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 extern kmem_cache_t *tcp_bucket_cachep;
157 extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head,
158 unsigned short snum);
159 extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb);
160 extern void tcp_bucket_unlock(struct sock *sk);
161 extern int tcp_port_rover;
163 /* These are AF independent. */
164 static __inline__ int tcp_bhashfn(__u16 lport)
166 return (lport & (tcp_bhash_size - 1));
169 extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb,
170 unsigned short snum);
172 #if (BITS_PER_LONG == 64)
173 #define TCP_ADDRCMP_ALIGN_BYTES 8
175 #define TCP_ADDRCMP_ALIGN_BYTES 4
178 /* This is a TIME_WAIT bucket. It works around the memory consumption
179 * problems of sockets in such a state on heavily loaded servers, but
180 * without violating the protocol specification.
182 struct tcp_tw_bucket {
184 * Now struct sock also uses sock_common, so please just
185 * don't add nothing before this first member (__tw_common) --acme
187 struct sock_common __tw_common;
188 #define tw_family __tw_common.skc_family
189 #define tw_state __tw_common.skc_state
190 #define tw_reuse __tw_common.skc_reuse
191 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
192 #define tw_node __tw_common.skc_node
193 #define tw_bind_node __tw_common.skc_bind_node
194 #define tw_refcnt __tw_common.skc_refcnt
195 volatile unsigned char tw_substate;
196 unsigned char tw_rcv_wscale;
198 /* Socket demultiplex comparisons on incoming packets. */
199 /* these five are in inet_opt */
201 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
205 /* And these are ours. */
212 long tw_ts_recent_stamp;
213 unsigned long tw_ttd;
214 struct tcp_bind_bucket *tw_tb;
215 struct hlist_node tw_death_node;
216 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
217 struct in6_addr tw_v6_daddr;
218 struct in6_addr tw_v6_rcv_saddr;
223 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
224 struct hlist_head *list)
226 hlist_add_head(&tw->tw_node, list);
229 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
230 struct hlist_head *list)
232 hlist_add_head(&tw->tw_bind_node, list);
235 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
237 return tw->tw_death_node.pprev != NULL;
240 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
242 tw->tw_death_node.pprev = NULL;
245 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
247 __hlist_del(&tw->tw_death_node);
248 tw_dead_node_init(tw);
251 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
253 if (tw_dead_hashed(tw)) {
254 __tw_del_dead_node(tw);
260 #define tw_for_each(tw, node, head) \
261 hlist_for_each_entry(tw, node, head, tw_node)
263 #define tw_for_each_inmate(tw, node, jail) \
264 hlist_for_each_entry(tw, node, jail, tw_death_node)
266 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
267 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
269 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
271 static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
273 return likely(sk->sk_state != TCP_TIME_WAIT) ?
274 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
277 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
278 static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
280 return likely(sk->sk_state != TCP_TIME_WAIT) ?
281 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
284 static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
286 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
289 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
291 static inline int tcp_v6_ipv6only(const struct sock *sk)
293 return likely(sk->sk_state != TCP_TIME_WAIT) ?
294 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
297 # define __tcp_v6_rcv_saddr(__sk) NULL
298 # define tcp_v6_rcv_saddr(__sk) NULL
299 # define tcptw_sk_ipv6only(__sk) 0
300 # define tcp_v6_ipv6only(__sk) 0
303 extern kmem_cache_t *tcp_timewait_cachep;
305 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
307 if (atomic_dec_and_test(&tw->tw_refcnt)) {
308 #ifdef INET_REFCNT_DEBUG
309 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
311 kmem_cache_free(tcp_timewait_cachep, tw);
315 extern atomic_t tcp_orphan_count;
316 extern int tcp_tw_count;
317 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
318 extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
319 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
322 /* Socket demux engine toys. */
324 #define TCP_COMBINED_PORTS(__sport, __dport) \
325 (((__u32)(__sport)<<16) | (__u32)(__dport))
326 #else /* __LITTLE_ENDIAN */
327 #define TCP_COMBINED_PORTS(__sport, __dport) \
328 (((__u32)(__dport)<<16) | (__u32)(__sport))
331 #if (BITS_PER_LONG == 64)
333 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
334 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
335 #else /* __LITTLE_ENDIAN */
336 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
337 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
338 #endif /* __BIG_ENDIAN */
339 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
340 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
341 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
342 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
343 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
344 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
345 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
346 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
347 #else /* 32-bit arch */
348 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
349 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
350 ((inet_sk(__sk)->daddr == (__saddr)) && \
351 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
352 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
353 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
354 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
355 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
356 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
357 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
358 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
359 #endif /* 64-bit arch */
361 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
362 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
363 ((__sk)->sk_family == AF_INET6) && \
364 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \
365 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
366 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
368 /* These can have wildcards, don't try too hard. */
369 static __inline__ int tcp_lhashfn(unsigned short num)
371 return num & (TCP_LHTABLE_SIZE - 1);
374 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
376 return tcp_lhashfn(inet_sk(sk)->num);
379 #define MAX_TCP_HEADER (128 + MAX_HEADER)
382 * Never offer a window over 32767 without using window scaling. Some
383 * poor stacks do signed 16bit maths!
385 #define MAX_TCP_WINDOW 32767U
387 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
388 #define TCP_MIN_MSS 88U
390 /* Minimal RCV_MSS. */
391 #define TCP_MIN_RCVMSS 536U
393 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
394 #define TCP_FASTRETRANS_THRESH 3
396 /* Maximal reordering. */
397 #define TCP_MAX_REORDERING 127
399 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
400 #define TCP_MAX_QUICKACKS 16U
402 /* urg_data states */
403 #define TCP_URG_VALID 0x0100
404 #define TCP_URG_NOTYET 0x0200
405 #define TCP_URG_READ 0x0400
407 #define TCP_RETR1 3 /*
408 * This is how many retries it does before it
409 * tries to figure out if the gateway is
410 * down. Minimal RFC value is 3; it corresponds
411 * to ~3sec-8min depending on RTO.
