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 /* Cancel timers, when they are not required. */
25 #undef TCP_CLEAR_TIMERS
27 #include <linux/config.h>
28 #include <linux/list.h>
29 #include <linux/tcp.h>
30 #include <linux/slab.h>
31 #include <linux/cache.h>
32 #include <linux/percpu.h>
33 #include <net/inet_hashtables.h>
34 #include <net/checksum.h>
35 #include <net/request_sock.h>
39 #include <net/tcp_states.h>
41 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
42 #include <linux/ipv6.h>
44 #include <linux/seq_file.h>
46 extern struct inet_hashinfo tcp_hashinfo
;
48 #if (BITS_PER_LONG == 64)
49 #define TCP_ADDRCMP_ALIGN_BYTES 8
51 #define TCP_ADDRCMP_ALIGN_BYTES 4
54 /* This is a TIME_WAIT bucket. It works around the memory consumption
55 * problems of sockets in such a state on heavily loaded servers, but
56 * without violating the protocol specification.
58 struct tcp_tw_bucket
{
60 * Now struct sock also uses sock_common, so please just
61 * don't add nothing before this first member (__tw_common) --acme
63 struct sock_common __tw_common
;
64 #define tw_family __tw_common.skc_family
65 #define tw_state __tw_common.skc_state
66 #define tw_reuse __tw_common.skc_reuse
67 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
68 #define tw_node __tw_common.skc_node
69 #define tw_bind_node __tw_common.skc_bind_node
70 #define tw_refcnt __tw_common.skc_refcnt
71 volatile unsigned char tw_substate
;
72 unsigned char tw_rcv_wscale
;
74 /* Socket demultiplex comparisons on incoming packets. */
75 /* these five are in inet_sock */
77 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES
)));
81 /* And these are ours. */
88 long tw_ts_recent_stamp
;
90 struct inet_bind_bucket
*tw_tb
;
91 struct hlist_node tw_death_node
;
92 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
93 struct in6_addr tw_v6_daddr
;
94 struct in6_addr tw_v6_rcv_saddr
;
99 static __inline__
void tw_add_node(struct tcp_tw_bucket
*tw
,
100 struct hlist_head
*list
)
102 hlist_add_head(&tw
->tw_node
, list
);
105 static __inline__
void tw_add_bind_node(struct tcp_tw_bucket
*tw
,
106 struct hlist_head
*list
)
108 hlist_add_head(&tw
->tw_bind_node
, list
);
111 static inline int tw_dead_hashed(struct tcp_tw_bucket
*tw
)
113 return tw
->tw_death_node
.pprev
!= NULL
;
116 static __inline__
void tw_dead_node_init(struct tcp_tw_bucket
*tw
)
118 tw
->tw_death_node
.pprev
= NULL
;
121 static __inline__
void __tw_del_dead_node(struct tcp_tw_bucket
*tw
)
123 __hlist_del(&tw
->tw_death_node
);
124 tw_dead_node_init(tw
);
127 static __inline__
int tw_del_dead_node(struct tcp_tw_bucket
*tw
)
129 if (tw_dead_hashed(tw
)) {
130 __tw_del_dead_node(tw
);
136 #define tw_for_each(tw, node, head) \
137 hlist_for_each_entry(tw, node, head, tw_node)
139 #define tw_for_each_inmate(tw, node, jail) \
140 hlist_for_each_entry(tw, node, jail, tw_death_node)
142 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
143 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
145 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
147 static inline u32
tcp_v4_rcv_saddr(const struct sock
*sk
)
149 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
150 inet_sk(sk
)->rcv_saddr
: tcptw_sk(sk
)->tw_rcv_saddr
;
153 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
154 static inline struct in6_addr
*__tcp_v6_rcv_saddr(const struct sock
*sk
)
156 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
157 &inet6_sk(sk
)->rcv_saddr
: &tcptw_sk(sk
)->tw_v6_rcv_saddr
;
160 static inline struct in6_addr
*tcp_v6_rcv_saddr(const struct sock
*sk
)
162 return sk
->sk_family
== AF_INET6
? __tcp_v6_rcv_saddr(sk
) : NULL
;
165 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
167 static inline int tcp_v6_ipv6only(const struct sock
*sk
)
169 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
170 ipv6_only_sock(sk
) : tcptw_sk_ipv6only(sk
);
173 # define __tcp_v6_rcv_saddr(__sk) NULL
174 # define tcp_v6_rcv_saddr(__sk) NULL
175 # define tcptw_sk_ipv6only(__sk) 0
176 # define tcp_v6_ipv6only(__sk) 0
179 extern kmem_cache_t
*tcp_timewait_cachep
;
181 static inline void tcp_tw_put(struct tcp_tw_bucket
*tw
)
183 if (atomic_dec_and_test(&tw
->tw_refcnt
)) {
184 #ifdef SOCK_REFCNT_DEBUG
185 printk(KERN_DEBUG
"tw_bucket %p released\n", tw
);
187 kmem_cache_free(tcp_timewait_cachep
, tw
);
191 extern atomic_t tcp_orphan_count
;
192 extern int tcp_tw_count
;
193 extern void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
194 extern void tcp_tw_deschedule(struct tcp_tw_bucket
*tw
);
197 /* Socket demux engine toys. */
199 #define TCP_COMBINED_PORTS(__sport, __dport) \
200 (((__u32)(__sport)<<16) | (__u32)(__dport))
201 #else /* __LITTLE_ENDIAN */
202 #define TCP_COMBINED_PORTS(__sport, __dport) \
203 (((__u32)(__dport)<<16) | (__u32)(__sport))
206 #if (BITS_PER_LONG == 64)
208 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
209 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
210 #else /* __LITTLE_ENDIAN */
211 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
212 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
213 #endif /* __BIG_ENDIAN */
214 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
215 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
216 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
217 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
218 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
219 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
220 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
221 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
222 #else /* 32-bit arch */
223 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
224 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
225 ((inet_sk(__sk)->daddr == (__saddr)) && \
226 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
227 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
228 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
229 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
230 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
231 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
232 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
233 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
234 #endif /* 64-bit arch */
236 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
237 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
238 ((__sk)->sk_family == AF_INET6) && \
239 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \
240 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
241 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
243 #define MAX_TCP_HEADER (128 + MAX_HEADER)
246 * Never offer a window over 32767 without using window scaling. Some
247 * poor stacks do signed 16bit maths!
