[INET]: Just rename the TCP hashtable functions/structs to inet_
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / include / net / tcp.h
CommitLineData
1da177e4
LT
1/*
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 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT
11 * 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>
2e6599cb 34#include <net/request_sock.h>
1da177e4
LT
35#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 */
0f7ff927 47struct inet_ehash_bucket {
1da177e4
LT
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. */
0f7ff927 53#define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
1da177e4
LT
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 */
0f7ff927 86struct inet_bind_bucket {
1da177e4
LT
87 unsigned short port;
88 signed short fastreuse;
89 struct hlist_node node;
90 struct hlist_head owners;
91};
92
0f7ff927
ACM
93#define inet_bind_bucket_for_each(tb, node, head) \
94 hlist_for_each_entry(tb, node, head, node)
1da177e4 95
0f7ff927 96struct inet_bind_hashbucket {
1da177e4
LT
97 spinlock_t lock;
98 struct hlist_head chain;
99};
100
0f7ff927 101struct inet_hashinfo {
1da177e4
LT
102 /* This is for sockets with full identity only. Sockets here will
103 * always be without wildcards and will have the following invariant:
104 *
105 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
106 *
107 * First half of the table is for sockets not in TIME_WAIT, second half
108 * is for TIME_WAIT sockets only.
109 */
0f7ff927 110 struct inet_ehash_bucket *ehash;
1da177e4
LT
111
112 /* Ok, let's try this, I give up, we do need a local binding
113 * TCP hash as well as the others for fast bind/connect.
114 */
0f7ff927 115 struct inet_bind_hashbucket *bhash;
1da177e4 116
0f7ff927
ACM
117 int bhash_size;
118 int ehash_size;
1da177e4
LT
119
120 /* All sockets in TCP_LISTEN state will be in here. This is the only
121 * table where wildcard'd TCP sockets can exist. Hash function here
122 * is just local port number.
123 */
0f7ff927 124 struct hlist_head listening_hash[INET_LHTABLE_SIZE];
1da177e4
LT
125
126 /* All the above members are written once at bootup and
127 * never written again _or_ are predominantly read-access.
128 *
129 * Now align to a new cache line as all the following members
130 * are often dirty.
131 */
0f7ff927
ACM
132 rwlock_t lhash_lock ____cacheline_aligned;
133 atomic_t lhash_users;
134 wait_queue_head_t lhash_wait;
135 spinlock_t portalloc_lock;
136};
137
138extern struct inet_hashinfo tcp_hashinfo;
139#define tcp_ehash (tcp_hashinfo.ehash)
140#define tcp_bhash (tcp_hashinfo.bhash)
141#define tcp_ehash_size (tcp_hashinfo.ehash_size)
142#define tcp_bhash_size (tcp_hashinfo.bhash_size)
143#define tcp_listening_hash (tcp_hashinfo.listening_hash)
144#define tcp_lhash_lock (tcp_hashinfo.lhash_lock)
145#define tcp_lhash_users (tcp_hashinfo.lhash_users)
146#define tcp_lhash_wait (tcp_hashinfo.lhash_wait)
147#define tcp_portalloc_lock (tcp_hashinfo.portalloc_lock)
1da177e4
LT
148
149extern kmem_cache_t *tcp_bucket_cachep;
0f7ff927
ACM
150extern struct inet_bind_bucket *
151 inet_bind_bucket_create(kmem_cache_t *cachep,
152 struct inet_bind_hashbucket *head,
153 const unsigned short snum);
154extern void inet_bind_bucket_destroy(kmem_cache_t *cachep,
155 struct inet_bind_bucket *tb);
1da177e4
LT
156extern int tcp_port_rover;
157
158/* These are AF independent. */
0f7ff927 159static inline int inet_bhashfn(const __u16 lport, const int bhash_size)
1da177e4 160{
0f7ff927 161 return lport & (bhash_size - 1);
1da177e4
LT
162}
163
0f7ff927 164extern void tcp_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
1da177e4
LT
165 unsigned short snum);
166
167#if (BITS_PER_LONG == 64)
168#define TCP_ADDRCMP_ALIGN_BYTES 8
169#else
170#define TCP_ADDRCMP_ALIGN_BYTES 4
171#endif
172
173/* This is a TIME_WAIT bucket. It works around the memory consumption
174 * problems of sockets in such a state on heavily loaded servers, but
175 * without violating the protocol specification.
