2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
22 #define FASTRETRANS_DEBUG 1
24 /* Cancel timers, when they are not required. */
25 #undef TCP_CLEAR_TIMERS
27 #include <linux/config.h>
28 #include <linux/list.h>
29 #include <linux/tcp.h>
30 #include <linux/slab.h>
31 #include <linux/cache.h>
32 #include <linux/percpu.h>
33 #include <net/checksum.h>
37 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
38 #include <linux/ipv6.h>
40 #include <linux/seq_file.h>
42 /* This is for all connections with a full identity, no wildcards.
43 * New scheme, half the table is for TIME_WAIT, the other half is
44 * for the rest. I'll experiment with dynamic table growth later.
46 struct tcp_ehash_bucket {
48 struct hlist_head chain;
49 } __attribute__((__aligned__(8)));
51 /* This is for listening sockets, thus all sockets which possess wildcards. */
52 #define TCP_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
54 /* There are a few simple rules, which allow for local port reuse by
55 * an application. In essence:
57 * 1) Sockets bound to different interfaces may share a local port.
58 * Failing that, goto test 2.
59 * 2) If all sockets have sk->sk_reuse set, and none of them are in
60 * TCP_LISTEN state, the port may be shared.
61 * Failing that, goto test 3.
62 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
63 * address, and none of them are the same, the port may be
65 * Failing this, the port cannot be shared.
67 * The interesting point, is test #2. This is what an FTP server does
68 * all day. To optimize this case we use a specific flag bit defined
69 * below. As we add sockets to a bind bucket list, we perform a
70 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
71 * As long as all sockets added to a bind bucket pass this test,
72 * the flag bit will be set.
73 * The resulting situation is that tcp_v[46]_verify_bind() can just check
74 * for this flag bit, if it is set and the socket trying to bind has
75 * sk->sk_reuse set, we don't even have to walk the owners list at all,
76 * we return that it is ok to bind this socket to the requested local port.
78 * Sounds like a lot of work, but it is worth it. In a more naive
79 * implementation (ie. current FreeBSD etc.) the entire list of ports
80 * must be walked for each data port opened by an ftp server. Needless
81 * to say, this does not scale at all. With a couple thousand FTP
82 * users logged onto your box, isn't it nice to know that new data
83 * ports are created in O(1) time? I thought so. ;-) -DaveM
85 struct tcp_bind_bucket {
87 signed short fastreuse;
88 struct hlist_node node;
89 struct hlist_head owners;
92 #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node)
94 struct tcp_bind_hashbucket {
96 struct hlist_head chain;
99 static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head)
101 return hlist_entry(head->chain.first, struct tcp_bind_bucket, node);
104 static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head)
106 return hlist_empty(&head->chain) ? NULL : __tb_head(head);
109 extern struct tcp_hashinfo {
110 /* This is for sockets with full identity only. Sockets here will
111 * always be without wildcards and will have the following invariant:
113 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
115 * First half of the table is for sockets not in TIME_WAIT, second half
116 * is for TIME_WAIT sockets only.
118 struct tcp_ehash_bucket *__tcp_ehash;
120 /* Ok, let's try this, I give up, we do need a local binding
121 * TCP hash as well as the others for fast bind/connect.
123 struct tcp_bind_hashbucket *__tcp_bhash;
125 int __tcp_bhash_size;
126 int __tcp_ehash_size;
128 /* All sockets in TCP_LISTEN state will be in here. This is the only
129 * table where wildcard'd TCP sockets can exist. Hash function here
130 * is just local port number.
132 struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE];
134 /* All the above members are written once at bootup and
135 * never written again _or_ are predominantly read-access.
137 * Now align to a new cache line as all the following members
140 rwlock_t __tcp_lhash_lock ____cacheline_aligned;
141 atomic_t __tcp_lhash_users;
142 wait_queue_head_t __tcp_lhash_wait;
143 spinlock_t __tcp_portalloc_lock;
146 #define tcp_ehash (tcp_hashinfo.__tcp_ehash)
147 #define tcp_bhash (tcp_hashinfo.__tcp_bhash)
148 #define tcp_ehash_size (tcp_hashinfo.__tcp_ehash_size)
149 #define tcp_bhash_size (tcp_hashinfo.__tcp_bhash_size)
150 #define tcp_listening_hash (tcp_hashinfo.__tcp_listening_hash)
151 #define tcp_lhash_lock (tcp_hashinfo.__tcp_lhash_lock)
152 #define tcp_lhash_users (tcp_hashinfo.__tcp_lhash_users)
153 #define tcp_lhash_wait (tcp_hashinfo.__tcp_lhash_wait)
154 #define tcp_portalloc_lock (tcp_hashinfo.__tcp_portalloc_lock)
156 /* SLAB cache for TCP socks */
157 extern kmem_cache_t *tcp_sk_cachep;
159 extern kmem_cache_t *tcp_bucket_cachep;
160 extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head,
161 unsigned short snum);
162 extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb);
163 extern void tcp_bucket_unlock(struct sock *sk);
164 extern int tcp_port_rover;
165 extern struct sock *tcp_v4_lookup_listener(u32 addr, unsigned short hnum, int dif);
167 /* These are AF independent. */
168 static __inline__ int tcp_bhashfn(__u16 lport)
170 return (lport & (tcp_bhash_size - 1));
173 extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb,
174 unsigned short snum);
176 #if (BITS_PER_LONG == 64)
177 #define TCP_ADDRCMP_ALIGN_BYTES 8
179 #define TCP_ADDRCMP_ALIGN_BYTES 4
182 /* This is a TIME_WAIT bucket. It works around the memory consumption
183 * problems of sockets in such a state on heavily loaded servers, but
184 * without violating the protocol specification.
186 struct tcp_tw_bucket {
188 * Now struct sock also uses sock_common, so please just
189 * don't add nothing before this first member (__tw_common) --acme
191 struct sock_common __tw_common;
192 #define tw_family __tw_common.skc_family
193 #define tw_state __tw_common.skc_state
194 #define tw_reuse __tw_common.skc_reuse
195 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
196 #define tw_node __tw_common.skc_node
197 #define tw_bind_node __tw_common.skc_bind_node
198 #define tw_refcnt __tw_common.skc_refcnt
199 #define tw_xid __tw_common.skc_xid
200 #define tw_vx_info __tw_common.skc_vx_info
201 #define tw_nid __tw_common.skc_nid
202 #define tw_nx_info __tw_common.skc_nx_info
203 volatile unsigned char tw_substate;
204 unsigned char tw_rcv_wscale;
206 /* Socket demultiplex comparisons on incoming packets. */
207 /* these five are in inet_opt */
209 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
213 /* And these are ours. */
220 long tw_ts_recent_stamp;
221 unsigned long tw_ttd;
222 struct tcp_bind_bucket *tw_tb;
223 struct hlist_node tw_death_node;
224 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
225 struct in6_addr tw_v6_daddr;
226 struct in6_addr tw_v6_rcv_saddr;
231 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
232 struct hlist_head *list)
234 hlist_add_head(&tw->tw_node, list);
237 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
238 struct hlist_head *list)
240 hlist_add_head(&tw->tw_bind_node, list);
243 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
245 return tw->tw_death_node.pprev != NULL;
248 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
250 tw->tw_death_node.pprev = NULL;
253 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
255 __hlist_del(&tw->tw_death_node);
256 tw_dead_node_init(tw);
259 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
261 if (tw_dead_hashed(tw)) {
262 __tw_del_dead_node(tw);
268 #define tw_for_each(tw, node, head) \
269 hlist_for_each_entry(tw, node, head, tw_node)
271 #define tw_for_each_inmate(tw, node, jail) \
272 hlist_for_each_entry(tw, node, jail, tw_death_node)
274 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
275 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
277 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
279 static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
281 return likely(sk->sk_state != TCP_TIME_WAIT) ?
