2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
22 #define FASTRETRANS_DEBUG 1
24 /* Cancel timers, when they are not required. */
25 #undef TCP_CLEAR_TIMERS
27 #include <linux/config.h>
28 #include <linux/list.h>
29 #include <linux/tcp.h>
30 #include <linux/slab.h>
31 #include <linux/cache.h>
32 #include <linux/percpu.h>
33 #include <net/checksum.h>
37 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
38 #include <linux/ipv6.h>
40 #include <linux/seq_file.h>
42 /* This is for all connections with a full identity, no wildcards.
43 * New scheme, half the table is for TIME_WAIT, the other half is
44 * for the rest. I'll experiment with dynamic table growth later.
46 struct tcp_ehash_bucket {
48 struct hlist_head chain;
49 } __attribute__((__aligned__(8)));
51 /* This is for listening sockets, thus all sockets which possess wildcards. */
52 #define TCP_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
54 /* There are a few simple rules, which allow for local port reuse by
55 * an application. In essence:
57 * 1) Sockets bound to different interfaces may share a local port.
58 * Failing that, goto test 2.
59 * 2) If all sockets have sk->sk_reuse set, and none of them are in
60 * TCP_LISTEN state, the port may be shared.
61 * Failing that, goto test 3.
62 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
63 * address, and none of them are the same, the port may be
65 * Failing this, the port cannot be shared.
67 * The interesting point, is test #2. This is what an FTP server does
68 * all day. To optimize this case we use a specific flag bit defined
69 * below. As we add sockets to a bind bucket list, we perform a
70 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
71 * As long as all sockets added to a bind bucket pass this test,
72 * the flag bit will be set.
73 * The resulting situation is that tcp_v[46]_verify_bind() can just check
74 * for this flag bit, if it is set and the socket trying to bind has
75 * sk->sk_reuse set, we don't even have to walk the owners list at all,
76 * we return that it is ok to bind this socket to the requested local port.
78 * Sounds like a lot of work, but it is worth it. In a more naive
79 * implementation (ie. current FreeBSD etc.) the entire list of ports
80 * must be walked for each data port opened by an ftp server. Needless
81 * to say, this does not scale at all. With a couple thousand FTP
82 * users logged onto your box, isn't it nice to know that new data
83 * ports are created in O(1) time? I thought so. ;-) -DaveM
85 struct tcp_bind_bucket {
87 signed short fastreuse;
88 struct hlist_node node;
89 struct hlist_head owners;
92 #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node)
94 struct tcp_bind_hashbucket {
96 struct hlist_head chain;
99 static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head)
101 return hlist_entry(head->chain.first, struct tcp_bind_bucket, node);
104 static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head)
106 return hlist_empty(&head->chain) ? NULL : __tb_head(head);
109 extern struct tcp_hashinfo {
110 /* This is for sockets with full identity only. Sockets here will
111 * always be without wildcards and will have the following invariant:
113 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
115 * First half of the table is for sockets not in TIME_WAIT, second half
116 * is for TIME_WAIT sockets only.
118 struct tcp_ehash_bucket *__tcp_ehash;
120 /* Ok, let's try this, I give up, we do need a local binding
121 * TCP hash as well as the others for fast bind/connect.
123 struct tcp_bind_hashbucket *__tcp_bhash;
125 int __tcp_bhash_size;
126 int __tcp_ehash_size;
128 /* All sockets in TCP_LISTEN state will be in here. This is the only
129 * table where wildcard'd TCP sockets can exist. Hash function here
130 * is just local port number.
132 struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE];
134 /* All the above members are written once at bootup and
135 * never written again _or_ are predominantly read-access.
137 * Now align to a new cache line as all the following members
140 rwlock_t __tcp_lhash_lock ____cacheline_aligned;
141 atomic_t __tcp_lhash_users;
142 wait_queue_head_t __tcp_lhash_wait;
143 spinlock_t __tcp_portalloc_lock;
146 #define tcp_ehash (tcp_hashinfo.__tcp_ehash)
147 #define tcp_bhash (tcp_hashinfo.__tcp_bhash)
148 #define tcp_ehash_size (tcp_hashinfo.__tcp_ehash_size)
149 #define tcp_bhash_size (tcp_hashinfo.__tcp_bhash_size)
150 #define tcp_listening_hash (tcp_hashinfo.__tcp_listening_hash)
151 #define tcp_lhash_lock (tcp_hashinfo.__tcp_lhash_lock)
152 #define tcp_lhash_users (tcp_hashinfo.__tcp_lhash_users)
153 #define tcp_lhash_wait (tcp_hashinfo.__tcp_lhash_wait)
154 #define tcp_portalloc_lock (tcp_hashinfo.__tcp_portalloc_lock)
156 extern kmem_cache_t *tcp_bucket_cachep;
157 extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head,
158 unsigned short snum);
159 extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb);
160 extern void tcp_bucket_unlock(struct sock *sk);
161 extern int tcp_port_rover;
163 /* These are AF independent. */
164 static __inline__ int tcp_bhashfn(__u16 lport)
166 return (lport & (tcp_bhash_size - 1));
169 extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb,
170 unsigned short snum);
172 #if (BITS_PER_LONG == 64)
173 #define TCP_ADDRCMP_ALIGN_BYTES 8
175 #define TCP_ADDRCMP_ALIGN_BYTES 4
178 /* This is a TIME_WAIT bucket. It works around the memory consumption
179 * problems of sockets in such a state on heavily loaded servers, but
180 * without violating the protocol specification.
