blob: ec9e20c27179af1f6446ece3e97b0dfb090b8403 [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
724static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp)
725{
726 if (tp->ack.quick && --tp->ack.quick == 0) {
727 /* Leaving quickack mode we deflate ATO. */
728 tp->ack.ato = TCP_ATO_MIN;
729 }
730}
731
732extern void tcp_enter_quickack_mode(struct tcp_sock *tp);
733
734static __inline__ void tcp_delack_init(struct tcp_sock *tp)
735{
736 memset(&tp->ack, 0, sizeof(tp->ack));
737}
738
739static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
740{
741 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
742}
743
744enum tcp_tw_status
745{
746 TCP_TW_SUCCESS = 0,
747 TCP_TW_RST = 1,
748 TCP_TW_ACK = 2,
749 TCP_TW_SYN = 3
750};
751
752
753extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
754 struct sk_buff *skb,
755 struct tcphdr *th,
756 unsigned len);
757
758extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700759 struct request_sock *req,
760 struct request_sock **prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761extern int tcp_child_process(struct sock *parent,
762 struct sock *child,
763 struct sk_buff *skb);
764extern void tcp_enter_frto(struct sock *sk);
765extern void tcp_enter_loss(struct sock *sk, int how);
766extern void tcp_clear_retrans(struct tcp_sock *tp);
767extern void tcp_update_metrics(struct sock *sk);
768
769extern void tcp_close(struct sock *sk,
770 long timeout);
771extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
772extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
773
774extern int tcp_getsockopt(struct sock *sk, int level,
775 int optname,
776 char __user *optval,
777 int __user *optlen);
778extern int tcp_setsockopt(struct sock *sk, int level,
779 int optname, char __user *optval,
780 int optlen);
781extern void tcp_set_keepalive(struct sock *sk, int val);
782extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
783 struct msghdr *msg,
784 size_t len, int nonblock,
785 int flags, int *addr_len);
786
787extern int tcp_listen_start(struct sock *sk);
788
789extern void tcp_parse_options(struct sk_buff *skb,
790 struct tcp_options_received *opt_rx,
791 int estab);
792
793/*
794 * TCP v4 functions exported for the inet6 API
795 */
796
797extern int tcp_v4_rebuild_header(struct sock *sk);
798
799extern int tcp_v4_build_header(struct sock *sk,
800 struct sk_buff *skb);
801
802extern void tcp_v4_send_check(struct sock *sk,
803 struct tcphdr *th, int len,
804 struct sk_buff *skb);
805
806extern int tcp_v4_conn_request(struct sock *sk,
807 struct sk_buff *skb);
808
809extern struct sock * tcp_create_openreq_child(struct sock *sk,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700810 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 struct sk_buff *skb);
812
813extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
814 struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700815 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 struct dst_entry *dst);
817
818extern int tcp_v4_do_rcv(struct sock *sk,
819 struct sk_buff *skb);
820
821extern int tcp_v4_connect(struct sock *sk,
822 struct sockaddr *uaddr,
823 int addr_len);
824
825extern int tcp_connect(struct sock *sk);
826
827extern struct sk_buff * tcp_make_synack(struct sock *sk,
828 struct dst_entry *dst,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700829 struct request_sock *req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831extern int tcp_disconnect(struct sock *sk, int flags);
832
833extern void tcp_unhash(struct sock *sk);
834
835extern int tcp_v4_hash_connecting(struct sock *sk);
836
837
838/* From syncookies.c */
839extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
840 struct ip_options *opt);
841extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
842 __u16 *mss);
843
844/* tcp_output.c */
845
846extern int tcp_write_xmit(struct sock *, int nonagle);
847extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
848extern void tcp_xmit_retransmit_queue(struct sock *);
849extern void tcp_simple_retransmit(struct sock *);
850extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
851
852extern void tcp_send_probe0(struct sock *);
853extern void tcp_send_partial(struct sock *);
854extern int tcp_write_wakeup(struct sock *);
855extern void tcp_send_fin(struct sock *sk);
856extern void tcp_send_active_reset(struct sock *sk, int priority);
857extern int tcp_send_synack(struct sock *);
858extern void tcp_push_one(struct sock *, unsigned mss_now);
859extern void tcp_send_ack(struct sock *sk);
860extern void tcp_send_delayed_ack(struct sock *sk);
861
862/* tcp_timer.c */
863extern void tcp_init_xmit_timers(struct sock *);
