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/checksum.h>
34 #include <net/request_sock.h>
38 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
39 #include <linux/ipv6.h>
41 #include <linux/seq_file.h>
43 /* This is for all connections with a full identity, no wildcards.
44 * New scheme, half the table is for TIME_WAIT, the other half is
45 * for the rest. I'll experiment with dynamic table growth later.
47 struct tcp_ehash_bucket
{
49 struct hlist_head chain
;
50 } __attribute__((__aligned__(8)));
52 /* This is for listening sockets, thus all sockets which possess wildcards. */
53 #define TCP_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
55 /* There are a few simple rules, which allow for local port reuse by
56 * an application. In essence:
58 * 1) Sockets bound to different interfaces may share a local port.
59 * Failing that, goto test 2.
60 * 2) If all sockets have sk->sk_reuse set, and none of them are in
61 * TCP_LISTEN state, the port may be shared.
62 * Failing that, goto test 3.
63 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
64 * address, and none of them are the same, the port may be
66 * Failing this, the port cannot be shared.
68 * The interesting point, is test #2. This is what an FTP server does
69 * all day. To optimize this case we use a specific flag bit defined
70 * below. As we add sockets to a bind bucket list, we perform a
71 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
72 * As long as all sockets added to a bind bucket pass this test,
73 * the flag bit will be set.
74 * The resulting situation is that tcp_v[46]_verify_bind() can just check
75 * for this flag bit, if it is set and the socket trying to bind has
76 * sk->sk_reuse set, we don't even have to walk the owners list at all,
77 * we return that it is ok to bind this socket to the requested local port.
79 * Sounds like a lot of work, but it is worth it. In a more naive
80 * implementation (ie. current FreeBSD etc.) the entire list of ports
81 * must be walked for each data port opened by an ftp server. Needless
82 * to say, this does not scale at all. With a couple thousand FTP
83 * users logged onto your box, isn't it nice to know that new data
84 * ports are created in O(1) time? I thought so. ;-) -DaveM
86 struct tcp_bind_bucket
{
88 signed short fastreuse
;
89 struct hlist_node node
;
90 struct hlist_head owners
;
93 #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node)
95 struct tcp_bind_hashbucket
{
97 struct hlist_head chain
;
100 static inline struct tcp_bind_bucket
*__tb_head(struct tcp_bind_hashbucket
*head
)
102 return hlist_entry(head
->chain
.first
, struct tcp_bind_bucket
, node
);
105 static inline struct tcp_bind_bucket
*tb_head(struct tcp_bind_hashbucket
*head
)
107 return hlist_empty(&head
->chain
) ? NULL
: __tb_head(head
);
110 extern struct tcp_hashinfo
{
111 /* This is for sockets with full identity only. Sockets here will
112 * always be without wildcards and will have the following invariant:
114 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
116 * First half of the table is for sockets not in TIME_WAIT, second half
117 * is for TIME_WAIT sockets only.
119 struct tcp_ehash_bucket
*__tcp_ehash
;
121 /* Ok, let's try this, I give up, we do need a local binding
122 * TCP hash as well as the others for fast bind/connect.
124 struct tcp_bind_hashbucket
*__tcp_bhash
;
126 int __tcp_bhash_size
;
127 int __tcp_ehash_size
;
129 /* All sockets in TCP_LISTEN state will be in here. This is the only
130 * table where wildcard'd TCP sockets can exist. Hash function here
131 * is just local port number.
133 struct hlist_head __tcp_listening_hash
[TCP_LHTABLE_SIZE
];
135 /* All the above members are written once at bootup and
136 * never written again _or_ are predominantly read-access.
138 * Now align to a new cache line as all the following members
141 rwlock_t __tcp_lhash_lock ____cacheline_aligned
;
142 atomic_t __tcp_lhash_users
;
143 wait_queue_head_t __tcp_lhash_wait
;
144 spinlock_t __tcp_portalloc_lock
;
147 #define tcp_ehash (tcp_hashinfo.__tcp_ehash)
148 #define tcp_bhash (tcp_hashinfo.__tcp_bhash)
149 #define tcp_ehash_size (tcp_hashinfo.__tcp_ehash_size)
150 #define tcp_bhash_size (tcp_hashinfo.__tcp_bhash_size)
151 #define tcp_listening_hash (tcp_hashinfo.__tcp_listening_hash)
152 #define tcp_lhash_lock (tcp_hashinfo.__tcp_lhash_lock)
153 #define tcp_lhash_users (tcp_hashinfo.__tcp_lhash_users)
154 #define tcp_lhash_wait (tcp_hashinfo.__tcp_lhash_wait)
155 #define tcp_portalloc_lock (tcp_hashinfo.__tcp_portalloc_lock)
157 extern kmem_cache_t
*tcp_bucket_cachep
;
158 extern struct tcp_bind_bucket
*tcp_bucket_create(struct tcp_bind_hashbucket
*head
,
159 unsigned short snum
);
160 extern void tcp_bucket_destroy(struct tcp_bind_bucket
*tb
);
161 extern void tcp_bucket_unlock(struct sock
*sk
);
162 extern int tcp_port_rover
;
164 /* These are AF independent. */
165 static __inline__
int tcp_bhashfn(__u16 lport
)
167 return (lport
& (tcp_bhash_size
- 1));
170 extern void tcp_bind_hash(struct sock
*sk
, struct tcp_bind_bucket
*tb
,
171 unsigned short snum
);
173 #if (BITS_PER_LONG == 64)
174 #define TCP_ADDRCMP_ALIGN_BYTES 8
176 #define TCP_ADDRCMP_ALIGN_BYTES 4
179 /* This is a TIME_WAIT bucket. It works around the memory consumption
180 * problems of sockets in such a state on heavily loaded servers, but
181 * without violating the protocol specification.
183 struct tcp_tw_bucket
{
185 * Now struct sock also uses sock_common, so please just
186 * don't add nothing before this first member (__tw_common) --acme
188 struct sock_common __tw_common
;
189 #define tw_family __tw_common.skc_family
190 #define tw_state __tw_common.skc_state
191 #define tw_reuse __tw_common.skc_reuse
192 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
193 #define tw_node __tw_common.skc_node
194 #define tw_bind_node __tw_common.skc_bind_node
195 #define tw_refcnt __tw_common.skc_refcnt
196 volatile unsigned char tw_substate
;
197 unsigned char tw_rcv_wscale
;
199 /* Socket demultiplex comparisons on incoming packets. */
200 /* these five are in inet_sock */
202 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES
)));
206 /* And these are ours. */
213 long tw_ts_recent_stamp
;
214 unsigned long tw_ttd
;
215 struct tcp_bind_bucket
*tw_tb
;
216 struct hlist_node tw_death_node
;
217 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
218 struct in6_addr tw_v6_daddr
;
219 struct in6_addr tw_v6_rcv_saddr
;
224 static __inline__
void tw_add_node(struct tcp_tw_bucket
*tw
,
225 struct hlist_head
*list
)
227 hlist_add_head(&tw
->tw_node
, list
);
230 static __inline__
void tw_add_bind_node(struct tcp_tw_bucket
*tw
,
231 struct hlist_head
*list
)
233 hlist_add_head(&tw
->tw_bind_node
, list
);
236 static inline int tw_dead_hashed(struct tcp_tw_bucket
*tw
)
238 return tw
->tw_death_node
.pprev
!= NULL
;
241 static __inline__
void tw_dead_node_init(struct tcp_tw_bucket
*tw
)
243 tw
->tw_death_node
.pprev
= NULL
;
246 static __inline__
void __tw_del_dead_node(struct tcp_tw_bucket
*tw
)
248 __hlist_del(&tw
->tw_death_node
);
249 tw_dead_node_init(tw
);
252 static __inline__
int tw_del_dead_node(struct tcp_tw_bucket
*tw
)
254 if (tw_dead_hashed(tw
)) {
255 __tw_del_dead_node(tw
);
261 #define tw_for_each(tw, node, head) \
262 hlist_for_each_entry(tw, node, head, tw_node)
264 #define tw_for_each_inmate(tw, node, jail) \
265 hlist_for_each_entry(tw, node, jail, tw_death_node)
267 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
268 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
270 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
272 static inline u32
tcp_v4_rcv_saddr(const struct sock
*sk
)
274 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
275 inet_sk(sk
)->rcv_saddr
: tcptw_sk(sk
)->tw_rcv_saddr
;
278 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
279 static inline struct in6_addr
*__tcp_v6_rcv_saddr(const struct sock
*sk
)
281 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
282 &inet6_sk(sk
)->rcv_saddr
: &tcptw_sk(sk
)->tw_v6_rcv_saddr
;
285 static inline struct in6_addr
*tcp_v6_rcv_saddr(const struct sock
*sk
)
287 return sk
->sk_family
== AF_INET6
? __tcp_v6_rcv_saddr(sk
) : NULL
;
290 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
292 static inline int tcp_v6_ipv6only(const struct sock
*sk
)
294 return likely(sk
->sk_state
!= TCP_TIME_WAIT
) ?
