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 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
40 #include <linux/ipv6.h>
42 #include <linux/seq_file.h>
44 extern struct inet_hashinfo tcp_hashinfo
;
46 #if (BITS_PER_LONG == 64)
47 #define TCP_ADDRCMP_ALIGN_BYTES 8
49 #define TCP_ADDRCMP_ALIGN_BYTES 4
52 /* This is a TIME_WAIT bucket. It works around the memory consumption
53 * problems of sockets in such a state on heavily loaded servers, but
54 * without violating the protocol specification.
56 struct tcp_tw_bucket
{
58 * Now struct sock also uses sock_common, so please just
59 * don't add nothing before this first member (__tw_common) --acme
61 struct sock_common __tw_common
;
62 #define tw_family __tw_common.skc_family
63 #define tw_state __tw_common.skc_state
64 #define tw_reuse __tw_common.skc_reuse
65 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
66 #define tw_node __tw_common.skc_node
67 #define tw_bind_node __tw_common.skc_bind_node
68 #define tw_refcnt __tw_common.skc_refcnt
69 volatile unsigned char tw_substate
;
70 unsigned char tw_rcv_wscale
;
72 /* Socket demultiplex comparisons on incoming packets. */
73 /* these five are in inet_sock */
75 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES
)));
79 /* And these are ours. */
86 long tw_ts_recent_stamp
;
88 struct inet_bind_bucket
*tw_tb
;
89 struct hlist_node tw_death_node
;
90 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
91 struct in6_addr tw_v6_daddr
;
92 struct in6_addr tw_v6_rcv_saddr
;
97 static __inline__
void tw_add_node(struct tcp_tw_bucket
*tw
,
98 struct hlist_head
*list
)
100 hlist_add_head(&tw
->tw_node
, list
);
103 static __inline__
void tw_add_bind_node(struct tcp_tw_bucket
*tw
,
104 struct hlist_head
*list
)
106 hlist_add_head(&tw
->tw_bind_node
, list
);
109 static inline int tw_dead_hashed(struct tcp_tw_bucket
*tw
)
111 return tw
->tw_death_node
.pprev
!= NULL
;
114 static __inline__
void tw_dead_node_init(struct tcp_tw_bucket
*tw
)
116 tw
->tw_death_node
.pprev
= NULL
;
119 static __inline__
void __tw_del_dead_node(struct tcp_tw_bucket
*tw
)
121 __hlist_del(&tw
->tw_death_node
);
122 tw_dead_node_init(tw
);
125 static __inline__
int tw_del_dead_node(struct tcp_tw_bucket
*tw
)
127 if (tw_dead_hashed(tw
)) {
128 __tw_del_dead_node(tw
);
134 #define tw_for_each(tw, node, head) \
135 hlist_for_each_entry(tw, node, head, tw_node)
137 #define tw_for_each_inmate(tw, node, jail) \
138 hlist_for_each_entry(tw, node, jail, tw_death_node)
140 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
141 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
143 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
145 static inline u32
tcp_v4_rcv_saddr(const struct sock
*sk
)
147 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
148 inet_sk(sk
)->rcv_saddr
: tcptw_sk(sk
)->tw_rcv_saddr
;
151 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
152 static inline struct in6_addr
*__tcp_v6_rcv_saddr(const struct sock
*sk
)
154 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
155 &inet6_sk(sk
)->rcv_saddr
: &tcptw_sk(sk
)->tw_v6_rcv_saddr
;
158 static inline struct in6_addr
*tcp_v6_rcv_saddr(const struct sock
*sk
)
160 return sk
->sk_family
== AF_INET6
? __tcp_v6_rcv_saddr(sk
) : NULL
;
163 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
165 static inline int tcp_v6_ipv6only(const struct sock
*sk
)
167 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
168 ipv6_only_sock(sk
) : tcptw_sk_ipv6only(sk
);
171 # define __tcp_v6_rcv_saddr(__sk) NULL
172 # define tcp_v6_rcv_saddr(__sk) NULL
173 # define tcptw_sk_ipv6only(__sk) 0
174 # define tcp_v6_ipv6only(__sk) 0
177 extern kmem_cache_t
*tcp_timewait_cachep
;
179 static inline void tcp_tw_put(struct tcp_tw_bucket
*tw
)
181 if (atomic_dec_and_test(&tw
->tw_refcnt
)) {
182 #ifdef SOCK_REFCNT_DEBUG
183 printk(KERN_DEBUG
"tw_bucket %p released\n", tw
);
185 kmem_cache_free(tcp_timewait_cachep
, tw
);
189 extern atomic_t tcp_orphan_count
;
190 extern int tcp_tw_count
;
191 extern void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
192 extern void tcp_tw_deschedule(struct tcp_tw_bucket
*tw
);
195 /* Socket demux engine toys. */
197 #define TCP_COMBINED_PORTS(__sport, __dport) \
198 (((__u32)(__sport)<<16) | (__u32)(__dport))
199 #else /* __LITTLE_ENDIAN */
200 #define TCP_COMBINED_PORTS(__sport, __dport) \
201 (((__u32)(__dport)<<16) | (__u32)(__sport))
204 #if (BITS_PER_LONG == 64)
206 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
207 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
208 #else /* __LITTLE_ENDIAN */
209 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
210 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
211 #endif /* __BIG_ENDIAN */
212 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
213 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
214 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
215 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
216 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
217 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
218 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
219 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
220 #else /* 32-bit arch */
221 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
222 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
223 ((inet_sk(__sk)->daddr == (__saddr)) && \
224 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
225 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
226 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
227 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
228 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
229 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
230 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
231 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
232 #endif /* 64-bit arch */
234 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
235 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
236 ((__sk)->sk_family == AF_INET6) && \
237 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \
238 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
239 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
241 #define MAX_TCP_HEADER (128 + MAX_HEADER)
244 * Never offer a window over 32767 without using window scaling. Some
245 * poor stacks do signed 16bit maths!
