]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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. | |
5 | * | |
6 | * Definitions for the TCP module. | |
7 | * | |
8 | * Version: @(#)tcp.h 1.0.5 05/23/93 | |
9 | * | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * | |
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. | |
17 | */ | |
18 | #ifndef _TCP_H | |
19 | #define _TCP_H | |
20 | ||
21 | #define TCP_DEBUG 1 | |
22 | #define FASTRETRANS_DEBUG 1 | |
23 | ||
24 | /* Cancel timers, when they are not required. */ | |
25 | #undef TCP_CLEAR_TIMERS | |
26 | ||
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> | |
2e6599cb | 34 | #include <net/request_sock.h> |
1da177e4 LT |
35 | #include <net/sock.h> |
36 | #include <net/snmp.h> | |
37 | #include <net/ip.h> | |
38 | #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) | |
39 | #include <linux/ipv6.h> | |
40 | #endif | |
41 | #include <linux/seq_file.h> | |
42 | ||
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. | |
46 | */ | |
47 | struct tcp_ehash_bucket { | |
48 | rwlock_t lock; | |
49 | struct hlist_head chain; | |
50 | } __attribute__((__aligned__(8))); | |
51 | ||
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. */ | |
54 | ||
55 | /* There are a few simple rules, which allow for local port reuse by | |
56 | * an application. In essence: | |
57 | * | |
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 | |
65 | * shared. | |
66 | * Failing this, the port cannot be shared. | |
67 | * | |
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. | |
78 | * | |
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 | |
85 | */ | |
86 | struct tcp_bind_bucket { | |
87 | unsigned short port; | |
88 | signed short fastreuse; | |
89 | struct hlist_node node; | |
90 | struct hlist_head owners; | |
91 | }; | |
92 | ||
93 | #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node) | |
94 | ||
95 | struct tcp_bind_hashbucket { | |
96 | spinlock_t lock; | |
97 | struct hlist_head chain; | |
98 | }; | |
99 | ||
100 | static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head) | |
101 | { | |
102 | return hlist_entry(head->chain.first, struct tcp_bind_bucket, node); | |
103 | } | |
104 | ||
105 | static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head) | |
106 | { | |
107 | return hlist_empty(&head->chain) ? NULL : __tb_head(head); | |
108 | } | |
109 | ||
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: | |
113 | * | |
114 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
115 | * | |
116 | * First half of the table is for sockets not in TIME_WAIT, second half | |
117 | * is for TIME_WAIT sockets only. | |
118 | */ | |
119 | struct tcp_ehash_bucket *__tcp_ehash; | |
120 | ||
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. | |
123 | */ | |
124 | struct tcp_bind_hashbucket *__tcp_bhash; | |
125 | ||
126 | int __tcp_bhash_size; | |
127 | int __tcp_ehash_size; | |
128 | ||
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. | |
132 | */ | |
133 | struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE]; | |
134 | ||
135 | /* All the above members are written once at bootup and | |
136 | * never written again _or_ are predominantly read-access. | |
137 | * | |
138 | * Now align to a new cache line as all the following members | |
139 | * are often dirty. | |
140 | */ | |
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; | |
145 | } tcp_hashinfo; | |
146 | ||
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) | |
156 | ||
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; | |
163 | ||
164 | /* These are AF independent. */ | |
165 | static __inline__ int tcp_bhashfn(__u16 lport) | |
166 | { | |
167 | return (lport & (tcp_bhash_size - 1)); | |
168 | } | |
169 | ||
170 | extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb, | |
171 | unsigned short snum); | |
172 | ||
173 | #if (BITS_PER_LONG == 64) | |
174 | #define TCP_ADDRCMP_ALIGN_BYTES 8 | |
175 | #else | |
176 | #define TCP_ADDRCMP_ALIGN_BYTES 4 | |
177 | #endif | |
178 | ||
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. | |
182 | */ | |
183 | struct tcp_tw_bucket { | |
184 | /* | |
185 | * Now struct sock also uses sock_common, so please just | |
186 | * don't add nothing before this first member (__tw_common) --acme | |
187 | */ | |
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; | |
198 | __u16 tw_sport; | |
199 | /* Socket demultiplex comparisons on incoming packets. */ | |
200 | /* these five are in inet_sock */ | |
201 | __u32 tw_daddr | |
202 | __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES))); | |
203 | __u32 tw_rcv_saddr; | |
204 | __u16 tw_dport; | |
205 | __u16 tw_num; | |
206 | /* And these are ours. */ | |
207 | int tw_hashent; | |
208 | int tw_timeout; | |
209 | __u32 tw_rcv_nxt; | |
210 | __u32 tw_snd_nxt; | |
211 | __u32 tw_rcv_wnd; | |
212 | __u32 tw_ts_recent; | |
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; | |
220 | int tw_v6_ipv6only; | |
221 | #endif | |
222 | }; | |
223 | ||
224 | static __inline__ void tw_add_node(struct tcp_tw_bucket *tw, | |
225 | struct hlist_head *list) | |
226 | { | |
227 | hlist_add_head(&tw->tw_node, list); | |
228 | } | |
229 | ||
230 | static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw, | |
231 | struct hlist_head *list) | |
232 | { | |
233 | hlist_add_head(&tw->tw_bind_node, list); | |
234 | } | |
235 | ||
236 | static inline int tw_dead_hashed(struct tcp_tw_bucket *tw) | |
237 | { | |
238 | return tw->tw_death_node.pprev != NULL; | |
239 | } | |
240 | ||
241 | static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw) | |
242 | { | |
243 | tw->tw_death_node.pprev = NULL; | |
244 | } | |
245 | ||
246 | static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw) | |
247 | { | |
248 | __hlist_del(&tw->tw_death_node); | |
249 | tw_dead_node_init(tw); | |
250 | } | |
251 | ||
252 | static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw) | |
253 | { | |
254 | if (tw_dead_hashed(tw)) { | |
255 | __tw_del_dead_node(tw); | |
256 | return 1; | |
257 | } | |
258 | return 0; | |
259 | } | |
260 | ||
261 | #define tw_for_each(tw, node, head) \ | |
262 | hlist_for_each_entry(tw, node, head, tw_node) | |
263 | ||
264 | #define tw_for_each_inmate(tw, node, jail) \ | |
265 | hlist_for_each_entry(tw, node, jail, tw_death_node) | |
266 | ||
267 | #define tw_for_each_inmate_safe(tw, node, safe, jail) \ | |
268 | hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node) | |
269 | ||
270 | #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk)) | |
271 | ||
272 | static inline u32 tcp_v4_rcv_saddr(const struct sock *sk) | |
273 | { | |
274 | return likely(sk->sk_state != TCP_TIME_WAIT) ? | |
275 | inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr; | |
276 | } | |
277 | ||
278 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
279 | static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk) | |
280 | { | |
281 | return likely(sk->sk_state != TCP_TIME_WAIT) ? | |
282 | &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr; | |
283 | } | |
284 | ||
285 | static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk) | |
286 | { | |
287 | return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL; | |
288 | } | |
289 | ||
290 | #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only) | |
291 | ||
292 | static inline int tcp_v6_ipv6only(const struct sock *sk) | |
293 | { | |
294 | return likely(sk->sk_state != TCP_TIME_WAIT) ? | |
295 | ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk); | |