864extern void tcp_clear_xmit_timers(struct sock *);
865
866extern void tcp_delete_keepalive_timer(struct sock *);
867extern void tcp_reset_keepalive_timer(struct sock *, unsigned long);
868extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
869extern unsigned int tcp_current_mss(struct sock *sk, int large);
870
871#ifdef TCP_DEBUG
872extern const char tcp_timer_bug_msg[];
873#endif
874
875/* tcp_diag.c */
876extern void tcp_get_info(struct sock *, struct tcp_info *);
877
878/* Read 'sendfile()'-style from a TCP socket */
879typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
880 unsigned int, size_t);
881extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
882 sk_read_actor_t recv_actor);
883
884static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
885{
886 struct tcp_sock *tp = tcp_sk(sk);
887
888 switch (what) {
889 case TCP_TIME_RETRANS:
890 case TCP_TIME_PROBE0:
891 tp->pending = 0;
892
893#ifdef TCP_CLEAR_TIMERS
894 sk_stop_timer(sk, &tp->retransmit_timer);
895#endif
896 break;
897 case TCP_TIME_DACK:
898 tp->ack.blocked = 0;
899 tp->ack.pending = 0;
900
901#ifdef TCP_CLEAR_TIMERS
902 sk_stop_timer(sk, &tp->delack_timer);
903#endif
904 break;
905 default:
906#ifdef TCP_DEBUG
907 printk(tcp_timer_bug_msg);
908#endif
909 return;
910 };
911
912}
913
914/*
915 * Reset the retransmission timer
916 */
917static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
918{
919 struct tcp_sock *tp = tcp_sk(sk);
920
921 if (when > TCP_RTO_MAX) {
922#ifdef TCP_DEBUG
923 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
924#endif
925 when = TCP_RTO_MAX;
926 }
927
928 switch (what) {
929 case TCP_TIME_RETRANS:
930 case TCP_TIME_PROBE0:
931 tp->pending = what;
932 tp->timeout = jiffies+when;
933 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
934 break;
935
936 case TCP_TIME_DACK:
937 tp->ack.pending |= TCP_ACK_TIMER;
938 tp->ack.timeout = jiffies+when;
939 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
940 break;
941
942 default:
943#ifdef TCP_DEBUG
944 printk(tcp_timer_bug_msg);
945#endif
946 return;
947 };
948}
949
950/* Initialize RCV_MSS value.
951 * RCV_MSS is an our guess about MSS used by the peer.
952 * We haven't any direct information about the MSS.
953 * It's better to underestimate the RCV_MSS rather than overestimate.
954 * Overestimations make us ACKing less frequently than needed.
955 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
956 */
957
958static inline void tcp_initialize_rcv_mss(struct sock *sk)
959{
960 struct tcp_sock *tp = tcp_sk(sk);
961 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
962
963 hint = min(hint, tp->rcv_wnd/2);
964 hint = min(hint, TCP_MIN_RCVMSS);
965 hint = max(hint, TCP_MIN_MSS);
966
967 tp->ack.rcv_mss = hint;
968}
969
970static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
971{
972 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
973 ntohl(TCP_FLAG_ACK) |
974 snd_wnd);
975}
976
977static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
978{
979 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
980}
981
982static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
983{
984 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
985 tp->rcv_wnd &&
986 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
987 !tp->urg_data)
988 tcp_fast_path_on(tp);
989}
990
991/* Compute the actual receive window we are currently advertising.
992 * Rcv_nxt can be after the window if our peer push more data
993 * than the offered window.
994 */
995static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
996{
997 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
998
999 if (win < 0)
1000 win = 0;
1001 return (u32) win;
1002}
1003
1004/* Choose a new window, without checks for shrinking, and without
1005 * scaling applied to the result. The caller does these things
1006 * if necessary. This is a "raw" window selection.
1007 */
1008extern u32 __tcp_select_window(struct sock *sk);
1009
1010/* TCP timestamps are only 32-bits, this causes a slight
1011 * complication on 64-bit systems since we store a snapshot
1012 * of jiffies in the buffer control blocks below. We decidely
1013 * only use of the low 32-bits of jiffies and hide the ugly
1014 * casts with the following macro.
1015 */
1016#define tcp_time_stamp ((__u32)(jiffies))
1017
1018/* This is what the send packet queueing engine uses to pass
1019 * TCP per-packet control information to the transmission
1020 * code. We also store the host-order sequence numbers in
1021 * here too. This is 36 bytes on 32-bit architectures,
1022 * 40 bytes on 64-bit machines, if this grows please adjust
1023 * skbuff.h:skbuff->cb[xxx] size appropriately.