282 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
285 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
286 static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
288 return likely(sk->sk_state != TCP_TIME_WAIT) ?
289 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
292 static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
294 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
297 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
299 static inline int tcp_v6_ipv6only(const struct sock *sk)
301 return likely(sk->sk_state != TCP_TIME_WAIT) ?
302 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
305 # define __tcp_v6_rcv_saddr(__sk) NULL
306 # define tcp_v6_rcv_saddr(__sk) NULL
307 # define tcptw_sk_ipv6only(__sk) 0
308 # define tcp_v6_ipv6only(__sk) 0
311 extern kmem_cache_t *tcp_timewait_cachep;
313 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
315 if (atomic_dec_and_test(&tw->tw_refcnt)) {
316 #ifdef INET_REFCNT_DEBUG
317 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
319 kmem_cache_free(tcp_timewait_cachep, tw);
323 extern atomic_t tcp_orphan_count;
324 extern int tcp_tw_count;
325 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
326 extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
327 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
330 /* Socket demux engine toys. */
332 #define TCP_COMBINED_PORTS(__sport, __dport) \
333 (((__u32)(__sport)<<16) | (__u32)(__dport))
334 #else /* __LITTLE_ENDIAN */
335 #define TCP_COMBINED_PORTS(__sport, __dport) \
336 (((__u32)(__dport)<<16) | (__u32)(__sport))
339 #if (BITS_PER_LONG == 64)
341 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
342 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
343 #else /* __LITTLE_ENDIAN */
344 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
345 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
346 #endif /* __BIG_ENDIAN */
347 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
348 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
349 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
350 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
351 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
352 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
353 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
354 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
355 #else /* 32-bit arch */
356 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
357 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
358 ((inet_sk(__sk)->daddr == (__saddr)) && \
359 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
360 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
361 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
362 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
363 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
364 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
365 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
366 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
367 #endif /* 64-bit arch */
369 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
370 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
371 ((__sk)->sk_family == AF_INET6) && \
372 !ipv6_addr_cmp(&inet6_sk(__sk)->daddr, (__saddr)) && \
373 !ipv6_addr_cmp(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
374 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
376 /* These can have wildcards, don't try too hard. */
377 static __inline__ int tcp_lhashfn(unsigned short num)
379 return num & (TCP_LHTABLE_SIZE - 1);
382 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
384 return tcp_lhashfn(inet_sk(sk)->num);
387 #define MAX_TCP_HEADER (128 + MAX_HEADER)
390 * Never offer a window over 32767 without using window scaling. Some
391 * poor stacks do signed 16bit maths!
393 #define MAX_TCP_WINDOW 32767U
395 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
396 #define TCP_MIN_MSS 88U
398 /* Minimal RCV_MSS. */
399 #define TCP_MIN_RCVMSS 536U
401 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
402 #define TCP_FASTRETRANS_THRESH 3
404 /* Maximal reordering. */
405 #define TCP_MAX_REORDERING 127
407 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
408 #define TCP_MAX_QUICKACKS 16U
410 /* urg_data states */
411 #define TCP_URG_VALID 0x0100
412 #define TCP_URG_NOTYET 0x0200
413 #define TCP_URG_READ 0x0400
415 #define TCP_RETR1 3 /*
416 * This is how many retries it does before it
417 * tries to figure out if the gateway is
418 * down. Minimal RFC value is 3; it corresponds
419 * to ~3sec-8min depending on RTO.
422 #define TCP_RETR2 15 /*
423 * This should take at least
424 * 90 minutes to time out.
425 * RFC1122 says that the limit is 100 sec.
426 * 15 is ~13-30min depending on RTO.
429 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
430 * connection: ~180sec is RFC minumum */
432 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
433 * connection: ~180sec is RFC minumum */
436 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
437 * socket. 7 is ~50sec-16min.
441 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
442 * state, about 60 seconds */
443 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
444 /* BSD style FIN_WAIT2 deadlock breaker.
445 * It used to be 3min, new value is 60sec,
446 * to combine FIN-WAIT-2 timeout with
450 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
452 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
453 #define TCP_ATO_MIN ((unsigned)(HZ/25))
455 #define TCP_DELACK_MIN 4U
456 #define TCP_ATO_MIN 4U
458 #define TCP_RTO_MAX ((unsigned)(120*HZ))
459 #define TCP_RTO_MIN ((unsigned)(HZ/5))
460 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
462 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
463 * for local resources.
466 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
467 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
468 #define TCP_KEEPALIVE_INTVL (75*HZ)
470 #define MAX_TCP_KEEPIDLE 32767
471 #define MAX_TCP_KEEPINTVL 32767
472 #define MAX_TCP_KEEPCNT 127
473 #define MAX_TCP_SYNCNT 127
475 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
476 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
478 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
479 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
480 * after this time. It should be equal
481 * (or greater than) TCP_TIMEWAIT_LEN
482 * to provide reliability equal to one
483 * provided by timewait state.
485 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
486 * timestamps. It must be less than
487 * minimal timewait lifetime.
490 #define TCP_TW_RECYCLE_SLOTS_LOG 5
491 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
493 /* If time > 4sec, it is "slow" path, no recycling is required,
494 so that we select tick to get range about 4 seconds.
