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;
162 extern struct sock *tcp_v4_lookup_listener(u32 addr, unsigned short hnum, int dif);
164 /* These are AF independent. */
165 static __inline__ int tcp_bhashfn(__u16 lport)
167 return (lport & (tcp_bhash_size - 1));
170 extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb,
171 unsigned short snum);
173 #if (BITS_PER_LONG == 64)
174 #define TCP_ADDRCMP_ALIGN_BYTES 8
176 #define TCP_ADDRCMP_ALIGN_BYTES 4
179 /* This is a TIME_WAIT bucket. It works around the memory consumption
180 * problems of sockets in such a state on heavily loaded servers, but
181 * without violating the protocol specification.
183 struct tcp_tw_bucket {
185 * Now struct sock also uses sock_common, so please just
186 * don't add nothing before this first member (__tw_common) --acme
188 struct sock_common __tw_common;
189 #define tw_family __tw_common.skc_family
190 #define tw_state __tw_common.skc_state
191 #define tw_reuse __tw_common.skc_reuse
192 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
193 #define tw_node __tw_common.skc_node
194 #define tw_bind_node __tw_common.skc_bind_node
195 #define tw_refcnt __tw_common.skc_refcnt
196 #define tw_xid __tw_common.skc_xid
197 #define tw_vx_info __tw_common.skc_vx_info
198 #define tw_nid __tw_common.skc_nid
199 #define tw_nx_info __tw_common.skc_nx_info
200 volatile unsigned char tw_substate;
201 unsigned char tw_rcv_wscale;
203 /* Socket demultiplex comparisons on incoming packets. */
204 /* these five are in inet_opt */
206 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
210 /* And these are ours. */
217 long tw_ts_recent_stamp;
218 unsigned long tw_ttd;
219 struct tcp_bind_bucket *tw_tb;
220 struct hlist_node tw_death_node;
221 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
222 struct in6_addr tw_v6_daddr;
223 struct in6_addr tw_v6_rcv_saddr;
228 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
229 struct hlist_head *list)
231 hlist_add_head(&tw->tw_node, list);
234 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
235 struct hlist_head *list)
237 hlist_add_head(&tw->tw_bind_node, list);
240 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
242 return tw->tw_death_node.pprev != NULL;
245 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
247 tw->tw_death_node.pprev = NULL;
250 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
252 __hlist_del(&tw->tw_death_node);
253 tw_dead_node_init(tw);
256 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
258 if (tw_dead_hashed(tw)) {
259 __tw_del_dead_node(tw);
265 #define tw_for_each(tw, node, head) \
266 hlist_for_each_entry(tw, node, head, tw_node)
268 #define tw_for_each_inmate(tw, node, jail) \
269 hlist_for_each_entry(tw, node, jail, tw_death_node)
271 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
272 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
274 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
276 static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
278 return likely(sk->sk_state != TCP_TIME_WAIT) ?
279 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
282 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
283 static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
285 return likely(sk->sk_state != TCP_TIME_WAIT) ?
286 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
289 static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
291 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
294 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
296 static inline int tcp_v6_ipv6only(const struct sock *sk)
298 return likely(sk->sk_state != TCP_TIME_WAIT) ?
299 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
302 # define __tcp_v6_rcv_saddr(__sk) NULL
303 # define tcp_v6_rcv_saddr(__sk) NULL
304 # define tcptw_sk_ipv6only(__sk) 0
305 # define tcp_v6_ipv6only(__sk) 0
308 extern kmem_cache_t *tcp_timewait_cachep;
310 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
312 if (atomic_dec_and_test(&tw->tw_refcnt)) {
313 #ifdef INET_REFCNT_DEBUG
314 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
316 kmem_cache_free(tcp_timewait_cachep, tw);
320 extern atomic_t tcp_orphan_count;
321 extern int tcp_tw_count;
322 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
323 extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
324 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
327 /* Socket demux engine toys. */
329 #define TCP_COMBINED_PORTS(__sport, __dport) \
330 (((__u32)(__sport)<<16) | (__u32)(__dport))
331 #else /* __LITTLE_ENDIAN */
332 #define TCP_COMBINED_PORTS(__sport, __dport) \
333 (((__u32)(__dport)<<16) | (__u32)(__sport))
336 #if (BITS_PER_LONG == 64)
338 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
339 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
340 #else /* __LITTLE_ENDIAN */
341 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
342 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
343 #endif /* __BIG_ENDIAN */
344 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
345 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
346 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
347 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
348 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
349 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
350 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
351 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
352 #else /* 32-bit arch */
353 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
354 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
355 ((inet_sk(__sk)->daddr == (__saddr)) && \
356 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
357 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
358 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
359 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
360 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
361 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
362 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
363 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
364 #endif /* 64-bit arch */
366 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
367 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
368 ((__sk)->sk_family == AF_INET6) && \
369 !ipv6_addr_cmp(&inet6_sk(__sk)->daddr, (__saddr)) && \
370 !ipv6_addr_cmp(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
371 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
373 /* These can have wildcards, don't try too hard. */
374 static __inline__ int tcp_lhashfn(unsigned short num)
376 return num & (TCP_LHTABLE_SIZE - 1);
379 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
381 return tcp_lhashfn(inet_sk(sk)->num);
384 #define MAX_TCP_HEADER (128 + MAX_HEADER)
387 * Never offer a window over 32767 without using window scaling. Some
388 * poor stacks do signed 16bit maths!
