2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
22 #define FASTRETRANS_DEBUG 1
24 /* Cancel timers, when they are not required. */
25 #undef TCP_CLEAR_TIMERS
27 #include <linux/config.h>
28 #include <linux/list.h>
29 #include <linux/tcp.h>
30 #include <linux/slab.h>
31 #include <linux/cache.h>
32 #include <linux/percpu.h>
33 #include <net/checksum.h>
37 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
38 #include <linux/ipv6.h>
40 #include <linux/seq_file.h>
42 /* This is for all connections with a full identity, no wildcards.
43 * New scheme, half the table is for TIME_WAIT, the other half is
44 * for the rest. I'll experiment with dynamic table growth later.
46 struct tcp_ehash_bucket {
48 struct hlist_head chain;
49 } __attribute__((__aligned__(8)));
51 /* This is for listening sockets, thus all sockets which possess wildcards. */
52 #define TCP_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
54 /* There are a few simple rules, which allow for local port reuse by
55 * an application. In essence:
57 * 1) Sockets bound to different interfaces may share a local port.
58 * Failing that, goto test 2.
59 * 2) If all sockets have sk->sk_reuse set, and none of them are in
60 * TCP_LISTEN state, the port may be shared.
61 * Failing that, goto test 3.
62 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
63 * address, and none of them are the same, the port may be
65 * Failing this, the port cannot be shared.
67 * The interesting point, is test #2. This is what an FTP server does
68 * all day. To optimize this case we use a specific flag bit defined
69 * below. As we add sockets to a bind bucket list, we perform a
70 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
71 * As long as all sockets added to a bind bucket pass this test,
72 * the flag bit will be set.
73 * The resulting situation is that tcp_v[46]_verify_bind() can just check
74 * for this flag bit, if it is set and the socket trying to bind has
75 * sk->sk_reuse set, we don't even have to walk the owners list at all,
76 * we return that it is ok to bind this socket to the requested local port.
78 * Sounds like a lot of work, but it is worth it. In a more naive
79 * implementation (ie. current FreeBSD etc.) the entire list of ports
80 * must be walked for each data port opened by an ftp server. Needless
81 * to say, this does not scale at all. With a couple thousand FTP
82 * users logged onto your box, isn't it nice to know that new data
83 * ports are created in O(1) time? I thought so. ;-) -DaveM
85 struct tcp_bind_bucket {
87 signed short fastreuse;
88 struct hlist_node node;
89 struct hlist_head owners;
92 #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node)
94 struct tcp_bind_hashbucket {
96 struct hlist_head chain;
99 static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head)
101 return hlist_entry(head->chain.first, struct tcp_bind_bucket, node);
104 static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head)
106 return hlist_empty(&head->chain) ? NULL : __tb_head(head);
109 extern struct tcp_hashinfo {
110 /* This is for sockets with full identity only. Sockets here will
111 * always be without wildcards and will have the following invariant:
113 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
115 * First half of the table is for sockets not in TIME_WAIT, second half
116 * is for TIME_WAIT sockets only.
118 struct tcp_ehash_bucket *__tcp_ehash;
120 /* Ok, let's try this, I give up, we do need a local binding
121 * TCP hash as well as the others for fast bind/connect.
123 struct tcp_bind_hashbucket *__tcp_bhash;
125 int __tcp_bhash_size;
126 int __tcp_ehash_size;
128 /* All sockets in TCP_LISTEN state will be in here. This is the only
129 * table where wildcard'd TCP sockets can exist. Hash function here
130 * is just local port number.
132 struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE];
134 /* All the above members are written once at bootup and
135 * never written again _or_ are predominantly read-access.
137 * Now align to a new cache line as all the following members
140 rwlock_t __tcp_lhash_lock ____cacheline_aligned;
141 atomic_t __tcp_lhash_users;
142 wait_queue_head_t __tcp_lhash_wait;
143 spinlock_t __tcp_portalloc_lock;
146 #define tcp_ehash (tcp_hashinfo.__tcp_ehash)
147 #define tcp_bhash (tcp_hashinfo.__tcp_bhash)
148 #define tcp_ehash_size (tcp_hashinfo.__tcp_ehash_size)
149 #define tcp_bhash_size (tcp_hashinfo.__tcp_bhash_size)
150 #define tcp_listening_hash (tcp_hashinfo.__tcp_listening_hash)
151 #define tcp_lhash_lock (tcp_hashinfo.__tcp_lhash_lock)
152 #define tcp_lhash_users (tcp_hashinfo.__tcp_lhash_users)
153 #define tcp_lhash_wait (tcp_hashinfo.__tcp_lhash_wait)
154 #define tcp_portalloc_lock (tcp_hashinfo.__tcp_portalloc_lock)
156 extern kmem_cache_t *tcp_bucket_cachep;
157 extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head,
158 unsigned short snum);
159 extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb);
160 extern void tcp_bucket_unlock(struct sock *sk);
161 extern int tcp_port_rover;
163 /* These are AF independent. */
164 static __inline__ int tcp_bhashfn(__u16 lport)
166 return (lport & (tcp_bhash_size - 1));
169 extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb,
170 unsigned short snum);
172 #if (BITS_PER_LONG == 64)
173 #define TCP_ADDRCMP_ALIGN_BYTES 8
175 #define TCP_ADDRCMP_ALIGN_BYTES 4
178 /* This is a TIME_WAIT bucket. It works around the memory consumption
179 * problems of sockets in such a state on heavily loaded servers, but
180 * without violating the protocol specification.
182 struct tcp_tw_bucket {
184 * Now struct sock also uses sock_common, so please just
185 * don't add nothing before this first member (__tw_common) --acme
187 struct sock_common __tw_common;
188 #define tw_family __tw_common.skc_family
189 #define tw_state __tw_common.skc_state
190 #define tw_reuse __tw_common.skc_reuse
191 #define tw_bound_dev_if __tw_common.skc_bound_dev_if
192 #define tw_node __tw_common.skc_node
193 #define tw_bind_node __tw_common.skc_bind_node
194 #define tw_refcnt __tw_common.skc_refcnt
195 volatile unsigned char tw_substate;
196 unsigned char tw_rcv_wscale;
198 /* Socket demultiplex comparisons on incoming packets. */
199 /* these five are in inet_opt */
201 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
205 /* And these are ours. */
212 long tw_ts_recent_stamp;
213 unsigned long tw_ttd;
214 struct tcp_bind_bucket *tw_tb;
215 struct hlist_node tw_death_node;
216 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
217 struct in6_addr tw_v6_daddr;
218 struct in6_addr tw_v6_rcv_saddr;
223 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
224 struct hlist_head *list)
226 hlist_add_head(&tw->tw_node, list);
229 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
230 struct hlist_head *list)
232 hlist_add_head(&tw->tw_bind_node, list);
235 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
237 return tw->tw_death_node.pprev != NULL;
240 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
242 tw->tw_death_node.pprev = NULL;
245 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
247 __hlist_del(&tw->tw_death_node);
248 tw_dead_node_init(tw);
251 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
253 if (tw_dead_hashed(tw)) {
254 __tw_del_dead_node(tw);
260 #define tw_for_each(tw, node, head) \
261 hlist_for_each_entry(tw, node, head, tw_node)
263 #define tw_for_each_inmate(tw, node, jail) \
264 hlist_for_each_entry(tw, node, jail, tw_death_node)
266 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
267 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
269 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
271 static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
273 return likely(sk->sk_state != TCP_TIME_WAIT) ?
