2 * ip_vs_proto_tcp.c: TCP load balancing support for IPVS
4 * Version: $Id: ip_vs_proto_tcp.c,v 1.3 2002/11/30 01:50:35 wensong Exp $
6 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
7 * Julian Anastasov <ja@ssi.bg>
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; either version
12 * 2 of the License, or (at your option) any later version.
18 #include <linux/kernel.h>
20 #include <linux/tcp.h> /* for tcphdr */
22 #include <net/tcp.h> /* for csum_tcpudp_magic */
23 #include <linux/netfilter_ipv4.h>
25 #include <net/ip_vs.h>
28 static struct ip_vs_conn
*
29 tcp_conn_in_get(const struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
30 const struct iphdr
*iph
, unsigned int proto_off
, int inverse
)
32 __be16 _ports
[2], *pptr
;
34 pptr
= skb_header_pointer(skb
, proto_off
, sizeof(_ports
), _ports
);
38 if (likely(!inverse
)) {
39 return ip_vs_conn_in_get(iph
->protocol
,
43 return ip_vs_conn_in_get(iph
->protocol
,
49 static struct ip_vs_conn
*
50 tcp_conn_out_get(const struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
51 const struct iphdr
*iph
, unsigned int proto_off
, int inverse
)
53 __be16 _ports
[2], *pptr
;
55 pptr
= skb_header_pointer(skb
, proto_off
, sizeof(_ports
), _ports
);
59 if (likely(!inverse
)) {
60 return ip_vs_conn_out_get(iph
->protocol
,
64 return ip_vs_conn_out_get(iph
->protocol
,
72 tcp_conn_schedule(struct sk_buff
*skb
,
73 struct ip_vs_protocol
*pp
,
74 int *verdict
, struct ip_vs_conn
**cpp
)
76 struct ip_vs_service
*svc
;
77 struct tcphdr _tcph
, *th
;
79 th
= skb_header_pointer(skb
, ip_hdrlen(skb
), sizeof(_tcph
), &_tcph
);
86 (svc
= ip_vs_service_get(skb
->mark
, skb
->nh
.iph
->protocol
,
87 skb
->nh
.iph
->daddr
, th
->dest
))) {
90 * It seems that we are very loaded.
91 * We have to drop this packet :(
93 ip_vs_service_put(svc
);
99 * Let the virtual server select a real server for the
100 * incoming connection, and create a connection entry.
102 *cpp
= ip_vs_schedule(svc
, skb
);
104 *verdict
= ip_vs_leave(svc
, skb
, pp
);
107 ip_vs_service_put(svc
);
114 tcp_fast_csum_update(struct tcphdr
*tcph
, __be32 oldip
, __be32 newip
,
115 __be16 oldport
, __be16 newport
)
118 csum_fold(ip_vs_check_diff4(oldip
, newip
,
119 ip_vs_check_diff2(oldport
, newport
,
120 ~csum_unfold(tcph
->check
))));
125 tcp_snat_handler(struct sk_buff
**pskb
,
126 struct ip_vs_protocol
*pp
, struct ip_vs_conn
*cp
)
129 const unsigned int tcphoff
= ip_hdrlen(*pskb
);
131 /* csum_check requires unshared skb */
132 if (!ip_vs_make_skb_writable(pskb
, tcphoff
+sizeof(*tcph
)))
135 if (unlikely(cp
->app
!= NULL
)) {
136 /* Some checks before mangling */
137 if (pp
->csum_check
&& !pp
->csum_check(*pskb
, pp
))
140 /* Call application helper if needed */
141 if (!ip_vs_app_pkt_out(cp
, pskb
))
145 tcph
= (void *)(*pskb
)->nh
.iph
+ tcphoff
;
146 tcph
->source
= cp
->vport
;
148 /* Adjust TCP checksums */
150 /* Only port and addr are changed, do fast csum update */
151 tcp_fast_csum_update(tcph
, cp
->daddr
, cp
->vaddr
,
152 cp
->dport
, cp
->vport
);
153 if ((*pskb
)->ip_summed
== CHECKSUM_COMPLETE
)
154 (*pskb
)->ip_summed
= CHECKSUM_NONE
;
156 /* full checksum calculation */
158 (*pskb
)->csum
= skb_checksum(*pskb
, tcphoff
,
159 (*pskb
)->len
- tcphoff
, 0);
160 tcph
->check
= csum_tcpudp_magic(cp
->vaddr
, cp
->caddr
,
161 (*pskb
)->len
- tcphoff
,
164 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
165 pp
->name
, tcph
->check
,
166 (char*)&(tcph
->check
) - (char*)tcph
);
173 tcp_dnat_handler(struct sk_buff
**pskb
,
174 struct ip_vs_protocol
*pp
, struct ip_vs_conn
*cp
)
177 const unsigned int tcphoff
= ip_hdrlen(*pskb
);
179 /* csum_check requires unshared skb */
180 if (!ip_vs_make_skb_writable(pskb
, tcphoff
+sizeof(*tcph
)))
183 if (unlikely(cp
->app
!= NULL
)) {
184 /* Some checks before mangling */
185 if (pp
->csum_check
&& !pp
->csum_check(*pskb
, pp
))
189 * Attempt ip_vs_app call.