295 ipv6_only_sock(sk
) : tcptw_sk_ipv6only(sk
);
298 # define __tcp_v6_rcv_saddr(__sk) NULL
299 # define tcp_v6_rcv_saddr(__sk) NULL
300 # define tcptw_sk_ipv6only(__sk) 0
301 # define tcp_v6_ipv6only(__sk) 0
304 extern kmem_cache_t
*tcp_timewait_cachep
;
306 static inline void tcp_tw_put(struct tcp_tw_bucket
*tw
)
308 if (atomic_dec_and_test(&tw
->tw_refcnt
)) {
309 #ifdef INET_REFCNT_DEBUG
310 printk(KERN_DEBUG
"tw_bucket %p released\n", tw
);
312 kmem_cache_free(tcp_timewait_cachep
, tw
);
316 extern atomic_t tcp_orphan_count
;
317 extern int tcp_tw_count
;
318 extern void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
319 extern void tcp_tw_deschedule(struct tcp_tw_bucket
*tw
);
322 /* Socket demux engine toys. */
324 #define TCP_COMBINED_PORTS(__sport, __dport) \
325 (((__u32)(__sport)<<16) | (__u32)(__dport))
326 #else /* __LITTLE_ENDIAN */
327 #define TCP_COMBINED_PORTS(__sport, __dport) \
328 (((__u32)(__dport)<<16) | (__u32)(__sport))
331 #if (BITS_PER_LONG == 64)
333 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
334 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
335 #else /* __LITTLE_ENDIAN */
336 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
337 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
338 #endif /* __BIG_ENDIAN */
339 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
340 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
341 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
342 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
343 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
344 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
345 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
346 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
347 #else /* 32-bit arch */
348 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
349 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
350 ((inet_sk(__sk)->daddr == (__saddr)) && \
351 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
352 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
353 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
354 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
355 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
356 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
357 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
358 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
359 #endif /* 64-bit arch */
361 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
362 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
363 ((__sk)->sk_family == AF_INET6) && \
364 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \
365 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
366 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
368 /* These can have wildcards, don't try too hard. */
369 static __inline__
int tcp_lhashfn(unsigned short num
)
371 return num
& (TCP_LHTABLE_SIZE
- 1);
374 static __inline__
int tcp_sk_listen_hashfn(struct sock
*sk
)
376 return tcp_lhashfn(inet_sk(sk
)->num
);
379 #define MAX_TCP_HEADER (128 + MAX_HEADER)
382 * Never offer a window over 32767 without using window scaling. Some
383 * poor stacks do signed 16bit maths!
385 #define MAX_TCP_WINDOW 32767U
387 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
388 #define TCP_MIN_MSS 88U
390 /* Minimal RCV_MSS. */
391 #define TCP_MIN_RCVMSS 536U
393 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
394 #define TCP_FASTRETRANS_THRESH 3
396 /* Maximal reordering. */
397 #define TCP_MAX_REORDERING 127
399 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
400 #define TCP_MAX_QUICKACKS 16U
402 /* urg_data states */
403 #define TCP_URG_VALID 0x0100
404 #define TCP_URG_NOTYET 0x0200
405 #define TCP_URG_READ 0x0400
407 #define TCP_RETR1 3 /*
408 * This is how many retries it does before it
409 * tries to figure out if the gateway is
410 * down. Minimal RFC value is 3; it corresponds
411 * to ~3sec-8min depending on RTO.
414 #define TCP_RETR2 15 /*
415 * This should take at least
416 * 90 minutes to time out.
417 * RFC1122 says that the limit is 100 sec.
418 * 15 is ~13-30min depending on RTO.
421 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
422 * connection: ~180sec is RFC minumum */
424 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
425 * connection: ~180sec is RFC minumum */
428 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
429 * socket. 7 is ~50sec-16min.
433 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
434 * state, about 60 seconds */
435 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
436 /* BSD style FIN_WAIT2 deadlock breaker.
437 * It used to be 3min, new value is 60sec,
438 * to combine FIN-WAIT-2 timeout with
442 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
444 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
445 #define TCP_ATO_MIN ((unsigned)(HZ/25))
447 #define TCP_DELACK_MIN 4U
448 #define TCP_ATO_MIN 4U
450 #define TCP_RTO_MAX ((unsigned)(120*HZ))
451 #define TCP_RTO_MIN ((unsigned)(HZ/5))
452 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
454 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
455 * for local resources.
458 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
459 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
460 #define TCP_KEEPALIVE_INTVL (75*HZ)
462 #define MAX_TCP_KEEPIDLE 32767
463 #define MAX_TCP_KEEPINTVL 32767
464 #define MAX_TCP_KEEPCNT 127
465 #define MAX_TCP_SYNCNT 127
467 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
468 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
470 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
471 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
472 * after this time. It should be equal
473 * (or greater than) TCP_TIMEWAIT_LEN
474 * to provide reliability equal to one
475 * provided by timewait state.
477 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
478 * timestamps. It must be less than
479 * minimal timewait lifetime.
482 #define TCP_TW_RECYCLE_SLOTS_LOG 5
483 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
485 /* If time > 4sec, it is "slow" path, no recycling is required,
486 so that we select tick to get range about 4 seconds.
