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>
36 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
37 #include <linux/ipv6.h>
39 #include <linux/seq_file.h>
41 /* This is for all connections with a full identity, no wildcards.
42 * New scheme, half the table is for TIME_WAIT, the other half is
43 * for the rest. I'll experiment with dynamic table growth later.
45 struct tcp_ehash_bucket {
47 struct hlist_head chain;
48 } __attribute__((__aligned__(8)));
50 /* This is for listening sockets, thus all sockets which possess wildcards. */
51 #define TCP_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
53 /* There are a few simple rules, which allow for local port reuse by
54 * an application. In essence:
56 * 1) Sockets bound to different interfaces may share a local port.
57 * Failing that, goto test 2.
58 * 2) If all sockets have sk->sk_reuse set, and none of them are in
59 * TCP_LISTEN state, the port may be shared.
60 * Failing that, goto test 3.
61 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
62 * address, and none of them are the same, the port may be
64 * Failing this, the port cannot be shared.
66 * The interesting point, is test #2. This is what an FTP server does
67 * all day. To optimize this case we use a specific flag bit defined
68 * below. As we add sockets to a bind bucket list, we perform a
69 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
70 * As long as all sockets added to a bind bucket pass this test,
71 * the flag bit will be set.
72 * The resulting situation is that tcp_v[46]_verify_bind() can just check
73 * for this flag bit, if it is set and the socket trying to bind has
74 * sk->sk_reuse set, we don't even have to walk the owners list at all,
75 * we return that it is ok to bind this socket to the requested local port.
77 * Sounds like a lot of work, but it is worth it. In a more naive
78 * implementation (ie. current FreeBSD etc.) the entire list of ports
79 * must be walked for each data port opened by an ftp server. Needless
80 * to say, this does not scale at all. With a couple thousand FTP
81 * users logged onto your box, isn't it nice to know that new data
82 * ports are created in O(1) time? I thought so. ;-) -DaveM
84 struct tcp_bind_bucket {
86 signed short fastreuse;
87 struct hlist_node node;
88 struct hlist_head owners;
91 #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node)
93 struct tcp_bind_hashbucket {
95 struct hlist_head chain;
98 static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head)
100 return hlist_entry(head->chain.first, struct tcp_bind_bucket, node);
103 static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head)
105 return hlist_empty(&head->chain) ? NULL : __tb_head(head);
108 extern struct tcp_hashinfo {
109 /* This is for sockets with full identity only. Sockets here will
110 * always be without wildcards and will have the following invariant:
112 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
114 * First half of the table is for sockets not in TIME_WAIT, second half
115 * is for TIME_WAIT sockets only.
117 struct tcp_ehash_bucket *__tcp_ehash;
119 /* Ok, let's try this, I give up, we do need a local binding
120 * TCP hash as well as the others for fast bind/connect.
122 struct tcp_bind_hashbucket *__tcp_bhash;
124 int __tcp_bhash_size;
125 int __tcp_ehash_size;
127 /* All sockets in TCP_LISTEN state will be in here. This is the only
128 * table where wildcard'd TCP sockets can exist. Hash function here
129 * is just local port number.
131 struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE];
133 /* All the above members are written once at bootup and
134 * never written again _or_ are predominantly read-access.
136 * Now align to a new cache line as all the following members
139 rwlock_t __tcp_lhash_lock ____cacheline_aligned;
140 atomic_t __tcp_lhash_users;
141 wait_queue_head_t __tcp_lhash_wait;
142 spinlock_t __tcp_portalloc_lock;
145 #define tcp_ehash (tcp_hashinfo.__tcp_ehash)
146 #define tcp_bhash (tcp_hashinfo.__tcp_bhash)
147 #define tcp_ehash_size (tcp_hashinfo.__tcp_ehash_size)
148 #define tcp_bhash_size (tcp_hashinfo.__tcp_bhash_size)
149 #define tcp_listening_hash (tcp_hashinfo.__tcp_listening_hash)
150 #define tcp_lhash_lock (tcp_hashinfo.__tcp_lhash_lock)
151 #define tcp_lhash_users (tcp_hashinfo.__tcp_lhash_users)
152 #define tcp_lhash_wait (tcp_hashinfo.__tcp_lhash_wait)
153 #define tcp_portalloc_lock (tcp_hashinfo.__tcp_portalloc_lock)
155 /* SLAB cache for TCP socks */
156 extern kmem_cache_t *tcp_sk_cachep;
158 extern kmem_cache_t *tcp_bucket_cachep;
159 extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head,
160 unsigned short snum);
161 extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb);
162 extern void tcp_bucket_unlock(struct sock *sk);
163 extern int tcp_port_rover;
164 extern struct sock *tcp_v4_lookup_listener(u32 addr, unsigned short hnum, int dif);
166 /* These are AF independent. */
167 static __inline__ int tcp_bhashfn(__u16 lport)
169 return (lport & (tcp_bhash_size - 1));
172 extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb,
173 unsigned short snum);
175 #if (BITS_PER_LONG == 64)
176 #define TCP_ADDRCMP_ALIGN_BYTES 8
178 #define TCP_ADDRCMP_ALIGN_BYTES 4
181 /* This is a TIME_WAIT bucket. It works around the memory consumption
182 * problems of sockets in such a state on heavily loaded servers, but
183 * without violating the protocol specification.
185 struct tcp_tw_bucket {
187 * Now struct sock also uses sock_common, so please just
188 * don't add nothing before this first member (__tw_common) --acme
190 struct sock_common __tw_common;
191 #define tw_family __tw_common.skc_family
192 #define tw_state __tw_common.skc_state
193 #define tw_reuse __tw_common.skc_reuse
194 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
195 #define tw_node __tw_common.skc_node
196 #define tw_bind_node __tw_common.skc_bind_node
197 #define tw_refcnt __tw_common.skc_refcnt
198 volatile unsigned char tw_substate;
199 unsigned char tw_rcv_wscale;
201 /* Socket demultiplex comparisons on incoming packets. */
202 /* these five are in inet_opt */
204 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
208 /* And these are ours. */
215 long tw_ts_recent_stamp;
216 unsigned long tw_ttd;
217 struct tcp_bind_bucket *tw_tb;
218 struct hlist_node tw_death_node;
219 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
220 struct in6_addr tw_v6_daddr;
221 struct in6_addr tw_v6_rcv_saddr;
226 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
227 struct hlist_head *list)
229 hlist_add_head(&tw->tw_node, list);
232 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
233 struct hlist_head *list)
235 hlist_add_head(&tw->tw_bind_node, list);
238 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
240 return tw->tw_death_node.pprev != NULL;
243 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
245 tw->tw_death_node.pprev = NULL;
248 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
250 __hlist_del(&tw->tw_death_node);
251 tw_dead_node_init(tw);
254 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
256 if (tw_dead_hashed(tw)) {
257 __tw_del_dead_node(tw);
263 #define tw_for_each(tw, node, head) \
264 hlist_for_each_entry(tw, node, head, tw_node)
266 #define tw_for_each_inmate(tw, node, jail) \
267 hlist_for_each_entry(tw, node, jail, tw_death_node)
269 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
270 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
272 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
274 static inline const u32 tcp_v4_rcv_saddr(const struct sock *sk)
276 return likely(sk->sk_state != TCP_TIME_WAIT) ?
