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;
617 extern int sysctl_tcp_default_win_scale;
618 extern int sysctl_tcp_moderate_rcvbuf;
620 extern atomic_t tcp_memory_allocated;
621 extern atomic_t tcp_sockets_allocated;
622 extern int tcp_memory_pressure;
626 struct or_calltable {
628 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
629 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
630 void (*destructor) (struct open_request *req);
631 void (*send_reset) (struct sk_buff *skb);
634 struct tcp_v4_open_req {
637 struct ip_options *opt;
640 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
641 struct tcp_v6_open_req {
642 struct in6_addr loc_addr;
643 struct in6_addr rmt_addr;
644 struct sk_buff *pktopts;
649 /* this structure is too big */
650 struct open_request {
651 struct open_request *dl_next; /* Must be first member! */
658 __u16 snd_wscale : 4,
665 /* The following two fields can be easily recomputed I think -AK */
666 __u32 window_clamp; /* window clamp at creation time */
667 __u32 rcv_wnd; /* rcv_wnd offered first time */
669 unsigned long expires;
670 struct or_calltable *class;
673 struct tcp_v4_open_req v4_req;
674 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
675 struct tcp_v6_open_req v6_req;
680 /* SLAB cache for open requests. */
681 extern kmem_cache_t *tcp_openreq_cachep;
683 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
684 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
686 static inline void tcp_openreq_free(struct open_request *req)
688 req->class->destructor(req);
689 tcp_openreq_fastfree(req);
692 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
693 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
695 #define TCP_INET_FAMILY(fam) 1
699 * Pointers to address related TCP functions
700 * (i.e. things that depend on the address family)
704 int (*queue_xmit) (struct sk_buff *skb,
707 void (*send_check) (struct sock *sk,
710 struct sk_buff *skb);
712 int (*rebuild_header) (struct sock *sk);
714 int (*conn_request) (struct sock *sk,
715 struct sk_buff *skb);
717 struct sock * (*syn_recv_sock) (struct sock *sk,
719 struct open_request *req,
720 struct dst_entry *dst);
722 int (*remember_stamp) (struct sock *sk);
724 __u16 net_header_len;
726 int (*setsockopt) (struct sock *sk,
732 int (*getsockopt) (struct sock *sk,
739 void (*addr2sockaddr) (struct sock *sk,
746 * The next routines deal with comparing 32 bit unsigned ints
747 * and worry about wraparound (automatic with unsigned arithmetic).
750 static inline int before(__u32 seq1, __u32 seq2)
752 return (__s32)(seq1-seq2) < 0;
755 static inline int after(__u32 seq1, __u32 seq2)
757 return (__s32)(seq2-seq1) < 0;
761 /* is s2<=s1<=s3 ? */
762 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
764 return seq3 - seq2 >= seq1 - seq2;
768 extern struct proto tcp_prot;
770 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
771 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
772 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
773 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
774 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
775 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
776 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
778 extern void tcp_put_port(struct sock *sk);
779 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
781 extern void tcp_v4_err(struct sk_buff *skb, u32);
783 extern void tcp_shutdown (struct sock *sk, int how);
785 extern int tcp_v4_rcv(struct sk_buff *skb);
787 extern int tcp_v4_remember_stamp(struct sock *sk);
789 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
791 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
792 struct msghdr *msg, size_t size);
793 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
795 extern int tcp_ioctl(struct sock *sk,
799 extern int tcp_rcv_state_process(struct sock *sk,
804 extern int tcp_rcv_established(struct sock *sk,
809 extern void tcp_rcv_space_adjust(struct sock *sk);
818 static inline void tcp_schedule_ack(struct tcp_opt *tp)
820 tp->ack.pending |= TCP_ACK_SCHED;
823 static inline int tcp_ack_scheduled(struct tcp_opt *tp)
825 return tp->ack.pending&TCP_ACK_SCHED;
828 static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
830 if (tp->ack.quick && --tp->ack.quick == 0) {
831 /* Leaving quickack mode we deflate ATO. */
832 tp->ack.ato = TCP_ATO_MIN;
836 extern void tcp_enter_quickack_mode(struct tcp_opt *tp);
838 static __inline__ void tcp_delack_init(struct tcp_opt *tp)
840 memset(&tp->ack, 0, sizeof(tp->ack));
843 static inline void tcp_clear_options(struct tcp_opt *tp)
845 tp->tstamp_ok = tp->sack_ok = tp->wscale_ok = tp->snd_wscale = 0;
857 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
862 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
863 struct open_request *req,
864 struct open_request **prev);