176 */
177struct tcp_tw_bucket {
178 /*
179 * Now struct sock also uses sock_common, so please just
180 * don't add nothing before this first member (__tw_common) --acme
181 */
182 struct sock_common __tw_common;
183#define tw_family __tw_common.skc_family
184#define tw_state __tw_common.skc_state
185#define tw_reuse __tw_common.skc_reuse
186#define tw_bound_dev_if __tw_common.skc_bound_dev_if
187#define tw_node __tw_common.skc_node
188#define tw_bind_node __tw_common.skc_bind_node
189#define tw_refcnt __tw_common.skc_refcnt
190 volatile unsigned char tw_substate;
191 unsigned char tw_rcv_wscale;
192 __u16 tw_sport;
193 /* Socket demultiplex comparisons on incoming packets. */
194 /* these five are in inet_sock */
195 __u32 tw_daddr
196 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
197 __u32 tw_rcv_saddr;
198 __u16 tw_dport;
199 __u16 tw_num;
200 /* And these are ours. */
201 int tw_hashent;
202 int tw_timeout;
203 __u32 tw_rcv_nxt;
204 __u32 tw_snd_nxt;
205 __u32 tw_rcv_wnd;
206 __u32 tw_ts_recent;
207 long tw_ts_recent_stamp;
208 unsigned long tw_ttd;
0f7ff927 209 struct inet_bind_bucket *tw_tb;
1da177e4
LT
210 struct hlist_node tw_death_node;
211#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
212 struct in6_addr tw_v6_daddr;
213 struct in6_addr tw_v6_rcv_saddr;
214 int tw_v6_ipv6only;
215#endif
216};
217
218static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
219 struct hlist_head *list)
220{
221 hlist_add_head(&tw->tw_node, list);
222}
223
224static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
225 struct hlist_head *list)
226{
227 hlist_add_head(&tw->tw_bind_node, list);
228}
229
230static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
231{
232 return tw->tw_death_node.pprev != NULL;
233}
234
235static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
236{
237 tw->tw_death_node.pprev = NULL;
238}
239
240static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
241{
242 __hlist_del(&tw->tw_death_node);
243 tw_dead_node_init(tw);
244}
245
246static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
247{
248 if (tw_dead_hashed(tw)) {
249 __tw_del_dead_node(tw);
250 return 1;
251 }
252 return 0;
253}
254
255#define tw_for_each(tw, node, head) \
256 hlist_for_each_entry(tw, node, head, tw_node)
257
258#define tw_for_each_inmate(tw, node, jail) \
259 hlist_for_each_entry(tw, node, jail, tw_death_node)
260
261#define tw_for_each_inmate_safe(tw, node, safe, jail) \
262 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
263
264#define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
265
266static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
267{
268 return likely(sk->sk_state != TCP_TIME_WAIT) ?
269 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
270}
271
272#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
273static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
274{
275 return likely(sk->sk_state != TCP_TIME_WAIT) ?
276 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
277}
278
279static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
280{
281 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
282}
283
284#define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
285
286static inline int tcp_v6_ipv6only(const struct sock *sk)
287{
288 return likely(sk->sk_state != TCP_TIME_WAIT) ?
289 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
290}
291#else
292# define __tcp_v6_rcv_saddr(__sk) NULL
293# define tcp_v6_rcv_saddr(__sk) NULL
294# define tcptw_sk_ipv6only(__sk) 0
295# define tcp_v6_ipv6only(__sk) 0
296#endif
297
298extern kmem_cache_t *tcp_timewait_cachep;
299
300static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
301{
302 if (atomic_dec_and_test(&tw->tw_refcnt)) {
e6848976 303#ifdef SOCK_REFCNT_DEBUG
1da177e4
LT
304 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
305#endif
306 kmem_cache_free(tcp_timewait_cachep, tw);
307 }
308}
309
310extern atomic_t tcp_orphan_count;
311extern int tcp_tw_count;
312extern void tcp_time_wait(struct sock *sk, int state, int timeo);
313extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
314
315
316/* Socket demux engine toys. */
317#ifdef __BIG_ENDIAN
318#define TCP_COMBINED_PORTS(__sport, __dport) \
319 (((__u32)(__sport)<<16) | (__u32)(__dport))
320#else /* __LITTLE_ENDIAN */
321#define TCP_COMBINED_PORTS(__sport, __dport) \
322 (((__u32)(__dport)<<16) | (__u32)(__sport))
323#endif
324
325#if (BITS_PER_LONG == 64)
326#ifdef __BIG_ENDIAN
327#define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
328 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
329#else /* __LITTLE_ENDIAN */
330#define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
331 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
332#endif /* __BIG_ENDIAN */
333#define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
334 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
335 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
336 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
337#define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
338 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
339 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
340 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
341#else /* 32-bit arch */
342#define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
343#define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
344 ((inet_sk(__sk)->daddr == (__saddr)) && \
345 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
346 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
347 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
348#define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
349 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
350 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
351 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
352 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
353#endif /* 64-bit arch */
354
355#define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
356 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
357 ((__sk)->sk_family == AF_INET6) && \
358 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \
359 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
360 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
361
362/* These can have wildcards, don't try too hard. */
0f7ff927 363static inline int inet_lhashfn(const unsigned short num)
1da177e4 364{
0f7ff927 365 return num & (INET_LHTABLE_SIZE - 1);
1da177e4
LT
366}
367
0f7ff927 368static inline int inet_sk_listen_hashfn(const struct sock *sk)
1da177e4 369{
0f7ff927 370 return inet_lhashfn(inet_sk(sk)->num);
1da177e4
LT
371}
372
373#define MAX_TCP_HEADER (128 + MAX_HEADER)
374
375/*
376 * Never offer a window over 32767 without using window scaling. Some
377 * poor stacks do signed 16bit maths!
