blob: a166918ca56d2f36f80f21d7678dd92b86c88454 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
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.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
21#define TCP_DEBUG 1
22#define FASTRETRANS_DEBUG 1
23
24/* Cancel timers, when they are not required. */
25#undef TCP_CLEAR_TIMERS
26
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>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070034#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <net/sock.h>
36#include <net/snmp.h>
37#include <net/ip.h>
38#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
39#include <linux/ipv6.h>
40#endif
41#include <linux/seq_file.h>
42
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.
46 */
47struct tcp_ehash_bucket {
48 rwlock_t lock;
49 struct hlist_head chain;
50} __attribute__((__aligned__(8)));
51
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. */
54
55/* There are a few simple rules, which allow for local port reuse by
56 * an application. In essence:
57 *
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
65 * shared.
66 * Failing this, the port cannot be shared.
67 *
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.
78 *
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
85 */
86struct tcp_bind_bucket {
87 unsigned short port;
88 signed short fastreuse;
89 struct hlist_node node;
90 struct hlist_head owners;
91};
92
93#define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node)
94
95struct tcp_bind_hashbucket {
96 spinlock_t lock;
97 struct hlist_head chain;
98};
99
100static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head)
101{
102 return hlist_entry(head->chain.first, struct tcp_bind_bucket, node);
103}
104
105static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head)
106{
107 return hlist_empty(&head->chain) ? NULL : __tb_head(head);
108}
109
110extern 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:
113 *
114 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
115 *
116 * First half of the table is for sockets not in TIME_WAIT, second half
117 * is for TIME_WAIT sockets only.
118 */
119 struct tcp_ehash_bucket *__tcp_ehash;
120
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.
123 */
124 struct tcp_bind_hashbucket *__tcp_bhash;
125
126 int __tcp_bhash_size;
127 int __tcp_ehash_size;
128
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.
132 */
133 struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE];
134
135 /* All the above members are written once at bootup and
136 * never written again _or_ are predominantly read-access.
137 *
138 * Now align to a new cache line as all the following members
139 * are often dirty.
140 */
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;
145} tcp_hashinfo;
146
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)
156
157extern kmem_cache_t *tcp_bucket_cachep;
158extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head,
159 unsigned short snum);
160extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb);
161extern void tcp_bucket_unlock(struct sock *sk);
162extern int tcp_port_rover;
163
164/* These are AF independent. */
165static __inline__ int tcp_bhashfn(__u16 lport)
166{
167 return (lport & (tcp_bhash_size - 1));
168}
169
170extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb,
171 unsigned short snum);
172
173#if (BITS_PER_LONG == 64)
174#define TCP_ADDRCMP_ALIGN_BYTES 8
175#else
176#define TCP_ADDRCMP_ALIGN_BYTES 4
177#endif
178
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.
182 */
183struct tcp_tw_bucket {
184 /*
185 * Now struct sock also uses sock_common, so please just
186 * don't add nothing before this first member (__tw_common) --acme
187 */
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;
198 __u16 tw_sport;
199 /* Socket demultiplex comparisons on incoming packets. */
200 /* these five are in inet_sock */
201 __u32 tw_daddr
202 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
203 __u32 tw_rcv_saddr;
204 __u16 tw_dport;
205 __u16 tw_num;
206 /* And these are ours. */
207 int tw_hashent;
208 int tw_timeout;
209 __u32 tw_rcv_nxt;
210 __u32 tw_snd_nxt;
211 __u32 tw_rcv_wnd;
212 __u32 tw_ts_recent;
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;
220 int tw_v6_ipv6only;
221#endif
222};
223
224static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
225 struct hlist_head *list)
226{
227 hlist_add_head(&tw->tw_node, list);
228}
229
230static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
231 struct hlist_head *list)
232{
233 hlist_add_head(&tw->tw_bind_node, list);
234}
235
236static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
237{
238 return tw->tw_death_node.pprev != NULL;
239}
240
241static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
242{
243 tw->tw_death_node.pprev = NULL;
244}
245
246static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
247{
248 __hlist_del(&tw->tw_death_node);
249 tw_dead_node_init(tw);
250}
251
252static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
253{
254 if (tw_dead_hashed(tw)) {
255 __tw_del_dead_node(tw);
256 return 1;
257 }
258 return 0;
259}
260
261#define tw_for_each(tw, node, head) \
262 hlist_for_each_entry(tw, node, head, tw_node)
263
264#define tw_for_each_inmate(tw, node, jail) \
265 hlist_for_each_entry(tw, node, jail, tw_death_node)
266
267#define tw_for_each_inmate_safe(tw, node, safe, jail) \
268 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
269
270#define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
271
272static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
273{
274 return likely(sk->sk_state != TCP_TIME_WAIT) ?
275 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
276}
277
278#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
279static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
280{
281 return likely(sk->sk_state != TCP_TIME_WAIT) ?
282 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
283}
284
285static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
286{
287 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
288}
289
290#define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
291
292static inline int tcp_v6_ipv6only(const struct sock *sk)
293{
294 return likely(sk->sk_state != TCP_TIME_WAIT) ?
295 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
296}
297#else
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
302#endif
303
304extern kmem_cache_t *tcp_timewait_cachep;
305
306static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
307{
308 if (atomic_dec_and_test(&tw->tw_refcnt)) {
309#ifdef INET_REFCNT_DEBUG
310 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
311#endif
312 kmem_cache_free(tcp_timewait_cachep, tw);
313 }
314}
315
316extern atomic_t tcp_orphan_count;
317extern int tcp_tw_count;
318extern void tcp_time_wait(struct sock *sk, int state, int timeo);
319extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
320
321
322/* Socket demux engine toys. */
323#ifdef __BIG_ENDIAN
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))
329#endif
330
331#if (BITS_PER_LONG == 64)
332#ifdef __BIG_ENDIAN
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 */
360
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))))
367
368/* These can have wildcards, don't try too hard. */
369static __inline__ int tcp_lhashfn(unsigned short num)
370{
371 return num & (TCP_LHTABLE_SIZE - 1);
372}
373
374static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
375{
376 return tcp_lhashfn(inet_sk(sk)->num);
377}
378
379#define MAX_TCP_HEADER (128 + MAX_HEADER)
380
381/*
382 * Never offer a window over 32767 without using window scaling. Some
383 * poor stacks do signed 16bit maths!