1024 */
1025struct tcp_skb_cb {
1026 union {
1027 struct inet_skb_parm h4;
1028#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1029 struct inet6_skb_parm h6;
1030#endif
1031 } header; /* For incoming frames */
1032 __u32 seq; /* Starting sequence number */
1033 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1034 __u32 when; /* used to compute rtt's */
1035 __u8 flags; /* TCP header flags. */
1036
1037 /* NOTE: These must match up to the flags byte in a
1038 * real TCP header.
1039 */
1040#define TCPCB_FLAG_FIN 0x01
1041#define TCPCB_FLAG_SYN 0x02
1042#define TCPCB_FLAG_RST 0x04
1043#define TCPCB_FLAG_PSH 0x08
1044#define TCPCB_FLAG_ACK 0x10
1045#define TCPCB_FLAG_URG 0x20
1046#define TCPCB_FLAG_ECE 0x40
1047#define TCPCB_FLAG_CWR 0x80
1048
1049 __u8 sacked; /* State flags for SACK/FACK. */
1050#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1051#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1052#define TCPCB_LOST 0x04 /* SKB is lost */
1053#define TCPCB_TAGBITS 0x07 /* All tag bits */
1054
1055#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1056#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1057
1058#define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1059
1060#define TCPCB_AT_TAIL (TCPCB_URG)
1061
1062 __u16 urg_ptr; /* Valid w/URG flags is set. */
1063 __u32 ack_seq; /* Sequence number ACK'd */
1064};
1065
1066#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1067
1068#include <net/tcp_ecn.h>
1069
1070/* Due to TSO, an SKB can be composed of multiple actual
1071 * packets. To keep these tracked properly, we use this.
1072 */
1073static inline int tcp_skb_pcount(const struct sk_buff *skb)
1074{
1075 return skb_shinfo(skb)->tso_segs;
1076}
1077
1078/* This is valid iff tcp_skb_pcount() > 1. */
1079static inline int tcp_skb_mss(const struct sk_buff *skb)
1080{
1081 return skb_shinfo(skb)->tso_size;
1082}
1083
1084static inline void tcp_dec_pcount_approx(__u32 *count,
1085 const struct sk_buff *skb)
1086{
1087 if (*count) {
1088 *count -= tcp_skb_pcount(skb);
1089 if ((int)*count < 0)
1090 *count = 0;
1091 }
1092}
1093
1094static inline void tcp_packets_out_inc(struct sock *sk,
1095 struct tcp_sock *tp,
1096 const struct sk_buff *skb)
1097{
1098 int orig = tp->packets_out;
1099
1100 tp->packets_out += tcp_skb_pcount(skb);
1101 if (!orig)
1102 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
1103}
1104
1105static inline void tcp_packets_out_dec(struct tcp_sock *tp,
1106 const struct sk_buff *skb)
1107{
1108 tp->packets_out -= tcp_skb_pcount(skb);
1109}
1110
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001111/* Events passed to congestion control interface */
1112enum tcp_ca_event {
1113 CA_EVENT_TX_START, /* first transmit when no packets in flight */
1114 CA_EVENT_CWND_RESTART, /* congestion window restart */
1115 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
1116 CA_EVENT_FRTO, /* fast recovery timeout */
1117 CA_EVENT_LOSS, /* loss timeout */
1118 CA_EVENT_FAST_ACK, /* in sequence ack */
1119 CA_EVENT_SLOW_ACK, /* other ack */
1120};
1121
1122/*
1123 * Interface for adding new TCP congestion control handlers
1124 */
1125#define TCP_CA_NAME_MAX 16
1126struct tcp_congestion_ops {
1127 struct list_head list;
1128
1129 /* initialize private data (optional) */
1130 void (*init)(struct tcp_sock *tp);
1131 /* cleanup private data (optional) */
1132 void (*release)(struct tcp_sock *tp);
1133
1134 /* return slow start threshold (required) */
1135 u32 (*ssthresh)(struct tcp_sock *tp);
1136 /* lower bound for congestion window (optional) */
1137 u32 (*min_cwnd)(struct tcp_sock *tp);
1138 /* do new cwnd calculation (required) */
1139 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
1140 u32 rtt, u32 in_flight, int good_ack);
1141 /* round trip time sample per acked packet (optional) */
1142 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
1143 /* call before changing ca_state (optional) */
1144 void (*set_state)(struct tcp_sock *tp, u8 new_state);
1145 /* call when cwnd event occurs (optional) */
1146 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
1147 /* new value of cwnd after loss (optional) */
1148 u32 (*undo_cwnd)(struct tcp_sock *tp);
1149 /* hook for packet ack accounting (optional) */
1150 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
1151 /* get info for tcp_diag (optional) */
1152 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
1153
1154 char name[TCP_CA_NAME_MAX];