247 #define MAX_TCP_WINDOW 32767U
249 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
250 #define TCP_MIN_MSS 88U
252 /* Minimal RCV_MSS. */
253 #define TCP_MIN_RCVMSS 536U
255 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
256 #define TCP_FASTRETRANS_THRESH 3
258 /* Maximal reordering. */
259 #define TCP_MAX_REORDERING 127
261 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
262 #define TCP_MAX_QUICKACKS 16U
264 /* urg_data states */
265 #define TCP_URG_VALID 0x0100
266 #define TCP_URG_NOTYET 0x0200
267 #define TCP_URG_READ 0x0400
269 #define TCP_RETR1 3 /*
270 * This is how many retries it does before it
271 * tries to figure out if the gateway is
272 * down. Minimal RFC value is 3; it corresponds
273 * to ~3sec-8min depending on RTO.
276 #define TCP_RETR2 15 /*
277 * This should take at least
278 * 90 minutes to time out.
279 * RFC1122 says that the limit is 100 sec.
280 * 15 is ~13-30min depending on RTO.
283 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
284 * connection: ~180sec is RFC minumum */
286 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
287 * connection: ~180sec is RFC minumum */
290 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
291 * socket. 7 is ~50sec-16min.
295 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
296 * state, about 60 seconds */
297 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
298 /* BSD style FIN_WAIT2 deadlock breaker.
299 * It used to be 3min, new value is 60sec,
300 * to combine FIN-WAIT-2 timeout with
304 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
306 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
307 #define TCP_ATO_MIN ((unsigned)(HZ/25))
309 #define TCP_DELACK_MIN 4U
310 #define TCP_ATO_MIN 4U
312 #define TCP_RTO_MAX ((unsigned)(120*HZ))
313 #define TCP_RTO_MIN ((unsigned)(HZ/5))
314 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
316 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
317 * for local resources.
320 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
321 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
322 #define TCP_KEEPALIVE_INTVL (75*HZ)
324 #define MAX_TCP_KEEPIDLE 32767
325 #define MAX_TCP_KEEPINTVL 32767
326 #define MAX_TCP_KEEPCNT 127
327 #define MAX_TCP_SYNCNT 127
329 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
330 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
332 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
333 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
334 * after this time. It should be equal
335 * (or greater than) TCP_TIMEWAIT_LEN
336 * to provide reliability equal to one
337 * provided by timewait state.
339 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
340 * timestamps. It must be less than
341 * minimal timewait lifetime.
344 #define TCP_TW_RECYCLE_SLOTS_LOG 5
345 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
347 /* If time > 4sec, it is "slow" path, no recycling is required,
348 so that we select tick to get range about 4 seconds.
351 #if HZ <= 16 || HZ > 4096
352 # error Unsupported: HZ <= 16 or HZ > 4096
354 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
356 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
358 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
360 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
362 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
364 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
366 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
368 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
374 #define TCPOPT_NOP 1 /* Padding */
375 #define TCPOPT_EOL 0 /* End of options */
376 #define TCPOPT_MSS 2 /* Segment size negotiating */
377 #define TCPOPT_WINDOW 3 /* Window scaling */
378 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
379 #define TCPOPT_SACK 5 /* SACK Block */
380 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
386 #define TCPOLEN_MSS 4
387 #define TCPOLEN_WINDOW 3
388 #define TCPOLEN_SACK_PERM 2
389 #define TCPOLEN_TIMESTAMP 10
391 /* But this is what stacks really send out. */
392 #define TCPOLEN_TSTAMP_ALIGNED 12
393 #define TCPOLEN_WSCALE_ALIGNED 4
394 #define TCPOLEN_SACKPERM_ALIGNED 4
395 #define TCPOLEN_SACK_BASE 2
396 #define TCPOLEN_SACK_BASE_ALIGNED 4
397 #define TCPOLEN_SACK_PERBLOCK 8
399 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
400 #define TCP_TIME_DACK 2 /* Delayed ack timer */
401 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
402 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
404 /* Flags in tp->nonagle */
405 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
406 #define TCP_NAGLE_CORK 2 /* Socket is corked */
407 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
409 /* sysctl variables for tcp */
410 extern int sysctl_tcp_timestamps
;
411 extern int sysctl_tcp_window_scaling
;
412 extern int sysctl_tcp_sack
;
413 extern int sysctl_tcp_fin_timeout
;
414 extern int sysctl_tcp_tw_recycle
;
415 extern int sysctl_tcp_keepalive_time
;
416 extern int sysctl_tcp_keepalive_probes
;
417 extern int sysctl_tcp_keepalive_intvl
;
418 extern int sysctl_tcp_syn_retries
;
419 extern int sysctl_tcp_synack_retries
;
420 extern int sysctl_tcp_retries1
;
421 extern int sysctl_tcp_retries2
;
422 extern int sysctl_tcp_orphan_retries
;
423 extern int sysctl_tcp_syncookies
;
424 extern int sysctl_tcp_retrans_collapse
;
425 extern int sysctl_tcp_stdurg
;
426 extern int sysctl_tcp_rfc1337
;
427 extern int sysctl_tcp_abort_on_overflow
;
428 extern int sysctl_tcp_max_orphans
;
429 extern int sysctl_tcp_max_tw_buckets
;
430 extern int sysctl_tcp_fack
;
431 extern int sysctl_tcp_reordering
;
432 extern int sysctl_tcp_ecn
;
433 extern int sysctl_tcp_dsack
;
434 extern int sysctl_tcp_mem
[3];
435 extern int sysctl_tcp_wmem
[3];
436 extern int sysctl_tcp_rmem
[3];
437 extern int sysctl_tcp_app_win
;
438 extern int sysctl_tcp_adv_win_scale
;
439 extern int sysctl_tcp_tw_reuse
;
440 extern int sysctl_tcp_frto
;
441 extern int sysctl_tcp_low_latency
;
442 extern int sysctl_tcp_nometrics_save
;
443 extern int sysctl_tcp_moderate_rcvbuf
;
444 extern int sysctl_tcp_tso_win_divisor
;
446 extern atomic_t tcp_memory_allocated
;
447 extern atomic_t tcp_sockets_allocated
;
448 extern int tcp_memory_pressure
;
450 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
451 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
453 #define TCP_INET_FAMILY(fam) 1
457 * Pointers to address related TCP functions
458 * (i.e. things that depend on the address family)
462 int (*queue_xmit
) (struct sk_buff
*skb
,
465 void (*send_check
) (struct sock
*sk
,
468 struct sk_buff
*skb
);
470 int (*rebuild_header
) (struct sock
*sk
);
472 int (*conn_request
) (struct sock
*sk
,
473 struct sk_buff
*skb
);
475 struct sock
* (*syn_recv_sock
) (struct sock
*sk
,
477 struct request_sock
*req
,
478 struct dst_entry
*dst
);
480 int (*remember_stamp
) (struct sock
*sk
);
482 __u16 net_header_len
;
484 int (*setsockopt
) (struct sock
*sk
,
490 int (*getsockopt
) (struct sock
*sk
,
497 void (*addr2sockaddr
) (struct sock
*sk
,
504 * The next routines deal with comparing 32 bit unsigned ints
505 * and worry about wraparound (automatic with unsigned arithmetic).