384 */
385#define MAX_TCP_WINDOW 32767U
386
387/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
388#define TCP_MIN_MSS 88U
389
390/* Minimal RCV_MSS. */
391#define TCP_MIN_RCVMSS 536U
392
393/* After receiving this amount of duplicate ACKs fast retransmit starts. */
394#define TCP_FASTRETRANS_THRESH 3
395
396/* Maximal reordering. */
397#define TCP_MAX_REORDERING 127
398
399/* Maximal number of ACKs sent quickly to accelerate slow-start. */
400#define TCP_MAX_QUICKACKS 16U
401
402/* urg_data states */
403#define TCP_URG_VALID 0x0100
404#define TCP_URG_NOTYET 0x0200
405#define TCP_URG_READ 0x0400
406
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.
412 */
413
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.
419 */
420
421#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
422 * connection: ~180sec is RFC minumum */
423
424#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
425 * connection: ~180sec is RFC minumum */
426
427
428#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
429 * socket. 7 is ~50sec-16min.
430 */
431
432
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
439 * TIME-WAIT timer.
440 */
441
442#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
443#if HZ >= 100
444#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
445#define TCP_ATO_MIN ((unsigned)(HZ/25))
446#else
447#define TCP_DELACK_MIN 4U
448#define TCP_ATO_MIN 4U
449#endif
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 */
453
454#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
455 * for local resources.
456 */
457
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)
461
462#define MAX_TCP_KEEPIDLE 32767
463#define MAX_TCP_KEEPINTVL 32767
464#define MAX_TCP_KEEPCNT 127
465#define MAX_TCP_SYNCNT 127
466
467#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
468#define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
469
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.
476 */
477#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
478 * timestamps. It must be less than
479 * minimal timewait lifetime.
480 */
481
482#define TCP_TW_RECYCLE_SLOTS_LOG 5
483#define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
484
485/* If time > 4sec, it is "slow" path, no recycling is required,
486 so that we select tick to get range about 4 seconds.
487 */
488
489#if HZ <= 16 || HZ > 4096
490# error Unsupported: HZ <= 16 or HZ > 4096
491#elif HZ <= 32
492# define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
493#elif HZ <= 64
494# define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
495#elif HZ <= 128
496# define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
497#elif HZ <= 256
498# define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
499#elif HZ <= 512
500# define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
501#elif HZ <= 1024
502# define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
503#elif HZ <= 2048
504# define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
505#else
506# define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
507#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508/*
509 * TCP option
510 */
511
512#define TCPOPT_NOP 1 /* Padding */
513#define TCPOPT_EOL 0 /* End of options */
514#define TCPOPT_MSS 2 /* Segment size negotiating */
515#define TCPOPT_WINDOW 3 /* Window scaling */
516#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
517#define TCPOPT_SACK 5 /* SACK Block */
518#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
519
520/*
521 * TCP option lengths
522 */
523
524#define TCPOLEN_MSS 4
525#define TCPOLEN_WINDOW 3
526#define TCPOLEN_SACK_PERM 2
527#define TCPOLEN_TIMESTAMP 10
528
529/* But this is what stacks really send out. */
530#define TCPOLEN_TSTAMP_ALIGNED 12
531#define TCPOLEN_WSCALE_ALIGNED 4
532#define TCPOLEN_SACKPERM_ALIGNED 4
533#define TCPOLEN_SACK_BASE 2
534#define TCPOLEN_SACK_BASE_ALIGNED 4
535#define TCPOLEN_SACK_PERBLOCK 8
536
537#define TCP_TIME_RETRANS 1 /* Retransmit timer */
538#define TCP_TIME_DACK 2 /* Delayed ack timer */
539#define TCP_TIME_PROBE0 3 /* Zero window probe timer */
540#define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
541
542/* Flags in tp->nonagle */
543#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
544#define TCP_NAGLE_CORK 2 /* Socket is corked */
545#define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
546
547/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548extern int sysctl_tcp_timestamps;
549extern int sysctl_tcp_window_scaling;
550extern int sysctl_tcp_sack;
551extern int sysctl_tcp_fin_timeout;
552extern int sysctl_tcp_tw_recycle;
553extern int sysctl_tcp_keepalive_time;
554extern int sysctl_tcp_keepalive_probes;
555extern int sysctl_tcp_keepalive_intvl;
556extern int sysctl_tcp_syn_retries;
557extern int sysctl_tcp_synack_retries;
558extern int sysctl_tcp_retries1;
559extern int sysctl_tcp_retries2;
560extern int sysctl_tcp_orphan_retries;
561extern int sysctl_tcp_syncookies;
562extern int sysctl_tcp_retrans_collapse;
563extern int sysctl_tcp_stdurg;
564extern int sysctl_tcp_rfc1337;
565extern int sysctl_tcp_abort_on_overflow;
566extern int sysctl_tcp_max_orphans;
567extern int sysctl_tcp_max_tw_buckets;
568extern int sysctl_tcp_fack;
569extern int sysctl_tcp_reordering;
570extern int sysctl_tcp_ecn;
571extern int sysctl_tcp_dsack;
572extern int sysctl_tcp_mem[3];
573extern int sysctl_tcp_wmem[3];
574extern int sysctl_tcp_rmem[3];
575extern int sysctl_tcp_app_win;
576extern int sysctl_tcp_adv_win_scale;
577extern int sysctl_tcp_tw_reuse;
578extern int sysctl_tcp_frto;
579extern int sysctl_tcp_low_latency;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581extern int sysctl_tcp_moderate_rcvbuf;