277 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
280 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
281 static inline const struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
283 return likely(sk->sk_state != TCP_TIME_WAIT) ?
284 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
287 static inline const struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
289 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
292 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
294 static inline int tcp_v6_ipv6only(const struct sock *sk)
296 return likely(sk->sk_state != TCP_TIME_WAIT) ?
297 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
300 # define __tcp_v6_rcv_saddr(__sk) NULL
301 # define tcp_v6_rcv_saddr(__sk) NULL
302 # define tcptw_sk_ipv6only(__sk) 0
303 # define tcp_v6_ipv6only(__sk) 0
306 extern kmem_cache_t *tcp_timewait_cachep;
308 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
310 if (atomic_dec_and_test(&tw->tw_refcnt)) {
311 #ifdef INET_REFCNT_DEBUG
312 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
314 kmem_cache_free(tcp_timewait_cachep, tw);
318 extern atomic_t tcp_orphan_count;
319 extern int tcp_tw_count;
320 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
321 extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
322 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
325 /* Socket demux engine toys. */
327 #define TCP_COMBINED_PORTS(__sport, __dport) \
328 (((__u32)(__sport)<<16) | (__u32)(__dport))
329 #else /* __LITTLE_ENDIAN */
330 #define TCP_COMBINED_PORTS(__sport, __dport) \
331 (((__u32)(__dport)<<16) | (__u32)(__sport))
334 #if (BITS_PER_LONG == 64)
336 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
337 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
338 #else /* __LITTLE_ENDIAN */
339 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
340 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
341 #endif /* __BIG_ENDIAN */
342 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
343 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
344 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
345 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
346 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
347 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
348 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
349 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
350 #else /* 32-bit arch */
351 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
352 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
353 ((inet_sk(__sk)->daddr == (__saddr)) && \
354 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
355 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
356 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
357 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
358 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
359 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
360 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
361 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
362 #endif /* 64-bit arch */
364 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
365 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
366 ((__sk)->sk_family == AF_INET6) && \
367 !ipv6_addr_cmp(&inet6_sk(__sk)->daddr, (__saddr)) && \
368 !ipv6_addr_cmp(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
369 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
371 /* These can have wildcards, don't try too hard. */
372 static __inline__ int tcp_lhashfn(unsigned short num)
374 return num & (TCP_LHTABLE_SIZE - 1);
377 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
379 return tcp_lhashfn(inet_sk(sk)->num);
382 #define MAX_TCP_HEADER (128 + MAX_HEADER)
385 * Never offer a window over 32767 without using window scaling. Some
386 * poor stacks do signed 16bit maths!
388 #define MAX_TCP_WINDOW 32767U
390 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
391 #define TCP_MIN_MSS 88U
393 /* Minimal RCV_MSS. */
394 #define TCP_MIN_RCVMSS 536U
396 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
397 #define TCP_FASTRETRANS_THRESH 3
399 /* Maximal reordering. */
400 #define TCP_MAX_REORDERING 127
402 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
403 #define TCP_MAX_QUICKACKS 16U
405 /* urg_data states */
406 #define TCP_URG_VALID 0x0100
407 #define TCP_URG_NOTYET 0x0200
408 #define TCP_URG_READ 0x0400
410 #define TCP_RETR1 3 /*
411 * This is how many retries it does before it
412 * tries to figure out if the gateway is
413 * down. Minimal RFC value is 3; it corresponds
414 * to ~3sec-8min depending on RTO.
417 #define TCP_RETR2 15 /*
418 * This should take at least
419 * 90 minutes to time out.
420 * RFC1122 says that the limit is 100 sec.
421 * 15 is ~13-30min depending on RTO.
424 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
425 * connection: ~180sec is RFC minumum */
427 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
428 * connection: ~180sec is RFC minumum */
431 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
432 * socket. 7 is ~50sec-16min.
436 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
437 * state, about 60 seconds */
438 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
439 /* BSD style FIN_WAIT2 deadlock breaker.
440 * It used to be 3min, new value is 60sec,
441 * to combine FIN-WAIT-2 timeout with
445 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
447 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
448 #define TCP_ATO_MIN ((unsigned)(HZ/25))
450 #define TCP_DELACK_MIN 4U
451 #define TCP_ATO_MIN 4U
453 #define TCP_RTO_MAX ((unsigned)(120*HZ))
454 #define TCP_RTO_MIN ((unsigned)(HZ/5))
455 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
457 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
458 * for local resources.
461 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
462 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
463 #define TCP_KEEPALIVE_INTVL (75*HZ)
465 #define MAX_TCP_KEEPIDLE 32767
466 #define MAX_TCP_KEEPINTVL 32767
467 #define MAX_TCP_KEEPCNT 127
468 #define MAX_TCP_SYNCNT 127
470 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
471 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
473 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
474 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
475 * after this time. It should be equal
476 * (or greater than) TCP_TIMEWAIT_LEN
477 * to provide reliability equal to one
478 * provided by timewait state.
480 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
481 * timestamps. It must be less than
482 * minimal timewait lifetime.
485 #define TCP_TW_RECYCLE_SLOTS_LOG 5
486 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
488 /* If time > 4sec, it is "slow" path, no recycling is required,
489 so that we select tick to get range about 4 seconds.