414 #define TCP_RETR2 15 /*
415 * This should take at least
416 * 90 minutes to time out.
417 * RFC1122 says that the limit is 100 sec.
418 * 15 is ~13-30min depending on RTO.
421 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
422 * connection: ~180sec is RFC minumum */
424 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
425 * connection: ~180sec is RFC minumum */
428 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
429 * socket. 7 is ~50sec-16min.
433 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
434 * state, about 60 seconds */
435 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
436 /* BSD style FIN_WAIT2 deadlock breaker.
437 * It used to be 3min, new value is 60sec,
438 * to combine FIN-WAIT-2 timeout with
442 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
444 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
445 #define TCP_ATO_MIN ((unsigned)(HZ/25))
447 #define TCP_DELACK_MIN 4U
448 #define TCP_ATO_MIN 4U
450 #define TCP_RTO_MAX ((unsigned)(120*HZ))
451 #define TCP_RTO_MIN ((unsigned)(HZ/5))
452 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
454 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
455 * for local resources.
458 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
459 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
460 #define TCP_KEEPALIVE_INTVL (75*HZ)
462 #define MAX_TCP_KEEPIDLE 32767
463 #define MAX_TCP_KEEPINTVL 32767
464 #define MAX_TCP_KEEPCNT 127
465 #define MAX_TCP_SYNCNT 127
467 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
468 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
470 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
471 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
472 * after this time. It should be equal
473 * (or greater than) TCP_TIMEWAIT_LEN
474 * to provide reliability equal to one
475 * provided by timewait state.
477 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
478 * timestamps. It must be less than
479 * minimal timewait lifetime.
482 #define TCP_TW_RECYCLE_SLOTS_LOG 5
483 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
485 /* If time > 4sec, it is "slow" path, no recycling is required,
486 so that we select tick to get range about 4 seconds.
489 #if HZ <= 16 || HZ > 4096
490 # error Unsupported: HZ <= 16 or HZ > 4096
492 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
494 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
496 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
498 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
500 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
502 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
504 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
506 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
509 #define BICTCP_1_OVER_BETA 8 /*
511 * multiplicative decrease factor
513 #define BICTCP_MAX_INCREMENT 32 /*
514 * Limit on the amount of
515 * increment allowed during
518 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
519 * log(B/Smin)/log(B/(B-1))+1,
523 #define BICTCP_B 4 /*
525 * go to point (max+min)/N
532 #define TCPOPT_NOP 1 /* Padding */
533 #define TCPOPT_EOL 0 /* End of options */
534 #define TCPOPT_MSS 2 /* Segment size negotiating */
535 #define TCPOPT_WINDOW 3 /* Window scaling */
536 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
537 #define TCPOPT_SACK 5 /* SACK Block */
538 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
544 #define TCPOLEN_MSS 4
545 #define TCPOLEN_WINDOW 3
546 #define TCPOLEN_SACK_PERM 2
547 #define TCPOLEN_TIMESTAMP 10
549 /* But this is what stacks really send out. */
550 #define TCPOLEN_TSTAMP_ALIGNED 12
551 #define TCPOLEN_WSCALE_ALIGNED 4
552 #define TCPOLEN_SACKPERM_ALIGNED 4
553 #define TCPOLEN_SACK_BASE 2
554 #define TCPOLEN_SACK_BASE_ALIGNED 4
555 #define TCPOLEN_SACK_PERBLOCK 8
557 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
558 #define TCP_TIME_DACK 2 /* Delayed ack timer */
559 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
560 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
562 /* Flags in tp->nonagle */
563 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
564 #define TCP_NAGLE_CORK 2 /* Socket is corked */
565 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
567 /* sysctl variables for tcp */
568 extern int sysctl_max_syn_backlog;
569 extern int sysctl_tcp_timestamps;
570 extern int sysctl_tcp_window_scaling;
571 extern int sysctl_tcp_sack;
572 extern int sysctl_tcp_fin_timeout;
573 extern int sysctl_tcp_tw_recycle;
574 extern int sysctl_tcp_keepalive_time;
575 extern int sysctl_tcp_keepalive_probes;
576 extern int sysctl_tcp_keepalive_intvl;
577 extern int sysctl_tcp_syn_retries;
578 extern int sysctl_tcp_synack_retries;
579 extern int sysctl_tcp_retries1;
580 extern int sysctl_tcp_retries2;
581 extern int sysctl_tcp_orphan_retries;
582 extern int sysctl_tcp_syncookies;
583 extern int sysctl_tcp_retrans_collapse;
584 extern int sysctl_tcp_stdurg;
585 extern int sysctl_tcp_rfc1337;
586 extern int sysctl_tcp_abort_on_overflow;
587 extern int sysctl_tcp_max_orphans;
588 extern int sysctl_tcp_max_tw_buckets;
589 extern int sysctl_tcp_fack;
590 extern int sysctl_tcp_reordering;
591 extern int sysctl_tcp_ecn;
592 extern int sysctl_tcp_dsack;
593 extern int sysctl_tcp_mem[3];
594 extern int sysctl_tcp_wmem[3];
595 extern int sysctl_tcp_rmem[3];
596 extern int sysctl_tcp_app_win;
597 extern int sysctl_tcp_adv_win_scale;
598 extern int sysctl_tcp_tw_reuse;
599 extern int sysctl_tcp_frto;
600 extern int sysctl_tcp_low_latency;
601 extern int sysctl_tcp_westwood;
602 extern int sysctl_tcp_vegas_cong_avoid;
603 extern int sysctl_tcp_vegas_alpha;
604 extern int sysctl_tcp_vegas_beta;
605 extern int sysctl_tcp_vegas_gamma;
606 extern int sysctl_tcp_nometrics_save;
607 extern int sysctl_tcp_bic;
608 extern int sysctl_tcp_bic_fast_convergence;
609 extern int sysctl_tcp_bic_low_window;
610 extern int sysctl_tcp_moderate_rcvbuf;
611 extern int sysctl_tcp_tso_win_divisor;
613 extern atomic_t tcp_memory_allocated;
614 extern atomic_t tcp_sockets_allocated;
615 extern int tcp_memory_pressure;
619 struct or_calltable {
621 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
622 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
623 void (*destructor) (struct open_request *req);
624 void (*send_reset) (struct sk_buff *skb);
627 struct tcp_v4_open_req {
630 struct ip_options *opt;
633 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
634 struct tcp_v6_open_req {
635 struct in6_addr loc_addr;
636 struct in6_addr rmt_addr;