249 #define MAX_TCP_WINDOW 32767U
251 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
252 #define TCP_MIN_MSS 88U
254 /* Minimal RCV_MSS. */
255 #define TCP_MIN_RCVMSS 536U
257 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
258 #define TCP_FASTRETRANS_THRESH 3
260 /* Maximal reordering. */
261 #define TCP_MAX_REORDERING 127
263 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
264 #define TCP_MAX_QUICKACKS 16U
266 /* urg_data states */
267 #define TCP_URG_VALID 0x0100
268 #define TCP_URG_NOTYET 0x0200
269 #define TCP_URG_READ 0x0400
271 #define TCP_RETR1 3 /*
272 * This is how many retries it does before it
273 * tries to figure out if the gateway is
274 * down. Minimal RFC value is 3; it corresponds
275 * to ~3sec-8min depending on RTO.
278 #define TCP_RETR2 15 /*
279 * This should take at least
280 * 90 minutes to time out.
281 * RFC1122 says that the limit is 100 sec.
282 * 15 is ~13-30min depending on RTO.
285 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
286 * connection: ~180sec is RFC minumum */
288 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
289 * connection: ~180sec is RFC minumum */
292 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
293 * socket. 7 is ~50sec-16min.
297 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
298 * state, about 60 seconds */
299 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
300 /* BSD style FIN_WAIT2 deadlock breaker.
301 * It used to be 3min, new value is 60sec,
302 * to combine FIN-WAIT-2 timeout with
306 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
308 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
309 #define TCP_ATO_MIN ((unsigned)(HZ/25))
311 #define TCP_DELACK_MIN 4U
312 #define TCP_ATO_MIN 4U
314 #define TCP_RTO_MAX ((unsigned)(120*HZ))
315 #define TCP_RTO_MIN ((unsigned)(HZ/5))
316 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
318 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
319 * for local resources.
322 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
323 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
324 #define TCP_KEEPALIVE_INTVL (75*HZ)
326 #define MAX_TCP_KEEPIDLE 32767
327 #define MAX_TCP_KEEPINTVL 32767
328 #define MAX_TCP_KEEPCNT 127
329 #define MAX_TCP_SYNCNT 127
331 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
332 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
334 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
335 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
336 * after this time. It should be equal
337 * (or greater than) TCP_TIMEWAIT_LEN
338 * to provide reliability equal to one
339 * provided by timewait state.
341 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
342 * timestamps. It must be less than
343 * minimal timewait lifetime.
346 #define TCP_TW_RECYCLE_SLOTS_LOG 5
347 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
349 /* If time > 4sec, it is "slow" path, no recycling is required,
350 so that we select tick to get range about 4 seconds.
353 #if HZ <= 16 || HZ > 4096
354 # error Unsupported: HZ <= 16 or HZ > 4096
356 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
358 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
360 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
362 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
364 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
366 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
368 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
370 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
376 #define TCPOPT_NOP 1 /* Padding */
377 #define TCPOPT_EOL 0 /* End of options */
378 #define TCPOPT_MSS 2 /* Segment size negotiating */
379 #define TCPOPT_WINDOW 3 /* Window scaling */
380 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
381 #define TCPOPT_SACK 5 /* SACK Block */
382 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
388 #define TCPOLEN_MSS 4
389 #define TCPOLEN_WINDOW 3
390 #define TCPOLEN_SACK_PERM 2
391 #define TCPOLEN_TIMESTAMP 10
393 /* But this is what stacks really send out. */
394 #define TCPOLEN_TSTAMP_ALIGNED 12
395 #define TCPOLEN_WSCALE_ALIGNED 4
396 #define TCPOLEN_SACKPERM_ALIGNED 4
397 #define TCPOLEN_SACK_BASE 2
398 #define TCPOLEN_SACK_BASE_ALIGNED 4
399 #define TCPOLEN_SACK_PERBLOCK 8
401 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
402 #define TCP_TIME_DACK 2 /* Delayed ack timer */
403 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
404 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
406 /* Flags in tp->nonagle */
407 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
408 #define TCP_NAGLE_CORK 2 /* Socket is corked */
409 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
411 /* sysctl variables for tcp */
412 extern int sysctl_tcp_timestamps
;
413 extern int sysctl_tcp_window_scaling
;
414 extern int sysctl_tcp_sack
;
415 extern int sysctl_tcp_fin_timeout
;
416 extern int sysctl_tcp_tw_recycle
;
417 extern int sysctl_tcp_keepalive_time
;
418 extern int sysctl_tcp_keepalive_probes
;
419 extern int sysctl_tcp_keepalive_intvl
;
420 extern int sysctl_tcp_syn_retries
;
421 extern int sysctl_tcp_synack_retries
;
422 extern int sysctl_tcp_retries1
;
423 extern int sysctl_tcp_retries2
;
424 extern int sysctl_tcp_orphan_retries
;
425 extern int sysctl_tcp_syncookies
;
426 extern int sysctl_tcp_retrans_collapse
;
427 extern int sysctl_tcp_stdurg
;
428 extern int sysctl_tcp_rfc1337
;
429 extern int sysctl_tcp_abort_on_overflow
;
430 extern int sysctl_tcp_max_orphans