497 #if HZ <= 16 || HZ > 4096
498 # error Unsupported: HZ <= 16 or HZ > 4096
500 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
502 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
504 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
506 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
508 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
510 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
512 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
514 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
517 #define BICTCP_1_OVER_BETA 8 /*
519 * multiplicative decrease factor
521 #define BICTCP_MAX_INCREMENT 32 /*
522 * Limit on the amount of
523 * increment allowed during
526 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
527 * log(B/Smin)/log(B/(B-1))+1,
531 #define BICTCP_B 4 /*
533 * go to point (max+min)/N
540 #define TCPOPT_NOP 1 /* Padding */
541 #define TCPOPT_EOL 0 /* End of options */
542 #define TCPOPT_MSS 2 /* Segment size negotiating */
543 #define TCPOPT_WINDOW 3 /* Window scaling */
544 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
545 #define TCPOPT_SACK 5 /* SACK Block */
546 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
552 #define TCPOLEN_MSS 4
553 #define TCPOLEN_WINDOW 3
554 #define TCPOLEN_SACK_PERM 2
555 #define TCPOLEN_TIMESTAMP 10
557 /* But this is what stacks really send out. */
558 #define TCPOLEN_TSTAMP_ALIGNED 12
559 #define TCPOLEN_WSCALE_ALIGNED 4
560 #define TCPOLEN_SACKPERM_ALIGNED 4
561 #define TCPOLEN_SACK_BASE 2
562 #define TCPOLEN_SACK_BASE_ALIGNED 4
563 #define TCPOLEN_SACK_PERBLOCK 8
565 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
566 #define TCP_TIME_DACK 2 /* Delayed ack timer */
567 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
568 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
570 /* Flags in tp->nonagle */
571 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
572 #define TCP_NAGLE_CORK 2 /* Socket is corked */
573 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
575 /* sysctl variables for tcp */
576 extern int sysctl_max_syn_backlog;
577 extern int sysctl_tcp_timestamps;
578 extern int sysctl_tcp_window_scaling;
579 extern int sysctl_tcp_sack;
580 extern int sysctl_tcp_fin_timeout;
581 extern int sysctl_tcp_tw_recycle;
582 extern int sysctl_tcp_keepalive_time;
583 extern int sysctl_tcp_keepalive_probes;
584 extern int sysctl_tcp_keepalive_intvl;
585 extern int sysctl_tcp_syn_retries;
586 extern int sysctl_tcp_synack_retries;
587 extern int sysctl_tcp_retries1;
588 extern int sysctl_tcp_retries2;
589 extern int sysctl_tcp_orphan_retries;
590 extern int sysctl_tcp_syncookies;
591 extern int sysctl_tcp_retrans_collapse;
592 extern int sysctl_tcp_stdurg;
593 extern int sysctl_tcp_rfc1337;
594 extern int sysctl_tcp_abort_on_overflow;
595 extern int sysctl_tcp_max_orphans;
596 extern int sysctl_tcp_max_tw_buckets;
597 extern int sysctl_tcp_fack;
598 extern int sysctl_tcp_reordering;
599 extern int sysctl_tcp_ecn;
600 extern int sysctl_tcp_dsack;
601 extern int sysctl_tcp_mem[3];
602 extern int sysctl_tcp_wmem[3];
603 extern int sysctl_tcp_rmem[3];
604 extern int sysctl_tcp_app_win;
605 extern int sysctl_tcp_adv_win_scale;
606 extern int sysctl_tcp_tw_reuse;
607 extern int sysctl_tcp_frto;
608 extern int sysctl_tcp_low_latency;
609 extern int sysctl_tcp_westwood;
610 extern int sysctl_tcp_vegas_cong_avoid;
611 extern int sysctl_tcp_vegas_alpha;
612 extern int sysctl_tcp_vegas_beta;
613 extern int sysctl_tcp_vegas_gamma;
614 extern int sysctl_tcp_nometrics_save;
615 extern int sysctl_tcp_bic;
616 extern int sysctl_tcp_bic_fast_convergence;
617 extern int sysctl_tcp_bic_low_window;
618 extern int sysctl_tcp_default_win_scale;
619 extern int sysctl_tcp_moderate_rcvbuf;
621 extern atomic_t tcp_memory_allocated;
622 extern atomic_t tcp_sockets_allocated;
623 extern int tcp_memory_pressure;
627 struct or_calltable {
629 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
630 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
631 void (*destructor) (struct open_request *req);
632 void (*send_reset) (struct sk_buff *skb);
635 struct tcp_v4_open_req {
638 struct ip_options *opt;
641 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
642 struct tcp_v6_open_req {
643 struct in6_addr loc_addr;
644 struct in6_addr rmt_addr;
645 struct sk_buff *pktopts;
650 /* this structure is too big */
651 struct open_request {
652 struct open_request *dl_next; /* Must be first member! */
659 __u16 snd_wscale : 4,
666 /* The following two fields can be easily recomputed I think -AK */
667 __u32 window_clamp; /* window clamp at creation time */
668 __u32 rcv_wnd; /* rcv_wnd offered first time */
670 unsigned long expires;
671 struct or_calltable *class;
674 struct tcp_v4_open_req v4_req;
675 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
676 struct tcp_v6_open_req v6_req;
679 #ifdef CONFIG_ACCEPT_QUEUES
680 unsigned long acceptq_time_stamp;
685 /* SLAB cache for open requests. */
686 extern kmem_cache_t *tcp_openreq_cachep;
688 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
689 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
691 static inline void tcp_openreq_free(struct open_request *req)
693 req->class->destructor(req);
694 tcp_openreq_fastfree(req);
697 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
698 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
700 #define TCP_INET_FAMILY(fam) 1
704 * Pointers to address related TCP functions
705 * (i.e. things that depend on the address family)
709 int (*queue_xmit) (struct sk_buff *skb,
712 void (*send_check) (struct sock *sk,
715 struct sk_buff *skb);
717 int (*rebuild_header) (struct sock *sk);
719 int (*conn_request) (struct sock *sk,
720 struct sk_buff *skb);
722 struct sock * (*syn_recv_sock) (struct sock *sk,
724 struct open_request *req,
725 struct dst_entry *dst);
727 int (*remember_stamp) (struct sock *sk);
729 __u16 net_header_len;
731 int (*setsockopt) (struct sock *sk,
737 int (*getsockopt) (struct sock *sk,
744 void (*addr2sockaddr) (struct sock *sk,
751 * The next routines deal with comparing 32 bit unsigned ints
752 * and worry about wraparound (automatic with unsigned arithmetic).