190 * It will fix ip_vs_conn and iph ack_seq stuff
192 if (!ip_vs_app_pkt_in(cp
, pskb
))
196 tcph
= (void *)(*pskb
)->nh
.iph
+ tcphoff
;
197 tcph
->dest
= cp
->dport
;
200 * Adjust TCP checksums
203 /* Only port and addr are changed, do fast csum update */
204 tcp_fast_csum_update(tcph
, cp
->vaddr
, cp
->daddr
,
205 cp
->vport
, cp
->dport
);
206 if ((*pskb
)->ip_summed
== CHECKSUM_COMPLETE
)
207 (*pskb
)->ip_summed
= CHECKSUM_NONE
;
209 /* full checksum calculation */
211 (*pskb
)->csum
= skb_checksum(*pskb
, tcphoff
,
212 (*pskb
)->len
- tcphoff
, 0);
213 tcph
->check
= csum_tcpudp_magic(cp
->caddr
, cp
->daddr
,
214 (*pskb
)->len
- tcphoff
,
217 (*pskb
)->ip_summed
= CHECKSUM_UNNECESSARY
;
224 tcp_csum_check(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
)
226 const unsigned int tcphoff
= ip_hdrlen(skb
);
228 switch (skb
->ip_summed
) {
230 skb
->csum
= skb_checksum(skb
, tcphoff
, skb
->len
- tcphoff
, 0);
231 case CHECKSUM_COMPLETE
:
232 if (csum_tcpudp_magic(skb
->nh
.iph
->saddr
, skb
->nh
.iph
->daddr
,
234 skb
->nh
.iph
->protocol
, skb
->csum
)) {
235 IP_VS_DBG_RL_PKT(0, pp
, skb
, 0,
236 "Failed checksum for");
241 /* No need to checksum. */
249 #define TCP_DIR_INPUT 0
250 #define TCP_DIR_OUTPUT 4
251 #define TCP_DIR_INPUT_ONLY 8
253 static const int tcp_state_off
[IP_VS_DIR_LAST
] = {
254 [IP_VS_DIR_INPUT
] = TCP_DIR_INPUT
,
255 [IP_VS_DIR_OUTPUT
] = TCP_DIR_OUTPUT
,
256 [IP_VS_DIR_INPUT_ONLY
] = TCP_DIR_INPUT_ONLY
,
260 * Timeout table[state]
262 static int tcp_timeouts
[IP_VS_TCP_S_LAST
+1] = {
263 [IP_VS_TCP_S_NONE
] = 2*HZ
,
264 [IP_VS_TCP_S_ESTABLISHED
] = 15*60*HZ
,
265 [IP_VS_TCP_S_SYN_SENT
] = 2*60*HZ
,
266 [IP_VS_TCP_S_SYN_RECV
] = 1*60*HZ
,
267 [IP_VS_TCP_S_FIN_WAIT
] = 2*60*HZ
,
268 [IP_VS_TCP_S_TIME_WAIT
] = 2*60*HZ
,
269 [IP_VS_TCP_S_CLOSE
] = 10*HZ
,
270 [IP_VS_TCP_S_CLOSE_WAIT
] = 60*HZ
,
271 [IP_VS_TCP_S_LAST_ACK
] = 30*HZ
,
272 [IP_VS_TCP_S_LISTEN
] = 2*60*HZ
,
273 [IP_VS_TCP_S_SYNACK
] = 120*HZ
,
274 [IP_VS_TCP_S_LAST
] = 2*HZ
,
277 static char * tcp_state_name_table
[IP_VS_TCP_S_LAST
+1] = {
278 [IP_VS_TCP_S_NONE
] = "NONE",
279 [IP_VS_TCP_S_ESTABLISHED
] = "ESTABLISHED",
280 [IP_VS_TCP_S_SYN_SENT
] = "SYN_SENT",
281 [IP_VS_TCP_S_SYN_RECV
] = "SYN_RECV",
282 [IP_VS_TCP_S_FIN_WAIT
] = "FIN_WAIT",
283 [IP_VS_TCP_S_TIME_WAIT
] = "TIME_WAIT",
284 [IP_VS_TCP_S_CLOSE
] = "CLOSE",
285 [IP_VS_TCP_S_CLOSE_WAIT
] = "CLOSE_WAIT",
286 [IP_VS_TCP_S_LAST_ACK
] = "LAST_ACK",
287 [IP_VS_TCP_S_LISTEN
] = "LISTEN",
288 [IP_VS_TCP_S_SYNACK
] = "SYNACK",
289 [IP_VS_TCP_S_LAST
] = "BUG!",
292 #define sNO IP_VS_TCP_S_NONE
293 #define sES IP_VS_TCP_S_ESTABLISHED
294 #define sSS IP_VS_TCP_S_SYN_SENT
295 #define sSR IP_VS_TCP_S_SYN_RECV
296 #define sFW IP_VS_TCP_S_FIN_WAIT
297 #define sTW IP_VS_TCP_S_TIME_WAIT
298 #define sCL IP_VS_TCP_S_CLOSE
299 #define sCW IP_VS_TCP_S_CLOSE_WAIT
300 #define sLA IP_VS_TCP_S_LAST_ACK
301 #define sLI IP_VS_TCP_S_LISTEN
302 #define sSA IP_VS_TCP_S_SYNACK
304 struct tcp_states_t
{
305 int next_state
[IP_VS_TCP_S_LAST
];