;
431 extern int sysctl_tcp_max_tw_buckets
;
432 extern int sysctl_tcp_fack
;
433 extern int sysctl_tcp_reordering
;
434 extern int sysctl_tcp_ecn
;
435 extern int sysctl_tcp_dsack
;
436 extern int sysctl_tcp_mem
[3];
437 extern int sysctl_tcp_wmem
[3];
438 extern int sysctl_tcp_rmem
[3];
439 extern int sysctl_tcp_app_win
;
440 extern int sysctl_tcp_adv_win_scale
;
441 extern int sysctl_tcp_tw_reuse
;
442 extern int sysctl_tcp_frto
;
443 extern int sysctl_tcp_low_latency
;
444 extern int sysctl_tcp_nometrics_save
;
445 extern int sysctl_tcp_moderate_rcvbuf
;
446 extern int sysctl_tcp_tso_win_divisor
;
448 extern atomic_t tcp_memory_allocated
;
449 extern atomic_t tcp_sockets_allocated
;
450 extern int tcp_memory_pressure
;
452 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
453 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
455 #define TCP_INET_FAMILY(fam) 1
459 * Pointers to address related TCP functions
460 * (i.e. things that depend on the address family)
464 int (*queue_xmit
) (struct sk_buff
*skb
,
467 void (*send_check
) (struct sock
*sk
,
470 struct sk_buff
*skb
);
472 int (*rebuild_header
) (struct sock
*sk
);
474 int (*conn_request
) (struct sock
*sk
,
475 struct sk_buff
*skb
);
477 struct sock
* (*syn_recv_sock
) (struct sock
*sk
,
479 struct request_sock
*req
,
480 struct dst_entry
*dst
);
482 int (*remember_stamp
) (struct sock
*sk
);
484 __u16 net_header_len
;
486 int (*setsockopt
) (struct sock
*sk
,
492 int (*getsockopt
) (struct sock
*sk
,
499 void (*addr2sockaddr
) (struct sock
*sk
,
506 * The next routines deal with comparing 32 bit unsigned ints
507 * and worry about wraparound (automatic with unsigned arithmetic).
510 static inline int before(__u32 seq1
, __u32 seq2
)
512 return (__s32
)(seq1
-seq2
) < 0;
515 static inline int after(__u32 seq1
, __u32 seq2
)
517 return (__s32
)(seq2
-seq1
) < 0;
521 /* is s2<=s1<=s3 ? */
522 static inline int between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
524 return seq3
- seq2
>= seq1
- seq2
;
528 extern struct proto tcp_prot
;
530 DECLARE_SNMP_STAT(struct tcp_mib
, tcp_statistics
);
531 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
532 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
533 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
534 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
535 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
536 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
538 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
540 extern void tcp_shutdown (struct sock
*sk
, int how
);
542 extern int tcp_v4_rcv(struct sk_buff
*skb
);
544 extern int tcp_v4_remember_stamp(struct sock
*sk
);
546 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket
*tw
);
548 extern int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
549 struct msghdr
*msg
, size_t size
);
550 extern ssize_t
tcp_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
);
552 extern int tcp_ioctl(struct sock
*sk
,
556 extern int tcp_rcv_state_process(struct sock
*sk
,
561 extern int tcp_rcv_established(struct sock
*sk
,
566 extern void tcp_rcv_space_adjust(struct sock
*sk
);
575 static inline void tcp_schedule_ack(struct tcp_sock
*tp
)
577 tp
->ack
.pending
|= TCP_ACK_SCHED
;
580 static inline int tcp_ack_scheduled(struct tcp_sock
*tp
)
582 return tp
->ack
.pending
&TCP_ACK_SCHED
;
585 static __inline__
void tcp_dec_quickack_mode(struct tcp_sock
*tp
, unsigned int pkts
)
588 if (pkts
>= tp
->ack
.quick
) {
591 /* Leaving quickack mode we deflate ATO. */
592 tp
->ack
.ato
= TCP_ATO_MIN
;
594 tp
->ack
.quick
-= pkts
;
598 extern void tcp_enter_quickack_mode(struct tcp_sock
*tp
);
600 static __inline__
void tcp_delack_init(struct tcp_sock
*tp
)
602 memset(&tp
->ack
, 0, sizeof(tp
->ack
));
605 static inline void tcp_clear_options(struct tcp_options_received
*rx_opt
)
607 rx_opt
->tstamp_ok
= rx_opt
->sack_ok
= rx_opt
->wscale_ok
= rx_opt
->snd_wscale
= 0;
619 extern enum tcp_tw_status
tcp_timewait_state_process(struct tcp_tw_bucket
*tw
,
624 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
625 struct request_sock
*req
,
626 struct request_sock
**prev
);
627 extern int tcp_child_process(struct sock
*parent
,
629 struct sk_buff
*skb
);
630 extern void tcp_enter_frto(struct sock
*sk
);
631 extern void tcp_enter_loss(struct sock
*sk
, int how
);
632 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
633 extern void tcp_update_metrics(struct sock
*sk
);
635 extern void tcp_close(struct sock
*sk
,
637 extern struct sock
* tcp_accept(struct sock
*sk
, int flags
, int *err
);
638 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
, struct poll_table_struct
*wait
);
640 extern int tcp_getsockopt(struct sock
*sk
, int level
,
644 extern int tcp_setsockopt(struct sock
*sk
, int level
,
645 int optname
, char __user
*optval
,
647 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
648 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
650 size_t len
, int nonblock
,
651 int flags
, int *addr_len
);
653 extern int tcp_listen_start(struct sock
*sk
);
655 extern void tcp_parse_options(struct sk_buff
*skb
,
656 struct tcp_options_received
*opt_rx
,
660 * TCP v4 functions exported for the inet6 API
663 extern void tcp_v4_send_check(struct sock
*sk
,
664 struct tcphdr
*th
, int len
,
665 struct sk_buff
*skb
);
667 extern int tcp_v4_conn_request(struct sock
*sk
,
668 struct sk_buff
*skb
);
670 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
671 struct request_sock
*req
,
672 struct sk_buff