182 struct tcp_tw_bucket {
184 * Now struct sock also uses sock_common, so please just
185 * don't add nothing before this first member (__tw_common) --acme
187 struct sock_common __tw_common;
188 #define tw_family __tw_common.skc_family
189 #define tw_state __tw_common.skc_state
190 #define tw_reuse __tw_common.skc_reuse
191 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
192 #define tw_node __tw_common.skc_node
193 #define tw_bind_node __tw_common.skc_bind_node
194 #define tw_refcnt __tw_common.skc_refcnt
195 #define tw_xid __tw_common.skc_xid
196 #define tw_vx_info __tw_common.skc_vx_info
197 #define tw_nid __tw_common.skc_nid
198 #define tw_nx_info __tw_common.skc_nx_info
199 volatile unsigned char tw_substate;
200 unsigned char tw_rcv_wscale;
202 /* Socket demultiplex comparisons on incoming packets. */
203 /* these five are in inet_opt */
205 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
209 /* And these are ours. */
216 long tw_ts_recent_stamp;
217 unsigned long tw_ttd;
218 struct tcp_bind_bucket *tw_tb;
219 struct hlist_node tw_death_node;
220 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
221 struct in6_addr tw_v6_daddr;
222 struct in6_addr tw_v6_rcv_saddr;
227 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
228 struct hlist_head *list)
230 hlist_add_head(&tw->tw_node, list);
233 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
234 struct hlist_head *list)
236 hlist_add_head(&tw->tw_bind_node, list);
239 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
241 return tw->tw_death_node.pprev != NULL;
244 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
246 tw->tw_death_node.pprev = NULL;
249 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
251 __hlist_del(&tw->tw_death_node);
252 tw_dead_node_init(tw);
255 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
257 if (tw_dead_hashed(tw)) {
258 __tw_del_dead_node(tw);
264 #define tw_for_each(tw, node, head) \
265 hlist_for_each_entry(tw, node, head, tw_node)
267 #define tw_for_each_inmate(tw, node, jail) \
268 hlist_for_each_entry(tw, node, jail, tw_death_node)
270 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
271 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
273 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
275 static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
277 return likely(sk->sk_state != TCP_TIME_WAIT) ?
278 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
281 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
282 static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
284 return likely(sk->sk_state != TCP_TIME_WAIT) ?
285 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
288 static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
290 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
293 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
295 static inline int tcp_v6_ipv6only(const struct sock *sk)
297 return likely(sk->sk_state != TCP_TIME_WAIT) ?
298 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
301 # define __tcp_v6_rcv_saddr(__sk) NULL
302 # define tcp_v6_rcv_saddr(__sk) NULL
303 # define tcptw_sk_ipv6only(__sk) 0
304 # define tcp_v6_ipv6only(__sk) 0
307 extern kmem_cache_t *tcp_timewait_cachep;
309 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
311 if (atomic_dec_and_test(&tw->tw_refcnt)) {
312 #ifdef INET_REFCNT_DEBUG
313 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
315 kmem_cache_free(tcp_timewait_cachep, tw);
319 extern atomic_t tcp_orphan_count;
320 extern int tcp_tw_count;
321 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
322 extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
323 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
326 /* Socket demux engine toys. */
328 #define TCP_COMBINED_PORTS(__sport, __dport) \
329 (((__u32)(__sport)<<16) | (__u32)(__dport))
330 #else /* __LITTLE_ENDIAN */
331 #define TCP_COMBINED_PORTS(__sport, __dport) \
332 (((__u32)(__dport)<<16) | (__u32)(__sport))
335 #if (BITS_PER_LONG == 64)
337 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
338 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
339 #else /* __LITTLE_ENDIAN */
340 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
341 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
342 #endif /* __BIG_ENDIAN */
343 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
344 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
345 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
346 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
347 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
348 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
349 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
350 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
351 #else /* 32-bit arch */
352 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
353 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
354 ((inet_sk(__sk)->daddr == (__saddr)) && \
355 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
356 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
357 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
358 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
359 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
360 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
361 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
362 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
363 #endif /* 64-bit arch */
365 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
366 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
367 ((__sk)->sk_family == AF_INET6) && \
368 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \
369 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
370 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
372 /* These can have wildcards, don't try too hard. */
373 static __inline__ int tcp_lhashfn(unsigned short num)
375 return num & (TCP_LHTABLE_SIZE - 1);
378 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
380 return tcp_lhashfn(inet_sk(sk)->num);
383 #define MAX_TCP_HEADER (128 + MAX_HEADER)
386 * Never offer a window over 32767 without using window scaling. Some
387 * poor stacks do signed 16bit maths!
389 #define MAX_TCP_WINDOW 32767U
391 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
392 #define TCP_MIN_MSS 88U
394 /* Minimal RCV_MSS. */
395 #define TCP_MIN_RCVMSS 536U
397 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
398 #define TCP_FASTRETRANS_THRESH 3
400 /* Maximal reordering. */
401 #define TCP_MAX_REORDERING 127
403 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
404 #define TCP_MAX_QUICKACKS 16U
406 /* urg_data states */
407 #define TCP_URG_VALID 0x0100
408 #define TCP_URG_NOTYET 0x0200
409 #define TCP_URG_READ 0x0400
411 #define TCP_RETR1 3 /*
412 * This is how many retries it does before it
413 * tries to figure out if the gateway is
414 * down. Minimal RFC value is 3; it corresponds
415 * to ~3sec-8min depending on RTO.
418 #define TCP_RETR2 15 /*
419 * This should take at least
420 * 90 minutes to time out.
421 * RFC1122 says that the limit is 100 sec.
422 * 15 is ~13-30min depending on RTO.
425 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
426 * connection: ~180sec is RFC minumum */
428 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
429 * connection: ~180sec is RFC minumum */
432 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
433 * socket. 7 is ~50sec-16min.
437 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
438 * state, about 60 seconds */
439 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
440 /* BSD style FIN_WAIT2 deadlock breaker.
441 * It used to be 3min, new value is 60sec,
442 * to combine FIN-WAIT-2 timeout with
446 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
448 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
449 #define TCP_ATO_MIN ((unsigned)(HZ/25))
451 #define TCP_DELACK_MIN 4U
452 #define TCP_ATO_MIN 4U
454 #define TCP_RTO_MAX ((unsigned)(120*HZ))
455 #define TCP_RTO_MIN ((unsigned)(HZ/5))
456 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
458 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
459 * for local resources.
462 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
463 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
464 #define TCP_KEEPALIVE_INTVL (75*HZ)
466 #define MAX_TCP_KEEPIDLE 32767
467 #define MAX_TCP_KEEPINTVL 32767
468 #define MAX_TCP_KEEPCNT 127
469 #define MAX_TCP_SYNCNT 127
471 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
472 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
474 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
475 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
476 * after this time. It should be equal
477 * (or greater than) TCP_TIMEWAIT_LEN
478 * to provide reliability equal to one
479 * provided by timewait state.
481 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
482 * timestamps. It must be less than
483 * minimal timewait lifetime.
486 #define TCP_TW_RECYCLE_SLOTS_LOG 5
487 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
489 /* If time > 4sec, it is "slow" path, no recycling is required,
490 so that we select tick to get range about 4 seconds.