308 static const char * tcp_state_name(int state
)
310 if (state
>= IP_VS_TCP_S_LAST
)
312 return tcp_state_name_table
[state
] ? tcp_state_name_table
[state
] : "?";
315 static struct tcp_states_t tcp_states
[] = {
317 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
318 /*syn*/ {{sSR
, sES
, sES
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
}},
319 /*fin*/ {{sCL
, sCW
, sSS
, sTW
, sTW
, sTW
, sCL
, sCW
, sLA
, sLI
, sTW
}},
320 /*ack*/ {{sCL
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sES
}},
321 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sSR
}},
324 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
325 /*syn*/ {{sSS
, sES
, sSS
, sSR
, sSS
, sSS
, sSS
, sSS
, sSS
, sLI
, sSR
}},
326 /*fin*/ {{sTW
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sTW
, sLA
, sLI
, sTW
}},
327 /*ack*/ {{sES
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sLA
, sES
, sES
}},
328 /*rst*/ {{sCL
, sCL
, sSS
, sCL
, sCL
, sTW
, sCL
, sCL
, sCL
, sCL
, sCL
}},
331 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
332 /*syn*/ {{sSR
, sES
, sES
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
}},
333 /*fin*/ {{sCL
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sCW
, sLA
, sLI
, sTW
}},
334 /*ack*/ {{sCL
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sES
}},
335 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sCL
}},
338 static struct tcp_states_t tcp_states_dos
[] = {
340 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
341 /*syn*/ {{sSR
, sES
, sES
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSA
}},
342 /*fin*/ {{sCL
, sCW
, sSS
, sTW
, sTW
, sTW
, sCL
, sCW
, sLA
, sLI
, sSA
}},
343 /*ack*/ {{sCL
, sES
, sSS
, sSR
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sSA
}},
344 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sCL
}},
347 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
348 /*syn*/ {{sSS
, sES
, sSS
, sSA
, sSS
, sSS
, sSS
, sSS
, sSS
, sLI
, sSA
}},
349 /*fin*/ {{sTW
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sTW
, sLA
, sLI
, sTW
}},
350 /*ack*/ {{sES
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sLA
, sES
, sES
}},
351 /*rst*/ {{sCL
, sCL
, sSS
, sCL
, sCL
, sTW
, sCL
, sCL
, sCL
, sCL
, sCL
}},
354 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
355 /*syn*/ {{sSA
, sES
, sES
, sSR
, sSA
, sSA
, sSA
, sSA
, sSA
, sSA
, sSA
}},
356 /*fin*/ {{sCL
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sCW
, sLA
, sLI
, sTW
}},
357 /*ack*/ {{sCL
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sES
}},
358 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sCL
}},
361 static struct tcp_states_t
*tcp_state_table
= tcp_states
;
364 static void tcp_timeout_change(struct ip_vs_protocol
*pp
, int flags
)
366 int on
= (flags
& 1); /* secure_tcp */
369 ** FIXME: change secure_tcp to independent sysctl var
370 ** or make it per-service or per-app because it is valid
371 ** for most if not for all of the applications. Something
372 ** like "capabilities" (flags) for each object.
