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
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>
34 #include <net/request_sock.h>
38 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
39 #include <linux/ipv6.h>
41 #include <linux/seq_file.h>
43 /* This is for all connections with a full identity, no wildcards.
44 * New scheme, half the table is for TIME_WAIT, the other half is
45 * for the rest. I'll experiment with dynamic table growth later.
47 struct tcp_ehash_bucket {
49 struct hlist_head chain;
50 } __attribute__((__aligned__(8)));
52 /* This is for listening sockets, thus all sockets which possess wildcards. */
53 #define TCP_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
55 /* There are a few simple rules, which allow for local port reuse by
56 * an application. In essence:
58 * 1) Sockets bound to different interfaces may share a local port.
59 * Failing that, goto test 2.
60 * 2) If all sockets have sk->sk_reuse set, and none of them are in
61 * TCP_LISTEN state, the port may be shared.
62 * Failing that, goto test 3.
63 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
64 * address, and none of them are the same, the port may be
66 * Failing this, the port cannot be shared.
68 * The interesting point, is test #2. This is what an FTP server does
69 * all day. To optimize this case we use a specific flag bit defined
70 * below. As we add sockets to a bind bucket list, we perform a
71 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
72 * As long as all sockets added to a bind bucket pass this test,
73 * the flag bit will be set.
74 * The resulting situation is that tcp_v[46]_verify_bind() can just check
75 * for this flag bit, if it is set and the socket trying to bind has
76 * sk->sk_reuse set, we don't even have to walk the owners list at all,
77 * we return that it is ok to bind this socket to the requested local port.
79 * Sounds like a lot of work, but it is worth it. In a more naive
80 * implementation (ie. current FreeBSD etc.) the entire list of ports
81 * must be walked for each data port opened by an ftp server. Needless
82 * to say, this does not scale at all. With a couple thousand FTP
83 * users logged onto your box, isn't it nice to know that new data
84 * ports are created in O(1) time? I thought so. ;-) -DaveM
86 struct tcp_bind_bucket {
88 signed short fastreuse;
89 struct hlist_node node;
90 struct hlist_head owners;
93 #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node)
95 struct tcp_bind_hashbucket {
97 struct hlist_head chain;
100 static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head)
102 return hlist_entry(head->chain.first, struct tcp_bind_bucket, node);
105 static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head)
107 return hlist_empty(&head->chain) ? NULL : __tb_head(head);
110 extern struct tcp_hashinfo {
111 /* This is for sockets with full identity only. Sockets here will
112 * always be without wildcards and will have the following invariant:
114 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
116 * First half of the table is for sockets not in TIME_WAIT, second half
117 * is for TIME_WAIT sockets only.
119 struct tcp_ehash_bucket *__tcp_ehash;
121 /* Ok, let's try this, I give up, we do need a local binding
122 * TCP hash as well as the others for fast bind/connect.
124 struct tcp_bind_hashbucket *__tcp_bhash;
126 int __tcp_bhash_size;
127 int __tcp_ehash_size;
129 /* All sockets in TCP_LISTEN state will be in here. This is the only
130 * table where wildcard'd TCP sockets can exist. Hash function here
131 * is just local port number.
133 struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE];
135 /* All the above members are written once at bootup and
136 * never written again _or_ are predominantly read-access.
138 * Now align to a new cache line as all the following members
141 rwlock_t __tcp_lhash_lock ____cacheline_aligned;
142 atomic_t __tcp_lhash_users;
143 wait_queue_head_t __tcp_lhash_wait;
144 spinlock_t __tcp_portalloc_lock;
147 #define tcp_ehash (tcp_hashinfo.__tcp_ehash)
148 #define tcp_bhash (tcp_hashinfo.__tcp_bhash)
149 #define tcp_ehash_size (tcp_hashinfo.__tcp_ehash_size)
150 #define tcp_bhash_size (tcp_hashinfo.__tcp_bhash_size)
151 #define tcp_listening_hash (tcp_hashinfo.__tcp_listening_hash)
152 #define tcp_lhash_lock (tcp_hashinfo.__tcp_lhash_lock)
153 #define tcp_lhash_users (tcp_hashinfo.__tcp_lhash_users)
154 #define tcp_lhash_wait (tcp_hashinfo.__tcp_lhash_wait)
155 #define tcp_portalloc_lock (tcp_hashinfo.__tcp_portalloc_lock)
157 extern kmem_cache_t *tcp_bucket_cachep;
158 extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head,
159 unsigned short snum);
160 extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb);
161 extern void tcp_bucket_unlock(struct sock *sk);
162 extern int tcp_port_rover;
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 volatile unsigned char tw_substate;
197 unsigned char tw_rcv_wscale;
199 /* Socket demultiplex comparisons on incoming packets. */
200 /* these five are in inet_sock */
202 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES)));
206 /* And these are ours. */
213 long tw_ts_recent_stamp;
214 unsigned long tw_ttd;
215 struct tcp_bind_bucket *tw_tb;
216 struct hlist_node tw_death_node;
217 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
218 struct in6_addr tw_v6_daddr;
219 struct in6_addr tw_v6_rcv_saddr;
224 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw,
225 struct hlist_head *list)
227 hlist_add_head(&tw->tw_node, list);
230 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw,
231 struct hlist_head *list)
233 hlist_add_head(&tw->tw_bind_node, list);
236 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw)
238 return tw->tw_death_node.pprev != NULL;
241 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw)
243 tw->tw_death_node.pprev = NULL;
246 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw)
248 __hlist_del(&tw->tw_death_node);
249 tw_dead_node_init(tw);
252 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw)
254 if (tw_dead_hashed(tw)) {
255 __tw_del_dead_node(tw);
261 #define tw_for_each(tw, node, head) \
262 hlist_for_each_entry(tw, node, head, tw_node)
264 #define tw_for_each_inmate(tw, node, jail) \
265 hlist_for_each_entry(tw, node, jail, tw_death_node)
267 #define tw_for_each_inmate_safe(tw, node, safe, jail) \
268 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
270 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk))
272 static inline u32 tcp_v4_rcv_saddr(const struct sock *sk)
274 return likely(sk->sk_state != TCP_TIME_WAIT) ?