274 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
277 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
278 static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
280 return likely(sk->sk_state != TCP_TIME_WAIT) ?
281 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
284 static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
286 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
289 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
291 static inline int tcp_v6_ipv6only(const struct sock *sk)
293 return likely(sk->sk_state != TCP_TIME_WAIT) ?
294 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
297 # define __tcp_v6_rcv_saddr(__sk) NULL
298 # define tcp_v6_rcv_saddr(__sk) NULL
299 # define tcptw_sk_ipv6only(__sk) 0
300 # define tcp_v6_ipv6only(__sk) 0
303 extern kmem_cache_t *tcp_timewait_cachep;
305 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
307 if (atomic_dec_and_test(&tw->tw_refcnt)) {
308 #ifdef INET_REFCNT_DEBUG
309 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
311 kmem_cache_free(tcp_timewait_cachep, tw);
315 extern atomic_t tcp_orphan_count;
316 extern int tcp_tw_count;
317 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
318 extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
319 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
322 /* Socket demux engine toys. */
324 #define TCP_COMBINED_PORTS(__sport, __dport) \
325 (((__u32)(__sport)<<16) | (__u32)(__dport))
326 #else /* __LITTLE_ENDIAN */
327 #define TCP_COMBINED_PORTS(__sport, __dport) \
328 (((__u32)(__dport)<<16) | (__u32)(__sport))
331 #if (BITS_PER_LONG == 64)
333 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
334 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr));
335 #else /* __LITTLE_ENDIAN */
336 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \
337 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr));
338 #endif /* __BIG_ENDIAN */
339 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
340 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
341 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
342 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
343 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
344 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \
345 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
346 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
347 #else /* 32-bit arch */
348 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr)
349 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
350 ((inet_sk(__sk)->daddr == (__saddr)) && \
351 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
352 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
353 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
354 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
355 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \
356 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \
357 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \
358 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
359 #endif /* 64-bit arch */
361 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \
362 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \
363 ((__sk)->sk_family == AF_INET6) && \
364 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \
365 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \
366 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
368 /* These can have wildcards, don't try too hard. */
369 static __inline__ int tcp_lhashfn(unsigned short num)
371 return num & (TCP_LHTABLE_SIZE - 1);
374 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk)
376 return tcp_lhashfn(inet_sk(sk)->num);
379 #define MAX_TCP_HEADER (128 + MAX_HEADER)
382 * Never offer a window over 32767 without using window scaling. Some
383 * poor stacks do signed 16bit maths!
385 #define MAX_TCP_WINDOW 32767U
387 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
388 #define TCP_MIN_MSS 88U
390 /* Minimal RCV_MSS. */
391 #define TCP_MIN_RCVMSS 536U
393 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
394 #define TCP_FASTRETRANS_THRESH 3
396 /* Maximal reordering. */
397 #define TCP_MAX_REORDERING 127
399 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
400 #define TCP_MAX_QUICKACKS 16U
402 /* urg_data states */
403 #define TCP_URG_VALID 0x0100
404 #define TCP_URG_NOTYET 0x0200
405 #define TCP_URG_READ 0x0400
407 #define TCP_RETR1 3 /*
408 * This is how many retries it does before it
409 * tries to figure out if the gateway is
410 * down. Minimal RFC value is 3; it corresponds
411 * to ~3sec-8min depending on RTO.
414 #define TCP_RETR2 15 /*
415 * This should take at least
416 * 90 minutes to time out.
417 * RFC1122 says that the limit is 100 sec.
418 * 15 is ~13-30min depending on RTO.
421 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
422 * connection: ~180sec is RFC minumum */
424 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
425 * connection: ~180sec is RFC minumum */
428 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
429 * socket. 7 is ~50sec-16min.
433 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
434 * state, about 60 seconds */
435 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
436 /* BSD style FIN_WAIT2 deadlock breaker.
437 * It used to be 3min, new value is 60sec,
438 * to combine FIN-WAIT-2 timeout with
442 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
444 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
445 #define TCP_ATO_MIN ((unsigned)(HZ/25))
447 #define TCP_DELACK_MIN 4U
448 #define TCP_ATO_MIN 4U
450 #define TCP_RTO_MAX ((unsigned)(120*HZ))
451 #define TCP_RTO_MIN ((unsigned)(HZ/5))
452 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
454 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
455 * for local resources.
458 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
459 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
460 #define TCP_KEEPALIVE_INTVL (75*HZ)
462 #define MAX_TCP_KEEPIDLE 32767
463 #define MAX_TCP_KEEPINTVL 32767
464 #define MAX_TCP_KEEPCNT 127
465 #define MAX_TCP_SYNCNT 127
467 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
468 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
470 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
471 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
472 * after this time. It should be equal
473 * (or greater than) TCP_TIMEWAIT_LEN
474 * to provide reliability equal to one
475 * provided by timewait state.
477 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
478 * timestamps. It must be less than
479 * minimal timewait lifetime.
482 #define TCP_TW_RECYCLE_SLOTS_LOG 5
483 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
485 /* If time > 4sec, it is "slow" path, no recycling is required,
486 so that we select tick to get range about 4 seconds.