378 */
379#define MAX_TCP_WINDOW 32767U
380
381/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
382#define TCP_MIN_MSS 88U
383
384/* Minimal RCV_MSS. */
385#define TCP_MIN_RCVMSS 536U
386
387/* After receiving this amount of duplicate ACKs fast retransmit starts. */
388#define TCP_FASTRETRANS_THRESH 3
389
390/* Maximal reordering. */
391#define TCP_MAX_REORDERING 127
392
393/* Maximal number of ACKs sent quickly to accelerate slow-start. */
394#define TCP_MAX_QUICKACKS 16U
395
396/* urg_data states */
397#define TCP_URG_VALID 0x0100
398#define TCP_URG_NOTYET 0x0200
399#define TCP_URG_READ 0x0400
400
401#define TCP_RETR1 3 /*
402 * This is how many retries it does before it
403 * tries to figure out if the gateway is
404 * down. Minimal RFC value is 3; it corresponds
405 * to ~3sec-8min depending on RTO.
406 */
407
408#define TCP_RETR2 15 /*
409 * This should take at least
410 * 90 minutes to time out.
411 * RFC1122 says that the limit is 100 sec.
412 * 15 is ~13-30min depending on RTO.
413 */
414
415#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
416 * connection: ~180sec is RFC minumum */
417
418#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
419 * connection: ~180sec is RFC minumum */
420
421
422#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
423 * socket. 7 is ~50sec-16min.
424 */
425
426
427#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
428 * state, about 60 seconds */
429#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
430 /* BSD style FIN_WAIT2 deadlock breaker.
431 * It used to be 3min, new value is 60sec,
432 * to combine FIN-WAIT-2 timeout with
433 * TIME-WAIT timer.
434 */
435
436#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
437#if HZ >= 100
438#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
439#define TCP_ATO_MIN ((unsigned)(HZ/25))
440#else
441#define TCP_DELACK_MIN 4U
442#define TCP_ATO_MIN 4U
443#endif
444#define TCP_RTO_MAX ((unsigned)(120*HZ))
445#define TCP_RTO_MIN ((unsigned)(HZ/5))
446#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
447
448#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
449 * for local resources.
450 */
451
452#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
453#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
454#define TCP_KEEPALIVE_INTVL (75*HZ)
455
456#define MAX_TCP_KEEPIDLE 32767
457#define MAX_TCP_KEEPINTVL 32767
458#define MAX_TCP_KEEPCNT 127
459#define MAX_TCP_SYNCNT 127
460
461#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
462#define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
463
464#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
465#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
466 * after this time. It should be equal
467 * (or greater than) TCP_TIMEWAIT_LEN
468 * to provide reliability equal to one
469 * provided by timewait state.
470 */
471#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
472 * timestamps. It must be less than
473 * minimal timewait lifetime.
474 */
475
476#define TCP_TW_RECYCLE_SLOTS_LOG 5
477#define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
478
479/* If time > 4sec, it is "slow" path, no recycling is required,
480 so that we select tick to get range about 4 seconds.
481 */
482
483#if HZ <= 16 || HZ > 4096
484# error Unsupported: HZ <= 16 or HZ > 4096
485#elif HZ <= 32
486# define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
487#elif HZ <= 64
488# define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
489#elif HZ <= 128
490# define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
491#elif HZ <= 256
492# define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
493#elif HZ <= 512
494# define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
495#elif HZ <= 1024
496# define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
497#elif HZ <= 2048
498# define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
499#else
500# define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
501#endif
1da177e4
LT
502/*
503 * TCP option
504 */
505
506#define TCPOPT_NOP 1 /* Padding */
507#define TCPOPT_EOL 0 /* End of options */
508#define TCPOPT_MSS 2 /* Segment size negotiating */
509#define TCPOPT_WINDOW 3 /* Window scaling */
510#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
511#define TCPOPT_SACK 5 /* SACK Block */
512#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
513
514/*
515 * TCP option lengths
516 */
517
518#define TCPOLEN_MSS 4
519#define TCPOLEN_WINDOW 3
520#define TCPOLEN_SACK_PERM 2
521#define TCPOLEN_TIMESTAMP 10
522
523/* But this is what stacks really send out. */
524#define TCPOLEN_TSTAMP_ALIGNED 12
525#define TCPOLEN_WSCALE_ALIGNED 4
526#define TCPOLEN_SACKPERM_ALIGNED 4
527#define TCPOLEN_SACK_BASE 2
528#define TCPOLEN_SACK_BASE_ALIGNED 4
529#define TCPOLEN_SACK_PERBLOCK 8
530
531#define TCP_TIME_RETRANS 1 /* Retransmit timer */
532#define TCP_TIME_DACK 2 /* Delayed ack timer */
533#define TCP_TIME_PROBE0 3 /* Zero window probe timer */
534#define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
535
536/* Flags in tp->nonagle */
537#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
538#define TCP_NAGLE_CORK 2 /* Socket is corked */
539#define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
540
541/* sysctl variables for tcp */
1da177e4
LT
542extern int sysctl_tcp_timestamps;
543extern int sysctl_tcp_window_scaling;
544extern int sysctl_tcp_sack;
545extern int sysctl_tcp_fin_timeout;
546extern int sysctl_tcp_tw_recycle;
547extern int sysctl_tcp_keepalive_time;