865 extern int tcp_child_process(struct sock *parent,
867 struct sk_buff *skb);
868 extern void tcp_enter_frto(struct sock *sk);
869 extern void tcp_enter_loss(struct sock *sk, int how);
870 extern void tcp_clear_retrans(struct tcp_opt *tp);
871 extern void tcp_update_metrics(struct sock *sk);
873 extern void tcp_close(struct sock *sk,
875 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
876 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
877 extern void tcp_write_space(struct sock *sk);
879 extern int tcp_getsockopt(struct sock *sk, int level,
883 extern int tcp_setsockopt(struct sock *sk, int level,
884 int optname, char __user *optval,
886 extern void tcp_set_keepalive(struct sock *sk, int val);
887 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
889 size_t len, int nonblock,
890 int flags, int *addr_len);
892 extern int tcp_listen_start(struct sock *sk);
894 extern void tcp_parse_options(struct sk_buff *skb,
899 * TCP v4 functions exported for the inet6 API
902 extern int tcp_v4_rebuild_header(struct sock *sk);
904 extern int tcp_v4_build_header(struct sock *sk,
905 struct sk_buff *skb);
907 extern void tcp_v4_send_check(struct sock *sk,
908 struct tcphdr *th, int len,
909 struct sk_buff *skb);
911 extern int tcp_v4_conn_request(struct sock *sk,
912 struct sk_buff *skb);
914 extern struct sock * tcp_create_openreq_child(struct sock *sk,
915 struct open_request *req,
916 struct sk_buff *skb);
918 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
920 struct open_request *req,
921 struct dst_entry *dst);
923 extern int tcp_v4_do_rcv(struct sock *sk,
924 struct sk_buff *skb);
926 extern int tcp_v4_connect(struct sock *sk,
927 struct sockaddr *uaddr,
930 extern int tcp_connect(struct sock *sk);
932 extern struct sk_buff * tcp_make_synack(struct sock *sk,
933 struct dst_entry *dst,
934 struct open_request *req);
936 extern int tcp_disconnect(struct sock *sk, int flags);
938 extern void tcp_unhash(struct sock *sk);
940 extern int tcp_v4_hash_connecting(struct sock *sk);
943 /* From syncookies.c */
944 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
945 struct ip_options *opt);
946 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
951 extern int tcp_write_xmit(struct sock *, int nonagle);
952 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
953 extern void tcp_xmit_retransmit_queue(struct sock *);
954 extern void tcp_simple_retransmit(struct sock *);
956 extern void tcp_send_probe0(struct sock *);
957 extern void tcp_send_partial(struct sock *);
958 extern int tcp_write_wakeup(struct sock *);
959 extern void tcp_send_fin(struct sock *sk);
960 extern void tcp_send_active_reset(struct sock *sk, int priority);
961 extern int tcp_send_synack(struct sock *);
962 extern int tcp_transmit_skb(struct sock *, struct sk_buff *);
963 extern void tcp_push_one(struct sock *, unsigned mss_now);
964 extern void tcp_send_ack(struct sock *sk);
965 extern void tcp_send_delayed_ack(struct sock *sk);
966 extern void cleanup_rbuf(struct sock *sk, int copied);
969 extern void tcp_init_xmit_timers(struct sock *);
970 extern void tcp_clear_xmit_timers(struct sock *);
972 extern void tcp_delete_keepalive_timer (struct sock *);
973 extern void tcp_reset_keepalive_timer (struct sock *, unsigned long);
974 extern int tcp_sync_mss(struct sock *sk, u32 pmtu);
976 extern const char timer_bug_msg[];
978 /* Read 'sendfile()'-style from a TCP socket */
979 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
980 unsigned int, size_t);
981 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
982 sk_read_actor_t recv_actor);
984 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
986 struct tcp_opt *tp = tcp_sk(sk);
989 case TCP_TIME_RETRANS:
990 case TCP_TIME_PROBE0:
993 #ifdef TCP_CLEAR_TIMERS
994 if (timer_pending(&tp->retransmit_timer) &&
995 del_timer(&tp->retransmit_timer))
1000 tp->ack.blocked = 0;
1001 tp->ack.pending = 0;
1003 #ifdef TCP_CLEAR_TIMERS
1004 if (timer_pending(&tp->delack_timer) &&
1005 del_timer(&tp->delack_timer))
1010 printk(timer_bug_msg);
1017 * Reset the retransmission timer
1019 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
1021 struct tcp_opt *tp = tcp_sk(sk);
1023 if (when > TCP_RTO_MAX) {
1025 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
1031 case TCP_TIME_RETRANS:
1032 case TCP_TIME_PROBE0:
1034 tp->timeout = jiffies+when;
1035 if (!mod_timer(&tp->retransmit_timer, tp->timeout))
1040 tp->ack.pending |= TCP_ACK_TIMER;
1041 tp->ack.timeout = jiffies+when;
1042 if (!mod_timer(&tp->delack_timer, tp->ack.timeout))
1047 printk(KERN_DEBUG "bug: unknown timer value\n");
1051 /* Compute the current effective MSS, taking SACKs and IP options,
1052 * and even PMTU discovery events into account.
1054 * LARGESEND note: !urg_mode is overkill, only frames up to snd_up
1055 * cannot be large. However, taking into account rare use of URG, this
1056 * is not a big flaw.