489 #if HZ <= 16 || HZ > 4096
490 # error Unsupported: HZ <= 16 or HZ > 4096
492 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
494 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
496 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
498 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
500 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
502 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
504 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
506 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
509 #define BICTCP_1_OVER_BETA 8 /*
511 * multiplicative decrease factor
513 #define BICTCP_MAX_INCREMENT 32 /*
514 * Limit on the amount of
515 * increment allowed during
518 #define BICTCP_FUNC_OF_MIN_INCR 11 /*
519 * log(B/Smin)/log(B/(B-1))+1,
523 #define BICTCP_B 4 /*
525 * go to point (max+min)/N
532 #define TCPOPT_NOP 1 /* Padding */
533 #define TCPOPT_EOL 0 /* End of options */
534 #define TCPOPT_MSS 2 /* Segment size negotiating */
535 #define TCPOPT_WINDOW 3 /* Window scaling */
536 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
537 #define TCPOPT_SACK 5 /* SACK Block */
538 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
544 #define TCPOLEN_MSS 4
545 #define TCPOLEN_WINDOW 3
546 #define TCPOLEN_SACK_PERM 2
547 #define TCPOLEN_TIMESTAMP 10
549 /* But this is what stacks really send out. */
550 #define TCPOLEN_TSTAMP_ALIGNED 12
551 #define TCPOLEN_WSCALE_ALIGNED 4
552 #define TCPOLEN_SACKPERM_ALIGNED 4
553 #define TCPOLEN_SACK_BASE 2
554 #define TCPOLEN_SACK_BASE_ALIGNED 4
555 #define TCPOLEN_SACK_PERBLOCK 8
557 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
558 #define TCP_TIME_DACK 2 /* Delayed ack timer */
559 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
560 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
562 /* Flags in tp->nonagle */
563 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
564 #define TCP_NAGLE_CORK 2 /* Socket is corked */
565 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
567 /* sysctl variables for tcp */
568 extern int sysctl_max_syn_backlog;
569 extern int sysctl_tcp_timestamps;
570 extern int sysctl_tcp_window_scaling;
571 extern int sysctl_tcp_sack;
572 extern int sysctl_tcp_fin_timeout;
573 extern int sysctl_tcp_tw_recycle;
574 extern int sysctl_tcp_keepalive_time;
575 extern int sysctl_tcp_keepalive_probes;
576 extern int sysctl_tcp_keepalive_intvl;
577 extern int sysctl_tcp_syn_retries;
578 extern int sysctl_tcp_synack_retries;
579 extern int sysctl_tcp_retries1;
580 extern int sysctl_tcp_retries2;
581 extern int sysctl_tcp_orphan_retries;
582 extern int sysctl_tcp_syncookies;
583 extern int sysctl_tcp_retrans_collapse;
584 extern int sysctl_tcp_stdurg;
585 extern int sysctl_tcp_rfc1337;
586 extern int sysctl_tcp_abort_on_overflow;
587 extern int sysctl_tcp_max_orphans;
588 extern int sysctl_tcp_max_tw_buckets;
589 extern int sysctl_tcp_fack;
590 extern int sysctl_tcp_reordering;
591 extern int sysctl_tcp_ecn;
592 extern int sysctl_tcp_dsack;
593 extern int sysctl_tcp_mem[3];
594 extern int sysctl_tcp_wmem[3];
595 extern int sysctl_tcp_rmem[3];
596 extern int sysctl_tcp_app_win;
597 extern int sysctl_tcp_adv_win_scale;
598 extern int sysctl_tcp_tw_reuse;
599 extern int sysctl_tcp_frto;
600 extern int sysctl_tcp_low_latency;
601 extern int sysctl_tcp_westwood;
602 extern int sysctl_tcp_vegas_cong_avoid;
603 extern int sysctl_tcp_vegas_alpha;
604 extern int sysctl_tcp_vegas_beta;
605 extern int sysctl_tcp_vegas_gamma;
606 extern int sysctl_tcp_nometrics_save;
607 extern int sysctl_tcp_bic;
608 extern int sysctl_tcp_bic_fast_convergence;
609 extern int sysctl_tcp_bic_low_window;
610 extern int sysctl_tcp_moderate_rcvbuf;
611 extern int sysctl_tcp_tso_win_divisor;
613 extern atomic_t tcp_memory_allocated;
614 extern atomic_t tcp_sockets_allocated;
615 extern int tcp_memory_pressure;
619 struct or_calltable {
621 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*);
622 void (*send_ack) (struct sk_buff *skb, struct open_request *req);
623 void (*destructor) (struct open_request *req);
624 void (*send_reset) (struct sk_buff *skb);
627 struct tcp_v4_open_req {
630 struct ip_options *opt;
633 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
634 struct tcp_v6_open_req {
635 struct in6_addr loc_addr;
636 struct in6_addr rmt_addr;
637 struct sk_buff *pktopts;
642 /* this structure is too big */
643 struct open_request {
644 struct open_request *dl_next; /* Must be first member! */
651 __u16 snd_wscale : 4,
658 /* The following two fields can be easily recomputed I think -AK */
659 __u32 window_clamp; /* window clamp at creation time */
660 __u32 rcv_wnd; /* rcv_wnd offered first time */
662 unsigned long expires;
663 struct or_calltable *class;
666 struct tcp_v4_open_req v4_req;
667 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
668 struct tcp_v6_open_req v6_req;
673 /* SLAB cache for open requests. */
674 extern kmem_cache_t *tcp_openreq_cachep;
676 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
677 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req)
679 static inline void tcp_openreq_free(struct open_request *req)
681 req->class->destructor(req);
682 tcp_openreq_fastfree(req);
685 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
686 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
688 #define TCP_INET_FAMILY(fam) 1
692 * Pointers to address related TCP functions
693 * (i.e. things that depend on the address family)
697 int (*queue_xmit) (struct sk_buff *skb,
700 void (*send_check) (struct sock *sk,
703 struct sk_buff *skb);
705 int (*rebuild_header) (struct sock *sk);
707 int (*conn_request) (struct sock *sk,
708 struct sk_buff *skb);
710 struct sock * (*syn_recv_sock) (struct sock *sk,
712 struct open_request *req,
713 struct dst_entry *dst);
715 int (*remember_stamp) (struct sock *sk);
717 __u16 net_header_len;
719 int (*setsockopt) (struct sock *sk,
725 int (*getsockopt) (struct sock *sk,
732 void (*addr2sockaddr) (struct sock *sk,
739 * The next routines deal with comparing 32 bit unsigned ints
740 * and worry about wraparound (automatic with unsigned arithmetic).