1155 struct module *owner;
1156};
1157
1158extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1159extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1160
1161extern void tcp_init_congestion_control(struct tcp_sock *tp);
1162extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
1163extern int tcp_set_default_congestion_control(const char *name);
1164extern void tcp_get_default_congestion_control(char *name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07001165extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001166
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07001167extern struct tcp_congestion_ops tcp_init_congestion_ops;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001168extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
1169extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
1170 u32 rtt, u32 in_flight, int flag);
1171extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
David S. Millera8acfba2005-06-23 23:45:02 -07001172extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001173
1174static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
1175{
1176 if (tp->ca_ops->set_state)
1177 tp->ca_ops->set_state(tp, ca_state);
1178 tp->ca_state = ca_state;
1179}
1180
1181static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
1182{
1183 if (tp->ca_ops->cwnd_event)
1184 tp->ca_ops->cwnd_event(tp, event);
1185}
1186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187/* This determines how many packets are "in the network" to the best
1188 * of our knowledge. In many cases it is conservative, but where
1189 * detailed information is available from the receiver (via SACK
1190 * blocks etc.) we can make more aggressive calculations.
1191 *
1192 * Use this for decisions involving congestion control, use just
1193 * tp->packets_out to determine if the send queue is empty or not.
1194 *
1195 * Read this equation as:
1196 *
1197 * "Packets sent once on transmission queue" MINUS
1198 * "Packets left network, but not honestly ACKed yet" PLUS
1199 * "Packets fast retransmitted"
1200 */
1201static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1202{
1203 return (tp->packets_out - tp->left_out + tp->retrans_out);
1204}
1205
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1207 * The exception is rate halving phase, when cwnd is decreasing towards
1208 * ssthresh.
1209 */
1210static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
1211{
1212 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1213 return tp->snd_ssthresh;
1214 else
1215 return max(tp->snd_ssthresh,
1216 ((tp->snd_cwnd >> 1) +
1217 (tp->snd_cwnd >> 2)));
1218}
1219
1220static inline void tcp_sync_left_out(struct tcp_sock *tp)
1221{
1222 if (tp->rx_opt.sack_ok &&
1223 (tp->sacked_out >= tp->packets_out - tp->lost_out))
1224 tp->sacked_out = tp->packets_out - tp->lost_out;
1225 tp->left_out = tp->sacked_out + tp->lost_out;
1226}
1227
1228extern void tcp_cwnd_application_limited(struct sock *sk);
1229
1230/* Congestion window validation. (RFC2861) */
1231
1232static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_sock *tp)
1233{
1234 __u32 packets_out = tp->packets_out;
1235
1236 if (packets_out >= tp->snd_cwnd) {
1237 /* Network is feed fully. */
1238 tp->snd_cwnd_used = 0;
1239 tp->snd_cwnd_stamp = tcp_time_stamp;
1240 } else {
1241 /* Network starves. */
1242 if (tp->packets_out > tp->snd_cwnd_used)
1243 tp->snd_cwnd_used = tp->packets_out;
1244
1245 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1246 tcp_cwnd_application_limited(sk);
1247 }
1248}
1249
1250/* Set slow start threshould and cwnd not falling to slow start */
1251static inline void __tcp_enter_cwr(struct tcp_sock *tp)
1252{
1253 tp->undo_marker = 0;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001254 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 tp->snd_cwnd = min(tp->snd_cwnd,
1256 tcp_packets_in_flight(tp) + 1U);
1257 tp->snd_cwnd_cnt = 0;
1258 tp->high_seq = tp->snd_nxt;
1259 tp->snd_cwnd_stamp = tcp_time_stamp;
1260 TCP_ECN_queue_cwr(tp);
1261}
1262
1263static inline void tcp_enter_cwr(struct tcp_sock *tp)
1264{
1265 tp->prior_ssthresh = 0;
1266 if (tp->ca_state < TCP_CA_CWR) {
1267 __tcp_enter_cwr(tp);
1268 tcp_set_ca_state(tp, TCP_CA_CWR);
1269 }
1270}
1271
1272extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
1273
1274/* Slow start with delack produces 3 packets of burst, so that
1275 * it is safe "de facto".
