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 #define tw_xid __tw_common.skc_xid
199 #define tw_vx_info __tw_common.skc_vx_info
200 #define tw_nid __tw_common.skc_nid
201 #define tw_nx_info __tw_common.skc_nx_info
202 volatile unsigned char tw_substate;
203 unsigned char tw_rcv_wscale;
205 /* Socket demultiplex comparisons on incoming packets. */
206 /* these five are in inet_opt */
208 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
212 /* And these are ours. */
219 long tw_ts_recent_stamp;
220 unsigned long tw_ttd;
221 struct tcp_bind_bucket *tw_tb;
222 struct hlist_node tw_death_node;
223 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
224 struct in6_addr tw_v6_daddr;
225 struct in6_addr tw_v6_rcv_saddr;
230 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
231 struct hlist_head *list)
233 hlist_add_head(&tw->tw_node, list);
236 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
237 struct hlist_head *list)
239 hlist_add_head(&tw->tw_bind_node, list);
242 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
244 return tw->tw_death_node.pprev != NULL;
247 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
249 tw->tw_death_node.pprev = NULL;
252 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
254 __hlist_del(&tw->tw_death_node);
255 tw_dead_node_init(tw);
258 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
260 if (tw_dead_hashed(tw)) {
261 __tw_del_dead_node(tw);
267 #define tw_for_each(tw, node, head) \
268 hlist_for_each_entry(tw, node, head, tw_node)
270 #define tw_for_each_inmate(tw, node, jail) \
271 hlist_for_each_entry(tw, node, jail, tw_death_node)
273 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
274 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
276 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
278 static inline const u32 tcp_v4_rcv_saddr(const struct sock *sk)
280 return likely(sk->sk_state != TCP_TIME_WAIT) ?
281 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
284 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
285 static inline const struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
287 return likely(sk->sk_state != TCP_TIME_WAIT) ?
288 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
291 static inline const struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
293 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
296 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
298 static inline int tcp_v6_ipv6only(const struct sock *sk)
300 return likely(sk->sk_state != TCP_TIME_WAIT) ?
301 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
304 # define __tcp_v6_rcv_saddr(__sk) NULL
305 # define tcp_v6_rcv_saddr(__sk) NULL
306 # define tcptw_sk_ipv6only(__sk) 0
307 # define tcp_v6_ipv6only(__sk) 0
310 extern kmem_cache_t *tcp_timewait_cachep;
312 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
314 if (atomic_dec_and_test(&tw->tw_refcnt)) {
315 #ifdef INET_REFCNT_DEBUG
316 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
318 kmem_cache_free(tcp_timewait_cachep, tw);
322 extern atomic_t tcp_orphan_count;
323 extern int tcp_tw_count;
324 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
325 extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
326 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
329 /* Socket demux engine toys. */
331 #define TCP_COMBINED_PORTS(__sport, __dport) \
332 (((__u32)(__sport)<<16) | (__u32)(__dport))
333 #else /* __LITTLE_ENDIAN */
334 #define TCP_COMBINED_PORTS(__sport, __dport) \
335 (((__u32)(__dport)<<16) | (__u32)(__sport))
338 #if (BITS_PER_LONG == 64)
340 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
341 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
342 #else /* __LITTLE_ENDIAN */
343 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
344 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
345 #endif /* __BIG_ENDIAN */
346 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
347 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
348 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
349 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
350 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
351 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
352 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
353 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
354 #else /* 32-bit arch */
355 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
356 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
357 ((inet_sk(__sk)->daddr == (__saddr)) && \
358 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
359 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
360 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
361 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
362 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
363 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
364 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
365 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
366 #endif /* 64-bit arch */
368 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
369 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
370 ((__sk)->sk_family == AF_INET6) && \
371 !ipv6_addr_cmp(&inet6_sk(__sk)->daddr, (__saddr)) && \
372 !ipv6_addr_cmp(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
373 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
375 /* These can have wildcards, don't try too hard. */
376 static __inline__ int tcp_lhashfn(unsigned short num)
378 return num & (TCP_LHTABLE_SIZE - 1);
381 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
383 return tcp_lhashfn(inet_sk(sk)->num);
386 #define MAX_TCP_HEADER (128 + MAX_HEADER)
389 * Never offer a window over 32767 without using window scaling. Some
390 * poor stacks do signed 16bit maths!
392 #define MAX_TCP_WINDOW 32767U
394 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
395 #define TCP_MIN_MSS 88U
397 /* Minimal RCV_MSS. */
398 #define TCP_MIN_RCVMSS 536U
400 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
401 #define TCP_FASTRETRANS_THRESH 3
403 /* Maximal reordering. */
404 #define TCP_MAX_REORDERING 127
406 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
407 #define TCP_MAX_QUICKACKS 16U
409 /* urg_data states */
410 #define TCP_URG_VALID 0x0100
411 #define TCP_URG_NOTYET 0x0200
412 #define TCP_URG_READ 0x0400
414 #define TCP_RETR1 3 /*
415 * This is how many retries it does before it
416 * tries to figure out if the gateway is
417 * down. Minimal RFC value is 3; it corresponds
418 * to ~3sec-8min depending on RTO.
421 #define TCP_RETR2 15 /*
422 * This should take at least
423 * 90 minutes to time out.
424 * RFC1122 says that the limit is 100 sec.
425 * 15 is ~13-30min depending on RTO.
428 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
429 * connection: ~180sec is RFC minumum */
431 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
432 * connection: ~180sec is RFC minumum */
435 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
436 * socket. 7 is ~50sec-16min.
440 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
441 * state, about 60 seconds */
442 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
443 /* BSD style FIN_WAIT2 deadlock breaker.
444 * It used to be 3min, new value is 60sec,
445 * to combine FIN-WAIT-2 timeout with
449 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
451 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
452 #define TCP_ATO_MIN ((unsigned)(HZ/25))
454 #define TCP_DELACK_MIN 4U
455 #define TCP_ATO_MIN 4U
457 #define TCP_RTO_MAX ((unsigned)(120*HZ))
458 #define TCP_RTO_MIN ((unsigned)(HZ/5))
459 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
461 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
462 * for local resources.