492 #if HZ <= 16 || HZ > 4096
493 # error Unsupported: HZ <= 16 or HZ > 4096
495 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
497 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
499 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
501 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
503 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
505 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
507 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
509 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
512 #define BICTCP_1_OVER_BETA 8 /*
514 * multiplicative decrease factor
516 #define BICTCP_MAX_INCREMENT 32 /*
517 * Limit on the amount of
518 * increment allowed during
521 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
522 * log(B/Smin)/log(B/(B-1))+1,
526 #define BICTCP_B 4 /*
528 * go to point (max+min)/N
535 #define TCPOPT_NOP 1 /* Padding */
536 #define TCPOPT_EOL 0 /* End of options */
537 #define TCPOPT_MSS 2 /* Segment size negotiating */
538 #define TCPOPT_WINDOW 3 /* Window scaling */
539 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
540 #define TCPOPT_SACK 5 /* SACK Block */
541 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
547 #define TCPOLEN_MSS 4
548 #define TCPOLEN_WINDOW 3
549 #define TCPOLEN_SACK_PERM 2
550 #define TCPOLEN_TIMESTAMP 10
552 /* But this is what stacks really send out. */
553 #define TCPOLEN_TSTAMP_ALIGNED 12
554 #define TCPOLEN_WSCALE_ALIGNED 4
555 #define TCPOLEN_SACKPERM_ALIGNED 4
556 #define TCPOLEN_SACK_BASE 2
557 #define TCPOLEN_SACK_BASE_ALIGNED 4
558 #define TCPOLEN_SACK_PERBLOCK 8
560 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
561 #define TCP_TIME_DACK 2 /* Delayed ack timer */
562 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
563 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
565 /* Flags in tp->nonagle */
566 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
567 #define TCP_NAGLE_CORK 2 /* Socket is corked */
568 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
570 /* sysctl variables for tcp */
571 extern int sysctl_max_syn_backlog;
572 extern int sysctl_tcp_timestamps;
573 extern int sysctl_tcp_window_scaling;
574 extern int sysctl_tcp_sack;
575 extern int sysctl_tcp_fin_timeout;
576 extern int sysctl_tcp_tw_recycle;
577 extern int sysctl_tcp_keepalive_time;
578 extern int sysctl_tcp_keepalive_probes;
579 extern int sysctl_tcp_keepalive_intvl;
580 extern int sysctl_tcp_syn_retries;
581 extern int sysctl_tcp_synack_retries;
582 extern int sysctl_tcp_retries1;
583 extern int sysctl_tcp_retries2;
584 extern int sysctl_tcp_orphan_retries;
585 extern int sysctl_tcp_syncookies;
586 extern int sysctl_tcp_retrans_collapse;
587 extern int sysctl_tcp_stdurg;
588 extern int sysctl_tcp_rfc1337;
589 extern int sysctl_tcp_abort_on_overflow;
590 extern int sysctl_tcp_max_orphans;
591 extern int sysctl_tcp_max_tw_buckets;
592 extern int sysctl_tcp_fack;
593 extern int sysctl_tcp_reordering;
594 extern int sysctl_tcp_ecn;
595 extern int sysctl_tcp_dsack;
596 extern int sysctl_tcp_mem[3];
597 extern int sysctl_tcp_wmem[3];
598 extern int sysctl_tcp_rmem[3];
599 extern int sysctl_tcp_app_win;
600 extern int sysctl_tcp_adv_win_scale;
601 extern int sysctl_tcp_tw_reuse;
602 extern int sysctl_tcp_frto;
603 extern int sysctl_tcp_low_latency;
604 extern int sysctl_tcp_westwood;
605 extern int sysctl_tcp_vegas_cong_avoid;
606 extern int sysctl_tcp_vegas_alpha;
607 extern int sysctl_tcp_vegas_beta;
608 extern int sysctl_tcp_vegas_gamma;
609 extern int sysctl_tcp_nometrics_save;
610 extern int sysctl_tcp_bic;
611 extern int sysctl_tcp_bic_fast_convergence;
612 extern int sysctl_tcp_bic_low_window;
613 extern int sysctl_tcp_default_win_scale;
614 extern int sysctl_tcp_moderate_rcvbuf;
616 extern atomic_t tcp_memory_allocated;
617 extern atomic_t tcp_sockets_allocated;
618 extern int tcp_memory_pressure;
622 struct or_calltable {
624 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
625 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
626 void (*destructor) (struct open_request *req);
627 void (*send_reset) (struct sk_buff *skb);
630 struct tcp_v4_open_req {
633 struct ip_options *opt;
636 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
637 struct tcp_v6_open_req {
638 struct in6_addr loc_addr;
639 struct in6_addr rmt_addr;
640 struct sk_buff *pktopts;
645 /* this structure is too big */
646 struct open_request {
647 struct open_request *dl_next; /* Must be first member! */
654 __u16 snd_wscale : 4,
661 /* The following two fields can be easily recomputed I think -AK */
662 __u32 window_clamp; /* window clamp at creation time */
663 __u32 rcv_wnd; /* rcv_wnd offered first time */
665 unsigned long expires;
666 struct or_calltable *class;
669 struct tcp_v4_open_req v4_req;
670 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
671 struct tcp_v6_open_req v6_req;
676 /* SLAB cache for open requests. */
677 extern kmem_cache_t *tcp_openreq_cachep;
679 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
680 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
682 static inline void tcp_openreq_free(struct open_request *req)
684 req->class->destructor(req);
685 tcp_openreq_fastfree(req);
688 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
689 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
691 #define TCP_INET_FAMILY(fam) 1
695 * Pointers to address related TCP functions
696 * (i.e. things that depend on the address family)
700 int (*queue_xmit) (struct sk_buff *skb,
703 void (*send_check) (struct sock *sk,
706 struct sk_buff *skb);
708 int (*rebuild_header) (struct sock *sk);
710 int (*conn_request) (struct sock *sk,
711 struct sk_buff *skb);
713 struct sock * (*syn_recv_sock) (struct sock *sk,
715 struct open_request *req,
716 struct dst_entry *dst);
718 int (*remember_stamp) (struct sock *sk);
720 __u16 net_header_len;
722 int (*setsockopt) (struct sock *sk,
728 int (*getsockopt) (struct sock *sk,
735 void (*addr2sockaddr) (struct sock *sk,
742 * The next routines deal with comparing 32 bit unsigned ints
743 * and worry about wraparound (automatic with unsigned arithmetic).