489 #if HZ <= 16 || HZ > 4096
490 # error Unsupported: HZ <= 16 or HZ > 4096
492 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
494 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
496 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
498 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
500 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
502 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
504 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
506 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
509 #define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
510 * max_cwnd = snd_cwnd * beta
512 #define BICTCP_MAX_INCREMENT 32 /*
513 * Limit on the amount of
514 * increment allowed during
517 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
518 * log(B/Smin)/log(B/(B-1))+1,
522 #define BICTCP_B 4 /*
524 * go to point (max+min)/N
531 #define TCPOPT_NOP 1 /* Padding */
532 #define TCPOPT_EOL 0 /* End of options */
533 #define TCPOPT_MSS 2 /* Segment size negotiating */
534 #define TCPOPT_WINDOW 3 /* Window scaling */
535 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
536 #define TCPOPT_SACK 5 /* SACK Block */
537 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
543 #define TCPOLEN_MSS 4
544 #define TCPOLEN_WINDOW 3
545 #define TCPOLEN_SACK_PERM 2
546 #define TCPOLEN_TIMESTAMP 10
548 /* But this is what stacks really send out. */
549 #define TCPOLEN_TSTAMP_ALIGNED 12
550 #define TCPOLEN_WSCALE_ALIGNED 4
551 #define TCPOLEN_SACKPERM_ALIGNED 4
552 #define TCPOLEN_SACK_BASE 2
553 #define TCPOLEN_SACK_BASE_ALIGNED 4
554 #define TCPOLEN_SACK_PERBLOCK 8
556 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
557 #define TCP_TIME_DACK 2 /* Delayed ack timer */
558 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
559 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
561 /* Flags in tp->nonagle */
562 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
563 #define TCP_NAGLE_CORK 2 /* Socket is corked */
564 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
566 /* sysctl variables for tcp */
567 extern int sysctl_max_syn_backlog
;
568 extern int sysctl_tcp_timestamps
;
569 extern int sysctl_tcp_window_scaling
;
570 extern int sysctl_tcp_sack
;
571 extern int sysctl_tcp_fin_timeout
;
572 extern int sysctl_tcp_tw_recycle
;
573 extern int sysctl_tcp_keepalive_time
;
574 extern int sysctl_tcp_keepalive_probes
;
575 extern int sysctl_tcp_keepalive_intvl
;
576 extern int sysctl_tcp_syn_retries
;
577 extern int sysctl_tcp_synack_retries
;
578 extern int sysctl_tcp_retries1
;
579 extern int sysctl_tcp_retries2
;
580 extern int sysctl_tcp_orphan_retries
;
581 extern int sysctl_tcp_syncookies
;
582 extern int sysctl_tcp_retrans_collapse
;
583 extern int sysctl_tcp_stdurg
;
584 extern int sysctl_tcp_rfc1337
;
585 extern int sysctl_tcp_abort_on_overflow
;
586 extern int sysctl_tcp_max_orphans
;
587 extern int sysctl_tcp_max_tw_buckets
;
588 extern int sysctl_tcp_fack
;
589 extern int sysctl_tcp_reordering
;
590 extern int sysctl_tcp_ecn
;
591 extern int sysctl_tcp_dsack
;
592 extern int sysctl_tcp_mem
[3];
593 extern int sysctl_tcp_wmem
[3];
594 extern int sysctl_tcp_rmem
[3];
595 extern int sysctl_tcp_app_win
;
596 extern int sysctl_tcp_adv_win_scale
;
597 extern int sysctl_tcp_tw_reuse
;
598 extern int sysctl_tcp_frto
;
599 extern int sysctl_tcp_low_latency
;
600 extern int sysctl_tcp_westwood
;
601 extern int sysctl_tcp_vegas_cong_avoid
;
602 extern int sysctl_tcp_vegas_alpha
;
603 extern int sysctl_tcp_vegas_beta
;
604 extern int sysctl_tcp_vegas_gamma
;
605 extern int sysctl_tcp_nometrics_save
;
606 extern int sysctl_tcp_bic
;
607 extern int sysctl_tcp_bic_fast_convergence
;
608 extern int sysctl_tcp_bic_low_window
;
609 extern int sysctl_tcp_bic_beta
;
610 extern int sysctl_tcp_moderate_rcvbuf
;
611 extern int sysctl_tcp_tso_win_divisor
;
613 extern atomic_t tcp_memory_allocated
;
614 extern atomic_t tcp_sockets_allocated
;
615 extern int tcp_memory_pressure
;
617 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
618 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
620 #define TCP_INET_FAMILY(fam) 1
624 * Pointers to address related TCP functions
625 * (i.e. things that depend on the address family)
629 int (*queue_xmit
) (struct sk_buff
*skb
,
632 void (*send_check
) (struct sock
*sk
,
635 struct sk_buff
*skb
);
637 int (*rebuild_header
) (struct sock
*sk
);
639 int (*conn_request
) (struct sock
*sk
,
640 struct sk_buff
*skb
);
642 struct sock
* (*syn_recv_sock
) (struct sock
*sk
,
644 struct open_request
*req
,
645 struct dst_entry
*dst
);
647 int (*remember_stamp
) (struct sock
*sk
);
649 __u16 net_header_len
;
651 int (*setsockopt
) (struct sock
*sk
,
657 int (*getsockopt
) (struct sock
*sk
,
664 void (*addr2sockaddr
) (struct sock
*sk
,
671 * The next routines deal with comparing 32 bit unsigned ints
672 * and worry about wraparound (automatic with unsigned arithmetic).
675 static inline int before(__u32 seq1
, __u32 seq2
)
677 return (__s32
)(seq1
-seq2
) < 0;
680 static inline int after(__u32 seq1
, __u32 seq2
)
682 return (__s32
)(seq2
-seq1
) < 0;
686 /* is s2<=s1<=s3 ? */
687 static inline int between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
689 return seq3
- seq2
>= seq1
- seq2
;
693 extern struct proto tcp_prot
;
695 DECLARE_SNMP_STAT(struct tcp_mib
, tcp_statistics
);
696 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
697 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
698 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
699 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
700 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
701 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
703 extern void tcp_put_port(struct sock
*sk
);
704 extern void tcp_inherit_port(struct sock
*sk
, struct sock
*child
);
706 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
708 extern void tcp_shutdown (struct sock
*sk
, int how
);
710 extern int tcp_v4_rcv(struct sk_buff
*skb
);
712 extern int tcp_v4_remember_stamp(struct sock
*sk
);
714 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket
*tw
);
716 extern int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
717 struct msghdr
*msg
, size_t size
);
718 extern ssize_t
tcp_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
);
720 extern int tcp_ioctl(struct sock
*sk
,
724 extern int tcp_rcv_state_process(struct sock
*sk
,
729 extern int tcp_rcv_established(struct sock
*sk
,
734 extern void tcp_rcv_space_adjust(struct sock
*sk
);
743 static inline void tcp_schedule_ack(struct tcp_sock
*tp
)
745 tp
->ack
.pending
|= TCP_ACK_SCHED
;
748 static inline int tcp_ack_scheduled(struct tcp_sock
*tp
)
750 return tp
->ack
.pending
&TCP_ACK_SCHED
;
753 static __inline__
void tcp_dec_quickack_mode(struct tcp_sock
*tp
)
755 if (tp
->ack
.quick
&& --tp
->ack
.quick
== 0) {
756 /* Leaving quickack mode we deflate ATO. */
757 tp
->ack
.ato
= TCP_ATO_MIN
;
761 extern void tcp_enter_quickack_mode(struct tcp_sock
*tp
);
763 static __inline__
void tcp_delack_init(struct tcp_sock
*tp
)
765 memset(&tp
->ack
, 0, sizeof(tp
->ack
));
768 static inline void tcp_clear_options(struct tcp_options_received
*rx_opt
)
770 rx_opt
->tstamp_ok
= rx_opt
->sack_ok
= rx_opt
->wscale_ok
= rx_opt
->snd_wscale
= 0;
782 extern enum tcp_tw_status
tcp_timewait_state_process(struct tcp_tw_bucket