637 struct sk_buff *pktopts;
642 /* this structure is too big */
643 struct open_request {
644 struct open_request *dl_next; /* Must be first member! */
651 __u16 snd_wscale : 4,
658 /* The following two fields can be easily recomputed I think -AK */
659 __u32 window_clamp; /* window clamp at creation time */
660 __u32 rcv_wnd; /* rcv_wnd offered first time */
662 unsigned long expires;
663 struct or_calltable *class;
666 struct tcp_v4_open_req v4_req;
667 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
668 struct tcp_v6_open_req v6_req;
671 #ifdef CONFIG_ACCEPT_QUEUES
672 unsigned long acceptq_time_stamp;
677 /* SLAB cache for open requests. */
678 extern kmem_cache_t *tcp_openreq_cachep;
680 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
681 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
683 static inline void tcp_openreq_free(struct open_request *req)
685 req->class->destructor(req);
686 tcp_openreq_fastfree(req);
689 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
690 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
692 #define TCP_INET_FAMILY(fam) 1
696 * Pointers to address related TCP functions
697 * (i.e. things that depend on the address family)
701 int (*queue_xmit) (struct sk_buff *skb,
704 void (*send_check) (struct sock *sk,
707 struct sk_buff *skb);
709 int (*rebuild_header) (struct sock *sk);
711 int (*conn_request) (struct sock *sk,
712 struct sk_buff *skb);
714 struct sock * (*syn_recv_sock) (struct sock *sk,
716 struct open_request *req,
717 struct dst_entry *dst);
719 int (*remember_stamp) (struct sock *sk);
721 __u16 net_header_len;
723 int (*setsockopt) (struct sock *sk,
729 int (*getsockopt) (struct sock *sk,
736 void (*addr2sockaddr) (struct sock *sk,
743 * The next routines deal with comparing 32 bit unsigned ints
744 * and worry about wraparound (automatic with unsigned arithmetic).
747 static inline int before(__u32 seq1, __u32 seq2)
749 return (__s32)(seq1-seq2) < 0;
752 static inline int after(__u32 seq1, __u32 seq2)
754 return (__s32)(seq2-seq1) < 0;
758 /* is s2<=s1<=s3 ? */
759 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
761 return seq3 - seq2 >= seq1 - seq2;
765 extern struct proto tcp_prot;
767 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
768 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
769 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
770 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
771 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
772 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
773 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
775 extern void tcp_put_port(struct sock *sk);
776 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
778 extern void tcp_v4_err(struct sk_buff *skb, u32);
780 extern void tcp_shutdown (struct sock *sk, int how);
782 extern int tcp_v4_rcv(struct sk_buff *skb);
784 extern int tcp_v4_remember_stamp(struct sock *sk);
786 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
788 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
789 struct msghdr *msg, size_t size);
790 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
792 extern int tcp_ioctl(struct sock *sk,
796 extern int tcp_rcv_state_process(struct sock *sk,
801 extern int tcp_rcv_established(struct sock *sk,
806 extern void tcp_rcv_space_adjust(struct sock *sk);
815 static inline void tcp_schedule_ack(struct tcp_opt *tp)
817 tp->ack.pending |= TCP_ACK_SCHED;
820 static inline int tcp_ack_scheduled(struct tcp_opt *tp)
822 return tp->ack.pending&TCP_ACK_SCHED;
825 static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
827 if (tp->ack.quick && --tp->ack.quick == 0) {
828 /* Leaving quickack mode we deflate ATO. */
829 tp->ack.ato = TCP_ATO_MIN;
833 extern void tcp_enter_quickack_mode(struct tcp_opt *tp);
835 static __inline__ void tcp_delack_init(struct tcp_opt *tp)
837 memset(&tp->ack, 0, sizeof(tp->ack));
840 static inline void tcp_clear_options(struct tcp_opt *tp)
842 tp->tstamp_ok = tp->sack_ok = tp->wscale_ok = tp->snd_wscale = 0;
854 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
859 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
860 struct open_request *req,
861 struct open_request **prev);
862 extern int tcp_child_process(struct sock *parent,
864 struct sk_buff *skb);
865 extern void tcp_enter_frto(struct sock *sk);
866 extern void tcp_enter_loss(struct sock *sk, int how);
867 extern void tcp_clear_retrans(struct tcp_opt *tp);
868 extern void tcp_update_metrics(struct sock *sk);
870 extern void tcp_close(struct sock *sk,
872 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
873 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
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 *);
951 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
953 extern void tcp_send_probe0(struct sock *);
954 extern void tcp_send_partial(struct sock *);
955 extern int tcp_write_wakeup(struct sock *);
956 extern void tcp_send_fin(struct sock *sk);
957 extern void tcp_send_active_reset(struct sock *sk, int priority);
958 extern int tcp_send_synack(struct sock *);
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 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
970 extern unsigned int tcp_current_mss(struct sock *sk, int large);
973 extern const char tcp_timer_bug_msg[];
977 extern void tcp_get_info(struct sock *, struct tcp_info *);
979 /* Read 'sendfile()'-style from a TCP socket */
980 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
981 unsigned int, size_t);
982 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
983 sk_read_actor_t recv_actor);
985 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
987 struct tcp_opt *tp = tcp_sk(sk);
990 case TCP_TIME_RETRANS:
991 case TCP_TIME_PROBE0:
994 #ifdef TCP_CLEAR_TIMERS
995 sk_stop_timer(sk, &tp->retransmit_timer);
1000 tp->ack.pending = 0;
1002 #ifdef TCP_CLEAR_TIMERS
1003 sk_stop_timer(sk, &tp->delack_timer);
1008 printk(tcp_timer_bug_msg);
1016 * Reset the retransmission timer
1018 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
1020 struct tcp_opt *tp = tcp_sk(sk);
1022 if (when > TCP_RTO_MAX) {
1024 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
1030 case TCP_TIME_RETRANS:
1031 case TCP_TIME_PROBE0:
1033 tp->timeout = jiffies+when;
1034 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
1038 tp->ack.pending |= TCP_ACK_TIMER;
1039 tp->ack.timeout = jiffies+when;
1040 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
1045 printk(tcp_timer_bug_msg);
1050 /* Initialize RCV_MSS value.