493 #if HZ <= 16 || HZ > 4096
494 # error Unsupported: HZ <= 16 or HZ > 4096
496 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
498 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
500 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
502 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
504 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
506 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
508 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
510 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
513 #define BICTCP_1_OVER_BETA 8 /*
515 * multiplicative decrease factor
517 #define BICTCP_MAX_INCREMENT 32 /*
518 * Limit on the amount of
519 * increment allowed during
522 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
523 * log(B/Smin)/log(B/(B-1))+1,
527 #define BICTCP_B 4 /*
529 * go to point (max+min)/N
536 #define TCPOPT_NOP 1 /* Padding */
537 #define TCPOPT_EOL 0 /* End of options */
538 #define TCPOPT_MSS 2 /* Segment size negotiating */
539 #define TCPOPT_WINDOW 3 /* Window scaling */
540 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
541 #define TCPOPT_SACK 5 /* SACK Block */
542 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
548 #define TCPOLEN_MSS 4
549 #define TCPOLEN_WINDOW 3
550 #define TCPOLEN_SACK_PERM 2
551 #define TCPOLEN_TIMESTAMP 10
553 /* But this is what stacks really send out. */
554 #define TCPOLEN_TSTAMP_ALIGNED 12
555 #define TCPOLEN_WSCALE_ALIGNED 4
556 #define TCPOLEN_SACKPERM_ALIGNED 4
557 #define TCPOLEN_SACK_BASE 2
558 #define TCPOLEN_SACK_BASE_ALIGNED 4
559 #define TCPOLEN_SACK_PERBLOCK 8
561 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
562 #define TCP_TIME_DACK 2 /* Delayed ack timer */
563 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
564 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
566 /* Flags in tp->nonagle */
567 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
568 #define TCP_NAGLE_CORK 2 /* Socket is corked */
569 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
571 /* sysctl variables for tcp */
572 extern int sysctl_max_syn_backlog;
573 extern int sysctl_tcp_timestamps;
574 extern int sysctl_tcp_window_scaling;
575 extern int sysctl_tcp_sack;
576 extern int sysctl_tcp_fin_timeout;
577 extern int sysctl_tcp_tw_recycle;
578 extern int sysctl_tcp_keepalive_time;
579 extern int sysctl_tcp_keepalive_probes;
580 extern int sysctl_tcp_keepalive_intvl;
581 extern int sysctl_tcp_syn_retries;
582 extern int sysctl_tcp_synack_retries;
583 extern int sysctl_tcp_retries1;
584 extern int sysctl_tcp_retries2;
585 extern int sysctl_tcp_orphan_retries;
586 extern int sysctl_tcp_syncookies;
587 extern int sysctl_tcp_retrans_collapse;
588 extern int sysctl_tcp_stdurg;
589 extern int sysctl_tcp_rfc1337;
590 extern int sysctl_tcp_abort_on_overflow;
591 extern int sysctl_tcp_max_orphans;
592 extern int sysctl_tcp_max_tw_buckets;
593 extern int sysctl_tcp_fack;
594 extern int sysctl_tcp_reordering;
595 extern int sysctl_tcp_ecn;
596 extern int sysctl_tcp_dsack;
597 extern int sysctl_tcp_mem[3];
598 extern int sysctl_tcp_wmem[3];
599 extern int sysctl_tcp_rmem[3];
600 extern int sysctl_tcp_app_win;
601 extern int sysctl_tcp_adv_win_scale;
602 extern int sysctl_tcp_tw_reuse;
603 extern int sysctl_tcp_frto;
604 extern int sysctl_tcp_low_latency;
605 extern int sysctl_tcp_westwood;
606 extern int sysctl_tcp_vegas_cong_avoid;
607 extern int sysctl_tcp_vegas_alpha;
608 extern int sysctl_tcp_vegas_beta;
609 extern int sysctl_tcp_vegas_gamma;
610 extern int sysctl_tcp_nometrics_save;
611 extern int sysctl_tcp_bic;
612 extern int sysctl_tcp_bic_fast_convergence;
613 extern int sysctl_tcp_bic_low_window;
614 extern int sysctl_tcp_moderate_rcvbuf;
615 extern int sysctl_tcp_tso_win_divisor;
617 extern atomic_t tcp_memory_allocated;
618 extern atomic_t tcp_sockets_allocated;
619 extern int tcp_memory_pressure;
623 struct or_calltable {
625 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
626 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
627 void (*destructor) (struct open_request *req);
628 void (*send_reset) (struct sk_buff *skb);
631 struct tcp_v4_open_req {
634 struct ip_options *opt;
637 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
638 struct tcp_v6_open_req {
639 struct in6_addr loc_addr;
640 struct in6_addr rmt_addr;
641 struct sk_buff *pktopts;
646 /* this structure is too big */
647 struct open_request {
648 struct open_request *dl_next; /* Must be first member! */
655 __u16 snd_wscale : 4,
662 /* The following two fields can be easily recomputed I think -AK */
663 __u32 window_clamp; /* window clamp at creation time */
664 __u32 rcv_wnd; /* rcv_wnd offered first time */
666 unsigned long expires;
667 struct or_calltable *class;
670 struct tcp_v4_open_req v4_req;
671 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
672 struct tcp_v6_open_req v6_req;
677 /* SLAB cache for open requests. */
678 extern kmem_cache_t *tcp_openreq_cachep;
680 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
681 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
683 static inline void tcp_openreq_free(struct open_request *req)
685 req->class->destructor(req);
686 tcp_openreq_fastfree(req);
689 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
690 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
692 #define TCP_INET_FAMILY(fam) 1
696 * Pointers to address related TCP functions
697 * (i.e. things that depend on the address family)
701 int (*queue_xmit) (struct sk_buff *skb,
704 void (*send_check) (struct sock *sk,
707 struct sk_buff *skb);
709 int (*rebuild_header) (struct sock *sk);
711 int (*conn_request) (struct sock *sk,
712 struct sk_buff *skb);
714 struct sock * (*syn_recv_sock) (struct sock *sk,
716 struct open_request *req,
717 struct dst_entry *dst);
719 int (*remember_stamp) (struct sock *sk);
721 __u16 net_header_len;
723 int (*setsockopt) (struct sock *sk,
729 int (*getsockopt) (struct sock *sk,
736 void (*addr2sockaddr) (struct sock *sk,
743 * The next routines deal with comparing 32 bit unsigned ints
744 * and worry about wraparound (automatic with unsigned arithmetic).