465 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
466 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
467 #define TCP_KEEPALIVE_INTVL (75*HZ)
469 #define MAX_TCP_KEEPIDLE 32767
470 #define MAX_TCP_KEEPINTVL 32767
471 #define MAX_TCP_KEEPCNT 127
472 #define MAX_TCP_SYNCNT 127
474 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
475 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
477 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
478 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
479 * after this time. It should be equal
480 * (or greater than) TCP_TIMEWAIT_LEN
481 * to provide reliability equal to one
482 * provided by timewait state.
484 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
485 * timestamps. It must be less than
486 * minimal timewait lifetime.
489 #define TCP_TW_RECYCLE_SLOTS_LOG 5
490 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
492 /* If time > 4sec, it is "slow" path, no recycling is required,
493 so that we select tick to get range about 4 seconds.
496 #if HZ <= 16 || HZ > 4096
497 # error Unsupported: HZ <= 16 or HZ > 4096
499 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
501 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
503 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
505 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
507 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
509 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
511 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
513 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
516 #define BICTCP_1_OVER_BETA 8 /*
518 * multiplicative decrease factor
520 #define BICTCP_MAX_INCREMENT 32 /*
521 * Limit on the amount of
522 * increment allowed during
525 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
526 * log(B/Smin)/log(B/(B-1))+1,
530 #define BICTCP_B 4 /*
532 * go to point (max+min)/N
539 #define TCPOPT_NOP 1 /* Padding */
540 #define TCPOPT_EOL 0 /* End of options */
541 #define TCPOPT_MSS 2 /* Segment size negotiating */
542 #define TCPOPT_WINDOW 3 /* Window scaling */
543 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
544 #define TCPOPT_SACK 5 /* SACK Block */
545 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
551 #define TCPOLEN_MSS 4
552 #define TCPOLEN_WINDOW 3
553 #define TCPOLEN_SACK_PERM 2
554 #define TCPOLEN_TIMESTAMP 10
556 /* But this is what stacks really send out. */
557 #define TCPOLEN_TSTAMP_ALIGNED 12
558 #define TCPOLEN_WSCALE_ALIGNED 4
559 #define TCPOLEN_SACKPERM_ALIGNED 4
560 #define TCPOLEN_SACK_BASE 2
561 #define TCPOLEN_SACK_BASE_ALIGNED 4
562 #define TCPOLEN_SACK_PERBLOCK 8
564 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
565 #define TCP_TIME_DACK 2 /* Delayed ack timer */
566 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
567 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
569 /* Flags in tp->nonagle */
570 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
571 #define TCP_NAGLE_CORK 2 /* Socket is corked */
572 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
574 /* sysctl variables for tcp */
575 extern int sysctl_max_syn_backlog;
576 extern int sysctl_tcp_timestamps;
577 extern int sysctl_tcp_window_scaling;
578 extern int sysctl_tcp_sack;
579 extern int sysctl_tcp_fin_timeout;
580 extern int sysctl_tcp_tw_recycle;
581 extern int sysctl_tcp_keepalive_time;
582 extern int sysctl_tcp_keepalive_probes;
583 extern int sysctl_tcp_keepalive_intvl;
584 extern int sysctl_tcp_syn_retries;
585 extern int sysctl_tcp_synack_retries;
586 extern int sysctl_tcp_retries1;
587 extern int sysctl_tcp_retries2;
588 extern int sysctl_tcp_orphan_retries;
589 extern int sysctl_tcp_syncookies;
590 extern int sysctl_tcp_retrans_collapse;
591 extern int sysctl_tcp_stdurg;
592 extern int sysctl_tcp_rfc1337;
593 extern int sysctl_tcp_abort_on_overflow;
594 extern int sysctl_tcp_max_orphans;
595 extern int sysctl_tcp_max_tw_buckets;
596 extern int sysctl_tcp_fack;
597 extern int sysctl_tcp_reordering;
598 extern int sysctl_tcp_ecn;
599 extern int sysctl_tcp_dsack;
600 extern int sysctl_tcp_mem[3];
601 extern int sysctl_tcp_wmem[3];
602 extern int sysctl_tcp_rmem[3];
603 extern int sysctl_tcp_app_win;
604 extern int sysctl_tcp_adv_win_scale;
605 extern int sysctl_tcp_tw_reuse;
606 extern int sysctl_tcp_frto;
607 extern int sysctl_tcp_low_latency;
608 extern int sysctl_tcp_westwood;
609 extern int sysctl_tcp_vegas_cong_avoid;
610 extern int sysctl_tcp_vegas_alpha;
611 extern int sysctl_tcp_vegas_beta;
612 extern int sysctl_tcp_vegas_gamma;
613 extern int sysctl_tcp_nometrics_save;
614 extern int sysctl_tcp_bic;
615 extern int sysctl_tcp_bic_fast_convergence;
616 extern int sysctl_tcp_bic_low_window;
618 extern atomic_t tcp_memory_allocated;
619 extern atomic_t tcp_sockets_allocated;
620 extern int tcp_memory_pressure;
624 struct or_calltable {
626 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
627 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
628 void (*destructor) (struct open_request *req);
629 void (*send_reset) (struct sk_buff *skb);
632 struct tcp_v4_open_req {
635 struct ip_options *opt;
638 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
639 struct tcp_v6_open_req {
640 struct in6_addr loc_addr;
641 struct in6_addr rmt_addr;
642 struct sk_buff *pktopts;
647 /* this structure is too big */
648 struct open_request {
649 struct open_request *dl_next; /* Must be first member! */
656 __u16 snd_wscale : 4,
663 /* The following two fields can be easily recomputed I think -AK */
664 __u32 window_clamp; /* window clamp at creation time */
665 __u32 rcv_wnd; /* rcv_wnd offered first time */
667 unsigned long expires;
668 struct or_calltable *class;
671 struct tcp_v4_open_req v4_req;
672 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
673 struct tcp_v6_open_req v6_req;
678 /* SLAB cache for open requests. */
679 extern kmem_cache_t *tcp_openreq_cachep;
681 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
682 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
684 static inline void tcp_openreq_free(struct open_request *req)
686 req->class->destructor(req);
687 tcp_openreq_fastfree(req);
690 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
691 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
693 #define TCP_INET_FAMILY(fam) 1
697 * Pointers to address related TCP functions
698 * (i.e. things that depend on the address family)
702 int (*queue_xmit) (struct sk_buff *skb,
705 void (*send_check) (struct sock *sk,
708 struct sk_buff *skb);
710 int (*rebuild_header) (struct sock *sk);
712 int (*conn_request) (struct sock *sk,
713 struct sk_buff *skb);
715 struct sock * (*syn_recv_sock) (struct sock *sk,
717 struct open_request *req,
718 struct dst_entry *dst);
720 int (*remember_stamp) (struct sock *sk);
722 __u16 net_header_len;
724 int (*setsockopt) (struct sock *sk,
730 int (*getsockopt) (struct sock *sk,
737 void (*addr2sockaddr) (struct sock *sk,
744 * The next routines deal with comparing 32 bit unsigned ints
745 * and worry about wraparound (automatic with unsigned arithmetic).