390 #define MAX_TCP_WINDOW 32767U
392 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
393 #define TCP_MIN_MSS 88U
395 /* Minimal RCV_MSS. */
396 #define TCP_MIN_RCVMSS 536U
398 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
399 #define TCP_FASTRETRANS_THRESH 3
401 /* Maximal reordering. */
402 #define TCP_MAX_REORDERING 127
404 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
405 #define TCP_MAX_QUICKACKS 16U
407 /* urg_data states */
408 #define TCP_URG_VALID 0x0100
409 #define TCP_URG_NOTYET 0x0200
410 #define TCP_URG_READ 0x0400
412 #define TCP_RETR1 3 /*
413 * This is how many retries it does before it
414 * tries to figure out if the gateway is
415 * down. Minimal RFC value is 3; it corresponds
416 * to ~3sec-8min depending on RTO.
419 #define TCP_RETR2 15 /*
420 * This should take at least
421 * 90 minutes to time out.
422 * RFC1122 says that the limit is 100 sec.
423 * 15 is ~13-30min depending on RTO.
426 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
427 * connection: ~180sec is RFC minumum */
429 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
430 * connection: ~180sec is RFC minumum */
433 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
434 * socket. 7 is ~50sec-16min.
438 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
439 * state, about 60 seconds */
440 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
441 /* BSD style FIN_WAIT2 deadlock breaker.
442 * It used to be 3min, new value is 60sec,
443 * to combine FIN-WAIT-2 timeout with
447 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
449 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
450 #define TCP_ATO_MIN ((unsigned)(HZ/25))
452 #define TCP_DELACK_MIN 4U
453 #define TCP_ATO_MIN 4U
455 #define TCP_RTO_MAX ((unsigned)(120*HZ))
456 #define TCP_RTO_MIN ((unsigned)(HZ/5))
457 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
459 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
460 * for local resources.
463 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
464 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
465 #define TCP_KEEPALIVE_INTVL (75*HZ)
467 #define MAX_TCP_KEEPIDLE 32767
468 #define MAX_TCP_KEEPINTVL 32767
469 #define MAX_TCP_KEEPCNT 127
470 #define MAX_TCP_SYNCNT 127
472 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
473 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
475 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
476 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
477 * after this time. It should be equal
478 * (or greater than) TCP_TIMEWAIT_LEN
479 * to provide reliability equal to one
480 * provided by timewait state.
482 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
483 * timestamps. It must be less than
484 * minimal timewait lifetime.
487 #define TCP_TW_RECYCLE_SLOTS_LOG 5
488 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
490 /* If time > 4sec, it is "slow" path, no recycling is required,
491 so that we select tick to get range about 4 seconds.
494 #if HZ <= 16 || HZ > 4096
495 # error Unsupported: HZ <= 16 or HZ > 4096
497 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
499 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
501 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
503 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
505 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
507 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
509 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
511 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
514 #define BICTCP_1_OVER_BETA 8 /*
516 * multiplicative decrease factor
518 #define BICTCP_MAX_INCREMENT 32 /*
519 * Limit on the amount of
520 * increment allowed during
523 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
524 * log(B/Smin)/log(B/(B-1))+1,
528 #define BICTCP_B 4 /*
530 * go to point (max+min)/N
537 #define TCPOPT_NOP 1 /* Padding */
538 #define TCPOPT_EOL 0 /* End of options */
539 #define TCPOPT_MSS 2 /* Segment size negotiating */
540 #define TCPOPT_WINDOW 3 /* Window scaling */
541 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
542 #define TCPOPT_SACK 5 /* SACK Block */
543 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
549 #define TCPOLEN_MSS 4
550 #define TCPOLEN_WINDOW 3
551 #define TCPOLEN_SACK_PERM 2
552 #define TCPOLEN_TIMESTAMP 10
554 /* But this is what stacks really send out. */
555 #define TCPOLEN_TSTAMP_ALIGNED 12
556 #define TCPOLEN_WSCALE_ALIGNED 4
557 #define TCPOLEN_SACKPERM_ALIGNED 4
558 #define TCPOLEN_SACK_BASE 2
559 #define TCPOLEN_SACK_BASE_ALIGNED 4
560 #define TCPOLEN_SACK_PERBLOCK 8
562 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
563 #define TCP_TIME_DACK 2 /* Delayed ack timer */
564 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
565 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
567 /* Flags in tp->nonagle */
568 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
569 #define TCP_NAGLE_CORK 2 /* Socket is corked */
570 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
572 /* sysctl variables for tcp */
573 extern int sysctl_max_syn_backlog;
574 extern int sysctl_tcp_timestamps;
575 extern int sysctl_tcp_window_scaling;
576 extern int sysctl_tcp_sack;
577 extern int sysctl_tcp_fin_timeout;
578 extern int sysctl_tcp_tw_recycle;
579 extern int sysctl_tcp_keepalive_time;
580 extern int sysctl_tcp_keepalive_probes;
581 extern int sysctl_tcp_keepalive_intvl;
582 extern int sysctl_tcp_syn_retries;
583 extern int sysctl_tcp_synack_retries;
584 extern int sysctl_tcp_retries1;
585 extern int sysctl_tcp_retries2;
586 extern int sysctl_tcp_orphan_retries;
587 extern int sysctl_tcp_syncookies;
588 extern int sysctl_tcp_retrans_collapse;
589 extern int sysctl_tcp_stdurg;
590 extern int sysctl_tcp_rfc1337;
591 extern int sysctl_tcp_abort_on_overflow;
592 extern int sysctl_tcp_max_orphans;
593 extern int sysctl_tcp_max_tw_buckets;
594 extern int sysctl_tcp_fack;
595 extern int sysctl_tcp_reordering;