296 | } | |
297 | #else | |
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 | |
302 | #endif | |
303 | ||
304 | extern kmem_cache_t *tcp_timewait_cachep; | |
305 | ||
306 | static inline void tcp_tw_put(struct tcp_tw_bucket *tw) | |
307 | { | |
308 | if (atomic_dec_and_test(&tw->tw_refcnt)) { | |
309 | #ifdef INET_REFCNT_DEBUG | |
310 | printk(KERN_DEBUG "tw_bucket %p released\n", tw); | |
311 | #endif | |
312 | kmem_cache_free(tcp_timewait_cachep, tw); | |
313 | } | |
314 | } | |
315 | ||
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); | |
320 | ||
321 | ||
322 | /* Socket demux engine toys. */ | |
323 | #ifdef __BIG_ENDIAN | |
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)) | |
329 | #endif | |
330 | ||
331 | #if (BITS_PER_LONG == 64) | |
332 | #ifdef __BIG_ENDIAN | |
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 */ | |
360 | ||
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)))) | |
367 | ||
368 | /* These can have wildcards, don't try too hard. */ | |
369 | static __inline__ int tcp_lhashfn(unsigned short num) | |
370 | { | |
371 | return num & (TCP_LHTABLE_SIZE - 1); | |
372 | } | |
373 | ||
374 | static __inline__ int tcp_sk_listen_hashfn(struct sock *sk) | |
375 | { | |
376 | return tcp_lhashfn(inet_sk(sk)->num); | |
377 | } | |
378 | ||
379 | #define MAX_TCP_HEADER (128 + MAX_HEADER) | |
380 | ||
381 | /* | |
382 | * Never offer a window over 32767 without using window scaling. Some | |
383 | * poor stacks do signed 16bit maths! | |
384 | */ | |
385 | #define MAX_TCP_WINDOW 32767U | |
386 | ||
387 | /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ | |
388 | #define TCP_MIN_MSS 88U | |
389 | ||
390 | /* Minimal RCV_MSS. */ | |
391 | #define TCP_MIN_RCVMSS 536U | |
392 | ||
393 | /* After receiving this amount of duplicate ACKs fast retransmit starts. */ | |
394 | #define TCP_FASTRETRANS_THRESH 3 | |
395 | ||
396 | /* Maximal reordering. */ | |
397 | #define TCP_MAX_REORDERING 127 | |
398 | ||
399 | /* Maximal number of ACKs sent quickly to accelerate slow-start. */ | |
400 | #define TCP_MAX_QUICKACKS 16U | |
401 | ||
402 | /* urg_data states */ | |
403 | #define TCP_URG_VALID 0x0100 | |
404 | #define TCP_URG_NOTYET 0x0200 | |
405 | #define TCP_URG_READ 0x0400 | |
406 | ||
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. | |
412 | */ | |
413 | ||
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. | |
419 | */ | |
420 | ||
421 | #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a | |
422 | * connection: ~180sec is RFC minumum */ | |
423 | ||
424 | #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a | |
425 | * connection: ~180sec is RFC minumum */ | |
426 | ||
427 | ||
428 | #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned | |
429 | * socket. 7 is ~50sec-16min. | |
430 | */ | |
431 | ||
432 | ||
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 | |
439 | * TIME-WAIT timer. | |
440 | */ | |
441 | ||
442 | #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ | |
443 | #if HZ >= 100 | |
444 | #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ | |
445 | #define TCP_ATO_MIN ((unsigned)(HZ/25)) | |
446 | #else | |
447 | #define TCP_DELACK_MIN 4U | |
448 | #define TCP_ATO_MIN 4U | |
449 | #endif | |
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 */ | |
453 | ||
454 | #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes | |
455 | * for local resources. | |
456 | */ | |
457 | ||
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) | |
461 | ||
462 | #define MAX_TCP_KEEPIDLE 32767 | |
463 | #define MAX_TCP_KEEPINTVL 32767 | |
464 | #define MAX_TCP_KEEPCNT 127 | |
465 | #define MAX_TCP_SYNCNT 127 | |
466 | ||
467 | #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ | |
468 | #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */ | |
469 | ||
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. | |
476 | */ | |
477 | #define TCP_PAWS_WINDOW 1 /* Replay window for per-host | |
478 | * timestamps. It must be less than | |
479 | * minimal timewait lifetime. | |
480 | */ | |
481 | ||
482 | #define TCP_TW_RECYCLE_SLOTS_LOG 5 | |
483 | #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG) | |
484 | ||
485 | /* If time > 4sec, it is "slow" path, no recycling is required, | |
486 | so that we select tick to get range about 4 seconds. | |
487 | */ | |
488 | ||
489 | #if HZ <= 16 || HZ > 4096 | |
490 | # error Unsupported: HZ <= 16 or HZ > 4096 | |
491 | #elif HZ <= 32 | |
492 | # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
493 | #elif HZ <= 64 | |
494 | # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
495 | #elif HZ <= 128 | |
496 | # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
497 | #elif HZ <= 256 | |
498 | # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
499 | #elif HZ <= 512 | |
500 | # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
501 | #elif HZ <= 1024 | |
502 | # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
503 | #elif HZ <= 2048 | |
504 | # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
505 | #else | |
506 | # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
507 | #endif | |
1da177e4 LT |
508 | /* |
509 | * TCP option | |
510 | */ | |
511 | ||
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 */ | |
519 | ||
520 | /* | |
521 | * TCP option lengths | |
522 | */ | |
523 | ||
524 | #define TCPOLEN_MSS 4 | |
525 | #define TCPOLEN_WINDOW 3 | |
526 | #define TCPOLEN_SACK_PERM 2 | |
527 | #define TCPOLEN_TIMESTAMP 10 | |
528 | ||
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 | |
536 | ||
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 */ | |
541 | ||
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 */ | |
546 | ||
547 | /* sysctl variables for tcp */ | |
1da177e4 LT |
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; | |
1da177e4 | 580 | extern int sysctl_tcp_nometrics_save; |
1da177e4 LT |
581 | extern int sysctl_tcp_moderate_rcvbuf; |
582 | extern int sysctl_tcp_tso_win_divisor; | |
583 | ||
584 | extern atomic_t tcp_memory_allocated; | |
585 | extern atomic_t tcp_sockets_allocated; | |
586 | extern int tcp_memory_pressure; | |
587 | ||
1da177e4 LT |
588 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
589 | #define TCP_INET_FAMILY(fam) ((fam) == AF_INET) | |
590 | #else | |
591 | #define TCP_INET_FAMILY(fam) 1 | |
592 | #endif | |
593 | ||
594 | /* | |
595 | * Pointers to address related TCP functions | |
596 | * (i.e. things that depend on the address family) | |
597 | */ | |
598 | ||
599 | struct tcp_func { | |
600 | int (*queue_xmit) (struct sk_buff *skb, | |
601 | int ipfragok); | |
602 | ||
603 | void (*send_check) (struct sock *sk, | |
604 | struct tcphdr *th, | |
605 | int len, | |
606 | struct sk_buff *skb); | |
607 | ||
608 | int (*rebuild_header) (struct sock *sk); | |
609 | ||
610 | int (*conn_request) (struct sock *sk, | |
611 | struct sk_buff *skb); | |
612 | ||
613 | struct sock * (*syn_recv_sock) (struct sock *sk, | |
614 | struct sk_buff *skb, | |
60236fdd | 615 | struct request_sock *req, |
1da177e4 LT |
616 | struct dst_entry *dst); |
617 | ||
618 | int (*remember_stamp) (struct sock *sk); | |
619 | ||
620 | __u16 net_header_len; | |
621 | ||
622 | int (*setsockopt) (struct sock *sk, | |
623 | int level, | |
624 | int optname, | |
625 | char __user *optval, | |
626 | int optlen); | |
627 | ||
628 | int (*getsockopt) (struct sock *sk, | |
629 | int level, | |
630 | int optname, | |
631 | char __user *optval, | |
632 | int __user *optlen); | |
633 | ||
634 | ||