*skb
);
674 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
,
676 struct request_sock
*req
,
677 struct dst_entry
*dst
);
679 extern int tcp_v4_do_rcv(struct sock
*sk
,
680 struct sk_buff
*skb
);
682 extern int tcp_v4_connect(struct sock
*sk
,
683 struct sockaddr
*uaddr
,
686 extern int tcp_connect(struct sock
*sk
);
688 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
,
689 struct dst_entry
*dst
,
690 struct request_sock
*req
);
692 extern int tcp_disconnect(struct sock
*sk
, int flags
);
694 extern void tcp_unhash(struct sock
*sk
);
696 extern int tcp_v4_hash_connecting(struct sock
*sk
);
699 /* From syncookies.c */
700 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
701 struct ip_options
*opt
);
702 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
707 extern void __tcp_push_pending_frames(struct sock
*sk
, struct tcp_sock
*tp
,
708 unsigned int cur_mss
, int nonagle
);
709 extern int tcp_may_send_now(struct sock
*sk
, struct tcp_sock
*tp
);
710 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
711 extern void tcp_xmit_retransmit_queue(struct sock
*);
712 extern void tcp_simple_retransmit(struct sock
*);
713 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
715 extern void tcp_send_probe0(struct sock
*);
716 extern void tcp_send_partial(struct sock
*);
717 extern int tcp_write_wakeup(struct sock
*);
718 extern void tcp_send_fin(struct sock
*sk
);
719 extern void tcp_send_active_reset(struct sock
*sk
,
720 unsigned int __nocast priority
);
721 extern int tcp_send_synack(struct sock
*);
722 extern void tcp_push_one(struct sock
*, unsigned int mss_now
);
723 extern void tcp_send_ack(struct sock
*sk
);
724 extern void tcp_send_delayed_ack(struct sock
*sk
);
727 extern void tcp_cwnd_application_limited(struct sock
*sk
);
730 extern void tcp_init_xmit_timers(struct sock
*);
731 extern void tcp_clear_xmit_timers(struct sock
*);
733 extern void tcp_delete_keepalive_timer(struct sock
*);
734 extern void tcp_reset_keepalive_timer(struct sock
*, unsigned long);
735 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
736 extern unsigned int tcp_current_mss(struct sock
*sk
, int large
);
739 extern const char tcp_timer_bug_msg
[];
743 extern void tcp_get_info(struct sock
*, struct tcp_info
*);
745 /* Read 'sendfile()'-style from a TCP socket */
746 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
747 unsigned int, size_t);
748 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
749 sk_read_actor_t recv_actor
);
751 static inline void tcp_clear_xmit_timer(struct sock
*sk
, int what
)
753 struct tcp_sock
*tp
= tcp_sk(sk
);
756 case TCP_TIME_RETRANS
:
757 case TCP_TIME_PROBE0
:
760 #ifdef TCP_CLEAR_TIMERS
761 sk_stop_timer(sk
, &tp
->retransmit_timer
);
768 #ifdef TCP_CLEAR_TIMERS
769 sk_stop_timer(sk
, &tp
->delack_timer
);
774 printk(tcp_timer_bug_msg
);
782 * Reset the retransmission timer
784 static inline void tcp_reset_xmit_timer(struct sock
*sk
, int what
, unsigned long when
)
786 struct tcp_sock
*tp
= tcp_sk(sk
);
788 if (when
> TCP_RTO_MAX
) {
790 printk(KERN_DEBUG
"reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk
, what
, when
, current_text_addr());
796 case TCP_TIME_RETRANS
:
797 case TCP_TIME_PROBE0
:
799 tp
->timeout
= jiffies
+when
;
800 sk_reset_timer(sk
, &tp
->retransmit_timer
, tp
->timeout
);
804 tp
->ack
.pending
|= TCP_ACK_TIMER
;
805 tp
->ack
.timeout
= jiffies
+when
;
806 sk_reset_timer(sk
, &tp
->delack_timer
, tp
->ack
.timeout
);
811 printk(tcp_timer_bug_msg
);
817 /* Initialize RCV_MSS value.
818 * RCV_MSS is an our guess about MSS used by the peer.
819 * We haven't any direct information about the MSS.
820 * It's better to underestimate the RCV_MSS rather than overestimate.
821 * Overestimations make us ACKing less frequently than needed.
822 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
825 static inline void tcp_initialize_rcv_mss(struct sock
*sk
)
827 struct tcp_sock
*tp
= tcp_sk(sk
);
828 unsigned int hint
= min_t(unsigned int, tp
->advmss
, tp
->mss_cache
);
830 hint
= min(hint
, tp
->rcv_wnd
/2);
831 hint
= min(hint
, TCP_MIN_RCVMSS
);
832 hint
= max(hint
, TCP_MIN_MSS
);
834 tp
->ack
.rcv_mss
= hint
;
837 static __inline__
void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
839 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
840 ntohl(TCP_FLAG_ACK
) |
844 static __inline__
void tcp_fast_path_on(struct tcp_sock
*tp
)
846 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
849 static inline void tcp_fast_path_check(struct sock
*sk
, struct tcp_sock
*tp
)
851 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
853 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
855 tcp_fast_path_on(tp
);
858 /* Compute the actual receive window we are currently advertising.
859 * Rcv_nxt can be after the window if our peer push more data
860 * than the offered window.
862 static __inline__ u32
tcp_receive_window(const struct tcp_sock
*tp
)
864 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
871 /* Choose a new window, without checks for shrinking, and without
872 * scaling applied to the result. The caller does these things
873 * if necessary. This is a "raw" window selection.
875 extern u32
__tcp_select_window(struct sock
*sk
);
877 /* TCP timestamps are only 32-bits, this causes a slight
878 * complication on 64-bit systems since we store a snapshot
879 * of jiffies in the buffer control blocks below. We decidely
880 * only use of the low 32-bits of jiffies and hide the ugly
881 * casts with the following macro.