1051 * RCV_MSS is an our guess about MSS used by the peer.
1052 * We haven't any direct information about the MSS.
1053 * It's better to underestimate the RCV_MSS rather than overestimate.
1054 * Overestimations make us ACKing less frequently than needed.
1055 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1058 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1060 struct tcp_opt *tp = tcp_sk(sk);
1061 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1063 hint = min(hint, tp->rcv_wnd/2);
1064 hint = min(hint, TCP_MIN_RCVMSS);
1065 hint = max(hint, TCP_MIN_MSS);
1067 tp->ack.rcv_mss = hint;
1070 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1072 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1073 ntohl(TCP_FLAG_ACK) |
1077 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1079 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1082 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1084 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1086 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1088 tcp_fast_path_on(tp);
1091 /* Compute the actual receive window we are currently advertising.
1092 * Rcv_nxt can be after the window if our peer push more data
1093 * than the offered window.
1095 static __inline__ u32 tcp_receive_window(const struct tcp_opt *tp)
1097 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1104 /* Choose a new window, without checks for shrinking, and without
1105 * scaling applied to the result. The caller does these things
1106 * if necessary. This is a "raw" window selection.
1108 extern u32 __tcp_select_window(struct sock *sk);
1110 /* TCP timestamps are only 32-bits, this causes a slight
1111 * complication on 64-bit systems since we store a snapshot
1112 * of jiffies in the buffer control blocks below. We decidely
1113 * only use of the low 32-bits of jiffies and hide the ugly
1114 * casts with the following macro.
1116 #define tcp_time_stamp ((__u32)(jiffies))
1118 /* This is what the send packet queueing engine uses to pass
1119 * TCP per-packet control information to the transmission
1120 * code. We also store the host-order sequence numbers in
1121 * here too. This is 36 bytes on 32-bit architectures,
1122 * 40 bytes on 64-bit machines, if this grows please adjust
1123 * skbuff.h:skbuff->cb[xxx] size appropriately.
1127 struct inet_skb_parm h4;
1128 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1129 struct inet6_skb_parm h6;
1131 } header; /* For incoming frames */
1132 __u32 seq; /* Starting sequence number */
1133 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1134 __u32 when; /* used to compute rtt's */
1135 __u8 flags; /* TCP header flags. */
1137 /* NOTE: These must match up to the flags byte in a
1140 #define TCPCB_FLAG_FIN 0x01
1141 #define TCPCB_FLAG_SYN 0x02
1142 #define TCPCB_FLAG_RST 0x04
1143 #define TCPCB_FLAG_PSH 0x08
1144 #define TCPCB_FLAG_ACK 0x10
1145 #define TCPCB_FLAG_URG 0x20
1146 #define TCPCB_FLAG_ECE 0x40
1147 #define TCPCB_FLAG_CWR 0x80
1149 __u8 sacked; /* State flags for SACK/FACK. */
1150 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1151 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1152 #define TCPCB_LOST 0x04 /* SKB is lost */
1153 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1155 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1156 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1158 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1160 #define TCPCB_AT_TAIL (TCPCB_URG)
1162 __u16 urg_ptr; /* Valid w/URG flags is set. */
1163 __u32 ack_seq; /* Sequence number ACK'd */
1166 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1168 #include <net/tcp_ecn.h>
1170 /* Due to TSO, an SKB can be composed of multiple actual
1171 * packets. To keep these tracked properly, we use this.
1173 static inline int tcp_skb_pcount(const struct sk_buff *skb)
1175 return skb_shinfo(skb)->tso_segs;
1178 /* This is valid iff tcp_skb_pcount() > 1. */
1179 static inline int tcp_skb_mss(const struct sk_buff *skb)
1181 return skb_shinfo(skb)->tso_size;
1184 static inline void tcp_inc_pcount(tcp_pcount_t *count,
1185 const struct sk_buff *skb)
1187 count->val += tcp_skb_pcount(skb);
1190 static inline void tcp_inc_pcount_explicit(tcp_pcount_t *count, int amt)
1195 static inline void tcp_dec_pcount_explicit(tcp_pcount_t *count, int amt)
1200 static inline void tcp_dec_pcount(tcp_pcount_t *count,
1201 const struct sk_buff *skb)
1203 count->val -= tcp_skb_pcount(skb);
1206 static inline void tcp_dec_pcount_approx(tcp_pcount_t *count,
1207 const struct sk_buff *skb)
1210 count->val -= tcp_skb_pcount(skb);
1211 if ((int)count->val < 0)
1216 static inline __u32 tcp_get_pcount(const tcp_pcount_t *count)
1221 static inline void tcp_set_pcount(tcp_pcount_t *count, __u32 val)
1226 static inline void tcp_packets_out_inc(struct sock *sk,
1228 const struct sk_buff *skb)
1230 int orig = tcp_get_pcount(&tp->packets_out);
1232 tcp_inc_pcount(&tp->packets_out, skb);
1234 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
1237 static inline void tcp_packets_out_dec(struct tcp_opt *tp,
1238 const struct sk_buff *skb)
1240 tcp_dec_pcount(&tp->packets_out, skb);
1243 /* This determines how many packets are "in the network" to the best
1244 * of our knowledge. In many cases it is conservative, but where
1245 * detailed information is available from the receiver (via SACK
1246 * blocks etc.) we can make more aggressive calculations.