743 static inline int before(__u32 seq1, __u32 seq2)
745 return (__s32)(seq1-seq2) < 0;
748 static inline int after(__u32 seq1, __u32 seq2)
750 return (__s32)(seq2-seq1) < 0;
754 /* is s2<=s1<=s3 ? */
755 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
757 return seq3 - seq2 >= seq1 - seq2;
761 extern struct proto tcp_prot;
763 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
764 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
765 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
766 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
767 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
768 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
769 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
771 extern void tcp_put_port(struct sock *sk);
772 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
774 extern void tcp_v4_err(struct sk_buff *skb, u32);
776 extern void tcp_shutdown (struct sock *sk, int how);
778 extern int tcp_v4_rcv(struct sk_buff *skb);
780 extern struct sock * tcp_v4_lookup_listener(u32 daddr, unsigned short hnum, int dif);
782 extern int tcp_v4_remember_stamp(struct sock *sk);
784 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
786 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
787 struct msghdr *msg, size_t size);
788 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
790 extern int tcp_ioctl(struct sock *sk,
794 extern int tcp_rcv_state_process(struct sock *sk,
799 extern int tcp_rcv_established(struct sock *sk,
804 extern void tcp_rcv_space_adjust(struct sock *sk);
813 static inline void tcp_schedule_ack(struct tcp_opt *tp)
815 tp->ack.pending |= TCP_ACK_SCHED;
818 static inline int tcp_ack_scheduled(struct tcp_opt *tp)
820 return tp->ack.pending&TCP_ACK_SCHED;
823 static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
825 if (tp->ack.quick && --tp->ack.quick == 0) {
826 /* Leaving quickack mode we deflate ATO. */
827 tp->ack.ato = TCP_ATO_MIN;
831 extern void tcp_enter_quickack_mode(struct tcp_opt *tp);
833 static __inline__ void tcp_delack_init(struct tcp_opt *tp)
835 memset(&tp->ack, 0, sizeof(tp->ack));
838 static inline void tcp_clear_options(struct tcp_opt *tp)
840 tp->tstamp_ok = tp->sack_ok = tp->wscale_ok = tp->snd_wscale = 0;
852 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
857 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
858 struct open_request *req,
859 struct open_request **prev);
860 extern int tcp_child_process(struct sock *parent,
862 struct sk_buff *skb);
863 extern void tcp_enter_frto(struct sock *sk);
864 extern void tcp_enter_loss(struct sock *sk, int how);
865 extern void tcp_clear_retrans(struct tcp_opt *tp);
866 extern void tcp_update_metrics(struct sock *sk);
868 extern void tcp_close(struct sock *sk,
870 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
871 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
873 extern int tcp_getsockopt(struct sock *sk, int level,
877 extern int tcp_setsockopt(struct sock *sk, int level,
878 int optname, char __user *optval,
880 extern void tcp_set_keepalive(struct sock *sk, int val);
881 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
883 size_t len, int nonblock,
884 int flags, int *addr_len);
886 extern int tcp_listen_start(struct sock *sk);
888 extern void tcp_parse_options(struct sk_buff *skb,
893 * TCP v4 functions exported for the inet6 API
896 extern int tcp_v4_rebuild_header(struct sock *sk);
898 extern int tcp_v4_build_header(struct sock *sk,
899 struct sk_buff *skb);
901 extern void tcp_v4_send_check(struct sock *sk,
902 struct tcphdr *th, int len,
903 struct sk_buff *skb);
905 extern int tcp_v4_conn_request(struct sock *sk,
906 struct sk_buff *skb);
908 extern struct sock * tcp_create_openreq_child(struct sock *sk,
909 struct open_request *req,
910 struct sk_buff *skb);
912 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
914 struct open_request *req,
915 struct dst_entry *dst);
917 extern int tcp_v4_do_rcv(struct sock *sk,
918 struct sk_buff *skb);
920 extern int tcp_v4_connect(struct sock *sk,
921 struct sockaddr *uaddr,
924 extern int tcp_connect(struct sock *sk);
926 extern struct sk_buff * tcp_make_synack(struct sock *sk,
927 struct dst_entry *dst,
928 struct open_request *req);
930 extern int tcp_disconnect(struct sock *sk, int flags);
932 extern void tcp_unhash(struct sock *sk);
934 extern int tcp_v4_hash_connecting(struct sock *sk);
937 /* From syncookies.c */
938 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
939 struct ip_options *opt);
940 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
945 extern int tcp_write_xmit(struct sock *, int nonagle);
946 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
947 extern void tcp_xmit_retransmit_queue(struct sock *);
948 extern void tcp_simple_retransmit(struct sock *);
949 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
951 extern void tcp_send_probe0(struct sock *);
952 extern void tcp_send_partial(struct sock *);
953 extern int tcp_write_wakeup(struct sock *);
954 extern void tcp_send_fin(struct sock *sk);
955 extern void tcp_send_active_reset(struct sock *sk, int priority);
956 extern int tcp_send_synack(struct sock *);
957 extern void tcp_push_one(struct sock *, unsigned mss_now);
958 extern void tcp_send_ack(struct sock *sk);
959 extern void tcp_send_delayed_ack(struct sock *sk);
960 extern void cleanup_rbuf(struct sock *sk, int copied);
963 extern void tcp_init_xmit_timers(struct sock *);
964 extern void tcp_clear_xmit_timers(struct sock *);
966 extern void tcp_delete_keepalive_timer(struct sock *);
967 extern void tcp_reset_keepalive_timer(struct sock *, unsigned long);
968 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
969 extern unsigned int tcp_current_mss(struct sock *sk, int large);
972 extern const char tcp_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);
1007 printk(tcp_timer_bug_msg);
1015 * Reset the retransmission timer
1017 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
1019 struct tcp_opt *tp = tcp_sk(sk);
1021 if (when > TCP_RTO_MAX) {
1023 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
1029 case TCP_TIME_RETRANS:
1030 case TCP_TIME_PROBE0:
1032 tp->timeout = jiffies+when;
1033 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
1037 tp->ack.pending |= TCP_ACK_TIMER;
1038 tp->ack.timeout = jiffies+when;
1039 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
1044 printk(tcp_timer_bug_msg);
1049 /* Initialize RCV_MSS value.
1050 * RCV_MSS is an our guess about MSS used by the peer.
1051 * We haven't any direct information about the MSS.
1052 * It's better to underestimate the RCV_MSS rather than overestimate.
1053 * Overestimations make us ACKing less frequently than needed.
1054 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
1057 static inline void tcp_initialize_rcv_mss(struct sock *sk)
1059 struct tcp_opt *tp = tcp_sk(sk);
1060 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
1062 hint = min(hint, tp->rcv_wnd/2);
1063 hint = min(hint, TCP_MIN_RCVMSS);
1064 hint = max(hint, TCP_MIN_MSS);
1066 tp->ack.rcv_mss = hint;
1069 static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
1071 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1072 ntohl(TCP_FLAG_ACK) |
1076 static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
1078 __tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
1081 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_opt *tp)
1083 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
1085 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1087 tcp_fast_path_on(tp);
1090 /* Compute the actual receive window we are currently advertising.
1091 * Rcv_nxt can be after the window if our peer push more data
1092 * than the offered window.