755 static inline int before(__u32 seq1, __u32 seq2)
757 return (__s32)(seq1-seq2) < 0;
760 static inline int after(__u32 seq1, __u32 seq2)
762 return (__s32)(seq2-seq1) < 0;
766 /* is s2<=s1<=s3 ? */
767 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
769 return seq3 - seq2 >= seq1 - seq2;
773 extern struct proto tcp_prot;
775 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
776 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
777 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
778 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
779 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
780 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
781 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
783 extern void tcp_put_port(struct sock *sk);
784 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
786 extern void tcp_v4_err(struct sk_buff *skb, u32);
788 extern void tcp_shutdown (struct sock *sk, int how);
790 extern int tcp_v4_rcv(struct sk_buff *skb);
792 extern int tcp_v4_remember_stamp(struct sock *sk);
794 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
796 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
797 struct msghdr *msg, size_t size);
798 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
800 extern int tcp_ioctl(struct sock *sk,
804 extern int tcp_rcv_state_process(struct sock *sk,
809 extern int tcp_rcv_established(struct sock *sk,
814 extern void tcp_rcv_space_adjust(struct sock *sk);
823 static inline void tcp_schedule_ack(struct tcp_opt *tp)
825 tp->ack.pending |= TCP_ACK_SCHED;
828 static inline int tcp_ack_scheduled(struct tcp_opt *tp)
830 return tp->ack.pending&TCP_ACK_SCHED;
833 static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
835 if (tp->ack.quick && --tp->ack.quick == 0) {
836 /* Leaving quickack mode we deflate ATO. */
837 tp->ack.ato = TCP_ATO_MIN;
841 extern void tcp_enter_quickack_mode(struct tcp_opt *tp);
843 static __inline__ void tcp_delack_init(struct tcp_opt *tp)
845 memset(&tp->ack, 0, sizeof(tp->ack));
848 static inline void tcp_clear_options(struct tcp_opt *tp)
850 tp->tstamp_ok = tp->sack_ok = tp->wscale_ok = tp->snd_wscale = 0;
862 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
867 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
868 struct open_request *req,
869 struct open_request **prev);
870 extern int tcp_child_process(struct sock *parent,
872 struct sk_buff *skb);
873 extern void tcp_enter_frto(struct sock *sk);
874 extern void tcp_enter_loss(struct sock *sk, int how);
875 extern void tcp_clear_retrans(struct tcp_opt *tp);
876 extern void tcp_update_metrics(struct sock *sk);
878 extern void tcp_close(struct sock *sk,
880 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
881 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
883 extern int tcp_getsockopt(struct sock *sk, int level,
887 extern int tcp_setsockopt(struct sock *sk, int level,
888 int optname, char __user *optval,
890 extern void tcp_set_keepalive(struct sock *sk, int val);
891 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
893 size_t len, int nonblock,
894 int flags, int *addr_len);
896 extern int tcp_listen_start(struct sock *sk);
898 extern void tcp_parse_options(struct sk_buff *skb,
903 * TCP v4 functions exported for the inet6 API
906 extern int tcp_v4_rebuild_header(struct sock *sk);
908 extern int tcp_v4_build_header(struct sock *sk,
909 struct sk_buff *skb);
911 extern void tcp_v4_send_check(struct sock *sk,
912 struct tcphdr *th, int len,
913 struct sk_buff *skb);
915 extern int tcp_v4_conn_request(struct sock *sk,
916 struct sk_buff *skb);
918 extern struct sock * tcp_create_openreq_child(struct sock *sk,
919 struct open_request *req,
920 struct sk_buff *skb);
922 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
924 struct open_request *req,
925 struct dst_entry *dst);
927 extern int tcp_v4_do_rcv(struct sock *sk,
928 struct sk_buff *skb);
930 extern int tcp_v4_connect(struct sock *sk,
931 struct sockaddr *uaddr,
934 extern int tcp_connect(struct sock *sk);
936 extern struct sk_buff * tcp_make_synack(struct sock *sk,
937 struct dst_entry *dst,
938 struct open_request *req);
940 extern int tcp_disconnect(struct sock *sk, int flags);
942 extern void tcp_unhash(struct sock *sk);
944 extern int tcp_v4_hash_connecting(struct sock *sk);
947 /* From syncookies.c */
948 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
949 struct ip_options *opt);
950 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
955 extern int tcp_write_xmit(struct sock *, int nonagle);
956 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
957 extern void tcp_xmit_retransmit_queue(struct sock *);
958 extern void tcp_simple_retransmit(struct sock *);
960 extern void tcp_send_probe0(struct sock *);
961 extern void tcp_send_partial(struct sock *);
962 extern int tcp_write_wakeup(struct sock *);
963 extern void tcp_send_fin(struct sock *sk);
964 extern void tcp_send_active_reset(struct sock *sk, int priority);
965 extern int tcp_send_synack(struct sock *);
966 extern int tcp_transmit_skb(struct sock *, struct sk_buff *);
967 extern void tcp_push_one(struct sock *, unsigned mss_now);
968 extern void tcp_send_ack(struct sock *sk);
969 extern void tcp_send_delayed_ack(struct sock *sk);
970 extern void cleanup_rbuf(struct sock *sk, int copied);
973 extern void tcp_init_xmit_timers(struct sock *);
974 extern void tcp_clear_xmit_timers(struct sock *);
976 extern void tcp_delete_keepalive_timer (struct sock *);
977 extern void tcp_reset_keepalive_timer (struct sock *, unsigned long);
978 extern int tcp_sync_mss(struct sock *sk, u32 pmtu);
980 extern const char timer_bug_msg[];
983 extern void tcp_get_info(struct sock *, struct tcp_info *);
985 /* Read 'sendfile()'-style from a TCP socket */
986 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
987 unsigned int, size_t);
988 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
989 sk_read_actor_t recv_actor);
991 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
993 struct tcp_opt *tp = tcp_sk(sk);
996 case TCP_TIME_RETRANS:
997 case TCP_TIME_PROBE0:
1000 #ifdef TCP_CLEAR_TIMERS
1001 sk_stop_timer(sk, &tp->retransmit_timer);
1005 tp->ack.blocked = 0;
1006 tp->ack.pending = 0;
1008 #ifdef TCP_CLEAR_TIMERS
1009 sk_stop_timer(sk, &tp->delack_timer);
1013 printk(timer_bug_msg);
1020 * Reset the retransmission timer
1022 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
1024 struct tcp_opt *tp = tcp_sk(sk);
1026 if (when > TCP_RTO_MAX) {
1028 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
1034 case TCP_TIME_RETRANS:
1035 case TCP_TIME_PROBE0:
1037 tp->timeout = jiffies+when;
1038 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
1042 tp->ack.pending |= TCP_ACK_TIMER;
1043 tp->ack.timeout = jiffies+when;
1044 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
1048 printk(timer_bug_msg);
1052 /* Compute the current effective MSS, taking SACKs and IP options,
1053 * and even PMTU discovery events into account.
1055 * LARGESEND note: !urg_mode is overkill, only frames up to snd_up
1056 * cannot be large. However, taking into account rare use of URG, this
1057 * is not a big flaw.
1060 static __inline__ unsigned int tcp_current_mss(struct sock *sk, int large)
1062 struct tcp_opt *tp = tcp_sk(sk);
1063 struct dst_entry *dst = __sk_dst_get(sk);
1064 int mss_now = large && (sk->sk_route_caps & NETIF_F_TSO) &&
1066 tp->mss_cache : tp->mss_cache_std;
1069 u32 mtu = dst_pmtu(dst);
1070 if (mtu != tp->pmtu_cookie ||
1071 tp->ext2_header_len != dst->header_len)
1072 mss_now = tcp_sync_mss(sk, mtu);
1075 mss_now -= (TCPOLEN_SACK_BASE_ALIGNED +
1076 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK));
1080 /* Initialize RCV_MSS value.