747 static inline int before(__u32 seq1, __u32 seq2)
749 return (__s32)(seq1-seq2) < 0;
752 static inline int after(__u32 seq1, __u32 seq2)
754 return (__s32)(seq2-seq1) < 0;
758 /* is s2<=s1<=s3 ? */
759 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
761 return seq3 - seq2 >= seq1 - seq2;
765 extern struct proto tcp_prot;
767 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
768 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
769 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
770 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
771 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
772 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
773 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
775 extern void tcp_put_port(struct sock *sk);
776 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
778 extern void tcp_v4_err(struct sk_buff *skb, u32);
780 extern void tcp_shutdown (struct sock *sk, int how);
782 extern int tcp_v4_rcv(struct sk_buff *skb);
784 extern struct sock * tcp_v4_lookup_listener(u32 daddr, unsigned short hnum, int dif);
786 extern int tcp_v4_remember_stamp(struct sock *sk);
788 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
790 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
791 struct msghdr *msg, size_t size);
792 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
794 extern int tcp_ioctl(struct sock *sk,
798 extern int tcp_rcv_state_process(struct sock *sk,
803 extern int tcp_rcv_established(struct sock *sk,
808 extern void tcp_rcv_space_adjust(struct sock *sk);
817 static inline void tcp_schedule_ack(struct tcp_opt *tp)
819 tp->ack.pending |= TCP_ACK_SCHED;
822 static inline int tcp_ack_scheduled(struct tcp_opt *tp)
824 return tp->ack.pending&TCP_ACK_SCHED;
827 static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
829 if (tp->ack.quick && --tp->ack.quick == 0) {
830 /* Leaving quickack mode we deflate ATO. */
831 tp->ack.ato = TCP_ATO_MIN;
835 extern void tcp_enter_quickack_mode(struct tcp_opt *tp);
837 static __inline__ void tcp_delack_init(struct tcp_opt *tp)
839 memset(&tp->ack, 0, sizeof(tp->ack));
842 static inline void tcp_clear_options(struct tcp_opt *tp)
844 tp->tstamp_ok = tp->sack_ok = tp->wscale_ok = tp->snd_wscale = 0;
856 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
861 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
862 struct open_request *req,
863 struct open_request **prev);
864 extern int tcp_child_process(struct sock *parent,
866 struct sk_buff *skb);
867 extern void tcp_enter_frto(struct sock *sk);
868 extern void tcp_enter_loss(struct sock *sk, int how);
869 extern void tcp_clear_retrans(struct tcp_opt *tp);
870 extern void tcp_update_metrics(struct sock *sk);
872 extern void tcp_close(struct sock *sk,
874 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
875 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
877 extern int tcp_getsockopt(struct sock *sk, int level,
881 extern int tcp_setsockopt(struct sock *sk, int level,
882 int optname, char __user *optval,
884 extern void tcp_set_keepalive(struct sock *sk, int val);
885 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
887 size_t len, int nonblock,
888 int flags, int *addr_len);
890 extern int tcp_listen_start(struct sock *sk);
892 extern void tcp_parse_options(struct sk_buff *skb,
897 * TCP v4 functions exported for the inet6 API
900 extern int tcp_v4_rebuild_header(struct sock *sk);
902 extern int tcp_v4_build_header(struct sock *sk,
903 struct sk_buff *skb);
905 extern void tcp_v4_send_check(struct sock *sk,
906 struct tcphdr *th, int len,
907 struct sk_buff *skb);
909 extern int tcp_v4_conn_request(struct sock *sk,
910 struct sk_buff *skb);
912 extern struct sock * tcp_create_openreq_child(struct sock *sk,
913 struct open_request *req,
914 struct sk_buff *skb);
916 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
918 struct open_request *req,
919 struct dst_entry *dst);
921 extern int tcp_v4_do_rcv(struct sock *sk,
922 struct sk_buff *skb);
924 extern int tcp_v4_connect(struct sock *sk,
925 struct sockaddr *uaddr,
928 extern int tcp_connect(struct sock *sk);
930 extern struct sk_buff * tcp_make_synack(struct sock *sk,
931 struct dst_entry *dst,
932 struct open_request *req);
934 extern int tcp_disconnect(struct sock *sk, int flags);
936 extern void tcp_unhash(struct sock *sk);
938 extern int tcp_v4_hash_connecting(struct sock *sk);
941 /* From syncookies.c */
942 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
943 struct ip_options *opt);
944 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
949 extern int tcp_write_xmit(struct sock *, int nonagle);
950 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
951 extern void tcp_xmit_retransmit_queue(struct sock *);
952 extern void tcp_simple_retransmit(struct sock *);
953 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
955 extern void tcp_send_probe0(struct sock *);
956 extern void tcp_send_partial(struct sock *);
957 extern int tcp_write_wakeup(struct sock *);
958 extern void tcp_send_fin(struct sock *sk);
959 extern void tcp_send_active_reset(struct sock *sk, int priority);
960 extern int tcp_send_synack(struct sock *);
961 extern void tcp_push_one(struct sock *, unsigned mss_now);
962 extern void tcp_send_ack(struct sock *sk);
963 extern void tcp_send_delayed_ack(struct sock *sk);
964 extern void cleanup_rbuf(struct sock *sk, int copied);
967 extern void tcp_init_xmit_timers(struct sock *);
968 extern void tcp_clear_xmit_timers(struct sock *);
970 extern void tcp_delete_keepalive_timer(struct sock *);
971 extern void tcp_reset_keepalive_timer(struct sock *, unsigned long);
972 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
973 extern unsigned int tcp_current_mss(struct sock *sk, int large);
976 extern const char tcp_timer_bug_msg[];
980 extern void tcp_get_info(struct sock *, struct tcp_info *);
982 /* Read 'sendfile()'-style from a TCP socket */
983 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
984 unsigned int, size_t);
985 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
986 sk_read_actor_t recv_actor);
988 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
990 struct tcp_opt *tp = tcp_sk(sk);
993 case TCP_TIME_RETRANS:
994 case TCP_TIME_PROBE0:
997 #ifdef TCP_CLEAR_TIMERS
998 sk_stop_timer(sk, &tp->retransmit_timer);
1002 tp->ack.blocked = 0;
1003 tp->ack.pending = 0;
1005 #ifdef TCP_CLEAR_TIMERS
1006 sk_stop_timer(sk, &tp->delack_timer);
1011 printk(tcp_timer_bug_msg);
1019 * Reset the retransmission timer
1021 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
1023 struct tcp_opt *tp = tcp_sk(sk);
1025 if (when > TCP_RTO_MAX) {
1027 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
1033 case TCP_TIME_RETRANS:
1034 case TCP_TIME_PROBE0:
1036 tp->timeout = jiffies+when;
1037 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
1041 tp->ack.pending |= TCP_ACK_TIMER;
1042 tp->ack.timeout = jiffies+when;
1043 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
1048 printk(tcp_timer_bug_msg);
1053 /* Initialize RCV_MSS value.