508 static inline int before(__u32 seq1
, __u32 seq2
)
510 return (__s32
)(seq1
-seq2
) < 0;
513 static inline int after(__u32 seq1
, __u32 seq2
)
515 return (__s32
)(seq2
-seq1
) < 0;
519 /* is s2<=s1<=s3 ? */
520 static inline int between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
522 return seq3
- seq2
>= seq1
- seq2
;
526 extern struct proto tcp_prot
;
528 DECLARE_SNMP_STAT(struct tcp_mib
, tcp_statistics
);
529 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
530 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
531 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
532 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
533 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
534 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
536 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
538 extern void tcp_shutdown (struct sock
*sk
, int how
);
540 extern int tcp_v4_rcv(struct sk_buff
*skb
);
542 extern int tcp_v4_remember_stamp(struct sock
*sk
);
544 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket
*tw
);
546 extern int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
547 struct msghdr
*msg
, size_t size
);
548 extern ssize_t
tcp_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
);
550 extern int tcp_ioctl(struct sock
*sk
,
554 extern int tcp_rcv_state_process(struct sock
*sk
,
559 extern int tcp_rcv_established(struct sock
*sk
,
564 extern void tcp_rcv_space_adjust(struct sock
*sk
);
573 static inline void tcp_schedule_ack(struct tcp_sock
*tp
)
575 tp
->ack
.pending
|= TCP_ACK_SCHED
;
578 static inline int tcp_ack_scheduled(struct tcp_sock
*tp
)
580 return tp
->ack
.pending
&TCP_ACK_SCHED
;
583 static __inline__
void tcp_dec_quickack_mode(struct tcp_sock
*tp
, unsigned int pkts
)
586 if (pkts
>= tp
->ack
.quick
) {
589 /* Leaving quickack mode we deflate ATO. */
590 tp
->ack
.ato
= TCP_ATO_MIN
;
592 tp
->ack
.quick
-= pkts
;
596 extern void tcp_enter_quickack_mode(struct tcp_sock
*tp
);
598 static __inline__
void tcp_delack_init(struct tcp_sock
*tp
)
600 memset(&tp
->ack
, 0, sizeof(tp
->ack
));
603 static inline void tcp_clear_options(struct tcp_options_received
*rx_opt
)
605 rx_opt
->tstamp_ok
= rx_opt
->sack_ok
= rx_opt
->wscale_ok
= rx_opt
->snd_wscale
= 0;
617 extern enum tcp_tw_status
tcp_timewait_state_process(struct tcp_tw_bucket
*tw
,
622 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
623 struct request_sock
*req
,
624 struct request_sock
**prev
);
625 extern int tcp_child_process(struct sock
*parent
,
627 struct sk_buff
*skb
);
628 extern void tcp_enter_frto(struct sock
*sk
);
629 extern void tcp_enter_loss(struct sock
*sk
, int how
);
630 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
631 extern void tcp_update_metrics(struct sock
*sk
);
633 extern void tcp_close(struct sock
*sk
,
635 extern struct sock
* tcp_accept(struct sock
*sk
, int flags
, int *err
);
636 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
, struct poll_table_struct
*wait
);
638 extern int tcp_getsockopt(struct sock
*sk
, int level
,
642 extern int tcp_setsockopt(struct sock
*sk
, int level
,
643 int optname
, char __user
*optval
,
645 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
646 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
648 size_t len
, int nonblock
,
649 int flags
, int *addr_len
);
651 extern int tcp_listen_start(struct sock
*sk
);
653 extern void tcp_parse_options(struct sk_buff
*skb
,
654 struct tcp_options_received
*opt_rx
,
658 * TCP v4 functions exported for the inet6 API
661 extern void tcp_v4_send_check(struct sock
*sk
,
662 struct tcphdr
*th
, int len
,
663 struct sk_buff
*skb
);
665 extern int tcp_v4_conn_request(struct sock
*sk
,
666 struct sk_buff
*skb
);
668 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
669 struct request_sock
*req
,
670 struct sk_buff
*skb
);
672 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
,
674 struct request_sock
*req
,
675 struct dst_entry
*dst
);
677 extern int tcp_v4_do_rcv(struct sock
*sk
,
678 struct sk_buff
*skb
);
680 extern int tcp_v4_connect(struct sock
*sk
,
681 struct sockaddr
*uaddr
,
684 extern int tcp_connect(struct sock
*sk
);
686 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
,
687 struct dst_entry
*dst
,
688 struct request_sock
*req
);
690 extern int tcp_disconnect(struct sock
*sk
, int flags
);
692 extern void tcp_unhash(struct sock
*sk
);
694 extern int tcp_v4_hash_connecting(struct sock
*sk
);
697 /* From syncookies.c */
698 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
699 struct ip_options
*opt
);
700 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
705 extern void __tcp_push_pending_frames(struct sock
*sk
, struct tcp_sock
*tp
,
706 unsigned int cur_mss
, int nonagle
);
707 extern int tcp_may_send_now(struct sock
*sk
, struct tcp_sock
*tp
);
708 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