548extern int sysctl_tcp_keepalive_probes;
549extern int sysctl_tcp_keepalive_intvl;
550extern int sysctl_tcp_syn_retries;
551extern int sysctl_tcp_synack_retries;
552extern int sysctl_tcp_retries1;
553extern int sysctl_tcp_retries2;
554extern int sysctl_tcp_orphan_retries;
555extern int sysctl_tcp_syncookies;
556extern int sysctl_tcp_retrans_collapse;
557extern int sysctl_tcp_stdurg;
558extern int sysctl_tcp_rfc1337;
559extern int sysctl_tcp_abort_on_overflow;
560extern int sysctl_tcp_max_orphans;
561extern int sysctl_tcp_max_tw_buckets;
562extern int sysctl_tcp_fack;
563extern int sysctl_tcp_reordering;
564extern int sysctl_tcp_ecn;
565extern int sysctl_tcp_dsack;
566extern int sysctl_tcp_mem[3];
567extern int sysctl_tcp_wmem[3];
568extern int sysctl_tcp_rmem[3];
569extern int sysctl_tcp_app_win;
570extern int sysctl_tcp_adv_win_scale;
571extern int sysctl_tcp_tw_reuse;
572extern int sysctl_tcp_frto;
573extern int sysctl_tcp_low_latency;
1da177e4 574extern int sysctl_tcp_nometrics_save;
1da177e4
LT
575extern int sysctl_tcp_moderate_rcvbuf;
576extern int sysctl_tcp_tso_win_divisor;
577
578extern atomic_t tcp_memory_allocated;
579extern atomic_t tcp_sockets_allocated;
580extern int tcp_memory_pressure;
581
1da177e4
LT
582#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
583#define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
584#else
585#define TCP_INET_FAMILY(fam) 1
586#endif
587
588/*
589 * Pointers to address related TCP functions
590 * (i.e. things that depend on the address family)
591 */
592
593struct tcp_func {
594 int (*queue_xmit) (struct sk_buff *skb,
595 int ipfragok);
596
597 void (*send_check) (struct sock *sk,
598 struct tcphdr *th,
599 int len,
600 struct sk_buff *skb);
601
602 int (*rebuild_header) (struct sock *sk);
603
604 int (*conn_request) (struct sock *sk,
605 struct sk_buff *skb);
606
607 struct sock * (*syn_recv_sock) (struct sock *sk,
608 struct sk_buff *skb,
60236fdd 609 struct request_sock *req,
1da177e4
LT
610 struct dst_entry *dst);
611
612 int (*remember_stamp) (struct sock *sk);
613
614 __u16 net_header_len;
615
616 int (*setsockopt) (struct sock *sk,
617 int level,
618 int optname,
619 char __user *optval,
620 int optlen);
621
622 int (*getsockopt) (struct sock *sk,
623 int level,
624 int optname,
625 char __user *optval,
626 int __user *optlen);
627
628
629 void (*addr2sockaddr) (struct sock *sk,
630 struct sockaddr *);
631
632 int sockaddr_len;
633};
634
635/*
636 * The next routines deal with comparing 32 bit unsigned ints
637 * and worry about wraparound (automatic with unsigned arithmetic).
638 */
639
640static inline int before(__u32 seq1, __u32 seq2)
641{
642 return (__s32)(seq1-seq2) < 0;
643}
644
645static inline int after(__u32 seq1, __u32 seq2)
646{
647 return (__s32)(seq2-seq1) < 0;
648}
649
650
651/* is s2<=s1<=s3 ? */
652static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
653{
654 return seq3 - seq2 >= seq1 - seq2;
655}
656
657
658extern struct proto tcp_prot;
659
660DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
661#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
662#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
663#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
664#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
665#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
666#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
667
668extern void tcp_put_port(struct sock *sk);
669extern void tcp_inherit_port(struct sock *sk, struct sock *child);
670
671extern void tcp_v4_err(struct sk_buff *skb, u32);
672
673extern void tcp_shutdown (struct sock *sk, int how);
674
675extern int tcp_v4_rcv(struct sk_buff *skb);
676
677extern int tcp_v4_remember_stamp(struct sock *sk);
678
679extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
680
681extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
682 struct msghdr *msg, size_t size);
683extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
684
685extern int tcp_ioctl(struct sock *sk,
686 int cmd,
687 unsigned long arg);
688
689extern int tcp_rcv_state_process(struct sock *sk,
690 struct sk_buff *skb,
691 struct tcphdr *th,
692 unsigned len);
693
694extern int tcp_rcv_established(struct sock *sk,
695 struct sk_buff *skb,
696 struct tcphdr *th,
697 unsigned len);
698
699extern void tcp_rcv_space_adjust(struct sock *sk);
700
701enum tcp_ack_state_t
702{
703 TCP_ACK_SCHED = 1,
704 TCP_ACK_TIMER = 2,
705 TCP_ACK_PUSHED= 4
706};
707
708static inline void tcp_schedule_ack(struct tcp_sock *tp)
709{
710 tp->ack.pending |= TCP_ACK_SCHED;
711}
712
713static inline int tcp_ack_scheduled(struct tcp_sock *tp)
714{
715 return tp->ack.pending&TCP_ACK_SCHED;
716}
717
fc6415bc 718static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp, unsigned int pkts)
1da177e4 719{
fc6415bc
DM
720 if (tp->ack.quick) {
721 if (pkts >= tp->ack.quick) {
722 tp->ack.quick = 0;
723
724 /* Leaving quickack mode we deflate ATO. */
725 tp->ack.ato = TCP_ATO_MIN;
726 } else
727 tp->ack.quick -= pkts;
1da177e4
LT
728 }
729}
730
731extern void tcp_enter_quickack_mode(struct tcp_sock *tp);
732
733static __inline__ void tcp_delack_init(struct tcp_sock *tp)
734{
735 memset(&tp->ack, 0, sizeof(tp->ack));
736}
737
738static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
739{
740 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
741}
742
743enum tcp_tw_status
744{
745 TCP_TW_SUCCESS = 0,
746 TCP_TW_RST = 1,
747 TCP_TW_ACK = 2,
748 TCP_TW_SYN = 3
749};
750
751
752extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