1054 * RCV_MSS is an our guess about MSS used by the peer.
1055 * We haven't any direct information about the MSS.
1056 * It's better to underestimate the RCV_MSS rather than overestimate.
1057 * Overestimations make us ACKing less frequently than needed.
1058 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1061 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1063 struct tcp_opt *tp = tcp_sk(sk);
1064 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1066 hint = min(hint, tp->rcv_wnd/2);
1067 hint = min(hint, TCP_MIN_RCVMSS);
1068 hint = max(hint, TCP_MIN_MSS);
1070 tp->ack.rcv_mss = hint;
1073 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1075 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1076 ntohl(TCP_FLAG_ACK) |
1080 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1082 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1085 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1087 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1089 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1091 tcp_fast_path_on(tp);
1094 /* Compute the actual receive window we are currently advertising.
1095 * Rcv_nxt can be after the window if our peer push more data
1096 * than the offered window.
1098 static __inline__ u32 tcp_receive_window(const struct tcp_opt *tp)
1100 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1107 /* Choose a new window, without checks for shrinking, and without
1108 * scaling applied to the result. The caller does these things
1109 * if necessary. This is a "raw" window selection.
1111 extern u32 __tcp_select_window(struct sock *sk);
1113 /* TCP timestamps are only 32-bits, this causes a slight
1114 * complication on 64-bit systems since we store a snapshot
1115 * of jiffies in the buffer control blocks below. We decidely
1116 * only use of the low 32-bits of jiffies and hide the ugly
1117 * casts with the following macro.
1119 #define tcp_time_stamp ((__u32)(jiffies))
1121 /* This is what the send packet queueing engine uses to pass
1122 * TCP per-packet control information to the transmission
1123 * code. We also store the host-order sequence numbers in
1124 * here too. This is 36 bytes on 32-bit architectures,
1125 * 40 bytes on 64-bit machines, if this grows please adjust
1126 * skbuff.h:skbuff->cb[xxx] size appropriately.
1130 struct inet_skb_parm h4;
1131 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1132 struct inet6_skb_parm h6;
1134 } header; /* For incoming frames */
1135 __u32 seq; /* Starting sequence number */
1136 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1137 __u32 when; /* used to compute rtt's */
1138 __u8 flags; /* TCP header flags. */
1140 /* NOTE: These must match up to the flags byte in a
1143 #define TCPCB_FLAG_FIN 0x01
1144 #define TCPCB_FLAG_SYN 0x02
1145 #define TCPCB_FLAG_RST 0x04
1146 #define TCPCB_FLAG_PSH 0x08
1147 #define TCPCB_FLAG_ACK 0x10
1148 #define TCPCB_FLAG_URG 0x20
1149 #define TCPCB_FLAG_ECE 0x40
1150 #define TCPCB_FLAG_CWR 0x80
1152 __u8 sacked; /* State flags for SACK/FACK. */
1153 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1154 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1155 #define TCPCB_LOST 0x04 /* SKB is lost */
1156 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1158 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1159 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1161 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1163 #define TCPCB_AT_TAIL (TCPCB_URG)
1165 __u16 urg_ptr; /* Valid w/URG flags is set. */
1166 __u32 ack_seq; /* Sequence number ACK'd */
1169 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1171 #include <net/tcp_ecn.h>
1173 /* Due to TSO, an SKB can be composed of multiple actual
1174 * packets. To keep these tracked properly, we use this.
1176 static inline int tcp_skb_pcount(const struct sk_buff *skb)
1178 return skb_shinfo(skb)->tso_segs;
1181 /* This is valid iff tcp_skb_pcount() > 1. */
1182 static inline int tcp_skb_mss(const struct sk_buff *skb)
1184 return skb_shinfo(skb)->tso_size;
1187 static inline void tcp_inc_pcount(tcp_pcount_t *count,
1188 const struct sk_buff *skb)
1190 count->val += tcp_skb_pcount(skb);
1193 static inline void tcp_inc_pcount_explicit(tcp_pcount_t *count, int amt)
1198 static inline void tcp_dec_pcount_explicit(tcp_pcount_t *count, int amt)
1203 static inline void tcp_dec_pcount(tcp_pcount_t *count,
1204 const struct sk_buff *skb)
1206 count->val -= tcp_skb_pcount(skb);
1209 static inline void tcp_dec_pcount_approx(tcp_pcount_t *count,
1210 const struct sk_buff *skb)
1213 count->val -= tcp_skb_pcount(skb);
1214 if ((int)count->val < 0)
1219 static inline __u32 tcp_get_pcount(const tcp_pcount_t *count)
1224 static inline void tcp_set_pcount(tcp_pcount_t *count, __u32 val)
1229 static inline void tcp_packets_out_inc(struct sock *sk,
1231 const struct sk_buff *skb)
1233 int orig = tcp_get_pcount(&tp->packets_out);
1235 tcp_inc_pcount(&tp->packets_out, skb);
1237 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
1240 static inline void tcp_packets_out_dec(struct tcp_opt *tp,
1241 const struct sk_buff *skb)
1243 tcp_dec_pcount(&tp->packets_out, skb);
1246 /* This determines how many packets are "in the network" to the best
1247 * of our knowledge. In many cases it is conservative, but where
1248 * detailed information is available from the receiver (via SACK
1249 * blocks etc.) we can make more aggressive calculations.
1251 * Use this for decisions involving congestion control, use just
1252 * tp->packets_out to determine if the send queue is empty or not.
1254 * Read this equation as:
1256 * "Packets sent once on transmission queue" MINUS
1257 * "Packets left network, but not honestly ACKed yet" PLUS
1258 * "Packets fast retransmitted"
1260 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_opt *tp)
1262 return (tcp_get_pcount(&tp->packets_out) -
1263 tcp_get_pcount(&tp->left_out) +
1264 tcp_get_pcount(&tp->retrans_out));
1268 * Which congestion algorithim is in use on the connection.