1276 */
1277static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1278{
1279 return 3;
1280}
1281
1282static __inline__ int tcp_minshall_check(const struct tcp_sock *tp)
1283{
1284 return after(tp->snd_sml,tp->snd_una) &&
1285 !after(tp->snd_sml, tp->snd_nxt);
1286}
1287
1288static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
1289 const struct sk_buff *skb)
1290{
1291 if (skb->len < mss)
1292 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1293}
1294
1295/* Return 0, if packet can be sent now without violation Nagle's rules:
1296 1. It is full sized.
1297 2. Or it contains FIN.
1298 3. Or TCP_NODELAY was set.
1299 4. Or TCP_CORK is not set, and all sent packets are ACKed.
1300 With Minshall's modification: all sent small packets are ACKed.
1301 */
1302
1303static __inline__ int
1304tcp_nagle_check(const struct tcp_sock *tp, const struct sk_buff *skb,
1305 unsigned mss_now, int nonagle)
1306{
1307 return (skb->len < mss_now &&
1308 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1309 ((nonagle&TCP_NAGLE_CORK) ||
1310 (!nonagle &&
1311 tp->packets_out &&
1312 tcp_minshall_check(tp))));
1313}
1314
David S. Millerd5ac99a2005-04-24 19:12:33 -07001315extern void tcp_set_skb_tso_segs(struct sock *, struct sk_buff *);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317/* This checks if the data bearing packet SKB (usually sk->sk_send_head)
1318 * should be put on the wire right now.
1319 */
David S. Millerd5ac99a2005-04-24 19:12:33 -07001320static __inline__ int tcp_snd_test(struct sock *sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 struct sk_buff *skb,
1322 unsigned cur_mss, int nonagle)
1323{
David S. Millerd5ac99a2005-04-24 19:12:33 -07001324 struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 int pkts = tcp_skb_pcount(skb);
1326
1327 if (!pkts) {
David S. Millerd5ac99a2005-04-24 19:12:33 -07001328 tcp_set_skb_tso_segs(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 pkts = tcp_skb_pcount(skb);
1330 }
1331
1332 /* RFC 1122 - section 4.2.3.4
1333 *
1334 * We must queue if
1335 *
1336 * a) The right edge of this frame exceeds the window
1337 * b) There are packets in flight and we have a small segment
1338 * [SWS avoidance and Nagle algorithm]
1339 * (part of SWS is done on packetization)
1340 * Minshall version sounds: there are no _small_
1341 * segments in flight. (tcp_nagle_check)
1342 * c) We have too many packets 'in flight'
1343 *
1344 * Don't use the nagle rule for urgent data (or
1345 * for the final FIN -DaveM).
1346 *
1347 * Also, Nagle rule does not apply to frames, which
1348 * sit in the middle of queue (they have no chances
1349 * to get new data) and if room at tail of skb is
1350 * not enough to save something seriously (<32 for now).
1351 */
1352
1353 /* Don't be strict about the congestion window for the
1354 * final FIN frame. -DaveM
1355 */
1356 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1357 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1358 (((tcp_packets_in_flight(tp) + (pkts-1)) < tp->snd_cwnd) ||
1359 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1360 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1361}
1362
1363static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
1364{
1365 if (!tp->packets_out && !tp->pending)
1366 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1367}
1368
1369static __inline__ int tcp_skb_is_last(const struct sock *sk,
1370 const struct sk_buff *skb)
1371{
1372 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1373}
1374
1375/* Push out any pending frames which were held back due to
1376 * TCP_CORK or attempt at coalescing tiny packets.
1377 * The socket must be locked by the caller.