753 struct sk_buff *skb,
754 struct tcphdr *th,
755 unsigned len);
756
757extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM
758 struct request_sock *req,
759 struct request_sock **prev);
1da177e4
LT
760extern int tcp_child_process(struct sock *parent,
761 struct sock *child,
762 struct sk_buff *skb);
763extern void tcp_enter_frto(struct sock *sk);
764extern void tcp_enter_loss(struct sock *sk, int how);
765extern void tcp_clear_retrans(struct tcp_sock *tp);
766extern void tcp_update_metrics(struct sock *sk);
767
768extern void tcp_close(struct sock *sk,
769 long timeout);
770extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
771extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
772
773extern int tcp_getsockopt(struct sock *sk, int level,
774 int optname,
775 char __user *optval,
776 int __user *optlen);
777extern int tcp_setsockopt(struct sock *sk, int level,
778 int optname, char __user *optval,
779 int optlen);
780extern void tcp_set_keepalive(struct sock *sk, int val);
781extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
782 struct msghdr *msg,
783 size_t len, int nonblock,
784 int flags, int *addr_len);
785
786extern int tcp_listen_start(struct sock *sk);
787
788extern void tcp_parse_options(struct sk_buff *skb,
789 struct tcp_options_received *opt_rx,
790 int estab);
791
792/*
793 * TCP v4 functions exported for the inet6 API
794 */
795
1da177e4
LT
796extern void tcp_v4_send_check(struct sock *sk,
797 struct tcphdr *th, int len,
798 struct sk_buff *skb);
799
800extern int tcp_v4_conn_request(struct sock *sk,
801 struct sk_buff *skb);
802
803extern struct sock * tcp_create_openreq_child(struct sock *sk,
60236fdd 804 struct request_sock *req,
1da177e4
LT
805 struct sk_buff *skb);
806
807extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
808 struct sk_buff *skb,
60236fdd 809 struct request_sock *req,
1da177e4
LT
810 struct dst_entry *dst);
811
812extern int tcp_v4_do_rcv(struct sock *sk,
813 struct sk_buff *skb);
814
815extern int tcp_v4_connect(struct sock *sk,
816 struct sockaddr *uaddr,
817 int addr_len);
818
819extern int tcp_connect(struct sock *sk);
820
821extern struct sk_buff * tcp_make_synack(struct sock *sk,
822 struct dst_entry *dst,
60236fdd 823 struct request_sock *req);
1da177e4
LT
824
825extern int tcp_disconnect(struct sock *sk, int flags);
826
827extern void tcp_unhash(struct sock *sk);
828
829extern int tcp_v4_hash_connecting(struct sock *sk);
830
831
832/* From syncookies.c */
833extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
834 struct ip_options *opt);
835extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
836 __u16 *mss);
837
838/* tcp_output.c */
839
f6302d1d 840extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
a2e2a59c 841 unsigned int cur_mss, int nonagle);
f6302d1d 842extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
1da177e4
LT
843extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
844extern void tcp_xmit_retransmit_queue(struct sock *);
845extern void tcp_simple_retransmit(struct sock *);
846extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
847
848extern void tcp_send_probe0(struct sock *);
849extern void tcp_send_partial(struct sock *);
850extern int tcp_write_wakeup(struct sock *);
851extern void tcp_send_fin(struct sock *sk);
86a76caf
VF
852extern void tcp_send_active_reset(struct sock *sk,
853 unsigned int __nocast priority);
1da177e4 854extern int tcp_send_synack(struct sock *);
c1b4a7e6 855extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT
856extern void tcp_send_ack(struct sock *sk);
857extern void tcp_send_delayed_ack(struct sock *sk);
858
a762a980
DM
859/* tcp_input.c */
860extern void tcp_cwnd_application_limited(struct sock *sk);
861
1da177e4
LT
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);
c1b4a7e6 961 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
1da177e4
LT
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{
b03efcfb 984 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT
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
317a76f9
SH
1111/* 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);
5f8ef48d 1165extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
317a76f9 1166
5f8ef48d 1167extern struct tcp_congestion_ops tcp_init_congestion_ops;
317a76f9
SH
1168extern 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);
a8acfbac 1172extern struct tcp_congestion_ops tcp_reno;
317a76f9
SH
1173
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
1da177e4
LT
1187/* 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
1da177e4
LT
1206/* 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
d1b04c08 1228/* Set slow start threshold and cwnd not falling to slow start */
1da177e4
LT
1229static inline void __tcp_enter_cwr(struct tcp_sock *tp)
1230{
1231 tp->undo_marker = 0;
317a76f9 1232 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
1da177e4
LT
1233 tp->snd_cwnd = min(tp->snd_cwnd,
1234 tcp_packets_in_flight(tp) + 1U);
1235 tp->snd_cwnd_cnt = 0;
1236 tp->high_seq = tp->snd_nxt;
1237 tp->snd_cwnd_stamp = tcp_time_stamp;
1238 TCP_ECN_queue_cwr(tp);
1239}
1240
1241static inline void tcp_enter_cwr(struct tcp_sock *tp)
1242{
1243 tp->prior_ssthresh = 0;
1244 if (tp->ca_state < TCP_CA_CWR) {
1245 __tcp_enter_cwr(tp);
1246 tcp_set_ca_state(tp, TCP_CA_CWR);
1247 }
1248}
1249
1250extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
1251
1252/* Slow start with delack produces 3 packets of burst, so that
1253 * it is safe "de facto".