1270 #define tcp_is_vegas(__tp) ((__tp)->adv_cong == TCP_VEGAS)
1271 #define tcp_is_westwood(__tp) ((__tp)->adv_cong == TCP_WESTWOOD)
1272 #define tcp_is_bic(__tp) ((__tp)->adv_cong == TCP_BIC)
1274 /* Recalculate snd_ssthresh, we want to set it to:
1277 * one half the current congestion window, but no
1278 * less than two segments
1281 * behave like Reno until low_window is reached,
1282 * then increase congestion window slowly
1284 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1286 if (tcp_is_bic(tp)) {
1287 if (sysctl_tcp_bic_fast_convergence &&
1288 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1289 tp->bictcp.last_max_cwnd
1290 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1291 / (BICTCP_1_OVER_BETA/2);
1293 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1295 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1296 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1300 return max(tp->snd_cwnd >> 1U, 2U);
1303 /* Stop taking Vegas samples for now. */
1304 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1306 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1308 /* There are several situations when we must "re-start" Vegas:
1310 * o when a connection is established
1312 * o after fast recovery
1313 * o when we send a packet and there is no outstanding
1314 * unacknowledged data (restarting an idle connection)
1316 * In these circumstances we cannot do a Vegas calculation at the
1317 * end of the first RTT, because any calculation we do is using
1318 * stale info -- both the saved cwnd and congestion feedback are
1321 * Instead we must wait until the completion of an RTT during
1322 * which we actually receive ACKs.
1325 /* Begin taking Vegas samples next time we send something. */
1326 tp->vegas.doing_vegas_now = 1;
1328 /* Set the beginning of the next send window. */
1329 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1331 tp->vegas.cntRTT = 0;
1332 tp->vegas.minRTT = 0x7fffffff;
1335 /* Should we be taking Vegas samples right now? */
1336 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1338 extern void tcp_ca_init(struct tcp_opt *tp);
1340 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1342 if (tcp_is_vegas(tp)) {
1343 if (ca_state == TCP_CA_Open)
1344 tcp_vegas_enable(tp);
1346 tcp_vegas_disable(tp);
1348 tp->ca_state = ca_state;
1351 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1352 * The exception is rate halving phase, when cwnd is decreasing towards
1355 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1357 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1358 return tp->snd_ssthresh;
1360 return max(tp->snd_ssthresh,
1361 ((tp->snd_cwnd >> 1) +
1362 (tp->snd_cwnd >> 2)));
1365 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1368 (tcp_get_pcount(&tp->sacked_out) >=
1369 tcp_get_pcount(&tp->packets_out) - tcp_get_pcount(&tp->lost_out)))
1370 tcp_set_pcount(&tp->sacked_out,
1371 (tcp_get_pcount(&tp->packets_out) -
1372 tcp_get_pcount(&tp->lost_out)));
1373 tcp_set_pcount(&tp->left_out,
1374 (tcp_get_pcount(&tp->sacked_out) +
1375 tcp_get_pcount(&tp->lost_out)));
1378 extern void tcp_cwnd_application_limited(struct sock *sk);
1380 /* Congestion window validation. (RFC2861) */
1382 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1384 __u32 packets_out = tcp_get_pcount(&tp->packets_out);
1386 if (packets_out >= tp->snd_cwnd) {
1387 /* Network is feed fully. */
1388 tp->snd_cwnd_used = 0;
1389 tp->snd_cwnd_stamp = tcp_time_stamp;
1391 /* Network starves. */
1392 if (tcp_get_pcount(&tp->packets_out) > tp->snd_cwnd_used)
1393 tp->snd_cwnd_used = tcp_get_pcount(&tp->packets_out);
1395 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1396 tcp_cwnd_application_limited(sk);
1400 /* Set slow start threshould and cwnd not falling to slow start */
1401 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1403 tp->undo_marker = 0;
1404 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1405 tp->snd_cwnd = min(tp->snd_cwnd,
1406 tcp_packets_in_flight(tp) + 1U);
1407 tp->snd_cwnd_cnt = 0;
1408 tp->high_seq = tp->snd_nxt;
1409 tp->snd_cwnd_stamp = tcp_time_stamp;
1410 TCP_ECN_queue_cwr(tp);
1413 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1415 tp->prior_ssthresh = 0;
1416 if (tp->ca_state < TCP_CA_CWR) {
1417 __tcp_enter_cwr(tp);
1418 tcp_set_ca_state(tp, TCP_CA_CWR);
1422 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1424 /* Slow start with delack produces 3 packets of burst, so that
1425 * it is safe "de facto".
1427 static __inline__ __u32 tcp_max_burst(const struct tcp_opt *tp)
1432 static __inline__ int tcp_minshall_check(const struct tcp_opt *tp)
1434 return after(tp->snd_sml,tp->snd_una) &&
1435 !after(tp->snd_sml, tp->snd_nxt);
1438 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss,
1439 const struct sk_buff *skb)
1442 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1445 /* Return 0, if packet can be sent now without violation Nagle's rules:
1446 1. It is full sized.
1447 2. Or it contains FIN.
1448 3. Or higher layers meant to force a packet boundary, hence the PSH bit.
1449 4. Or TCP_NODELAY was set.
1450 5. Or TCP_CORK is not set, and all sent packets are ACKed.
1451 With Minshall's modification: all sent small packets are ACKed.
1454 static __inline__ int
1455 tcp_nagle_check(const struct tcp_opt *tp, const struct sk_buff *skb,
1456 unsigned mss_now, int nonagle)
1458 return (skb->len < mss_now &&
1459 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1460 ((nonagle&TCP_NAGLE_CORK) ||
1462 tcp_get_pcount(&tp->packets_out) &&
1463 tcp_minshall_check(tp))));
1466 extern void tcp_set_skb_tso_segs(struct sk_buff *, unsigned int);
1468 /* This checks if the data bearing packet SKB (usually sk->sk_send_head)
1469 * should be put on the wire right now.
1471 static __inline__ int tcp_snd_test(const struct tcp_opt *tp,
1472 struct sk_buff *skb,
1473 unsigned cur_mss, int nonagle)
1475 int pkts = tcp_skb_pcount(skb);
1478 tcp_set_skb_tso_segs(skb, tp->mss_cache_std);
1479 pkts = tcp_skb_pcount(skb);
1482 /* RFC 1122 - section 4.2.3.4
1486 * a) The right edge of this frame exceeds the window
1487 * b) There are packets in flight and we have a small segment
1488 * [SWS avoidance and Nagle algorithm]
1489 * (part of SWS is done on packetization)
1490 * Minshall version sounds: there are no _small_
1491 * segments in flight. (tcp_nagle_check)
1492 * c) We have too many packets 'in flight'
1494 * Don't use the nagle rule for urgent data (or
1495 * for the final FIN -DaveM).