746 static inline int before(__u32 seq1, __u32 seq2)
748 return (__s32)(seq1-seq2) < 0;
751 static inline int after(__u32 seq1, __u32 seq2)
753 return (__s32)(seq2-seq1) < 0;
757 /* is s2<=s1<=s3 ? */
758 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
760 return seq3 - seq2 >= seq1 - seq2;
764 extern struct proto tcp_prot;
766 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
767 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
768 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
769 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
770 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
771 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
772 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
774 extern void tcp_put_port(struct sock *sk);
775 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
777 extern void tcp_v4_err(struct sk_buff *skb, u32);
779 extern void tcp_shutdown (struct sock *sk, int how);
781 extern int tcp_v4_rcv(struct sk_buff *skb);
783 extern int tcp_v4_remember_stamp(struct sock *sk);
785 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
787 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
788 struct msghdr *msg, size_t size);
789 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
791 extern int tcp_ioctl(struct sock *sk,
795 extern int tcp_rcv_state_process(struct sock *sk,
800 extern int tcp_rcv_established(struct sock *sk,
805 extern void tcp_rcv_space_adjust(struct sock *sk);
814 static inline void tcp_schedule_ack(struct tcp_opt *tp)
816 tp->ack.pending |= TCP_ACK_SCHED;
819 static inline int tcp_ack_scheduled(struct tcp_opt *tp)
821 return tp->ack.pending&TCP_ACK_SCHED;
824 static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
826 if (tp->ack.quick && --tp->ack.quick == 0) {
827 /* Leaving quickack mode we deflate ATO. */
828 tp->ack.ato = TCP_ATO_MIN;
832 extern void tcp_enter_quickack_mode(struct tcp_opt *tp);
834 static __inline__ void tcp_delack_init(struct tcp_opt *tp)
836 memset(&tp->ack, 0, sizeof(tp->ack));
839 static inline void tcp_clear_options(struct tcp_opt *tp)
841 tp->tstamp_ok = tp->sack_ok = tp->wscale_ok = tp->snd_wscale = 0;
853 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
858 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
859 struct open_request *req,
860 struct open_request **prev);
861 extern int tcp_child_process(struct sock *parent,
863 struct sk_buff *skb);
864 extern void tcp_enter_frto(struct sock *sk);
865 extern void tcp_enter_loss(struct sock *sk, int how);
866 extern void tcp_clear_retrans(struct tcp_opt *tp);
867 extern void tcp_update_metrics(struct sock *sk);
869 extern void tcp_close(struct sock *sk,
871 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
872 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
873 extern void tcp_write_space(struct sock *sk);
875 extern int tcp_getsockopt(struct sock *sk, int level,
879 extern int tcp_setsockopt(struct sock *sk, int level,
880 int optname, char __user *optval,
882 extern void tcp_set_keepalive(struct sock *sk, int val);
883 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
885 size_t len, int nonblock,
886 int flags, int *addr_len);
888 extern int tcp_listen_start(struct sock *sk);
890 extern void tcp_parse_options(struct sk_buff *skb,
895 * TCP v4 functions exported for the inet6 API
898 extern int tcp_v4_rebuild_header(struct sock *sk);
900 extern int tcp_v4_build_header(struct sock *sk,
901 struct sk_buff *skb);
903 extern void tcp_v4_send_check(struct sock *sk,
904 struct tcphdr *th, int len,
905 struct sk_buff *skb);
907 extern int tcp_v4_conn_request(struct sock *sk,
908 struct sk_buff *skb);
910 extern struct sock * tcp_create_openreq_child(struct sock *sk,
911 struct open_request *req,
912 struct sk_buff *skb);
914 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
916 struct open_request *req,
917 struct dst_entry *dst);
919 extern int tcp_v4_do_rcv(struct sock *sk,
920 struct sk_buff *skb);
922 extern int tcp_v4_connect(struct sock *sk,
923 struct sockaddr *uaddr,
926 extern int tcp_connect(struct sock *sk);
928 extern struct sk_buff * tcp_make_synack(struct sock *sk,
929 struct dst_entry *dst,
930 struct open_request *req);
932 extern int tcp_disconnect(struct sock *sk, int flags);
934 extern void tcp_unhash(struct sock *sk);
936 extern int tcp_v4_hash_connecting(struct sock *sk);
939 /* From syncookies.c */
940 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
941 struct ip_options *opt);
942 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
947 extern int tcp_write_xmit(struct sock *, int nonagle);
948 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
949 extern void tcp_xmit_retransmit_queue(struct sock *);
950 extern void tcp_simple_retransmit(struct sock *);
952 extern void tcp_send_probe0(struct sock *);
953 extern void tcp_send_partial(struct sock *);
954 extern int tcp_write_wakeup(struct sock *);
955 extern void tcp_send_fin(struct sock *sk);
956 extern void tcp_send_active_reset(struct sock *sk, int priority);
957 extern int tcp_send_synack(struct sock *);
958 extern int tcp_transmit_skb(struct sock *, struct sk_buff *);
959 extern void tcp_push_one(struct sock *, unsigned mss_now);
960 extern void tcp_send_ack(struct sock *sk);
961 extern void tcp_send_delayed_ack(struct sock *sk);
964 extern void tcp_init_xmit_timers(struct sock *);
965 extern void tcp_clear_xmit_timers(struct sock *);
967 extern void tcp_delete_keepalive_timer (struct sock *);
968 extern void tcp_reset_keepalive_timer (struct sock *, unsigned long);
969 extern int tcp_sync_mss(struct sock *sk, u32 pmtu);
971 extern const char timer_bug_msg[];
974 extern void tcp_get_info(struct sock *, struct tcp_info *);
976 /* Read 'sendfile()'-style from a TCP socket */
977 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
978 unsigned int, size_t);
979 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
980 sk_read_actor_t recv_actor);
982 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
984 struct tcp_opt *tp = tcp_sk(sk);
987 case TCP_TIME_RETRANS:
988 case TCP_TIME_PROBE0:
991 #ifdef TCP_CLEAR_TIMERS
992 sk_stop_timer(sk, &tp->retransmit_timer);
999 #ifdef TCP_CLEAR_TIMERS
1000 sk_stop_timer(sk, &tp->delack_timer);
1004 printk(timer_bug_msg);
1011 * Reset the retransmission timer
1013 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
1015 struct tcp_opt *tp = tcp_sk(sk);
1017 if (when > TCP_RTO_MAX) {
1019 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
1025 case TCP_TIME_RETRANS:
1026 case TCP_TIME_PROBE0:
1028 tp->timeout = jiffies+when;
1029 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
1033 tp->ack.pending |= TCP_ACK_TIMER;
1034 tp->ack.timeout = jiffies+when;
1035 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
1039 printk(KERN_DEBUG "bug: unknown timer value\n");
1043 /* Compute the current effective MSS, taking SACKs and IP options,
1044 * and even PMTU discovery events into account.
1046 * LARGESEND note: !urg_mode is overkill, only frames up to snd_up
1047 * cannot be large. However, taking into account rare use of URG, this
1048 * is not a big flaw.
1051 static __inline__ unsigned int tcp_current_mss(struct sock *sk, int large)
1053 struct tcp_opt *tp = tcp_sk(sk);
1054 struct dst_entry *dst = __sk_dst_get(sk);
1055 int mss_now = large && (sk->sk_route_caps & NETIF_F_TSO) &&
1057 tp->mss_cache : tp->mss_cache_std;
1060 u32 mtu = dst_pmtu(dst);
1061 if (mtu != tp->pmtu_cookie ||
1062 tp->ext2_header_len != dst->header_len)
1063 mss_now = tcp_sync_mss(sk, mtu);
1066 mss_now -= (TCPOLEN_SACK_BASE_ALIGNED +
1067 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK));
1071 /* Initialize RCV_MSS value.
1072 * RCV_MSS is an our guess about MSS used by the peer.
1073 * We haven't any direct information about the MSS.
1074 * It's better to underestimate the RCV_MSS rather than overestimate.
1075 * Overestimations make us ACKing less frequently than needed.
1076 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1079 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1081 struct tcp_opt *tp = tcp_sk(sk);
1082 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1084 hint = min(hint, tp->rcv_wnd/2);
1085 hint = min(hint, TCP_MIN_RCVMSS);
1086 hint = max(hint, TCP_MIN_MSS);
1088 tp->ack.rcv_mss = hint;
1091 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1093 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1094 ntohl(TCP_FLAG_ACK) |
1098 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1100 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1103 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1105 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1107 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1109 tcp_fast_path_on(tp);
1112 /* Compute the actual receive window we are currently advertising.