635 | void (*addr2sockaddr) (struct sock *sk, | |
636 | struct sockaddr *); | |
637 | ||
638 | int sockaddr_len; | |
639 | }; | |
640 | ||
641 | /* | |
642 | * The next routines deal with comparing 32 bit unsigned ints | |
643 | * and worry about wraparound (automatic with unsigned arithmetic). | |
644 | */ | |
645 | ||
646 | static inline int before(__u32 seq1, __u32 seq2) | |
647 | { | |
648 | return (__s32)(seq1-seq2) < 0; | |
649 | } | |
650 | ||
651 | static inline int after(__u32 seq1, __u32 seq2) | |
652 | { | |
653 | return (__s32)(seq2-seq1) < 0; | |
654 | } | |
655 | ||
656 | ||
657 | /* is s2<=s1<=s3 ? */ | |
658 | static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) | |
659 | { | |
660 | return seq3 - seq2 >= seq1 - seq2; | |
661 | } | |
662 | ||
663 | ||
664 | extern struct proto tcp_prot; | |
665 | ||
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) | |
673 | ||
674 | extern void tcp_put_port(struct sock *sk); | |
675 | extern void tcp_inherit_port(struct sock *sk, struct sock *child); | |
676 | ||
677 | extern void tcp_v4_err(struct sk_buff *skb, u32); | |
678 | ||
679 | extern void tcp_shutdown (struct sock *sk, int how); | |
680 | ||
681 | extern int tcp_v4_rcv(struct sk_buff *skb); | |
682 | ||
683 | extern int tcp_v4_remember_stamp(struct sock *sk); | |
684 | ||
685 | extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw); | |
686 | ||
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); | |
690 | ||
691 | extern int tcp_ioctl(struct sock *sk, | |
692 | int cmd, | |
693 | unsigned long arg); | |
694 | ||
695 | extern int tcp_rcv_state_process(struct sock *sk, | |
696 | struct sk_buff *skb, | |
697 | struct tcphdr *th, | |
698 | unsigned len); | |
699 | ||
700 | extern int tcp_rcv_established(struct sock *sk, | |
701 | struct sk_buff *skb, | |
702 | struct tcphdr *th, | |
703 | unsigned len); | |
704 | ||
705 | extern void tcp_rcv_space_adjust(struct sock *sk); | |
706 | ||
707 | enum tcp_ack_state_t | |
708 | { | |
709 | TCP_ACK_SCHED = 1, | |
710 | TCP_ACK_TIMER = 2, | |
711 | TCP_ACK_PUSHED= 4 | |
712 | }; | |
713 | ||
714 | static inline void tcp_schedule_ack(struct tcp_sock *tp) | |
715 | { | |
716 | tp->ack.pending |= TCP_ACK_SCHED; | |
717 | } | |
718 | ||
719 | static inline int tcp_ack_scheduled(struct tcp_sock *tp) | |
720 | { | |
721 | return tp->ack.pending&TCP_ACK_SCHED; | |
722 | } | |
723 | ||
fc6415bc | 724 | static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp, unsigned int pkts) |
1da177e4 | 725 | { |
fc6415bc DM |
726 | if (tp->ack.quick) { |
727 | if (pkts >= tp->ack.quick) { | |
728 | tp->ack.quick = 0; | |
729 | ||
730 | /* Leaving quickack mode we deflate ATO. */ | |
731 | tp->ack.ato = TCP_ATO_MIN; | |
732 | } else | |
733 | tp->ack.quick -= pkts; | |
1da177e4 LT |
734 | } |
735 | } | |
736 | ||
737 | extern void tcp_enter_quickack_mode(struct tcp_sock *tp); | |
738 | ||
739 | static __inline__ void tcp_delack_init(struct tcp_sock *tp) | |
740 | { | |
741 | memset(&tp->ack, 0, sizeof(tp->ack)); | |
742 | } | |
743 | ||
744 | static inline void tcp_clear_options(struct tcp_options_received *rx_opt) | |
745 | { | |
746 | rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; | |
747 | } | |
748 | ||
749 | enum tcp_tw_status | |
750 | { | |
751 | TCP_TW_SUCCESS = 0, | |
752 | TCP_TW_RST = 1, | |
753 | TCP_TW_ACK = 2, | |
754 | TCP_TW_SYN = 3 | |
755 | }; | |
756 | ||
757 | ||
758 | extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw, | |
759 | struct sk_buff *skb, | |
760 | struct tcphdr *th, | |
761 | unsigned len); | |
762 | ||
763 | extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, | |
60236fdd ACM |
764 | struct request_sock *req, |
765 | struct request_sock **prev); | |
1da177e4 LT |
766 | extern int tcp_child_process(struct sock *parent, |
767 | struct sock *child, | |
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); | |
773 | ||
774 | extern void tcp_close(struct sock *sk, | |
775 | long timeout); | |
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); | |
778 | ||
779 | extern int tcp_getsockopt(struct sock *sk, int level, | |
780 | int optname, | |
781 | char __user *optval, | |
782 | int __user *optlen); | |
783 | extern int tcp_setsockopt(struct sock *sk, int level, | |
784 | int optname, char __user *optval, | |
785 | int optlen); | |
786 | extern void tcp_set_keepalive(struct sock *sk, int val); | |
787 | extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, | |
788 | struct msghdr *msg, | |
789 | size_t len, int nonblock, | |
790 | int flags, int *addr_len); | |
791 | ||
792 | extern int tcp_listen_start(struct sock *sk); | |
793 | ||
794 | extern void tcp_parse_options(struct sk_buff *skb, | |
795 | struct tcp_options_received *opt_rx, | |
796 | int estab); | |
797 | ||
798 | /* | |
799 | * TCP v4 functions exported for the inet6 API | |
800 | */ | |
801 | ||
802 | extern int tcp_v4_rebuild_header(struct sock *sk); | |
803 | ||
804 | extern int tcp_v4_build_header(struct sock *sk, | |
805 | struct sk_buff *skb); | |
806 | ||
807 | extern void tcp_v4_send_check(struct sock *sk, | |
808 | struct tcphdr *th, int len, | |
809 | struct sk_buff *skb); | |
810 | ||
811 | extern int tcp_v4_conn_request(struct sock *sk, | |
812 | struct sk_buff *skb); | |
813 | ||
814 | extern struct sock * tcp_create_openreq_child(struct sock *sk, | |
60236fdd | 815 | struct request_sock *req, |
1da177e4 LT |
816 | struct sk_buff *skb); |
817 | ||
818 | extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, | |
819 | struct sk_buff *skb, | |
60236fdd | 820 | struct request_sock *req, |
1da177e4 LT |
821 | struct dst_entry *dst); |
822 | ||
823 | extern int tcp_v4_do_rcv(struct sock *sk, | |
824 | struct sk_buff *skb); | |
825 | ||
826 | extern int tcp_v4_connect(struct sock *sk, | |
827 | struct sockaddr *uaddr, | |
828 | int addr_len); | |
829 | ||
830 | extern int tcp_connect(struct sock *sk); | |
831 | ||
832 | extern struct sk_buff * tcp_make_synack(struct sock *sk, | |
833 | struct dst_entry *dst, | |
60236fdd | 834 | struct request_sock *req); |
1da177e4 LT |
835 | |
836 | extern int tcp_disconnect(struct sock *sk, int flags); | |
837 | ||
838 | extern void tcp_unhash(struct sock *sk); | |
839 | ||
840 | extern int tcp_v4_hash_connecting(struct sock *sk); | |
841 | ||
842 | ||
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, | |
847 | __u16 *mss); | |
848 | ||
849 | /* tcp_output.c */ | |
850 | ||
851 | extern int tcp_write_xmit(struct sock *, int nonagle); | |
852 | extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); | |
853 | extern void tcp_xmit_retransmit_queue(struct sock *); | |
854 | extern void tcp_simple_retransmit(struct sock *); | |
855 | extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); | |
856 | ||
857 | extern void tcp_send_probe0(struct sock *); | |
858 | extern void tcp_send_partial(struct sock *); | |
859 | extern int tcp_write_wakeup(struct sock *); | |
860 | extern void tcp_send_fin(struct sock *sk); | |
861 | extern void tcp_send_active_reset(struct sock *sk, int priority); | |
862 | extern int tcp_send_synack(struct sock *); | |
863 | extern void tcp_push_one(struct sock *, unsigned mss_now); | |
864 | extern void tcp_send_ack(struct sock *sk); | |
865 | extern void tcp_send_delayed_ack(struct sock *sk); | |
866 | ||
867 | /* tcp_timer.c */ | |
868 | extern void tcp_init_xmit_timers(struct sock *); | |
869 | extern void tcp_clear_xmit_timers(struct sock *); | |
870 | ||
871 | extern void tcp_delete_keepalive_timer(struct sock *); | |