582extern int sysctl_tcp_tso_win_divisor;
583
584extern atomic_t tcp_memory_allocated;
585extern atomic_t tcp_sockets_allocated;
586extern int tcp_memory_pressure;
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
589#define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
590#else
591#define TCP_INET_FAMILY(fam) 1
592#endif
593
594/*
595 * Pointers to address related TCP functions
596 * (i.e. things that depend on the address family)
597 */
598
599struct tcp_func {
600 int (*queue_xmit) (struct sk_buff *skb,
601 int ipfragok);
602
603 void (*send_check) (struct sock *sk,
604 struct tcphdr *th,
605 int len,
606 struct sk_buff *skb);
607
608 int (*rebuild_header) (struct sock *sk);
609
610 int (*conn_request) (struct sock *sk,
611 struct sk_buff *skb);
612
613 struct sock * (*syn_recv_sock) (struct sock *sk,
614 struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700615 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 struct dst_entry *dst);
617
618 int (*remember_stamp) (struct sock *sk);
619
620 __u16 net_header_len;
621
622 int (*setsockopt) (struct sock *sk,
623 int level,
624 int optname,
625 char __user *optval,
626 int optlen);
627
628 int (*getsockopt) (struct sock *sk,
629 int level,
630 int optname,
631 char __user *optval,
632 int __user *optlen);
633
634
635 void (*addr2sockaddr) (struct sock *sk,
636 struct sockaddr *);
637
638 int sockaddr_len;
639};
640
641/*
642 * The next routines deal with comparing 32 bit unsigned ints
643 * and worry about wraparound (automatic with unsigned arithmetic).
644 */
645
646static inline int before(__u32 seq1, __u32 seq2)
647{
648 return (__s32)(seq1-seq2) < 0;
649}
650
651static inline int after(__u32 seq1, __u32 seq2)
652{
653 return (__s32)(seq2-seq1) < 0;
654}
655
656
657/* is s2<=s1<=s3 ? */
658static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
659{
660 return seq3 - seq2 >= seq1 - seq2;
661}
662
663
664extern struct proto tcp_prot;
665
666DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
667#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
668#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
669#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
670#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
671#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
672#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
673
674extern void tcp_put_port(struct sock *sk);
675extern void tcp_inherit_port(struct sock *sk, struct sock *child);
676
677extern void tcp_v4_err(struct sk_buff *skb, u32);
678
679extern void tcp_shutdown (struct sock *sk, int how);
680
681extern int tcp_v4_rcv(struct sk_buff *skb);
682
683extern int tcp_v4_remember_stamp(struct sock *sk);
684
685extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
686
687extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
688 struct msghdr *msg, size_t size);
689extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
690
691extern int tcp_ioctl(struct sock *sk,
692 int cmd,
693 unsigned long arg);
694
695extern int tcp_rcv_state_process(struct sock *sk,
696 struct sk_buff *skb,
697 struct tcphdr *th,
698 unsigned len);
699
700extern int tcp_rcv_established(struct sock *sk,
701 struct sk_buff *skb,
702 struct tcphdr *th,
703 unsigned len);
704
705extern void tcp_rcv_space_adjust(struct sock *sk);
706
707enum tcp_ack_state_t
708{
709 TCP_ACK_SCHED = 1,
710 TCP_ACK_TIMER = 2,
711 TCP_ACK_PUSHED= 4
712};
713
714static inline void tcp_schedule_ack(struct tcp_sock *tp)
715{
716 tp->ack.pending |= TCP_ACK_SCHED;
717}
718
719static inline int tcp_ack_scheduled(struct tcp_sock *tp)
720{
721 return tp->ack.pending&TCP_ACK_SCHED;
722}
723
David S. Millerfc6415bc2005-07-05 15:17:45 -0700724static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp, unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
David S. Millerfc6415bc2005-07-05 15:17:45 -0700726 if (tp->ack.quick) {
727 if (pkts >= tp->ack.quick) {
728 tp->ack.quick = 0;
729
730 /* Leaving quickack mode we deflate ATO. */
731 tp->ack.ato = TCP_ATO_MIN;
732 } else
733 tp->ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
735}
736
737extern void tcp_enter_quickack_mode(struct tcp_sock *tp);
738
739static __inline__ void tcp_delack_init(struct tcp_sock *tp)
740{
741 memset(&tp->ack, 0, sizeof(tp->ack));
742}
743
744static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
745{
746 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
747}
748
749enum tcp_tw_status
750{
751 TCP_TW_SUCCESS = 0,
752 TCP_TW_RST = 1,
753 TCP_TW_ACK = 2,
754 TCP_TW_SYN = 3
755};
756
757
758extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
759 struct sk_buff *skb,
760 struct tcphdr *th,
761 unsigned len);
762
763extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700764 struct request_sock *req,
765 struct request_sock **prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766extern int tcp_child_process(struct sock *parent,
767 struct sock *child,
768 struct sk_buff *skb);
769extern void tcp_enter_frto(struct sock *sk);
770extern void tcp_enter_loss(struct sock *sk, int how);
771extern void tcp_clear_retrans(struct tcp_sock *tp);
772extern void tcp_update_metrics(struct sock *sk);
773
774extern void tcp_close(struct sock *sk,
775 long timeout);
776extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
777extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
778
779extern int tcp_getsockopt(struct sock *sk, int level,
780 int optname,
781 char __user *optval,
782 int __user *optlen);
783extern int tcp_setsockopt(struct sock *sk, int level,
784 int optname, char __user *optval,
785 int optlen);
786extern void tcp_set_keepalive(struct sock *sk, int val);
787extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
788 struct msghdr *msg,
789 size_t len, int nonblock,
790 int flags, int *addr_len);
791
792extern int tcp_listen_start(struct sock *sk);
793
794extern void tcp_parse_options(struct sk_buff *skb,
795 struct tcp_options_received *opt_rx,
796 int estab);
797
798/*
799 * TCP v4 functions exported for the inet6 API
800 */
801
802extern int tcp_v4_rebuild_header(struct sock *sk);
803
804extern int tcp_v4_build_header(struct sock *sk,
805 struct sk_buff *skb);
806
807extern void tcp_v4_send_check(struct sock *sk,
808 struct tcphdr *th, int len,
809 struct sk_buff *skb);
810
811extern int tcp_v4_conn_request(struct sock *sk,
812 struct sk_buff *skb);
813
814extern struct sock * tcp_create_openreq_child(struct sock *sk,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700815 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 struct sk_buff *skb);
817
818extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
819 struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700820 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 struct dst_entry *dst);
822
823extern int tcp_v4_do_rcv(struct sock *sk,
824 struct sk_buff *skb);
825
826extern int tcp_v4_connect(struct sock *sk,
827 struct sockaddr *uaddr,
828 int addr_len);
829
830extern int tcp_connect(struct sock *sk);
831
832extern struct sk_buff * tcp_make_synack(struct sock *sk,
833 struct dst_entry *dst,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700834 struct request_sock *req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836extern int tcp_disconnect(struct sock *sk, int flags);
837
838extern void tcp_unhash(struct sock *sk);
839
840extern int tcp_v4_hash_connecting(struct sock *sk);
841
842
843/* From syncookies.c */
844extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
845 struct ip_options *opt);
846extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
847 __u16 *mss);
848
849/* tcp_output.c */
850
David S. Millerf6302d12005-07-05 15:18:03 -0700851extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
David S. Millera2e2a592005-07-05 15:19:23 -0700852 unsigned int cur_mss, int nonagle);
David S. Millerf6302d12005-07-05 15:18:03 -0700853extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
855extern void tcp_xmit_retransmit_queue(struct sock *);
856extern void tcp_simple_retransmit(struct sock *);
857extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
858
859extern void tcp_send_probe0(struct sock *);
860extern void tcp_send_partial(struct sock *);
861extern int tcp_write_wakeup(struct sock *);
862extern void tcp_send_fin(struct sock *sk);
863extern void tcp_send_active_reset(struct sock *sk, int priority);
864extern int tcp_send_synack(struct sock *);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700865extern void tcp_push_one(struct sock *, unsigned int mss_now);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866extern void tcp_send_ack(struct sock *sk);
867extern void tcp_send_delayed_ack(struct sock *sk);
868
David S. Millera762a982005-07-05 15:18:51 -0700869/* tcp_input.c */
870extern void tcp_cwnd_application_limited(struct sock *sk);
871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872/* tcp_timer.c */
873extern void tcp_init_xmit_timers(struct sock *);
874extern void tcp_clear_xmit_timers(struct sock *);
875
876extern void tcp_delete_keepalive_timer(struct sock *);
877extern void tcp_reset_keepalive_timer(struct sock *, unsigned long);
878extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
879extern unsigned int tcp_current_mss(struct sock *sk, int large);
880
881#ifdef TCP_DEBUG
882extern const char tcp_timer_bug_msg[];
883#endif
884
885/* tcp_diag.c */
886extern void tcp_get_info(struct sock *, struct tcp_info *);
887
888/* Read 'sendfile()'-style from a TCP socket */
889typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
890 unsigned int, size_t);
891extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
892 sk_read_actor_t recv_actor);
893
894static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
895{
896 struct tcp_sock *tp = tcp_sk(sk);
897
898 switch (what) {
899 case TCP_TIME_RETRANS:
900 case TCP_TIME_PROBE0:
901 tp->pending = 0;
902
903#ifdef TCP_CLEAR_TIMERS
904 sk_stop_timer(sk, &tp->retransmit_timer);
905#endif
906 break;
907 case TCP_TIME_DACK:
908 tp->ack.blocked = 0;
909 tp->ack.pending = 0;
910
911#ifdef TCP_CLEAR_TIMERS
912 sk_stop_timer(sk, &tp->delack_timer);
913#endif
914 break;
915 default:
916#ifdef TCP_DEBUG
917 printk(tcp_timer_bug_msg);
918#endif
919 return;
920 };
921
922}
923
924/*
925 * Reset the retransmission timer
926 */
927static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
928{
929 struct tcp_sock *tp = tcp_sk(sk);
930
931 if (when > TCP_RTO_MAX) {
932#ifdef TCP_DEBUG
933 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
934#endif
935 when = TCP_RTO_MAX;
936 }
937
938 switch (what) {
939 case TCP_TIME_RETRANS:
940 case TCP_TIME_PROBE0:
941 tp->pending = what;
942 tp->timeout = jiffies+when;
943 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
944 break;
945
946 case TCP_TIME_DACK:
947 tp->ack.pending |= TCP_ACK_TIMER;
948 tp->ack.timeout = jiffies+when;
949 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
950 break;
951
952 default:
953#ifdef TCP_DEBUG
954 printk(tcp_timer_bug_msg);
955#endif
956 return;
957 };
958}
959
960/* Initialize RCV_MSS value.
961 * RCV_MSS is an our guess about MSS used by the peer.
962 * We haven't any direct information about the MSS.
963 * It's better to underestimate the RCV_MSS rather than overestimate.
964 * Overestimations make us ACKing less frequently than needed.