1094 static __inline__ u32 tcp_receive_window(const struct tcp_opt *tp)
1096 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1103 /* Choose a new window, without checks for shrinking, and without
1104 * scaling applied to the result. The caller does these things
1105 * if necessary. This is a "raw" window selection.
1107 extern u32 __tcp_select_window(struct sock *sk);
1109 /* TCP timestamps are only 32-bits, this causes a slight
1110 * complication on 64-bit systems since we store a snapshot
1111 * of jiffies in the buffer control blocks below. We decidely
1112 * only use of the low 32-bits of jiffies and hide the ugly
1113 * casts with the following macro.
1115 #define tcp_time_stamp ((__u32)(jiffies))
1117 /* This is what the send packet queueing engine uses to pass
1118 * TCP per-packet control information to the transmission
1119 * code. We also store the host-order sequence numbers in
1120 * here too. This is 36 bytes on 32-bit architectures,
1121 * 40 bytes on 64-bit machines, if this grows please adjust
1122 * skbuff.h:skbuff->cb[xxx] size appropriately.
1126 struct inet_skb_parm h4;
1127 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1128 struct inet6_skb_parm h6;
1130 } header; /* For incoming frames */
1131 __u32 seq; /* Starting sequence number */
1132 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1133 __u32 when; /* used to compute rtt's */
1134 __u8 flags; /* TCP header flags. */
1136 /* NOTE: These must match up to the flags byte in a
1139 #define TCPCB_FLAG_FIN 0x01
1140 #define TCPCB_FLAG_SYN 0x02
1141 #define TCPCB_FLAG_RST 0x04
1142 #define TCPCB_FLAG_PSH 0x08
1143 #define TCPCB_FLAG_ACK 0x10
1144 #define TCPCB_FLAG_URG 0x20
1145 #define TCPCB_FLAG_ECE 0x40
1146 #define TCPCB_FLAG_CWR 0x80
1148 __u8 sacked; /* State flags for SACK/FACK. */
1149 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1150 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1151 #define TCPCB_LOST 0x04 /* SKB is lost */
1152 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1154 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1155 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1157 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1159 #define TCPCB_AT_TAIL (TCPCB_URG)
1161 __u16 urg_ptr; /* Valid w/URG flags is set. */
1162 __u32 ack_seq; /* Sequence number ACK'd */
1165 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1167 #include <net/tcp_ecn.h>
1169 /* Due to TSO, an SKB can be composed of multiple actual
1170 * packets. To keep these tracked properly, we use this.
1172 static inline int tcp_skb_pcount(const struct sk_buff *skb)
1174 return skb_shinfo(skb)->tso_segs;
1177 /* This is valid iff tcp_skb_pcount() > 1. */
1178 static inline int tcp_skb_mss(const struct sk_buff *skb)
1180 return skb_shinfo(skb)->tso_size;
1183 static inline void tcp_inc_pcount(tcp_pcount_t *count,
1184 const struct sk_buff *skb)
1186 count->val += tcp_skb_pcount(skb);
1189 static inline void tcp_inc_pcount_explicit(tcp_pcount_t *count, int amt)
1194 static inline void tcp_dec_pcount_explicit(tcp_pcount_t *count, int amt)
1199 static inline void tcp_dec_pcount(tcp_pcount_t *count,
1200 const struct sk_buff *skb)
1202 count->val -= tcp_skb_pcount(skb);
1205 static inline void tcp_dec_pcount_approx(tcp_pcount_t *count,
1206 const struct sk_buff *skb)
1209 count->val -= tcp_skb_pcount(skb);
1210 if ((int)count->val < 0)
1215 static inline __u32 tcp_get_pcount(const tcp_pcount_t *count)
1220 static inline void tcp_set_pcount(tcp_pcount_t *count, __u32 val)
1225 static inline void tcp_packets_out_inc(struct sock *sk,
1227 const struct sk_buff *skb)
1229 int orig = tcp_get_pcount(&tp->packets_out);
1231 tcp_inc_pcount(&tp->packets_out, skb);
1233 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
1236 static inline void tcp_packets_out_dec(struct tcp_opt *tp,
1237 const struct sk_buff *skb)
1239 tcp_dec_pcount(&tp->packets_out, skb);
1242 /* This determines how many packets are "in the network" to the best
1243 * of our knowledge. In many cases it is conservative, but where
1244 * detailed information is available from the receiver (via SACK
1245 * blocks etc.) we can make more aggressive calculations.
1247 * Use this for decisions involving congestion control, use just
1248 * tp->packets_out to determine if the send queue is empty or not.
1250 * Read this equation as:
1252 * "Packets sent once on transmission queue" MINUS
1253 * "Packets left network, but not honestly ACKed yet" PLUS
1254 * "Packets fast retransmitted"
1256 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_opt *tp)
1258 return (tcp_get_pcount(&tp->packets_out) -
1259 tcp_get_pcount(&tp->left_out) +
1260 tcp_get_pcount(&tp->retrans_out));
1264 * Which congestion algorithim is in use on the connection.
1266 #define tcp_is_vegas(__tp) ((__tp)->adv_cong == TCP_VEGAS)
1267 #define tcp_is_westwood(__tp) ((__tp)->adv_cong == TCP_WESTWOOD)
1268 #define tcp_is_bic(__tp) ((__tp)->adv_cong == TCP_BIC)
1270 /* Recalculate snd_ssthresh, we want to set it to:
1273 * one half the current congestion window, but no
1274 * less than two segments
1277 * behave like Reno until low_window is reached,
1278 * then increase congestion window slowly
1280 static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
1282 if (tcp_is_bic(tp)) {
1283 if (sysctl_tcp_bic_fast_convergence &&
1284 tp->snd_cwnd < tp->bictcp.last_max_cwnd)
1285 tp->bictcp.last_max_cwnd
1286 = (tp->snd_cwnd * (2*BICTCP_1_OVER_BETA-1))
1287 / (BICTCP_1_OVER_BETA/2);
1289 tp->bictcp.last_max_cwnd = tp->snd_cwnd;
1291 if (tp->snd_cwnd > sysctl_tcp_bic_low_window)
1292 return max(tp->snd_cwnd - (tp->snd_cwnd/BICTCP_1_OVER_BETA),
1296 return max(tp->snd_cwnd >> 1U, 2U);
1299 /* Stop taking Vegas samples for now. */
1300 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0)
1302 static inline void tcp_vegas_enable(struct tcp_opt *tp)
1304 /* There are several situations when we must "re-start" Vegas:
1306 * o when a connection is established
1308 * o after fast recovery
1309 * o when we send a packet and there is no outstanding
1310 * unacknowledged data (restarting an idle connection)
1312 * In these circumstances we cannot do a Vegas calculation at the
1313 * end of the first RTT, because any calculation we do is using
1314 * stale info -- both the saved cwnd and congestion feedback are
1317 * Instead we must wait until the completion of an RTT during
1318 * which we actually receive ACKs.