748 static inline int before(__u32 seq1, __u32 seq2)
750 return (__s32)(seq1-seq2) < 0;
753 static inline int after(__u32 seq1, __u32 seq2)
755 return (__s32)(seq2-seq1) < 0;
759 /* is s2<=s1<=s3 ? */
760 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
762 return seq3 - seq2 >= seq1 - seq2;
766 extern struct proto tcp_prot;
768 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
769 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
770 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
771 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
772 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
773 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
774 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
776 extern void tcp_put_port(struct sock *sk);
777 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
779 extern void tcp_v4_err(struct sk_buff *skb, u32);
781 extern void tcp_shutdown (struct sock *sk, int how);
783 extern int tcp_v4_rcv(struct sk_buff *skb);
785 extern int tcp_v4_remember_stamp(struct sock *sk);
787 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
789 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
790 struct msghdr *msg, size_t size);
791 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
793 extern int tcp_ioctl(struct sock *sk,
797 extern int tcp_rcv_state_process(struct sock *sk,
802 extern int tcp_rcv_established(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,
876 int optname, char *optval,
878 extern int tcp_setsockopt(struct sock *sk, int level,
879 int optname, char *optval,
881 extern void tcp_set_keepalive(struct sock *sk, int val);
882 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
884 size_t len, int nonblock,
885 int flags, int *addr_len);
887 extern int tcp_listen_start(struct sock *sk);
889 extern void tcp_parse_options(struct sk_buff *skb,
894 * TCP v4 functions exported for the inet6 API
897 extern int tcp_v4_rebuild_header(struct sock *sk);
899 extern int tcp_v4_build_header(struct sock *sk,
900 struct sk_buff *skb);
902 extern void tcp_v4_send_check(struct sock *sk,
903 struct tcphdr *th, int len,
904 struct sk_buff *skb);
906 extern int tcp_v4_conn_request(struct sock *sk,
907 struct sk_buff *skb);
909 extern struct sock * tcp_create_openreq_child(struct sock *sk,
910 struct open_request *req,
911 struct sk_buff *skb);
913 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
915 struct open_request *req,
916 struct dst_entry *dst);
918 extern int tcp_v4_do_rcv(struct sock *sk,
919 struct sk_buff *skb);
921 extern int tcp_v4_connect(struct sock *sk,
922 struct sockaddr *uaddr,
925 extern int tcp_connect(struct sock *sk);
927 extern struct sk_buff * tcp_make_synack(struct sock *sk,
928 struct dst_entry *dst,
929 struct open_request *req);
931 extern int tcp_disconnect(struct sock *sk, int flags);
933 extern void tcp_unhash(struct sock *sk);
935 extern int tcp_v4_hash_connecting(struct sock *sk);
938 /* From syncookies.c */
939 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
940 struct ip_options *opt);
941 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
946 extern int tcp_write_xmit(struct sock *, int nonagle);
947 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
948 extern void tcp_xmit_retransmit_queue(struct sock *);
949 extern void tcp_simple_retransmit(struct sock *);
951 extern void tcp_send_probe0(struct sock *);
952 extern void tcp_send_partial(struct sock *);
953 extern int tcp_write_wakeup(struct sock *);
954 extern void tcp_send_fin(struct sock *sk);
955 extern void tcp_send_active_reset(struct sock *sk, int priority);
956 extern int tcp_send_synack(struct sock *);
957 extern int tcp_transmit_skb(struct sock *, struct sk_buff *);
958 extern void tcp_push_one(struct sock *, unsigned mss_now);
959 extern void tcp_send_ack(struct sock *sk);
960 extern void tcp_send_delayed_ack(struct sock *sk);
963 extern void tcp_init_xmit_timers(struct sock *);
964 extern void tcp_clear_xmit_timers(struct sock *);
966 extern void tcp_delete_keepalive_timer (struct sock *);
967 extern void tcp_reset_keepalive_timer (struct sock *, unsigned long);
968 extern int tcp_sync_mss(struct sock *sk, u32 pmtu);
970 extern const char timer_bug_msg[];
972 /* Read 'sendfile()'-style from a TCP socket */
973 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
974 unsigned int, size_t);
975 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
976 sk_read_actor_t recv_actor);
978 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
980 struct tcp_opt *tp = tcp_sk(sk);
983 case TCP_TIME_RETRANS:
984 case TCP_TIME_PROBE0:
987 #ifdef TCP_CLEAR_TIMERS
988 if (timer_pending(&tp->retransmit_timer) &&
989 del_timer(&tp->retransmit_timer))
997 #ifdef TCP_CLEAR_TIMERS
998 if (timer_pending(&tp->delack_timer) &&
999 del_timer(&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 if (!mod_timer(&tp->retransmit_timer, tp->timeout))
1034 tp->ack.pending |= TCP_ACK_TIMER;
1035 tp->ack.timeout = jiffies+when;
1036 if (!mod_timer(&tp->delack_timer, tp->ack.timeout))
1041 printk(KERN_DEBUG "bug: unknown timer value\n");
1045 /* Compute the current effective MSS, taking SACKs and IP options,
1046 * and even PMTU discovery events into account.