275 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr;
278 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
279 static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk)
281 return likely(sk->sk_state != TCP_TIME_WAIT) ?
282 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr;
285 static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk)
287 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL;
290 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only)
292 static inline int tcp_v6_ipv6only(const struct sock *sk)
294 return likely(sk->sk_state != TCP_TIME_WAIT) ?
295 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk);
298 # define __tcp_v6_rcv_saddr(__sk) NULL
299 # define tcp_v6_rcv_saddr(__sk) NULL
300 # define tcptw_sk_ipv6only(__sk) 0
301 # define tcp_v6_ipv6only(__sk) 0
304 extern kmem_cache_t *tcp_timewait_cachep;
306 static inline void tcp_tw_put(struct tcp_tw_bucket *tw)
308 if (atomic_dec_and_test(&tw->tw_refcnt)) {
309 #ifdef INET_REFCNT_DEBUG
310 printk(KERN_DEBUG "tw_bucket %p released\n", tw);
312 kmem_cache_free(tcp_timewait_cachep, tw);
316 extern atomic_t tcp_orphan_count;
317 extern int tcp_tw_count;
318 extern void tcp_time_wait(struct sock *sk, int state, 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)
512 #define TCPOPT_NOP 1 /* Padding */
513 #define TCPOPT_EOL 0 /* End of options */
514 #define TCPOPT_MSS 2 /* Segment size negotiating */
515 #define TCPOPT_WINDOW 3 /* Window scaling */
516 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
517 #define TCPOPT_SACK 5 /* SACK Block */
518 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
524 #define TCPOLEN_MSS 4
525 #define TCPOLEN_WINDOW 3
526 #define TCPOLEN_SACK_PERM 2
527 #define TCPOLEN_TIMESTAMP 10
529 /* But this is what stacks really send out. */
530 #define TCPOLEN_TSTAMP_ALIGNED 12
531 #define TCPOLEN_WSCALE_ALIGNED 4
532 #define TCPOLEN_SACKPERM_ALIGNED 4
533 #define TCPOLEN_SACK_BASE 2
534 #define TCPOLEN_SACK_BASE_ALIGNED 4
535 #define TCPOLEN_SACK_PERBLOCK 8
537 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
538 #define TCP_TIME_DACK 2 /* Delayed ack timer */
539 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
540 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
542 /* Flags in tp->nonagle */
543 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
544 #define TCP_NAGLE_CORK 2 /* Socket is corked */
545 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
547 /* sysctl variables for tcp */
548 extern int sysctl_tcp_timestamps;
549 extern int sysctl_tcp_window_scaling;
550 extern int sysctl_tcp_sack;
551 extern int sysctl_tcp_fin_timeout;
552 extern int sysctl_tcp_tw_recycle;
553 extern int sysctl_tcp_keepalive_time;
554 extern int sysctl_tcp_keepalive_probes;
555 extern int sysctl_tcp_keepalive_intvl;
556 extern int sysctl_tcp_syn_retries;
557 extern int sysctl_tcp_synack_retries;
558 extern int sysctl_tcp_retries1;
559 extern int sysctl_tcp_retries2;
560 extern int sysctl_tcp_orphan_retries;
561 extern int sysctl_tcp_syncookies;
562 extern int sysctl_tcp_retrans_collapse;
563 extern int sysctl_tcp_stdurg;
564 extern int sysctl_tcp_rfc1337;
565 extern int sysctl_tcp_abort_on_overflow;
566 extern int sysctl_tcp_max_orphans;
567 extern int sysctl_tcp_max_tw_buckets;
568 extern int sysctl_tcp_fack;
569 extern int sysctl_tcp_reordering;
570 extern int sysctl_tcp_ecn;
571 extern int sysctl_tcp_dsack;
572 extern int sysctl_tcp_mem[3];
573 extern int sysctl_tcp_wmem[3];
574 extern int sysctl_tcp_rmem[3];
575 extern int sysctl_tcp_app_win;
576 extern int sysctl_tcp_adv_win_scale;
577 extern int sysctl_tcp_tw_reuse;
578 extern int sysctl_tcp_frto;
579 extern int sysctl_tcp_low_latency;
580 extern int sysctl_tcp_nometrics_save;
581 extern int sysctl_tcp_moderate_rcvbuf;
582 extern int sysctl_tcp_tso_win_divisor;
584 extern atomic_t tcp_memory_allocated;
585 extern atomic_t tcp_sockets_allocated;
586 extern int tcp_memory_pressure;
588 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
589 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
591 #define TCP_INET_FAMILY(fam) 1
595 * Pointers to address related TCP functions
596 * (i.e. things that depend on the address family)
600 int (*queue_xmit) (struct sk_buff *skb,
603 void (*send_check) (struct sock *sk,
606 struct sk_buff *skb);
608 int (*rebuild_header) (struct sock *sk);
610 int (*conn_request) (struct sock *sk,
611 struct sk_buff *skb);
613 struct sock * (*syn_recv_sock) (struct sock *sk,
615 struct request_sock *req,