596 extern int sysctl_tcp_ecn;
597 extern int sysctl_tcp_dsack;
598 extern int sysctl_tcp_mem[3];
599 extern int sysctl_tcp_wmem[3];
600 extern int sysctl_tcp_rmem[3];
601 extern int sysctl_tcp_app_win;
602 extern int sysctl_tcp_adv_win_scale;
603 extern int sysctl_tcp_tw_reuse;
604 extern int sysctl_tcp_frto;
605 extern int sysctl_tcp_low_latency;
606 extern int sysctl_tcp_westwood;
607 extern int sysctl_tcp_vegas_cong_avoid;
608 extern int sysctl_tcp_vegas_alpha;
609 extern int sysctl_tcp_vegas_beta;
610 extern int sysctl_tcp_vegas_gamma;
611 extern int sysctl_tcp_nometrics_save;
612 extern int sysctl_tcp_bic;
613 extern int sysctl_tcp_bic_fast_convergence;
614 extern int sysctl_tcp_bic_low_window;
615 extern int sysctl_tcp_moderate_rcvbuf;
616 extern int sysctl_tcp_tso_win_divisor;
618 extern atomic_t tcp_memory_allocated;
619 extern atomic_t tcp_sockets_allocated;
620 extern int tcp_memory_pressure;
624 struct or_calltable {
626 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
627 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
628 void (*destructor) (struct open_request *req);
629 void (*send_reset) (struct sk_buff *skb);
632 struct tcp_v4_open_req {
635 struct ip_options *opt;
638 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
639 struct tcp_v6_open_req {
640 struct in6_addr loc_addr;
641 struct in6_addr rmt_addr;
642 struct sk_buff *pktopts;
647 /* this structure is too big */
648 struct open_request {
649 struct open_request *dl_next; /* Must be first member! */
656 __u16 snd_wscale : 4,
663 /* The following two fields can be easily recomputed I think -AK */
664 __u32 window_clamp; /* window clamp at creation time */
665 __u32 rcv_wnd; /* rcv_wnd offered first time */
667 unsigned long expires;
668 struct or_calltable *class;
671 struct tcp_v4_open_req v4_req;
672 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
673 struct tcp_v6_open_req v6_req;
678 /* SLAB cache for open requests. */
679 extern kmem_cache_t *tcp_openreq_cachep;
681 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
682 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
684 static inline void tcp_openreq_free(struct open_request *req)
686 req->class->destructor(req);
687 tcp_openreq_fastfree(req);
690 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
691 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
693 #define TCP_INET_FAMILY(fam) 1
697 * Pointers to address related TCP functions
698 * (i.e. things that depend on the address family)
702 int (*queue_xmit) (struct sk_buff *skb,
705 void (*send_check) (struct sock *sk,
708 struct sk_buff *skb);
710 int (*rebuild_header) (struct sock *sk);
712 int (*conn_request) (struct sock *sk,
713 struct sk_buff *skb);
715 struct sock * (*syn_recv_sock) (struct sock *sk,
717 struct open_request *req,
718 struct dst_entry *dst);
720 int (*remember_stamp) (struct sock *sk);
722 __u16 net_header_len;
724 int (*setsockopt) (struct sock *sk,
730 int (*getsockopt) (struct sock *sk,
737 void (*addr2sockaddr) (struct sock *sk,
744 * The next routines deal with comparing 32 bit unsigned ints
745 * and worry about wraparound (automatic with unsigned arithmetic).
748 static inline int before(__u32 seq1, __u32 seq2)
750 return (__s32)(seq1-seq2) < 0;
753 static inline int after(__u32 seq1, __u32 seq2)
755 return (__s32)(seq2-seq1) < 0;
759 /* is s2<=s1<=s3 ? */
760 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
762 return seq3 - seq2 >= seq1 - seq2;
766 extern struct proto tcp_prot;
768 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
769 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
770 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
771 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
772 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
773 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
774 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
776 extern void tcp_put_port(struct sock *sk);
777 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
779 extern void tcp_v4_err(struct sk_buff *skb, u32);
781 extern void tcp_shutdown (struct sock *sk, int how);
783 extern int tcp_v4_rcv(struct sk_buff *skb);
785 extern int tcp_v4_remember_stamp(struct sock *sk);
787 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
789 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
790 struct msghdr *msg, size_t size);
791 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
793 extern int tcp_ioctl(struct sock *sk,
797 extern int tcp_rcv_state_process(struct sock *sk,
802 extern int tcp_rcv_established(struct sock *sk,
807 extern void tcp_rcv_space_adjust(struct sock *sk);
816 static inline void tcp_schedule_ack(struct tcp_opt *tp)
818 tp->ack.pending |= TCP_ACK_SCHED;
821 static inline int tcp_ack_scheduled(struct tcp_opt *tp)
823 return tp->ack.pending&TCP_ACK_SCHED;
826 static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
828 if (tp->ack.quick && --tp->ack.quick == 0) {
829 /* Leaving quickack mode we deflate ATO. */
830 tp->ack.ato = TCP_ATO_MIN;
834 extern void tcp_enter_quickack_mode(struct tcp_opt *tp);
836 static __inline__ void tcp_delack_init(struct tcp_opt *tp)
838 memset(&tp->ack, 0, sizeof(tp->ack));
841 static inline void tcp_clear_options(struct tcp_opt *tp)
843 tp->tstamp_ok = tp->sack_ok = tp->wscale_ok = tp->snd_wscale = 0;
855 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
860 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
861 struct open_request *req,
862 struct open_request **prev);
863 extern int tcp_child_process(struct sock *parent,
865 struct sk_buff *skb);
866 extern void tcp_enter_frto(struct sock *sk);
867 extern void tcp_enter_loss(struct sock *sk, int how);
868 extern void tcp_clear_retrans(struct tcp_opt *tp);
869 extern void tcp_update_metrics(struct sock *sk);
871 extern void tcp_close(struct sock *sk,
873 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
874 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
876 extern int tcp_getsockopt(struct sock *sk, int level,