709 extern void tcp_xmit_retransmit_queue(struct sock
*);
710 extern void tcp_simple_retransmit(struct sock
*);
711 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
713 extern void tcp_send_probe0(struct sock
*);
714 extern void tcp_send_partial(struct sock
*);
715 extern int tcp_write_wakeup(struct sock
*);
716 extern void tcp_send_fin(struct sock
*sk
);
717 extern void tcp_send_active_reset(struct sock
*sk
,
718 unsigned int __nocast priority
);
719 extern int tcp_send_synack(struct sock
*);
720 extern void tcp_push_one(struct sock
*, unsigned int mss_now
);
721 extern void tcp_send_ack(struct sock
*sk
);
722 extern void tcp_send_delayed_ack(struct sock
*sk
);
725 extern void tcp_cwnd_application_limited(struct sock
*sk
);
728 extern void tcp_init_xmit_timers(struct sock
*);
729 extern void tcp_clear_xmit_timers(struct sock
*);
731 extern void tcp_delete_keepalive_timer(struct sock
*);
732 extern void tcp_reset_keepalive_timer(struct sock
*, unsigned long);
733 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
734 extern unsigned int tcp_current_mss(struct sock
*sk
, int large
);
737 extern const char tcp_timer_bug_msg
[];
741 extern void tcp_get_info(struct sock
*, struct tcp_info
*);
743 /* Read 'sendfile()'-style from a TCP socket */
744 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
745 unsigned int, size_t);
746 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
747 sk_read_actor_t recv_actor
);
749 static inline void tcp_clear_xmit_timer(struct sock
*sk
, int what
)
751 struct tcp_sock
*tp
= tcp_sk(sk
);
754 case TCP_TIME_RETRANS
:
755 case TCP_TIME_PROBE0
:
758 #ifdef TCP_CLEAR_TIMERS
759 sk_stop_timer(sk
, &tp
->retransmit_timer
);
766 #ifdef TCP_CLEAR_TIMERS
767 sk_stop_timer(sk
, &tp
->delack_timer
);
772 printk(tcp_timer_bug_msg
);
780 * Reset the retransmission timer
782 static inline void tcp_reset_xmit_timer(struct sock
*sk
, int what
, unsigned long when
)
784 struct tcp_sock
*tp
= tcp_sk(sk
);
786 if (when
> TCP_RTO_MAX
) {
788 printk(KERN_DEBUG
"reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk
, what
, when
, current_text_addr());
794 case TCP_TIME_RETRANS
:
795 case TCP_TIME_PROBE0
:
797 tp
->timeout
= jiffies
+when
;
798 sk_reset_timer(sk
, &tp
->retransmit_timer
, tp
->timeout
);
802 tp
->ack
.pending
|= TCP_ACK_TIMER
;
803 tp
->ack
.timeout
= jiffies
+when
;
804 sk_reset_timer(sk
, &tp
->delack_timer
, tp
->ack
.timeout
);
809 printk(tcp_timer_bug_msg
);
815 /* Initialize RCV_MSS value.
816 * RCV_MSS is an our guess about MSS used by the peer.
817 * We haven't any direct information about the MSS.
818 * It's better to underestimate the RCV_MSS rather than overestimate.
819 * Overestimations make us ACKing less frequently than needed.
820 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
823 static inline void tcp_initialize_rcv_mss(struct sock
*sk
)
825 struct tcp_sock
*tp
= tcp_sk(sk
);
826 unsigned int hint
= min_t(unsigned int, tp
->advmss
, tp
->mss_cache
);
828 hint
= min(hint
, tp
->rcv_wnd
/2);
829 hint
= min(hint
, TCP_MIN_RCVMSS
);
830 hint
= max(hint
, TCP_MIN_MSS
);
832 tp
->ack
.rcv_mss
= hint
;
835 static __inline__
void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
837 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
838 ntohl(TCP_FLAG_ACK
) |
842 static __inline__
void tcp_fast_path_on(struct tcp_sock
*tp
)
844 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
847 static inline void tcp_fast_path_check(struct sock
*sk
, struct tcp_sock
*tp
)
849 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
851 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
853 tcp_fast_path_on(tp
);
856 /* Compute the actual receive window we are currently advertising.
857 * Rcv_nxt can be after the window if our peer push more data
858 * than the offered window.
860 static __inline__ u32
tcp_receive_window(const struct tcp_sock
*tp
)
862 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
869 /* Choose a new window, without checks for shrinking, and without
870 * scaling applied to the result. The caller does these things
871 * if necessary. This is a "raw" window selection.
873 extern u32
__tcp_select_window(struct sock
*sk
);
875 /* TCP timestamps are only 32-bits, this causes a slight
876 * complication on 64-bit systems since we store a snapshot
877 * of jiffies in the buffer control blocks below. We decidely
878 * only use of the low 32-bits of jiffies and hide the ugly
879 * casts with the following macro.
881 #define tcp_time_stamp ((__u32)(jiffies))
883 /* This is what the send packet queueing engine uses to pass
884 * TCP per-packet control information to the transmission
885 * code. We also store the host-order sequence numbers in
886 * here too. This is 36 bytes on 32-bit architectures,
887 * 40 bytes on 64-bit machines, if this grows please adjust
888 * skbuff.h:skbuff->cb[xxx] size appropriately.