1081 * RCV_MSS is an our guess about MSS used by the peer.
1082 * We haven't any direct information about the MSS.
1083 * It's better to underestimate the RCV_MSS rather than overestimate.
1084 * Overestimations make us ACKing less frequently than needed.
1085 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1088 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1090 struct tcp_opt *tp = tcp_sk(sk);
1091 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1093 hint = min(hint, tp->rcv_wnd/2);
1094 hint = min(hint, TCP_MIN_RCVMSS);
1095 hint = max(hint, TCP_MIN_MSS);
1097 tp->ack.rcv_mss = hint;
1100 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1102 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1103 ntohl(TCP_FLAG_ACK) |
1107 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1109 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1112 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1114 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1116 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1118 tcp_fast_path_on(tp);
1121 /* Compute the actual receive window we are currently advertising.
1122 * Rcv_nxt can be after the window if our peer push more data
1123 * than the offered window.
1125 static __inline__ u32 tcp_receive_window(struct tcp_opt *tp)
1127 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1134 /* Choose a new window, without checks for shrinking, and without
1135 * scaling applied to the result. The caller does these things
1136 * if necessary. This is a "raw" window selection.
1138 extern u32 __tcp_select_window(struct sock *sk);
1140 /* TCP timestamps are only 32-bits, this causes a slight
1141 * complication on 64-bit systems since we store a snapshot
1142 * of jiffies in the buffer control blocks below. We decidely
1143 * only use of the low 32-bits of jiffies and hide the ugly
1144 * casts with the following macro.
1146 #define tcp_time_stamp ((__u32)(jiffies))
1148 /* This is what the send packet queueing engine uses to pass
1149 * TCP per-packet control information to the transmission
1150 * code. We also store the host-order sequence numbers in
1151 * here too. This is 36 bytes on 32-bit architectures,
1152 * 40 bytes on 64-bit machines, if this grows please adjust
1153 * skbuff.h:skbuff->cb[xxx] size appropriately.
1157 struct inet_skb_parm h4;
1158 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1159 struct inet6_skb_parm h6;
1161 } header; /* For incoming frames */
1162 __u32 seq; /* Starting sequence number */
1163 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1164 __u32 when; /* used to compute rtt's */
1165 __u8 flags; /* TCP header flags. */
1167 /* NOTE: These must match up to the flags byte in a
1170 #define TCPCB_FLAG_FIN 0x01
1171 #define TCPCB_FLAG_SYN 0x02
1172 #define TCPCB_FLAG_RST 0x04
1173 #define TCPCB_FLAG_PSH 0x08
1174 #define TCPCB_FLAG_ACK 0x10
1175 #define TCPCB_FLAG_URG 0x20
1176 #define TCPCB_FLAG_ECE 0x40
1177 #define TCPCB_FLAG_CWR 0x80
1179 __u8 sacked; /* State flags for SACK/FACK. */
1180 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1181 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1182 #define TCPCB_LOST 0x04 /* SKB is lost */
1183 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1185 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1186 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1188 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1190 #define TCPCB_AT_TAIL (TCPCB_URG)
1192 __u16 urg_ptr; /* Valid w/URG flags is set. */
1193 __u32 ack_seq; /* Sequence number ACK'd */
1196 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1198 #include <net/tcp_ecn.h>
1200 /* This determines how many packets are "in the network" to the best
1201 * of our knowledge. In many cases it is conservative, but where
1202 * detailed information is available from the receiver (via SACK
1203 * blocks etc.) we can make more aggressive calculations.
1205 * Use this for decisions involving congestion control, use just
1206 * tp->packets_out to determine if the send queue is empty or not.
1208 * Read this equation as:
1210 * "Packets sent once on transmission queue" MINUS
1211 * "Packets left network, but not honestly ACKed yet" PLUS
1212 * "Packets fast retransmitted"
1214 static __inline__ unsigned int tcp_packets_in_flight(struct tcp_opt *tp)
1216 return tp->packets_out - tp->left_out + tp->retrans_out;
1219 /* Recalculate snd_ssthresh, we want to set it to:
1222 * one half the current congestion window, but no
1223 * less than two segments
1226 * behave like Reno until low_window is reached,
1227 * then increase congestion window slowly
1229 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1231 if (sysctl_tcp_bic) {
1232 if (sysctl_tcp_bic_fast_convergence &&
1233 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1234 tp->bictcp.last_max_cwnd
1235 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1236 / (BICTCP_1_OVER_BETA/2);
1238 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1240 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1241 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1245 return max(tp->snd_cwnd >> 1U, 2U);
1248 /* Stop taking Vegas samples for now. */
1249 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1251 /* Is this TCP connection using Vegas (regardless of whether it is taking
1252 * Vegas measurements at the current time)?
1254 #define tcp_is_vegas(__tp) ((__tp)->vegas.do_vegas)
1256 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1258 /* There are several situations when we must "re-start" Vegas:
1260 * o when a connection is established
1262 * o after fast recovery
1263 * o when we send a packet and there is no outstanding
1264 * unacknowledged data (restarting an idle connection)
1266 * In these circumstances we cannot do a Vegas calculation at the
1267 * end of the first RTT, because any calculation we do is using
1268 * stale info -- both the saved cwnd and congestion feedback are
1271 * Instead we must wait until the completion of an RTT during
1272 * which we actually receive ACKs.