880 extern int tcp_setsockopt(struct sock *sk, int level,
881 int optname, char __user *optval,
883 extern void tcp_set_keepalive(struct sock *sk, int val);
884 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
886 size_t len, int nonblock,
887 int flags, int *addr_len);
889 extern int tcp_listen_start(struct sock *sk);
891 extern void tcp_parse_options(struct sk_buff *skb,
896 * TCP v4 functions exported for the inet6 API
899 extern int tcp_v4_rebuild_header(struct sock *sk);
901 extern int tcp_v4_build_header(struct sock *sk,
902 struct sk_buff *skb);
904 extern void tcp_v4_send_check(struct sock *sk,
905 struct tcphdr *th, int len,
906 struct sk_buff *skb);
908 extern int tcp_v4_conn_request(struct sock *sk,
909 struct sk_buff *skb);
911 extern struct sock * tcp_create_openreq_child(struct sock *sk,
912 struct open_request *req,
913 struct sk_buff *skb);
915 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
917 struct open_request *req,
918 struct dst_entry *dst);
920 extern int tcp_v4_do_rcv(struct sock *sk,
921 struct sk_buff *skb);
923 extern int tcp_v4_connect(struct sock *sk,
924 struct sockaddr *uaddr,
927 extern int tcp_connect(struct sock *sk);
929 extern struct sk_buff * tcp_make_synack(struct sock *sk,
930 struct dst_entry *dst,
931 struct open_request *req);
933 extern int tcp_disconnect(struct sock *sk, int flags);
935 extern void tcp_unhash(struct sock *sk);
937 extern int tcp_v4_hash_connecting(struct sock *sk);
940 /* From syncookies.c */
941 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
942 struct ip_options *opt);
943 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
948 extern int tcp_write_xmit(struct sock *, int nonagle);
949 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
950 extern void tcp_xmit_retransmit_queue(struct sock *);
951 extern void tcp_simple_retransmit(struct sock *);
952 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
954 extern void tcp_send_probe0(struct sock *);
955 extern void tcp_send_partial(struct sock *);
956 extern int tcp_write_wakeup(struct sock *);
957 extern void tcp_send_fin(struct sock *sk);
958 extern void tcp_send_active_reset(struct sock *sk, int priority);
959 extern int tcp_send_synack(struct sock *);
960 extern void tcp_push_one(struct sock *, unsigned mss_now);
961 extern void tcp_send_ack(struct sock *sk);
962 extern void tcp_send_delayed_ack(struct sock *sk);
965 extern void tcp_init_xmit_timers(struct sock *);
966 extern void tcp_clear_xmit_timers(struct sock *);
968 extern void tcp_delete_keepalive_timer (struct sock *);
969 extern void tcp_reset_keepalive_timer (struct sock *, unsigned long);
970 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
971 extern unsigned int tcp_current_mss(struct sock *sk, int large);
973 extern const char timer_bug_msg[];
976 extern void tcp_get_info(struct sock *, struct tcp_info *);
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 sk_stop_timer(sk, &tp->retransmit_timer);
1001 #ifdef TCP_CLEAR_TIMERS
1002 sk_stop_timer(sk, &tp->delack_timer);
1006 printk(timer_bug_msg);
1013 * Reset the retransmission timer
1015 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
1017 struct tcp_opt *tp = tcp_sk(sk);
1019 if (when > TCP_RTO_MAX) {
1021 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
1027 case TCP_TIME_RETRANS:
1028 case TCP_TIME_PROBE0:
1030 tp->timeout = jiffies+when;
1031 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
1035 tp->ack.pending |= TCP_ACK_TIMER;
1036 tp->ack.timeout = jiffies+when;
1037 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
1041 printk(timer_bug_msg);
1045 /* Initialize RCV_MSS value.
1046 * RCV_MSS is an our guess about MSS used by the peer.
1047 * We haven't any direct information about the MSS.
1048 * It's better to underestimate the RCV_MSS rather than overestimate.
1049 * Overestimations make us ACKing less frequently than needed.
1050 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1053 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1055 struct tcp_opt *tp = tcp_sk(sk);
1056 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1058 hint = min(hint, tp->rcv_wnd/2);
1059 hint = min(hint, TCP_MIN_RCVMSS);
1060 hint = max(hint, TCP_MIN_MSS);
1062 tp->ack.rcv_mss = hint;
1065 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1067 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1068 ntohl(TCP_FLAG_ACK) |
1072 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1074 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1077 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1079 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1081 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1083 tcp_fast_path_on(tp);
1086 /* Compute the actual receive window we are currently advertising.
1087 * Rcv_nxt can be after the window if our peer push more data
1088 * than the offered window.
1090 static __inline__ u32 tcp_receive_window(struct tcp_opt *tp)
1092 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1099 /* Choose a new window, without checks for shrinking, and without
1100 * scaling applied to the result. The caller does these things
1101 * if necessary. This is a "raw" window selection.
1103 extern u32 __tcp_select_window(struct sock *sk);
1105 /* TCP timestamps are only 32-bits, this causes a slight
1106 * complication on 64-bit systems since we store a snapshot
1107 * of jiffies in the buffer control blocks below. We decidely
1108 * only use of the low 32-bits of jiffies and hide the ugly
1109 * casts with the following macro.
1111 #define tcp_time_stamp ((__u32)(jiffies))
1113 /* This is what the send packet queueing engine uses to pass
1114 * TCP per-packet control information to the transmission
1115 * code. We also store the host-order sequence numbers in
1116 * here too. This is 36 bytes on 32-bit architectures,
1117 * 40 bytes on 64-bit machines, if this grows please adjust
1118 * skbuff.h:skbuff->cb[xxx] size appropriately.