1321 /* Begin taking Vegas samples next time we send something. */
1322 tp->vegas.doing_vegas_now = 1;
1324 /* Set the beginning of the next send window. */
1325 tp->vegas.beg_snd_nxt = tp->snd_nxt;
1327 tp->vegas.cntRTT = 0;
1328 tp->vegas.minRTT = 0x7fffffff;
1331 /* Should we be taking Vegas samples right now? */
1332 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now)
1334 extern void tcp_ca_init(struct tcp_opt *tp);
1336 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
1338 if (tcp_is_vegas(tp)) {
1339 if (ca_state == TCP_CA_Open)
1340 tcp_vegas_enable(tp);
1342 tcp_vegas_disable(tp);
1344 tp->ca_state = ca_state;
1347 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1348 * The exception is rate halving phase, when cwnd is decreasing towards
1351 static inline __u32 tcp_current_ssthresh(struct tcp_opt *tp)
1353 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1354 return tp->snd_ssthresh;
1356 return max(tp->snd_ssthresh,
1357 ((tp->snd_cwnd >> 1) +
1358 (tp->snd_cwnd >> 2)));
1361 static inline void tcp_sync_left_out(struct tcp_opt *tp)
1364 (tcp_get_pcount(&tp->sacked_out) >=
1365 tcp_get_pcount(&tp->packets_out) - tcp_get_pcount(&tp->lost_out)))
1366 tcp_set_pcount(&tp->sacked_out,
1367 (tcp_get_pcount(&tp->packets_out) -
1368 tcp_get_pcount(&tp->lost_out)));
1369 tcp_set_pcount(&tp->left_out,
1370 (tcp_get_pcount(&tp->sacked_out) +
1371 tcp_get_pcount(&tp->lost_out)));
1374 extern void tcp_cwnd_application_limited(struct sock *sk);
1376 /* Congestion window validation. (RFC2861) */
1378 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_opt *tp)
1380 __u32 packets_out = tcp_get_pcount(&tp->packets_out);
1382 if (packets_out >= tp->snd_cwnd) {
1383 /* Network is feed fully. */
1384 tp->snd_cwnd_used = 0;
1385 tp->snd_cwnd_stamp = tcp_time_stamp;
1387 /* Network starves. */
1388 if (tcp_get_pcount(&tp->packets_out) > tp->snd_cwnd_used)
1389 tp->snd_cwnd_used = tcp_get_pcount(&tp->packets_out);
1391 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1392 tcp_cwnd_application_limited(sk);
1396 /* Set slow start threshould and cwnd not falling to slow start */
1397 static inline void __tcp_enter_cwr(struct tcp_opt *tp)
1399 tp->undo_marker = 0;
1400 tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
1401 tp->snd_cwnd = min(tp->snd_cwnd,
1402 tcp_packets_in_flight(tp) + 1U);
1403 tp->snd_cwnd_cnt = 0;
1404 tp->high_seq = tp->snd_nxt;
1405 tp->snd_cwnd_stamp = tcp_time_stamp;
1406 TCP_ECN_queue_cwr(tp);
1409 static inline void tcp_enter_cwr(struct tcp_opt *tp)
1411 tp->prior_ssthresh = 0;
1412 if (tp->ca_state < TCP_CA_CWR) {
1413 __tcp_enter_cwr(tp);
1414 tcp_set_ca_state(tp, TCP_CA_CWR);
1418 extern __u32 tcp_init_cwnd(struct tcp_opt *tp, struct dst_entry *dst);
1420 /* Slow start with delack produces 3 packets of burst, so that
1421 * it is safe "de facto".
1423 static __inline__ __u32 tcp_max_burst(const struct tcp_opt *tp)
1428 static __inline__ int tcp_minshall_check(const struct tcp_opt *tp)
1430 return after(tp->snd_sml,tp->snd_una) &&
1431 !after(tp->snd_sml, tp->snd_nxt);
1434 static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss,
1435 const struct sk_buff *skb)
1438 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1441 /* Return 0, if packet can be sent now without violation Nagle's rules:
1442 1. It is full sized.
1443 2. Or it contains FIN.
1444 3. Or higher layers meant to force a packet boundary, hence the PSH bit.
1445 4. Or TCP_NODELAY was set.
1446 5. Or TCP_CORK is not set, and all sent packets are ACKed.
1447 With Minshall's modification: all sent small packets are ACKed.
1450 static __inline__ int
1451 tcp_nagle_check(const struct tcp_opt *tp, const struct sk_buff *skb,
1452 unsigned mss_now, int nonagle)
1454 return (skb->len < mss_now &&
1455 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
1456 ((nonagle&TCP_NAGLE_CORK) ||
1458 tcp_get_pcount(&tp->packets_out) &&
1459 tcp_minshall_check(tp))));
1462 extern void tcp_set_skb_tso_segs(struct sk_buff *, unsigned int);
1464 /* This checks if the data bearing packet SKB (usually sk->sk_send_head)
1465 * should be put on the wire right now.
1467 static __inline__ int tcp_snd_test(const struct tcp_opt *tp,
1468 struct sk_buff *skb,
1469 unsigned cur_mss, int nonagle)
1471 int pkts = tcp_skb_pcount(skb);
1474 tcp_set_skb_tso_segs(skb, tp->mss_cache_std);
1475 pkts = tcp_skb_pcount(skb);
1478 /* RFC 1122 - section 4.2.3.4
1482 * a) The right edge of this frame exceeds the window
1483 * b) There are packets in flight and we have a small segment
1484 * [SWS avoidance and Nagle algorithm]
1485 * (part of SWS is done on packetization)
1486 * Minshall version sounds: there are no _small_
1487 * segments in flight. (tcp_nagle_check)
1488 * c) We have too many packets 'in flight'
1490 * Don't use the nagle rule for urgent data (or
1491 * for the final FIN -DaveM).
1493 * Also, Nagle rule does not apply to frames, which
1494 * sit in the middle of queue (they have no chances
1495 * to get new data) and if room at tail of skb is
1496 * not enough to save something seriously (<32 for now).
1499 /* Don't be strict about the congestion window for the
1500 * final FIN frame. -DaveM
1502 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode
1503 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) &&
1504 (((tcp_packets_in_flight(tp) + (pkts-1)) < tp->snd_cwnd) ||
1505 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
1506 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd));
1509 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
1511 if (!tcp_get_pcount(&tp->packets_out) && !tp->pending)
1512 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1515 static __inline__ int tcp_skb_is_last(const struct sock *sk,
1516 const struct sk_buff *skb)
1518 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1521 /* Push out any pending frames which were held back due to
1522 * TCP_CORK or attempt at coalescing tiny packets.