1275 /* Begin taking Vegas samples next time we send something. */
1276 tp->vegas.doing_vegas_now = 1;
1278 /* Set the beginning of the next send window. */
1279 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1281 tp->vegas.cntRTT = 0;
1282 tp->vegas.minRTT = 0x7fffffff;
1285 /* Should we be taking Vegas samples right now? */
1286 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1288 extern void tcp_vegas_init(struct tcp_opt *tp);
1290 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1292 if (tcp_is_vegas(tp)) {
1293 if (ca_state == TCP_CA_Open)
1294 tcp_vegas_enable(tp);
1296 tcp_vegas_disable(tp);
1298 tp->ca_state = ca_state;
1301 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1302 * The exception is rate halving phase, when cwnd is decreasing towards
1305 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1307 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1308 return tp->snd_ssthresh;
1310 return max(tp->snd_ssthresh,
1311 ((tp->snd_cwnd >> 1) +
1312 (tp->snd_cwnd >> 2)));
1315 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1317 if (tp->sack_ok && tp->sacked_out >= tp->packets_out - tp->lost_out)
1318 tp->sacked_out = tp->packets_out - tp->lost_out;
1319 tp->left_out = tp->sacked_out + tp->lost_out;
1322 extern void tcp_cwnd_application_limited(struct sock *sk);
1324 /* Congestion window validation. (RFC2861) */
1326 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1328 if (tp->packets_out >= tp->snd_cwnd) {
1329 /* Network is feed fully. */
1330 tp->snd_cwnd_used = 0;
1331 tp->snd_cwnd_stamp = tcp_time_stamp;
1333 /* Network starves. */
1334 if (tp->packets_out > tp->snd_cwnd_used)
1335 tp->snd_cwnd_used = tp->packets_out;
1337 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1338 tcp_cwnd_application_limited(sk);
1342 /* Set slow start threshould and cwnd not falling to slow start */
1343 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1345 tp->undo_marker = 0;
1346 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1347 tp->snd_cwnd = min(tp->snd_cwnd,
1348 tcp_packets_in_flight(tp) + 1U);
1349 tp->snd_cwnd_cnt = 0;
1350 tp->high_seq = tp->snd_nxt;
1351 tp->snd_cwnd_stamp = tcp_time_stamp;
1352 TCP_ECN_queue_cwr(tp);
1355 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1357 tp->prior_ssthresh = 0;
1358 if (tp->ca_state < TCP_CA_CWR) {
1359 __tcp_enter_cwr(tp);
1360 tcp_set_ca_state(tp, TCP_CA_CWR);
1364 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1366 /* Slow start with delack produces 3 packets of burst, so that
1367 * it is safe "de facto".
1369 static __inline__ __u32 tcp_max_burst(struct tcp_opt *tp)
1374 static __inline__ int tcp_minshall_check(struct tcp_opt *tp)
1376 return after(tp->snd_sml,tp->snd_una) &&
1377 !after(tp->snd_sml, tp->snd_nxt);
1380 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss, struct sk_buff *skb)
1383 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1386 /* Return 0, if packet can be sent now without violation Nagle's rules:
1387 1. It is full sized.
1388 2. Or it contains FIN.
1389 3. Or higher layers meant to force a packet boundary, hence the PSH bit.
1390 4. Or TCP_NODELAY was set.
1391 5. Or TCP_CORK is not set, and all sent packets are ACKed.
1392 With Minshall's modification: all sent small packets are ACKed.
1395 static __inline__ int
1396 tcp_nagle_check(struct tcp_opt *tp, struct sk_buff *skb, unsigned mss_now, int nonagle)
1398 return (skb->len < mss_now &&
1399 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1400 ((nonagle&TCP_NAGLE_CORK) ||
1403 tcp_minshall_check(tp))));
1406 /* This checks if the data bearing packet SKB (usually sk->sk_send_head)
1407 * should be put on the wire right now.
1409 static __inline__ int tcp_snd_test(struct tcp_opt *tp, struct sk_buff *skb,
1410 unsigned cur_mss, int nonagle)
1412 /* RFC 1122 - section 4.2.3.4
1416 * a) The right edge of this frame exceeds the window
1417 * b) There are packets in flight and we have a small segment
1418 * [SWS avoidance and Nagle algorithm]
1419 * (part of SWS is done on packetization)
1420 * Minshall version sounds: there are no _small_
1421 * segments in flight. (tcp_nagle_check)
1422 * c) We have too many packets 'in flight'
1424 * Don't use the nagle rule for urgent data (or
1425 * for the final FIN -DaveM).
1427 * Also, Nagle rule does not apply to frames, which
1428 * sit in the middle of queue (they have no chances
1429 * to get new data) and if room at tail of skb is
1430 * not enough to save something seriously (<32 for now).
1433 /* Don't be strict about the congestion window for the
1434 * final FIN frame. -DaveM
1436 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1437 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1438 ((tcp_packets_in_flight(tp) < tp->snd_cwnd) ||
1439 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1440 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1443 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1445 if (!tp->packets_out && !tp->pending)
1446 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1449 static __inline__ int tcp_skb_is_last(struct sock *sk, struct sk_buff *skb)
1451 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1454 /* Push out any pending frames which were held back due to
1455 * TCP_CORK or attempt at coalescing tiny packets.
1456 * The socket must be locked by the caller.
1458 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1463 struct sk_buff *skb = sk->sk_send_head;
1466 if (!tcp_skb_is_last(sk, skb))
1467 nonagle = TCP_NAGLE_PUSH;
1468 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1469 tcp_write_xmit(sk, nonagle))
1470 tcp_check_probe_timer(sk, tp);
1472 tcp_cwnd_validate(sk, tp);
1475 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1478 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1481 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1483 struct sk_buff *skb = sk->sk_send_head;
1486 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1487 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1490 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1495 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1500 extern void tcp_destroy_sock(struct sock *sk);
1504 * Calculate(/check) TCP checksum
1506 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1507 unsigned long saddr, unsigned long daddr,
1510 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1513 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1515 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1518 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1520 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1521 __tcp_checksum_complete(skb);
1524 /* Prequeue for VJ style copy to user, combined with checksumming. */
1526 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1528 tp->ucopy.task = NULL;
1530 tp->ucopy.memory = 0;
1531 skb_queue_head_init(&tp->ucopy.prequeue);
1534 /* Packet is added to VJ-style prequeue for processing in process
1535 * context, if a reader task is waiting. Apparently, this exciting
1536 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1537 * failed somewhere. Latency? Burstiness? Well, at least now we will
1538 * see, why it failed. 8)8) --ANK
1540 * NOTE: is this not too big to inline?
1542 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1544 struct tcp_opt *tp = tcp_sk(sk);
1546 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1547 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1548 tp->ucopy.memory += skb->truesize;
1549 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1550 struct sk_buff *skb1;
1552 BUG_ON(sock_owned_by_user(sk));
1554 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1555 sk->sk_backlog_rcv(sk, skb1);
1556 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1559 tp->ucopy.memory = 0;
1560 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1561 wake_up_interruptible(sk->sk_sleep);
1562 if (!tcp_ack_scheduled(tp))
1563 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1574 static char *statename[]={
1575 "Unused","Established","Syn Sent","Syn Recv",
1576 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1577 "Close Wait","Last ACK","Listen","Closing"
1581 static __inline__ void tcp_set_state(struct sock *sk, int state)
1583 int oldstate = sk->sk_state;
1586 case TCP_ESTABLISHED:
1587 if (oldstate != TCP_ESTABLISHED)
1588 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1592 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1593 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1595 sk->sk_prot->unhash(sk);
1596 if (tcp_sk(sk)->bind_hash &&
1597 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1601 if (oldstate==TCP_ESTABLISHED)
1602 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1605 /* Change state AFTER socket is unhashed to avoid closed
1606 * socket sitting in hash tables.