1048 * LARGESEND note: !urg_mode is overkill, only frames up to snd_up
1049 * cannot be large. However, taking into account rare use of URG, this
1050 * is not a big flaw.
1053 static __inline__ unsigned int tcp_current_mss(struct sock *sk, int large)
1055 struct tcp_opt *tp = tcp_sk(sk);
1056 struct dst_entry *dst = __sk_dst_get(sk);
1057 int mss_now = large && (sk->sk_route_caps & NETIF_F_TSO) &&
1059 tp->mss_cache : tp->mss_cache_std;
1062 u32 mtu = dst_pmtu(dst);
1063 if (mtu != tp->pmtu_cookie ||
1064 tp->ext2_header_len != dst->header_len)
1065 mss_now = tcp_sync_mss(sk, mtu);
1068 mss_now -= (TCPOLEN_SACK_BASE_ALIGNED +
1069 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK));
1073 /* Initialize RCV_MSS value.
1074 * RCV_MSS is an our guess about MSS used by the peer.
1075 * We haven't any direct information about the MSS.
1076 * It's better to underestimate the RCV_MSS rather than overestimate.
1077 * Overestimations make us ACKing less frequently than needed.
1078 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1081 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1083 struct tcp_opt *tp = tcp_sk(sk);
1084 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1086 hint = min(hint, tp->rcv_wnd/2);
1087 hint = min(hint, TCP_MIN_RCVMSS);
1088 hint = max(hint, TCP_MIN_MSS);
1090 tp->ack.rcv_mss = hint;
1093 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1095 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1096 ntohl(TCP_FLAG_ACK) |
1100 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1102 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1105 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1107 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1109 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1111 tcp_fast_path_on(tp);
1114 /* Compute the actual receive window we are currently advertising.
1115 * Rcv_nxt can be after the window if our peer push more data
1116 * than the offered window.
1118 static __inline__ u32 tcp_receive_window(struct tcp_opt *tp)
1120 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1127 /* Choose a new window, without checks for shrinking, and without
1128 * scaling applied to the result. The caller does these things
1129 * if necessary. This is a "raw" window selection.
1131 extern u32 __tcp_select_window(struct sock *sk);
1133 /* TCP timestamps are only 32-bits, this causes a slight
1134 * complication on 64-bit systems since we store a snapshot
1135 * of jiffies in the buffer control blocks below. We decidely
1136 * only use of the low 32-bits of jiffies and hide the ugly
1137 * casts with the following macro.
1139 #define tcp_time_stamp ((__u32)(jiffies))
1141 /* This is what the send packet queueing engine uses to pass
1142 * TCP per-packet control information to the transmission
1143 * code. We also store the host-order sequence numbers in
1144 * here too. This is 36 bytes on 32-bit architectures,
1145 * 40 bytes on 64-bit machines, if this grows please adjust
1146 * skbuff.h:skbuff->cb[xxx] size appropriately.
1150 struct inet_skb_parm h4;
1151 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1152 struct inet6_skb_parm h6;
1154 } header; /* For incoming frames */
1155 __u32 seq; /* Starting sequence number */
1156 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1157 __u32 when; /* used to compute rtt's */
1158 __u8 flags; /* TCP header flags. */
1160 /* NOTE: These must match up to the flags byte in a
1163 #define TCPCB_FLAG_FIN 0x01
1164 #define TCPCB_FLAG_SYN 0x02
1165 #define TCPCB_FLAG_RST 0x04
1166 #define TCPCB_FLAG_PSH 0x08
1167 #define TCPCB_FLAG_ACK 0x10
1168 #define TCPCB_FLAG_URG 0x20
1169 #define TCPCB_FLAG_ECE 0x40
1170 #define TCPCB_FLAG_CWR 0x80
1172 __u8 sacked; /* State flags for SACK/FACK. */
1173 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1174 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1175 #define TCPCB_LOST 0x04 /* SKB is lost */
1176 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1178 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1179 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1181 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1183 #define TCPCB_AT_TAIL (TCPCB_URG)
1185 __u16 urg_ptr; /* Valid w/URG flags is set. */
1186 __u32 ack_seq; /* Sequence number ACK'd */
1189 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1191 #define for_retrans_queue(skb, sk, tp) \
1192 for (skb = (sk)->sk_write_queue.next; \
1193 (skb != (tp)->send_head) && \
1194 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1198 #include <net/tcp_ecn.h>
1202 * Compute minimal free write space needed to queue new packets.
1204 static inline int tcp_min_write_space(struct sock *sk)
1206 return sk->sk_wmem_queued / 2;
1209 static inline int tcp_wspace(struct sock *sk)
1211 return sk->sk_sndbuf - sk->sk_wmem_queued;
1215 /* This determines how many packets are "in the network" to the best
1216 * of our knowledge. In many cases it is conservative, but where
1217 * detailed information is available from the receiver (via SACK
1218 * blocks etc.) we can make more aggressive calculations.
1220 * Use this for decisions involving congestion control, use just
1221 * tp->packets_out to determine if the send queue is empty or not.