883 #define tcp_time_stamp ((__u32)(jiffies))
885 /* This is what the send packet queueing engine uses to pass
886 * TCP per-packet control information to the transmission
887 * code. We also store the host-order sequence numbers in
888 * here too. This is 36 bytes on 32-bit architectures,
889 * 40 bytes on 64-bit machines, if this grows please adjust
890 * skbuff.h:skbuff->cb[xxx] size appropriately.
894 struct inet_skb_parm h4
;
895 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
896 struct inet6_skb_parm h6
;
898 } header
; /* For incoming frames */
899 __u32 seq
; /* Starting sequence number */
900 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
901 __u32 when
; /* used to compute rtt's */
902 __u8 flags
; /* TCP header flags. */
904 /* NOTE: These must match up to the flags byte in a
907 #define TCPCB_FLAG_FIN 0x01
908 #define TCPCB_FLAG_SYN 0x02
909 #define TCPCB_FLAG_RST 0x04
910 #define TCPCB_FLAG_PSH 0x08
911 #define TCPCB_FLAG_ACK 0x10
912 #define TCPCB_FLAG_URG 0x20
913 #define TCPCB_FLAG_ECE 0x40
914 #define TCPCB_FLAG_CWR 0x80
916 __u8 sacked
; /* State flags for SACK/FACK. */
917 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
918 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
919 #define TCPCB_LOST 0x04 /* SKB is lost */
920 #define TCPCB_TAGBITS 0x07 /* All tag bits */
922 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
923 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
925 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
927 #define TCPCB_AT_TAIL (TCPCB_URG)
929 __u16 urg_ptr
; /* Valid w/URG flags is set. */
930 __u32 ack_seq
; /* Sequence number ACK'd */
933 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
935 #include <net/tcp_ecn.h>
937 /* Due to TSO, an SKB can be composed of multiple actual
938 * packets. To keep these tracked properly, we use this.
940 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
942 return skb_shinfo(skb
)->tso_segs
;
945 /* This is valid iff tcp_skb_pcount() > 1. */
946 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
948 return skb_shinfo(skb
)->tso_size
;
951 static inline void tcp_dec_pcount_approx(__u32
*count
,
952 const struct sk_buff
*skb
)
955 *count
-= tcp_skb_pcount(skb
);
961 static inline void tcp_packets_out_inc(struct sock
*sk
,
963 const struct sk_buff
*skb
)
965 int orig
= tp
->packets_out
;
967 tp
->packets_out
+= tcp_skb_pcount(skb
);
969 tcp_reset_xmit_timer(sk
, TCP_TIME_RETRANS
, tp
->rto
);
972 static inline void tcp_packets_out_dec(struct tcp_sock
*tp
,
973 const struct sk_buff
*skb
)
975 tp
->packets_out
-= tcp_skb_pcount(skb
);
978 /* Events passed to congestion control interface */
980 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
981 CA_EVENT_CWND_RESTART
, /* congestion window restart */
982 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
983 CA_EVENT_FRTO
, /* fast recovery timeout */
984 CA_EVENT_LOSS
, /* loss timeout */
985 CA_EVENT_FAST_ACK
, /* in sequence ack */
986 CA_EVENT_SLOW_ACK
, /* other ack */
990 * Interface for adding new TCP congestion control handlers
992 #define TCP_CA_NAME_MAX 16
993 struct tcp_congestion_ops
{
994 struct list_head list
;
996 /* initialize private data (optional) */
997 void (*init
)(struct tcp_sock
*tp
);
998 /* cleanup private data (optional) */
999 void (*release
)(struct tcp_sock
*tp
);
1001 /* return slow start threshold (required) */
1002 u32 (*ssthresh
)(struct tcp_sock
*tp
);
1003 /* lower bound for congestion window (optional) */
1004 u32 (*min_cwnd
)(struct tcp_sock
*tp
);
1005 /* do new cwnd calculation (required) */
1006 void (*cong_avoid
)(struct tcp_sock
*tp
, u32 ack
,
1007 u32 rtt
, u32 in_flight
, int good_ack
);
1008 /* round trip time sample per acked packet (optional) */
1009 void (*rtt_sample
)(struct tcp_sock
*tp
, u32 usrtt
);
1010 /* call before changing ca_state (optional) */
1011 void (*set_state
)(struct tcp_sock
*tp
, u8 new_state
);
1012 /* call when cwnd event occurs (optional) */
1013 void (*cwnd_event
)(struct tcp_sock
*tp
, enum tcp_ca_event ev
);
1014 /* new value of cwnd after loss (optional) */
1015 u32 (*undo_cwnd
)(struct tcp_sock
*tp
);
1016 /* hook for packet ack accounting (optional) */
1017 void (*pkts_acked
)(struct tcp_sock
*tp
, u32 num_acked
);
1018 /* get info for tcp_diag (optional) */
1019 void (*get_info
)(struct tcp_sock
*tp
, u32 ext
, struct sk_buff
*skb
);
1021 char name
[TCP_CA_NAME_MAX
];
1022 struct module
*owner
;
1025 extern int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1026 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1028 extern void tcp_init_congestion_control(struct tcp_sock
*tp
);
1029 extern void tcp_cleanup_congestion_control(struct tcp_sock
*tp
);
1030 extern int tcp_set_default_congestion_control(const char *name
);
1031 extern void tcp_get_default_congestion_control(char *name
);
1032 extern int tcp_set_congestion_control(struct tcp_sock
*tp
, const char *name
);
1034 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
1035 extern u32
tcp_reno_ssthresh(struct tcp_sock
*tp
);
1036 extern void tcp_reno_cong_avoid(struct tcp_sock
*tp
, u32 ack
,
1037 u32 rtt
, u32 in_flight
, int flag
);
1038 extern u32
tcp_reno_min_cwnd(struct tcp_sock
*tp
);
1039 extern struct tcp_congestion_ops tcp_reno
;
1041 static inline void tcp_set_ca_state(struct tcp_sock
*tp
, u8 ca_state
)
1043 if (tp
->ca_ops
->set_state
)
1044 tp
->ca_ops
->set_state(tp
, ca_state
);
1045 tp
->ca_state
= ca_state
;
1048 static inline void tcp_ca_event(struct tcp_sock
*tp
, enum tcp_ca_event event
)
1050 if (tp
->ca_ops
->cwnd_event
)
1051 tp
->ca_ops
->cwnd_event(tp
, event
);
1054 /* This determines how many packets are "in the network" to the best
1055 * of our knowledge. In many cases it is conservative, but where
1056 * detailed information is available from the receiver (via SACK
1057 * blocks etc.) we can make more aggressive calculations.