1523 * The socket must be locked by the caller.
1525 static __inline__ void __tcp_push_pending_frames(struct sock *sk,
1530 struct sk_buff *skb = sk->sk_send_head;
1533 if (!tcp_skb_is_last(sk, skb))
1534 nonagle = TCP_NAGLE_PUSH;
1535 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) ||
1536 tcp_write_xmit(sk, nonagle))
1537 tcp_check_probe_timer(sk, tp);
1539 tcp_cwnd_validate(sk, tp);
1542 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1545 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1548 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_opt *tp)
1550 struct sk_buff *skb = sk->sk_send_head;
1553 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1),
1554 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle));
1557 static __inline__ void tcp_init_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1562 static __inline__ void tcp_update_wl(struct tcp_opt *tp, u32 ack, u32 seq)
1567 extern void tcp_destroy_sock(struct sock *sk);
1571 * Calculate(/check) TCP checksum
1573 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1574 unsigned long saddr, unsigned long daddr,
1577 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1580 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1582 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1585 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1587 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1588 __tcp_checksum_complete(skb);
1591 /* Prequeue for VJ style copy to user, combined with checksumming. */
1593 static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
1595 tp->ucopy.task = NULL;
1597 tp->ucopy.memory = 0;
1598 skb_queue_head_init(&tp->ucopy.prequeue);
1601 /* Packet is added to VJ-style prequeue for processing in process
1602 * context, if a reader task is waiting. Apparently, this exciting
1603 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1604 * failed somewhere. Latency? Burstiness? Well, at least now we will
1605 * see, why it failed. 8)8) --ANK
1607 * NOTE: is this not too big to inline?
1609 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1611 struct tcp_opt *tp = tcp_sk(sk);
1613 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1614 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1615 tp->ucopy.memory += skb->truesize;
1616 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1617 struct sk_buff *skb1;
1619 BUG_ON(sock_owned_by_user(sk));
1621 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1622 sk->sk_backlog_rcv(sk, skb1);
1623 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1626 tp->ucopy.memory = 0;
1627 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1628 wake_up_interruptible(sk->sk_sleep);
1629 if (!tcp_ack_scheduled(tp))
1630 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1641 static const char *statename[]={
1642 "Unused","Established","Syn Sent","Syn Recv",
1643 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1644 "Close Wait","Last ACK","Listen","Closing"
1648 static __inline__ void tcp_set_state(struct sock *sk, int state)
1650 int oldstate = sk->sk_state;
1653 case TCP_ESTABLISHED:
1654 if (oldstate != TCP_ESTABLISHED)
1655 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1659 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1660 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1662 sk->sk_prot->unhash(sk);
1663 if (tcp_sk(sk)->bind_hash &&
1664 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1668 if (oldstate==TCP_ESTABLISHED)
1669 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1672 /* Change state AFTER socket is unhashed to avoid closed
1673 * socket sitting in hash tables.
1675 sk->sk_state = state;
1678 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1682 static __inline__ void tcp_done(struct sock *sk)
1684 tcp_set_state(sk, TCP_CLOSE);
1685 tcp_clear_xmit_timers(sk);
1687 sk->sk_shutdown = SHUTDOWN_MASK;
1689 if (!sock_flag(sk, SOCK_DEAD))
1690 sk->sk_state_change(sk);
1692 tcp_destroy_sock(sk);
1695 static __inline__ void tcp_sack_reset(struct tcp_opt *tp)
1702 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_opt *tp, __u32 tstamp)
1704 if (tp->tstamp_ok) {
1705 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1706 (TCPOPT_NOP << 16) |
1707 (TCPOPT_TIMESTAMP << 8) |
1709 *ptr++ = htonl(tstamp);
1710 *ptr++ = htonl(tp->ts_recent);
1712 if (tp->eff_sacks) {
1713 struct tcp_sack_block *sp = tp->dsack ? tp->duplicate_sack : tp->selective_acks;
1716 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1717 (TCPOPT_NOP << 16) |
1718 (TCPOPT_SACK << 8) |
1719 (TCPOLEN_SACK_BASE +
1720 (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1721 for(this_sack = 0; this_sack < tp->eff_sacks; this_sack++) {
1722 *ptr++ = htonl(sp[this_sack].start_seq);
1723 *ptr++ = htonl(sp[this_sack].end_seq);
1732 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1733 * If this is every changed make sure to change the definition of
1734 * MAX_SYN_SIZE to match the new maximum number of options that you
1737 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1738 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1740 /* We always get an MSS option.
1741 * The option bytes which will be seen in normal data
1742 * packets should timestamps be used, must be in the MSS
1743 * advertised. But we subtract them from tp->mss_cache so
1744 * that calculations in tcp_sendmsg are simpler etc.
1745 * So account for this fact here if necessary. If we
1746 * don't do this correctly, as a receiver we won't
1747 * recognize data packets as being full sized when we
1748 * should, and thus we won't abide by the delayed ACK
1750 * SACKs don't matter, we never delay an ACK when we
1751 * have any of those going out.
1753 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1756 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1757 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1759 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1760 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1761 *ptr++ = htonl(tstamp); /* TSVAL */
1762 *ptr++ = htonl(ts_recent); /* TSECR */
1764 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1765 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1767 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1770 /* Determine a window scaling and initial window to offer. */
1771 extern void tcp_select_initial_window(int __space, __u32 mss,
1772 __u32 *rcv_wnd, __u32 *window_clamp,
1773 int wscale_ok, __u8 *rcv_wscale);
1775 static inline int tcp_win_from_space(int space)
1777 return sysctl_tcp_adv_win_scale<=0 ?