1059 static __inline__ unsigned int tcp_current_mss(struct sock *sk, int large)
1061 struct tcp_opt *tp = tcp_sk(sk);
1062 struct dst_entry *dst = __sk_dst_get(sk);
1063 int mss_now = large && (sk->sk_route_caps & NETIF_F_TSO) &&
1065 tp->mss_cache : tp->mss_cache_std;
1068 u32 mtu = dst_pmtu(dst);
1069 if (mtu != tp->pmtu_cookie ||
1070 tp->ext2_header_len != dst->header_len)
1071 mss_now = tcp_sync_mss(sk, mtu);
1074 mss_now -= (TCPOLEN_SACK_BASE_ALIGNED +
1075 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK));
1079 /* Initialize RCV_MSS value.
1080 * RCV_MSS is an our guess about MSS used by the peer.
1081 * We haven't any direct information about the MSS.
1082 * It's better to underestimate the RCV_MSS rather than overestimate.
1083 * Overestimations make us ACKing less frequently than needed.
1084 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1087 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1089 struct tcp_opt *tp = tcp_sk(sk);
1090 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1092 hint = min(hint, tp->rcv_wnd/2);
1093 hint = min(hint, TCP_MIN_RCVMSS);
1094 hint = max(hint, TCP_MIN_MSS);
1096 tp->ack.rcv_mss = hint;
1099 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1101 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1102 ntohl(TCP_FLAG_ACK) |
1106 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1108 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1111 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1113 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1115 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1117 tcp_fast_path_on(tp);
1120 /* Compute the actual receive window we are currently advertising.
1121 * Rcv_nxt can be after the window if our peer push more data
1122 * than the offered window.
1124 static __inline__ u32 tcp_receive_window(struct tcp_opt *tp)
1126 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1133 /* Choose a new window, without checks for shrinking, and without
1134 * scaling applied to the result. The caller does these things
1135 * if necessary. This is a "raw" window selection.
1137 extern u32 __tcp_select_window(struct sock *sk);
1139 /* TCP timestamps are only 32-bits, this causes a slight
1140 * complication on 64-bit systems since we store a snapshot
1141 * of jiffies in the buffer control blocks below. We decidely
1142 * only use of the low 32-bits of jiffies and hide the ugly
1143 * casts with the following macro.
1145 #define tcp_time_stamp ((__u32)(jiffies))
1147 /* This is what the send packet queueing engine uses to pass
1148 * TCP per-packet control information to the transmission
1149 * code. We also store the host-order sequence numbers in
1150 * here too. This is 36 bytes on 32-bit architectures,
1151 * 40 bytes on 64-bit machines, if this grows please adjust
1152 * skbuff.h:skbuff->cb[xxx] size appropriately.
1156 struct inet_skb_parm h4;
1157 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1158 struct inet6_skb_parm h6;
1160 } header; /* For incoming frames */
1161 __u32 seq; /* Starting sequence number */
1162 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1163 __u32 when; /* used to compute rtt's */
1164 __u8 flags; /* TCP header flags. */
1166 /* NOTE: These must match up to the flags byte in a
1169 #define TCPCB_FLAG_FIN 0x01
1170 #define TCPCB_FLAG_SYN 0x02
1171 #define TCPCB_FLAG_RST 0x04
1172 #define TCPCB_FLAG_PSH 0x08
1173 #define TCPCB_FLAG_ACK 0x10
1174 #define TCPCB_FLAG_URG 0x20
1175 #define TCPCB_FLAG_ECE 0x40
1176 #define TCPCB_FLAG_CWR 0x80
1178 __u8 sacked; /* State flags for SACK/FACK. */
1179 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1180 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1181 #define TCPCB_LOST 0x04 /* SKB is lost */
1182 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1184 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1185 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1187 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1189 #define TCPCB_AT_TAIL (TCPCB_URG)
1191 __u16 urg_ptr; /* Valid w/URG flags is set. */
1192 __u32 ack_seq; /* Sequence number ACK'd */
1195 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1197 #define for_retrans_queue(skb, sk, tp) \
1198 for (skb = (sk)->sk_write_queue.next; \
1199 (skb != (tp)->send_head) && \
1200 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1204 #include <net/tcp_ecn.h>
1208 * Compute minimal free write space needed to queue new packets.
1210 static inline int tcp_min_write_space(struct sock *sk)
1212 return sk->sk_wmem_queued / 2;
1215 static inline int tcp_wspace(struct sock *sk)
1217 return sk->sk_sndbuf - sk->sk_wmem_queued;
1221 /* This determines how many packets are "in the network" to the best
1222 * of our knowledge. In many cases it is conservative, but where
1223 * detailed information is available from the receiver (via SACK
1224 * blocks etc.) we can make more aggressive calculations.
1226 * Use this for decisions involving congestion control, use just
1227 * tp->packets_out to determine if the send queue is empty or not.