1254 */
1255static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1256{
1257 return 3;
1258}
1259
1da177e4
LT
1260static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
1261 const struct sk_buff *skb)
1262{
1263 if (skb->len < mss)
1264 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1265}
1266
1da177e4
LT
1267static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
1268{
1269 if (!tp->packets_out && !tp->pending)
1270 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1271}
1272
1da177e4
LT
1273static __inline__ void tcp_push_pending_frames(struct sock *sk,
1274 struct tcp_sock *tp)
1275{
1276 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1277}
1278
1da177e4
LT
1279static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1280{
1281 tp->snd_wl1 = seq;
1282}
1283
1284static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1285{
1286 tp->snd_wl1 = seq;
1287}
1288
1289extern void tcp_destroy_sock(struct sock *sk);
1290
1291
1292/*
1293 * Calculate(/check) TCP checksum
1294 */
1295static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1296 unsigned long saddr, unsigned long daddr,
1297 unsigned long base)
1298{
1299 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1300}
1301
1302static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1303{
1304 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1305}
1306
1307static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1308{
1309 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1310 __tcp_checksum_complete(skb);
1311}
1312
1313/* Prequeue for VJ style copy to user, combined with checksumming. */
1314
1315static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
1316{
1317 tp->ucopy.task = NULL;
1318 tp->ucopy.len = 0;
1319 tp->ucopy.memory = 0;
1320 skb_queue_head_init(&tp->ucopy.prequeue);
1321}
1322
1323/* Packet is added to VJ-style prequeue for processing in process
1324 * context, if a reader task is waiting. Apparently, this exciting
1325 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1326 * failed somewhere. Latency? Burstiness? Well, at least now we will
1327 * see, why it failed. 8)8) --ANK
1328 *
1329 * NOTE: is this not too big to inline?
1330 */
1331static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1332{
1333 struct tcp_sock *tp = tcp_sk(sk);
1334
1335 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1336 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1337 tp->ucopy.memory += skb->truesize;
1338 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1339 struct sk_buff *skb1;
1340
1341 BUG_ON(sock_owned_by_user(sk));
1342
1343 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1344 sk->sk_backlog_rcv(sk, skb1);
1345 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1346 }
1347
1348 tp->ucopy.memory = 0;
1349 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1350 wake_up_interruptible(sk->sk_sleep);
1351 if (!tcp_ack_scheduled(tp))
1352 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1353 }
1354 return 1;
1355 }
1356 return 0;
1357}
1358
1359
1360#undef STATE_TRACE
1361
1362#ifdef STATE_TRACE
1363static const char *statename[]={
1364 "Unused","Established","Syn Sent","Syn Recv",
1365 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1366 "Close Wait","Last ACK","Listen","Closing"
1367};
1368#endif
1369
1370static __inline__ void tcp_set_state(struct sock *sk, int state)
1371{
1372 int oldstate = sk->sk_state;
1373
1374 switch (state) {
1375 case TCP_ESTABLISHED:
1376 if (oldstate != TCP_ESTABLISHED)
1377 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1378 break;
1379
1380 case TCP_CLOSE:
1381 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1382 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1383
1384 sk->sk_prot->unhash(sk);
1385 if (tcp_sk(sk)->bind_hash &&
1386 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1387 tcp_put_port(sk);
1388 /* fall through */
1389 default:
1390 if (oldstate==TCP_ESTABLISHED)
1391 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1392 }
1393
1394 /* Change state AFTER socket is unhashed to avoid closed
1395 * socket sitting in hash tables.