1113 * Rcv_nxt can be after the window if our peer push more data
1114 * than the offered window.
1116 static __inline__ u32 tcp_receive_window(struct tcp_opt *tp)
1118 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1125 /* Choose a new window, without checks for shrinking, and without
1126 * scaling applied to the result. The caller does these things
1127 * if necessary. This is a "raw" window selection.
1129 extern u32 __tcp_select_window(struct sock *sk);
1131 /* TCP timestamps are only 32-bits, this causes a slight
1132 * complication on 64-bit systems since we store a snapshot
1133 * of jiffies in the buffer control blocks below. We decidely
1134 * only use of the low 32-bits of jiffies and hide the ugly
1135 * casts with the following macro.
1137 #define tcp_time_stamp ((__u32)(jiffies))
1139 /* This is what the send packet queueing engine uses to pass
1140 * TCP per-packet control information to the transmission
1141 * code. We also store the host-order sequence numbers in
1142 * here too. This is 36 bytes on 32-bit architectures,
1143 * 40 bytes on 64-bit machines, if this grows please adjust
1144 * skbuff.h:skbuff->cb[xxx] size appropriately.
1148 struct inet_skb_parm h4;
1149 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1150 struct inet6_skb_parm h6;
1152 } header; /* For incoming frames */
1153 __u32 seq; /* Starting sequence number */
1154 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1155 __u32 when; /* used to compute rtt's */
1156 __u8 flags; /* TCP header flags. */
1158 /* NOTE: These must match up to the flags byte in a
1161 #define TCPCB_FLAG_FIN 0x01
1162 #define TCPCB_FLAG_SYN 0x02
1163 #define TCPCB_FLAG_RST 0x04
1164 #define TCPCB_FLAG_PSH 0x08
1165 #define TCPCB_FLAG_ACK 0x10
1166 #define TCPCB_FLAG_URG 0x20
1167 #define TCPCB_FLAG_ECE 0x40
1168 #define TCPCB_FLAG_CWR 0x80
1170 __u8 sacked; /* State flags for SACK/FACK. */
1171 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1172 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1173 #define TCPCB_LOST 0x04 /* SKB is lost */
1174 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1176 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1177 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1179 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1181 #define TCPCB_AT_TAIL (TCPCB_URG)
1183 __u16 urg_ptr; /* Valid w/URG flags is set. */
1184 __u32 ack_seq; /* Sequence number ACK'd */
1187 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1189 #define for_retrans_queue(skb, sk, tp) \
1190 for (skb = (sk)->sk_write_queue.next; \
1191 (skb != (tp)->send_head) && \
1192 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1196 #include <net/tcp_ecn.h>
1200 * Compute minimal free write space needed to queue new packets.
1202 static inline int tcp_min_write_space(struct sock *sk)
1204 return sk->sk_wmem_queued / 2;
1207 static inline int tcp_wspace(struct sock *sk)
1209 return sk->sk_sndbuf - sk->sk_wmem_queued;
1213 /* This determines how many packets are "in the network" to the best
1214 * of our knowledge. In many cases it is conservative, but where
1215 * detailed information is available from the receiver (via SACK
1216 * blocks etc.) we can make more aggressive calculations.
1218 * Use this for decisions involving congestion control, use just
1219 * tp->packets_out to determine if the send queue is empty or not.
1221 * Read this equation as:
1223 * "Packets sent once on transmission queue" MINUS
1224 * "Packets left network, but not honestly ACKed yet" PLUS
1225 * "Packets fast retransmitted"
1227 static __inline__ unsigned int tcp_packets_in_flight(struct tcp_opt *tp)
1229 return tp->packets_out - tp->left_out + tp->retrans_out;
1232 /* Recalculate snd_ssthresh, we want to set it to:
1235 * one half the current congestion window, but no
1236 * less than two segments
1239 * behave like Reno until low_window is reached,
1240 * then increase congestion window slowly
1242 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1244 if (sysctl_tcp_bic) {
1245 if (sysctl_tcp_bic_fast_convergence &&
1246 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1247 tp->bictcp.last_max_cwnd
1248 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1249 / (BICTCP_1_OVER_BETA/2);
1251 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1253 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1254 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1258 return max(tp->snd_cwnd >> 1U, 2U);
1261 /* Stop taking Vegas samples for now. */
1262 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1264 /* Is this TCP connection using Vegas (regardless of whether it is taking
1265 * Vegas measurements at the current time)?
1267 #define tcp_is_vegas(__tp) ((__tp)->vegas.do_vegas)
1269 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1271 /* There are several situations when we must "re-start" Vegas:
1273 * o when a connection is established
1275 * o after fast recovery
1276 * o when we send a packet and there is no outstanding
1277 * unacknowledged data (restarting an idle connection)
1279 * In these circumstances we cannot do a Vegas calculation at the
1280 * end of the first RTT, because any calculation we do is using
1281 * stale info -- both the saved cwnd and congestion feedback are
1284 * Instead we must wait until the completion of an RTT during
1285 * which we actually receive ACKs.
1288 /* Begin taking Vegas samples next time we send something. */
1289 tp->vegas.doing_vegas_now = 1;
1291 /* Set the beginning of the next send window. */
1292 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1294 tp->vegas.cntRTT = 0;
1295 tp->vegas.minRTT = 0x7fffffff;
1298 /* Should we be taking Vegas samples right now? */
1299 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1301 extern void tcp_vegas_init(struct tcp_opt *tp);
1303 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1305 if (tcp_is_vegas(tp)) {
1306 if (ca_state == TCP_CA_Open)
1307 tcp_vegas_enable(tp);
1309 tcp_vegas_disable(tp);
1311 tp->ca_state = ca_state;
1314 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1315 * The exception is rate halving phase, when cwnd is decreasing towards
1318 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1320 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1321 return tp->snd_ssthresh;
1323 return max(tp->snd_ssthresh,
1324 ((tp->snd_cwnd >> 1) +
1325 (tp->snd_cwnd >> 2)));
1328 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1330 if (tp->sack_ok && tp->sacked_out >= tp->packets_out - tp->lost_out)
1331 tp->sacked_out = tp->packets_out - tp->lost_out;
1332 tp->left_out = tp->sacked_out + tp->lost_out;
1335 extern void tcp_cwnd_application_limited(struct sock *sk);
1337 /* Congestion window validation. (RFC2861) */
1339 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1341 if (tp->packets_out >= tp->snd_cwnd) {
1342 /* Network is feed fully. */
1343 tp->snd_cwnd_used = 0;
1344 tp->snd_cwnd_stamp = tcp_time_stamp;
1346 /* Network starves. */
1347 if (tp->packets_out > tp->snd_cwnd_used)
1348 tp->snd_cwnd_used = tp->packets_out;
1350 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1351 tcp_cwnd_application_limited(sk);
1355 /* Set slow start threshould and cwnd not falling to slow start */
1356 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1358 tp->undo_marker = 0;
1359 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1360 tp->snd_cwnd = min(tp->snd_cwnd,
1361 tcp_packets_in_flight(tp) + 1U);
1362 tp->snd_cwnd_cnt = 0;
1363 tp->high_seq = tp->snd_nxt;
1364 tp->snd_cwnd_stamp = tcp_time_stamp;
1365 TCP_ECN_queue_cwr(tp);
1368 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1370 tp->prior_ssthresh = 0;
1371 if (tp->ca_state < TCP_CA_CWR) {
1372 __tcp_enter_cwr(tp);
1373 tcp_set_ca_state(tp, TCP_CA_CWR);
1377 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1379 /* Slow start with delack produces 3 packets of burst, so that
1380 * it is safe "de facto".