892 struct inet_skb_parm h4
;
893 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
894 struct inet6_skb_parm h6
;
896 } header
; /* For incoming frames */
897 __u32 seq
; /* Starting sequence number */
898 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
899 __u32 when
; /* used to compute rtt's */
900 __u8 flags
; /* TCP header flags. */
902 /* NOTE: These must match up to the flags byte in a
905 #define TCPCB_FLAG_FIN 0x01
906 #define TCPCB_FLAG_SYN 0x02
907 #define TCPCB_FLAG_RST 0x04
908 #define TCPCB_FLAG_PSH 0x08
909 #define TCPCB_FLAG_ACK 0x10
910 #define TCPCB_FLAG_URG 0x20
911 #define TCPCB_FLAG_ECE 0x40
912 #define TCPCB_FLAG_CWR 0x80
914 __u8 sacked
; /* State flags for SACK/FACK. */
915 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
916 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
917 #define TCPCB_LOST 0x04 /* SKB is lost */
918 #define TCPCB_TAGBITS 0x07 /* All tag bits */
920 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
921 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
923 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
925 #define TCPCB_AT_TAIL (TCPCB_URG)
927 __u16 urg_ptr
; /* Valid w/URG flags is set. */
928 __u32 ack_seq
; /* Sequence number ACK'd */
931 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
933 #include <net/tcp_ecn.h>
935 /* Due to TSO, an SKB can be composed of multiple actual
936 * packets. To keep these tracked properly, we use this.
938 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
940 return skb_shinfo(skb
)->tso_segs
;
943 /* This is valid iff tcp_skb_pcount() > 1. */
944 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
946 return skb_shinfo(skb
)->tso_size
;
949 static inline void tcp_dec_pcount_approx(__u32
*count
,
950 const struct sk_buff
*skb
)
953 *count
-= tcp_skb_pcount(skb
);
959 static inline void tcp_packets_out_inc(struct sock
*sk
,
961 const struct sk_buff
*skb
)
963 int orig
= tp
->packets_out
;
965 tp
->packets_out
+= tcp_skb_pcount(skb
);
967 tcp_reset_xmit_timer(sk
, TCP_TIME_RETRANS
, tp
->rto
);
970 static inline void tcp_packets_out_dec(struct tcp_sock
*tp
,
971 const struct sk_buff
*skb
)
973 tp
->packets_out
-= tcp_skb_pcount(skb
);
976 /* Events passed to congestion control interface */
978 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
979 CA_EVENT_CWND_RESTART
, /* congestion window restart */
980 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
981 CA_EVENT_FRTO
, /* fast recovery timeout */
982 CA_EVENT_LOSS
, /* loss timeout */
983 CA_EVENT_FAST_ACK
, /* in sequence ack */
984 CA_EVENT_SLOW_ACK
, /* other ack */
988 * Interface for adding new TCP congestion control handlers
990 #define TCP_CA_NAME_MAX 16
991 struct tcp_congestion_ops
{
992 struct list_head list
;
994 /* initialize private data (optional) */
995 void (*init
)(struct tcp_sock
*tp
);
996 /* cleanup private data (optional) */
997 void (*release
)(struct tcp_sock
*tp
);
999 /* return slow start threshold (required) */
1000 u32 (*ssthresh
)(struct tcp_sock
*tp
);
1001 /* lower bound for congestion window (optional) */
1002 u32 (*min_cwnd
)(struct tcp_sock
*tp
);
1003 /* do new cwnd calculation (required) */
1004 void (*cong_avoid
)(struct tcp_sock
*tp
, u32 ack
,
1005 u32 rtt
, u32 in_flight
, int good_ack
);
1006 /* round trip time sample per acked packet (optional) */
1007 void (*rtt_sample
)(struct tcp_sock
*tp
, u32 usrtt
);
1008 /* call before changing ca_state (optional) */
1009 void (*set_state
)(struct tcp_sock
*tp
, u8 new_state
);
1010 /* call when cwnd event occurs (optional) */
1011 void (*cwnd_event
)(struct tcp_sock
*tp
, enum tcp_ca_event ev
);
1012 /* new value of cwnd after loss (optional) */
1013 u32 (*undo_cwnd
)(struct tcp_sock
*tp
);
1014 /* hook for packet ack accounting (optional) */
1015 void (*pkts_acked
)(struct tcp_sock
*tp
, u32 num_acked
);
1016 /* get info for tcp_diag (optional) */
1017 void (*get_info
)(struct tcp_sock
*tp
, u32 ext
, struct sk_buff
*skb
);
1019 char name
[TCP_CA_NAME_MAX
];
1020 struct module
*owner
;
1023 extern int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
1024 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
1026 extern void tcp_init_congestion_control(struct tcp_sock
*tp
);
1027 extern void tcp_cleanup_congestion_control(struct tcp_sock
*tp
);
1028 extern int tcp_set_default_congestion_control(const char *name
);
1029 extern void tcp_get_default_congestion_control(char *name
);
1030 extern int tcp_set_congestion_control(struct tcp_sock
*tp
, const char *name
);
1032 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
1033 extern u32
tcp_reno_ssthresh(struct tcp_sock
*tp
);
1034 extern void tcp_reno_cong_avoid(struct tcp_sock
*tp
, u32 ack
,
1035 u32 rtt
, u32 in_flight
, int flag
);
1036 extern u32
tcp_reno_min_cwnd(struct tcp_sock
*tp
);
1037 extern struct tcp_congestion_ops tcp_reno
;
1039 static inline void tcp_set_ca_state(struct tcp_sock
*tp
, u8 ca_state
)
1041 if (tp
->ca_ops
->set_state
)
1042 tp
->ca_ops
->set_state(tp
, ca_state
);
1043 tp
->ca_state
= ca_state
;
1046 static inline void tcp_ca_event(struct tcp_sock
*tp
, enum tcp_ca_event event
)
1048 if (tp
->ca_ops
->cwnd_event
)
1049 tp
->ca_ops
->cwnd_event(tp
, event
);
1052 /* This determines how many packets are "in the network" to the best
1053 * of our knowledge. In many cases it is conservative, but where
1054 * detailed information is available from the receiver (via SACK
1055 * blocks etc.) we can make more aggressive calculations.