1248 * Use this for decisions involving congestion control, use just
1249 * tp->packets_out to determine if the send queue is empty or not.
1251 * Read this equation as:
1253 * "Packets sent once on transmission queue" MINUS
1254 * "Packets left network, but not honestly ACKed yet" PLUS
1255 * "Packets fast retransmitted"
1257 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_opt *tp)
1259 return (tcp_get_pcount(&tp->packets_out) -
1260 tcp_get_pcount(&tp->left_out) +
1261 tcp_get_pcount(&tp->retrans_out));
1265 * Which congestion algorithim is in use on the connection.
1267 #define tcp_is_vegas(__tp) ((__tp)->adv_cong == TCP_VEGAS)
1268 #define tcp_is_westwood(__tp) ((__tp)->adv_cong == TCP_WESTWOOD)
1269 #define tcp_is_bic(__tp) ((__tp)->adv_cong == TCP_BIC)
1271 /* Recalculate snd_ssthresh, we want to set it to:
1274 * one half the current congestion window, but no
1275 * less than two segments
1278 * behave like Reno until low_window is reached,
1279 * then increase congestion window slowly
1281 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1283 if (tcp_is_bic(tp)) {
1284 if (sysctl_tcp_bic_fast_convergence &&
1285 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1286 tp->bictcp.last_max_cwnd
1287 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1288 / (BICTCP_1_OVER_BETA/2);
1290 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1292 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1293 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1297 return max(tp->snd_cwnd >> 1U, 2U);
1300 /* Stop taking Vegas samples for now. */
1301 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1303 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1305 /* There are several situations when we must "re-start" Vegas:
1307 * o when a connection is established
1309 * o after fast recovery
1310 * o when we send a packet and there is no outstanding
1311 * unacknowledged data (restarting an idle connection)
1313 * In these circumstances we cannot do a Vegas calculation at the
1314 * end of the first RTT, because any calculation we do is using
1315 * stale info -- both the saved cwnd and congestion feedback are
1318 * Instead we must wait until the completion of an RTT during
1319 * which we actually receive ACKs.
1322 /* Begin taking Vegas samples next time we send something. */
1323 tp->vegas.doing_vegas_now = 1;
1325 /* Set the beginning of the next send window. */
1326 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1328 tp->vegas.cntRTT = 0;
1329 tp->vegas.minRTT = 0x7fffffff;
1332 /* Should we be taking Vegas samples right now? */
1333 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1335 extern void tcp_ca_init(struct tcp_opt *tp);
1337 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1339 if (tcp_is_vegas(tp)) {
1340 if (ca_state == TCP_CA_Open)
1341 tcp_vegas_enable(tp);
1343 tcp_vegas_disable(tp);
1345 tp->ca_state = ca_state;
1348 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1349 * The exception is rate halving phase, when cwnd is decreasing towards
1352 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1354 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1355 return tp->snd_ssthresh;
1357 return max(tp->snd_ssthresh,
1358 ((tp->snd_cwnd >> 1) +
1359 (tp->snd_cwnd >> 2)));
1362 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1365 (tcp_get_pcount(&tp->sacked_out) >=
1366 tcp_get_pcount(&tp->packets_out) - tcp_get_pcount(&tp->lost_out)))
1367 tcp_set_pcount(&tp->sacked_out,
1368 (tcp_get_pcount(&tp->packets_out) -
1369 tcp_get_pcount(&tp->lost_out)));
1370 tcp_set_pcount(&tp->left_out,
1371 (tcp_get_pcount(&tp->sacked_out) +
1372 tcp_get_pcount(&tp->lost_out)));
1375 extern void tcp_cwnd_application_limited(struct sock *sk);
1377 /* Congestion window validation. (RFC2861) */
1379 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1381 __u32 packets_out = tcp_get_pcount(&tp->packets_out);
1383 if (packets_out >= tp->snd_cwnd) {
1384 /* Network is feed fully. */
1385 tp->snd_cwnd_used = 0;
1386 tp->snd_cwnd_stamp = tcp_time_stamp;
1388 /* Network starves. */
1389 if (tcp_get_pcount(&tp->packets_out) > tp->snd_cwnd_used)
1390 tp->snd_cwnd_used = tcp_get_pcount(&tp->packets_out);
1392 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1393 tcp_cwnd_application_limited(sk);
1397 /* Set slow start threshould and cwnd not falling to slow start */
1398 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1400 tp->undo_marker = 0;
1401 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1402 tp->snd_cwnd = min(tp->snd_cwnd,
1403 tcp_packets_in_flight(tp) + 1U);
1404 tp->snd_cwnd_cnt = 0;
1405 tp->high_seq = tp->snd_nxt;
1406 tp->snd_cwnd_stamp = tcp_time_stamp;
1407 TCP_ECN_queue_cwr(tp);
1410 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1412 tp->prior_ssthresh = 0;
1413 if (tp->ca_state < TCP_CA_CWR) {
1414 __tcp_enter_cwr(tp);
1415 tcp_set_ca_state(tp, TCP_CA_CWR);
1419 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1421 /* Slow start with delack produces 3 packets of burst, so that
1422 * it is safe "de facto".