1608 sk->sk_state = state;
1611 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1615 static __inline__ void tcp_done(struct sock *sk)
1617 tcp_set_state(sk, TCP_CLOSE);
1618 tcp_clear_xmit_timers(sk);
1620 sk->sk_shutdown = SHUTDOWN_MASK;
1622 if (!sock_flag(sk, SOCK_DEAD))
1623 sk->sk_state_change(sk);
1625 tcp_destroy_sock(sk);
1628 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1635 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1637 if (tp->tstamp_ok) {
1638 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1639 (TCPOPT_NOP << 16) |
1640 (TCPOPT_TIMESTAMP << 8) |
1642 *ptr++ = htonl(tstamp);
1643 *ptr++ = htonl(tp->ts_recent);
1645 if (tp->eff_sacks) {
1646 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1649 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1650 (TCPOPT_NOP << 16) |
1651 (TCPOPT_SACK << 8) |
1652 (TCPOLEN_SACK_BASE +
1653 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1654 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1655 *ptr++ = htonl(sp[this_sack].start_seq);
1656 *ptr++ = htonl(sp[this_sack].end_seq);
1665 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1666 * If this is every changed make sure to change the definition of
1667 * MAX_SYN_SIZE to match the new maximum number of options that you
1670 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1671 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1673 /* We always get an MSS option.
1674 * The option bytes which will be seen in normal data
1675 * packets should timestamps be used, must be in the MSS
1676 * advertised. But we subtract them from tp->mss_cache so
1677 * that calculations in tcp_sendmsg are simpler etc.
1678 * So account for this fact here if necessary. If we
1679 * don't do this correctly, as a receiver we won't
1680 * recognize data packets as being full sized when we
1681 * should, and thus we won't abide by the delayed ACK
1683 * SACKs don't matter, we never delay an ACK when we
1684 * have any of those going out.
1686 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1689 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1690 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1692 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1693 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1694 *ptr++ = htonl(tstamp); /* TSVAL */
1695 *ptr++ = htonl(ts_recent); /* TSECR */
1697 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1698 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1700 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1703 /* Determine a window scaling and initial window to offer.
1704 * Based on the assumption that the given amount of space
1705 * will be offered. Store the results in the tp structure.
1706 * NOTE: for smooth operation initial space offering should
1707 * be a multiple of mss if possible. We assume here that mss >= 1.
1708 * This MUST be enforced by all callers.
1710 static inline void tcp_select_initial_window(int __space, __u32 mss,
1712 __u32 *window_clamp,
1716 unsigned int space = (__space < 0 ? 0 : __space);
1718 /* If no clamp set the clamp to the max possible scaled window */
1719 if (*window_clamp == 0)
1720 (*window_clamp) = (65535 << 14);
1721 space = min(*window_clamp, space);
1723 /* Quantize space offering to a multiple of mss if possible. */
1725 space = (space / mss) * mss;
1727 /* NOTE: offering an initial window larger than 32767
1728 * will break some buggy TCP stacks. We try to be nice.
1729 * If we are not window scaling, then this truncates
1730 * our initial window offering to 32k. There should also
1731 * be a sysctl option to stop being nice.
1733 (*rcv_wnd) = min(space, MAX_TCP_WINDOW);
1736 /* See RFC1323 for an explanation of the limit to 14 */
1737 while (space > 65535 && (*rcv_wscale) < 14) {
1741 if (*rcv_wscale && sysctl_tcp_app_win && space>=mss &&
1742 space - max((space>>sysctl_tcp_app_win), mss>>*rcv_wscale) < 65536/2)
1745 *rcv_wscale = max((__u8)sysctl_tcp_default_win_scale,
1749 /* Set initial window to value enough for senders,
1750 * following RFC1414. Senders, not following this RFC,
1751 * will be satisfied with 2.
1753 if (mss > (1<<*rcv_wscale)) {
1757 else if (mss > 1460)
1759 if (*rcv_wnd > init_cwnd*mss)
1760 *rcv_wnd = init_cwnd*mss;
1762 /* Set the clamp no higher than max representable value */
1763 (*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
1766 static inline int tcp_win_from_space(int space)
1768 return sysctl_tcp_adv_win_scale<=0 ?
1769 (space>>(-sysctl_tcp_adv_win_scale)) :
1770 space - (space>>sysctl_tcp_adv_win_scale);
1773 /* Note: caller must be prepared to deal with negative returns */
1774 static inline int tcp_space(struct sock *sk)
1776 return tcp_win_from_space(sk->sk_rcvbuf -
1777 atomic_read(&sk->sk_rmem_alloc));
1780 static inline int tcp_full_space( struct sock *sk)
1782 return tcp_win_from_space(sk->sk_rcvbuf);
1785 struct tcp_listen_opt
1787 u8 max_qlen_log; /* log_2 of maximal queued SYNs */
1789 #ifdef CONFIG_ACCEPT_QUEUES
1790 int qlen_young[NUM_ACCEPT_QUEUES];
1796 struct open_request *syn_table[TCP_SYNQ_HSIZE];
1799 #ifdef CONFIG_ACCEPT_QUEUES
1800 static inline void sk_acceptq_removed(struct sock *sk, int class)
1802 tcp_sk(sk)->acceptq[class].aq_backlog--;
1805 static inline void sk_acceptq_added(struct sock *sk, int class)
1807 tcp_sk(sk)->acceptq[class].aq_backlog++;
1810 static inline int sk_acceptq_is_full(struct sock *sk, int class)
1812 return tcp_sk(sk)->acceptq[class].aq_backlog >
1813 sk->sk_max_ack_backlog;
1816 static inline void tcp_set_acceptq(struct tcp_opt *tp, struct open_request *req)
1818 int class = req->acceptq_class;
1821 if (!tp->acceptq[class].aq_ratio) {
1822 req->acceptq_class = 0;
1826 tp->acceptq[class].aq_qcount++;
1827 req->acceptq_time_stamp = jiffies;
1829 if (tp->acceptq[class].aq_tail) {
1830 req->dl_next = tp->acceptq[class].aq_tail->dl_next;
1831 tp->acceptq[class].aq_tail->dl_next = req;
1832 tp->acceptq[class].aq_tail = req;
1833 } else { /* if first request in the class */
1834 tp->acceptq[class].aq_head = req;
1835 tp->acceptq[class].aq_tail = req;
1837 prev_class = class - 1;
1838 while (prev_class >= 0) {
1839 if (tp->acceptq[prev_class].aq_tail)
1843 if (prev_class < 0) {
1844 req->dl_next = tp->accept_queue;
1845 tp->accept_queue = req;
1848 req->dl_next = tp->acceptq[prev_class].aq_tail->dl_next;
1849 tp->acceptq[prev_class].aq_tail->dl_next = req;
1853 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1856 tcp_set_acceptq(tcp_sk(sk),req);
1858 sk_acceptq_added(sk,req->acceptq_class);
1862 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1865 struct tcp_opt *tp = tcp_sk(sk);
1868 sk_acceptq_added(sk);
1870 if (!tp->accept_queue_tail) {
1871 tp->accept_queue = req;
1873 tp->accept_queue_tail->dl_next = req;
1875 tp->accept_queue_tail = req;
1876 req->dl_next = NULL;
1882 #ifdef CONFIG_ACCEPT_QUEUES
1884 tcp_synq_removed(struct sock *sk, struct open_request *req)