872 | extern void tcp_reset_keepalive_timer(struct sock *, unsigned long); | |
873 | extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); | |
874 | extern unsigned int tcp_current_mss(struct sock *sk, int large); | |
875 | ||
876 | #ifdef TCP_DEBUG | |
877 | extern const char tcp_timer_bug_msg[]; | |
878 | #endif | |
879 | ||
880 | /* tcp_diag.c */ | |
881 | extern void tcp_get_info(struct sock *, struct tcp_info *); | |
882 | ||
883 | /* Read 'sendfile()'-style from a TCP socket */ | |
884 | typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, | |
885 | unsigned int, size_t); | |
886 | extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
887 | sk_read_actor_t recv_actor); | |
888 | ||
889 | static inline void tcp_clear_xmit_timer(struct sock *sk, int what) | |
890 | { | |
891 | struct tcp_sock *tp = tcp_sk(sk); | |
892 | ||
893 | switch (what) { | |
894 | case TCP_TIME_RETRANS: | |
895 | case TCP_TIME_PROBE0: | |
896 | tp->pending = 0; | |
897 | ||
898 | #ifdef TCP_CLEAR_TIMERS | |
899 | sk_stop_timer(sk, &tp->retransmit_timer); | |
900 | #endif | |
901 | break; | |
902 | case TCP_TIME_DACK: | |
903 | tp->ack.blocked = 0; | |
904 | tp->ack.pending = 0; | |
905 | ||
906 | #ifdef TCP_CLEAR_TIMERS | |
907 | sk_stop_timer(sk, &tp->delack_timer); | |
908 | #endif | |
909 | break; | |
910 | default: | |
911 | #ifdef TCP_DEBUG | |
912 | printk(tcp_timer_bug_msg); | |
913 | #endif | |
914 | return; | |
915 | }; | |
916 | ||
917 | } | |
918 | ||
919 | /* | |
920 | * Reset the retransmission timer | |
921 | */ | |
922 | static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when) | |
923 | { | |
924 | struct tcp_sock *tp = tcp_sk(sk); | |
925 | ||
926 | if (when > TCP_RTO_MAX) { | |
927 | #ifdef TCP_DEBUG | |
928 | printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr()); | |
929 | #endif | |
930 | when = TCP_RTO_MAX; | |
931 | } | |
932 | ||
933 | switch (what) { | |
934 | case TCP_TIME_RETRANS: | |
935 | case TCP_TIME_PROBE0: | |
936 | tp->pending = what; | |
937 | tp->timeout = jiffies+when; | |
938 | sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout); | |
939 | break; | |
940 | ||
941 | case TCP_TIME_DACK: | |
942 | tp->ack.pending |= TCP_ACK_TIMER; | |
943 | tp->ack.timeout = jiffies+when; | |
944 | sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout); | |
945 | break; | |
946 | ||
947 | default: | |
948 | #ifdef TCP_DEBUG | |
949 | printk(tcp_timer_bug_msg); | |
950 | #endif | |
951 | return; | |
952 | }; | |
953 | } | |
954 | ||
955 | /* Initialize RCV_MSS value. | |
956 | * RCV_MSS is an our guess about MSS used by the peer. | |
957 | * We haven't any direct information about the MSS. | |
958 | * It's better to underestimate the RCV_MSS rather than overestimate. | |
959 | * Overestimations make us ACKing less frequently than needed. | |
960 | * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss(). | |
961 | */ | |
962 | ||
963 | static inline void tcp_initialize_rcv_mss(struct sock *sk) | |
964 | { | |
965 | struct tcp_sock *tp = tcp_sk(sk); | |
966 | unsigned int hint = min(tp->advmss, tp->mss_cache_std); | |
967 | ||
968 | hint = min(hint, tp->rcv_wnd/2); | |
969 | hint = min(hint, TCP_MIN_RCVMSS); | |
970 | hint = max(hint, TCP_MIN_MSS); | |
971 | ||
972 | tp->ack.rcv_mss = hint; | |
973 | } | |
974 | ||
975 | static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) | |
976 | { | |
977 | tp->pred_flags = htonl((tp->tcp_header_len << 26) | | |
978 | ntohl(TCP_FLAG_ACK) | | |
979 | snd_wnd); | |
980 | } | |
981 | ||
982 | static __inline__ void tcp_fast_path_on(struct tcp_sock *tp) | |
983 | { | |
984 | __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); | |
985 | } | |
986 | ||
987 | static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp) | |
988 | { | |
989 | if (skb_queue_len(&tp->out_of_order_queue) == 0 && | |
990 | tp->rcv_wnd && | |
991 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && | |
992 | !tp->urg_data) | |
993 | tcp_fast_path_on(tp); | |
994 | } | |
995 | ||
996 | /* Compute the actual receive window we are currently advertising. | |
997 | * Rcv_nxt can be after the window if our peer push more data | |
998 | * than the offered window. | |
999 | */ | |
1000 | static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp) | |
1001 | { | |
1002 | s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; | |
1003 | ||
1004 | if (win < 0) | |
1005 | win = 0; | |
1006 | return (u32) win; | |
1007 | } | |
1008 | ||
1009 | /* Choose a new window, without checks for shrinking, and without | |
1010 | * scaling applied to the result. The caller does these things | |
1011 | * if necessary. This is a "raw" window selection. | |
1012 | */ | |
1013 | extern u32 __tcp_select_window(struct sock *sk); | |
1014 | ||
1015 | /* TCP timestamps are only 32-bits, this causes a slight | |
1016 | * complication on 64-bit systems since we store a snapshot | |
1017 | * of jiffies in the buffer control blocks below. We decidely | |
1018 | * only use of the low 32-bits of jiffies and hide the ugly | |
1019 | * casts with the following macro. | |
1020 | */ | |
1021 | #define tcp_time_stamp ((__u32)(jiffies)) | |
1022 | ||
1023 | /* This is what the send packet queueing engine uses to pass | |
1024 | * TCP per-packet control information to the transmission | |
1025 | * code. We also store the host-order sequence numbers in | |
1026 | * here too. This is 36 bytes on 32-bit architectures, | |
1027 | * 40 bytes on 64-bit machines, if this grows please adjust | |
1028 | * skbuff.h:skbuff->cb[xxx] size appropriately. | |
1029 | */ | |
1030 | struct tcp_skb_cb { | |
1031 | union { | |
1032 | struct inet_skb_parm h4; | |
1033 | #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) | |
1034 | struct inet6_skb_parm h6; | |
1035 | #endif | |
1036 | } header; /* For incoming frames */ | |
1037 | __u32 seq; /* Starting sequence number */ | |
1038 | __u32 end_seq; /* SEQ + FIN + SYN + datalen */ | |
1039 | __u32 when; /* used to compute rtt's */ | |
1040 | __u8 flags; /* TCP header flags. */ | |
1041 | ||
1042 | /* NOTE: These must match up to the flags byte in a | |
1043 | * real TCP header. | |
1044 | */ | |
1045 | #define TCPCB_FLAG_FIN 0x01 | |
1046 | #define TCPCB_FLAG_SYN 0x02 | |
1047 | #define TCPCB_FLAG_RST 0x04 | |
1048 | #define TCPCB_FLAG_PSH 0x08 | |
1049 | #define TCPCB_FLAG_ACK 0x10 | |
1050 | #define TCPCB_FLAG_URG 0x20 | |
1051 | #define TCPCB_FLAG_ECE 0x40 | |
1052 | #define TCPCB_FLAG_CWR 0x80 | |
1053 | ||
1054 | __u8 sacked; /* State flags for SACK/FACK. */ | |
1055 | #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ | |
1056 | #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ | |
1057 | #define TCPCB_LOST 0x04 /* SKB is lost */ | |
1058 | #define TCPCB_TAGBITS 0x07 /* All tag bits */ | |
1059 | ||
1060 | #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ | |
1061 | #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) | |
1062 | ||
1063 | #define TCPCB_URG 0x20 /* Urgent pointer advenced here */ | |
1064 | ||
1065 | #define TCPCB_AT_TAIL (TCPCB_URG) | |
1066 | ||
1067 | __u16 urg_ptr; /* Valid w/URG flags is set. */ | |
1068 | __u32 ack_seq; /* Sequence number ACK'd */ | |
1069 | }; | |
1070 | ||
1071 | #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) | |
1072 | ||
1073 | #include <net/tcp_ecn.h> | |
1074 | ||
1075 | /* Due to TSO, an SKB can be composed of multiple actual | |
1076 | * packets. To keep these tracked properly, we use this. | |
1077 | */ | |
1078 | static inline int tcp_skb_pcount(const struct sk_buff *skb) | |