1229 * Read this equation as:
1231 * "Packets sent once on transmission queue" MINUS
1232 * "Packets left network, but not honestly ACKed yet" PLUS
1233 * "Packets fast retransmitted"
1235 static __inline__ unsigned int tcp_packets_in_flight(struct tcp_opt *tp)
1237 return tp->packets_out - tp->left_out + tp->retrans_out;
1240 /* Recalculate snd_ssthresh, we want to set it to:
1243 * one half the current congestion window, but no
1244 * less than two segments
1247 * behave like Reno until low_window is reached,
1248 * then increase congestion window slowly
1250 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1252 if (sysctl_tcp_bic) {
1253 if (sysctl_tcp_bic_fast_convergence &&
1254 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1255 tp->bictcp.last_max_cwnd
1256 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1257 / (BICTCP_1_OVER_BETA/2);
1259 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1261 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1262 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1266 return max(tp->snd_cwnd >> 1U, 2U);
1269 /* Stop taking Vegas samples for now. */
1270 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1272 /* Is this TCP connection using Vegas (regardless of whether it is taking
1273 * Vegas measurements at the current time)?
1275 #define tcp_is_vegas(__tp) ((__tp)->vegas.do_vegas)
1277 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1279 /* There are several situations when we must "re-start" Vegas:
1281 * o when a connection is established
1283 * o after fast recovery
1284 * o when we send a packet and there is no outstanding
1285 * unacknowledged data (restarting an idle connection)
1287 * In these circumstances we cannot do a Vegas calculation at the
1288 * end of the first RTT, because any calculation we do is using
1289 * stale info -- both the saved cwnd and congestion feedback are
1292 * Instead we must wait until the completion of an RTT during
1293 * which we actually receive ACKs.
1296 /* Begin taking Vegas samples next time we send something. */
1297 tp->vegas.doing_vegas_now = 1;
1299 /* Set the beginning of the next send window. */
1300 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1302 tp->vegas.cntRTT = 0;
1303 tp->vegas.minRTT = 0x7fffffff;
1306 /* Should we be taking Vegas samples right now? */
1307 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1309 extern void tcp_vegas_init(struct tcp_opt *tp);
1311 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1313 if (tcp_is_vegas(tp)) {
1314 if (ca_state == TCP_CA_Open)
1315 tcp_vegas_enable(tp);
1317 tcp_vegas_disable(tp);
1319 tp->ca_state = ca_state;
1322 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1323 * The exception is rate halving phase, when cwnd is decreasing towards
1326 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1328 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1329 return tp->snd_ssthresh;
1331 return max(tp->snd_ssthresh,
1332 ((tp->snd_cwnd >> 1) +
1333 (tp->snd_cwnd >> 2)));
1336 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1338 if (tp->sack_ok && tp->sacked_out >= tp->packets_out - tp->lost_out)
1339 tp->sacked_out = tp->packets_out - tp->lost_out;
1340 tp->left_out = tp->sacked_out + tp->lost_out;
1343 extern void tcp_cwnd_application_limited(struct sock *sk);
1345 /* Congestion window validation. (RFC2861) */
1347 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1349 if (tp->packets_out >= tp->snd_cwnd) {
1350 /* Network is feed fully. */
1351 tp->snd_cwnd_used = 0;
1352 tp->snd_cwnd_stamp = tcp_time_stamp;
1354 /* Network starves. */
1355 if (tp->packets_out > tp->snd_cwnd_used)
1356 tp->snd_cwnd_used = tp->packets_out;
1358 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1359 tcp_cwnd_application_limited(sk);
1363 /* Set slow start threshould and cwnd not falling to slow start */
1364 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1366 tp->undo_marker = 0;
1367 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1368 tp->snd_cwnd = min(tp->snd_cwnd,
1369 tcp_packets_in_flight(tp) + 1U);
1370 tp->snd_cwnd_cnt = 0;
1371 tp->high_seq = tp->snd_nxt;
1372 tp->snd_cwnd_stamp = tcp_time_stamp;
1373 TCP_ECN_queue_cwr(tp);
1376 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1378 tp->prior_ssthresh = 0;
1379 if (tp->ca_state < TCP_CA_CWR) {
1380 __tcp_enter_cwr(tp);
1381 tcp_set_ca_state(tp, TCP_CA_CWR);
1385 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1387 /* Slow start with delack produces 3 packets of burst, so that
1388 * it is safe "de facto".
1390 static __inline__ __u32 tcp_max_burst(struct tcp_opt *tp)
1395 static __inline__ int tcp_minshall_check(struct tcp_opt *tp)
1397 return after(tp->snd_sml,tp->snd_una) &&
1398 !after(tp->snd_sml, tp->snd_nxt);
1401 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss, struct sk_buff *skb)
1404 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1407 /* Return 0, if packet can be sent now without violation Nagle's rules:
1408 1. It is full sized.