*tw
,
787 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
788 struct open_request
*req
,
789 struct open_request
**prev
);
790 extern int tcp_child_process(struct sock
*parent
,
792 struct sk_buff
*skb
);
793 extern void tcp_enter_frto(struct sock
*sk
);
794 extern void tcp_enter_loss(struct sock
*sk
, int how
);
795 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
796 extern void tcp_update_metrics(struct sock
*sk
);
798 extern void tcp_close(struct sock
*sk
,
800 extern struct sock
* tcp_accept(struct sock
*sk
, int flags
, int *err
);
801 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
, struct poll_table_struct
*wait
);
803 extern int tcp_getsockopt(struct sock
*sk
, int level
,
807 extern int tcp_setsockopt(struct sock
*sk
, int level
,
808 int optname
, char __user
*optval
,
810 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
811 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
813 size_t len
, int nonblock
,
814 int flags
, int *addr_len
);
816 extern int tcp_listen_start(struct sock
*sk
);
818 extern void tcp_parse_options(struct sk_buff
*skb
,
819 struct tcp_options_received
*opt_rx
,
823 * TCP v4 functions exported for the inet6 API
826 extern int tcp_v4_rebuild_header(struct sock
*sk
);
828 extern int tcp_v4_build_header(struct sock
*sk
,
829 struct sk_buff
*skb
);
831 extern void tcp_v4_send_check(struct sock
*sk
,
832 struct tcphdr
*th
, int len
,
833 struct sk_buff
*skb
);
835 extern int tcp_v4_conn_request(struct sock
*sk
,
836 struct sk_buff
*skb
);
838 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
839 struct open_request
*req
,
840 struct sk_buff
*skb
);
842 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
,
844 struct open_request
*req
,
845 struct dst_entry
*dst
);
847 extern int tcp_v4_do_rcv(struct sock
*sk
,
848 struct sk_buff
*skb
);
850 extern int tcp_v4_connect(struct sock
*sk
,
851 struct sockaddr
*uaddr
,
854 extern int tcp_connect(struct sock
*sk
);
856 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
,
857 struct dst_entry
*dst
,
858 struct open_request
*req
);
860 extern int tcp_disconnect(struct sock
*sk
, int flags
);
862 extern void tcp_unhash(struct sock
*sk
);
864 extern int tcp_v4_hash_connecting(struct sock
*sk
);
867 /* From syncookies.c */
868 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
869 struct ip_options
*opt
);
870 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
875 extern int tcp_write_xmit(struct sock
*, int nonagle
);
876 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
877 extern void tcp_xmit_retransmit_queue(struct sock
*);
878 extern void tcp_simple_retransmit(struct sock
*);
879 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
881 extern void tcp_send_probe0(struct sock
*);
882 extern void tcp_send_partial(struct sock
*);
883 extern int tcp_write_wakeup(struct sock
*);
884 extern void tcp_send_fin(struct sock
*sk
);
885 extern void tcp_send_active_reset(struct sock
*sk
, int priority
);
886 extern int tcp_send_synack(struct sock
*);
887 extern void tcp_push_one(struct sock
*, unsigned mss_now
);
888 extern void tcp_send_ack(struct sock
*sk
);
889 extern void tcp_send_delayed_ack(struct sock
*sk
);
892 extern void tcp_init_xmit_timers(struct sock
*);
893 extern void tcp_clear_xmit_timers(struct sock
*);
895 extern void tcp_delete_keepalive_timer(struct sock
*);
896 extern void tcp_reset_keepalive_timer(struct sock
*, unsigned long);
897 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
898 extern unsigned int tcp_current_mss(struct sock
*sk
, int large
);
901 extern const char tcp_timer_bug_msg
[];
905 extern void tcp_get_info(struct sock
*, struct tcp_info
*);
907 /* Read 'sendfile()'-style from a TCP socket */
908 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
909 unsigned int, size_t);
910 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
911 sk_read_actor_t recv_actor
);
913 static inline void tcp_clear_xmit_timer(struct sock
*sk
, int what
)
915 struct tcp_sock
*tp
= tcp_sk(sk
);
918 case TCP_TIME_RETRANS
:
919 case TCP_TIME_PROBE0
:
922 #ifdef TCP_CLEAR_TIMERS
923 sk_stop_timer(sk
, &tp
->retransmit_timer
);
930 #ifdef TCP_CLEAR_TIMERS
931 sk_stop_timer(sk
, &tp
->delack_timer
);
936 printk(tcp_timer_bug_msg
);
944 * Reset the retransmission timer
946 static inline void tcp_reset_xmit_timer(struct sock
*sk
, int what
, unsigned long when
)
948 struct tcp_sock
*tp
= tcp_sk(sk
);
950 if (when
> TCP_RTO_MAX
) {
952 printk(KERN_DEBUG
"reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk
, what
, when
, current_text_addr());
958 case TCP_TIME_RETRANS
:
959 case TCP_TIME_PROBE0
:
961 tp
->timeout
= jiffies
+when
;
962 sk_reset_timer(sk
, &tp
->retransmit_timer
, tp
->timeout
);
966 tp
->ack
.pending
|= TCP_ACK_TIMER
;
967 tp
->ack
.timeout
= jiffies
+when
;
968 sk_reset_timer(sk
, &tp
->delack_timer
, tp
->ack
.timeout
);
973 printk(tcp_timer_bug_msg
);
979 /* Initialize RCV_MSS value.
980 * RCV_MSS is an our guess about MSS used by the peer.
981 * We haven't any direct information about the MSS.
982 * It's better to underestimate the RCV_MSS rather than overestimate.
983 * Overestimations make us ACKing less frequently than needed.
984 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
987 static inline void tcp_initialize_rcv_mss(struct sock
*sk
)
989 struct tcp_sock
*tp
= tcp_sk(sk
);
990 unsigned int hint
= min(tp
->advmss
, tp
->mss_cache_std
);
992 hint
= min(hint
, tp
->rcv_wnd
/2);
993 hint
= min(hint
, TCP_MIN_RCVMSS
);
994 hint
= max(hint
, TCP_MIN_MSS
);
996 tp
->ack
.rcv_mss
= hint
;
999 static __inline__
void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
1001 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
1002 ntohl(TCP_FLAG_ACK
) |
1006 static __inline__
void tcp_fast_path_on(struct tcp_sock
*tp
)
1008 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
1011 static inline void tcp_fast_path_check(struct sock
*sk
, struct tcp_sock
*tp
)
1013 if (skb_queue_len(&tp
->out_of_order_queue
) == 0 &&
1015 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
1017 tcp_fast_path_on(tp
);
1020 /* Compute the actual receive window we are currently advertising.
1021 * Rcv_nxt can be after the window if our peer push more data
1022 * than the offered window.
1024 static __inline__ u32
tcp_receive_window(const struct tcp_sock
*tp
)
1026 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
1033 /* Choose a new window, without checks for shrinking, and without
1034 * scaling applied to the result. The caller does these things
1035 * if necessary. This is a "raw" window selection.
1037 extern u32
__tcp_select_window(struct sock
*sk
);
1039 /* TCP timestamps are only 32-bits, this causes a slight
1040 * complication on 64-bit systems since we store a snapshot
1041 * of jiffies in the buffer control blocks below. We decidely
1042 * only use of the low 32-bits of jiffies and hide the ugly
1043 * casts with the following macro.
1045 #define tcp_time_stamp ((__u32)(jiffies))
1047 /* This is what the send packet queueing engine uses to pass
1048 * TCP per-packet control information to the transmission
1049 * code. We also store the host-order sequence numbers in
1050 * here too. This is 36 bytes on 32-bit architectures,
1051 * 40 bytes on 64-bit machines, if this grows please adjust
1052 * skbuff.h:skbuff->cb[xxx] size appropriately.