1059 * Use this for decisions involving congestion control, use just
1060 * tp->packets_out to determine if the send queue is empty or not.
1062 * Read this equation as:
1064 * "Packets sent once on transmission queue" MINUS
1065 * "Packets left network, but not honestly ACKed yet" PLUS
1066 * "Packets fast retransmitted"
1068 static __inline__
unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1070 return (tp
->packets_out
- tp
->left_out
+ tp
->retrans_out
);
1073 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1074 * The exception is rate halving phase, when cwnd is decreasing towards
1077 static inline __u32
tcp_current_ssthresh(struct tcp_sock
*tp
)
1079 if ((1<<tp
->ca_state
)&(TCPF_CA_CWR
|TCPF_CA_Recovery
))
1080 return tp
->snd_ssthresh
;
1082 return max(tp
->snd_ssthresh
,
1083 ((tp
->snd_cwnd
>> 1) +
1084 (tp
->snd_cwnd
>> 2)));
1087 static inline void tcp_sync_left_out(struct tcp_sock
*tp
)
1089 if (tp
->rx_opt
.sack_ok
&&
1090 (tp
->sacked_out
>= tp
->packets_out
- tp
->lost_out
))
1091 tp
->sacked_out
= tp
->packets_out
- tp
->lost_out
;
1092 tp
->left_out
= tp
->sacked_out
+ tp
->lost_out
;
1095 /* Set slow start threshold and cwnd not falling to slow start */
1096 static inline void __tcp_enter_cwr(struct tcp_sock
*tp
)
1098 tp
->undo_marker
= 0;
1099 tp
->snd_ssthresh
= tp
->ca_ops
->ssthresh(tp
);
1100 tp
->snd_cwnd
= min(tp
->snd_cwnd
,
1101 tcp_packets_in_flight(tp
) + 1U);
1102 tp
->snd_cwnd_cnt
= 0;
1103 tp
->high_seq
= tp
->snd_nxt
;
1104 tp
->snd_cwnd_stamp
= tcp_time_stamp
;
1105 TCP_ECN_queue_cwr(tp
);
1108 static inline void tcp_enter_cwr(struct tcp_sock
*tp
)
1110 tp
->prior_ssthresh
= 0;
1111 if (tp
->ca_state
< TCP_CA_CWR
) {
1112 __tcp_enter_cwr(tp
);
1113 tcp_set_ca_state(tp
, TCP_CA_CWR
);
1117 extern __u32
tcp_init_cwnd(struct tcp_sock
*tp
, struct dst_entry
*dst
);
1119 /* Slow start with delack produces 3 packets of burst, so that
1120 * it is safe "de facto".
1122 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
1127 static __inline__
void tcp_minshall_update(struct tcp_sock
*tp
, int mss
,
1128 const struct sk_buff
*skb
)
1131 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
1134 static __inline__
void tcp_check_probe_timer(struct sock
*sk
, struct tcp_sock
*tp
)
1136 if (!tp
->packets_out
&& !tp
->pending
)
1137 tcp_reset_xmit_timer(sk
, TCP_TIME_PROBE0
, tp
->rto
);
1140 static __inline__
void tcp_push_pending_frames(struct sock
*sk
,
1141 struct tcp_sock
*tp
)
1143 __tcp_push_pending_frames(sk
, tp
, tcp_current_mss(sk
, 1), tp
->nonagle
);
1146 static __inline__
void tcp_init_wl(struct tcp_sock
*tp
, u32 ack
, u32 seq
)
1151 static __inline__
void tcp_update_wl(struct tcp_sock
*tp
, u32 ack
, u32 seq
)
1156 extern void tcp_destroy_sock(struct sock
*sk
);
1160 * Calculate(/check) TCP checksum
1162 static __inline__ u16
tcp_v4_check(struct tcphdr
*th
, int len
,
1163 unsigned long saddr
, unsigned long daddr
,
1166 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1169 static __inline__
int __tcp_checksum_complete(struct sk_buff
*skb
)
1171 return (unsigned short)csum_fold(skb_checksum(skb
, 0, skb
->len
, skb
->csum
));
1174 static __inline__
int tcp_checksum_complete(struct sk_buff
*skb
)
1176 return skb
->ip_summed
!= CHECKSUM_UNNECESSARY
&&
1177 __tcp_checksum_complete(skb
);
1180 /* Prequeue for VJ style copy to user, combined with checksumming. */
1182 static __inline__
void tcp_prequeue_init(struct tcp_sock
*tp
)
1184 tp
->ucopy
.task
= NULL
;
1186 tp
->ucopy
.memory
= 0;
1187 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1190 /* Packet is added to VJ-style prequeue for processing in process
1191 * context, if a reader task is waiting. Apparently, this exciting
1192 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1193 * failed somewhere. Latency? Burstiness? Well, at least now we will
1194 * see, why it failed. 8)8) --ANK
1196 * NOTE: is this not too big to inline?