616 struct dst_entry *dst);
618 int (*remember_stamp) (struct sock *sk);
620 __u16 net_header_len;
622 int (*setsockopt) (struct sock *sk,
628 int (*getsockopt) (struct sock *sk,
635 void (*addr2sockaddr) (struct sock *sk,
642 * The next routines deal with comparing 32 bit unsigned ints
643 * and worry about wraparound (automatic with unsigned arithmetic).
646 static inline int before(__u32 seq1, __u32 seq2)
648 return (__s32)(seq1-seq2) < 0;
651 static inline int after(__u32 seq1, __u32 seq2)
653 return (__s32)(seq2-seq1) < 0;
657 /* is s2<=s1<=s3 ? */
658 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
660 return seq3 - seq2 >= seq1 - seq2;
664 extern struct proto tcp_prot;
666 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
667 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
668 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
669 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
670 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
671 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
672 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
674 extern void tcp_put_port(struct sock *sk);
675 extern void tcp_inherit_port(struct sock *sk, struct sock *child);
677 extern void tcp_v4_err(struct sk_buff *skb, u32);
679 extern void tcp_shutdown (struct sock *sk, int how);
681 extern int tcp_v4_rcv(struct sk_buff *skb);
683 extern int tcp_v4_remember_stamp(struct sock *sk);
685 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
687 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
688 struct msghdr *msg, size_t size);
689 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
691 extern int tcp_ioctl(struct sock *sk,
695 extern int tcp_rcv_state_process(struct sock *sk,
700 extern int tcp_rcv_established(struct sock *sk,
705 extern void tcp_rcv_space_adjust(struct sock *sk);
714 static inline void tcp_schedule_ack(struct tcp_sock *tp)
716 tp->ack.pending |= TCP_ACK_SCHED;
719 static inline int tcp_ack_scheduled(struct tcp_sock *tp)
721 return tp->ack.pending&TCP_ACK_SCHED;
724 static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp, unsigned int pkts)
727 if (pkts >= tp->ack.quick) {
730 /* Leaving quickack mode we deflate ATO. */
731 tp->ack.ato = TCP_ATO_MIN;
733 tp->ack.quick -= pkts;
737 extern void tcp_enter_quickack_mode(struct tcp_sock *tp);
739 static __inline__ void tcp_delack_init(struct tcp_sock *tp)
741 memset(&tp->ack, 0, sizeof(tp->ack));
744 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
746 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
758 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw,
763 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
764 struct request_sock *req,
765 struct request_sock **prev);
766 extern int tcp_child_process(struct sock *parent,
768 struct sk_buff *skb);
769 extern void tcp_enter_frto(struct sock *sk);
770 extern void tcp_enter_loss(struct sock *sk, int how);
771 extern void tcp_clear_retrans(struct tcp_sock *tp);
772 extern void tcp_update_metrics(struct sock *sk);
774 extern void tcp_close(struct sock *sk,
776 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
777 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
779 extern int tcp_getsockopt(struct sock *sk, int level,
783 extern int tcp_setsockopt(struct sock *sk, int level,
784 int optname, char __user *optval,
786 extern void tcp_set_keepalive(struct sock *sk, int val);
787 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
789 size_t len, int nonblock,
790 int flags, int *addr_len);
792 extern int tcp_listen_start(struct sock *sk);
794 extern void tcp_parse_options(struct sk_buff *skb,
795 struct tcp_options_received *opt_rx,
799 * TCP v4 functions exported for the inet6 API
802 extern int tcp_v4_rebuild_header(struct sock *sk);
804 extern int tcp_v4_build_header(struct sock *sk,
805 struct sk_buff *skb);
807 extern void tcp_v4_send_check(struct sock *sk,
808 struct tcphdr *th, int len,
809 struct sk_buff *skb);
811 extern int tcp_v4_conn_request(struct sock *sk,
812 struct sk_buff *skb);
814 extern struct sock * tcp_create_openreq_child(struct sock *sk,
815 struct request_sock *req,
816 struct sk_buff *skb);
818 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
820 struct request_sock *req,
821 struct dst_entry *dst);
823 extern int tcp_v4_do_rcv(struct sock *sk,
824 struct sk_buff *skb);
826 extern int tcp_v4_connect(struct sock *sk,
827 struct sockaddr *uaddr,
830 extern int tcp_connect(struct sock *sk);