1056 struct inet_skb_parm h4
;
1057 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1058 struct inet6_skb_parm h6
;
1060 } header
; /* For incoming frames */
1061 __u32 seq
; /* Starting sequence number */
1062 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
1063 __u32 when
; /* used to compute rtt's */
1064 __u8 flags
; /* TCP header flags. */
1066 /* NOTE: These must match up to the flags byte in a
1069 #define TCPCB_FLAG_FIN 0x01
1070 #define TCPCB_FLAG_SYN 0x02
1071 #define TCPCB_FLAG_RST 0x04
1072 #define TCPCB_FLAG_PSH 0x08
1073 #define TCPCB_FLAG_ACK 0x10
1074 #define TCPCB_FLAG_URG 0x20
1075 #define TCPCB_FLAG_ECE 0x40
1076 #define TCPCB_FLAG_CWR 0x80
1078 __u8 sacked
; /* State flags for SACK/FACK. */
1079 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1080 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1081 #define TCPCB_LOST 0x04 /* SKB is lost */
1082 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1084 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1085 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1087 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1089 #define TCPCB_AT_TAIL (TCPCB_URG)
1091 __u16 urg_ptr
; /* Valid w/URG flags is set. */
1092 __u32 ack_seq
; /* Sequence number ACK'd */
1095 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1097 #include <net/tcp_ecn.h>
1099 /* Due to TSO, an SKB can be composed of multiple actual
1100 * packets. To keep these tracked properly, we use this.
1102 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
1104 return skb_shinfo(skb
)->tso_segs
;
1107 /* This is valid iff tcp_skb_pcount() > 1. */
1108 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
1110 return skb_shinfo(skb
)->tso_size
;
1113 static inline void tcp_dec_pcount_approx(__u32
*count
,
1114 const struct sk_buff
*skb
)
1117 *count
-= tcp_skb_pcount(skb
);
1118 if ((int)*count
< 0)
1123 static inline void tcp_packets_out_inc(struct sock
*sk
,
1124 struct tcp_sock
*tp
,
1125 const struct sk_buff
*skb
)
1127 int orig
= tp
->packets_out
;
1129 tp
->packets_out
+= tcp_skb_pcount(skb
);
1131 tcp_reset_xmit_timer(sk
, TCP_TIME_RETRANS
, tp
->rto
);
1134 static inline void tcp_packets_out_dec(struct tcp_sock
*tp
,
1135 const struct sk_buff
*skb
)
1137 tp
->packets_out
-= tcp_skb_pcount(skb
);
1140 /* This determines how many packets are "in the network" to the best
1141 * of our knowledge. In many cases it is conservative, but where
1142 * detailed information is available from the receiver (via SACK
1143 * blocks etc.) we can make more aggressive calculations.
1145 * Use this for decisions involving congestion control, use just
1146 * tp->packets_out to determine if the send queue is empty or not.
1148 * Read this equation as:
1150 * "Packets sent once on transmission queue" MINUS
1151 * "Packets left network, but not honestly ACKed yet" PLUS
1152 * "Packets fast retransmitted"
1154 static __inline__
unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
1156 return (tp
->packets_out
- tp
->left_out
+ tp
->retrans_out
);
1160 * Which congestion algorithim is in use on the connection.
1162 #define tcp_is_vegas(__tp) ((__tp)->adv_cong == TCP_VEGAS)
1163 #define tcp_is_westwood(__tp) ((__tp)->adv_cong == TCP_WESTWOOD)
1164 #define tcp_is_bic(__tp) ((__tp)->adv_cong == TCP_BIC)
1166 /* Recalculate snd_ssthresh, we want to set it to:
1169 * one half the current congestion window, but no
1170 * less than two segments
1173 * behave like Reno until low_window is reached,
1174 * then increase congestion window slowly
1176 static inline __u32
tcp_recalc_ssthresh(struct tcp_sock
*tp
)
1178 if (tcp_is_bic(tp
)) {
1179 if (sysctl_tcp_bic_fast_convergence
&&
1180 tp
->snd_cwnd
< tp
->bictcp
.last_max_cwnd
)
1181 tp
->bictcp
.last_max_cwnd
= (tp
->snd_cwnd
*
1183 + sysctl_tcp_bic_beta
))
1184 / (2 * BICTCP_BETA_SCALE
);
1186 tp
->bictcp
.last_max_cwnd
= tp
->snd_cwnd
;
1188 if (tp
->snd_cwnd
> sysctl_tcp_bic_low_window
)
1189 return max((tp
->snd_cwnd
* sysctl_tcp_bic_beta
)
1190 / BICTCP_BETA_SCALE
, 2U);
1193 return max(tp
->snd_cwnd
>> 1U, 2U);
1196 /* Stop taking Vegas samples for now. */
1197 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1199 static inline void tcp_vegas_enable(struct tcp_sock
*tp
)
1201 /* There are several situations when we must "re-start" Vegas:
1203 * o when a connection is established
1205 * o after fast recovery
1206 * o when we send a packet and there is no outstanding
1207 * unacknowledged data (restarting an idle connection)
1209 * In these circumstances we cannot do a Vegas calculation at the
1210 * end of the first RTT, because any calculation we do is using
1211 * stale info -- both the saved cwnd and congestion feedback are
1214 * Instead we must wait until the completion of an RTT during
1215 * which we actually receive ACKs.
1218 /* Begin taking Vegas samples next time we send something. */
1219 tp
->vegas
.doing_vegas_now
= 1;
1221 /* Set the beginning of the next send window. */
1222 tp
->vegas
.beg_snd_nxt
= tp
->snd_nxt
;
1224 tp
->vegas
.cntRTT
= 0;
1225 tp
->vegas
.minRTT
= 0x7fffffff;
1228 /* Should we be taking Vegas samples right now? */
1229 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1231 extern void tcp_ca_init(struct tcp_sock
*tp
);
1233 static inline void tcp_set_ca_state(struct tcp_sock
*tp
, u8 ca_state
)
1235 if (tcp_is_vegas(tp
)) {
1236 if (ca_state
== TCP_CA_Open
)
1237 tcp_vegas_enable(tp
);
1239 tcp_vegas_disable(tp
);
1241 tp
->ca_state
= ca_state
;
1244 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1245 * The exception is rate halving phase, when cwnd is decreasing towards
1248 static inline __u32
tcp_current_ssthresh(struct tcp_sock
*tp
)
1250 if ((1<<tp
->ca_state
)&(TCPF_CA_CWR
|TCPF_CA_Recovery
))
1251 return tp
->snd_ssthresh
;
1253 return max(tp
->snd_ssthresh
,
1254 ((tp
->snd_cwnd
>> 1) +
1255 (tp
->snd_cwnd
>> 2)));
1258 static inline void tcp_sync_left_out(struct tcp_sock
*tp
)
1260 if (tp
->rx_opt
.sack_ok
&&
1261 (tp
->sacked_out
>= tp
->packets_out
- tp
->lost_out
))
1262 tp
->sacked_out
= tp
->packets_out
- tp
->lost_out
;
1263 tp
->left_out
= tp
->sacked_out
+ tp
->lost_out
;
1266 extern void tcp_cwnd_application_limited(struct sock
*sk
);
1268 /* Congestion window validation. (RFC2861) */
1270 static inline void tcp_cwnd_validate(struct sock
*sk
, struct tcp_sock
*tp
)
1272 __u32 packets_out
= tp
->packets_out
;
1274 if (packets_out
>= tp
->snd_cwnd
) {
1275 /* Network is feed fully. */
1276 tp
->snd_cwnd_used
= 0;
1277 tp
->snd_cwnd_stamp
= tcp_time_stamp
;
1279 /* Network starves. */
1280 if (tp
->packets_out
> tp
->snd_cwnd_used
)
1281 tp
->snd_cwnd_used
= tp
->packets_out
;
1283 if ((s32
)(tcp_time_stamp
- tp
->snd_cwnd_stamp
) >= tp
->rto
)
1284 tcp_cwnd_application_limited(sk
);
1288 /* Set slow start threshould and cwnd not falling to slow start */
1289 static inline void __tcp_enter_cwr(struct tcp_sock
*tp
)
1291 tp
->undo_marker
= 0;
1292 tp
->snd_ssthresh
= tcp_recalc_ssthresh(tp
);
1293 tp
->snd_cwnd
= min(tp
->snd_cwnd
,
1294 tcp_packets_in_flight(tp
) + 1U);
1295 tp
->snd_cwnd_cnt
= 0;
1296 tp
->high_seq
= tp
->snd_nxt
;
1297 tp
->snd_cwnd_stamp
= tcp_time_stamp
;
1298 TCP_ECN_queue_cwr(tp
);
1301 static inline void tcp_enter_cwr(struct tcp_sock
*tp
)
1303 tp
->prior_ssthresh
= 0;
1304 if (tp
->ca_state
< TCP_CA_CWR
) {
1305 __tcp_enter_cwr(tp
);
1306 tcp_set_ca_state(tp
, TCP_CA_CWR
);
1310 extern __u32
tcp_init_cwnd(struct tcp_sock
*tp
, struct dst_entry
*dst
);
1312 /* Slow start with delack produces 3 packets of burst, so that
1313 * it is safe "de facto".