1223 * Read this equation as:
1225 * "Packets sent once on transmission queue" MINUS
1226 * "Packets left network, but not honestly ACKed yet" PLUS
1227 * "Packets fast retransmitted"
1229 static __inline__ unsigned int tcp_packets_in_flight(struct tcp_opt *tp)
1231 return tp->packets_out - tp->left_out + tp->retrans_out;
1234 /* Recalculate snd_ssthresh, we want to set it to:
1237 * one half the current congestion window, but no
1238 * less than two segments
1241 * behave like Reno until low_window is reached,
1242 * then increase congestion window slowly
1244 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1246 if (sysctl_tcp_bic) {
1247 if (sysctl_tcp_bic_fast_convergence &&
1248 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1249 tp->bictcp.last_max_cwnd
1250 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1251 / (BICTCP_1_OVER_BETA/2);
1253 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1255 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1256 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1260 return max(tp->snd_cwnd >> 1U, 2U);
1263 /* Stop taking Vegas samples for now. */
1264 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1266 /* Is this TCP connection using Vegas (regardless of whether it is taking
1267 * Vegas measurements at the current time)?
1269 #define tcp_is_vegas(__tp) ((__tp)->vegas.do_vegas)
1271 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1273 /* There are several situations when we must "re-start" Vegas:
1275 * o when a connection is established
1277 * o after fast recovery
1278 * o when we send a packet and there is no outstanding
1279 * unacknowledged data (restarting an idle connection)
1281 * In these circumstances we cannot do a Vegas calculation at the
1282 * end of the first RTT, because any calculation we do is using
1283 * stale info -- both the saved cwnd and congestion feedback are
1286 * Instead we must wait until the completion of an RTT during
1287 * which we actually receive ACKs.
1290 /* Begin taking Vegas samples next time we send something. */
1291 tp->vegas.doing_vegas_now = 1;
1293 /* Set the beginning of the next send window. */
1294 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1296 tp->vegas.cntRTT = 0;
1297 tp->vegas.minRTT = 0x7fffffff;
1300 /* Should we be taking Vegas samples right now? */
1301 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1303 extern void tcp_vegas_init(struct tcp_opt *tp);
1305 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1307 if (tcp_is_vegas(tp)) {
1308 if (ca_state == TCP_CA_Open)
1309 tcp_vegas_enable(tp);
1311 tcp_vegas_disable(tp);
1313 tp->ca_state = ca_state;
1316 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1317 * The exception is rate halving phase, when cwnd is decreasing towards
1320 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1322 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1323 return tp->snd_ssthresh;
1325 return max(tp->snd_ssthresh,
1326 ((tp->snd_cwnd >> 1) +
1327 (tp->snd_cwnd >> 2)));
1330 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1332 if (tp->sack_ok && tp->sacked_out >= tp->packets_out - tp->lost_out)
1333 tp->sacked_out = tp->packets_out - tp->lost_out;
1334 tp->left_out = tp->sacked_out + tp->lost_out;
1337 extern void tcp_cwnd_application_limited(struct sock *sk);
1339 /* Congestion window validation. (RFC2861) */
1341 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1343 if (tp->packets_out >= tp->snd_cwnd) {
1344 /* Network is feed fully. */
1345 tp->snd_cwnd_used = 0;
1346 tp->snd_cwnd_stamp = tcp_time_stamp;
1348 /* Network starves. */
1349 if (tp->packets_out > tp->snd_cwnd_used)
1350 tp->snd_cwnd_used = tp->packets_out;
1352 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1353 tcp_cwnd_application_limited(sk);
1357 /* Set slow start threshould and cwnd not falling to slow start */
1358 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1360 tp->undo_marker = 0;
1361 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1362 tp->snd_cwnd = min(tp->snd_cwnd,
1363 tcp_packets_in_flight(tp) + 1U);
1364 tp->snd_cwnd_cnt = 0;
1365 tp->high_seq = tp->snd_nxt;
1366 tp->snd_cwnd_stamp = tcp_time_stamp;
1367 TCP_ECN_queue_cwr(tp);
1370 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1372 tp->prior_ssthresh = 0;
1373 if (tp->ca_state < TCP_CA_CWR) {
1374 __tcp_enter_cwr(tp);
1375 tcp_set_ca_state(tp, TCP_CA_CWR);
1379 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1381 /* Slow start with delack produces 3 packets of burst, so that
1382 * it is safe "de facto".
1384 static __inline__ __u32 tcp_max_burst(struct tcp_opt *tp)
1389 static __inline__ int tcp_minshall_check(struct tcp_opt *tp)
1391 return after(tp->snd_sml,tp->snd_una) &&
1392 !after(tp->snd_sml, tp->snd_nxt);
1395 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss, struct sk_buff *skb)
1398 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1401 /* Return 0, if packet can be sent now without violation Nagle's rules:
1402 1. It is full sized.
1403 2. Or it contains FIN.
1404 3. Or TCP_NODELAY was set.
1405 4. Or TCP_CORK is not set, and all sent packets are ACKed.
1406 With Minshall's modification: all sent small packets are ACKed.
1409 static __inline__ int
1410 tcp_nagle_check(struct tcp_opt *tp, struct sk_buff *skb, unsigned mss_now, int nonagle)
1412 return (skb->len < mss_now &&
1413 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1414 ((nonagle&TCP_NAGLE_CORK) ||
1417 tcp_minshall_check(tp))));
1420 /* This checks if the data bearing packet SKB (usually tp->send_head)
1421 * should be put on the wire right now.
1423 static __inline__ int tcp_snd_test(struct tcp_opt *tp, struct sk_buff *skb,
1424 unsigned cur_mss, int nonagle)
1426 /* RFC 1122 - section 4.2.3.4
1430 * a) The right edge of this frame exceeds the window
1431 * b) There are packets in flight and we have a small segment
1432 * [SWS avoidance and Nagle algorithm]
1433 * (part of SWS is done on packetization)
1434 * Minshall version sounds: there are no _small_
1435 * segments in flight. (tcp_nagle_check)
1436 * c) We have too many packets 'in flight'
1438 * Don't use the nagle rule for urgent data (or
1439 * for the final FIN -DaveM).
1441 * Also, Nagle rule does not apply to frames, which
1442 * sit in the middle of queue (they have no chances
1443 * to get new data) and if room at tail of skb is
1444 * not enough to save something seriously (<32 for now).