1409 2. Or it contains FIN.
1410 3. Or higher layers meant to force a packet boundary, hence the PSH bit.
1411 4. Or TCP_NODELAY was set.
1412 5. Or TCP_CORK is not set, and all sent packets are ACKed.
1413 With Minshall's modification: all sent small packets are ACKed.
1416 static __inline__ int
1417 tcp_nagle_check(struct tcp_opt *tp, struct sk_buff *skb, unsigned mss_now, int nonagle)
1419 return (skb->len < mss_now &&
1420 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1421 ((nonagle&TCP_NAGLE_CORK) ||
1424 tcp_minshall_check(tp))));
1427 /* This checks if the data bearing packet SKB (usually tp->send_head)
1428 * should be put on the wire right now.
1430 static __inline__ int tcp_snd_test(struct tcp_opt *tp, struct sk_buff *skb,
1431 unsigned cur_mss, int nonagle)
1433 /* RFC 1122 - section 4.2.3.4
1437 * a) The right edge of this frame exceeds the window
1438 * b) There are packets in flight and we have a small segment
1439 * [SWS avoidance and Nagle algorithm]
1440 * (part of SWS is done on packetization)
1441 * Minshall version sounds: there are no _small_
1442 * segments in flight. (tcp_nagle_check)
1443 * c) We have too many packets 'in flight'
1445 * Don't use the nagle rule for urgent data (or
1446 * for the final FIN -DaveM).
1448 * Also, Nagle rule does not apply to frames, which
1449 * sit in the middle of queue (they have no chances
1450 * to get new data) and if room at tail of skb is
1451 * not enough to save something seriously (<32 for now).
1454 /* Don't be strict about the congestion window for the
1455 * final FIN frame. -DaveM
1457 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1458 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1459 ((tcp_packets_in_flight(tp) < tp->snd_cwnd) ||
1460 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1461 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1464 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1466 if (!tp->packets_out && !tp->pending)
1467 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1470 static __inline__ int tcp_skb_is_last(struct sock *sk, struct sk_buff *skb)
1472 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1475 /* Push out any pending frames which were held back due to
1476 * TCP_CORK or attempt at coalescing tiny packets.
1477 * The socket must be locked by the caller.
1479 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1484 struct sk_buff *skb = tp->send_head;
1487 if (!tcp_skb_is_last(sk, skb))
1488 nonagle = TCP_NAGLE_PUSH;
1489 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1490 tcp_write_xmit(sk, nonagle))
1491 tcp_check_probe_timer(sk, tp);
1493 tcp_cwnd_validate(sk, tp);
1496 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1499 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1502 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1504 struct sk_buff *skb = tp->send_head;
1507 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1508 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1511 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1516 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1521 extern void tcp_destroy_sock(struct sock *sk);
1525 * Calculate(/check) TCP checksum
1527 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1528 unsigned long saddr, unsigned long daddr,
1531 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1534 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1536 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1539 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1541 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1542 __tcp_checksum_complete(skb);
1545 /* Prequeue for VJ style copy to user, combined with checksumming. */
1547 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1549 tp->ucopy.task = NULL;
1551 tp->ucopy.memory = 0;
1552 skb_queue_head_init(&tp->ucopy.prequeue);
1555 /* Packet is added to VJ-style prequeue for processing in process
1556 * context, if a reader task is waiting. Apparently, this exciting
1557 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1558 * failed somewhere. Latency? Burstiness? Well, at least now we will
1559 * see, why it failed. 8)8) --ANK
1561 * NOTE: is this not too big to inline?
1563 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1565 struct tcp_opt *tp = tcp_sk(sk);
1567 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1568 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1569 tp->ucopy.memory += skb->truesize;
1570 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1571 struct sk_buff *skb1;
1573 BUG_ON(sock_owned_by_user(sk));
1575 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1576 sk->sk_backlog_rcv(sk, skb1);
1577 NET_INC_STATS_BH(TCPPrequeueDropped);
1580 tp->ucopy.memory = 0;
1581 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1582 wake_up_interruptible(sk->sk_sleep);
1583 if (!tcp_ack_scheduled(tp))
1584 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1595 static char *statename[]={
1596 "Unused","Established","Syn Sent","Syn Recv",
1597 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1598 "Close Wait","Last ACK","Listen","Closing"
1602 static __inline__ void tcp_set_state(struct sock *sk, int state)
1604 int oldstate = sk->sk_state;
1607 case TCP_ESTABLISHED:
1608 if (oldstate != TCP_ESTABLISHED)
1609 TCP_INC_STATS(TcpCurrEstab);
1613 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1614 TCP_INC_STATS(TcpEstabResets);
1616 sk->sk_prot->unhash(sk);
1617 if (tcp_sk(sk)->bind_hash &&
1618 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1622 if (oldstate==TCP_ESTABLISHED)
1623 TCP_DEC_STATS(TcpCurrEstab);
1626 /* Change state AFTER socket is unhashed to avoid closed
1627 * socket sitting in hash tables.