1497 * Also, Nagle rule does not apply to frames, which
1498 * sit in the middle of queue (they have no chances
1499 * to get new data) and if room at tail of skb is
1500 * not enough to save something seriously (<32 for now).
1503 /* Don't be strict about the congestion window for the
1504 * final FIN frame. -DaveM
1506 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1507 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1508 (((tcp_packets_in_flight(tp) + (pkts-1)) < tp->snd_cwnd) ||
1509 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1510 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1513 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1515 if (!tcp_get_pcount(&tp->packets_out) && !tp->pending)
1516 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1519 static __inline__ int tcp_skb_is_last(const struct sock *sk,
1520 const struct sk_buff *skb)
1522 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1525 /* Push out any pending frames which were held back due to
1526 * TCP_CORK or attempt at coalescing tiny packets.
1527 * The socket must be locked by the caller.
1529 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1534 struct sk_buff *skb = sk->sk_send_head;
1537 if (!tcp_skb_is_last(sk, skb))
1538 nonagle = TCP_NAGLE_PUSH;
1539 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1540 tcp_write_xmit(sk, nonagle))
1541 tcp_check_probe_timer(sk, tp);
1543 tcp_cwnd_validate(sk, tp);
1546 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1549 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1552 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1554 struct sk_buff *skb = sk->sk_send_head;
1557 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1558 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1561 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1566 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1571 extern void tcp_destroy_sock(struct sock *sk);
1575 * Calculate(/check) TCP checksum
1577 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1578 unsigned long saddr, unsigned long daddr,
1581 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1584 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1586 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1589 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1591 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1592 __tcp_checksum_complete(skb);
1595 /* Prequeue for VJ style copy to user, combined with checksumming. */
1597 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1599 tp->ucopy.task = NULL;
1601 tp->ucopy.memory = 0;
1602 skb_queue_head_init(&tp->ucopy.prequeue);
1605 /* Packet is added to VJ-style prequeue for processing in process
1606 * context, if a reader task is waiting. Apparently, this exciting
1607 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1608 * failed somewhere. Latency? Burstiness? Well, at least now we will
1609 * see, why it failed. 8)8) --ANK
1611 * NOTE: is this not too big to inline?
1613 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1615 struct tcp_opt *tp = tcp_sk(sk);
1617 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1618 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1619 tp->ucopy.memory += skb->truesize;
1620 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1621 struct sk_buff *skb1;
1623 BUG_ON(sock_owned_by_user(sk));
1625 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1626 sk->sk_backlog_rcv(sk, skb1);
1627 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1630 tp->ucopy.memory = 0;
1631 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1632 wake_up_interruptible(sk->sk_sleep);
1633 if (!tcp_ack_scheduled(tp))
1634 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1645 static const char *statename[]={
1646 "Unused","Established","Syn Sent","Syn Recv",
1647 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1648 "Close Wait","Last ACK","Listen","Closing"
1652 static __inline__ void tcp_set_state(struct sock *sk, int state)
1654 int oldstate = sk->sk_state;
1657 case TCP_ESTABLISHED:
1658 if (oldstate != TCP_ESTABLISHED)
1659 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1663 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1664 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1666 sk->sk_prot->unhash(sk);
1667 if (tcp_sk(sk)->bind_hash &&
1668 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1672 if (oldstate==TCP_ESTABLISHED)
1673 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1676 /* Change state AFTER socket is unhashed to avoid closed
1677 * socket sitting in hash tables.
1679 sk->sk_state = state;
1682 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1686 static __inline__ void tcp_done(struct sock *sk)
1688 tcp_set_state(sk, TCP_CLOSE);
1689 tcp_clear_xmit_timers(sk);
1691 sk->sk_shutdown = SHUTDOWN_MASK;
1693 if (!sock_flag(sk, SOCK_DEAD))
1694 sk->sk_state_change(sk);
1696 tcp_destroy_sock(sk);
1699 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1706 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1708 if (tp->tstamp_ok) {
1709 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1710 (TCPOPT_NOP << 16) |
1711 (TCPOPT_TIMESTAMP << 8) |
1713 *ptr++ = htonl(tstamp);
1714 *ptr++ = htonl(tp->ts_recent);
1716 if (tp->eff_sacks) {
1717 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1720 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1721 (TCPOPT_NOP << 16) |
1722 (TCPOPT_SACK << 8) |
1723 (TCPOLEN_SACK_BASE +
1724 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1725 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1726 *ptr++ = htonl(sp[this_sack].start_seq);
1727 *ptr++ = htonl(sp[this_sack].end_seq);
1736 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1737 * If this is every changed make sure to change the definition of
1738 * MAX_SYN_SIZE to match the new maximum number of options that you
1741 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1742 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1744 /* We always get an MSS option.
1745 * The option bytes which will be seen in normal data
1746 * packets should timestamps be used, must be in the MSS
1747 * advertised. But we subtract them from tp->mss_cache so
1748 * that calculations in tcp_sendmsg are simpler etc.
1749 * So account for this fact here if necessary. If we
1750 * don't do this correctly, as a receiver we won't
1751 * recognize data packets as being full sized when we
1752 * should, and thus we won't abide by the delayed ACK
1754 * SACKs don't matter, we never delay an ACK when we
1755 * have any of those going out.
1757 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1760 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1761 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1763 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1764 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1765 *ptr++ = htonl(tstamp); /* TSVAL */
1766 *ptr++ = htonl(ts_recent); /* TSECR */
1768 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1769 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1771 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1774 /* Determine a window scaling and initial window to offer. */
1775 extern void tcp_select_initial_window(int __space, __u32 mss,
1776 __u32 *rcv_wnd, __u32 *window_clamp,
1777 int wscale_ok, __u8 *rcv_wscale);
1779 static inline int tcp_win_from_space(int space)
1781 return sysctl_tcp_adv_win_scale<=0 ?