1378 */
1379static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1380 struct tcp_sock *tp,
1381 unsigned cur_mss,
1382 int nonagle)
1383{
1384 struct sk_buff *skb = sk->sk_send_head;
1385
1386 if (skb) {
1387 if (!tcp_skb_is_last(sk, skb))
1388 nonagle = TCP_NAGLE_PUSH;
David S. Millerd5ac99a2005-04-24 19:12:33 -07001389 if (!tcp_snd_test(sk, skb, cur_mss, nonagle) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 tcp_write_xmit(sk, nonagle))
1391 tcp_check_probe_timer(sk, tp);
1392 }
1393 tcp_cwnd_validate(sk, tp);
1394}
1395
1396static __inline__ void tcp_push_pending_frames(struct sock *sk,
1397 struct tcp_sock *tp)
1398{
1399 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1400}
1401
1402static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp)
1403{
1404 struct sk_buff *skb = sk->sk_send_head;
1405
1406 return (skb &&
David S. Millerd5ac99a2005-04-24 19:12:33 -07001407 tcp_snd_test(sk, skb, tcp_current_mss(sk, 1),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1409}
1410
1411static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1412{
1413 tp->snd_wl1 = seq;
1414}
1415
1416static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1417{
1418 tp->snd_wl1 = seq;
1419}
1420
1421extern void tcp_destroy_sock(struct sock *sk);
1422
1423
1424/*
1425 * Calculate(/check) TCP checksum
1426 */
1427static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1428 unsigned long saddr, unsigned long daddr,
1429 unsigned long base)
1430{
1431 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1432}
1433
1434static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1435{
1436 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1437}
1438
1439static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1440{
1441 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1442 __tcp_checksum_complete(skb);
1443}
1444
1445/* Prequeue for VJ style copy to user, combined with checksumming. */
1446
1447static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
1448{
1449 tp->ucopy.task = NULL;
1450 tp->ucopy.len = 0;
1451 tp->ucopy.memory = 0;
1452 skb_queue_head_init(&tp->ucopy.prequeue);
1453}
1454
1455/* Packet is added to VJ-style prequeue for processing in process
1456 * context, if a reader task is waiting. Apparently, this exciting
1457 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1458 * failed somewhere. Latency? Burstiness? Well, at least now we will
1459 * see, why it failed. 8)8) --ANK
1460 *
1461 * NOTE: is this not too big to inline?
1462 */
1463static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1464{
1465 struct tcp_sock *tp = tcp_sk(sk);
1466
1467 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1468 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1469 tp->ucopy.memory += skb->truesize;
1470 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1471 struct sk_buff *skb1;
1472
1473 BUG_ON(sock_owned_by_user(sk));
1474
1475 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1476 sk->sk_backlog_rcv(sk, skb1);
1477 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1478 }
1479
1480 tp->ucopy.memory = 0;
1481 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1482 wake_up_interruptible(sk->sk_sleep);
1483 if (!tcp_ack_scheduled(tp))
1484 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1485 }
1486 return 1;
1487 }
1488 return 0;
1489}
1490
1491
1492#undef STATE_TRACE
1493
1494#ifdef STATE_TRACE
1495static const char *statename[]={
1496 "Unused","Established","Syn Sent","Syn Recv",
1497 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1498 "Close Wait","Last ACK","Listen","Closing"
1499};
1500#endif
1501
1502static __inline__ void tcp_set_state(struct sock *sk, int state)
1503{
1504 int oldstate = sk->sk_state;
1505
1506 switch (state) {
1507 case TCP_ESTABLISHED:
1508 if (oldstate != TCP_ESTABLISHED)
1509 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1510 break;
1511
1512 case TCP_CLOSE:
1513 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1514 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1515
1516 sk->sk_prot->unhash(sk);
1517 if (tcp_sk(sk)->bind_hash &&
1518 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1519 tcp_put_port(sk);
1520 /* fall through */
1521 default:
1522 if (oldstate==TCP_ESTABLISHED)
1523 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1524 }
1525
1526 /* Change state AFTER socket is unhashed to avoid closed
1527 * socket sitting in hash tables.
1528 */
1529 sk->sk_state = state;
1530
1531#ifdef STATE_TRACE
1532 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1533#endif
1534}
1535
1536static __inline__ void tcp_done(struct sock *sk)
1537{
1538 tcp_set_state(sk, TCP_CLOSE);
1539 tcp_clear_xmit_timers(sk);
1540
1541 sk->sk_shutdown = SHUTDOWN_MASK;
1542
1543 if (!sock_flag(sk, SOCK_DEAD))
1544 sk->sk_state_change(sk);
1545 else
1546 tcp_destroy_sock(sk);
1547}
1548
1549static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
1550{
1551 rx_opt->dsack = 0;
1552 rx_opt->eff_sacks = 0;
1553 rx_opt->num_sacks = 0;
1554}
1555
1556static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1557{
1558 if (tp->rx_opt.tstamp_ok) {
1559 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1560 (TCPOPT_NOP << 16) |
1561 (TCPOPT_TIMESTAMP << 8) |
1562 TCPOLEN_TIMESTAMP);
1563 *ptr++ = htonl(tstamp);
1564 *ptr++ = htonl(tp->rx_opt.ts_recent);
1565 }
1566 if (tp->rx_opt.eff_sacks) {
1567 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1568 int this_sack;
1569
1570 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1571 (TCPOPT_NOP << 16) |
1572 (TCPOPT_SACK << 8) |
1573 (TCPOLEN_SACK_BASE +
1574 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1575 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1576 *ptr++ = htonl(sp[this_sack].start_seq);
1577 *ptr++ = htonl(sp[this_sack].end_seq);
1578 }
1579 if (tp->rx_opt.dsack) {
1580 tp->rx_opt.dsack = 0;
1581 tp->rx_opt.eff_sacks--;
1582 }
1583 }
1584}
1585
1586/* Construct a tcp options header for a SYN or SYN_ACK packet.