1057 * Use this for decisions involving congestion control, use just
1058 * tp->packets_out to determine if the send queue is empty or not.
1060 * Read this equation as:
1062 * "Packets sent once on transmission queue" MINUS
1063 * "Packets left network, but not honestly ACKed yet" PLUS
1064 * "Packets fast retransmitted"
1066 static __inline__
unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1068 return (tp
->packets_out
- tp
->left_out
+ tp
->retrans_out
);
1071 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1072 * The exception is rate halving phase, when cwnd is decreasing towards
1075 static inline __u32
tcp_current_ssthresh(struct tcp_sock
*tp
)
1077 if ((1<<tp
->ca_state
)&(TCPF_CA_CWR
|TCPF_CA_Recovery
))
1078 return tp
->snd_ssthresh
;
1080 return max(tp
->snd_ssthresh
,
1081 ((tp
->snd_cwnd
>> 1) +
1082 (tp
->snd_cwnd
>> 2)));
1085 static inline void tcp_sync_left_out(struct tcp_sock
*tp
)
1087 if (tp
->rx_opt
.sack_ok
&&
1088 (tp
->sacked_out
>= tp
->packets_out
- tp
->lost_out
))
1089 tp
->sacked_out
= tp
->packets_out
- tp
->lost_out
;
1090 tp
->left_out
= tp
->sacked_out
+ tp
->lost_out
;
1093 /* Set slow start threshold and cwnd not falling to slow start */
1094 static inline void __tcp_enter_cwr(struct tcp_sock
*tp
)
1096 tp
->undo_marker
= 0;
1097 tp
->snd_ssthresh
= tp
->ca_ops
->ssthresh(tp
);
1098 tp
->snd_cwnd
= min(tp
->snd_cwnd
,
1099 tcp_packets_in_flight(tp
) + 1U);
1100 tp
->snd_cwnd_cnt
= 0;
1101 tp
->high_seq
= tp
->snd_nxt
;
1102 tp
->snd_cwnd_stamp
= tcp_time_stamp
;
1103 TCP_ECN_queue_cwr(tp
);
1106 static inline void tcp_enter_cwr(struct tcp_sock
*tp
)
1108 tp
->prior_ssthresh
= 0;
1109 if (tp
->ca_state
< TCP_CA_CWR
) {
1110 __tcp_enter_cwr(tp
);
1111 tcp_set_ca_state(tp
, TCP_CA_CWR
);
1115 extern __u32
tcp_init_cwnd(struct tcp_sock
*tp
, struct dst_entry
*dst
);
1117 /* Slow start with delack produces 3 packets of burst, so that
1118 * it is safe "de facto".
1120 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
1125 static __inline__
void tcp_minshall_update(struct tcp_sock
*tp
, int mss
,
1126 const struct sk_buff
*skb
)
1129 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
1132 static __inline__
void tcp_check_probe_timer(struct sock
*sk
, struct tcp_sock
*tp
)
1134 if (!tp
->packets_out
&& !tp
->pending
)
1135 tcp_reset_xmit_timer(sk
, TCP_TIME_PROBE0
, tp
->rto
);
1138 static __inline__
void tcp_push_pending_frames(struct sock
*sk
,
1139 struct tcp_sock
*tp
)
1141 __tcp_push_pending_frames(sk
, tp
, tcp_current_mss(sk
, 1), tp
->nonagle
);
1144 static __inline__
void tcp_init_wl(struct tcp_sock
*tp
, u32 ack
, u32 seq
)
1149 static __inline__
void tcp_update_wl(struct tcp_sock
*tp
, u32 ack
, u32 seq
)
1154 extern void tcp_destroy_sock(struct sock
*sk
);
1158 * Calculate(/check) TCP checksum
1160 static __inline__ u16
tcp_v4_check(struct tcphdr
*th
, int len
,
1161 unsigned long saddr
, unsigned long daddr
,
1164 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1167 static __inline__
int __tcp_checksum_complete(struct sk_buff
*skb
)
1169 return (unsigned short)csum_fold(skb_checksum(skb
, 0, skb
->len
, skb
->csum
));
1172 static __inline__
int tcp_checksum_complete(struct sk_buff
*skb
)
1174 return skb
->ip_summed
!= CHECKSUM_UNNECESSARY
&&
1175 __tcp_checksum_complete(skb
);
1178 /* Prequeue for VJ style copy to user, combined with checksumming. */
1180 static __inline__
void tcp_prequeue_init(struct tcp_sock
*tp
)
1182 tp
->ucopy
.task
= NULL
;
1184 tp
->ucopy
.memory
= 0;
1185 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1188 /* Packet is added to VJ-style prequeue for processing in process
1189 * context, if a reader task is waiting. Apparently, this exciting
1190 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1191 * failed somewhere. Latency? Burstiness? Well, at least now we will
1192 * see, why it failed. 8)8) --ANK
1194 * NOTE: is this not too big to inline?