965 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
966 */
967
968static inline void tcp_initialize_rcv_mss(struct sock *sk)
969{
970 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700971 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 hint = min(hint, tp->rcv_wnd/2);
974 hint = min(hint, TCP_MIN_RCVMSS);
975 hint = max(hint, TCP_MIN_MSS);
976
977 tp->ack.rcv_mss = hint;
978}
979
980static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
981{
982 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
983 ntohl(TCP_FLAG_ACK) |
984 snd_wnd);
985}
986
987static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
988{
989 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
990}
991
992static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
993{
994 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
995 tp->rcv_wnd &&
996 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
997 !tp->urg_data)
998 tcp_fast_path_on(tp);
999}
1000
1001/* Compute the actual receive window we are currently advertising.
1002 * Rcv_nxt can be after the window if our peer push more data
1003 * than the offered window.
1004 */
1005static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
1006{
1007 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1008
1009 if (win < 0)
1010 win = 0;
1011 return (u32) win;
1012}
1013
1014/* Choose a new window, without checks for shrinking, and without
1015 * scaling applied to the result. The caller does these things
1016 * if necessary. This is a "raw" window selection.
1017 */
1018extern u32 __tcp_select_window(struct sock *sk);
1019
1020/* TCP timestamps are only 32-bits, this causes a slight
1021 * complication on 64-bit systems since we store a snapshot
1022 * of jiffies in the buffer control blocks below. We decidely
1023 * only use of the low 32-bits of jiffies and hide the ugly
1024 * casts with the following macro.
1025 */
1026#define tcp_time_stamp ((__u32)(jiffies))
1027
1028/* This is what the send packet queueing engine uses to pass
1029 * TCP per-packet control information to the transmission
1030 * code. We also store the host-order sequence numbers in
1031 * here too. This is 36 bytes on 32-bit architectures,
1032 * 40 bytes on 64-bit machines, if this grows please adjust
1033 * skbuff.h:skbuff->cb[xxx] size appropriately.
1034 */
1035struct tcp_skb_cb {
1036 union {
1037 struct inet_skb_parm h4;
1038#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1039 struct inet6_skb_parm h6;
1040#endif
1041 } header; /* For incoming frames */
1042 __u32 seq; /* Starting sequence number */
1043 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1044 __u32 when; /* used to compute rtt's */
1045 __u8 flags; /* TCP header flags. */
1046
1047 /* NOTE: These must match up to the flags byte in a
1048 * real TCP header.
1049 */
1050#define TCPCB_FLAG_FIN 0x01
1051#define TCPCB_FLAG_SYN 0x02
1052#define TCPCB_FLAG_RST 0x04
1053#define TCPCB_FLAG_PSH 0x08
1054#define TCPCB_FLAG_ACK 0x10
1055#define TCPCB_FLAG_URG 0x20
1056#define TCPCB_FLAG_ECE 0x40
1057#define TCPCB_FLAG_CWR 0x80
1058
1059 __u8 sacked; /* State flags for SACK/FACK. */
1060#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1061#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1062#define TCPCB_LOST 0x04 /* SKB is lost */
1063#define TCPCB_TAGBITS 0x07 /* All tag bits */
1064
1065#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1066#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1067
1068#define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1069
1070#define TCPCB_AT_TAIL (TCPCB_URG)
1071
1072 __u16 urg_ptr; /* Valid w/URG flags is set. */
1073 __u32 ack_seq; /* Sequence number ACK'd */
1074};
1075
1076#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1077
1078#include <net/tcp_ecn.h>
1079
1080/* Due to TSO, an SKB can be composed of multiple actual
1081 * packets. To keep these tracked properly, we use this.
1082 */
1083static inline int tcp_skb_pcount(const struct sk_buff *skb)
1084{
1085 return skb_shinfo(skb)->tso_segs;
1086}
1087
1088/* This is valid iff tcp_skb_pcount() > 1. */
1089static inline int tcp_skb_mss(const struct sk_buff *skb)
1090{
1091 return skb_shinfo(skb)->tso_size;
1092}
1093
1094static inline void tcp_dec_pcount_approx(__u32 *count,
1095 const struct sk_buff *skb)
1096{
1097 if (*count) {
1098 *count -= tcp_skb_pcount(skb);
1099 if ((int)*count < 0)
1100 *count = 0;
1101 }
1102}
1103
1104static inline void tcp_packets_out_inc(struct sock *sk,
1105 struct tcp_sock *tp,
1106 const struct sk_buff *skb)
1107{
1108 int orig = tp->packets_out;
1109
1110 tp->packets_out += tcp_skb_pcount(skb);
1111 if (!orig)
1112 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
1113}
1114
1115static inline void tcp_packets_out_dec(struct tcp_sock *tp,
1116 const struct sk_buff *skb)
1117{
1118 tp->packets_out -= tcp_skb_pcount(skb);
1119}
1120
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001121/* Events passed to congestion control interface */
1122enum tcp_ca_event {
1123 CA_EVENT_TX_START, /* first transmit when no packets in flight */
1124 CA_EVENT_CWND_RESTART, /* congestion window restart */
1125 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
1126 CA_EVENT_FRTO, /* fast recovery timeout */
1127 CA_EVENT_LOSS, /* loss timeout */
1128 CA_EVENT_FAST_ACK, /* in sequence ack */
1129 CA_EVENT_SLOW_ACK, /* other ack */
1130};
1131
1132/*
1133 * Interface for adding new TCP congestion control handlers
1134 */
1135#define TCP_CA_NAME_MAX 16
1136struct tcp_congestion_ops {
1137 struct list_head list;
1138
1139 /* initialize private data (optional) */
1140 void (*init)(struct tcp_sock *tp);
1141 /* cleanup private data (optional) */
1142 void (*release)(struct tcp_sock *tp);
1143
1144 /* return slow start threshold (required) */
1145 u32 (*ssthresh)(struct tcp_sock *tp);
1146 /* lower bound for congestion window (optional) */
1147 u32 (*min_cwnd)(struct tcp_sock *tp);
1148 /* do new cwnd calculation (required) */
1149 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
1150 u32 rtt, u32 in_flight, int good_ack);
1151 /* round trip time sample per acked packet (optional) */
1152 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
1153 /* call before changing ca_state (optional) */
1154 void (*set_state)(struct tcp_sock *tp, u8 new_state);
1155 /* call when cwnd event occurs (optional) */
