 static inline int tcp_fin_time(const struct sock
*sk
)
1000 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1001 const int rto
= inet_csk(sk
)->icsk_rto
;
1003 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1004 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1009 static inline int tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1012 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1014 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1020 static inline int tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1023 if (tcp_paws_check(rx_opt
, 0))
1026 /* RST segments are not recommended to carry timestamp,
1027 and, if they do, it is recommended to ignore PAWS because
1028 "their cleanup function should take precedence over timestamps."
1029 Certainly, it is mistake. It is necessary to understand the reasons
1030 of this constraint to relax it: if peer reboots, clock may go
1031 out-of-sync and half-open connections will not be reset.
1032 Actually, the problem would be not existing if all
1033 the implementations followed draft about maintaining clock
1034 via reboots. Linux-2.2 DOES NOT!
1036 However, we can relax time bounds for RST segments to MSL.
1038 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1043 #define TCP_CHECK_TIMER(sk) do { } while (0)
1045 static inline void tcp_mib_init(struct net
*net
)
1048 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1049 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1050 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1051 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1055 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1057 tp
->lost_skb_hint
= NULL
;
1058 tp
->scoreboard_skb_hint
= NULL
;
1061 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1063 tcp_clear_retrans_hints_partial(tp
);
1064 tp
->retransmit_skb_hint
= NULL
;
1070 /* - key database */
1071 struct tcp_md5sig_key
{
1076 struct tcp4_md5sig_key
{
1077 struct tcp_md5sig_key base
;
1081 struct tcp6_md5sig_key
{
1082 struct tcp_md5sig_key base
;
1084 u32 scope_id
; /* XXX */
1086 struct in6_addr addr
;
1090 struct tcp_md5sig_info
{
1091 struct tcp4_md5sig_key
*keys4
;
1092 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1093 struct tcp6_md5sig_key
*keys6
;
1101 /* - pseudo header */
1102 struct tcp4_pseudohdr
{
1110 struct tcp6_pseudohdr
{
1111 struct in6_addr saddr
;
1112 struct in6_addr daddr
;
1114 __be32 protocol
; /* including padding */
1117 union tcp_md5sum_block
{
1118 struct tcp4_pseudohdr ip4
;
1119 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1120 struct tcp6_pseudohdr ip6
;
1124 /* - pool: digest algorithm, hash description and scratch buffer */
1125 struct tcp_md5sig_pool
{
1126 struct hash_desc md5_desc
;
1127 union tcp_md5sum_block md5_blk
;
1130 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1133 extern int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1134 struct sock
*sk
, struct request_sock
*req
,
1135 struct sk_buff
*skb
);
1136 extern struct tcp_md5sig_key
* tcp_v4_md5_lookup(struct sock
*sk
,
1137 struct sock
*addr_sk
);
1138 extern int tcp_v4_md5_do_add(struct sock
*sk
, __be32 addr
, u8
*newkey
,
1140 extern int tcp_v4_md5_do_del(struct sock
*sk
, __be32 addr
);
1142 #ifdef CONFIG_TCP_MD5SIG
1143 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1144 &(struct tcp_md5sig_key) { \
1145 .key = (twsk)->tw_md5_key, \
1146 .keylen = (twsk)->tw_md5_keylen, \
1149 #define tcp_twsk_md5_key(twsk) NULL
1152 extern struct tcp_md5sig_pool
* __percpu
*tcp_alloc_md5sig_pool(struct sock
*);
1153 extern void tcp_free_md5sig_pool(void);
1155 extern struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1156 extern void tcp_put_md5sig_pool(void);
1158 extern int tcp_md5_hash_header(struct tcp_md5sig_pool
*, struct tcphdr
*);
1159 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, struct sk_buff
*,
1160 unsigned header_len
);
1161 extern int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1162 struct tcp_md5sig_key
*key
);
1164 /* write queue abstraction */
1165 static inline void tcp_write_queue_purge(struct sock
*sk
)
1167 struct sk_buff
*skb
;
1169 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1170 sk_wmem_free_skb(sk