1196 static __inline__
int tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1198 struct tcp_sock
*tp
= tcp_sk(sk
);
1200 if (!sysctl_tcp_low_latency
&& tp
->ucopy
.task
) {
1201 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1202 tp
->ucopy
.memory
+= skb
->truesize
;
1203 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1204 struct sk_buff
*skb1
;
1206 BUG_ON(sock_owned_by_user(sk
));
1208 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1209 sk
->sk_backlog_rcv(sk
, skb1
);
1210 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED
);
1213 tp
->ucopy
.memory
= 0;
1214 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1215 wake_up_interruptible(sk
->sk_sleep
);
1216 if (!tcp_ack_scheduled(tp
))
1217 tcp_reset_xmit_timer(sk
, TCP_TIME_DACK
, (3*TCP_RTO_MIN
)/4);
1228 static const char *statename
[]={
1229 "Unused","Established","Syn Sent","Syn Recv",
1230 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1231 "Close Wait","Last ACK","Listen","Closing"
1235 static __inline__
void tcp_set_state(struct sock
*sk
, int state
)
1237 int oldstate
= sk
->sk_state
;
1240 case TCP_ESTABLISHED
:
1241 if (oldstate
!= TCP_ESTABLISHED
)
1242 TCP_INC_STATS(TCP_MIB_CURRESTAB
);
1246 if (oldstate
== TCP_CLOSE_WAIT
|| oldstate
== TCP_ESTABLISHED
)
1247 TCP_INC_STATS(TCP_MIB_ESTABRESETS
);
1249 sk
->sk_prot
->unhash(sk
);
1250 if (inet_sk(sk
)->bind_hash
&&
1251 !(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
))
1252 inet_put_port(&tcp_hashinfo
, sk
);
1255 if (oldstate
==TCP_ESTABLISHED
)
1256 TCP_DEC_STATS(TCP_MIB_CURRESTAB
);
1259 /* Change state AFTER socket is unhashed to avoid closed
1260 * socket sitting in hash tables.
1262 sk
->sk_state
= state
;
1265 SOCK_DEBUG(sk
, "TCP sk=%p, State %s -> %s\n",sk
, statename
[oldstate
],statename
[state
]);
1269 static __inline__
void tcp_done(struct sock
*sk
)
1271 tcp_set_state(sk
, TCP_CLOSE
);
1272 tcp_clear_xmit_timers(sk
);
1274 sk
->sk_shutdown
= SHUTDOWN_MASK
;
1276 if (!sock_flag(sk
, SOCK_DEAD
))
1277 sk
->sk_state_change(sk
);
1279 tcp_destroy_sock(sk
);
1282 static __inline__
void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1285 rx_opt
->eff_sacks
= 0;
1286 rx_opt
->num_sacks
= 0;
1289 static __inline__
void tcp_build_and_update_options(__u32
*ptr
, struct tcp_sock
*tp
, __u32 tstamp
)
1291 if (tp
->rx_opt
.tstamp_ok
) {
1292 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) |
1293 (TCPOPT_NOP
<< 16) |
1294 (TCPOPT_TIMESTAMP
<< 8) |
1296 *ptr
++ = htonl(tstamp
);
1297 *ptr
++ = htonl(tp
->rx_opt
.ts_recent
);
1299 if (tp
->rx_opt
.eff_sacks
) {
1300 struct tcp_sack_block
*sp
= tp
->rx_opt
.dsack
? tp
->duplicate_sack
: tp
->selective_acks
;
1303 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) |
1304 (TCPOPT_NOP
<< 16) |
1305 (TCPOPT_SACK
<< 8) |
1306 (TCPOLEN_SACK_BASE
+
1307 (tp
->rx_opt
.eff_sacks
* TCPOLEN_SACK_PERBLOCK
)));
1308 for(this_sack
= 0; this_sack
< tp
->rx_opt
.eff_sacks
; this_sack
++) {
1309 *ptr
++ = htonl(sp
[this_sack
].start_seq
);
1310 *ptr
++ = htonl(sp
[this_sack
].end_seq
);
1312 if (tp
->rx_opt
.dsack
) {
1313 tp
->rx_opt
.dsack
= 0;
1314 tp
->rx_opt
.eff_sacks
--;
1319 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1320 * If this is every changed make sure to change the definition of
1321 * MAX_SYN_SIZE to match the new maximum number of options that you
1324 static inline void tcp_syn_build_options(__u32
*ptr
, int mss
, int ts
, int sack
,
1325 int offer_wscale
, int wscale
, __u32 tstamp
, __u32 ts_recent
)
1327 /* We always get an MSS option.
1328 * The option bytes which will be seen in normal data
1329 * packets should timestamps be used, must be in the MSS
1330 * advertised. But we subtract them from tp->mss_cache so
1331 * that calculations in tcp_sendmsg are simpler etc.
1332 * So account for this fact here if necessary. If we
1333 * don't do this correctly, as a receiver we won't
1334 * recognize data packets as being full sized when we
1335 * should, and thus we won't abide by the delayed ACK
1337 * SACKs don't matter, we never delay an ACK when we
1338 * have any of those going out.
1340 *ptr
++ = htonl((TCPOPT_MSS
<< 24) | (TCPOLEN_MSS
<< 16) | mss
);
1343 *ptr
++ = __constant_htonl((TCPOPT_SACK_PERM
<< 24) | (TCPOLEN_SACK_PERM
<< 16) |
1344 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
1346 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
1347 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
1348 *ptr
++ = htonl(tstamp
); /* TSVAL */
1349 *ptr
++ = htonl(ts_recent
); /* TSECR */
1351 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
1352 (TCPOPT_SACK_PERM
<< 8) | TCPOLEN_SACK_PERM
);
1354 *ptr
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_WINDOW
<< 16) | (TCPOLEN_WINDOW
<< 8) | (wscale
));
1357 /* Determine a window scaling and initial window to offer. */
1358 extern void tcp_select_initial_window(int __space
, __u32 mss
,
1359 __u32
*rcv_wnd
, __u32
*window_clamp
,
1360 int wscale_ok
, __u8
*rcv_wscale
);
1362 static inline int tcp_win_from_space(int space
)
1364 return sysctl_tcp_adv_win_scale
<=0 ?