1079 | { | |
1080 | return skb_shinfo(skb)->tso_segs; | |
1081 | } | |
1082 | ||
1083 | /* This is valid iff tcp_skb_pcount() > 1. */ | |
1084 | static inline int tcp_skb_mss(const struct sk_buff *skb) | |
1085 | { | |
1086 | return skb_shinfo(skb)->tso_size; | |
1087 | } | |
1088 | ||
1089 | static inline void tcp_dec_pcount_approx(__u32 *count, | |
1090 | const struct sk_buff *skb) | |
1091 | { | |
1092 | if (*count) { | |
1093 | *count -= tcp_skb_pcount(skb); | |
1094 | if ((int)*count < 0) | |
1095 | *count = 0; | |
1096 | } | |
1097 | } | |
1098 | ||
1099 | static inline void tcp_packets_out_inc(struct sock *sk, | |
1100 | struct tcp_sock *tp, | |
1101 | const struct sk_buff *skb) | |
1102 | { | |
1103 | int orig = tp->packets_out; | |
1104 | ||
1105 | tp->packets_out += tcp_skb_pcount(skb); | |
1106 | if (!orig) | |
1107 | tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto); | |
1108 | } | |
1109 | ||
1110 | static inline void tcp_packets_out_dec(struct tcp_sock *tp, | |
1111 | const struct sk_buff *skb) | |
1112 | { | |
1113 | tp->packets_out -= tcp_skb_pcount(skb); | |
1114 | } | |
1115 | ||
317a76f9 SH |
1116 | /* Events passed to congestion control interface */ |
1117 | enum tcp_ca_event { | |
1118 | CA_EVENT_TX_START, /* first transmit when no packets in flight */ | |
1119 | CA_EVENT_CWND_RESTART, /* congestion window restart */ | |
1120 | CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ | |
1121 | CA_EVENT_FRTO, /* fast recovery timeout */ | |
1122 | CA_EVENT_LOSS, /* loss timeout */ | |
1123 | CA_EVENT_FAST_ACK, /* in sequence ack */ | |
1124 | CA_EVENT_SLOW_ACK, /* other ack */ | |
1125 | }; | |
1126 | ||
1127 | /* | |
1128 | * Interface for adding new TCP congestion control handlers | |
1129 | */ | |
1130 | #define TCP_CA_NAME_MAX 16 | |
1131 | struct tcp_congestion_ops { | |
1132 | struct list_head list; | |
1133 | ||
1134 | /* initialize private data (optional) */ | |
1135 | void (*init)(struct tcp_sock *tp); | |
1136 | /* cleanup private data (optional) */ | |
1137 | void (*release)(struct tcp_sock *tp); | |
1138 | ||
1139 | /* return slow start threshold (required) */ | |
1140 | u32 (*ssthresh)(struct tcp_sock *tp); | |
1141 | /* lower bound for congestion window (optional) */ | |
1142 | u32 (*min_cwnd)(struct tcp_sock *tp); | |
1143 | /* do new cwnd calculation (required) */ | |
1144 | void (*cong_avoid)(struct tcp_sock *tp, u32 ack, | |
1145 | u32 rtt, u32 in_flight, int good_ack); | |
1146 | /* round trip time sample per acked packet (optional) */ | |
1147 | void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt); | |
1148 | /* call before changing ca_state (optional) */ | |
1149 | void (*set_state)(struct tcp_sock *tp, u8 new_state); | |
1150 | /* call when cwnd event occurs (optional) */ | |
1151 | void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev); | |
1152 | /* new value of cwnd after loss (optional) */ | |
1153 | u32 (*undo_cwnd)(struct tcp_sock *tp); | |
1154 | /* hook for packet ack accounting (optional) */ | |
1155 | void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked); | |
1156 | /* get info for tcp_diag (optional) */ | |
1157 | void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb); | |
1158 | ||
1159 | char name[TCP_CA_NAME_MAX]; | |
1160 | struct module *owner; | |
1161 | }; | |
1162 | ||
1163 | extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); | |
1164 | extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); | |
1165 | ||
1166 | extern void tcp_init_congestion_control(struct tcp_sock *tp); | |
1167 | extern void tcp_cleanup_congestion_control(struct tcp_sock *tp); | |
1168 | extern int tcp_set_default_congestion_control(const char *name); | |
1169 | extern void tcp_get_default_congestion_control(char *name); | |
5f8ef48d | 1170 | extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name); |
317a76f9 | 1171 | |
5f8ef48d | 1172 | extern struct tcp_congestion_ops tcp_init_congestion_ops; |
317a76f9 SH |
1173 | extern u32 tcp_reno_ssthresh(struct tcp_sock *tp); |
1174 | extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack, | |
1175 | u32 rtt, u32 in_flight, int flag); | |
1176 | extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp); | |
a8acfbac | 1177 | extern struct tcp_congestion_ops tcp_reno; |
317a76f9 SH |
1178 | |
1179 | static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state) | |
1180 | { | |
1181 | if (tp->ca_ops->set_state) | |
1182 | tp->ca_ops->set_state(tp, ca_state); | |
1183 | tp->ca_state = ca_state; | |
1184 | } | |
1185 | ||
1186 | static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event) | |
1187 | { | |
1188 | if (tp->ca_ops->cwnd_event) | |
1189 | tp->ca_ops->cwnd_event(tp, event); | |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* This determines how many packets are "in the network" to the best |
1193 | * of our knowledge. In many cases it is conservative, but where | |
1194 | * detailed information is available from the receiver (via SACK | |
1195 | * blocks etc.) we can make more aggressive calculations. | |
1196 | * | |
1197 | * Use this for decisions involving congestion control, use just | |
1198 | * tp->packets_out to determine if the send queue is empty or not. | |
1199 | * | |
1200 | * Read this equation as: | |
1201 | * | |
1202 | * "Packets sent once on transmission queue" MINUS | |
1203 | * "Packets left network, but not honestly ACKed yet" PLUS | |
1204 | * "Packets fast retransmitted" | |
1205 | */ | |
1206 | static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) | |
1207 | { | |
1208 | return (tp->packets_out - tp->left_out + tp->retrans_out); | |
1209 | } | |
1210 | ||
1da177e4 LT |
1211 | /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. |
1212 | * The exception is rate halving phase, when cwnd is decreasing towards | |
1213 | * ssthresh. | |
1214 | */ | |
1215 | static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp) | |
1216 | { | |
1217 | if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery)) | |
1218 | return tp->snd_ssthresh; | |
1219 | else | |
1220 | return max(tp->snd_ssthresh, | |
1221 | ((tp->snd_cwnd >> 1) + | |
1222 | (tp->snd_cwnd >> 2))); | |
1223 | } | |
1224 | ||
1225 | static inline void tcp_sync_left_out(struct tcp_sock *tp) | |
1226 | { | |
1227 | if (tp->rx_opt.sack_ok && | |
1228 | (tp->sacked_out >= tp->packets_out - tp->lost_out)) | |
1229 | tp->sacked_out = tp->packets_out - tp->lost_out; | |
1230 | tp->left_out = tp->sacked_out + tp->lost_out; | |
1231 | } | |
1232 | ||
1233 | extern void tcp_cwnd_application_limited(struct sock *sk); | |
1234 | ||
1235 | /* Congestion window validation. (RFC2861) */ | |
1236 | ||
1237 | static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_sock *tp) | |
1238 | { | |
1239 | __u32 packets_out = tp->packets_out; | |
1240 | ||
1241 | if (packets_out >= tp->snd_cwnd) { | |
1242 | /* Network is feed fully. */ | |
1243 | tp->snd_cwnd_used = 0; | |
1244 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
1245 | } else { | |
1246 | /* Network starves. */ | |
1247 | if (tp->packets_out > tp->snd_cwnd_used) | |
1248 | tp->snd_cwnd_used = tp->packets_out; | |
1249 | ||
1250 | if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto) | |
1251 | tcp_cwnd_application_limited(sk); | |
1252 | } | |
1253 | } | |
1254 | ||
1255 | /* Set slow start threshould and cwnd not falling to slow start */ | |
1256 | static inline void __tcp_enter_cwr(struct tcp_sock *tp) | |
1257 | { | |
1258 | tp->undo_marker = 0; | |