1315 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
1320 static __inline__
int tcp_minshall_check(const struct tcp_sock
*tp
)
1322 return after(tp
->snd_sml
,tp
->snd_una
) &&
1323 !after(tp
->snd_sml
, tp
->snd_nxt
);
1326 static __inline__
void tcp_minshall_update(struct tcp_sock
*tp
, int mss
,
1327 const struct sk_buff
*skb
)
1330 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
1333 /* Return 0, if packet can be sent now without violation Nagle's rules:
1334 1. It is full sized.
1335 2. Or it contains FIN.
1336 3. Or TCP_NODELAY was set.
1337 4. Or TCP_CORK is not set, and all sent packets are ACKed.
1338 With Minshall's modification: all sent small packets are ACKed.
1341 static __inline__
int
1342 tcp_nagle_check(const struct tcp_sock
*tp
, const struct sk_buff
*skb
,
1343 unsigned mss_now
, int nonagle
)
1345 return (skb
->len
< mss_now
&&
1346 !(TCP_SKB_CB(skb
)->flags
& TCPCB_FLAG_FIN
) &&
1347 ((nonagle
&TCP_NAGLE_CORK
) ||
1350 tcp_minshall_check(tp
))));
1353 extern void tcp_set_skb_tso_segs(struct sock
*, struct sk_buff
*);
1355 /* This checks if the data bearing packet SKB (usually sk->sk_send_head)
1356 * should be put on the wire right now.
1358 static __inline__
int tcp_snd_test(struct sock
*sk
,
1359 struct sk_buff
*skb
,
1360 unsigned cur_mss
, int nonagle
)
1362 struct tcp_sock
*tp
= tcp_sk(sk
);
1363 int pkts
= tcp_skb_pcount(skb
);
1366 tcp_set_skb_tso_segs(sk
, skb
);
1367 pkts
= tcp_skb_pcount(skb
);
1370 /* RFC 1122 - section 4.2.3.4
1374 * a) The right edge of this frame exceeds the window
1375 * b) There are packets in flight and we have a small segment
1376 * [SWS avoidance and Nagle algorithm]
1377 * (part of SWS is done on packetization)
1378 * Minshall version sounds: there are no _small_
1379 * segments in flight. (tcp_nagle_check)
1380 * c) We have too many packets 'in flight'
1382 * Don't use the nagle rule for urgent data (or
1383 * for the final FIN -DaveM).
1385 * Also, Nagle rule does not apply to frames, which
1386 * sit in the middle of queue (they have no chances
1387 * to get new data) and if room at tail of skb is
1388 * not enough to save something seriously (<32 for now).
1391 /* Don't be strict about the congestion window for the
1392 * final FIN frame. -DaveM
1394 return (((nonagle
&TCP_NAGLE_PUSH
) || tp
->urg_mode
1395 || !tcp_nagle_check(tp
, skb
, cur_mss
, nonagle
)) &&
1396 (((tcp_packets_in_flight(tp
) + (pkts
-1)) < tp
->snd_cwnd
) ||
1397 (TCP_SKB_CB(skb
)->flags
& TCPCB_FLAG_FIN
)) &&
1398 !after(TCP_SKB_CB(skb
)->end_seq
, tp
->snd_una
+ tp
->snd_wnd
));
1401 static __inline__
void tcp_check_probe_timer(struct sock
*sk
, struct tcp_sock
*tp
)
1403 if (!tp
->packets_out
&& !tp
->pending
)
1404 tcp_reset_xmit_timer(sk
, TCP_TIME_PROBE0
, tp
->rto
);
1407 static __inline__
int tcp_skb_is_last(const struct sock
*sk
,
1408 const struct sk_buff
*skb
)
1410 return skb
->next
== (struct sk_buff
*)&sk
->sk_write_queue
;
1413 /* Push out any pending frames which were held back due to
1414 * TCP_CORK or attempt at coalescing tiny packets.
1415 * The socket must be locked by the caller.
1417 static __inline__
void __tcp_push_pending_frames(struct sock
*sk
,
1418 struct tcp_sock
*tp
,
1422 struct sk_buff
*skb
= sk
->sk_send_head
;
1425 if (!tcp_skb_is_last(sk
, skb
))
1426 nonagle
= TCP_NAGLE_PUSH
;
1427 if (!tcp_snd_test(sk
, skb
, cur_mss
, nonagle
) ||
1428 tcp_write_xmit(sk
, nonagle
))
1429 tcp_check_probe_timer(sk
, tp
);
1431 tcp_cwnd_validate(sk
, tp
);
1434 static __inline__
void tcp_push_pending_frames(struct sock
*sk
,
1435 struct tcp_sock
*tp
)
1437 __tcp_push_pending_frames(sk
, tp
, tcp_current_mss(sk
, 1), tp
->nonagle
);
1440 static __inline__
int tcp_may_send_now(struct sock
*sk
, struct tcp_sock
*tp
)
1442 struct sk_buff
*skb
= sk
->sk_send_head
;
1445 tcp_snd_test(sk
, skb
, tcp_current_mss(sk
, 1),
1446 tcp_skb_is_last(sk
, skb
) ? TCP_NAGLE_PUSH
: tp
->nonagle
));
1449 static __inline__
void tcp_init_wl(struct tcp_sock
*tp
, u32 ack
, u32 seq
)
1454 static __inline__
void tcp_update_wl(struct tcp_sock
*tp
, u32 ack
, u32 seq
)
1459 extern void tcp_destroy_sock(struct sock
*sk
);
1463 * Calculate(/check) TCP checksum
1465 static __inline__ u16
tcp_v4_check(struct tcphdr
*th
, int len
,
1466 unsigned long saddr
, unsigned long daddr
,
1469 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
1472 static __inline__
int __tcp_checksum_complete(struct sk_buff
*skb
)
1474 return (unsigned short)csum_fold(skb_checksum(skb
, 0, skb
->len
, skb
->csum
));
1477 static __inline__
int tcp_checksum_complete(struct sk_buff
*skb
)
1479 return skb
->ip_summed
!= CHECKSUM_UNNECESSARY
&&
1480 __tcp_checksum_complete(skb
);
1483 /* Prequeue for VJ style copy to user, combined with checksumming. */
1485 static __inline__
void tcp_prequeue_init(struct tcp_sock
*tp
)
1487 tp
->ucopy
.task
= NULL
;
1489 tp
->ucopy
.memory
= 0;
1490 skb_queue_head_init(&tp
->ucopy
.prequeue
);
1493 /* Packet is added to VJ-style prequeue for processing in process
1494 * context, if a reader task is waiting. Apparently, this exciting
1495 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1496 * failed somewhere. Latency? Burstiness? Well, at least now we will
1497 * see, why it failed. 8)8) --ANK
1499 * NOTE: is this not too big to inline?