1365 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1366 space
- (space
>>sysctl_tcp_adv_win_scale
);
1369 /* Note: caller must be prepared to deal with negative returns */
1370 static inline int tcp_space(const struct sock
*sk
)
1372 return tcp_win_from_space(sk
->sk_rcvbuf
-
1373 atomic_read(&sk
->sk_rmem_alloc
));
1376 static inline int tcp_full_space(const struct sock
*sk
)
1378 return tcp_win_from_space(sk
->sk_rcvbuf
);
1381 static inline void tcp_acceptq_queue(struct sock
*sk
, struct request_sock
*req
,
1384 reqsk_queue_add(&tcp_sk(sk
)->accept_queue
, req
, sk
, child
);
1388 tcp_synq_removed(struct sock
*sk
, struct request_sock
*req
)
1390 if (reqsk_queue_removed(&tcp_sk(sk
)->accept_queue
, req
) == 0)
1391 tcp_delete_keepalive_timer(sk
);
1394 static inline void tcp_synq_added(struct sock
*sk
)
1396 if (reqsk_queue_added(&tcp_sk(sk
)->accept_queue
) == 0)
1397 tcp_reset_keepalive_timer(sk
, TCP_TIMEOUT_INIT
);
1400 static inline int tcp_synq_len(struct sock
*sk
)
1402 return reqsk_queue_len(&tcp_sk(sk
)->accept_queue
);
1405 static inline int tcp_synq_young(struct sock
*sk
)
1407 return reqsk_queue_len_young(&tcp_sk(sk
)->accept_queue
);
1410 static inline int tcp_synq_is_full(struct sock
*sk
)
1412 return reqsk_queue_is_full(&tcp_sk(sk
)->accept_queue
);
1415 static inline void tcp_synq_unlink(struct tcp_sock
*tp
, struct request_sock
*req
,
1416 struct request_sock
**prev
)
1418 reqsk_queue_unlink(&tp
->accept_queue
, req
, prev
);
1421 static inline void tcp_synq_drop(struct sock
*sk
, struct request_sock
*req
,
1422 struct request_sock
**prev
)
1424 tcp_synq_unlink(tcp_sk(sk
), req
, prev
);
1425 tcp_synq_removed(sk
, req
);
1429 static __inline__
void tcp_openreq_init(struct request_sock
*req
,
1430 struct tcp_options_received
*rx_opt
,
1431 struct sk_buff
*skb
)
1433 struct inet_request_sock
*ireq
= inet_rsk(req
);
1435 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1436 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1437 req
->mss
= rx_opt
->mss_clamp
;
1438 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1439 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1440 ireq
->sack_ok
= rx_opt
->sack_ok
;
1441 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1442 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1445 ireq
->rmt_port
= skb
->h
.th
->source
;
1448 extern void tcp_enter_memory_pressure(void);
1450 extern void tcp_listen_wlock(void);
1452 /* - We may sleep inside this lock.
1453 * - If sleeping is not required (or called from BH),
1454 * use plain read_(un)lock(&inet_hashinfo.lhash_lock).
1457 static inline void tcp_listen_lock(void)
1459 /* read_lock synchronizes to candidates to writers */
1460 read_lock(&tcp_hashinfo
.lhash_lock
);
1461 atomic_inc(&tcp_hashinfo
.lhash_users
);
1462 read_unlock(&tcp_hashinfo
.lhash_lock
);
1465 static inline void tcp_listen_unlock(void)
1467 if (atomic_dec_and_test(&tcp_hashinfo
.lhash_users
))
1468 wake_up(&tcp_hashinfo
.lhash_wait
);
1471 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1473 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1476 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1478 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1481 static inline int tcp_fin_time(const struct tcp_sock
*tp
)
1483 int fin_timeout
= tp
->linger2
? : sysctl_tcp_fin_timeout
;
1485 if (fin_timeout
< (tp
->rto
<<2) - (tp
->rto
>>1))
1486 fin_timeout
= (tp
->rto
<<2) - (tp
->rto
>>1);
1491 static inline int tcp_paws_check(const struct tcp_options_received
*rx_opt
, int rst
)
1493 if ((s32
)(rx_opt
->rcv_tsval
- rx_opt
->ts_recent
) >= 0)
1495 if (xtime
.tv_sec
>= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
)
1498 /* RST segments are not recommended to carry timestamp,
1499 and, if they do, it is recommended to ignore PAWS because
1500 "their cleanup function should take precedence over timestamps."
1501 Certainly, it is mistake. It is necessary to understand the reasons
1502 of this constraint to relax it: if peer reboots, clock may go
1503 out-of-sync and half-open connections will not be reset.
1504 Actually, the problem would be not existing if all
1505 the implementations followed draft about maintaining clock
1506 via reboots. Linux-2.2 DOES NOT!
1508 However, we can relax time bounds for RST segments to MSL.
1510 if (rst
&& xtime
.tv_sec
>= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1515 #define TCP_CHECK_TIMER(sk) do { } while (0)
1517 static inline int tcp_use_frto(const struct sock
*sk
)
1519 const struct tcp_sock
*tp
= tcp_sk(sk
);
1521 /* F-RTO must be activated in sysctl and there must be some
1522 * unsent new data, and the advertised window should allow
1525 return (sysctl_tcp_frto
&& sk
->sk_send_head
&&
1526 !after(TCP_SKB_CB(sk
->sk_send_head
)->end_seq
,
1527 tp
->snd_una
+ tp
->snd_wnd
));
1530 static inline void tcp_mib_init(void)
1533 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM
, 1);
1534 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1535 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1536 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN
, -1);
1540 enum tcp_seq_states
{
1541 TCP_SEQ_STATE_LISTENING
,
1542 TCP_SEQ_STATE_OPENREQ
,
1543 TCP_SEQ_STATE_ESTABLISHED
,
1544 TCP_SEQ_STATE_TIME_WAIT
,
1547 struct tcp_seq_afinfo
{
1548 struct module
*owner
;
1551 int (*seq_show
) (struct seq_file
*m
, void *v
);
1552 struct file_operations
*seq_fops
;
1555 struct tcp_iter_state
{
1557 enum tcp_seq_states state
;
1558 struct sock
*syn_wait_sk
;
1559 int bucket
, sbucket
, num
, uid
;
1560 struct seq_operations seq_ops
;
1563 extern int tcp_proc_register(struct tcp_seq_afinfo
*afinfo
);
1564 extern void tcp_proc_unregister(struct tcp_seq_afinfo
*afinfo
);