832 extern struct sk_buff * tcp_make_synack(struct sock *sk,
833 struct dst_entry *dst,
834 struct request_sock *req);
836 extern int tcp_disconnect(struct sock *sk, int flags);
838 extern void tcp_unhash(struct sock *sk);
840 extern int tcp_v4_hash_connecting(struct sock *sk);
843 /* From syncookies.c */
844 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
845 struct ip_options *opt);
846 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
851 extern int tcp_write_xmit(struct sock *, int nonagle);
852 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
853 unsigned cur_mss, int nonagle);
854 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
855 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
856 extern void tcp_xmit_retransmit_queue(struct sock *);
857 extern void tcp_simple_retransmit(struct sock *);
858 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
860 extern void tcp_send_probe0(struct sock *);
861 extern void tcp_send_partial(struct sock *);
862 extern int tcp_write_wakeup(struct sock *);
863 extern void tcp_send_fin(struct sock *sk);
864 extern void tcp_send_active_reset(struct sock *sk, int priority);
865 extern int tcp_send_synack(struct sock *);
866 extern void tcp_push_one(struct sock *, unsigned mss_now);
867 extern void tcp_send_ack(struct sock *sk);
868 extern void tcp_send_delayed_ack(struct sock *sk);
871 extern void tcp_init_xmit_timers(struct sock *);
872 extern void tcp_clear_xmit_timers(struct sock *);
874 extern void tcp_delete_keepalive_timer(struct sock *);
875 extern void tcp_reset_keepalive_timer(struct sock *, unsigned long);
876 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
877 extern unsigned int tcp_current_mss(struct sock *sk, int large);
880 extern const char tcp_timer_bug_msg[];
884 extern void tcp_get_info(struct sock *, struct tcp_info *);
886 /* Read 'sendfile()'-style from a TCP socket */
887 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
888 unsigned int, size_t);
889 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
890 sk_read_actor_t recv_actor);
892 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
894 struct tcp_sock *tp = tcp_sk(sk);
897 case TCP_TIME_RETRANS:
898 case TCP_TIME_PROBE0:
901 #ifdef TCP_CLEAR_TIMERS
902 sk_stop_timer(sk, &tp->retransmit_timer);
909 #ifdef TCP_CLEAR_TIMERS
910 sk_stop_timer(sk, &tp->delack_timer);
915 printk(tcp_timer_bug_msg);
923 * Reset the retransmission timer
925 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
927 struct tcp_sock *tp = tcp_sk(sk);
929 if (when > TCP_RTO_MAX) {
931 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
937 case TCP_TIME_RETRANS:
938 case TCP_TIME_PROBE0:
940 tp->timeout = jiffies+when;
941 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
945 tp->ack.pending |= TCP_ACK_TIMER;
946 tp->ack.timeout = jiffies+when;
947 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
952 printk(tcp_timer_bug_msg);
958 /* Initialize RCV_MSS value.
959 * RCV_MSS is an our guess about MSS used by the peer.
960 * We haven't any direct information about the MSS.
961 * It's better to underestimate the RCV_MSS rather than overestimate.
962 * Overestimations make us ACKing less frequently than needed.
963 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
966 static inline void tcp_initialize_rcv_mss(struct sock *sk)
968 struct tcp_sock *tp = tcp_sk(sk);
969 unsigned int hint = min(tp->advmss, tp->mss_cache_std);
971 hint = min(hint, tp->rcv_wnd/2);
972 hint = min(hint, TCP_MIN_RCVMSS);
973 hint = max(hint, TCP_MIN_MSS);
975 tp->ack.rcv_mss = hint;
978 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
980 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
981 ntohl(TCP_FLAG_ACK) |
985 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
987 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
990 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
992 if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
994 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
996 tcp_fast_path_on(tp);
999 /* Compute the actual receive window we are currently advertising.
1000 * Rcv_nxt can be after the window if our peer push more data
1001 * than the offered window.
1003 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
1005 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
1012 /* Choose a new window, without checks for shrinking, and without
1013 * scaling applied to the result. The caller does these things
1014 * if necessary. This is a "raw" window selection.