1156 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
1157 /* new value of cwnd after loss (optional) */
1158 u32 (*undo_cwnd)(struct tcp_sock *tp);
1159 /* hook for packet ack accounting (optional) */
1160 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
1161 /* get info for tcp_diag (optional) */
1162 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
1163
1164 char name[TCP_CA_NAME_MAX];
1165 struct module *owner;
1166};
1167
1168extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1169extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1170
1171extern void tcp_init_congestion_control(struct tcp_sock *tp);
1172extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
1173extern int tcp_set_default_congestion_control(const char *name);
1174extern void tcp_get_default_congestion_control(char *name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07001175extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001176
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07001177extern struct tcp_congestion_ops tcp_init_congestion_ops;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001178extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
1179extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
1180 u32 rtt, u32 in_flight, int flag);
1181extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
David S. Millera8acfba2005-06-23 23:45:02 -07001182extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001183
1184static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
1185{
1186 if (tp->ca_ops->set_state)
1187 tp->ca_ops->set_state(tp, ca_state);
1188 tp->ca_state = ca_state;
1189}
1190
1191static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
1192{
1193 if (tp->ca_ops->cwnd_event)
1194 tp->ca_ops->cwnd_event(tp, event);
1195}
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197/* This determines how many packets are "in the network" to the best
1198 * of our knowledge. In many cases it is conservative, but where
1199 * detailed information is available from the receiver (via SACK
1200 * blocks etc.) we can make more aggressive calculations.
1201 *
1202 * Use this for decisions involving congestion control, use just
1203 * tp->packets_out to determine if the send queue is empty or not.
1204 *
1205 * Read this equation as:
1206 *
1207 * "Packets sent once on transmission queue" MINUS
1208 * "Packets left network, but not honestly ACKed yet" PLUS
1209 * "Packets fast retransmitted"
1210 */
1211static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1212{
1213 return (tp->packets_out - tp->left_out + tp->retrans_out);
1214}
1215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1217 * The exception is rate halving phase, when cwnd is decreasing towards
1218 * ssthresh.
1219 */
1220static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
1221{
1222 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1223 return tp->snd_ssthresh;
1224 else
1225 return max(tp->snd_ssthresh,
1226 ((tp->snd_cwnd >> 1) +
1227 (tp->snd_cwnd >> 2)));
1228}
1229
1230static inline void tcp_sync_left_out(struct tcp_sock *tp)
1231{
1232 if (tp->rx_opt.sack_ok &&
1233 (tp->sacked_out >= tp->packets_out - tp->lost_out))
1234 tp->sacked_out = tp->packets_out - tp->lost_out;
1235 tp->left_out = tp->sacked_out + tp->lost_out;
1236}
1237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238/* Set slow start threshould and cwnd not falling to slow start */
1239static inline void __tcp_enter_cwr(struct tcp_sock *tp)
1240{
1241 tp->undo_marker = 0;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001242 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 tp->snd_cwnd = min(tp->snd_cwnd,
1244 tcp_packets_in_flight(tp) + 1U);
1245 tp->snd_cwnd_cnt = 0;
1246 tp->high_seq = tp->snd_nxt;
1247 tp->snd_cwnd_stamp = tcp_time_stamp;
1248 TCP_ECN_queue_cwr(tp);
1249}
1250
1251static inline void tcp_enter_cwr(struct tcp_sock *tp)
1252{
1253 tp->prior_ssthresh = 0;
1254 if (tp->ca_state < TCP_CA_CWR) {
1255 __tcp_enter_cwr(tp);
1256 tcp_set_ca_state(tp, TCP_CA_CWR);
1257 }
1258}
1259
1260extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
1261
1262/* Slow start with delack produces 3 packets of burst, so that
1263 * it is safe "de facto".
1264 */
1265static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1266{
1267 return 3;
1268}
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
1271 const struct sk_buff *skb)
1272{
1273 if (skb->len < mss)
1274 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1275}
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
1278{
1279 if (!tp->packets_out && !tp->pending)
1280 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1281}
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283static __inline__ void tcp_push_pending_frames(struct sock *sk,
1284 struct tcp_sock *tp)
1285{
1286 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1287}
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1290{
1291 tp->snd_wl1 = seq;
1292}
1293
1294static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1295{
1296 tp->snd_wl1 = seq;
1297}
1298
1299extern void tcp_destroy_sock(struct sock *sk);
1300
1301
1302/*
1303 * Calculate(/check) TCP checksum
1304 */
1305static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1306 unsigned long saddr, unsigned long daddr,
1307 unsigned long base)
1308{
1309 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1310}
1311
1312static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1313{
1314 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1315}
1316
1317static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1318{
1319 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1320 __tcp_checksum_complete(skb);
1321}
1322
1323/* Prequeue for VJ style copy to user, combined with checksumming. */
1324
1325static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
1326{
1327 tp->ucopy.task = NULL;
1328 tp->ucopy.len = 0;
1329 tp->ucopy.memory = 0;
1330 skb_queue_head_init(&tp->ucopy.prequeue);
1331}
1332
1333/* Packet is added to VJ-style prequeue for processing in process
1334 * context, if a reader task is waiting. Apparently, this exciting
1335 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1336 * failed somewhere. Latency? Burstiness? Well, at least now we will
1337 * see, why it failed. 8)8) --ANK
1338 *
1339 * NOTE: is this not too big to inline?