1396 */
1397 sk->sk_state = state;
1398
1399#ifdef STATE_TRACE
1400 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1401#endif
1402}
1403
1404static __inline__ void tcp_done(struct sock *sk)
1405{
1406 tcp_set_state(sk, TCP_CLOSE);
1407 tcp_clear_xmit_timers(sk);
1408
1409 sk->sk_shutdown = SHUTDOWN_MASK;
1410
1411 if (!sock_flag(sk, SOCK_DEAD))
1412 sk->sk_state_change(sk);
1413 else
1414 tcp_destroy_sock(sk);
1415}
1416
1417static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
1418{
1419 rx_opt->dsack = 0;
1420 rx_opt->eff_sacks = 0;
1421 rx_opt->num_sacks = 0;
1422}
1423
1424static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1425{
1426 if (tp->rx_opt.tstamp_ok) {
1427 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1428 (TCPOPT_NOP << 16) |
1429 (TCPOPT_TIMESTAMP << 8) |
1430 TCPOLEN_TIMESTAMP);
1431 *ptr++ = htonl(tstamp);
1432 *ptr++ = htonl(tp->rx_opt.ts_recent);
1433 }
1434 if (tp->rx_opt.eff_sacks) {
1435 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1436 int this_sack;
1437
1438 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1439 (TCPOPT_NOP << 16) |
1440 (TCPOPT_SACK << 8) |
1441 (TCPOLEN_SACK_BASE +
1442 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1443 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1444 *ptr++ = htonl(sp[this_sack].start_seq);
1445 *ptr++ = htonl(sp[this_sack].end_seq);
1446 }
1447 if (tp->rx_opt.dsack) {
1448 tp->rx_opt.dsack = 0;
1449 tp->rx_opt.eff_sacks--;
1450 }
1451 }
1452}
1453
1454/* Construct a tcp options header for a SYN or SYN_ACK packet.
1455 * If this is every changed make sure to change the definition of
1456 * MAX_SYN_SIZE to match the new maximum number of options that you
1457 * can generate.
1458 */
1459static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1460 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1461{
1462 /* We always get an MSS option.
1463 * The option bytes which will be seen in normal data
1464 * packets should timestamps be used, must be in the MSS
1465 * advertised. But we subtract them from tp->mss_cache so
1466 * that calculations in tcp_sendmsg are simpler etc.
1467 * So account for this fact here if necessary. If we
1468 * don't do this correctly, as a receiver we won't
1469 * recognize data packets as being full sized when we
1470 * should, and thus we won't abide by the delayed ACK
1471 * rules correctly.
1472 * SACKs don't matter, we never delay an ACK when we
1473 * have any of those going out.
1474 */
1475 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1476 if (ts) {
1477 if(sack)
1478 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1479 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1480 else
1481 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1482 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1483 *ptr++ = htonl(tstamp); /* TSVAL */
1484 *ptr++ = htonl(ts_recent); /* TSECR */
1485 } else if(sack)
1486 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1487 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1488 if (offer_wscale)
1489 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1490}
1491
1492/* Determine a window scaling and initial window to offer. */
1493extern void tcp_select_initial_window(int __space, __u32 mss,
1494 __u32 *rcv_wnd, __u32 *window_clamp,
1495 int wscale_ok, __u8 *rcv_wscale);
1496
1497static inline int tcp_win_from_space(int space)
1498{
1499 return sysctl_tcp_adv_win_scale<=0 ?
1500 (space>>(-sysctl_tcp_adv_win_scale)) :
1501 space - (space>>sysctl_tcp_adv_win_scale);
1502}
1503
1504/* Note: caller must be prepared to deal with negative returns */
1505static inline int tcp_space(const struct sock *sk)
1506{
1507 return tcp_win_from_space(sk->sk_rcvbuf -
1508 atomic_read(&sk->sk_rmem_alloc));
1509}
1510
1511static inline int tcp_full_space(const struct sock *sk)
1512{
1513 return tcp_win_from_space(sk->sk_rcvbuf);
1514}
1515
60236fdd 1516static inline void tcp_acceptq_queue(struct sock *sk, struct request_sock *req,
1da177e4
LT
1517 struct sock *child)
1518{
0e87506f 1519 reqsk_queue_add(&tcp_sk(sk)->accept_queue, req, sk, child);
1da177e4
LT
1520}
1521
1da177e4 1522static inline void
60236fdd 1523tcp_synq_removed(struct sock *sk, struct request_sock *req)
1da177e4 1524{
0e87506f 1525 if (reqsk_queue_removed(&tcp_sk(sk)->accept_queue, req) == 0)
1da177e4 1526 tcp_delete_keepalive_timer(sk);
1da177e4
LT
1527}
1528
1529static inline void tcp_synq_added(struct sock *sk)
1530{
0e87506f 1531 if (reqsk_queue_added(&tcp_sk(sk)->accept_queue) == 0)
1da177e4 1532 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1da177e4
LT
1533}
1534
1535static inline int tcp_synq_len(struct sock *sk)
1536{
0e87506f 1537 return reqsk_queue_len(&tcp_sk(sk)->accept_queue);
1da177e4
LT
1538}
1539
1540static inline int tcp_synq_young(struct sock *sk)
1541{
0e87506f 1542 return reqsk_queue_len_young(&tcp_sk(sk)->accept_queue);
1da177e4
LT
1543}
1544
1545static inline int tcp_synq_is_full(struct sock *sk)
1546{