1447 /* Don't be strict about the congestion window for the
1448 * final FIN frame. -DaveM
1450 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1451 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1452 ((tcp_packets_in_flight(tp) < tp->snd_cwnd) ||
1453 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1454 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1457 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1459 if (!tp->packets_out && !tp->pending)
1460 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1463 static __inline__ int tcp_skb_is_last(struct sock *sk, struct sk_buff *skb)
1465 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1468 /* Push out any pending frames which were held back due to
1469 * TCP_CORK or attempt at coalescing tiny packets.
1470 * The socket must be locked by the caller.
1472 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1477 struct sk_buff *skb = tp->send_head;
1480 if (!tcp_skb_is_last(sk, skb))
1481 nonagle = TCP_NAGLE_PUSH;
1482 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1483 tcp_write_xmit(sk, nonagle))
1484 tcp_check_probe_timer(sk, tp);
1486 tcp_cwnd_validate(sk, tp);
1489 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1492 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1495 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1497 struct sk_buff *skb = tp->send_head;
1500 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1501 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1504 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1509 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1514 extern void tcp_destroy_sock(struct sock *sk);
1518 * Calculate(/check) TCP checksum
1520 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1521 unsigned long saddr, unsigned long daddr,
1524 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1527 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1529 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1532 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1534 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1535 __tcp_checksum_complete(skb);
1538 /* Prequeue for VJ style copy to user, combined with checksumming. */
1540 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1542 tp->ucopy.task = NULL;
1544 tp->ucopy.memory = 0;
1545 skb_queue_head_init(&tp->ucopy.prequeue);
1548 /* Packet is added to VJ-style prequeue for processing in process
1549 * context, if a reader task is waiting. Apparently, this exciting
1550 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1551 * failed somewhere. Latency? Burstiness? Well, at least now we will
1552 * see, why it failed. 8)8) --ANK
1554 * NOTE: is this not too big to inline?
1556 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1558 struct tcp_opt *tp = tcp_sk(sk);
1560 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1561 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1562 tp->ucopy.memory += skb->truesize;
1563 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1564 struct sk_buff *skb1;
1566 BUG_ON(sock_owned_by_user(sk));
1568 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1569 sk->sk_backlog_rcv(sk, skb1);
1570 NET_INC_STATS_BH(TCPPrequeueDropped);
1573 tp->ucopy.memory = 0;
1574 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1575 wake_up_interruptible(sk->sk_sleep);
1576 if (!tcp_ack_scheduled(tp))
1577 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1588 static char *statename[]={
1589 "Unused","Established","Syn Sent","Syn Recv",
1590 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1591 "Close Wait","Last ACK","Listen","Closing"
1595 static __inline__ void tcp_set_state(struct sock *sk, int state)
1597 int oldstate = sk->sk_state;
1600 case TCP_ESTABLISHED:
1601 if (oldstate != TCP_ESTABLISHED)
1602 TCP_INC_STATS(TcpCurrEstab);
1606 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1607 TCP_INC_STATS(TcpEstabResets);
1609 sk->sk_prot->unhash(sk);
1610 if (tcp_sk(sk)->bind_hash &&
1611 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1615 if (oldstate==TCP_ESTABLISHED)
1616 TCP_DEC_STATS(TcpCurrEstab);
1619 /* Change state AFTER socket is unhashed to avoid closed
1620 * socket sitting in hash tables.
1622 sk->sk_state = state;
1625 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1629 static __inline__ void tcp_done(struct sock *sk)
1631 tcp_set_state(sk, TCP_CLOSE);
1632 tcp_clear_xmit_timers(sk);
1634 sk->sk_shutdown = SHUTDOWN_MASK;
1636 if (!sock_flag(sk, SOCK_DEAD))
1637 sk->sk_state_change(sk);
1639 tcp_destroy_sock(sk);
1642 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1649 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1651 if (tp->tstamp_ok) {
1652 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1653 (TCPOPT_NOP << 16) |
1654 (TCPOPT_TIMESTAMP << 8) |
1656 *ptr++ = htonl(tstamp);
1657 *ptr++ = htonl(tp->ts_recent);
1659 if (tp->eff_sacks) {
1660 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1663 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1664 (TCPOPT_NOP << 16) |
1665 (TCPOPT_SACK << 8) |
1666 (TCPOLEN_SACK_BASE +
1667 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1668 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1669 *ptr++ = htonl(sp[this_sack].start_seq);
1670 *ptr++ = htonl(sp[this_sack].end_seq);
1679 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1680 * If this is every changed make sure to change the definition of
1681 * MAX_SYN_SIZE to match the new maximum number of options that you
1684 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1685 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1687 /* We always get an MSS option.
1688 * The option bytes which will be seen in normal data
1689 * packets should timestamps be used, must be in the MSS
1690 * advertised. But we subtract them from tp->mss_cache so
1691 * that calculations in tcp_sendmsg are simpler etc.
1692 * So account for this fact here if necessary. If we
1693 * don't do this correctly, as a receiver we won't
1694 * recognize data packets as being full sized when we
1695 * should, and thus we won't abide by the delayed ACK
1697 * SACKs don't matter, we never delay an ACK when we
1698 * have any of those going out.
1700 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1703 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1704 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1706 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1707 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1708 *ptr++ = htonl(tstamp); /* TSVAL */
1709 *ptr++ = htonl(ts_recent); /* TSECR */
1711 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1712 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1714 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1717 /* Determine a window scaling and initial window to offer.
1718 * Based on the assumption that the given amount of space
1719 * will be offered. Store the results in the tp structure.
1720 * NOTE: for smooth operation initial space offering should
1721 * be a multiple of mss if possible. We assume here that mss >= 1.
1722 * This MUST be enforced by all callers.