1382 static __inline__ __u32 tcp_max_burst(struct tcp_opt *tp)
1387 static __inline__ int tcp_minshall_check(struct tcp_opt *tp)
1389 return after(tp->snd_sml,tp->snd_una) &&
1390 !after(tp->snd_sml, tp->snd_nxt);
1393 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss, struct sk_buff *skb)
1396 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1399 /* Return 0, if packet can be sent now without violation Nagle's rules:
1400 1. It is full sized.
1401 2. Or it contains FIN.
1402 3. Or TCP_NODELAY was set.
1403 4. Or TCP_CORK is not set, and all sent packets are ACKed.
1404 With Minshall's modification: all sent small packets are ACKed.
1407 static __inline__ int
1408 tcp_nagle_check(struct tcp_opt *tp, struct sk_buff *skb, unsigned mss_now, int nonagle)
1410 return (skb->len < mss_now &&
1411 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1412 ((nonagle&TCP_NAGLE_CORK) ||
1415 tcp_minshall_check(tp))));
1418 /* This checks if the data bearing packet SKB (usually tp->send_head)
1419 * should be put on the wire right now.
1421 static __inline__ int tcp_snd_test(struct tcp_opt *tp, struct sk_buff *skb,
1422 unsigned cur_mss, int nonagle)
1424 /* RFC 1122 - section 4.2.3.4
1428 * a) The right edge of this frame exceeds the window
1429 * b) There are packets in flight and we have a small segment
1430 * [SWS avoidance and Nagle algorithm]
1431 * (part of SWS is done on packetization)
1432 * Minshall version sounds: there are no _small_
1433 * segments in flight. (tcp_nagle_check)
1434 * c) We have too many packets 'in flight'
1436 * Don't use the nagle rule for urgent data (or
1437 * for the final FIN -DaveM).
1439 * Also, Nagle rule does not apply to frames, which
1440 * sit in the middle of queue (they have no chances
1441 * to get new data) and if room at tail of skb is
1442 * not enough to save something seriously (<32 for now).
1445 /* Don't be strict about the congestion window for the
1446 * final FIN frame. -DaveM
1448 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1449 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1450 ((tcp_packets_in_flight(tp) < tp->snd_cwnd) ||
1451 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1452 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1455 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1457 if (!tp->packets_out && !tp->pending)
1458 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1461 static __inline__ int tcp_skb_is_last(struct sock *sk, struct sk_buff *skb)
1463 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1466 /* Push out any pending frames which were held back due to
1467 * TCP_CORK or attempt at coalescing tiny packets.
1468 * The socket must be locked by the caller.
1470 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1475 struct sk_buff *skb = tp->send_head;
1478 if (!tcp_skb_is_last(sk, skb))
1479 nonagle = TCP_NAGLE_PUSH;
1480 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1481 tcp_write_xmit(sk, nonagle))
1482 tcp_check_probe_timer(sk, tp);
1484 tcp_cwnd_validate(sk, tp);
1487 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1490 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1493 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1495 struct sk_buff *skb = tp->send_head;
1498 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1499 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1502 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1507 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1512 extern void tcp_destroy_sock(struct sock *sk);
1516 * Calculate(/check) TCP checksum
1518 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1519 unsigned long saddr, unsigned long daddr,
1522 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1525 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1527 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1530 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1532 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1533 __tcp_checksum_complete(skb);
1536 /* Prequeue for VJ style copy to user, combined with checksumming. */
1538 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1540 tp->ucopy.task = NULL;
1542 tp->ucopy.memory = 0;
1543 skb_queue_head_init(&tp->ucopy.prequeue);
1546 /* Packet is added to VJ-style prequeue for processing in process
1547 * context, if a reader task is waiting. Apparently, this exciting
1548 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1549 * failed somewhere. Latency? Burstiness? Well, at least now we will
1550 * see, why it failed. 8)8) --ANK
1552 * NOTE: is this not too big to inline?
1554 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1556 struct tcp_opt *tp = tcp_sk(sk);
1558 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1559 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1560 tp->ucopy.memory += skb->truesize;
1561 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1562 struct sk_buff *skb1;
1564 BUG_ON(sock_owned_by_user(sk));
1566 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1567 sk->sk_backlog_rcv(sk, skb1);
1568 NET_INC_STATS_BH(TCPPrequeueDropped);
1571 tp->ucopy.memory = 0;
1572 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1573 wake_up_interruptible(sk->sk_sleep);
1574 if (!tcp_ack_scheduled(tp))
1575 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1586 static char *statename[]={
1587 "Unused","Established","Syn Sent","Syn Recv",
1588 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1589 "Close Wait","Last ACK","Listen","Closing"
1593 static __inline__ void tcp_set_state(struct sock *sk, int state)
1595 int oldstate = sk->sk_state;
1598 case TCP_ESTABLISHED:
1599 if (oldstate != TCP_ESTABLISHED)
1600 TCP_INC_STATS(TcpCurrEstab);
1604 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1605 TCP_INC_STATS(TcpEstabResets);
1607 sk->sk_prot->unhash(sk);
1608 if (tcp_sk(sk)->bind_hash &&
1609 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1613 if (oldstate==TCP_ESTABLISHED)
1614 TCP_DEC_STATS(TcpCurrEstab);
1617 /* Change state AFTER socket is unhashed to avoid closed
1618 * socket sitting in hash tables.