0e87506f 1547 return reqsk_queue_is_full(&tcp_sk(sk)->accept_queue);
1da177e4
LT
1548}
1549
60236fdd 1550static inline void tcp_synq_unlink(struct tcp_sock *tp, struct request_sock *req,
0e87506f 1551 struct request_sock **prev)
1da177e4 1552{
0e87506f 1553 reqsk_queue_unlink(&tp->accept_queue, req, prev);
1da177e4
LT
1554}
1555
60236fdd
ACM
1556static inline void tcp_synq_drop(struct sock *sk, struct request_sock *req,
1557 struct request_sock **prev)
1da177e4
LT
1558{
1559 tcp_synq_unlink(tcp_sk(sk), req, prev);
1560 tcp_synq_removed(sk, req);
60236fdd 1561 reqsk_free(req);
1da177e4
LT
1562}
1563
60236fdd 1564static __inline__ void tcp_openreq_init(struct request_sock *req,
1da177e4
LT
1565 struct tcp_options_received *rx_opt,
1566 struct sk_buff *skb)
1567{
2e6599cb
ACM
1568 struct inet_request_sock *ireq = inet_rsk(req);
1569
1da177e4 1570 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
2e6599cb 1571 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT
1572 req->mss = rx_opt->mss_clamp;
1573 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM
1574 ireq->tstamp_ok = rx_opt->tstamp_ok;
1575 ireq->sack_ok = rx_opt->sack_ok;
1576 ireq->snd_wscale = rx_opt->snd_wscale;
1577 ireq->wscale_ok = rx_opt->wscale_ok;
1578 ireq->acked = 0;
1579 ireq->ecn_ok = 0;
1580 ireq->rmt_port = skb->h.th->source;
1da177e4
LT
1581}
1582
1583extern void tcp_enter_memory_pressure(void);
1584
1585extern void tcp_listen_wlock(void);
1586
1587/* - We may sleep inside this lock.
1588 * - If sleeping is not required (or called from BH),
1589 * use plain read_(un)lock(&tcp_lhash_lock).
1590 */
1591
1592static inline void tcp_listen_lock(void)
1593{
1594 /* read_lock synchronizes to candidates to writers */
1595 read_lock(&tcp_lhash_lock);
1596 atomic_inc(&tcp_lhash_users);
1597 read_unlock(&tcp_lhash_lock);
1598}
1599
1600static inline void tcp_listen_unlock(void)
1601{
1602 if (atomic_dec_and_test(&tcp_lhash_users))
1603 wake_up(&tcp_lhash_wait);
1604}
1605
1606static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1607{
1608 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1609}
1610
1611static inline int keepalive_time_when(const struct tcp_sock *tp)
1612{
1613 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1614}
1615
1616static inline int tcp_fin_time(const struct tcp_sock *tp)
1617{
1618 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1619
1620 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1621 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1622
1623 return fin_timeout;
1624}
1625
1626static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1627{
1628 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1629 return 0;
1630 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1631 return 0;
1632
1633 /* RST segments are not recommended to carry timestamp,
1634 and, if they do, it is recommended to ignore PAWS because
1635 "their cleanup function should take precedence over timestamps."
1636 Certainly, it is mistake. It is necessary to understand the reasons
1637 of this constraint to relax it: if peer reboots, clock may go
1638 out-of-sync and half-open connections will not be reset.
1639 Actually, the problem would be not existing if all
1640 the implementations followed draft about maintaining clock
1641 via reboots. Linux-2.2 DOES NOT!
1642
1643 However, we can relax time bounds for RST segments to MSL.
1644 */
1645 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1646 return 0;
1647 return 1;
1648}
1649
1da177e4
LT
1650#define TCP_CHECK_TIMER(sk) do { } while (0)
1651
1652static inline int tcp_use_frto(const struct sock *sk)
1653{
1654 const struct tcp_sock *tp = tcp_sk(sk);
1655
1656 /* F-RTO must be activated in sysctl and there must be some
1657 * unsent new data, and the advertised window should allow
1658 * sending it.
1659 */
1660 return (sysctl_tcp_frto && sk->sk_send_head &&
1661 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1662 tp->snd_una + tp->snd_wnd));
1663}
1664
1665static inline void tcp_mib_init(void)
1666{
1667 /* See RFC 2012 */
1668 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1669 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1670 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1671 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1672}
1673
1674/* /proc */
1675enum tcp_seq_states {
1676 TCP_SEQ_STATE_LISTENING,
1677 TCP_SEQ_STATE_OPENREQ,
1678 TCP_SEQ_STATE_ESTABLISHED,
1679 TCP_SEQ_STATE_TIME_WAIT,
1680};
1681
1682struct tcp_seq_afinfo {
1683 struct module *owner;
1684 char *name;
1685 sa_family_t family;
1686 int (*seq_show) (struct seq_file *m, void *v);
1687 struct file_operations *seq_fops;
1688};
1689
1690struct tcp_iter_state {
1691 sa_family_t family;
1692 enum tcp_seq_states state;
1693 struct sock *syn_wait_sk;
1694 int bucket, sbucket, num, uid;
1695 struct seq_operations seq_ops;
1696};
1697
1698extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1699extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1700
1da177e4 1701#endif /* _TCP_H */