374 tcp_state_table
= (on
? tcp_states_dos
: tcp_states
);
378 tcp_set_state_timeout(struct ip_vs_protocol
*pp
, char *sname
, int to
)
380 return ip_vs_set_state_timeout(pp
->timeout_table
, IP_VS_TCP_S_LAST
,
381 tcp_state_name_table
, sname
, to
);
384 static inline int tcp_state_idx(struct tcphdr
*th
)
398 set_tcp_state(struct ip_vs_protocol
*pp
, struct ip_vs_conn
*cp
,
399 int direction
, struct tcphdr
*th
)
402 int new_state
= IP_VS_TCP_S_CLOSE
;
403 int state_off
= tcp_state_off
[direction
];
406 * Update state offset to INPUT_ONLY if necessary
407 * or delete NO_OUTPUT flag if output packet detected
409 if (cp
->flags
& IP_VS_CONN_F_NOOUTPUT
) {
410 if (state_off
== TCP_DIR_OUTPUT
)
411 cp
->flags
&= ~IP_VS_CONN_F_NOOUTPUT
;
413 state_off
= TCP_DIR_INPUT_ONLY
;
416 if ((state_idx
= tcp_state_idx(th
)) < 0) {
417 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx
);
421 new_state
= tcp_state_table
[state_off
+state_idx
].next_state
[cp
->state
];
424 if (new_state
!= cp
->state
) {
425 struct ip_vs_dest
*dest
= cp
->dest
;
427 IP_VS_DBG(8, "%s %s [%c%c%c%c] %u.%u.%u.%u:%d->"
428 "%u.%u.%u.%u:%d state: %s->%s conn->refcnt:%d\n",
430 (state_off
==TCP_DIR_OUTPUT
)?"output ":"input ",
435 NIPQUAD(cp
->daddr
), ntohs(cp
->dport
),
436 NIPQUAD(cp
->caddr
), ntohs(cp
->cport
),
437 tcp_state_name(cp
->state
),
438 tcp_state_name(new_state
),
439 atomic_read(&cp
->refcnt
));
441 if (!(cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
442 (new_state
!= IP_VS_TCP_S_ESTABLISHED
)) {
443 atomic_dec(&dest
->activeconns
);
444 atomic_inc(&dest
->inactconns
);
445 cp
->flags
|= IP_VS_CONN_F_INACTIVE
;
446 } else if ((cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
447 (new_state
== IP_VS_TCP_S_ESTABLISHED
)) {
448 atomic_inc(&dest
->activeconns
);
449 atomic_dec(&dest
->inactconns
);
450 cp
->flags
&= ~IP_VS_CONN_F_INACTIVE
;
455 cp
->timeout
= pp
->timeout_table
[cp
->state
= new_state
];
460 * Handle state transitions
463 tcp_state_transition(struct ip_vs_conn
*cp
, int direction
,
464 const struct sk_buff
*skb
,
465 struct ip_vs_protocol
*pp
)
467 struct tcphdr _tcph
, *th
;
469 th
= skb_header_pointer(skb
, ip_hdrlen(skb
), sizeof(_tcph
), &_tcph
);
473 spin_lock(&cp
->lock
);
474 set_tcp_state(pp
, cp
, direction
, th
);
475 spin_unlock(&cp
->lock
);
482 * Hash table for TCP application incarnations
484 #define TCP_APP_TAB_BITS 4
485 #define TCP_APP_TAB_SIZE (1 << TCP_APP_TAB_BITS)
486 #define TCP_APP_TAB_MASK (TCP_APP_TAB_SIZE - 1)
488 static struct list_head tcp_apps
[TCP_APP_TAB_SIZE
];
489 static DEFINE_SPINLOCK(tcp_app_lock
);
491 static inline __u16
tcp_app_hashkey(__be16 port
)