1620 sk->sk_state = state;
1623 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1627 static __inline__ void tcp_done(struct sock *sk)
1629 tcp_set_state(sk, TCP_CLOSE);
1630 tcp_clear_xmit_timers(sk);
1632 sk->sk_shutdown = SHUTDOWN_MASK;
1634 if (!sock_flag(sk, SOCK_DEAD))
1635 sk->sk_state_change(sk);
1637 tcp_destroy_sock(sk);
1640 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1647 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1649 if (tp->tstamp_ok) {
1650 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1651 (TCPOPT_NOP << 16) |
1652 (TCPOPT_TIMESTAMP << 8) |
1654 *ptr++ = htonl(tstamp);
1655 *ptr++ = htonl(tp->ts_recent);
1657 if (tp->eff_sacks) {
1658 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1661 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1662 (TCPOPT_NOP << 16) |
1663 (TCPOPT_SACK << 8) |
1664 (TCPOLEN_SACK_BASE +
1665 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1666 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1667 *ptr++ = htonl(sp[this_sack].start_seq);
1668 *ptr++ = htonl(sp[this_sack].end_seq);
1677 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1678 * If this is every changed make sure to change the definition of
1679 * MAX_SYN_SIZE to match the new maximum number of options that you
1682 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1683 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1685 /* We always get an MSS option.
1686 * The option bytes which will be seen in normal data
1687 * packets should timestamps be used, must be in the MSS
1688 * advertised. But we subtract them from tp->mss_cache so
1689 * that calculations in tcp_sendmsg are simpler etc.
1690 * So account for this fact here if necessary. If we
1691 * don't do this correctly, as a receiver we won't
1692 * recognize data packets as being full sized when we
1693 * should, and thus we won't abide by the delayed ACK
1695 * SACKs don't matter, we never delay an ACK when we
1696 * have any of those going out.
1698 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1701 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1702 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1704 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1705 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1706 *ptr++ = htonl(tstamp); /* TSVAL */
1707 *ptr++ = htonl(ts_recent); /* TSECR */
1709 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1710 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1712 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1715 /* Determine a window scaling and initial window to offer.
1716 * Based on the assumption that the given amount of space
1717 * will be offered. Store the results in the tp structure.
1718 * NOTE: for smooth operation initial space offering should
1719 * be a multiple of mss if possible. We assume here that mss >= 1.
1720 * This MUST be enforced by all callers.
1722 static inline void tcp_select_initial_window(int __space, __u32 mss,
1724 __u32 *window_clamp,
1728 unsigned int space = (__space < 0 ? 0 : __space);
1730 /* If no clamp set the clamp to the max possible scaled window */
1731 if (*window_clamp == 0)
1732 (*window_clamp) = (65535 << 14);
1733 space = min(*window_clamp, space);
1735 /* Quantize space offering to a multiple of mss if possible. */
1737 space = (space / mss) * mss;
1739 /* NOTE: offering an initial window larger than 32767
1740 * will break some buggy TCP stacks. We try to be nice.
1741 * If we are not window scaling, then this truncates
1742 * our initial window offering to 32k. There should also
1743 * be a sysctl option to stop being nice.
1745 (*rcv_wnd) = min(space, MAX_TCP_WINDOW);
1748 /* See RFC1323 for an explanation of the limit to 14 */
1749 while (space > 65535 && (*rcv_wscale) < 14) {
1753 if (*rcv_wscale && sysctl_tcp_app_win && space>=mss &&
1754 space - max((space>>sysctl_tcp_app_win), mss>>*rcv_wscale) < 65536/2)
1757 *rcv_wscale = max((__u8)sysctl_tcp_default_win_scale,
1761 /* Set initial window to value enough for senders,
1762 * following RFC1414. Senders, not following this RFC,
1763 * will be satisfied with 2.
1765 if (mss > (1<<*rcv_wscale)) {
1769 else if (mss > 1460)
1771 if (*rcv_wnd > init_cwnd*mss)
1772 *rcv_wnd = init_cwnd*mss;
1774 /* Set the clamp no higher than max representable value */
1775 (*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
1778 static inline int tcp_win_from_space(int space)
1780 return sysctl_tcp_adv_win_scale<=0 ?
1781 (space>>(-sysctl_tcp_adv_win_scale)) :
1782 space - (space>>sysctl_tcp_adv_win_scale);
1785 /* Note: caller must be prepared to deal with negative returns */
1786 static inline int tcp_space(struct sock *sk)
1788 return tcp_win_from_space(sk->sk_rcvbuf -
1789 atomic_read(&sk->sk_rmem_alloc));
1792 static inline int tcp_full_space( struct sock *sk)
1794 return tcp_win_from_space(sk->sk_rcvbuf);
1797 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1800 struct tcp_opt *tp = tcp_sk(sk);
1803 sk_acceptq_added(sk);
1805 if (!tp->accept_queue_tail) {
1806 tp->accept_queue = req;
1808 tp->accept_queue_tail->dl_next = req;
1810 tp->accept_queue_tail = req;
1811 req->dl_next = NULL;
1814 struct tcp_listen_opt
1816 u8 max_qlen_log; /* log_2 of maximal queued SYNs */
1821 struct open_request *syn_table[TCP_SYNQ_HSIZE];
1825 tcp_synq_removed(struct sock *sk, struct open_request *req)
1827 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1829 if (--lopt->qlen == 0)
1830 tcp_delete_keepalive_timer(sk);
1831 if (req->retrans == 0)
1835 static inline void tcp_synq_added(struct sock *sk)
1837 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1839 if (lopt->qlen++ == 0)
1840 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1844 static inline int tcp_synq_len(struct sock *sk)
1846 return tcp_sk(sk)->listen_opt->qlen;
1849 static inline int tcp_synq_young(struct sock *sk)
1851 return tcp_sk(sk)->listen_opt->qlen_young;
1854 static inline int tcp_synq_is_full(struct sock *sk)
1856 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1859 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1860 struct open_request **prev)
1862 write_lock(&tp->syn_wait_lock);
1863 *prev = req->dl_next;
1864 write_unlock(&tp->syn_wait_lock);
1867 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1868 struct open_request **prev)
1870 tcp_synq_unlink(tcp_sk(sk), req, prev);
1871 tcp_synq_removed(sk, req);
1872 tcp_openreq_free(req);
1875 static __inline__ void tcp_openreq_init(struct open_request *req,
1877 struct sk_buff *skb)
1879 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1880 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1881 req->mss = tp->mss_clamp;
1882 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1883 req->tstamp_ok = tp->tstamp_ok;
1884 req->sack_ok = tp->sack_ok;
1885 req->snd_wscale = tp->snd_wscale;
1886 req->wscale_ok = tp->wscale_ok;
1889 req->rmt_port = skb->h.th->source;