1340 */
1341static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1342{
1343 struct tcp_sock *tp = tcp_sk(sk);
1344
1345 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1346 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1347 tp->ucopy.memory += skb->truesize;
1348 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1349 struct sk_buff *skb1;
1350
1351 BUG_ON(sock_owned_by_user(sk));
1352
1353 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1354 sk->sk_backlog_rcv(sk, skb1);
1355 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1356 }
1357
1358 tp->ucopy.memory = 0;
1359 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1360 wake_up_interruptible(sk->sk_sleep);
1361 if (!tcp_ack_scheduled(tp))
1362 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1363 }
1364 return 1;
1365 }
1366 return 0;
1367}
1368
1369
1370#undef STATE_TRACE
1371
1372#ifdef STATE_TRACE
1373static const char *statename[]={
1374 "Unused","Established","Syn Sent","Syn Recv",
1375 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1376 "Close Wait","Last ACK","Listen","Closing"
1377};
1378#endif
1379
1380static __inline__ void tcp_set_state(struct sock *sk, int state)
1381{
1382 int oldstate = sk->sk_state;
1383
1384 switch (state) {
1385 case TCP_ESTABLISHED:
1386 if (oldstate != TCP_ESTABLISHED)
1387 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1388 break;
1389
1390 case TCP_CLOSE:
1391 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1392 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1393
1394 sk->sk_prot->unhash(sk);
1395 if (tcp_sk(sk)->bind_hash &&
1396 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1397 tcp_put_port(sk);
1398 /* fall through */
1399 default:
1400 if (oldstate==TCP_ESTABLISHED)
1401 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1402 }
1403
1404 /* Change state AFTER socket is unhashed to avoid closed
1405 * socket sitting in hash tables.
1406 */
1407 sk->sk_state = state;
1408
1409#ifdef STATE_TRACE
1410 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1411#endif
1412}
1413
1414static __inline__ void tcp_done(struct sock *sk)
1415{
1416 tcp_set_state(sk, TCP_CLOSE);
1417 tcp_clear_xmit_timers(sk);
1418
1419 sk->sk_shutdown = SHUTDOWN_MASK;
1420
1421 if (!sock_flag(sk, SOCK_DEAD))
1422 sk->sk_state_change(sk);
1423 else
1424 tcp_destroy_sock(sk);
1425}
1426
1427static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
1428{
1429 rx_opt->dsack = 0;
1430 rx_opt->eff_sacks = 0;
1431 rx_opt->num_sacks = 0;
1432}
1433
1434static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1435{
1436 if (tp->rx_opt.tstamp_ok) {
1437 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1438 (TCPOPT_NOP << 16) |
1439 (TCPOPT_TIMESTAMP << 8) |
1440 TCPOLEN_TIMESTAMP);
1441 *ptr++ = htonl(tstamp);
1442 *ptr++ = htonl(tp->rx_opt.ts_recent);
1443 }
1444 if (tp->rx_opt.eff_sacks) {
1445 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1446 int this_sack;
1447
1448 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1449 (TCPOPT_NOP << 16) |
1450 (TCPOPT_SACK << 8) |
1451 (TCPOLEN_SACK_BASE +
1452 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1453 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1454 *ptr++ = htonl(sp[this_sack].start_seq);
1455 *ptr++ = htonl(sp[this_sack].end_seq);
1456 }
1457 if (tp->rx_opt.dsack) {
1458 tp->rx_opt.dsack = 0;
1459 tp->rx_opt.eff_sacks--;
1460 }
1461 }
1462}
1463
1464/* Construct a tcp options header for a SYN or SYN_ACK packet.
1465 * If this is every changed make sure to change the definition of
1466 * MAX_SYN_SIZE to match the new maximum number of options that you
1467 * can generate.
1468 */
1469static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1470 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1471{
1472 /* We always get an MSS option.
1473 * The option bytes which will be seen in normal data
1474 * packets should timestamps be used, must be in the MSS
1475 * advertised. But we subtract them from tp->mss_cache so
1476 * that calculations in tcp_sendmsg are simpler etc.
1477 * So account for this fact here if necessary. If we
1478 * don't do this correctly, as a receiver we won't
1479 * recognize data packets as being full sized when we
1480 * should, and thus we won't abide by the delayed ACK
1481 * rules correctly.
1482 * SACKs don't matter, we never delay an ACK when we
1483 * have any of those going out.
1484 */
1485 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1486 if (ts) {
1487 if(sack)
1488 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1489 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1490 else
1491 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1492 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1493 *ptr++ = htonl(tstamp); /* TSVAL */
1494 *ptr++ = htonl(ts_recent); /* TSECR */
1495 } else if(sack)
1496 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1497 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1498 if (offer_wscale)
1499 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1500}
1501
1502/* Determine a window scaling and initial window to offer. */
1503extern void tcp_select_initial_window(int __space, __u32 mss,
1504 __u32 *rcv_wnd, __u32 *window_clamp,
1505 int wscale_ok, __u8 *rcv_wscale);
1506
1507static inline int tcp_win_from_space(int space)
1508{
1509 return sysctl_tcp_adv_win_scale<=0 ?