1782 (space>>(-sysctl_tcp_adv_win_scale)) :
1783 space - (space>>sysctl_tcp_adv_win_scale);
1786 /* Note: caller must be prepared to deal with negative returns */
1787 static inline int tcp_space(const struct sock *sk)
1789 return tcp_win_from_space(sk->sk_rcvbuf -
1790 atomic_read(&sk->sk_rmem_alloc));
1793 static inline int tcp_full_space(const struct sock *sk)
1795 return tcp_win_from_space(sk->sk_rcvbuf);
1798 struct tcp_listen_opt
1800 u8 max_qlen_log; /* log_2 of maximal queued SYNs */
1805 struct open_request *syn_table[TCP_SYNQ_HSIZE];
1808 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1811 struct tcp_opt *tp = tcp_sk(sk);
1814 sk_acceptq_added(sk);
1816 if (!tp->accept_queue_tail) {
1817 tp->accept_queue = req;
1819 tp->accept_queue_tail->dl_next = req;
1821 tp->accept_queue_tail = req;
1822 req->dl_next = NULL;
1827 tcp_synq_removed(struct sock *sk, struct open_request *req)
1829 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1831 if (--lopt->qlen == 0)
1832 tcp_delete_keepalive_timer(sk);
1833 if (req->retrans == 0)
1837 static inline void tcp_synq_added(struct sock *sk)
1839 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1841 if (lopt->qlen++ == 0)
1842 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1846 static inline int tcp_synq_len(struct sock *sk)
1848 return tcp_sk(sk)->listen_opt->qlen;
1851 static inline int tcp_synq_young(struct sock *sk)
1853 return tcp_sk(sk)->listen_opt->qlen_young;
1856 static inline int tcp_synq_is_full(struct sock *sk)
1858 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1861 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1862 struct open_request **prev)
1864 write_lock(&tp->syn_wait_lock);
1865 *prev = req->dl_next;
1866 write_unlock(&tp->syn_wait_lock);
1869 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1870 struct open_request **prev)
1872 tcp_synq_unlink(tcp_sk(sk), req, prev);
1873 tcp_synq_removed(sk, req);
1874 tcp_openreq_free(req);
1877 static __inline__ void tcp_openreq_init(struct open_request *req,
1879 struct sk_buff *skb)
1881 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1882 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1883 req->mss = tp->mss_clamp;
1884 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1885 req->tstamp_ok = tp->tstamp_ok;
1886 req->sack_ok = tp->sack_ok;
1887 req->snd_wscale = tp->snd_wscale;
1888 req->wscale_ok = tp->wscale_ok;
1891 req->rmt_port = skb->h.th->source;
1894 extern void tcp_enter_memory_pressure(void);
1896 extern void tcp_listen_wlock(void);
1898 /* - We may sleep inside this lock.
1899 * - If sleeping is not required (or called from BH),
1900 * use plain read_(un)lock(&tcp_lhash_lock).
1903 static inline void tcp_listen_lock(void)
1905 /* read_lock synchronizes to candidates to writers */
1906 read_lock(&tcp_lhash_lock);
1907 atomic_inc(&tcp_lhash_users);
1908 read_unlock(&tcp_lhash_lock);
1911 static inline void tcp_listen_unlock(void)
1913 if (atomic_dec_and_test(&tcp_lhash_users))
1914 wake_up(&tcp_lhash_wait);
1917 static inline int keepalive_intvl_when(const struct tcp_opt *tp)
1919 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1922 static inline int keepalive_time_when(const struct tcp_opt *tp)
1924 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1927 static inline int tcp_fin_time(const struct tcp_opt *tp)
1929 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1931 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1932 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1937 static inline int tcp_paws_check(const struct tcp_opt *tp, int rst)
1939 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
1941 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
1944 /* RST segments are not recommended to carry timestamp,
1945 and, if they do, it is recommended to ignore PAWS because
1946 "their cleanup function should take precedence over timestamps."
1947 Certainly, it is mistake. It is necessary to understand the reasons
1948 of this constraint to relax it: if peer reboots, clock may go
1949 out-of-sync and half-open connections will not be reset.
1950 Actually, the problem would be not existing if all
1951 the implementations followed draft about maintaining clock
1952 via reboots. Linux-2.2 DOES NOT!
1954 However, we can relax time bounds for RST segments to MSL.
1956 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
1961 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
1963 sk->sk_route_caps = dst->dev->features;
1964 if (sk->sk_route_caps & NETIF_F_TSO) {
1965 if (sk->sk_no_largesend || dst->header_len)
1966 sk->sk_route_caps &= ~NETIF_F_TSO;
1970 #define TCP_CHECK_TIMER(sk) do { } while (0)
1972 static inline int tcp_use_frto(const struct sock *sk)
1974 const struct tcp_opt *tp = tcp_sk(sk);
1976 /* F-RTO must be activated in sysctl and there must be some
1977 * unsent new data, and the advertised window should allow
1980 return (sysctl_tcp_frto && sk->sk_send_head &&
1981 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1982 tp->snd_una + tp->snd_wnd));
1985 static inline void tcp_mib_init(void)
1988 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1989 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1990 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1991 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1995 enum tcp_seq_states {
1996 TCP_SEQ_STATE_LISTENING,
1997 TCP_SEQ_STATE_OPENREQ,
1998 TCP_SEQ_STATE_ESTABLISHED,
1999 TCP_SEQ_STATE_TIME_WAIT,
2002 struct tcp_seq_afinfo {
2003 struct module *owner;
2006 int (*seq_show) (struct seq_file *m, void *v);
2007 struct file_operations *seq_fops;
2010 struct tcp_iter_state {
2012 enum tcp_seq_states state;
2013 struct sock *syn_wait_sk;
2014 int bucket, sbucket, num, uid;
2015 struct seq_operations seq_ops;
2018 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2019 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2021 /* TCP Westwood functions and constants */
2023 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2024 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2026 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2028 if (tcp_is_westwood(tp))
2029 tp->westwood.rtt = rtt_seq;
2032 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2033 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2035 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2037 if (tcp_is_westwood(tcp_sk(sk)))
2038 __tcp_westwood_fast_bw(sk, skb);
2041 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2043 if (tcp_is_westwood(tcp_sk(sk)))
2044 __tcp_westwood_slow_bw(sk, skb);
2047 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2049 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2050 (__u32) (tp->mss_cache_std),
2054 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2056 return tcp_is_westwood(tp) ? __tcp_westwood_bw_rttmin(tp) : 0;
2059 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2063 if (tcp_is_westwood(tp)) {
2064 ssthresh = __tcp_westwood_bw_rttmin(tp);
2066 tp->snd_ssthresh = ssthresh;
2069 return (ssthresh != 0);
2072 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2076 if (tcp_is_westwood(tp)) {
2077 cwnd = __tcp_westwood_bw_rttmin(tp);
2079 tp->snd_cwnd = cwnd;