1892 #define TCP_MEM_QUANTUM ((int)PAGE_SIZE)
1894 static inline void tcp_free_skb(struct sock *sk, struct sk_buff *skb)
1896 tcp_sk(sk)->queue_shrunk = 1;
1897 sk->sk_wmem_queued -= skb->truesize;
1898 sk->sk_forward_alloc += skb->truesize;
1902 static inline void tcp_charge_skb(struct sock *sk, struct sk_buff *skb)
1904 sk->sk_wmem_queued += skb->truesize;
1905 sk->sk_forward_alloc -= skb->truesize;
1908 extern void __tcp_mem_reclaim(struct sock *sk);
1909 extern int tcp_mem_schedule(struct sock *sk, int size, int kind);
1911 static inline void tcp_mem_reclaim(struct sock *sk)
1913 if (sk->sk_forward_alloc >= TCP_MEM_QUANTUM)
1914 __tcp_mem_reclaim(sk);
1917 static inline void tcp_enter_memory_pressure(void)
1919 if (!tcp_memory_pressure) {
1920 NET_INC_STATS(TCPMemoryPressures);
1921 tcp_memory_pressure = 1;
1925 static inline void tcp_moderate_sndbuf(struct sock *sk)
1927 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1928 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued / 2);
1929 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1933 static inline struct sk_buff *tcp_alloc_pskb(struct sock *sk, int size, int mem, int gfp)
1935 struct sk_buff *skb = alloc_skb(size+MAX_TCP_HEADER, gfp);
1938 skb->truesize += mem;
1939 if (sk->sk_forward_alloc >= (int)skb->truesize ||
1940 tcp_mem_schedule(sk, skb->truesize, 0)) {
1941 skb_reserve(skb, MAX_TCP_HEADER);
1946 tcp_enter_memory_pressure();
1947 tcp_moderate_sndbuf(sk);
1952 static inline struct sk_buff *tcp_alloc_skb(struct sock *sk, int size, int gfp)
1954 return tcp_alloc_pskb(sk, size, 0, gfp);
1957 static inline struct page * tcp_alloc_page(struct sock *sk)
1959 if (sk->sk_forward_alloc >= (int)PAGE_SIZE ||
1960 tcp_mem_schedule(sk, PAGE_SIZE, 0)) {
1961 struct page *page = alloc_pages(sk->sk_allocation, 0);
1965 tcp_enter_memory_pressure();
1966 tcp_moderate_sndbuf(sk);
1970 static inline void tcp_writequeue_purge(struct sock *sk)
1972 struct sk_buff *skb;
1974 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1975 tcp_free_skb(sk, skb);
1976 tcp_mem_reclaim(sk);
1979 extern void tcp_rfree(struct sk_buff *skb);
1981 static inline void tcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
1984 skb->destructor = tcp_rfree;
1985 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1986 sk->sk_forward_alloc -= skb->truesize;
1989 extern void tcp_listen_wlock(void);
1991 /* - We may sleep inside this lock.
1992 * - If sleeping is not required (or called from BH),
1993 * use plain read_(un)lock(&tcp_lhash_lock).
1996 static inline void tcp_listen_lock(void)
1998 /* read_lock synchronizes to candidates to writers */
1999 read_lock(&tcp_lhash_lock);
2000 atomic_inc(&tcp_lhash_users);
2001 read_unlock(&tcp_lhash_lock);
2004 static inline void tcp_listen_unlock(void)
2006 if (atomic_dec_and_test(&tcp_lhash_users))
2007 wake_up(&tcp_lhash_wait);
2010 static inline int keepalive_intvl_when(struct tcp_opt *tp)
2012 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
2015 static inline int keepalive_time_when(struct tcp_opt *tp)
2017 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
2020 static inline int tcp_fin_time(struct tcp_opt *tp)
2022 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
2024 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
2025 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
2030 static inline int tcp_paws_check(struct tcp_opt *tp, int rst)
2032 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
2034 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
2037 /* RST segments are not recommended to carry timestamp,
2038 and, if they do, it is recommended to ignore PAWS because
2039 "their cleanup function should take precedence over timestamps."
2040 Certainly, it is mistake. It is necessary to understand the reasons
2041 of this constraint to relax it: if peer reboots, clock may go
2042 out-of-sync and half-open connections will not be reset.
2043 Actually, the problem would be not existing if all
2044 the implementations followed draft about maintaining clock
2045 via reboots. Linux-2.2 DOES NOT!
2047 However, we can relax time bounds for RST segments to MSL.
2049 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
2054 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
2056 sk->sk_route_caps = dst->dev->features;
2057 if (sk->sk_route_caps & NETIF_F_TSO) {
2058 if (sk->sk_no_largesend || dst->header_len)
2059 sk->sk_route_caps &= ~NETIF_F_TSO;
2063 #define TCP_CHECK_TIMER(sk) do { } while (0)
2065 static inline int tcp_use_frto(const struct sock *sk)
2067 const struct tcp_opt *tp = tcp_sk(sk);
2069 /* F-RTO must be activated in sysctl and there must be some
2070 * unsent new data, and the advertised window should allow
2073 return (sysctl_tcp_frto && tp->send_head &&
2074 !after(TCP_SKB_CB(tp->send_head)->end_seq,
2075 tp->snd_una + tp->snd_wnd));
2078 static inline void tcp_mib_init(void)
2081 TCP_ADD_STATS_USER(TcpRtoAlgorithm, 1);
2082 TCP_ADD_STATS_USER(TcpRtoMin, TCP_RTO_MIN*1000/HZ);
2083 TCP_ADD_STATS_USER(TcpRtoMax, TCP_RTO_MAX*1000/HZ);
2084 TCP_ADD_STATS_USER(TcpMaxConn, -1);
2088 enum tcp_seq_states {
2089 TCP_SEQ_STATE_LISTENING,
2090 TCP_SEQ_STATE_OPENREQ,
2091 TCP_SEQ_STATE_ESTABLISHED,
2092 TCP_SEQ_STATE_TIME_WAIT,
2095 struct tcp_seq_afinfo {
2096 struct module *owner;
2099 int (*seq_show) (struct seq_file *m, void *v);
2100 struct file_operations *seq_fops;
2103 struct tcp_iter_state {
2105 enum tcp_seq_states state;
2106 struct sock *syn_wait_sk;
2107 int bucket, sbucket, num, uid;
2108 struct seq_operations seq_ops;
2111 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2112 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2114 /* TCP Westwood functions and constants */
2116 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2117 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2119 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2121 if (sysctl_tcp_westwood)
2122 tp->westwood.rtt = rtt_seq;
2125 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2126 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2128 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2130 if (sysctl_tcp_westwood)
2131 __tcp_westwood_fast_bw(sk, skb);
2134 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2136 if (sysctl_tcp_westwood)
2137 __tcp_westwood_slow_bw(sk, skb);
2140 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2142 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2143 (__u32) (tp->mss_cache),
2147 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2149 return sysctl_tcp_westwood ? __tcp_westwood_bw_rttmin(tp) : 0;
2152 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2156 if (sysctl_tcp_westwood) {
2157 ssthresh = __tcp_westwood_bw_rttmin(tp);
2159 tp->snd_ssthresh = ssthresh;
2162 return (ssthresh != 0);
2165 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2169 if (sysctl_tcp_westwood) {
2170 cwnd = __tcp_westwood_bw_rttmin(tp);
2172 tp->snd_cwnd = cwnd;