1886 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1888 if (--lopt->qlen == 0)
1889 tcp_delete_keepalive_timer(sk);
1890 if (req->retrans == 0)
1891 lopt->qlen_young[req->acceptq_class]--;
1894 static inline void tcp_synq_added(struct sock *sk, struct open_request *req)
1896 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1898 if (lopt->qlen++ == 0)
1899 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1900 lopt->qlen_young[req->acceptq_class]++;
1903 static inline int tcp_synq_len(struct sock *sk)
1905 return tcp_sk(sk)->listen_opt->qlen;
1908 static inline int tcp_synq_young(struct sock *sk, int class)
1910 return tcp_sk(sk)->listen_opt->qlen_young[class];
1916 tcp_synq_removed(struct sock *sk, struct open_request *req)
1918 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1920 if (--lopt->qlen == 0)
1921 tcp_delete_keepalive_timer(sk);
1922 if (req->retrans == 0)
1926 static inline void tcp_synq_added(struct sock *sk)
1928 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1930 if (lopt->qlen++ == 0)
1931 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1935 static inline int tcp_synq_len(struct sock *sk)
1937 return tcp_sk(sk)->listen_opt->qlen;
1940 static inline int tcp_synq_young(struct sock *sk)
1942 return tcp_sk(sk)->listen_opt->qlen_young;
1946 static inline int tcp_synq_is_full(struct sock *sk)
1948 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1951 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1952 struct open_request **prev)
1954 write_lock(&tp->syn_wait_lock);
1955 *prev = req->dl_next;
1956 write_unlock(&tp->syn_wait_lock);
1959 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1960 struct open_request **prev)
1962 tcp_synq_unlink(tcp_sk(sk), req, prev);
1963 tcp_synq_removed(sk, req);
1964 tcp_openreq_free(req);
1967 static __inline__ void tcp_openreq_init(struct open_request *req,
1969 struct sk_buff *skb)
1971 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1972 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1973 req->mss = tp->mss_clamp;
1974 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1975 req->tstamp_ok = tp->tstamp_ok;
1976 req->sack_ok = tp->sack_ok;
1977 req->snd_wscale = tp->snd_wscale;
1978 req->wscale_ok = tp->wscale_ok;
1981 req->rmt_port = skb->h.th->source;
1984 extern void tcp_enter_memory_pressure(void);
1986 extern void tcp_listen_wlock(void);
1988 /* - We may sleep inside this lock.
1989 * - If sleeping is not required (or called from BH),
1990 * use plain read_(un)lock(&tcp_lhash_lock).
1993 static inline void tcp_listen_lock(void)
1995 /* read_lock synchronizes to candidates to writers */
1996 read_lock(&tcp_lhash_lock);
1997 atomic_inc(&tcp_lhash_users);
1998 read_unlock(&tcp_lhash_lock);
2001 static inline void tcp_listen_unlock(void)
2003 if (atomic_dec_and_test(&tcp_lhash_users))
2004 wake_up(&tcp_lhash_wait);
2007 static inline int keepalive_intvl_when(struct tcp_opt *tp)
2009 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
2012 static inline int keepalive_time_when(struct tcp_opt *tp)
2014 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
2017 static inline int tcp_fin_time(struct tcp_opt *tp)
2019 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
2021 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
2022 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
2027 static inline int tcp_paws_check(struct tcp_opt *tp, int rst)
2029 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
2031 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
2034 /* RST segments are not recommended to carry timestamp,
2035 and, if they do, it is recommended to ignore PAWS because
2036 "their cleanup function should take precedence over timestamps."
2037 Certainly, it is mistake. It is necessary to understand the reasons
2038 of this constraint to relax it: if peer reboots, clock may go
2039 out-of-sync and half-open connections will not be reset.
2040 Actually, the problem would be not existing if all
2041 the implementations followed draft about maintaining clock
2042 via reboots. Linux-2.2 DOES NOT!
2044 However, we can relax time bounds for RST segments to MSL.
2046 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
2051 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
2053 sk->sk_route_caps = dst->dev->features;
2054 if (sk->sk_route_caps & NETIF_F_TSO) {
2055 if (sk->sk_no_largesend || dst->header_len)
2056 sk->sk_route_caps &= ~NETIF_F_TSO;
2060 #define TCP_CHECK_TIMER(sk) do { } while (0)
2062 static inline int tcp_use_frto(const struct sock *sk)
2064 const struct tcp_opt *tp = tcp_sk(sk);
2066 /* F-RTO must be activated in sysctl and there must be some
2067 * unsent new data, and the advertised window should allow
2070 return (sysctl_tcp_frto && sk->sk_send_head &&
2071 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
2072 tp->snd_una + tp->snd_wnd));
2075 static inline void tcp_mib_init(void)
2078 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
2079 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
2080 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
2081 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
2085 enum tcp_seq_states {
2086 TCP_SEQ_STATE_LISTENING,
2087 TCP_SEQ_STATE_OPENREQ,
2088 TCP_SEQ_STATE_ESTABLISHED,
2089 TCP_SEQ_STATE_TIME_WAIT,
2092 struct tcp_seq_afinfo {
2093 struct module *owner;
2096 int (*seq_show) (struct seq_file *m, void *v);
2097 struct file_operations *seq_fops;
2100 struct tcp_iter_state {
2102 enum tcp_seq_states state;
2103 struct sock *syn_wait_sk;
2104 int bucket, sbucket, num, uid;
2105 struct seq_operations seq_ops;
2108 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2109 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2111 /* TCP Westwood functions and constants */
2113 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2114 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2116 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2118 if (sysctl_tcp_westwood)
2119 tp->westwood.rtt = rtt_seq;
2122 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2123 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2125 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2127 if (sysctl_tcp_westwood)
2128 __tcp_westwood_fast_bw(sk, skb);
2131 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2133 if (sysctl_tcp_westwood)
2134 __tcp_westwood_slow_bw(sk, skb);
2137 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2139 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2140 (__u32) (tp->mss_cache),
2144 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2146 return sysctl_tcp_westwood ? __tcp_westwood_bw_rttmin(tp) : 0;
2149 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2153 if (sysctl_tcp_westwood) {
2154 ssthresh = __tcp_westwood_bw_rttmin(tp);
2156 tp->snd_ssthresh = ssthresh;
2159 return (ssthresh != 0);
2162 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2166 if (sysctl_tcp_westwood) {
2167 cwnd = __tcp_westwood_bw_rttmin(tp);
2169 tp->snd_cwnd = cwnd;