1629 sk->sk_state = state;
1632 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1636 static __inline__ void tcp_done(struct sock *sk)
1638 tcp_set_state(sk, TCP_CLOSE);
1639 tcp_clear_xmit_timers(sk);
1641 sk->sk_shutdown = SHUTDOWN_MASK;
1643 if (!sock_flag(sk, SOCK_DEAD))
1644 sk->sk_state_change(sk);
1646 tcp_destroy_sock(sk);
1649 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1656 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1658 if (tp->tstamp_ok) {
1659 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1660 (TCPOPT_NOP << 16) |
1661 (TCPOPT_TIMESTAMP << 8) |
1663 *ptr++ = htonl(tstamp);
1664 *ptr++ = htonl(tp->ts_recent);
1666 if (tp->eff_sacks) {
1667 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1670 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1671 (TCPOPT_NOP << 16) |
1672 (TCPOPT_SACK << 8) |
1673 (TCPOLEN_SACK_BASE +
1674 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1675 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1676 *ptr++ = htonl(sp[this_sack].start_seq);
1677 *ptr++ = htonl(sp[this_sack].end_seq);
1686 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1687 * If this is every changed make sure to change the definition of
1688 * MAX_SYN_SIZE to match the new maximum number of options that you
1691 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1692 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1694 /* We always get an MSS option.
1695 * The option bytes which will be seen in normal data
1696 * packets should timestamps be used, must be in the MSS
1697 * advertised. But we subtract them from tp->mss_cache so
1698 * that calculations in tcp_sendmsg are simpler etc.
1699 * So account for this fact here if necessary. If we
1700 * don't do this correctly, as a receiver we won't
1701 * recognize data packets as being full sized when we
1702 * should, and thus we won't abide by the delayed ACK
1704 * SACKs don't matter, we never delay an ACK when we
1705 * have any of those going out.
1707 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1710 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1711 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1713 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1714 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1715 *ptr++ = htonl(tstamp); /* TSVAL */
1716 *ptr++ = htonl(ts_recent); /* TSECR */
1718 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1719 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1721 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1724 /* Determine a window scaling and initial window to offer.
1725 * Based on the assumption that the given amount of space
1726 * will be offered. Store the results in the tp structure.
1727 * NOTE: for smooth operation initial space offering should
1728 * be a multiple of mss if possible. We assume here that mss >= 1.
1729 * This MUST be enforced by all callers.
1731 static inline void tcp_select_initial_window(int __space, __u32 mss,
1733 __u32 *window_clamp,
1737 unsigned int space = (__space < 0 ? 0 : __space);
1739 /* If no clamp set the clamp to the max possible scaled window */
1740 if (*window_clamp == 0)
1741 (*window_clamp) = (65535 << 14);
1742 space = min(*window_clamp, space);
1744 /* Quantize space offering to a multiple of mss if possible. */
1746 space = (space / mss) * mss;
1748 /* NOTE: offering an initial window larger than 32767
1749 * will break some buggy TCP stacks. We try to be nice.
1750 * If we are not window scaling, then this truncates
1751 * our initial window offering to 32k. There should also
1752 * be a sysctl option to stop being nice.
1754 (*rcv_wnd) = min(space, MAX_TCP_WINDOW);
1757 /* See RFC1323 for an explanation of the limit to 14 */
1758 while (space > 65535 && (*rcv_wscale) < 14) {
1762 if (*rcv_wscale && sysctl_tcp_app_win && space>=mss &&
1763 space - max((space>>sysctl_tcp_app_win), mss>>*rcv_wscale) < 65536/2)
1766 *rcv_wscale = max((__u8)sysctl_tcp_default_win_scale,
1770 /* Set initial window to value enough for senders,
1771 * following RFC1414. Senders, not following this RFC,
1772 * will be satisfied with 2.
1774 if (mss > (1<<*rcv_wscale)) {
1778 else if (mss > 1460)
1780 if (*rcv_wnd > init_cwnd*mss)
1781 *rcv_wnd = init_cwnd*mss;
1783 /* Set the clamp no higher than max representable value */
1784 (*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
1787 static inline int tcp_win_from_space(int space)
1789 return sysctl_tcp_adv_win_scale<=0 ?