1198 static __inline__
int tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1200 struct tcp_sock
*tp
= tcp_sk(sk
);
1202 if (!sysctl_tcp_low_latency
&& tp
->ucopy
.task
) {
1203 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1204 tp
->ucopy
.memory
+= skb
->truesize
;
1205 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1206 struct sk_buff
*skb1
;
1208 BUG_ON(sock_owned_by_user(sk
));
1210 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1211 sk
->sk_backlog_rcv(sk
, skb1
);
1212 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED
);
1215 tp
->ucopy
.memory
= 0;
1216 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1217 wake_up_interruptible(sk
->sk_sleep
);
1218 if (!tcp_ack_scheduled(tp
))
1219 tcp_reset_xmit_timer(sk
, TCP_TIME_DACK
, (3*TCP_RTO_MIN
)/4);
1230 static const char *statename
[]={
1231 "Unused","Established","Syn Sent","Syn Recv",
1232 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1233 "Close Wait","Last ACK","Listen","Closing"
1237 static __inline__
void tcp_set_state(struct sock
*sk
, int state
)
1239 int oldstate
= sk
->sk_state
;
1242 case TCP_ESTABLISHED
:
1243 if (oldstate
!= TCP_ESTABLISHED
)
1244 TCP_INC_STATS(TCP_MIB_CURRESTAB
);
1248 if (oldstate
== TCP_CLOSE_WAIT
|| oldstate
== TCP_ESTABLISHED
)
1249 TCP_INC_STATS(TCP_MIB_ESTABRESETS
);
1251 sk
->sk_prot
->unhash(sk
);
1252 if (inet_sk(sk
)->bind_hash
&&
1253 !(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
))
1254 inet_put_port(&tcp_hashinfo
, sk
);
1257 if (oldstate
==TCP_ESTABLISHED
)
1258 TCP_DEC_STATS(TCP_MIB_CURRESTAB
);
1261 /* Change state AFTER socket is unhashed to avoid closed
1262 * socket sitting in hash tables.
1264 sk
->sk_state
= state
;
1267 SOCK_DEBUG(sk
, "TCP sk=%p, State %s -> %s\n",sk
, statename
[oldstate
],statename
[state
]);
1271 static __inline__
void tcp_done(struct sock
*sk
)
1273 tcp_set_state(sk
, TCP_CLOSE
);
1274 tcp_clear_xmit_timers(sk
);
1276 sk
->sk_shutdown
= SHUTDOWN_MASK
;
1278 if (!sock_flag(sk
, SOCK_DEAD
))
1279 sk
->sk_state_change(sk
);
1281 tcp_destroy_sock(sk
);
1284 static __inline__
void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1287 rx_opt
->eff_sacks
= 0;
1288 rx_opt
->num_sacks
= 0;
1291 static __inline__
void tcp_build_and_update_options(__u32
*ptr
, struct tcp_sock
*tp
, __u32 tstamp
)
1293 if (tp
->rx_opt
.tstamp_ok
) {
1294 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) |
1295 (TCPOPT_NOP
<< 16) |
1296 (TCPOPT_TIMESTAMP
<< 8) |
1298 *ptr
++ = htonl(tstamp
);
1299 *ptr
++ = htonl(tp
->rx_opt
.ts_recent
);
1301 if (tp
->rx_opt
.eff_sacks
) {
1302 struct tcp_sack_block
*sp
= tp
->rx_opt
.dsack
? tp
->duplicate_sack
: tp
->selective_acks
;
1305 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) |
1306 (TCPOPT_NOP
<< 16) |
1307 (TCPOPT_SACK
<< 8) |
1308 (TCPOLEN_SACK_BASE
+
1309 (tp
->rx_opt
.eff_sacks
* TCPOLEN_SACK_PERBLOCK
)));
1310 for(this_sack
= 0; this_sack
< tp
->rx_opt
.eff_sacks
; this_sack
++) {
1311 *ptr
++ = htonl(sp
[this_sack
].start_seq
);
1312 *ptr
++ = htonl(sp
[this_sack
].end_seq
);
1314 if (tp
->rx_opt
.dsack
) {
1315 tp
->rx_opt
.dsack
= 0;
1316 tp
->rx_opt
.eff_sacks
--;
1321 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1322 * If this is every changed make sure to change the definition of
1323 * MAX_SYN_SIZE to match the new maximum number of options that you
1326 static inline void tcp_syn_build_options(__u32
*ptr
, int mss
, int ts
, int sack
,
1327 int offer_wscale
, int wscale
, __u32 tstamp
, __u32 ts_recent
)
1329 /* We always get an MSS option.
1330 * The option bytes which will be seen in normal data
1331 * packets should timestamps be used, must be in the MSS
1332 * advertised. But we subtract them from tp->mss_cache so
1333 * that calculations in tcp_sendmsg are simpler etc.
1334 * So account for this fact here if necessary. If we
1335 * don't do this correctly, as a receiver we won't
1336 * recognize data packets as being full sized when we
1337 * should, and thus we won't abide by the delayed ACK
1339 * SACKs don't matter, we never delay an ACK when we
1340 * have any of those going out.
1342 *ptr
++ = htonl((TCPOPT_MSS
<< 24) | (TCPOLEN_MSS
<< 16) | mss
);
1345 *ptr
++ = __constant_htonl((TCPOPT_SACK_PERM
<< 24) | (TCPOLEN_SACK_PERM
<< 16) |
1346 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
1348 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
1349 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
1350 *ptr
++ = htonl(tstamp
); /* TSVAL */
1351 *ptr
++ = htonl(ts_recent
); /* TSECR */
1353 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
1354 (TCPOPT_SACK_PERM
<< 8) | TCPOLEN_SACK_PERM
);
1356 *ptr
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_WINDOW
<< 16) | (TCPOLEN_WINDOW
<< 8) | (wscale
));
1359 /* Determine a window scaling and initial window to offer. */
1360 extern void tcp_select_initial_window(int __space
, __u32 mss
,
1361 __u32
*rcv_wnd
, __u32
*window_clamp
,
1362 int wscale_ok
, __u8
*rcv_wscale
);
1364 static inline int tcp_win_from_space(int space
)
1366 return sysctl_tcp_adv_win_scale
<=0 ?