1122 struct inet_skb_parm h4;
1123 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1124 struct inet6_skb_parm h6;
1126 } header; /* For incoming frames */
1127 __u32 seq; /* Starting sequence number */
1128 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1129 __u32 when; /* used to compute rtt's */
1130 __u8 flags; /* TCP header flags. */
1132 /* NOTE: These must match up to the flags byte in a
1135 #define TCPCB_FLAG_FIN 0x01
1136 #define TCPCB_FLAG_SYN 0x02
1137 #define TCPCB_FLAG_RST 0x04
1138 #define TCPCB_FLAG_PSH 0x08
1139 #define TCPCB_FLAG_ACK 0x10
1140 #define TCPCB_FLAG_URG 0x20
1141 #define TCPCB_FLAG_ECE 0x40
1142 #define TCPCB_FLAG_CWR 0x80
1144 __u8 sacked; /* State flags for SACK/FACK. */
1145 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1146 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1147 #define TCPCB_LOST 0x04 /* SKB is lost */
1148 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1150 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1151 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1153 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1155 #define TCPCB_AT_TAIL (TCPCB_URG)
1157 __u16 urg_ptr; /* Valid w/URG flags is set. */
1158 __u32 ack_seq; /* Sequence number ACK'd */
1161 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1163 #include <net/tcp_ecn.h>
1165 /* Due to TSO, an SKB can be composed of multiple actual
1166 * packets. To keep these tracked properly, we use this.
1168 static inline int tcp_skb_pcount(struct sk_buff *skb)
1170 return skb_shinfo(skb)->tso_segs;
1173 /* This is valid iff tcp_skb_pcount() > 1. */
1174 static inline int tcp_skb_mss(struct sk_buff *skb)
1176 return skb_shinfo(skb)->tso_size;
1179 static inline void tcp_inc_pcount(tcp_pcount_t *count, struct sk_buff *skb)
1181 count->val += tcp_skb_pcount(skb);
1184 static inline void tcp_inc_pcount_explicit(tcp_pcount_t *count, int amt)
1189 static inline void tcp_dec_pcount_explicit(tcp_pcount_t *count, int amt)
1194 static inline void tcp_dec_pcount(tcp_pcount_t *count, struct sk_buff *skb)
1196 count->val -= tcp_skb_pcount(skb);
1199 static inline void tcp_dec_pcount_approx(tcp_pcount_t *count,
1200 struct sk_buff *skb)
1203 count->val -= tcp_skb_pcount(skb);
1204 if ((int)count->val < 0)
1209 static inline __u32 tcp_get_pcount(tcp_pcount_t *count)
1214 static inline void tcp_set_pcount(tcp_pcount_t *count, __u32 val)
1219 static inline void tcp_packets_out_inc(struct sock *sk, struct tcp_opt *tp,
1220 struct sk_buff *skb)
1222 int orig = tcp_get_pcount(&tp->packets_out);
1224 tcp_inc_pcount(&tp->packets_out, skb);
1226 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
1229 static inline void tcp_packets_out_dec(struct tcp_opt *tp, struct sk_buff *skb)
1231 tcp_dec_pcount(&tp->packets_out, skb);
1234 /* This determines how many packets are "in the network" to the best
1235 * of our knowledge. In many cases it is conservative, but where
1236 * detailed information is available from the receiver (via SACK
1237 * blocks etc.) we can make more aggressive calculations.
1239 * Use this for decisions involving congestion control, use just
1240 * tp->packets_out to determine if the send queue is empty or not.
1242 * Read this equation as:
1244 * "Packets sent once on transmission queue" MINUS
1245 * "Packets left network, but not honestly ACKed yet" PLUS
1246 * "Packets fast retransmitted"
1248 static __inline__ unsigned int tcp_packets_in_flight(struct tcp_opt *tp)
1250 return (tcp_get_pcount(&tp->packets_out) -
1251 tcp_get_pcount(&tp->left_out) +
1252 tcp_get_pcount(&tp->retrans_out));
1256 * Which congestion algorithim is in use on the connection.
1258 #define tcp_is_vegas(__tp) ((__tp)->adv_cong == TCP_VEGAS)
1259 #define tcp_is_westwood(__tp) ((__tp)->adv_cong == TCP_WESTWOOD)
1260 #define tcp_is_bic(__tp) ((__tp)->adv_cong == TCP_BIC)
1262 /* Recalculate snd_ssthresh, we want to set it to:
1265 * one half the current congestion window, but no
1266 * less than two segments
1269 * behave like Reno until low_window is reached,
1270 * then increase congestion window slowly
1272 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1274 if (tcp_is_bic(tp)) {
1275 if (sysctl_tcp_bic_fast_convergence &&
1276 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1277 tp->bictcp.last_max_cwnd
1278 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1279 / (BICTCP_1_OVER_BETA/2);
1281 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1283 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1284 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1288 return max(tp->snd_cwnd >> 1U, 2U);
1291 /* Stop taking Vegas samples for now. */
1292 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1294 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1296 /* There are several situations when we must "re-start" Vegas:
1298 * o when a connection is established
1300 * o after fast recovery
1301 * o when we send a packet and there is no outstanding
1302 * unacknowledged data (restarting an idle connection)
1304 * In these circumstances we cannot do a Vegas calculation at the
1305 * end of the first RTT, because any calculation we do is using
1306 * stale info -- both the saved cwnd and congestion feedback are
1309 * Instead we must wait until the completion of an RTT during
1310 * which we actually receive ACKs.