1587 * If this is every changed make sure to change the definition of
1588 * MAX_SYN_SIZE to match the new maximum number of options that you
1589 * can generate.
1590 */
1591static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1592 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1593{
1594 /* We always get an MSS option.
1595 * The option bytes which will be seen in normal data
1596 * packets should timestamps be used, must be in the MSS
1597 * advertised. But we subtract them from tp->mss_cache so
1598 * that calculations in tcp_sendmsg are simpler etc.
1599 * So account for this fact here if necessary. If we
1600 * don't do this correctly, as a receiver we won't
1601 * recognize data packets as being full sized when we
1602 * should, and thus we won't abide by the delayed ACK
1603 * rules correctly.
1604 * SACKs don't matter, we never delay an ACK when we
1605 * have any of those going out.
1606 */
1607 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1608 if (ts) {
1609 if(sack)
1610 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1611 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1612 else
1613 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1614 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1615 *ptr++ = htonl(tstamp); /* TSVAL */
1616 *ptr++ = htonl(ts_recent); /* TSECR */
1617 } else if(sack)
1618 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1619 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1620 if (offer_wscale)
1621 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1622}
1623
1624/* Determine a window scaling and initial window to offer. */
1625extern void tcp_select_initial_window(int __space, __u32 mss,
1626 __u32 *rcv_wnd, __u32 *window_clamp,
1627 int wscale_ok, __u8 *rcv_wscale);
1628
1629static inline int tcp_win_from_space(int space)
1630{
1631 return sysctl_tcp_adv_win_scale<=0 ?
1632 (space>>(-sysctl_tcp_adv_win_scale)) :
1633 space - (space>>sysctl_tcp_adv_win_scale);
1634}
1635
1636/* Note: caller must be prepared to deal with negative returns */
1637static inline int tcp_space(const struct sock *sk)
1638{
1639 return tcp_win_from_space(sk->sk_rcvbuf -
1640 atomic_read(&sk->sk_rmem_alloc));
1641}
1642
1643static inline int tcp_full_space(const struct sock *sk)
1644{
1645 return tcp_win_from_space(sk->sk_rcvbuf);
1646}
1647
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001648static inline void tcp_acceptq_queue(struct sock *sk, struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 struct sock *child)
1650{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001651 reqsk_queue_add(&tcp_sk(sk)->accept_queue, req, sk, child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652}
1653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654static inline void
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001655tcp_synq_removed(struct sock *sk, struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001657 if (reqsk_queue_removed(&tcp_sk(sk)->accept_queue, req) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 tcp_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
1661static inline void tcp_synq_added(struct sock *sk)
1662{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001663 if (reqsk_queue_added(&tcp_sk(sk)->accept_queue) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665}
1666
1667static inline int tcp_synq_len(struct sock *sk)
1668{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001669 return reqsk_queue_len(&tcp_sk(sk)->accept_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670}
1671
1672static inline int tcp_synq_young(struct sock *sk)
1673{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001674 return reqsk_queue_len_young(&tcp_sk(sk)->accept_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675}
1676
1677static inline int tcp_synq_is_full(struct sock *sk)
1678{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001679 return reqsk_queue_is_full(&tcp_sk(sk)->accept_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680}
1681
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001682static inline void tcp_synq_unlink(struct tcp_sock *tp, struct request_sock *req,
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001683 struct request_sock **prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684{
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001685 reqsk_queue_unlink(&tp->accept_queue, req, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686}
1687
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001688static inline void tcp_synq_drop(struct sock *sk, struct request_sock *req,
1689 struct request_sock **prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690{
1691 tcp_synq_unlink(tcp_sk(sk), req, prev);
1692 tcp_synq_removed(sk, req);
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001693 reqsk_free(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694}
1695
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001696static __inline__ void tcp_openreq_init(struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 struct tcp_options_received *rx_opt,
1698 struct sk_buff *skb)
1699{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001700 struct inet_request_sock *ireq = inet_rsk(req);
1701
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001703 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 req->mss = rx_opt->mss_clamp;
1705 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001706 ireq->tstamp_ok = rx_opt->tstamp_ok;
1707 ireq->sack_ok = rx_opt->sack_ok;
1708 ireq->snd_wscale = rx_opt->snd_wscale;
1709 ireq->wscale_ok = rx_opt->wscale_ok;
1710 ireq->acked = 0;
1711 ireq->ecn_ok = 0;
1712 ireq->rmt_port = skb->h.th->source;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713}
1714
1715extern void tcp_enter_memory_pressure(void);
1716
1717extern void tcp_listen_wlock(void);
1718
1719/* - We may sleep inside this lock.