1424 static __inline__ __u32 tcp_max_burst(const struct tcp_opt *tp)
1429 static __inline__ int tcp_minshall_check(const struct tcp_opt *tp)
1431 return after(tp->snd_sml,tp->snd_una) &&
1432 !after(tp->snd_sml, tp->snd_nxt);
1435 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss,
1436 const struct sk_buff *skb)
1439 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1442 /* Return 0, if packet can be sent now without violation Nagle's rules:
1443 1. It is full sized.
1444 2. Or it contains FIN.
1445 3. Or TCP_NODELAY was set.
1446 4. Or TCP_CORK is not set, and all sent packets are ACKed.
1447 With Minshall's modification: all sent small packets are ACKed.
1450 static __inline__ int
1451 tcp_nagle_check(const struct tcp_opt *tp, const struct sk_buff *skb,
1452 unsigned mss_now, int nonagle)
1454 return (skb->len < mss_now &&
1455 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1456 ((nonagle&TCP_NAGLE_CORK) ||
1458 tcp_get_pcount(&tp->packets_out) &&
1459 tcp_minshall_check(tp))));
1462 extern void tcp_set_skb_tso_segs(struct sk_buff *, unsigned int);
1464 /* This checks if the data bearing packet SKB (usually sk->sk_send_head)
1465 * should be put on the wire right now.
1467 static __inline__ int tcp_snd_test(const struct tcp_opt *tp,
1468 struct sk_buff *skb,
1469 unsigned cur_mss, int nonagle)
1471 int pkts = tcp_skb_pcount(skb);
1474 tcp_set_skb_tso_segs(skb, tp->mss_cache_std);
1475 pkts = tcp_skb_pcount(skb);
1478 /* RFC 1122 - section 4.2.3.4
1482 * a) The right edge of this frame exceeds the window
1483 * b) There are packets in flight and we have a small segment
1484 * [SWS avoidance and Nagle algorithm]
1485 * (part of SWS is done on packetization)
1486 * Minshall version sounds: there are no _small_
1487 * segments in flight. (tcp_nagle_check)
1488 * c) We have too many packets 'in flight'
1490 * Don't use the nagle rule for urgent data (or
1491 * for the final FIN -DaveM).
1493 * Also, Nagle rule does not apply to frames, which
1494 * sit in the middle of queue (they have no chances
1495 * to get new data) and if room at tail of skb is
1496 * not enough to save something seriously (<32 for now).
1499 /* Don't be strict about the congestion window for the
1500 * final FIN frame. -DaveM
1502 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1503 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1504 (((tcp_packets_in_flight(tp) + (pkts-1)) < tp->snd_cwnd) ||
1505 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1506 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1509 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1511 if (!tcp_get_pcount(&tp->packets_out) && !tp->pending)
1512 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1515 static __inline__ int tcp_skb_is_last(const struct sock *sk,
1516 const struct sk_buff *skb)
1518 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1521 /* Push out any pending frames which were held back due to
1522 * TCP_CORK or attempt at coalescing tiny packets.
1523 * The socket must be locked by the caller.
1525 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1530 struct sk_buff *skb = sk->sk_send_head;
1533 if (!tcp_skb_is_last(sk, skb))
1534 nonagle = TCP_NAGLE_PUSH;
1535 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1536 tcp_write_xmit(sk, nonagle))
1537 tcp_check_probe_timer(sk, tp);
1539 tcp_cwnd_validate(sk, tp);
1542 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1545 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1548 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1550 struct sk_buff *skb = sk->sk_send_head;
1553 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1554 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1557 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1562 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1567 extern void tcp_destroy_sock(struct sock *sk);
1571 * Calculate(/check) TCP checksum
1573 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1574 unsigned long saddr, unsigned long daddr,
1577 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1580 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1582 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1585 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1587 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1588 __tcp_checksum_complete(skb);
1591 /* Prequeue for VJ style copy to user, combined with checksumming. */
1593 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1595 tp->ucopy.task = NULL;
1597 tp->ucopy.memory = 0;
1598 skb_queue_head_init(&tp->ucopy.prequeue);
1601 /* Packet is added to VJ-style prequeue for processing in process
1602 * context, if a reader task is waiting. Apparently, this exciting
1603 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1604 * failed somewhere. Latency? Burstiness? Well, at least now we will
1605 * see, why it failed. 8)8) --ANK
1607 * NOTE: is this not too big to inline?
1609 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1611 struct tcp_opt *tp = tcp_sk(sk);
1613 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1614 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1615 tp->ucopy.memory += skb->truesize;
1616 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1617 struct sk_buff *skb1;
1619 BUG_ON(sock_owned_by_user(sk));
1621 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1622 sk->sk_backlog_rcv(sk, skb1);
1623 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1626 tp->ucopy.memory = 0;
1627 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1628 wake_up_interruptible(sk->sk_sleep);
1629 if (!tcp_ack_scheduled(tp))
1630 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1641 static const char *statename[]={
1642 "Unused","Established","Syn Sent","Syn Recv",
1643 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1644 "Close Wait","Last ACK","Listen","Closing"
1648 static __inline__ void tcp_set_state(struct sock *sk, int state)
1650 int oldstate = sk->sk_state;
1653 case TCP_ESTABLISHED:
1654 if (oldstate != TCP_ESTABLISHED)
1655 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1659 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1660 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1662 sk->sk_prot->unhash(sk);
1663 if (tcp_sk(sk)->bind_hash &&
1664 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1668 if (oldstate==TCP_ESTABLISHED)
1669 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1672 /* Change state AFTER socket is unhashed to avoid closed
1673 * socket sitting in hash tables.