1790 (space>>(-sysctl_tcp_adv_win_scale)) :
1791 space - (space>>sysctl_tcp_adv_win_scale);
1794 /* Note: caller must be prepared to deal with negative returns */
1795 static inline int tcp_space(struct sock *sk)
1797 return tcp_win_from_space(sk->sk_rcvbuf -
1798 atomic_read(&sk->sk_rmem_alloc));
1801 static inline int tcp_full_space( struct sock *sk)
1803 return tcp_win_from_space(sk->sk_rcvbuf);
1806 static inline void tcp_acceptq_removed(struct sock *sk)
1808 sk->sk_ack_backlog--;
1811 static inline void tcp_acceptq_added(struct sock *sk)
1813 sk->sk_ack_backlog++;
1816 static inline int tcp_acceptq_is_full(struct sock *sk)
1818 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
1821 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1824 struct tcp_opt *tp = tcp_sk(sk);
1827 tcp_acceptq_added(sk);
1829 if (!tp->accept_queue_tail) {
1830 tp->accept_queue = req;
1832 tp->accept_queue_tail->dl_next = req;
1834 tp->accept_queue_tail = req;
1835 req->dl_next = NULL;
1838 struct tcp_listen_opt
1840 u8 max_qlen_log; /* log_2 of maximal queued SYNs */
1845 struct open_request *syn_table[TCP_SYNQ_HSIZE];
1849 tcp_synq_removed(struct sock *sk, struct open_request *req)
1851 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1853 if (--lopt->qlen == 0)
1854 tcp_delete_keepalive_timer(sk);
1855 if (req->retrans == 0)
1859 static inline void tcp_synq_added(struct sock *sk)
1861 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1863 if (lopt->qlen++ == 0)
1864 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1868 static inline int tcp_synq_len(struct sock *sk)
1870 return tcp_sk(sk)->listen_opt->qlen;
1873 static inline int tcp_synq_young(struct sock *sk)
1875 return tcp_sk(sk)->listen_opt->qlen_young;
1878 static inline int tcp_synq_is_full(struct sock *sk)
1880 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1883 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1884 struct open_request **prev)
1886 write_lock(&tp->syn_wait_lock);
1887 *prev = req->dl_next;
1888 write_unlock(&tp->syn_wait_lock);
1891 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1892 struct open_request **prev)
1894 tcp_synq_unlink(tcp_sk(sk), req, prev);
1895 tcp_synq_removed(sk, req);
1896 tcp_openreq_free(req);
1899 static __inline__ void tcp_openreq_init(struct open_request *req,
1901 struct sk_buff *skb)
1903 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1904 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1905 req->mss = tp->mss_clamp;
1906 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1907 req->tstamp_ok = tp->tstamp_ok;
1908 req->sack_ok = tp->sack_ok;
1909 req->snd_wscale = tp->snd_wscale;
1910 req->wscale_ok = tp->wscale_ok;
1913 req->rmt_port = skb->h.th->source;
1916 #define TCP_MEM_QUANTUM ((int)PAGE_SIZE)
1918 static inline void tcp_free_skb(struct sock *sk, struct sk_buff *skb)
1920 tcp_sk(sk)->queue_shrunk = 1;
1921 sk->sk_wmem_queued -= skb->truesize;
1922 sk->sk_forward_alloc += skb->truesize;
1926 static inline void tcp_charge_skb(struct sock *sk, struct sk_buff *skb)
1928 sk->sk_wmem_queued += skb->truesize;
1929 sk->sk_forward_alloc -= skb->truesize;
1932 extern void __tcp_mem_reclaim(struct sock *sk);
1933 extern int tcp_mem_schedule(struct sock *sk, int size, int kind);
1935 static inline void tcp_mem_reclaim(struct sock *sk)
1937 if (sk->sk_forward_alloc >= TCP_MEM_QUANTUM)
1938 __tcp_mem_reclaim(sk);
1941 static inline void tcp_enter_memory_pressure(void)
1943 if (!tcp_memory_pressure) {
1944 NET_INC_STATS(TCPMemoryPressures);
1945 tcp_memory_pressure = 1;
1949 static inline void tcp_moderate_sndbuf(struct sock *sk)
1951 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1952 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued / 2);
1953 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1957 static inline struct sk_buff *tcp_alloc_pskb(struct sock *sk, int size, int mem, int gfp)
1959 struct sk_buff *skb = alloc_skb(size+MAX_TCP_HEADER, gfp);
1962 skb->truesize += mem;
1963 if (sk->sk_forward_alloc >= (int)skb->truesize ||
1964 tcp_mem_schedule(sk, skb->truesize, 0)) {
1965 skb_reserve(skb, MAX_TCP_HEADER);
1970 tcp_enter_memory_pressure();
1971 tcp_moderate_sndbuf(sk);
1976 static inline struct sk_buff *tcp_alloc_skb(struct sock *sk, int size, int gfp)
1978 return tcp_alloc_pskb(sk, size, 0, gfp);
1981 static inline struct page * tcp_alloc_page(struct sock *sk)
1983 if (sk->sk_forward_alloc >= (int)PAGE_SIZE ||
1984 tcp_mem_schedule(sk, PAGE_SIZE, 0)) {
1985 struct page *page = alloc_pages(sk->sk_allocation, 0);
1989 tcp_enter_memory_pressure();
1990 tcp_moderate_sndbuf(sk);
1994 static inline void tcp_writequeue_purge(struct sock *sk)
1996 struct sk_buff *skb;
1998 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1999 tcp_free_skb(sk, skb);
2000 tcp_mem_reclaim(sk);
2003 extern void tcp_rfree(struct sk_buff *skb);
2005 static inline void tcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
2008 skb->destructor = tcp_rfree;
2009 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2010 sk->sk_forward_alloc -= skb->truesize;
2013 extern void tcp_listen_wlock(void);
2015 /* - We may sleep inside this lock.