1016 extern u32 __tcp_select_window(struct sock *sk);
1018 /* TCP timestamps are only 32-bits, this causes a slight
1019 * complication on 64-bit systems since we store a snapshot
1020 * of jiffies in the buffer control blocks below. We decidely
1021 * only use of the low 32-bits of jiffies and hide the ugly
1022 * casts with the following macro.
1024 #define tcp_time_stamp ((__u32)(jiffies))
1026 /* This is what the send packet queueing engine uses to pass
1027 * TCP per-packet control information to the transmission
1028 * code. We also store the host-order sequence numbers in
1029 * here too. This is 36 bytes on 32-bit architectures,
1030 * 40 bytes on 64-bit machines, if this grows please adjust
1031 * skbuff.h:skbuff->cb[xxx] size appropriately.
1035 struct inet_skb_parm h4;
1036 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
1037 struct inet6_skb_parm h6;
1039 } header; /* For incoming frames */
1040 __u32 seq; /* Starting sequence number */
1041 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1042 __u32 when; /* used to compute rtt's */
1043 __u8 flags; /* TCP header flags. */
1045 /* NOTE: These must match up to the flags byte in a
1048 #define TCPCB_FLAG_FIN 0x01
1049 #define TCPCB_FLAG_SYN 0x02
1050 #define TCPCB_FLAG_RST 0x04
1051 #define TCPCB_FLAG_PSH 0x08
1052 #define TCPCB_FLAG_ACK 0x10
1053 #define TCPCB_FLAG_URG 0x20
1054 #define TCPCB_FLAG_ECE 0x40
1055 #define TCPCB_FLAG_CWR 0x80
1057 __u8 sacked; /* State flags for SACK/FACK. */
1058 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
1059 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
1060 #define TCPCB_LOST 0x04 /* SKB is lost */
1061 #define TCPCB_TAGBITS 0x07 /* All tag bits */
1063 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
1064 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
1066 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
1068 #define TCPCB_AT_TAIL (TCPCB_URG)
1070 __u16 urg_ptr; /* Valid w/URG flags is set. */
1071 __u32 ack_seq; /* Sequence number ACK'd */
1074 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1076 #include <net/tcp_ecn.h>
1078 /* Due to TSO, an SKB can be composed of multiple actual
1079 * packets. To keep these tracked properly, we use this.
1081 static inline int tcp_skb_pcount(const struct sk_buff *skb)
1083 return skb_shinfo(skb)->tso_segs;
1086 /* This is valid iff tcp_skb_pcount() > 1. */
1087 static inline int tcp_skb_mss(const struct sk_buff *skb)
1089 return skb_shinfo(skb)->tso_size;
1092 static inline void tcp_dec_pcount_approx(__u32 *count,
1093 const struct sk_buff *skb)
1096 *count -= tcp_skb_pcount(skb);
1097 if ((int)*count < 0)
1102 static inline void tcp_packets_out_inc(struct sock *sk,
1103 struct tcp_sock *tp,
1104 const struct sk_buff *skb)
1106 int orig = tp->packets_out;
1108 tp->packets_out += tcp_skb_pcount(skb);
1110 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
1113 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
1114 const struct sk_buff *skb)
1116 tp->packets_out -= tcp_skb_pcount(skb);
1119 /* Events passed to congestion control interface */
1121 CA_EVENT_TX_START, /* first transmit when no packets in flight */
1122 CA_EVENT_CWND_RESTART, /* congestion window restart */
1123 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
1124 CA_EVENT_FRTO, /* fast recovery timeout */
1125 CA_EVENT_LOSS, /* loss timeout */
1126 CA_EVENT_FAST_ACK, /* in sequence ack */
1127 CA_EVENT_SLOW_ACK, /* other ack */
1131 * Interface for adding new TCP congestion control handlers
1133 #define TCP_CA_NAME_MAX 16
1134 struct tcp_congestion_ops {
1135 struct list_head list;
1137 /* initialize private data (optional) */
1138 void (*init)(struct tcp_sock *tp);
1139 /* cleanup private data (optional) */
1140 void (*release)(struct tcp_sock *tp);
1142 /* return slow start threshold (required) */
1143 u32 (*ssthresh)(struct tcp_sock *tp);
1144 /* lower bound for congestion window (optional) */
1145 u32 (*min_cwnd)(struct tcp_sock *tp);
1146 /* do new cwnd calculation (required) */
1147 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
1148 u32 rtt, u32 in_flight, int good_ack);
1149 /* round trip time sample per acked packet (optional) */
1150 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
1151 /* call before changing ca_state (optional) */
1152 void (*set_state)(struct tcp_sock *tp, u8 new_state);
1153 /* call when cwnd event occurs (optional) */
1154 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
1155 /* new value of cwnd after loss (optional) */
1156 u32 (*undo_cwnd)(struct tcp_sock *tp);
1157 /* hook for packet ack accounting (optional) */
1158 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
1159 /* get info for tcp_diag (optional) */
1160 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
1162 char name[TCP_CA_NAME_MAX];
1163 struct module *owner;
1166 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1167 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1169 extern void tcp_init_congestion_control(struct tcp_sock *tp);
1170 extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
1171 extern int tcp_set_default_congestion_control(const char *name);
1172 extern void tcp_get_default_congestion_control(char *name);
1173 extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
1175 extern struct tcp_congestion_ops tcp_init_congestion_ops;
1176 extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
1177 extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
1178 u32 rtt, u32 in_flight, int flag);
1179 extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
1180 extern struct tcp_congestion_ops tcp_reno;
1182 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
1184 if (tp->ca_ops->set_state)
1185 tp->ca_ops->set_state(tp, ca_state);
1186 tp->ca_state = ca_state;
1189 static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
1191 if (tp->ca_ops->cwnd_event)
1192 tp->ca_ops->cwnd_event(tp, event);
1195 /* This determines how many packets are "in the network" to the best
1196 * of our knowledge. In many cases it is conservative, but where
1197 * detailed information is available from the receiver (via SACK
1198 * blocks etc.) we can make more aggressive calculations.