1313 /* Begin taking Vegas samples next time we send something. */
1314 tp->vegas.doing_vegas_now = 1;
1316 /* Set the beginning of the next send window. */
1317 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1319 tp->vegas.cntRTT = 0;
1320 tp->vegas.minRTT = 0x7fffffff;
1323 /* Should we be taking Vegas samples right now? */
1324 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1326 extern void tcp_ca_init(struct tcp_opt *tp);
1328 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1330 if (tcp_is_vegas(tp)) {
1331 if (ca_state == TCP_CA_Open)
1332 tcp_vegas_enable(tp);
1334 tcp_vegas_disable(tp);
1336 tp->ca_state = ca_state;
1339 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1340 * The exception is rate halving phase, when cwnd is decreasing towards
1343 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1345 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1346 return tp->snd_ssthresh;
1348 return max(tp->snd_ssthresh,
1349 ((tp->snd_cwnd >> 1) +
1350 (tp->snd_cwnd >> 2)));
1353 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1356 (tcp_get_pcount(&tp->sacked_out) >=
1357 tcp_get_pcount(&tp->packets_out) - tcp_get_pcount(&tp->lost_out)))
1358 tcp_set_pcount(&tp->sacked_out,
1359 (tcp_get_pcount(&tp->packets_out) -
1360 tcp_get_pcount(&tp->lost_out)));
1361 tcp_set_pcount(&tp->left_out,
1362 (tcp_get_pcount(&tp->sacked_out) +
1363 tcp_get_pcount(&tp->lost_out)));
1366 extern void tcp_cwnd_application_limited(struct sock *sk);
1368 /* Congestion window validation. (RFC2861) */
1370 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1372 __u32 packets_out = tcp_get_pcount(&tp->packets_out);
1374 if (packets_out >= tp->snd_cwnd) {
1375 /* Network is feed fully. */
1376 tp->snd_cwnd_used = 0;
1377 tp->snd_cwnd_stamp = tcp_time_stamp;
1379 /* Network starves. */
1380 if (tcp_get_pcount(&tp->packets_out) > tp->snd_cwnd_used)
1381 tp->snd_cwnd_used = tcp_get_pcount(&tp->packets_out);
1383 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1384 tcp_cwnd_application_limited(sk);
1388 /* Set slow start threshould and cwnd not falling to slow start */
1389 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1391 tp->undo_marker = 0;
1392 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1393 tp->snd_cwnd = min(tp->snd_cwnd,
1394 tcp_packets_in_flight(tp) + 1U);
1395 tp->snd_cwnd_cnt = 0;
1396 tp->high_seq = tp->snd_nxt;
1397 tp->snd_cwnd_stamp = tcp_time_stamp;
1398 TCP_ECN_queue_cwr(tp);
1401 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1403 tp->prior_ssthresh = 0;
1404 if (tp->ca_state < TCP_CA_CWR) {
1405 __tcp_enter_cwr(tp);
1406 tcp_set_ca_state(tp, TCP_CA_CWR);
1410 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1412 /* Slow start with delack produces 3 packets of burst, so that
1413 * it is safe "de facto".
1415 static __inline__ __u32 tcp_max_burst(struct tcp_opt *tp)
1420 static __inline__ int tcp_minshall_check(struct tcp_opt *tp)
1422 return after(tp->snd_sml,tp->snd_una) &&
1423 !after(tp->snd_sml, tp->snd_nxt);
1426 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss, struct sk_buff *skb)
1429 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1432 /* Return 0, if packet can be sent now without violation Nagle's rules:
1433 1. It is full sized.
1434 2. Or it contains FIN.
1435 3. Or TCP_NODELAY was set.
1436 4. Or TCP_CORK is not set, and all sent packets are ACKed.
1437 With Minshall's modification: all sent small packets are ACKed.
1440 static __inline__ int
1441 tcp_nagle_check(struct tcp_opt *tp, struct sk_buff *skb, unsigned mss_now, int nonagle)
1443 return (skb->len < mss_now &&
1444 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1445 ((nonagle&TCP_NAGLE_CORK) ||
1447 tcp_get_pcount(&tp->packets_out) &&
1448 tcp_minshall_check(tp))));
1451 extern void tcp_set_skb_tso_segs(struct sk_buff *, unsigned int);
1453 /* This checks if the data bearing packet SKB (usually sk->sk_send_head)
1454 * should be put on the wire right now.
1456 static __inline__ int tcp_snd_test(struct tcp_opt *tp, struct sk_buff *skb,
1457 unsigned cur_mss, int nonagle)
1459 int pkts = tcp_skb_pcount(skb);
1462 tcp_set_skb_tso_segs(skb, tp->mss_cache_std);
1463 pkts = tcp_skb_pcount(skb);
1466 /* RFC 1122 - section 4.2.3.4
1470 * a) The right edge of this frame exceeds the window
1471 * b) There are packets in flight and we have a small segment
1472 * [SWS avoidance and Nagle algorithm]
1473 * (part of SWS is done on packetization)
1474 * Minshall version sounds: there are no _small_
1475 * segments in flight. (tcp_nagle_check)
1476 * c) We have too many packets 'in flight'
1478 * Don't use the nagle rule for urgent data (or
1479 * for the final FIN -DaveM).
1481 * Also, Nagle rule does not apply to frames, which
1482 * sit in the middle of queue (they have no chances
1483 * to get new data) and if room at tail of skb is
1484 * not enough to save something seriously (<32 for now).
1487 /* Don't be strict about the congestion window for the
1488 * final FIN frame. -DaveM
1490 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1491 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1492 (((tcp_packets_in_flight(tp) + (pkts-1)) < tp->snd_cwnd) ||
1493 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1494 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1497 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1499 if (!tcp_get_pcount(&tp->packets_out) && !tp->pending)
1500 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1503 static __inline__ int tcp_skb_is_last(struct sock *sk, struct sk_buff *skb)
1505 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1508 /* Push out any pending frames which were held back due to
1509 * TCP_CORK or attempt at coalescing tiny packets.
1510 * The socket must be locked by the caller.