1675 sk->sk_state = state;
1678 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1682 static __inline__ void tcp_done(struct sock *sk)
1684 tcp_set_state(sk, TCP_CLOSE);
1685 tcp_clear_xmit_timers(sk);
1687 sk->sk_shutdown = SHUTDOWN_MASK;
1689 if (!sock_flag(sk, SOCK_DEAD))
1690 sk->sk_state_change(sk);
1692 tcp_destroy_sock(sk);
1695 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1702 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1704 if (tp->tstamp_ok) {
1705 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1706 (TCPOPT_NOP << 16) |
1707 (TCPOPT_TIMESTAMP << 8) |
1709 *ptr++ = htonl(tstamp);
1710 *ptr++ = htonl(tp->ts_recent);
1712 if (tp->eff_sacks) {
1713 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1716 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1717 (TCPOPT_NOP << 16) |
1718 (TCPOPT_SACK << 8) |
1719 (TCPOLEN_SACK_BASE +
1720 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1721 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1722 *ptr++ = htonl(sp[this_sack].start_seq);
1723 *ptr++ = htonl(sp[this_sack].end_seq);
1732 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1733 * If this is every changed make sure to change the definition of
1734 * MAX_SYN_SIZE to match the new maximum number of options that you
1737 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1738 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1740 /* We always get an MSS option.
1741 * The option bytes which will be seen in normal data
1742 * packets should timestamps be used, must be in the MSS
1743 * advertised. But we subtract them from tp->mss_cache so
1744 * that calculations in tcp_sendmsg are simpler etc.
1745 * So account for this fact here if necessary. If we
1746 * don't do this correctly, as a receiver we won't
1747 * recognize data packets as being full sized when we
1748 * should, and thus we won't abide by the delayed ACK
1750 * SACKs don't matter, we never delay an ACK when we
1751 * have any of those going out.
1753 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1756 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1757 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1759 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1760 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1761 *ptr++ = htonl(tstamp); /* TSVAL */
1762 *ptr++ = htonl(ts_recent); /* TSECR */
1764 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1765 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1767 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1770 /* Determine a window scaling and initial window to offer. */
1771 extern void tcp_select_initial_window(int __space, __u32 mss,
1772 __u32 *rcv_wnd, __u32 *window_clamp,
1773 int wscale_ok, __u8 *rcv_wscale);
1775 static inline int tcp_win_from_space(int space)
1777 return sysctl_tcp_adv_win_scale<=0 ?
1778 (space>>(-sysctl_tcp_adv_win_scale)) :
1779 space - (space>>sysctl_tcp_adv_win_scale);
1782 /* Note: caller must be prepared to deal with negative returns */
1783 static inline int tcp_space(const struct sock *sk)
1785 return tcp_win_from_space(sk->sk_rcvbuf -
1786 atomic_read(&sk->sk_rmem_alloc));
1789 static inline int tcp_full_space(const struct sock *sk)
1791 return tcp_win_from_space(sk->sk_rcvbuf);
1794 struct tcp_listen_opt
1796 u8 max_qlen_log; /* log_2 of maximal queued SYNs */
1798 #ifdef CONFIG_ACCEPT_QUEUES
1799 int qlen_young[NUM_ACCEPT_QUEUES];
1805 struct open_request *syn_table[TCP_SYNQ_HSIZE];
1808 #ifdef CONFIG_ACCEPT_QUEUES
1809 static inline void sk_acceptq_removed(struct sock *sk, int class)
1811 tcp_sk(sk)->acceptq[class].aq_backlog--;
1814 static inline void sk_acceptq_added(struct sock *sk, int class)
1816 tcp_sk(sk)->acceptq[class].aq_backlog++;
1819 static inline int sk_acceptq_is_full(struct sock *sk, int class)
1821 return tcp_sk(sk)->acceptq[class].aq_backlog >
1822 sk->sk_max_ack_backlog;
1825 static inline void tcp_set_acceptq(struct tcp_opt *tp, struct open_request *req)
1827 int class = req->acceptq_class;
1830 if (!tp->acceptq[class].aq_ratio) {
1831 req->acceptq_class = 0;
1835 tp->acceptq[class].aq_qcount++;
1836 req->acceptq_time_stamp = jiffies;
1838 if (tp->acceptq[class].aq_tail) {
1839 req->dl_next = tp->acceptq[class].aq_tail->dl_next;
1840 tp->acceptq[class].aq_tail->dl_next = req;
1841 tp->acceptq[class].aq_tail = req;
1842 } else { /* if first request in the class */
1843 tp->acceptq[class].aq_head = req;
1844 tp->acceptq[class].aq_tail = req;
1846 prev_class = class - 1;
1847 while (prev_class >= 0) {
1848 if (tp->acceptq[prev_class].aq_tail)
1852 if (prev_class < 0) {
1853 req->dl_next = tp->accept_queue;
1854 tp->accept_queue = req;
1857 req->dl_next = tp->acceptq[prev_class].aq_tail->dl_next;
1858 tp->acceptq[prev_class].aq_tail->dl_next = req;
1862 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1865 tcp_set_acceptq(tcp_sk(sk),req);
1867 sk_acceptq_added(sk,req->acceptq_class);
1871 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1874 struct tcp_opt *tp = tcp_sk(sk);
1877 sk_acceptq_added(sk);
1879 if (!tp->accept_queue_tail) {
1880 tp->accept_queue = req;
1882 tp->accept_queue_tail->dl_next = req;
1884 tp->accept_queue_tail = req;
1885 req->dl_next = NULL;
1891 #ifdef CONFIG_ACCEPT_QUEUES
1893 tcp_synq_removed(struct sock *sk, struct open_request *req)
1895 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1897 if (--lopt->qlen == 0)
1898 tcp_delete_keepalive_timer(sk);
1899 if (req->retrans == 0)
1900 lopt->qlen_young[req->acceptq_class]--;
1903 static inline void tcp_synq_added(struct sock *sk, struct open_request *req)
1905 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1907 if (lopt->qlen++ == 0)
1908 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1909 lopt->qlen_young[req->acceptq_class]++;
1912 static inline int tcp_synq_len(struct sock *sk)
1914 return tcp_sk(sk)->listen_opt->qlen;
1917 static inline int tcp_synq_young(struct sock *sk, int class)
1919 return tcp_sk(sk)->listen_opt->qlen_young[class];
1925 tcp_synq_removed(struct sock *sk, struct open_request *req)
1927 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1929 if (--lopt->qlen == 0)
1930 tcp_delete_keepalive_timer(sk);
1931 if (req->retrans == 0)
1935 static inline void tcp_synq_added(struct sock *sk)
1937 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1939 if (lopt->qlen++ == 0)
1940 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1944 static inline int tcp_synq_len(struct sock *sk)
1946 return tcp_sk(sk)->listen_opt->qlen;
1949 static inline int tcp_synq_young(struct sock *sk)
1951 return tcp_sk(sk)->listen_opt->qlen_young;
1955 static inline int tcp_synq_is_full(struct sock *sk)
1957 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1960 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1961 struct open_request **prev)
1963 write_lock(&tp->syn_wait_lock);
1964 *prev = req->dl_next;
1965 write_unlock(&tp->syn_wait_lock);
1968 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1969 struct open_request **prev)
1971 tcp_synq_unlink(tcp_sk(sk), req, prev);
1972 tcp_synq_removed(sk, req);
1973 tcp_openreq_free(req);
1976 static __inline__ void tcp_openreq_init(struct open_request *req,
1978 struct sk_buff *skb)
1980 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1981 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1982 req->mss = tp->mss_clamp;
1983 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1984 req->tstamp_ok = tp->tstamp_ok;
1985 req->sack_ok = tp->sack_ok;
1986 req->snd_wscale = tp->snd_wscale;
1987 req->wscale_ok = tp->wscale_ok;
1990 req->rmt_port = skb->h.th->source;
1993 extern void tcp_enter_memory_pressure(void);
1995 extern void tcp_listen_wlock(void);
1997 /* - We may sleep inside this lock.