1200 * Use this for decisions involving congestion control, use just
1201 * tp->packets_out to determine if the send queue is empty or not.
1203 * Read this equation as:
1205 * "Packets sent once on transmission queue" MINUS
1206 * "Packets left network, but not honestly ACKed yet" PLUS
1207 * "Packets fast retransmitted"
1209 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1211 return (tp->packets_out - tp->left_out + tp->retrans_out);
1214 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1215 * The exception is rate halving phase, when cwnd is decreasing towards
1218 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
1220 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
1221 return tp->snd_ssthresh;
1223 return max(tp->snd_ssthresh,
1224 ((tp->snd_cwnd >> 1) +
1225 (tp->snd_cwnd >> 2)));
1228 static inline void tcp_sync_left_out(struct tcp_sock *tp)
1230 if (tp->rx_opt.sack_ok &&
1231 (tp->sacked_out >= tp->packets_out - tp->lost_out))
1232 tp->sacked_out = tp->packets_out - tp->lost_out;
1233 tp->left_out = tp->sacked_out + tp->lost_out;
1236 extern void tcp_cwnd_application_limited(struct sock *sk);
1238 /* Congestion window validation. (RFC2861) */
1240 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_sock *tp)
1242 __u32 packets_out = tp->packets_out;
1244 if (packets_out >= tp->snd_cwnd) {
1245 /* Network is feed fully. */
1246 tp->snd_cwnd_used = 0;
1247 tp->snd_cwnd_stamp = tcp_time_stamp;
1249 /* Network starves. */
1250 if (tp->packets_out > tp->snd_cwnd_used)
1251 tp->snd_cwnd_used = tp->packets_out;
1253 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto)
1254 tcp_cwnd_application_limited(sk);
1258 /* Set slow start threshould and cwnd not falling to slow start */
1259 static inline void __tcp_enter_cwr(struct tcp_sock *tp)
1261 tp->undo_marker = 0;
1262 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
1263 tp->snd_cwnd = min(tp->snd_cwnd,
1264 tcp_packets_in_flight(tp) + 1U);
1265 tp->snd_cwnd_cnt = 0;
1266 tp->high_seq = tp->snd_nxt;
1267 tp->snd_cwnd_stamp = tcp_time_stamp;
1268 TCP_ECN_queue_cwr(tp);
1271 static inline void tcp_enter_cwr(struct tcp_sock *tp)
1273 tp->prior_ssthresh = 0;
1274 if (tp->ca_state < TCP_CA_CWR) {
1275 __tcp_enter_cwr(tp);
1276 tcp_set_ca_state(tp, TCP_CA_CWR);
1280 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
1282 /* Slow start with delack produces 3 packets of burst, so that
1283 * it is safe "de facto".
1285 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1290 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
1291 const struct sk_buff *skb)
1294 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1297 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
1299 if (!tp->packets_out && !tp->pending)
1300 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
1303 static __inline__ void tcp_push_pending_frames(struct sock *sk,
1304 struct tcp_sock *tp)
1306 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
1309 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1314 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1319 extern void tcp_destroy_sock(struct sock *sk);
1323 * Calculate(/check) TCP checksum
1325 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
1326 unsigned long saddr, unsigned long daddr,
1329 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1332 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
1334 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
1337 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
1339 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1340 __tcp_checksum_complete(skb);
1343 /* Prequeue for VJ style copy to user, combined with checksumming. */
1345 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
1347 tp->ucopy.task = NULL;
1349 tp->ucopy.memory = 0;
1350 skb_queue_head_init(&tp->ucopy.prequeue);
1353 /* Packet is added to VJ-style prequeue for processing in process
1354 * context, if a reader task is waiting. Apparently, this exciting
1355 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1356 * failed somewhere. Latency? Burstiness? Well, at least now we will
1357 * see, why it failed. 8)8) --ANK
1359 * NOTE: is this not too big to inline?