1501 static __inline__
int tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1503 struct tcp_sock
*tp
= tcp_sk(sk
);
1505 if (!sysctl_tcp_low_latency
&& tp
->ucopy
.task
) {
1506 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1507 tp
->ucopy
.memory
+= skb
->truesize
;
1508 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1509 struct sk_buff
*skb1
;
1511 BUG_ON(sock_owned_by_user(sk
));
1513 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1514 sk
->sk_backlog_rcv(sk
, skb1
);
1515 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED
);
1518 tp
->ucopy
.memory
= 0;
1519 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1520 wake_up_interruptible(sk
->sk_sleep
);
1521 if (!tcp_ack_scheduled(tp
))
1522 tcp_reset_xmit_timer(sk
, TCP_TIME_DACK
, (3*TCP_RTO_MIN
)/4);
1533 static const char *statename
[]={
1534 "Unused","Established","Syn Sent","Syn Recv",
1535 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1536 "Close Wait","Last ACK","Listen","Closing"
1540 static __inline__
void tcp_set_state(struct sock
*sk
, int state
)
1542 int oldstate
= sk
->sk_state
;
1545 case TCP_ESTABLISHED
:
1546 if (oldstate
!= TCP_ESTABLISHED
)
1547 TCP_INC_STATS(TCP_MIB_CURRESTAB
);
1551 if (oldstate
== TCP_CLOSE_WAIT
|| oldstate
== TCP_ESTABLISHED
)
1552 TCP_INC_STATS(TCP_MIB_ESTABRESETS
);
1554 sk
->sk_prot
->unhash(sk
);
1555 if (tcp_sk(sk
)->bind_hash
&&
1556 !(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
))
1560 if (oldstate
==TCP_ESTABLISHED
)
1561 TCP_DEC_STATS(TCP_MIB_CURRESTAB
);
1564 /* Change state AFTER socket is unhashed to avoid closed
1565 * socket sitting in hash tables.
1567 sk
->sk_state
= state
;
1570 SOCK_DEBUG(sk
, "TCP sk=%p, State %s -> %s\n",sk
, statename
[oldstate
],statename
[state
]);
1574 static __inline__
void tcp_done(struct sock
*sk
)
1576 tcp_set_state(sk
, TCP_CLOSE
);
1577 tcp_clear_xmit_timers(sk
);
1579 sk
->sk_shutdown
= SHUTDOWN_MASK
;
1581 if (!sock_flag(sk
, SOCK_DEAD
))
1582 sk
->sk_state_change(sk
);
1584 tcp_destroy_sock(sk
);
1587 static __inline__
void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1590 rx_opt
->eff_sacks
= 0;
1591 rx_opt
->num_sacks
= 0;
1594 static __inline__
void tcp_build_and_update_options(__u32
*ptr
, struct tcp_sock
*tp
, __u32 tstamp
)
1596 if (tp
->rx_opt
.tstamp_ok
) {
1597 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) |
1598 (TCPOPT_NOP
<< 16) |
1599 (TCPOPT_TIMESTAMP
<< 8) |
1601 *ptr
++ = htonl(tstamp
);
1602 *ptr
++ = htonl(tp
->rx_opt
.ts_recent
);
1604 if (tp
->rx_opt
.eff_sacks
) {
1605 struct tcp_sack_block
*sp
= tp
->rx_opt
.dsack
? tp
->duplicate_sack
: tp
->selective_acks
;
1608 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) |
1609 (TCPOPT_NOP
<< 16) |
1610 (TCPOPT_SACK
<< 8) |
1611 (TCPOLEN_SACK_BASE
+
1612 (tp
->rx_opt
.eff_sacks
* TCPOLEN_SACK_PERBLOCK
)));
1613 for(this_sack
= 0; this_sack
< tp
->rx_opt
.eff_sacks
; this_sack
++) {
1614 *ptr
++ = htonl(sp
[this_sack
].start_seq
);
1615 *ptr
++ = htonl(sp
[this_sack
].end_seq
);
1617 if (tp
->rx_opt
.dsack
) {
1618 tp
->rx_opt
.dsack
= 0;
1619 tp
->rx_opt
.eff_sacks
--;
1624 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1625 * If this is every changed make sure to change the definition of
1626 * MAX_SYN_SIZE to match the new maximum number of options that you
1629 static inline void tcp_syn_build_options(__u32
*ptr
, int mss
, int ts
, int sack
,
1630 int offer_wscale
, int wscale
, __u32 tstamp
, __u32 ts_recent
)
1632 /* We always get an MSS option.
1633 * The option bytes which will be seen in normal data
1634 * packets should timestamps be used, must be in the MSS
1635 * advertised. But we subtract them from tp->mss_cache so
1636 * that calculations in tcp_sendmsg are simpler etc.
1637 * So account for this fact here if necessary. If we
1638 * don't do this correctly, as a receiver we won't
1639 * recognize data packets as being full sized when we
1640 * should, and thus we won't abide by the delayed ACK
1642 * SACKs don't matter, we never delay an ACK when we
1643 * have any of those going out.
1645 *ptr
++ = htonl((TCPOPT_MSS
<< 24) | (TCPOLEN_MSS
<< 16) | mss
);
1648 *ptr
++ = __constant_htonl((TCPOPT_SACK_PERM
<< 24) | (TCPOLEN_SACK_PERM
<< 16) |
1649 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
1651 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
1652 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
1653 *ptr
++ = htonl(tstamp
); /* TSVAL */
1654 *ptr
++ = htonl(ts_recent
); /* TSECR */
1656 *ptr
++ = __constant_htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
1657 (TCPOPT_SACK_PERM
<< 8) | TCPOLEN_SACK_PERM
);
1659 *ptr
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_WINDOW
<< 16) | (TCPOLEN_WINDOW
<< 8) | (wscale
));
1662 /* Determine a window scaling and initial window to offer. */
1663 extern void tcp_select_initial_window(int __space
, __u32 mss
,
1664 __u32
*rcv_wnd
, __u32
*window_clamp
,
1665 int wscale_ok
, __u8
*rcv_wscale
);
1667 static inline int tcp_win_from_space(int space
)
1669 return sysctl_tcp_adv_win_scale
<=0 ?