1724 static inline void tcp_select_initial_window(int __space, __u32 mss,
1726 __u32 *window_clamp,
1730 unsigned int space = (__space < 0 ? 0 : __space);
1732 /* If no clamp set the clamp to the max possible scaled window */
1733 if (*window_clamp == 0)
1734 (*window_clamp) = (65535 << 14);
1735 space = min(*window_clamp, space);
1737 /* Quantize space offering to a multiple of mss if possible. */
1739 space = (space / mss) * mss;
1741 /* NOTE: offering an initial window larger than 32767
1742 * will break some buggy TCP stacks. We try to be nice.
1743 * If we are not window scaling, then this truncates
1744 * our initial window offering to 32k. There should also
1745 * be a sysctl option to stop being nice.
1747 (*rcv_wnd) = min(space, MAX_TCP_WINDOW);
1750 /* See RFC1323 for an explanation of the limit to 14 */
1751 while (space > 65535 && (*rcv_wscale) < 14) {
1755 if (*rcv_wscale && sysctl_tcp_app_win && space>=mss &&
1756 space - max((space>>sysctl_tcp_app_win), mss>>*rcv_wscale) < 65536/2)
1760 /* Set initial window to value enough for senders,
1761 * following RFC1414. Senders, not following this RFC,
1762 * will be satisfied with 2.
1764 if (mss > (1<<*rcv_wscale)) {
1768 else if (mss > 1460)
1770 if (*rcv_wnd > init_cwnd*mss)
1771 *rcv_wnd = init_cwnd*mss;
1773 /* Set the clamp no higher than max representable value */
1774 (*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
1777 static inline int tcp_win_from_space(int space)
1779 return sysctl_tcp_adv_win_scale<=0 ?
1780 (space>>(-sysctl_tcp_adv_win_scale)) :
1781 space - (space>>sysctl_tcp_adv_win_scale);
1784 /* Note: caller must be prepared to deal with negative returns */
1785 static inline int tcp_space(struct sock *sk)
1787 return tcp_win_from_space(sk->sk_rcvbuf -
1788 atomic_read(&sk->sk_rmem_alloc));
1791 static inline int tcp_full_space( struct sock *sk)
1793 return tcp_win_from_space(sk->sk_rcvbuf);
1796 static inline void tcp_acceptq_removed(struct sock *sk)
1798 sk->sk_ack_backlog--;
1801 static inline void tcp_acceptq_added(struct sock *sk)
1803 sk->sk_ack_backlog++;
1806 static inline int tcp_acceptq_is_full(struct sock *sk)
1808 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
1811 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1814 struct tcp_opt *tp = tcp_sk(sk);
1817 tcp_acceptq_added(sk);
1819 if (!tp->accept_queue_tail) {
1820 tp->accept_queue = req;
1822 tp->accept_queue_tail->dl_next = req;
1824 tp->accept_queue_tail = req;
1825 req->dl_next = NULL;
1828 struct tcp_listen_opt
1830 u8 max_qlen_log; /* log_2 of maximal queued SYNs */
1835 struct open_request *syn_table[TCP_SYNQ_HSIZE];
1839 tcp_synq_removed(struct sock *sk, struct open_request *req)
1841 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1843 if (--lopt->qlen == 0)
1844 tcp_delete_keepalive_timer(sk);
1845 if (req->retrans == 0)
1849 static inline void tcp_synq_added(struct sock *sk)
1851 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1853 if (lopt->qlen++ == 0)
1854 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1858 static inline int tcp_synq_len(struct sock *sk)
1860 return tcp_sk(sk)->listen_opt->qlen;
1863 static inline int tcp_synq_young(struct sock *sk)
1865 return tcp_sk(sk)->listen_opt->qlen_young;
1868 static inline int tcp_synq_is_full(struct sock *sk)
1870 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1873 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1874 struct open_request **prev)
1876 write_lock(&tp->syn_wait_lock);
1877 *prev = req->dl_next;
1878 write_unlock(&tp->syn_wait_lock);
1881 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1882 struct open_request **prev)
1884 tcp_synq_unlink(tcp_sk(sk), req, prev);
1885 tcp_synq_removed(sk, req);
1886 tcp_openreq_free(req);
1889 static __inline__ void tcp_openreq_init(struct open_request *req,
1891 struct sk_buff *skb)
1893 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1894 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1895 req->mss = tp->mss_clamp;
1896 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1897 req->tstamp_ok = tp->tstamp_ok;
1898 req->sack_ok = tp->sack_ok;
1899 req->snd_wscale = tp->snd_wscale;
1900 req->wscale_ok = tp->wscale_ok;
1903 req->rmt_port = skb->h.th->source;
1906 #define TCP_MEM_QUANTUM ((int)PAGE_SIZE)
1908 static inline void tcp_free_skb(struct sock *sk, struct sk_buff *skb)
1910 tcp_sk(sk)->queue_shrunk = 1;
1911 sk->sk_wmem_queued -= skb->truesize;
1912 sk->sk_forward_alloc += skb->truesize;
1916 static inline void tcp_charge_skb(struct sock *sk, struct sk_buff *skb)
1918 sk->sk_wmem_queued += skb->truesize;
1919 sk->sk_forward_alloc -= skb->truesize;
1922 extern void __tcp_mem_reclaim(struct sock *sk);
1923 extern int tcp_mem_schedule(struct sock *sk, int size, int kind);
1925 static inline void tcp_mem_reclaim(struct sock *sk)
1927 if (sk->sk_forward_alloc >= TCP_MEM_QUANTUM)
1928 __tcp_mem_reclaim(sk);
1931 static inline void tcp_enter_memory_pressure(void)
1933 if (!tcp_memory_pressure) {
1934 NET_INC_STATS(TCPMemoryPressures);
1935 tcp_memory_pressure = 1;
1939 static inline void tcp_moderate_sndbuf(struct sock *sk)
1941 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1942 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued / 2);
1943 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1947 static inline struct sk_buff *tcp_alloc_pskb(struct sock *sk, int size, int mem, int gfp)
1949 struct sk_buff *skb = alloc_skb(size+MAX_TCP_HEADER, gfp);
1952 skb->truesize += mem;
1953 if (sk->sk_forward_alloc >= (int)skb->truesize ||
1954 tcp_mem_schedule(sk, skb->truesize, 0)) {
1955 skb_reserve(skb, MAX_TCP_HEADER);
1960 tcp_enter_memory_pressure();
1961 tcp_moderate_sndbuf(sk);
1966 static inline struct sk_buff *tcp_alloc_skb(struct sock *sk, int size, int gfp)
1968 return tcp_alloc_pskb(sk, size, 0, gfp);
1971 static inline struct page * tcp_alloc_page(struct sock *sk)
1973 if (sk->sk_forward_alloc >= (int)PAGE_SIZE ||
1974 tcp_mem_schedule(sk, PAGE_SIZE, 0)) {
1975 struct page *page = alloc_pages(sk->sk_allocation, 0);
1979 tcp_enter_memory_pressure();
1980 tcp_moderate_sndbuf(sk);
1984 static inline void tcp_writequeue_purge(struct sock *sk)
1986 struct sk_buff *skb;
1988 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1989 tcp_free_skb(sk, skb);
1990 tcp_mem_reclaim(sk);
1993 extern void tcp_rfree(struct sk_buff *skb);
1995 static inline void tcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
1998 skb->destructor = tcp_rfree;
1999 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2000 sk->sk_forward_alloc -= skb->truesize;
2003 extern void tcp_listen_wlock(void);
2005 /* - We may sleep inside this lock.