1720 * - If sleeping is not required (or called from BH),
1721 * use plain read_(un)lock(&tcp_lhash_lock).
1722 */
1723
1724static inline void tcp_listen_lock(void)
1725{
1726 /* read_lock synchronizes to candidates to writers */
1727 read_lock(&tcp_lhash_lock);
1728 atomic_inc(&tcp_lhash_users);
1729 read_unlock(&tcp_lhash_lock);
1730}
1731
1732static inline void tcp_listen_unlock(void)
1733{
1734 if (atomic_dec_and_test(&tcp_lhash_users))
1735 wake_up(&tcp_lhash_wait);
1736}
1737
1738static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1739{
1740 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1741}
1742
1743static inline int keepalive_time_when(const struct tcp_sock *tp)
1744{
1745 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1746}
1747
1748static inline int tcp_fin_time(const struct tcp_sock *tp)
1749{
1750 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1751
1752 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1753 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1754
1755 return fin_timeout;
1756}
1757
1758static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1759{
1760 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1761 return 0;
1762 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1763 return 0;
1764
1765 /* RST segments are not recommended to carry timestamp,
1766 and, if they do, it is recommended to ignore PAWS because
1767 "their cleanup function should take precedence over timestamps."
1768 Certainly, it is mistake. It is necessary to understand the reasons
1769 of this constraint to relax it: if peer reboots, clock may go
1770 out-of-sync and half-open connections will not be reset.
1771 Actually, the problem would be not existing if all
1772 the implementations followed draft about maintaining clock
1773 via reboots. Linux-2.2 DOES NOT!
1774
1775 However, we can relax time bounds for RST segments to MSL.
1776 */
1777 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1778 return 0;
1779 return 1;
1780}
1781
1782static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
1783{
1784 sk->sk_route_caps = dst->dev->features;
1785 if (sk->sk_route_caps & NETIF_F_TSO) {
1786 if (sock_flag(sk, SOCK_NO_LARGESEND) || dst->header_len)
1787 sk->sk_route_caps &= ~NETIF_F_TSO;
1788 }
1789}
1790
1791#define TCP_CHECK_TIMER(sk) do { } while (0)
1792
1793static inline int tcp_use_frto(const struct sock *sk)
1794{
1795 const struct tcp_sock *tp = tcp_sk(sk);
1796
1797 /* F-RTO must be activated in sysctl and there must be some
1798 * unsent new data, and the advertised window should allow
1799 * sending it.
1800 */
1801 return (sysctl_tcp_frto && sk->sk_send_head &&
1802 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1803 tp->snd_una + tp->snd_wnd));
1804}
1805
1806static inline void tcp_mib_init(void)
1807{
1808 /* See RFC 2012 */
1809 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1810 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1811 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1812 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1813}
1814
1815/* /proc */
1816enum tcp_seq_states {
1817 TCP_SEQ_STATE_LISTENING,
1818 TCP_SEQ_STATE_OPENREQ,
1819 TCP_SEQ_STATE_ESTABLISHED,
1820 TCP_SEQ_STATE_TIME_WAIT,
1821};
1822
1823struct tcp_seq_afinfo {
1824 struct module *owner;
1825 char *name;
1826 sa_family_t family;
1827 int (*seq_show) (struct seq_file *m, void *v);
1828 struct file_operations *seq_fops;
1829};
1830
1831struct tcp_iter_state {
1832 sa_family_t family;
1833 enum tcp_seq_states state;
1834 struct sock *syn_wait_sk;
1835 int bucket, sbucket, num, uid;
1836 struct seq_operations seq_ops;
1837};
1838
1839extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1840extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842#endif /* _TCP_H */