1670 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1671 space
- (space
>>sysctl_tcp_adv_win_scale
);
1674 /* Note: caller must be prepared to deal with negative returns */
1675 static inline int tcp_space(const struct sock
*sk
)
1677 return tcp_win_from_space(sk
->sk_rcvbuf
-
1678 atomic_read(&sk
->sk_rmem_alloc
));
1681 static inline int tcp_full_space(const struct sock
*sk
)
1683 return tcp_win_from_space(sk
->sk_rcvbuf
);
1686 static inline void tcp_acceptq_queue(struct sock
*sk
, struct open_request
*req
,
1689 struct tcp_sock
*tp
= tcp_sk(sk
);
1692 sk_acceptq_added(sk
);
1694 if (!tp
->accept_queue_tail
) {
1695 tp
->accept_queue
= req
;
1697 tp
->accept_queue_tail
->dl_next
= req
;
1699 tp
->accept_queue_tail
= req
;
1700 req
->dl_next
= NULL
;
1703 struct tcp_listen_opt
1705 u8 max_qlen_log
; /* log_2 of maximal queued SYNs */
1710 struct open_request
*syn_table
[TCP_SYNQ_HSIZE
];
1714 tcp_synq_removed(struct sock
*sk
, struct open_request
*req
)
1716 struct tcp_listen_opt
*lopt
= tcp_sk(sk
)->listen_opt
;
1718 if (--lopt
->qlen
== 0)
1719 tcp_delete_keepalive_timer(sk
);
1720 if (req
->retrans
== 0)
1724 static inline void tcp_synq_added(struct sock
*sk
)
1726 struct tcp_listen_opt
*lopt
= tcp_sk(sk
)->listen_opt
;
1728 if (lopt
->qlen
++ == 0)
1729 tcp_reset_keepalive_timer(sk
, TCP_TIMEOUT_INIT
);
1733 static inline int tcp_synq_len(struct sock
*sk
)
1735 return tcp_sk(sk
)->listen_opt
->qlen
;
1738 static inline int tcp_synq_young(struct sock
*sk
)
1740 return tcp_sk(sk
)->listen_opt
->qlen_young
;
1743 static inline int tcp_synq_is_full(struct sock
*sk
)
1745 return tcp_synq_len(sk
) >> tcp_sk(sk
)->listen_opt
->max_qlen_log
;
1748 static inline void tcp_synq_unlink(struct tcp_sock
*tp
, struct open_request
*req
,
1749 struct open_request
**prev
)
1751 write_lock(&tp
->syn_wait_lock
);
1752 *prev
= req
->dl_next
;
1753 write_unlock(&tp
->syn_wait_lock
);
1756 static inline void tcp_synq_drop(struct sock
*sk
, struct open_request
*req
,
1757 struct open_request
**prev
)
1759 tcp_synq_unlink(tcp_sk(sk
), req
, prev
);
1760 tcp_synq_removed(sk
, req
);
1761 tcp_openreq_free(req
);
1764 static __inline__
void tcp_openreq_init(struct open_request
*req
,
1765 struct tcp_options_received
*rx_opt
,
1766 struct sk_buff
*skb
)
1768 struct inet_request_sock
*ireq
= inet_rsk(req
);
1770 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1771 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1772 req
->mss
= rx_opt
->mss_clamp
;
1773 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1774 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1775 ireq
->sack_ok
= rx_opt
->sack_ok
;
1776 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1777 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1780 ireq
->rmt_port
= skb
->h
.th
->source
;
1783 extern void tcp_enter_memory_pressure(void);
1785 extern void tcp_listen_wlock(void);
1787 /* - We may sleep inside this lock.
1788 * - If sleeping is not required (or called from BH),
1789 * use plain read_(un)lock(&tcp_lhash_lock).
1792 static inline void tcp_listen_lock(void)
1794 /* read_lock synchronizes to candidates to writers */
1795 read_lock(&tcp_lhash_lock
);
1796 atomic_inc(&tcp_lhash_users
);
1797 read_unlock(&tcp_lhash_lock
);
1800 static inline void tcp_listen_unlock(void)
1802 if (atomic_dec_and_test(&tcp_lhash_users
))
1803 wake_up(&tcp_lhash_wait
);
1806 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1808 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1811 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1813 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1816 static inline int tcp_fin_time(const struct tcp_sock
*tp
)
1818 int fin_timeout
= tp
->linger2
? : sysctl_tcp_fin_timeout
;
1820 if (fin_timeout
< (tp
->rto
<<2) - (tp
->rto
>>1))
1821 fin_timeout
= (tp
->rto
<<2) - (tp
->rto
>>1);
1826 static inline int tcp_paws_check(const struct tcp_options_received
*rx_opt
, int rst
)
1828 if ((s32
)(rx_opt
->rcv_tsval
- rx_opt
->ts_recent
) >= 0)
1830 if (xtime
.tv_sec
>= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
)
1833 /* RST segments are not recommended to carry timestamp,
1834 and, if they do, it is recommended to ignore PAWS because
1835 "their cleanup function should take precedence over timestamps."
1836 Certainly, it is mistake. It is necessary to understand the reasons
1837 of this constraint to relax it: if peer reboots, clock may go
1838 out-of-sync and half-open connections will not be reset.
1839 Actually, the problem would be not existing if all
1840 the implementations followed draft about maintaining clock
1841 via reboots. Linux-2.2 DOES NOT!
1843 However, we can relax time bounds for RST segments to MSL.
1845 if (rst
&& xtime
.tv_sec
>= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1850 static inline void tcp_v4_setup_caps(struct sock
*sk
, struct dst_entry
*dst
)
1852 sk
->sk_route_caps
= dst
->dev
->features
;
1853 if (sk
->sk_route_caps
& NETIF_F_TSO
) {
1854 if (sock_flag(sk
, SOCK_NO_LARGESEND
) || dst
->header_len
)
1855 sk
->sk_route_caps
&= ~NETIF_F_TSO
;
1859 #define TCP_CHECK_TIMER(sk) do { } while (0)
1861 static inline int tcp_use_frto(const struct sock
*sk
)
1863 const struct tcp_sock
*tp
= tcp_sk(sk
);
1865 /* F-RTO must be activated in sysctl and there must be some
1866 * unsent new data, and the advertised window should allow
1869 return (sysctl_tcp_frto
&& sk
->sk_send_head
&&
1870 !after(TCP_SKB_CB(sk
->sk_send_head
)->end_seq
,
1871 tp
->snd_una
+ tp
->snd_wnd
));
1874 static inline void tcp_mib_init(void)
1877 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM
, 1);
1878 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1879 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1880 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN
, -1);
1884 enum tcp_seq_states
{
1885 TCP_SEQ_STATE_LISTENING
,
1886 TCP_SEQ_STATE_OPENREQ
,
1887 TCP_SEQ_STATE_ESTABLISHED
,
1888 TCP_SEQ_STATE_TIME_WAIT
,
1891 struct tcp_seq_afinfo
{
1892 struct module
*owner
;
1895 int (*seq_show
) (struct seq_file
*m
, void *v
);
1896 struct file_operations
*seq_fops
;
1899 struct tcp_iter_state
{
1901 enum tcp_seq_states state
;
1902 struct sock
*syn_wait_sk
;
1903 int bucket
, sbucket
, num
, uid
;
1904 struct seq_operations seq_ops
;
1907 extern int tcp_proc_register(struct tcp_seq_afinfo
*afinfo
);
1908 extern void tcp_proc_unregister(struct tcp_seq_afinfo
*afinfo
);
1910 /* TCP Westwood functions and constants */
1912 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
1913 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
1915 static inline void tcp_westwood_update_rtt(struct tcp_sock
*tp
, __u32 rtt_seq
)
1917 if (tcp_is_westwood(tp
))
1918 tp
->westwood
.rtt
= rtt_seq
;
1921 static inline __u32
__tcp_westwood_bw_rttmin(const struct tcp_sock
*tp
)
1923 return max((tp
->westwood
.bw_est
) * (tp
->westwood
.rtt_min
) /
1924 (__u32
) (tp
->mss_cache_std
),
1928 static inline __u32
tcp_westwood_bw_rttmin(const struct tcp_sock
*tp
)
1930 return tcp_is_westwood(tp
) ? __tcp_westwood_bw_rttmin(tp
) : 0;
1933 static inline int tcp_westwood_ssthresh(struct tcp_sock
*tp
)
1937 if (tcp_is_westwood(tp
)) {
1938 ssthresh
= __tcp_westwood_bw_rttmin(tp
);
1940 tp
->snd_ssthresh
= ssthresh
;
1943 return (ssthresh
!= 0);
1946 static inline int tcp_westwood_cwnd(struct tcp_sock
*tp
)
1950 if (tcp_is_westwood(tp
)) {
1951 cwnd
= __tcp_westwood_bw_rttmin(tp
);
1953 tp
->snd_cwnd
= cwnd
;