, skb
);
1172 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1175 static inline struct sk_buff
*tcp_write_queue_head(struct sock
*sk
)
1177 return skb_peek(&sk
->sk_write_queue
);
1180 static inline struct sk_buff
*tcp_write_queue_tail(struct sock
*sk
)
1182 return skb_peek_tail(&sk
->sk_write_queue
);
1185 static inline struct sk_buff
*tcp_write_queue_next(struct sock
*sk
, struct sk_buff
*skb
)
1187 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1190 static inline struct sk_buff
*tcp_write_queue_prev(struct sock
*sk
, struct sk_buff
*skb
)
1192 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1195 #define tcp_for_write_queue(skb, sk) \
1196 skb_queue_walk(&(sk)->sk_write_queue, skb)
1198 #define tcp_for_write_queue_from(skb, sk) \
1199 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1201 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1202 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1204 static inline struct sk_buff
*tcp_send_head(struct sock
*sk
)
1206 return sk
->sk_send_head
;
1209 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1210 const struct sk_buff
*skb
)
1212 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1215 static inline void tcp_advance_send_head(struct sock
*sk
, struct sk_buff
*skb
)
1217 if (tcp_skb_is_last(sk
, skb
))
1218 sk
->sk_send_head
= NULL
;
1220 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1223 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1225 if (sk
->sk_send_head
== skb_unlinked
)
1226 sk
->sk_send_head
= NULL
;
1229 static inline void tcp_init_send_head(struct sock
*sk
)
1231 sk
->sk_send_head
= NULL
;
1234 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1236 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1239 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1241 __tcp_add_write_queue_tail(sk
, skb
);
1243 /* Queue it, remembering where we must start sending. */
1244 if (sk
->sk_send_head
== NULL
) {
1245 sk
->sk_send_head
= skb
;
1247 if (tcp_sk(sk
)->highest_sack
== NULL
)
1248 tcp_sk(sk
)->highest_sack
= skb
;
1252 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1254 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1257 /* Insert buff after skb on the write queue of sk. */
1258 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1259 struct sk_buff
*buff
,
1262 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1265 /* Insert new before skb on the write queue of sk. */
1266 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1267 struct sk_buff
*skb
,
1270 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1272 if (sk
->sk_send_head
== skb
)
1273 sk
->sk_send_head
= new;
1276 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1278 __skb_unlink(skb
, &sk
->sk_write_queue
);
1281 static inline int tcp_write_queue_empty(struct sock
*sk
)
1283 return skb_queue_empty(&sk
->sk_write_queue
);
1286 static inline void tcp_push_pending_frames(struct sock
*sk
)
1288 if (tcp_send_head(sk
)) {
1289 struct tcp_sock
*tp
= tcp_sk(sk
);
1291 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1295 /* Start sequence of the highest skb with SACKed bit, valid only if
1296 * sacked > 0 or when the caller has ensured validity by itself.
1298 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1300 if (!tp
->sacked_out
)
1303 if (tp
->highest_sack
== NULL
)
1306 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1309 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1311 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1312 tcp_write_queue_next(sk
, skb
);
1315 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1317 return tcp_sk(sk
)->highest_sack
;
1320 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1322 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1325 /* Called when old skb is about to be deleted (to be combined with new skb) */
1326 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1327 struct sk_buff
*old
,
1328 struct sk_buff
*new)
1330 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1331 tcp_sk(sk
)->highest_sack
= new;
1334 /* Determines whether this is a thin stream (which may suffer from
1335 * increased latency). Used to trigger latency-reducing mechanisms.