317a76f9 | 1259 | tp->snd_ssthresh = tp->ca_ops->ssthresh(tp); |
1da177e4 LT |
1260 | tp->snd_cwnd = min(tp->snd_cwnd, |
1261 | tcp_packets_in_flight(tp) + 1U); | |
1262 | tp->snd_cwnd_cnt = 0; | |
1263 | tp->high_seq = tp->snd_nxt; | |
1264 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
1265 | TCP_ECN_queue_cwr(tp); | |
1266 | } | |
1267 | ||
1268 | static inline void tcp_enter_cwr(struct tcp_sock *tp) | |
1269 | { | |
1270 | tp->prior_ssthresh = 0; | |
1271 | if (tp->ca_state < TCP_CA_CWR) { | |
1272 | __tcp_enter_cwr(tp); | |
1273 | tcp_set_ca_state(tp, TCP_CA_CWR); | |
1274 | } | |
1275 | } | |
1276 | ||
1277 | extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); | |
1278 | ||
1279 | /* Slow start with delack produces 3 packets of burst, so that | |
1280 | * it is safe "de facto". | |
1281 | */ | |
1282 | static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) | |
1283 | { | |
1284 | return 3; | |
1285 | } | |
1286 | ||
1287 | static __inline__ int tcp_minshall_check(const struct tcp_sock *tp) | |
1288 | { | |
1289 | return after(tp->snd_sml,tp->snd_una) && | |
1290 | !after(tp->snd_sml, tp->snd_nxt); | |
1291 | } | |
1292 | ||
1293 | static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss, | |
1294 | const struct sk_buff *skb) | |
1295 | { | |
1296 | if (skb->len < mss) | |
1297 | tp->snd_sml = TCP_SKB_CB(skb)->end_seq; | |
1298 | } | |
1299 | ||
1300 | /* Return 0, if packet can be sent now without violation Nagle's rules: | |
1301 | 1. It is full sized. | |
1302 | 2. Or it contains FIN. | |
1303 | 3. Or TCP_NODELAY was set. | |
1304 | 4. Or TCP_CORK is not set, and all sent packets are ACKed. | |
1305 | With Minshall's modification: all sent small packets are ACKed. | |
1306 | */ | |
1307 | ||
1308 | static __inline__ int | |
1309 | tcp_nagle_check(const struct tcp_sock *tp, const struct sk_buff *skb, | |
1310 | unsigned mss_now, int nonagle) | |
1311 | { | |
1312 | return (skb->len < mss_now && | |
1313 | !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) && | |
1314 | ((nonagle&TCP_NAGLE_CORK) || | |
1315 | (!nonagle && | |
1316 | tp->packets_out && | |
1317 | tcp_minshall_check(tp)))); | |
1318 | } | |
1319 | ||
d5ac99a6 | 1320 | extern void tcp_set_skb_tso_segs(struct sock *, struct sk_buff *); |
1da177e4 LT |
1321 | |
1322 | /* This checks if the data bearing packet SKB (usually sk->sk_send_head) | |
1323 | * should be put on the wire right now. | |
1324 | */ | |
d5ac99a6 | 1325 | static __inline__ int tcp_snd_test(struct sock *sk, |
1da177e4 LT |
1326 | struct sk_buff *skb, |
1327 | unsigned cur_mss, int nonagle) | |
1328 | { | |
d5ac99a6 | 1329 | struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 LT |
1330 | int pkts = tcp_skb_pcount(skb); |
1331 | ||
1332 | if (!pkts) { | |
d5ac99a6 | 1333 | tcp_set_skb_tso_segs(sk, skb); |
1da177e4 LT |
1334 | pkts = tcp_skb_pcount(skb); |
1335 | } | |
1336 | ||
1337 | /* RFC 1122 - section 4.2.3.4 | |
1338 | * | |
1339 | * We must queue if | |
1340 | * | |
1341 | * a) The right edge of this frame exceeds the window | |
1342 | * b) There are packets in flight and we have a small segment | |
1343 | * [SWS avoidance and Nagle algorithm] | |
1344 | * (part of SWS is done on packetization) | |
1345 | * Minshall version sounds: there are no _small_ | |
1346 | * segments in flight. (tcp_nagle_check) | |
1347 | * c) We have too many packets 'in flight' | |
1348 | * | |
1349 | * Don't use the nagle rule for urgent data (or | |
1350 | * for the final FIN -DaveM). | |
1351 | * | |
1352 | * Also, Nagle rule does not apply to frames, which | |
1353 | * sit in the middle of queue (they have no chances | |
1354 | * to get new data) and if room at tail of skb is | |
1355 | * not enough to save something seriously (<32 for now). | |
1356 | */ | |
1357 | ||
1358 | /* Don't be strict about the congestion window for the | |
1359 | * final FIN frame. -DaveM | |
1360 | */ | |
1361 | return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode | |
1362 | || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) && | |
1363 | (((tcp_packets_in_flight(tp) + (pkts-1)) < tp->snd_cwnd) || | |
1364 | (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) && | |
1365 | !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd)); | |
1366 | } | |
1367 | ||
1368 | static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp) | |
1369 | { | |
1370 | if (!tp->packets_out && !tp->pending) | |
1371 | tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto); | |
1372 | } | |
1373 | ||
1374 | static __inline__ int tcp_skb_is_last(const struct sock *sk, | |
1375 | const struct sk_buff *skb) | |
1376 | { | |
1377 | return skb->next == (struct sk_buff *)&sk->sk_write_queue; | |
1378 | } | |
1379 | ||
1380 | /* Push out any pending frames which were held back due to | |
1381 | * TCP_CORK or attempt at coalescing tiny packets. | |
1382 | * The socket must be locked by the caller. | |
1383 | */ | |
1384 | static __inline__ void __tcp_push_pending_frames(struct sock *sk, | |
1385 | struct tcp_sock *tp, | |
1386 | unsigned cur_mss, | |
1387 | int nonagle) | |
1388 | { | |
1389 | struct sk_buff *skb = sk->sk_send_head; | |
1390 | ||
1391 | if (skb) { | |
1392 | if (!tcp_skb_is_last(sk, skb)) | |
1393 | nonagle = TCP_NAGLE_PUSH; | |
d5ac99a6 | 1394 | if (!tcp_snd_test(sk, skb, cur_mss, nonagle) || |
1da177e4 LT |
1395 | tcp_write_xmit(sk, nonagle)) |
1396 | tcp_check_probe_timer(sk, tp); | |
1397 | } | |
1398 | tcp_cwnd_validate(sk, tp); | |
1399 | } | |
1400 | ||
1401 | static __inline__ void tcp_push_pending_frames(struct sock *sk, | |
1402 | struct tcp_sock *tp) | |
1403 | { | |
1404 | __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle); | |
1405 | } | |
1406 | ||
1407 | static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp) | |
1408 | { | |
1409 | struct sk_buff *skb = sk->sk_send_head; | |
1410 | ||
1411 | return (skb && | |
d5ac99a6 | 1412 | tcp_snd_test(sk, skb, tcp_current_mss(sk, 1), |
1da177e4 LT |
1413 | tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle)); |
1414 | } | |
1415 | ||
1416 | static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) | |
1417 | { | |
1418 | tp->snd_wl1 = seq; | |
1419 | } | |
1420 | ||
1421 | static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) | |
1422 | { | |
1423 | tp->snd_wl1 = seq; | |
1424 | } | |
1425 | ||
1426 | extern void tcp_destroy_sock(struct sock *sk); | |
1427 | ||
1428 | ||
1429 | /* | |
1430 | * Calculate(/check) TCP checksum | |
1431 | */ | |
1432 | static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len, | |
1433 | unsigned long saddr, unsigned long daddr, | |
1434 | unsigned long base) | |
1435 | { | |
1436 | return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); | |
1437 | } | |
1438 | ||
1439 | static __inline__ int __tcp_checksum_complete(struct sk_buff *skb) | |
1440 | { | |
1441 | return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum)); | |
1442 | } | |
1443 | ||
1444 | static __inline__ int tcp_checksum_complete(struct sk_buff *skb) | |
1445 | { | |
1446 | return skb->ip_summed != CHECKSUM_UNNECESSARY && | |
1447 | __tcp_checksum_complete(skb); | |
1448 | } | |
1449 | ||
1450 | /* Prequeue for VJ style copy to user, combined with checksumming. */ | |
1451 | ||
1452 | static __inline__ void tcp_prequeue_init(struct tcp_sock *tp) | |
1453 | { | |
1454 | tp->ucopy.task = NULL; | |
1455 | tp->ucopy.len = 0; | |
1456 | tp->ucopy.memory = 0; | |
1457 | skb_queue_head_init(&tp->ucopy.prequeue); | |
1458 | } | |
1459 | ||