1361 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1363 struct tcp_sock *tp = tcp_sk(sk);
1365 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1366 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1367 tp->ucopy.memory += skb->truesize;
1368 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1369 struct sk_buff *skb1;
1371 BUG_ON(sock_owned_by_user(sk));
1373 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1374 sk->sk_backlog_rcv(sk, skb1);
1375 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1378 tp->ucopy.memory = 0;
1379 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1380 wake_up_interruptible(sk->sk_sleep);
1381 if (!tcp_ack_scheduled(tp))
1382 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1393 static const char *statename[]={
1394 "Unused","Established","Syn Sent","Syn Recv",
1395 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1396 "Close Wait","Last ACK","Listen","Closing"
1400 static __inline__ void tcp_set_state(struct sock *sk, int state)
1402 int oldstate = sk->sk_state;
1405 case TCP_ESTABLISHED:
1406 if (oldstate != TCP_ESTABLISHED)
1407 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1411 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1412 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1414 sk->sk_prot->unhash(sk);
1415 if (tcp_sk(sk)->bind_hash &&
1416 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1420 if (oldstate==TCP_ESTABLISHED)
1421 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1424 /* Change state AFTER socket is unhashed to avoid closed
1425 * socket sitting in hash tables.
1427 sk->sk_state = state;
1430 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1434 static __inline__ void tcp_done(struct sock *sk)
1436 tcp_set_state(sk, TCP_CLOSE);
1437 tcp_clear_xmit_timers(sk);
1439 sk->sk_shutdown = SHUTDOWN_MASK;
1441 if (!sock_flag(sk, SOCK_DEAD))
1442 sk->sk_state_change(sk);
1444 tcp_destroy_sock(sk);
1447 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
1450 rx_opt->eff_sacks = 0;
1451 rx_opt->num_sacks = 0;
1454 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1456 if (tp->rx_opt.tstamp_ok) {
1457 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1458 (TCPOPT_NOP << 16) |
1459 (TCPOPT_TIMESTAMP << 8) |
1461 *ptr++ = htonl(tstamp);
1462 *ptr++ = htonl(tp->rx_opt.ts_recent);
1464 if (tp->rx_opt.eff_sacks) {
1465 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1468 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1469 (TCPOPT_NOP << 16) |
1470 (TCPOPT_SACK << 8) |
1471 (TCPOLEN_SACK_BASE +
1472 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1473 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1474 *ptr++ = htonl(sp[this_sack].start_seq);
1475 *ptr++ = htonl(sp[this_sack].end_seq);
1477 if (tp->rx_opt.dsack) {
1478 tp->rx_opt.dsack = 0;
1479 tp->rx_opt.eff_sacks--;
1484 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1485 * If this is every changed make sure to change the definition of
1486 * MAX_SYN_SIZE to match the new maximum number of options that you
1489 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1490 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1492 /* We always get an MSS option.
1493 * The option bytes which will be seen in normal data
1494 * packets should timestamps be used, must be in the MSS
1495 * advertised. But we subtract them from tp->mss_cache so
1496 * that calculations in tcp_sendmsg are simpler etc.
1497 * So account for this fact here if necessary. If we
1498 * don't do this correctly, as a receiver we won't
1499 * recognize data packets as being full sized when we
1500 * should, and thus we won't abide by the delayed ACK
1502 * SACKs don't matter, we never delay an ACK when we
1503 * have any of those going out.
1505 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1508 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1509 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1511 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1512 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1513 *ptr++ = htonl(tstamp); /* TSVAL */
1514 *ptr++ = htonl(ts_recent); /* TSECR */
1516 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1517 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1519 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1522 /* Determine a window scaling and initial window to offer. */
1523 extern void tcp_select_initial_window(int __space, __u32 mss,
1524 __u32 *rcv_wnd, __u32 *window_clamp,
1525 int wscale_ok, __u8 *rcv_wscale);
1527 static inline int tcp_win_from_space(int space)
1529 return sysctl_tcp_adv_win_scale<=0 ?