1337 static inline unsigned int tcp_stream_is_thin(struct tcp_sock
*tp
)
1339 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1343 enum tcp_seq_states
{
1344 TCP_SEQ_STATE_LISTENING
,
1345 TCP_SEQ_STATE_OPENREQ
,
1346 TCP_SEQ_STATE_ESTABLISHED
,
1347 TCP_SEQ_STATE_TIME_WAIT
,
1350 struct tcp_seq_afinfo
{
1353 struct file_operations seq_fops
;
1354 struct seq_operations seq_ops
;
1357 struct tcp_iter_state
{
1358 struct seq_net_private p
;
1360 enum tcp_seq_states state
;
1361 struct sock
*syn_wait_sk
;
1362 int bucket
, offset
, sbucket
, num
, uid
;
1366 extern int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1367 extern void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1369 extern struct request_sock_ops tcp_request_sock_ops
;
1370 extern struct request_sock_ops tcp6_request_sock_ops
;
1372 extern void tcp_v4_destroy_sock(struct sock
*sk
);
1374 extern int tcp_v4_gso_send_check(struct sk_buff
*skb
);
1375 extern struct sk_buff
*tcp_tso_segment(struct sk_buff
*skb
, int features
);
1376 extern struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
,
1377 struct sk_buff
*skb
);
1378 extern struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
,
1379 struct sk_buff
*skb
);
1380 extern int tcp_gro_complete(struct sk_buff
*skb
);
1381 extern int tcp4_gro_complete(struct sk_buff
*skb
);
1383 #ifdef CONFIG_PROC_FS
1384 extern int tcp4_proc_init(void);
1385 extern void tcp4_proc_exit(void);
1388 /* TCP af-specific functions */
1389 struct tcp_sock_af_ops
{
1390 #ifdef CONFIG_TCP_MD5SIG
1391 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1392 struct sock
*addr_sk
);
1393 int (*calc_md5_hash
) (char *location
,
1394 struct tcp_md5sig_key
*md5
,
1396 struct request_sock
*req
,
1397 struct sk_buff
*skb
);
1398 int (*md5_add
) (struct sock
*sk
,
1399 struct sock
*addr_sk
,
1402 int (*md5_parse
) (struct sock
*sk
,
1403 char __user
*optval
,
1408 struct tcp_request_sock_ops
{
1409 #ifdef CONFIG_TCP_MD5SIG
1410 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1411 struct request_sock
*req
);
1412 int (*calc_md5_hash
) (char *location
,
1413 struct tcp_md5sig_key
*md5
,
1415 struct request_sock
*req
,
1416 struct sk_buff
*skb
);
1420 /* Using SHA1 for now, define some constants.
1422 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1423 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1424 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1426 extern int tcp_cookie_generator(u32
*bakery
);
1429 * struct tcp_cookie_values - each socket needs extra space for the
1430 * cookies, together with (optional) space for any SYN data.
1432 * A tcp_sock contains a pointer to the current value, and this is
1433 * cloned to the tcp_timewait_sock.
1435 * @cookie_pair: variable data from the option exchange.
1437 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1438 * indicates default (sysctl_tcp_cookie_size).
1439 * After cookie sent, remembers size of cookie.
1440 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1442 * @s_data_desired: user specified tcpct_s_data_desired. When the
1443 * constant payload is specified (@s_data_constant),
1444 * holds its length instead.
1445 * Range 0 to TCP_MSS_DESIRED.
1447 * @s_data_payload: constant data that is to be included in the
1448 * payload of SYN or SYNACK segments when the
1449 * cookie option is present.
1451 struct tcp_cookie_values
{
1453 u8 cookie_pair
[TCP_COOKIE_PAIR_SIZE
];
1454 u8 cookie_pair_size
;
1456 u16 s_data_desired
:11,
1461 u8 s_data_payload
[0];
1464 static inline void tcp_cookie_values_release(struct kref
*kref
)
1466 kfree(container_of(kref
, struct tcp_cookie_values
, kref
));
1469 /* The length of constant payload data. Note that s_data_desired is
1470 * overloaded, depending on s_data_constant: either the length of constant
1471 * data (returned here) or the limit on variable data.
1473 static inline int tcp_s_data_size(const struct tcp_sock
*tp
)
1475 return (tp
->cookie_values
!= NULL
&& tp
->cookie_values
->s_data_constant
)
1476 ? tp
->cookie_values
->s_data_desired
1481 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1483 * As tcp_request_sock has already been extended in other places, the
1484 * only remaining method is to pass stack values along as function
1485 * parameters. These parameters are not needed after sending SYNACK.
1487 * @cookie_bakery: cryptographic secret and message workspace.
1489 * @cookie_plus: bytes in authenticator/cookie option, copied from
1490 * struct tcp_options_received (above).
1492 struct tcp_extend_values
{
1493 struct request_values rv
;
1494 u32 cookie_bakery
[COOKIE_WORKSPACE_WORDS
];
1500 static inline struct tcp_extend_values
*tcp_xv(struct request_values
*rvp
)
1502 return (struct tcp_extend_values
*)rvp
;
1505 extern void tcp_v4_init(void);
1506 extern void tcp_init(void);