2006 * - If sleeping is not required (or called from BH),
2007 * use plain read_(un)lock(&tcp_lhash_lock).
2010 static inline void tcp_listen_lock(void)
2012 /* read_lock synchronizes to candidates to writers */
2013 read_lock(&tcp_lhash_lock);
2014 atomic_inc(&tcp_lhash_users);
2015 read_unlock(&tcp_lhash_lock);
2018 static inline void tcp_listen_unlock(void)
2020 if (atomic_dec_and_test(&tcp_lhash_users))
2021 wake_up(&tcp_lhash_wait);
2024 static inline int keepalive_intvl_when(struct tcp_opt *tp)
2026 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
2029 static inline int keepalive_time_when(struct tcp_opt *tp)
2031 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
2034 static inline int tcp_fin_time(struct tcp_opt *tp)
2036 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
2038 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
2039 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
2044 static inline int tcp_paws_check(struct tcp_opt *tp, int rst)
2046 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
2048 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
2051 /* RST segments are not recommended to carry timestamp,
2052 and, if they do, it is recommended to ignore PAWS because
2053 "their cleanup function should take precedence over timestamps."
2054 Certainly, it is mistake. It is necessary to understand the reasons
2055 of this constraint to relax it: if peer reboots, clock may go
2056 out-of-sync and half-open connections will not be reset.
2057 Actually, the problem would be not existing if all
2058 the implementations followed draft about maintaining clock
2059 via reboots. Linux-2.2 DOES NOT!
2061 However, we can relax time bounds for RST segments to MSL.
2063 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
2068 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
2070 sk->sk_route_caps = dst->dev->features;
2071 if (sk->sk_route_caps & NETIF_F_TSO) {
2072 if (sk->sk_no_largesend || dst->header_len)
2073 sk->sk_route_caps &= ~NETIF_F_TSO;
2077 #define TCP_CHECK_TIMER(sk) do { } while (0)
2079 static inline int tcp_use_frto(const struct sock *sk)
2081 const struct tcp_opt *tp = tcp_sk(sk);
2083 /* F-RTO must be activated in sysctl and there must be some
2084 * unsent new data, and the advertised window should allow
2087 return (sysctl_tcp_frto && tp->send_head &&
2088 !after(TCP_SKB_CB(tp->send_head)->end_seq,
2089 tp->snd_una + tp->snd_wnd));
2092 static inline void tcp_mib_init(void)
2095 TCP_ADD_STATS_USER(TcpRtoAlgorithm, 1);
2096 TCP_ADD_STATS_USER(TcpRtoMin, TCP_RTO_MIN*1000/HZ);
2097 TCP_ADD_STATS_USER(TcpRtoMax, TCP_RTO_MAX*1000/HZ);
2098 TCP_ADD_STATS_USER(TcpMaxConn, -1);
2102 enum tcp_seq_states {
2103 TCP_SEQ_STATE_LISTENING,
2104 TCP_SEQ_STATE_OPENREQ,
2105 TCP_SEQ_STATE_ESTABLISHED,
2106 TCP_SEQ_STATE_TIME_WAIT,
2109 struct tcp_seq_afinfo {
2110 struct module *owner;
2113 int (*seq_show) (struct seq_file *m, void *v);
2114 struct file_operations *seq_fops;
2117 struct tcp_iter_state {
2119 enum tcp_seq_states state;
2120 struct sock *syn_wait_sk;
2121 int bucket, sbucket, num, uid;
2122 struct seq_operations seq_ops;
2125 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2126 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2128 /* TCP Westwood functions and constants */
2130 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2131 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2133 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2135 if (sysctl_tcp_westwood)
2136 tp->westwood.rtt = rtt_seq;
2139 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2140 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2142 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2144 if (sysctl_tcp_westwood)
2145 __tcp_westwood_fast_bw(sk, skb);
2148 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2150 if (sysctl_tcp_westwood)
2151 __tcp_westwood_slow_bw(sk, skb);
2154 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2156 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2157 (__u32) (tp->mss_cache),
2161 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2163 return sysctl_tcp_westwood ? __tcp_westwood_bw_rttmin(tp) : 0;
2166 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2170 if (sysctl_tcp_westwood) {
2171 ssthresh = __tcp_westwood_bw_rttmin(tp);
2173 tp->snd_ssthresh = ssthresh;
2176 return (ssthresh != 0);
2179 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2183 if (sysctl_tcp_westwood) {
2184 cwnd = __tcp_westwood_bw_rttmin(tp);
2186 tp->snd_cwnd = cwnd;