1460 | /* Packet is added to VJ-style prequeue for processing in process | |
1461 | * context, if a reader task is waiting. Apparently, this exciting | |
1462 | * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) | |
1463 | * failed somewhere. Latency? Burstiness? Well, at least now we will | |
1464 | * see, why it failed. 8)8) --ANK | |
1465 | * | |
1466 | * NOTE: is this not too big to inline? | |
1467 | */ | |
1468 | static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb) | |
1469 | { | |
1470 | struct tcp_sock *tp = tcp_sk(sk); | |
1471 | ||
1472 | if (!sysctl_tcp_low_latency && tp->ucopy.task) { | |
1473 | __skb_queue_tail(&tp->ucopy.prequeue, skb); | |
1474 | tp->ucopy.memory += skb->truesize; | |
1475 | if (tp->ucopy.memory > sk->sk_rcvbuf) { | |
1476 | struct sk_buff *skb1; | |
1477 | ||
1478 | BUG_ON(sock_owned_by_user(sk)); | |
1479 | ||
1480 | while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { | |
1481 | sk->sk_backlog_rcv(sk, skb1); | |
1482 | NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); | |
1483 | } | |
1484 | ||
1485 | tp->ucopy.memory = 0; | |
1486 | } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { | |
1487 | wake_up_interruptible(sk->sk_sleep); | |
1488 | if (!tcp_ack_scheduled(tp)) | |
1489 | tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4); | |
1490 | } | |
1491 | return 1; | |
1492 | } | |
1493 | return 0; | |
1494 | } | |
1495 | ||
1496 | ||
1497 | #undef STATE_TRACE | |
1498 | ||
1499 | #ifdef STATE_TRACE | |
1500 | static const char *statename[]={ | |
1501 | "Unused","Established","Syn Sent","Syn Recv", | |
1502 | "Fin Wait 1","Fin Wait 2","Time Wait", "Close", | |
1503 | "Close Wait","Last ACK","Listen","Closing" | |
1504 | }; | |
1505 | #endif | |
1506 | ||
1507 | static __inline__ void tcp_set_state(struct sock *sk, int state) | |
1508 | { | |
1509 | int oldstate = sk->sk_state; | |
1510 | ||
1511 | switch (state) { | |
1512 | case TCP_ESTABLISHED: | |
1513 | if (oldstate != TCP_ESTABLISHED) | |
1514 | TCP_INC_STATS(TCP_MIB_CURRESTAB); | |
1515 | break; | |
1516 | ||
1517 | case TCP_CLOSE: | |
1518 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
1519 | TCP_INC_STATS(TCP_MIB_ESTABRESETS); | |
1520 | ||
1521 | sk->sk_prot->unhash(sk); | |
1522 | if (tcp_sk(sk)->bind_hash && | |
1523 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) | |
1524 | tcp_put_port(sk); | |
1525 | /* fall through */ | |
1526 | default: | |
1527 | if (oldstate==TCP_ESTABLISHED) | |
1528 | TCP_DEC_STATS(TCP_MIB_CURRESTAB); | |
1529 | } | |
1530 | ||
1531 | /* Change state AFTER socket is unhashed to avoid closed | |
1532 | * socket sitting in hash tables. | |
1533 | */ | |
1534 | sk->sk_state = state; | |
1535 | ||
1536 | #ifdef STATE_TRACE | |
1537 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]); | |
1538 | #endif | |
1539 | } | |
1540 | ||
1541 | static __inline__ void tcp_done(struct sock *sk) | |
1542 | { | |
1543 | tcp_set_state(sk, TCP_CLOSE); | |
1544 | tcp_clear_xmit_timers(sk); | |
1545 | ||
1546 | sk->sk_shutdown = SHUTDOWN_MASK; | |
1547 | ||
1548 | if (!sock_flag(sk, SOCK_DEAD)) | |
1549 | sk->sk_state_change(sk); | |
1550 | else | |
1551 | tcp_destroy_sock(sk); | |
1552 | } | |
1553 | ||
1554 | static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt) | |
1555 | { | |
1556 | rx_opt->dsack = 0; | |
1557 | rx_opt->eff_sacks = 0; | |
1558 | rx_opt->num_sacks = 0; | |
1559 | } | |
1560 | ||
1561 | static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp) | |
1562 | { | |
1563 | if (tp->rx_opt.tstamp_ok) { | |
1564 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | | |
1565 | (TCPOPT_NOP << 16) | | |
1566 | (TCPOPT_TIMESTAMP << 8) | | |
1567 | TCPOLEN_TIMESTAMP); | |
1568 | *ptr++ = htonl(tstamp); | |
1569 | *ptr++ = htonl(tp->rx_opt.ts_recent); | |
1570 | } | |
1571 | if (tp->rx_opt.eff_sacks) { | |
1572 | struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks; | |
1573 | int this_sack; | |
1574 | ||
1575 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | | |
1576 | (TCPOPT_NOP << 16) | | |
1577 | (TCPOPT_SACK << 8) | | |
1578 | (TCPOLEN_SACK_BASE + | |
1579 | (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK))); | |
1580 | for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) { | |
1581 | *ptr++ = htonl(sp[this_sack].start_seq); | |
1582 | *ptr++ = htonl(sp[this_sack].end_seq); | |
1583 | } | |
1584 | if (tp->rx_opt.dsack) { | |
1585 | tp->rx_opt.dsack = 0; | |
1586 | tp->rx_opt.eff_sacks--; | |
1587 | } | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | /* Construct a tcp options header for a SYN or SYN_ACK packet. | |
1592 | * If this is every changed make sure to change the definition of | |
1593 | * MAX_SYN_SIZE to match the new maximum number of options that you | |
1594 | * can generate. | |
1595 | */ | |
1596 | static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack, | |
1597 | int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent) | |
1598 | { | |
1599 | /* We always get an MSS option. | |
1600 | * The option bytes which will be seen in normal data | |
1601 | * packets should timestamps be used, must be in the MSS | |
1602 | * advertised. But we subtract them from tp->mss_cache so | |
1603 | * that calculations in tcp_sendmsg are simpler etc. | |
1604 | * So account for this fact here if necessary. If we | |
1605 | * don't do this correctly, as a receiver we won't | |
1606 | * recognize data packets as being full sized when we | |
1607 | * should, and thus we won't abide by the delayed ACK | |
1608 | * rules correctly. | |
1609 | * SACKs don't matter, we never delay an ACK when we | |
1610 | * have any of those going out. | |
1611 | */ | |
1612 | *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss); | |
1613 | if (ts) { | |
1614 | if(sack) | |
1615 | *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) | | |
1616 | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); | |
1617 | else | |
1618 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
1619 | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); | |
1620 | *ptr++ = htonl(tstamp); /* TSVAL */ | |
1621 | *ptr++ = htonl(ts_recent); /* TSECR */ | |
1622 | } else if(sack) | |
1623 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
1624 | (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM); | |
1625 | if (offer_wscale) | |
1626 | *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale)); | |
1627 | } | |
1628 | ||
1629 | /* Determine a window scaling and initial window to offer. */ | |
1630 | extern void tcp_select_initial_window(int __space, __u32 mss, | |
1631 | __u32 *rcv_wnd, __u32 *window_clamp, | |
1632 | int wscale_ok, __u8 *rcv_wscale); | |
1633 | ||
1634 | static inline int tcp_win_from_space(int space) | |
1635 | { | |
1636 | return sysctl_tcp_adv_win_scale<=0 ? | |
1637 | (space>>(-sysctl_tcp_adv_win_scale)) : | |
1638 | space - (space>>sysctl_tcp_adv_win_scale); | |
1639 | } | |
1640 | ||
1641 | /* Note: caller must be prepared to deal with negative returns */ | |
1642 | static inline int tcp_space(const struct sock *sk) | |
1643 | { | |
1644 | return tcp_win_from_space(sk->sk_rcvbuf - | |
1645 | atomic_read(&sk->sk_rmem_alloc)); | |
1646 | } | |
1647 | ||
1648 | static inline int tcp_full_space(const struct sock *sk) | |
1649 | { | |
1650 | return tcp_win_from_space(sk->sk_rcvbuf); | |
1651 | } | |
1652 | ||
60236fdd | 1653 | static inline void tcp_acceptq_queue(struct sock *sk, struct request_sock *req, |
1da177e4 LT |
1654 | struct sock *child) |
1655 | { | |