1367 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1368 space
- (space
>>sysctl_tcp_adv_win_scale
);
1371 /* Note: caller must be prepared to deal with negative returns */
1372 static inline int tcp_space(const struct sock
*sk
)
1374 return tcp_win_from_space(sk
->sk_rcvbuf
-
1375 atomic_read(&sk
->sk_rmem_alloc
));
1378 static inline int tcp_full_space(const struct sock
*sk
)
1380 return tcp_win_from_space(sk
->sk_rcvbuf
);
1383 static inline void tcp_acceptq_queue(struct sock
*sk
, struct request_sock
*req
,
1386 reqsk_queue_add(&tcp_sk(sk
)->accept_queue
, req
, sk
, child
);
1390 tcp_synq_removed(struct sock
*sk
, struct request_sock
*req
)
1392 if (reqsk_queue_removed(&tcp_sk(sk
)->accept_queue
, req
) == 0)
1393 tcp_delete_keepalive_timer(sk
);
1396 static inline void tcp_synq_added(struct sock
*sk
)
1398 if (reqsk_queue_added(&tcp_sk(sk
)->accept_queue
) == 0)
1399 tcp_reset_keepalive_timer(sk
, TCP_TIMEOUT_INIT
);
1402 static inline int tcp_synq_len(struct sock
*sk
)
1404 return reqsk_queue_len(&tcp_sk(sk
)->accept_queue
);
1407 static inline int tcp_synq_young(struct sock
*sk
)
1409 return reqsk_queue_len_young(&tcp_sk(sk
)->accept_queue
);
1412 static inline int tcp_synq_is_full(struct sock
*sk
)
1414 return reqsk_queue_is_full(&tcp_sk(sk
)->accept_queue
);
1417 static inline void tcp_synq_unlink(struct tcp_sock
*tp
, struct request_sock
*req
,
1418 struct request_sock
**prev
)
1420 reqsk_queue_unlink(&tp
->accept_queue
, req
, prev
);
1423 static inline void tcp_synq_drop(struct sock
*sk
, struct request_sock
*req
,
1424 struct request_sock
**prev
)
1426 tcp_synq_unlink(tcp_sk(sk
), req
, prev
);
1427 tcp_synq_removed(sk
, req
);
1431 static __inline__
void tcp_openreq_init(struct request_sock
*req
,
1432 struct tcp_options_received
*rx_opt
,
1433 struct sk_buff
*skb
)
1435 struct inet_request_sock
*ireq
= inet_rsk(req
);
1437 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1438 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1439 req
->mss
= rx_opt
->mss_clamp
;
1440 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1441 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1442 ireq
->sack_ok
= rx_opt
->sack_ok
;
1443 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1444 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1447 ireq
->rmt_port
= skb
->h
.th
->source
;
1450 extern void tcp_enter_memory_pressure(void);
1452 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1454 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1457 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1459 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1462 static inline int tcp_fin_time(const struct tcp_sock
*tp
)
1464 int fin_timeout
= tp
->linger2
? : sysctl_tcp_fin_timeout
;
1466 if (fin_timeout
< (tp
->rto
<<2) - (tp
->rto
>>1))
1467 fin_timeout
= (tp
->rto
<<2) - (tp
->rto
>>1);
1472 static inline int tcp_paws_check(const struct tcp_options_received
*rx_opt
, int rst
)
1474 if ((s32
)(rx_opt
->rcv_tsval
- rx_opt
->ts_recent
) >= 0)
1476 if (xtime
.tv_sec
>= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
)
1479 /* RST segments are not recommended to carry timestamp,
1480 and, if they do, it is recommended to ignore PAWS because
1481 "their cleanup function should take precedence over timestamps."
1482 Certainly, it is mistake. It is necessary to understand the reasons
1483 of this constraint to relax it: if peer reboots, clock may go
1484 out-of-sync and half-open connections will not be reset.
1485 Actually, the problem would be not existing if all
1486 the implementations followed draft about maintaining clock
1487 via reboots. Linux-2.2 DOES NOT!
1489 However, we can relax time bounds for RST segments to MSL.
1491 if (rst
&& xtime
.tv_sec
>= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1496 #define TCP_CHECK_TIMER(sk) do { } while (0)
1498 static inline int tcp_use_frto(const struct sock
*sk
)
1500 const struct tcp_sock
*tp
= tcp_sk(sk
);
1502 /* F-RTO must be activated in sysctl and there must be some
1503 * unsent new data, and the advertised window should allow
1506 return (sysctl_tcp_frto
&& sk
->sk_send_head
&&
1507 !after(TCP_SKB_CB(sk
->sk_send_head
)->end_seq
,
1508 tp
->snd_una
+ tp
->snd_wnd
));
1511 static inline void tcp_mib_init(void)
1514 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM
, 1);
1515 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1516 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1517 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN
, -1);
1521 enum tcp_seq_states
{
1522 TCP_SEQ_STATE_LISTENING
,
1523 TCP_SEQ_STATE_OPENREQ
,
1524 TCP_SEQ_STATE_ESTABLISHED
,
1525 TCP_SEQ_STATE_TIME_WAIT
,
1528 struct tcp_seq_afinfo
{
1529 struct module
*owner
;
1532 int (*seq_show
) (struct seq_file
*m
, void *v
);
1533 struct file_operations
*seq_fops
;
1536 struct tcp_iter_state
{
1538 enum tcp_seq_states state
;
1539 struct sock
*syn_wait_sk
;
1540 int bucket
, sbucket
, num
, uid
;
1541 struct seq_operations seq_ops
;
1544 extern int tcp_proc_register(struct tcp_seq_afinfo
*afinfo
);
1545 extern void tcp_proc_unregister(struct tcp_seq_afinfo
*afinfo
);