1512 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1517 struct sk_buff *skb = sk->sk_send_head;
1520 if (!tcp_skb_is_last(sk, skb))
1521 nonagle = TCP_NAGLE_PUSH;
1522 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1523 tcp_write_xmit(sk, nonagle))
1524 tcp_check_probe_timer(sk, tp);
1526 tcp_cwnd_validate(sk, tp);
1529 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1532 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1535 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1537 struct sk_buff *skb = sk->sk_send_head;
1540 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1541 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1544 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1549 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1554 extern void tcp_destroy_sock(struct sock *sk);
1558 * Calculate(/check) TCP checksum
1560 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1561 unsigned long saddr, unsigned long daddr,
1564 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1567 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1569 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1572 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1574 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1575 __tcp_checksum_complete(skb);
1578 /* Prequeue for VJ style copy to user, combined with checksumming. */
1580 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1582 tp->ucopy.task = NULL;
1584 tp->ucopy.memory = 0;
1585 skb_queue_head_init(&tp->ucopy.prequeue);
1588 /* Packet is added to VJ-style prequeue for processing in process
1589 * context, if a reader task is waiting. Apparently, this exciting
1590 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1591 * failed somewhere. Latency? Burstiness? Well, at least now we will
1592 * see, why it failed. 8)8) --ANK
1594 * NOTE: is this not too big to inline?
1596 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1598 struct tcp_opt *tp = tcp_sk(sk);
1600 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1601 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1602 tp->ucopy.memory += skb->truesize;
1603 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1604 struct sk_buff *skb1;
1606 BUG_ON(sock_owned_by_user(sk));
1608 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1609 sk->sk_backlog_rcv(sk, skb1);
1610 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1613 tp->ucopy.memory = 0;
1614 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1615 wake_up_interruptible(sk->sk_sleep);
1616 if (!tcp_ack_scheduled(tp))
1617 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1628 static char *statename[]={
1629 "Unused","Established","Syn Sent","Syn Recv",
1630 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1631 "Close Wait","Last ACK","Listen","Closing"
1635 static __inline__ void tcp_set_state(struct sock *sk, int state)
1637 int oldstate = sk->sk_state;
1640 case TCP_ESTABLISHED:
1641 if (oldstate != TCP_ESTABLISHED)
1642 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1646 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1647 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1649 sk->sk_prot->unhash(sk);
1650 if (tcp_sk(sk)->bind_hash &&
1651 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1655 if (oldstate==TCP_ESTABLISHED)
1656 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1659 /* Change state AFTER socket is unhashed to avoid closed
1660 * socket sitting in hash tables.
1662 sk->sk_state = state;
1665 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1669 static __inline__ void tcp_done(struct sock *sk)
1671 tcp_set_state(sk, TCP_CLOSE);
1672 tcp_clear_xmit_timers(sk);
1674 sk->sk_shutdown = SHUTDOWN_MASK;
1676 if (!sock_flag(sk, SOCK_DEAD))
1677 sk->sk_state_change(sk);
1679 tcp_destroy_sock(sk);
1682 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1689 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1691 if (tp->tstamp_ok) {
1692 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1693 (TCPOPT_NOP << 16) |
1694 (TCPOPT_TIMESTAMP << 8) |
1696 *ptr++ = htonl(tstamp);
1697 *ptr++ = htonl(tp->ts_recent);
1699 if (tp->eff_sacks) {
1700 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1703 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1704 (TCPOPT_NOP << 16) |
1705 (TCPOPT_SACK << 8) |
1706 (TCPOLEN_SACK_BASE +
1707 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1708 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1709 *ptr++ = htonl(sp[this_sack].start_seq);
1710 *ptr++ = htonl(sp[this_sack].end_seq);
1719 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1720 * If this is every changed make sure to change the definition of
1721 * MAX_SYN_SIZE to match the new maximum number of options that you
1724 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1725 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1727 /* We always get an MSS option.
1728 * The option bytes which will be seen in normal data
1729 * packets should timestamps be used, must be in the MSS
1730 * advertised. But we subtract them from tp->mss_cache so
1731 * that calculations in tcp_sendmsg are simpler etc.
1732 * So account for this fact here if necessary. If we
1733 * don't do this correctly, as a receiver we won't
1734 * recognize data packets as being full sized when we
1735 * should, and thus we won't abide by the delayed ACK
1737 * SACKs don't matter, we never delay an ACK when we
1738 * have any of those going out.
1740 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1743 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1744 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1746 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1747 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1748 *ptr++ = htonl(tstamp); /* TSVAL */
1749 *ptr++ = htonl(ts_recent); /* TSECR */
1751 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1752 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1754 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1757 /* Determine a window scaling and initial window to offer. */
1758 extern void tcp_select_initial_window(int __space, __u32 mss,
1759 __u32 *rcv_wnd, __u32 *window_clamp,
1760 int wscale_ok, __u8 *rcv_wscale);
1762 static inline int tcp_win_from_space(int space)
1764 return sysctl_tcp_adv_win_scale<=0 ?