2016 * - If sleeping is not required (or called from BH),
2017 * use plain read_(un)lock(&tcp_lhash_lock).
2020 static inline void tcp_listen_lock(void)
2022 /* read_lock synchronizes to candidates to writers */
2023 read_lock(&tcp_lhash_lock);
2024 atomic_inc(&tcp_lhash_users);
2025 read_unlock(&tcp_lhash_lock);
2028 static inline void tcp_listen_unlock(void)
2030 if (atomic_dec_and_test(&tcp_lhash_users))
2031 wake_up(&tcp_lhash_wait);
2034 static inline int keepalive_intvl_when(struct tcp_opt *tp)
2036 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
2039 static inline int keepalive_time_when(struct tcp_opt *tp)
2041 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
2044 static inline int tcp_fin_time(struct tcp_opt *tp)
2046 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
2048 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
2049 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
2054 static inline int tcp_paws_check(struct tcp_opt *tp, int rst)
2056 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
2058 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
2061 /* RST segments are not recommended to carry timestamp,
2062 and, if they do, it is recommended to ignore PAWS because
2063 "their cleanup function should take precedence over timestamps."
2064 Certainly, it is mistake. It is necessary to understand the reasons
2065 of this constraint to relax it: if peer reboots, clock may go
2066 out-of-sync and half-open connections will not be reset.
2067 Actually, the problem would be not existing if all
2068 the implementations followed draft about maintaining clock
2069 via reboots. Linux-2.2 DOES NOT!
2071 However, we can relax time bounds for RST segments to MSL.
2073 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
2078 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
2080 sk->sk_route_caps = dst->dev->features;
2081 if (sk->sk_route_caps & NETIF_F_TSO) {
2082 if (sk->sk_no_largesend || dst->header_len)
2083 sk->sk_route_caps &= ~NETIF_F_TSO;
2087 #define TCP_CHECK_TIMER(sk) do { } while (0)
2089 static inline int tcp_use_frto(const struct sock *sk)
2091 const struct tcp_opt *tp = tcp_sk(sk);
2093 /* F-RTO must be activated in sysctl and there must be some
2094 * unsent new data, and the advertised window should allow
2097 return (sysctl_tcp_frto && tp->send_head &&
2098 !after(TCP_SKB_CB(tp->send_head)->end_seq,
2099 tp->snd_una + tp->snd_wnd));
2102 static inline void tcp_mib_init(void)
2105 TCP_ADD_STATS_USER(TcpRtoAlgorithm, 1);
2106 TCP_ADD_STATS_USER(TcpRtoMin, TCP_RTO_MIN*1000/HZ);
2107 TCP_ADD_STATS_USER(TcpRtoMax, TCP_RTO_MAX*1000/HZ);
2108 TCP_ADD_STATS_USER(TcpMaxConn, -1);
2112 enum tcp_seq_states {
2113 TCP_SEQ_STATE_LISTENING,
2114 TCP_SEQ_STATE_OPENREQ,
2115 TCP_SEQ_STATE_ESTABLISHED,
2116 TCP_SEQ_STATE_TIME_WAIT,
2119 struct tcp_seq_afinfo {
2120 struct module *owner;
2123 int (*seq_show) (struct seq_file *m, void *v);
2124 struct file_operations *seq_fops;
2127 struct tcp_iter_state {
2129 enum tcp_seq_states state;
2130 struct sock *syn_wait_sk;
2131 int bucket, sbucket, num, uid;
2132 struct seq_operations seq_ops;
2135 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2136 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2138 /* TCP Westwood functions and constants */
2140 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2141 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2143 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2145 if (sysctl_tcp_westwood)
2146 tp->westwood.rtt = rtt_seq;
2149 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2150 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2152 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2154 if (sysctl_tcp_westwood)
2155 __tcp_westwood_fast_bw(sk, skb);
2158 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2160 if (sysctl_tcp_westwood)
2161 __tcp_westwood_slow_bw(sk, skb);
2164 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2166 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2167 (__u32) (tp->mss_cache),
2171 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2173 return sysctl_tcp_westwood ? __tcp_westwood_bw_rttmin(tp) : 0;
2176 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2180 if (sysctl_tcp_westwood) {
2181 ssthresh = __tcp_westwood_bw_rttmin(tp);
2183 tp->snd_ssthresh = ssthresh;
2186 return (ssthresh != 0);
2189 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2193 if (sysctl_tcp_westwood) {
2194 cwnd = __tcp_westwood_bw_rttmin(tp);
2196 tp->snd_cwnd = cwnd;