1778 (space>>(-sysctl_tcp_adv_win_scale)) :
1779 space - (space>>sysctl_tcp_adv_win_scale);
1782 /* Note: caller must be prepared to deal with negative returns */
1783 static inline int tcp_space(const struct sock *sk)
1785 return tcp_win_from_space(sk->sk_rcvbuf -
1786 atomic_read(&sk->sk_rmem_alloc));
1789 static inline int tcp_full_space(const struct sock *sk)
1791 return tcp_win_from_space(sk->sk_rcvbuf);
1794 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
1797 struct tcp_opt *tp = tcp_sk(sk);
1800 sk_acceptq_added(sk);
1802 if (!tp->accept_queue_tail) {
1803 tp->accept_queue = req;
1805 tp->accept_queue_tail->dl_next = req;
1807 tp->accept_queue_tail = req;
1808 req->dl_next = NULL;
1811 struct tcp_listen_opt
1813 u8 max_qlen_log; /* log_2 of maximal queued SYNs */
1818 struct open_request *syn_table[TCP_SYNQ_HSIZE];
1822 tcp_synq_removed(struct sock *sk, struct open_request *req)
1824 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1826 if (--lopt->qlen == 0)
1827 tcp_delete_keepalive_timer(sk);
1828 if (req->retrans == 0)
1832 static inline void tcp_synq_added(struct sock *sk)
1834 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt;
1836 if (lopt->qlen++ == 0)
1837 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1841 static inline int tcp_synq_len(struct sock *sk)
1843 return tcp_sk(sk)->listen_opt->qlen;
1846 static inline int tcp_synq_young(struct sock *sk)
1848 return tcp_sk(sk)->listen_opt->qlen_young;
1851 static inline int tcp_synq_is_full(struct sock *sk)
1853 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log;
1856 static inline void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
1857 struct open_request **prev)
1859 write_lock(&tp->syn_wait_lock);
1860 *prev = req->dl_next;
1861 write_unlock(&tp->syn_wait_lock);
1864 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req,
1865 struct open_request **prev)
1867 tcp_synq_unlink(tcp_sk(sk), req, prev);
1868 tcp_synq_removed(sk, req);
1869 tcp_openreq_free(req);
1872 static __inline__ void tcp_openreq_init(struct open_request *req,
1874 struct sk_buff *skb)
1876 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1877 req->rcv_isn = TCP_SKB_CB(skb)->seq;
1878 req->mss = tp->mss_clamp;
1879 req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
1880 req->tstamp_ok = tp->tstamp_ok;
1881 req->sack_ok = tp->sack_ok;
1882 req->snd_wscale = tp->snd_wscale;
1883 req->wscale_ok = tp->wscale_ok;
1886 req->rmt_port = skb->h.th->source;
1889 extern void tcp_enter_memory_pressure(void);
1891 extern void tcp_listen_wlock(void);
1893 /* - We may sleep inside this lock.
1894 * - If sleeping is not required (or called from BH),
1895 * use plain read_(un)lock(&tcp_lhash_lock).
1898 static inline void tcp_listen_lock(void)
1900 /* read_lock synchronizes to candidates to writers */
1901 read_lock(&tcp_lhash_lock);
1902 atomic_inc(&tcp_lhash_users);
1903 read_unlock(&tcp_lhash_lock);
1906 static inline void tcp_listen_unlock(void)
1908 if (atomic_dec_and_test(&tcp_lhash_users))
1909 wake_up(&tcp_lhash_wait);
1912 static inline int keepalive_intvl_when(const struct tcp_opt *tp)
1914 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1917 static inline int keepalive_time_when(const struct tcp_opt *tp)
1919 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1922 static inline int tcp_fin_time(const struct tcp_opt *tp)
1924 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1926 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1927 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1932 static inline int tcp_paws_check(const struct tcp_opt *tp, int rst)
1934 if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
1936 if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
1939 /* RST segments are not recommended to carry timestamp,
1940 and, if they do, it is recommended to ignore PAWS because
1941 "their cleanup function should take precedence over timestamps."
1942 Certainly, it is mistake. It is necessary to understand the reasons
1943 of this constraint to relax it: if peer reboots, clock may go
1944 out-of-sync and half-open connections will not be reset.
1945 Actually, the problem would be not existing if all
1946 the implementations followed draft about maintaining clock
1947 via reboots. Linux-2.2 DOES NOT!
1949 However, we can relax time bounds for RST segments to MSL.
1951 if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
1956 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
1958 sk->sk_route_caps = dst->dev->features;
1959 if (sk->sk_route_caps & NETIF_F_TSO) {
1960 if (sk->sk_no_largesend || dst->header_len)
1961 sk->sk_route_caps &= ~NETIF_F_TSO;
1965 #define TCP_CHECK_TIMER(sk) do { } while (0)
1967 static inline int tcp_use_frto(const struct sock *sk)
1969 const struct tcp_opt *tp = tcp_sk(sk);
1971 /* F-RTO must be activated in sysctl and there must be some
1972 * unsent new data, and the advertised window should allow
1975 return (sysctl_tcp_frto && sk->sk_send_head &&
1976 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1977 tp->snd_una + tp->snd_wnd));
1980 static inline void tcp_mib_init(void)
1983 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1984 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1985 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1986 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1990 enum tcp_seq_states {
1991 TCP_SEQ_STATE_LISTENING,
1992 TCP_SEQ_STATE_OPENREQ,
1993 TCP_SEQ_STATE_ESTABLISHED,
1994 TCP_SEQ_STATE_TIME_WAIT,
1997 struct tcp_seq_afinfo {
1998 struct module *owner;
2001 int (*seq_show) (struct seq_file *m, void *v);
2002 struct file_operations *seq_fops;
2005 struct tcp_iter_state {
2007 enum tcp_seq_states state;
2008 struct sock *syn_wait_sk;
2009 int bucket, sbucket, num, uid;
2010 struct seq_operations seq_ops;
2013 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
2014 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
2016 /* TCP Westwood functions and constants */
2018 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
2019 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
2021 static inline void tcp_westwood_update_rtt(struct tcp_opt *tp, __u32 rtt_seq)
2023 if (tcp_is_westwood(tp))
2024 tp->westwood.rtt = rtt_seq;
2027 void __tcp_westwood_fast_bw(struct sock *, struct sk_buff *);
2028 void __tcp_westwood_slow_bw(struct sock *, struct sk_buff *);
2030 static inline void tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb)
2032 if (tcp_is_westwood(tcp_sk(sk)))
2033 __tcp_westwood_fast_bw(sk, skb);
2036 static inline void tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb)
2038 if (tcp_is_westwood(tcp_sk(sk)))
2039 __tcp_westwood_slow_bw(sk, skb);
2042 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2044 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) /
2045 (__u32) (tp->mss_cache_std),
2049 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_opt *tp)
2051 return tcp_is_westwood(tp) ? __tcp_westwood_bw_rttmin(tp) : 0;
2054 static inline int tcp_westwood_ssthresh(struct tcp_opt *tp)
2058 if (tcp_is_westwood(tp)) {
2059 ssthresh = __tcp_westwood_bw_rttmin(tp);
2061 tp->snd_ssthresh = ssthresh;
2064 return (ssthresh != 0);
2067 static inline int tcp_westwood_cwnd(struct tcp_opt *tp)
2071 if (tcp_is_westwood(tp)) {
2072 cwnd = __tcp_westwood_bw_rttmin(tp);
2074 tp->snd_cwnd = cwnd;