1998 * - If sleeping is not required (or called from BH),
1999 * use plain read_(un)lock(&tcp_lhash_lock).
2002 static inline void tcp_listen_lock(void)
2004 /* read_lock synchronizes to candidates to writers */
2005 read_lock(&tcp_lhash_lock);
2006 atomic_inc(&tcp_lhash_users);
2007 read_unlock(&tcp_lhash_lock);
2010 static inline void tcp_listen_unlock(void)
2012 if (atomic_dec_and_test(&tcp_lhash_users))
2013 wake_up(&tcp_lhash_wait);
2016 static inline int keepalive_intvl_when(const struct tcp_opt *tp)
2018 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
2021 static inline int keepalive_time_when(const struct tcp_opt *tp)
2023 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
2026 static inline int tcp_fin_time(const struct tcp_opt *tp)
2028 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
2030 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
2031 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
2036 static inline int tcp_paws_check(const struct tcp_opt *tp, int rst)
2038 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
2040 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
2043 /* RST segments are not recommended to carry timestamp,
2044 and, if they do, it is recommended to ignore PAWS because
2045 "their cleanup function should take precedence over timestamps."
2046 Certainly, it is mistake. It is necessary to understand the reasons
2047 of this constraint to relax it: if peer reboots, clock may go
2048 out-of-sync and half-open connections will not be reset.
2049 Actually, the problem would be not existing if all
2050 the implementations followed draft about maintaining clock
2051 via reboots. Linux-2.2 DOES NOT!
2053 However, we can relax time bounds for RST segments to MSL.
2055 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
2060 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
2062 sk->sk_route_caps = dst->dev->features;
2063 if (sk->sk_route_caps & NETIF_F_TSO) {
2064 if (sk->sk_no_largesend || dst->header_len)
2065 sk->sk_route_caps &= ~NETIF_F_TSO;
2069 #define TCP_CHECK_TIMER(sk) do { } while (0)
2071 static inline int tcp_use_frto(const struct sock *sk)
2073 const struct tcp_opt *tp = tcp_sk(sk);
2075 /* F-RTO must be activated in sysctl and there must be some
2076 * unsent new data, and the advertised window should allow
2079 return (sysctl_tcp_frto && sk->sk_send_head &&
2080 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
2081 tp->snd_una + tp->snd_wnd));
2084 static inline void tcp_mib_init(void)
2087 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
2088 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
2089 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
2090 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
2094 enum tcp_seq_states {
2095 TCP_SEQ_STATE_LISTENING,
2096 TCP_SEQ_STATE_OPENREQ,
2097 TCP_SEQ_STATE_ESTABLISHED,
2098 TCP_SEQ_STATE_TIME_WAIT,
2101 struct tcp_seq_afinfo {
2102 struct module *owner;
2105 int (*seq_show) (struct seq_file *m, void *v);
2106 struct file_operations *seq_fops;
2109 struct tcp_iter_state {
2111 enum tcp_seq_states state;
2112 struct sock *syn_wait_sk;
2113 int bucket, sbucket, num, uid;
2114 struct seq_operations seq_ops;
2117 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2118 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2120 /* TCP Westwood functions and constants */
2122 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2123 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2125 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2127 if (tcp_is_westwood(tp))
2128 tp->westwood.rtt = rtt_seq;
2131 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2132 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2134 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2136 if (tcp_is_westwood(tcp_sk(sk)))
2137 __tcp_westwood_fast_bw(sk, skb);
2140 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2142 if (tcp_is_westwood(tcp_sk(sk)))
2143 __tcp_westwood_slow_bw(sk, skb);
2146 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2148 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2149 (__u32) (tp->mss_cache_std),
2153 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2155 return tcp_is_westwood(tp) ? __tcp_westwood_bw_rttmin(tp) : 0;
2158 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2162 if (tcp_is_westwood(tp)) {
2163 ssthresh = __tcp_westwood_bw_rttmin(tp);
2165 tp->snd_ssthresh = ssthresh;
2168 return (ssthresh != 0);
2171 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2175 if (tcp_is_westwood(tp)) {
2176 cwnd = __tcp_westwood_bw_rttmin(tp);
2178 tp->snd_cwnd = cwnd;