493 return (((__force u16
)port
>> TCP_APP_TAB_BITS
) ^ (__force u16
)port
)
498 static int tcp_register_app(struct ip_vs_app
*inc
)
502 __be16 port
= inc
->port
;
505 hash
= tcp_app_hashkey(port
);
507 spin_lock_bh(&tcp_app_lock
);
508 list_for_each_entry(i
, &tcp_apps
[hash
], p_list
) {
509 if (i
->port
== port
) {
514 list_add(&inc
->p_list
, &tcp_apps
[hash
]);
515 atomic_inc(&ip_vs_protocol_tcp
.appcnt
);
518 spin_unlock_bh(&tcp_app_lock
);
524 tcp_unregister_app(struct ip_vs_app
*inc
)
526 spin_lock_bh(&tcp_app_lock
);
527 atomic_dec(&ip_vs_protocol_tcp
.appcnt
);
528 list_del(&inc
->p_list
);
529 spin_unlock_bh(&tcp_app_lock
);
534 tcp_app_conn_bind(struct ip_vs_conn
*cp
)
537 struct ip_vs_app
*inc
;
540 /* Default binding: bind app only for NAT */
541 if (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)
544 /* Lookup application incarnations and bind the right one */
545 hash
= tcp_app_hashkey(cp
->vport
);
547 spin_lock(&tcp_app_lock
);
548 list_for_each_entry(inc
, &tcp_apps
[hash
], p_list
) {
549 if (inc
->port
== cp
->vport
) {
550 if (unlikely(!ip_vs_app_inc_get(inc
)))
552 spin_unlock(&tcp_app_lock
);
554 IP_VS_DBG(9, "%s: Binding conn %u.%u.%u.%u:%u->"
555 "%u.%u.%u.%u:%u to app %s on port %u\n",
557 NIPQUAD(cp
->caddr
), ntohs(cp
->cport
),
558 NIPQUAD(cp
->vaddr
), ntohs(cp
->vport
),
559 inc
->name
, ntohs(inc
->port
));
562 result
= inc
->init_conn(inc
, cp
);
566 spin_unlock(&tcp_app_lock
);
574 * Set LISTEN timeout. (ip_vs_conn_put will setup timer)
576 void ip_vs_tcp_conn_listen(struct ip_vs_conn
*cp
)
578 spin_lock(&cp
->lock
);
579 cp
->state
= IP_VS_TCP_S_LISTEN
;
580 cp
->timeout
= ip_vs_protocol_tcp
.timeout_table
[IP_VS_TCP_S_LISTEN
];
581 spin_unlock(&cp
->lock
);
585 static void ip_vs_tcp_init(struct ip_vs_protocol
*pp
)
587 IP_VS_INIT_HASH_TABLE(tcp_apps
);
588 pp
->timeout_table
= tcp_timeouts
;
592 static void ip_vs_tcp_exit(struct ip_vs_protocol
*pp
)
597 struct ip_vs_protocol ip_vs_protocol_tcp
= {
599 .protocol
= IPPROTO_TCP
,
601 .appcnt
= ATOMIC_INIT(0),
602 .init
= ip_vs_tcp_init
,
603 .exit
= ip_vs_tcp_exit
,
604 .register_app
= tcp_register_app
,
605 .unregister_app
= tcp_unregister_app
,
606 .conn_schedule
= tcp_conn_schedule
,
607 .conn_in_get
= tcp_conn_in_get
,
608 .conn_out_get
= tcp_conn_out_get
,
609 .snat_handler
= tcp_snat_handler
,
610 .dnat_handler
= tcp_dnat_handler
,
611 .csum_check
= tcp_csum_check
,
612 .state_name
= tcp_state_name
,
613 .state_transition
= tcp_state_transition
,
614 .app_conn_bind
= tcp_app_conn_bind
,
615 .debug_packet
= ip_vs_tcpudp_debug_packet
,
616 .timeout_change
= tcp_timeout_change
,
617 .set_state_timeout
= tcp_set_state_timeout
,