1530 (space>>(-sysctl_tcp_adv_win_scale)) :
1531 space - (space>>sysctl_tcp_adv_win_scale);
1534 /* Note: caller must be prepared to deal with negative returns */
1535 static inline int tcp_space(const struct sock *sk)
1537 return tcp_win_from_space(sk->sk_rcvbuf -
1538 atomic_read(&sk->sk_rmem_alloc));
1541 static inline int tcp_full_space(const struct sock *sk)
1543 return tcp_win_from_space(sk->sk_rcvbuf);
1546 static inline void tcp_acceptq_queue(struct sock *sk, struct request_sock *req,
1549 reqsk_queue_add(&tcp_sk(sk)->accept_queue, req, sk, child);
1553 tcp_synq_removed(struct sock *sk, struct request_sock *req)
1555 if (reqsk_queue_removed(&tcp_sk(sk)->accept_queue, req) == 0)
1556 tcp_delete_keepalive_timer(sk);
1559 static inline void tcp_synq_added(struct sock *sk)
1561 if (reqsk_queue_added(&tcp_sk(sk)->accept_queue) == 0)
1562 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1565 static inline int tcp_synq_len(struct sock *sk)
1567 return reqsk_queue_len(&tcp_sk(sk)->accept_queue);
1570 static inline int tcp_synq_young(struct sock *sk)
1572 return reqsk_queue_len_young(&tcp_sk(sk)->accept_queue);
1575 static inline int tcp_synq_is_full(struct sock *sk)
1577 return reqsk_queue_is_full(&tcp_sk(sk)->accept_queue);
1580 static inline void tcp_synq_unlink(struct tcp_sock *tp, struct request_sock *req,
1581 struct request_sock **prev)
1583 reqsk_queue_unlink(&tp->accept_queue, req, prev);
1586 static inline void tcp_synq_drop(struct sock *sk, struct request_sock *req,
1587 struct request_sock **prev)
1589 tcp_synq_unlink(tcp_sk(sk), req, prev);
1590 tcp_synq_removed(sk, req);
1594 static __inline__ void tcp_openreq_init(struct request_sock *req,
1595 struct tcp_options_received *rx_opt,
1596 struct sk_buff *skb)
1598 struct inet_request_sock *ireq = inet_rsk(req);
1600 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1601 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1602 req->mss = rx_opt->mss_clamp;
1603 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1604 ireq->tstamp_ok = rx_opt->tstamp_ok;
1605 ireq->sack_ok = rx_opt->sack_ok;
1606 ireq->snd_wscale = rx_opt->snd_wscale;
1607 ireq->wscale_ok = rx_opt->wscale_ok;
1610 ireq->rmt_port = skb->h.th->source;
1613 extern void tcp_enter_memory_pressure(void);
1615 extern void tcp_listen_wlock(void);
1617 /* - We may sleep inside this lock.
1618 * - If sleeping is not required (or called from BH),
1619 * use plain read_(un)lock(&tcp_lhash_lock).
1622 static inline void tcp_listen_lock(void)
1624 /* read_lock synchronizes to candidates to writers */
1625 read_lock(&tcp_lhash_lock);
1626 atomic_inc(&tcp_lhash_users);
1627 read_unlock(&tcp_lhash_lock);
1630 static inline void tcp_listen_unlock(void)
1632 if (atomic_dec_and_test(&tcp_lhash_users))
1633 wake_up(&tcp_lhash_wait);
1636 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1638 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1641 static inline int keepalive_time_when(const struct tcp_sock *tp)
1643 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1646 static inline int tcp_fin_time(const struct tcp_sock *tp)
1648 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1650 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1651 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1656 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1658 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1660 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1663 /* RST segments are not recommended to carry timestamp,
1664 and, if they do, it is recommended to ignore PAWS because
1665 "their cleanup function should take precedence over timestamps."
1666 Certainly, it is mistake. It is necessary to understand the reasons
1667 of this constraint to relax it: if peer reboots, clock may go
1668 out-of-sync and half-open connections will not be reset.
1669 Actually, the problem would be not existing if all
1670 the implementations followed draft about maintaining clock
1671 via reboots. Linux-2.2 DOES NOT!
1673 However, we can relax time bounds for RST segments to MSL.
1675 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1680 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst)
1682 sk->sk_route_caps = dst->dev->features;
1683 if (sk->sk_route_caps & NETIF_F_TSO) {
1684 if (sock_flag(sk, SOCK_NO_LARGESEND) || dst->header_len)
1685 sk->sk_route_caps &= ~NETIF_F_TSO;
1689 #define TCP_CHECK_TIMER(sk) do { } while (0)
1691 static inline int tcp_use_frto(const struct sock *sk)
1693 const struct tcp_sock *tp = tcp_sk(sk);
1695 /* F-RTO must be activated in sysctl and there must be some
1696 * unsent new data, and the advertised window should allow
1699 return (sysctl_tcp_frto && sk->sk_send_head &&
1700 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1701 tp->snd_una + tp->snd_wnd));
1704 static inline void tcp_mib_init(void)
1707 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1708 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1709 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1710 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1714 enum tcp_seq_states {
1715 TCP_SEQ_STATE_LISTENING,
1716 TCP_SEQ_STATE_OPENREQ,
1717 TCP_SEQ_STATE_ESTABLISHED,
1718 TCP_SEQ_STATE_TIME_WAIT,
1721 struct tcp_seq_afinfo {
1722 struct module *owner;
1725 int (*seq_show) (struct seq_file *m, void *v);
1726 struct file_operations *seq_fops;
1729 struct tcp_iter_state {
1731 enum tcp_seq_states state;
1732 struct sock *syn_wait_sk;
1733 int bucket, sbucket, num, uid;
1734 struct seq_operations seq_ops;
1737 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1738 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);