0e87506f | 1656 | reqsk_queue_add(&tcp_sk(sk)->accept_queue, req, sk, child); |
1da177e4 LT |
1657 | } |
1658 | ||
1da177e4 | 1659 | static inline void |
60236fdd | 1660 | tcp_synq_removed(struct sock *sk, struct request_sock *req) |
1da177e4 | 1661 | { |
0e87506f | 1662 | if (reqsk_queue_removed(&tcp_sk(sk)->accept_queue, req) == 0) |
1da177e4 | 1663 | tcp_delete_keepalive_timer(sk); |
1da177e4 LT |
1664 | } |
1665 | ||
1666 | static inline void tcp_synq_added(struct sock *sk) | |
1667 | { | |
0e87506f | 1668 | if (reqsk_queue_added(&tcp_sk(sk)->accept_queue) == 0) |
1da177e4 | 1669 | tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT); |
1da177e4 LT |
1670 | } |
1671 | ||
1672 | static inline int tcp_synq_len(struct sock *sk) | |
1673 | { | |
0e87506f | 1674 | return reqsk_queue_len(&tcp_sk(sk)->accept_queue); |
1da177e4 LT |
1675 | } |
1676 | ||
1677 | static inline int tcp_synq_young(struct sock *sk) | |
1678 | { | |
0e87506f | 1679 | return reqsk_queue_len_young(&tcp_sk(sk)->accept_queue); |
1da177e4 LT |
1680 | } |
1681 | ||
1682 | static inline int tcp_synq_is_full(struct sock *sk) | |
1683 | { | |
0e87506f | 1684 | return reqsk_queue_is_full(&tcp_sk(sk)->accept_queue); |
1da177e4 LT |
1685 | } |
1686 | ||
60236fdd | 1687 | static inline void tcp_synq_unlink(struct tcp_sock *tp, struct request_sock *req, |
0e87506f | 1688 | struct request_sock **prev) |
1da177e4 | 1689 | { |
0e87506f | 1690 | reqsk_queue_unlink(&tp->accept_queue, req, prev); |
1da177e4 LT |
1691 | } |
1692 | ||
60236fdd ACM |
1693 | static inline void tcp_synq_drop(struct sock *sk, struct request_sock *req, |
1694 | struct request_sock **prev) | |
1da177e4 LT |
1695 | { |
1696 | tcp_synq_unlink(tcp_sk(sk), req, prev); | |
1697 | tcp_synq_removed(sk, req); | |
60236fdd | 1698 | reqsk_free(req); |
1da177e4 LT |
1699 | } |
1700 | ||
60236fdd | 1701 | static __inline__ void tcp_openreq_init(struct request_sock *req, |
1da177e4 LT |
1702 | struct tcp_options_received *rx_opt, |
1703 | struct sk_buff *skb) | |
1704 | { | |
2e6599cb ACM |
1705 | struct inet_request_sock *ireq = inet_rsk(req); |
1706 | ||
1da177e4 | 1707 | req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ |
2e6599cb | 1708 | tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; |
1da177e4 LT |
1709 | req->mss = rx_opt->mss_clamp; |
1710 | req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; | |
2e6599cb ACM |
1711 | ireq->tstamp_ok = rx_opt->tstamp_ok; |
1712 | ireq->sack_ok = rx_opt->sack_ok; | |
1713 | ireq->snd_wscale = rx_opt->snd_wscale; | |
1714 | ireq->wscale_ok = rx_opt->wscale_ok; | |
1715 | ireq->acked = 0; | |
1716 | ireq->ecn_ok = 0; | |
1717 | ireq->rmt_port = skb->h.th->source; | |
1da177e4 LT |
1718 | } |
1719 | ||
1720 | extern void tcp_enter_memory_pressure(void); | |
1721 | ||
1722 | extern void tcp_listen_wlock(void); | |
1723 | ||
1724 | /* - We may sleep inside this lock. | |
1725 | * - If sleeping is not required (or called from BH), | |
1726 | * use plain read_(un)lock(&tcp_lhash_lock). | |
1727 | */ | |
1728 | ||
1729 | static inline void tcp_listen_lock(void) | |
1730 | { | |
1731 | /* read_lock synchronizes to candidates to writers */ | |
1732 | read_lock(&tcp_lhash_lock); | |
1733 | atomic_inc(&tcp_lhash_users); | |
1734 | read_unlock(&tcp_lhash_lock); | |
1735 | } | |
1736 | ||
1737 | static inline void tcp_listen_unlock(void) | |
1738 | { | |
1739 | if (atomic_dec_and_test(&tcp_lhash_users)) | |
1740 | wake_up(&tcp_lhash_wait); | |
1741 | } | |
1742 | ||
1743 | static inline int keepalive_intvl_when(const struct tcp_sock *tp) | |
1744 | { | |
1745 | return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; | |
1746 | } | |
1747 | ||
1748 | static inline int keepalive_time_when(const struct tcp_sock *tp) | |
1749 | { | |
1750 | return tp->keepalive_time ? : sysctl_tcp_keepalive_time; | |
1751 | } | |
1752 | ||
1753 | static inline int tcp_fin_time(const struct tcp_sock *tp) | |
1754 | { | |
1755 | int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout; | |
1756 | ||
1757 | if (fin_timeout < (tp->rto<<2) - (tp->rto>>1)) | |
1758 | fin_timeout = (tp->rto<<2) - (tp->rto>>1); | |
1759 | ||
1760 | return fin_timeout; | |
1761 | } | |
1762 | ||
1763 | static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) | |
1764 | { | |
1765 | if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) | |
1766 | return 0; | |
1767 | if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) | |
1768 | return 0; | |
1769 | ||
1770 | /* RST segments are not recommended to carry timestamp, | |
1771 | and, if they do, it is recommended to ignore PAWS because | |
1772 | "their cleanup function should take precedence over timestamps." | |
1773 | Certainly, it is mistake. It is necessary to understand the reasons | |
1774 | of this constraint to relax it: if peer reboots, clock may go | |
1775 | out-of-sync and half-open connections will not be reset. | |
1776 | Actually, the problem would be not existing if all | |
1777 | the implementations followed draft about maintaining clock | |
1778 | via reboots. Linux-2.2 DOES NOT! | |
1779 | ||
1780 | However, we can relax time bounds for RST segments to MSL. | |
1781 | */ | |
1782 | if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) | |
1783 | return 0; | |
1784 | return 1; | |
1785 | } | |
1786 | ||
1787 | static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst) | |
1788 | { | |
1789 | sk->sk_route_caps = dst->dev->features; | |
1790 | if (sk->sk_route_caps & NETIF_F_TSO) { | |
1791 | if (sock_flag(sk, SOCK_NO_LARGESEND) || dst->header_len) | |
1792 | sk->sk_route_caps &= ~NETIF_F_TSO; | |
1793 | } | |
1794 | } | |
1795 | ||
1796 | #define TCP_CHECK_TIMER(sk) do { } while (0) | |
1797 | ||
1798 | static inline int tcp_use_frto(const struct sock *sk) | |
1799 | { | |
1800 | const struct tcp_sock *tp = tcp_sk(sk); | |
1801 | ||
1802 | /* F-RTO must be activated in sysctl and there must be some | |
1803 | * unsent new data, and the advertised window should allow | |
1804 | * sending it. | |
1805 | */ | |
1806 | return (sysctl_tcp_frto && sk->sk_send_head && | |
1807 | !after(TCP_SKB_CB(sk->sk_send_head)->end_seq, | |
1808 | tp->snd_una + tp->snd_wnd)); | |
1809 | } | |
1810 | ||
1811 | static inline void tcp_mib_init(void) | |
1812 | { | |
1813 | /* See RFC 2012 */ | |
1814 | TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1); | |
1815 | TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); | |
1816 | TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); | |
1817 | TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1); | |
1818 | } | |
1819 | ||
1820 | /* /proc */ | |
1821 | enum tcp_seq_states { | |
1822 | TCP_SEQ_STATE_LISTENING, | |
1823 | TCP_SEQ_STATE_OPENREQ, | |
1824 | TCP_SEQ_STATE_ESTABLISHED, | |
1825 | TCP_SEQ_STATE_TIME_WAIT, | |
1826 | }; | |
1827 | ||
1828 | struct tcp_seq_afinfo { | |
1829 | struct module *owner; | |
1830 | char *name; | |
1831 | sa_family_t family; | |
1832 | int (*seq_show) (struct seq_file *m, void *v); | |
1833 | struct file_operations *seq_fops; | |
1834 | }; | |
1835 | ||
1836 | struct tcp_iter_state { | |
1837 | sa_family_t family; | |
1838 | enum tcp_seq_states state; | |
1839 | struct sock *syn_wait_sk; | |
1840 | int bucket, sbucket, num, uid; | |
1841 | struct seq_operations seq_ops; | |
1842 | }; | |
1843 | ||
1844 | extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo); | |
1845 | extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo); | |
1846 | ||
1da177e4 | 1847 | #endif /* _TCP_H */ |