1765 (space>>(-sysctl_tcp_adv_win_scale)) :
1766 space - (space>>sysctl_tcp_adv_win_scale);
1769 /* Note: caller must be prepared to deal with negative returns */
1770 static inline int tcp_space(const struct sock *sk)
1772 return tcp_win_from_space(sk->sk_rcvbuf -
1773 atomic_read(&sk->sk_rmem_alloc));
1776 static inline int tcp_full_space(const struct sock *sk)
1778 return tcp_win_from_space(sk->sk_rcvbuf);
1781 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1784 struct tcp_opt *tp = tcp_sk(sk);
1787 sk_acceptq_added(sk);
1789 if (!tp->accept_queue_tail) {
1790 tp->accept_queue = req;
1792 tp->accept_queue_tail->dl_next = req;
1794 tp->accept_queue_tail = req;
1795 req->dl_next = NULL;
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];
1809 tcp_synq_removed(struct sock *sk, struct open_request *req)
1811 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1813 if (--lopt->qlen == 0)
1814 tcp_delete_keepalive_timer(sk);
1815 if (req->retrans == 0)
1819 static inline void tcp_synq_added(struct sock *sk)
1821 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1823 if (lopt->qlen++ == 0)
1824 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1828 static inline int tcp_synq_len(struct sock *sk)
1830 return tcp_sk(sk)->listen_opt->qlen;
1833 static inline int tcp_synq_young(struct sock *sk)
1835 return tcp_sk(sk)->listen_opt->qlen_young;
1838 static inline int tcp_synq_is_full(struct sock *sk)
1840 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1843 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1844 struct open_request **prev)
1846 write_lock(&tp->syn_wait_lock);
1847 *prev = req->dl_next;
1848 write_unlock(&tp->syn_wait_lock);
1851 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1852 struct open_request **prev)
1854 tcp_synq_unlink(tcp_sk(sk), req, prev);
1855 tcp_synq_removed(sk, req);
1856 tcp_openreq_free(req);
1859 static __inline__ void tcp_openreq_init(struct open_request *req,
1861 struct sk_buff *skb)
1863 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1864 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1865 req->mss = tp->mss_clamp;
1866 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1867 req->tstamp_ok = tp->tstamp_ok;
1868 req->sack_ok = tp->sack_ok;
1869 req->snd_wscale = tp->snd_wscale;
1870 req->wscale_ok = tp->wscale_ok;
1873 req->rmt_port = skb->h.th->source;
1876 extern void tcp_enter_memory_pressure(void);
1878 extern void tcp_listen_wlock(void);
1880 /* - We may sleep inside this lock.
1881 * - If sleeping is not required (or called from BH),
1882 * use plain read_(un)lock(&tcp_lhash_lock).
1885 static inline void tcp_listen_lock(void)
1887 /* read_lock synchronizes to candidates to writers */
1888 read_lock(&tcp_lhash_lock);
1889 atomic_inc(&tcp_lhash_users);
1890 read_unlock(&tcp_lhash_lock);
1893 static inline void tcp_listen_unlock(void)
1895 if (atomic_dec_and_test(&tcp_lhash_users))
1896 wake_up(&tcp_lhash_wait);
1899 static inline int keepalive_intvl_when(struct tcp_opt *tp)
1901 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1904 static inline int keepalive_time_when(struct tcp_opt *tp)
1906 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1909 static inline int tcp_fin_time(struct tcp_opt *tp)
1911 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1913 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1914 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1919 static inline int tcp_paws_check(struct tcp_opt *tp, int rst)
1921 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
1923 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
1926 /* RST segments are not recommended to carry timestamp,
1927 and, if they do, it is recommended to ignore PAWS because
1928 "their cleanup function should take precedence over timestamps."
1929 Certainly, it is mistake. It is necessary to understand the reasons
1930 of this constraint to relax it: if peer reboots, clock may go
1931 out-of-sync and half-open connections will not be reset.
1932 Actually, the problem would be not existing if all
1933 the implementations followed draft about maintaining clock
1934 via reboots. Linux-2.2 DOES NOT!
1936 However, we can relax time bounds for RST segments to MSL.
1938 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
1943 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
1945 sk->sk_route_caps = dst->dev->features;
1946 if (sk->sk_route_caps & NETIF_F_TSO) {
1947 if (sk->sk_no_largesend || dst->header_len)
1948 sk->sk_route_caps &= ~NETIF_F_TSO;
1952 #define TCP_CHECK_TIMER(sk) do { } while (0)
1954 static inline int tcp_use_frto(const struct sock *sk)
1956 const struct tcp_opt *tp = tcp_sk(sk);
1958 /* F-RTO must be activated in sysctl and there must be some
1959 * unsent new data, and the advertised window should allow
1962 return (sysctl_tcp_frto && sk->sk_send_head &&
1963 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1964 tp->snd_una + tp->snd_wnd));
1967 static inline void tcp_mib_init(void)
1970 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1971 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1972 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1973 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1977 enum tcp_seq_states {
1978 TCP_SEQ_STATE_LISTENING,
1979 TCP_SEQ_STATE_OPENREQ,
1980 TCP_SEQ_STATE_ESTABLISHED,
1981 TCP_SEQ_STATE_TIME_WAIT,
1984 struct tcp_seq_afinfo {
1985 struct module *owner;
1988 int (*seq_show) (struct seq_file *m, void *v);
1989 struct file_operations *seq_fops;
1992 struct tcp_iter_state {
1994 enum tcp_seq_states state;
1995 struct sock *syn_wait_sk;
1996 int bucket, sbucket, num, uid;
1997 struct seq_operations seq_ops;
2000 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2001 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2003 /* TCP Westwood functions and constants */
2005 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2006 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2008 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2010 if (tcp_is_westwood(tp))
2011 tp->westwood.rtt = rtt_seq;
2014 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2015 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2017 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2019 if (tcp_is_westwood(tcp_sk(sk)))
2020 __tcp_westwood_fast_bw(sk, skb);
2023 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2025 if (tcp_is_westwood(tcp_sk(sk)))
2026 __tcp_westwood_slow_bw(sk, skb);
2029 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2031 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2032 (__u32) (tp->mss_cache_std),
2036 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2038 return tcp_is_westwood(tp) ? __tcp_westwood_bw_rttmin(tp) : 0;
2041 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2045 if (tcp_is_westwood(tp)) {
2046 ssthresh = __tcp_westwood_bw_rttmin(tp);
2048 tp->snd_ssthresh = ssthresh;
2051 return (ssthresh != 0);
2054 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2058 if (tcp_is_westwood(tp)) {
2059 cwnd = __tcp_westwood_bw_rttmin(tp);
2061 tp->snd_cwnd = cwnd;