1510 (space>>(-sysctl_tcp_adv_win_scale)) :
1511 space - (space>>sysctl_tcp_adv_win_scale);
1512}
1513
1514/* Note: caller must be prepared to deal with negative returns */
1515static inline int tcp_space(const struct sock *sk)
1516{
1517 return tcp_win_from_space(sk->sk_rcvbuf -
1518 atomic_read(&sk->sk_rmem_alloc));
1519}
1520
1521static inline int tcp_full_space(const struct sock *sk)
1522{
1523 return tcp_win_from_space(sk->sk_rcvbuf);
1524}
1525
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001526static inline void tcp_acceptq_queue(struct sock *sk, struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 struct sock *child)
1528{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001529 reqsk_queue_add(&tcp_sk(sk)->accept_queue, req, sk, child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532static inline void
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001533tcp_synq_removed(struct sock *sk, struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001535 if (reqsk_queue_removed(&tcp_sk(sk)->accept_queue, req) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 tcp_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537}
1538
1539static inline void tcp_synq_added(struct sock *sk)
1540{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001541 if (reqsk_queue_added(&tcp_sk(sk)->accept_queue) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543}
1544
1545static inline int tcp_synq_len(struct sock *sk)
1546{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001547 return reqsk_queue_len(&tcp_sk(sk)->accept_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548}
1549
1550static inline int tcp_synq_young(struct sock *sk)
1551{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001552 return reqsk_queue_len_young(&tcp_sk(sk)->accept_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553}
1554
1555static inline int tcp_synq_is_full(struct sock *sk)
1556{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001557 return reqsk_queue_is_full(&tcp_sk(sk)->accept_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558}
1559
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001560static inline void tcp_synq_unlink(struct tcp_sock *tp, struct request_sock *req,
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001561 struct request_sock **prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001563 reqsk_queue_unlink(&tp->accept_queue, req, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564}
1565
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001566static inline void tcp_synq_drop(struct sock *sk, struct request_sock *req,
1567 struct request_sock **prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568{
1569 tcp_synq_unlink(tcp_sk(sk), req, prev);
1570 tcp_synq_removed(sk, req);
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001571 reqsk_free(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572}
1573
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001574static __inline__ void tcp_openreq_init(struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 struct tcp_options_received *rx_opt,
1576 struct sk_buff *skb)
1577{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001578 struct inet_request_sock *ireq = inet_rsk(req);
1579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001581 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 req->mss = rx_opt->mss_clamp;
1583 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001584 ireq->tstamp_ok = rx_opt->tstamp_ok;
1585 ireq->sack_ok = rx_opt->sack_ok;
1586 ireq->snd_wscale = rx_opt->snd_wscale;
1587 ireq->wscale_ok = rx_opt->wscale_ok;
1588 ireq->acked = 0;
1589 ireq->ecn_ok = 0;
1590 ireq->rmt_port = skb->h.th->source;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591}
1592
1593extern void tcp_enter_memory_pressure(void);
1594
1595extern void tcp_listen_wlock(void);
1596
1597/* - We may sleep inside this lock.
1598 * - If sleeping is not required (or called from BH),
1599 * use plain read_(un)lock(&tcp_lhash_lock).
1600 */
1601
1602static inline void tcp_listen_lock(void)
1603{
1604 /* read_lock synchronizes to candidates to writers */
1605 read_lock(&tcp_lhash_lock);
1606 atomic_inc(&tcp_lhash_users);
1607 read_unlock(&tcp_lhash_lock);
1608}
1609
1610static inline void tcp_listen_unlock(void)
1611{
1612 if (atomic_dec_and_test(&tcp_lhash_users))
1613 wake_up(&tcp_lhash_wait);
1614}
1615
1616static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1617{
1618 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1619}
1620
1621static inline int keepalive_time_when(const struct tcp_sock *tp)
1622{
1623 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1624}
1625
1626static inline int tcp_fin_time(const struct tcp_sock *tp)
1627{
1628 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1629
1630 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1631 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1632
1633 return fin_timeout;
1634}
1635
1636static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1637{
1638 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1639 return 0;
1640 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1641 return 0;
1642
1643 /* RST segments are not recommended to carry timestamp,
1644 and, if they do, it is recommended to ignore PAWS because
1645 "their cleanup function should take precedence over timestamps."
1646 Certainly, it is mistake. It is necessary to understand the reasons
1647 of this constraint to relax it: if peer reboots, clock may go
1648 out-of-sync and half-open connections will not be reset.
1649 Actually, the problem would be not existing if all
1650 the implementations followed draft about maintaining clock
1651 via reboots. Linux-2.2 DOES NOT!
1652
1653 However, we can relax time bounds for RST segments to MSL.
1654 */
1655 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1656 return 0;
1657 return 1;
1658}
1659
1660static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
1661{
1662 sk->sk_route_caps = dst->dev->features;
1663 if (sk->sk_route_caps & NETIF_F_TSO) {
1664 if (sock_flag(sk, SOCK_NO_LARGESEND) || dst->header_len)
1665 sk->sk_route_caps &= ~NETIF_F_TSO;
1666 }
1667}
1668
1669#define TCP_CHECK_TIMER(sk) do { } while (0)
1670
1671static inline int tcp_use_frto(const struct sock *sk)
1672{
1673 const struct tcp_sock *tp = tcp_sk(sk);
1674
1675 /* F-RTO must be activated in sysctl and there must be some
1676 * unsent new data, and the advertised window should allow
1677 * sending it.
1678 */
1679 return (sysctl_tcp_frto && sk->sk_send_head &&
1680 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1681 tp->snd_una + tp->snd_wnd));
1682}
1683
1684static inline void tcp_mib_init(void)
1685{
1686 /* See RFC 2012 */
1687 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1688 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1689 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1690 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1691}
1692
1693/* /proc */
1694enum tcp_seq_states {
1695 TCP_SEQ_STATE_LISTENING,
1696 TCP_SEQ_STATE_OPENREQ,
1697 TCP_SEQ_STATE_ESTABLISHED,
1698 TCP_SEQ_STATE_TIME_WAIT,
1699};
1700
1701struct tcp_seq_afinfo {
1702 struct module *owner;
1703 char *name;
1704 sa_family_t family;
1705 int (*seq_show) (struct seq_file *m, void *v);
1706 struct file_operations *seq_fops;
1707};
1708
1709struct tcp_iter_state {
1710 sa_family_t family;
1711 enum tcp_seq_states state;
1712 struct sock *syn_wait_sk;
1713 int bucket, sbucket, num, uid;
1714 struct seq_operations seq_ops;
1715};
1716
1717extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1718extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1719
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720#endif /* _TCP_H */