2 * ip_vs_proto_tcp.c: TCP load balancing support for IPVS
4 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
5 * Julian Anastasov <ja@ssi.bg>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
12 * Changes: Hans Schillstrom <hans.schillstrom@ericsson.com>
14 * Network name space (netns) aware.
15 * Global data moved to netns i.e struct netns_ipvs
16 * tcp_timeouts table has copy per netns in a hash table per
17 * protocol ip_vs_proto_data and is handled by netns
20 #define KMSG_COMPONENT "IPVS"
21 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
23 #include <linux/kernel.h>
25 #include <linux/tcp.h> /* for tcphdr */
27 #include <net/tcp.h> /* for csum_tcpudp_magic */
28 #include <net/ip6_checksum.h>
29 #include <linux/netfilter.h>
30 #include <linux/netfilter_ipv4.h>
32 #include <net/ip_vs.h>
35 tcp_conn_schedule(struct netns_ipvs
*ipvs
, int af
, struct sk_buff
*skb
,
36 struct ip_vs_proto_data
*pd
,
37 int *verdict
, struct ip_vs_conn
**cpp
,
38 struct ip_vs_iphdr
*iph
)
40 struct ip_vs_service
*svc
;
41 struct tcphdr _tcph
, *th
;
42 __be16 _ports
[2], *ports
= NULL
;
44 /* In the event of icmp, we're only guaranteed to have the first 8
45 * bytes of the transport header, so we only check the rest of the
46 * TCP packet for non-ICMP packets
48 if (likely(!ip_vs_iph_icmp(iph
))) {
49 th
= skb_header_pointer(skb
, iph
->len
, sizeof(_tcph
), &_tcph
);
51 if (th
->rst
|| !(sysctl_sloppy_tcp(ipvs
) || th
->syn
))
56 ports
= skb_header_pointer(
57 skb
, iph
->len
, sizeof(_ports
), &_ports
);
65 /* No !th->ack check to allow scheduling on SYN+ACK for Active FTP */
68 if (likely(!ip_vs_iph_inverse(iph
)))
69 svc
= ip_vs_service_find(ipvs
, af
, skb
->mark
, iph
->protocol
,
70 &iph
->daddr
, ports
[1]);
72 svc
= ip_vs_service_find(ipvs
, af
, skb
->mark
, iph
->protocol
,
73 &iph
->saddr
, ports
[0]);
78 if (ip_vs_todrop(ipvs
)) {
80 * It seems that we are very loaded.
81 * We have to drop this packet :(
89 * Let the virtual server select a real server for the
90 * incoming connection, and create a connection entry.
92 *cpp
= ip_vs_schedule(svc
, skb
, pd
, &ignored
, iph
);
93 if (!*cpp
&& ignored
<= 0) {
95 *verdict
= ip_vs_leave(svc
, skb
, pd
, iph
);
109 tcp_fast_csum_update(int af
, struct tcphdr
*tcph
,
110 const union nf_inet_addr
*oldip
,
111 const union nf_inet_addr
*newip
,
112 __be16 oldport
, __be16 newport
)
114 #ifdef CONFIG_IP_VS_IPV6
117 csum_fold(ip_vs_check_diff16(oldip
->ip6
, newip
->ip6
,
118 ip_vs_check_diff2(oldport
, newport
,
119 ~csum_unfold(tcph
->check
))));
123 csum_fold(ip_vs_check_diff4(oldip
->ip
, newip
->ip
,
124 ip_vs_check_diff2(oldport
, newport
,
125 ~csum_unfold(tcph
->check
))));
130 tcp_partial_csum_update(int af
, struct tcphdr
*tcph
,
131 const union nf_inet_addr
*oldip
,
132 const union nf_inet_addr
*newip
,
133 __be16 oldlen
, __be16 newlen
)
135 #ifdef CONFIG_IP_VS_IPV6
138 ~csum_fold(ip_vs_check_diff16(oldip
->ip6
, newip
->ip6
,
139 ip_vs_check_diff2(oldlen
, newlen
,
140 csum_unfold(tcph
->check
))));
144 ~csum_fold(ip_vs_check_diff4(oldip
->ip
, newip
->ip
,
145 ip_vs_check_diff2(oldlen
, newlen
,
146 csum_unfold(tcph
->check
))));
151 tcp_snat_handler(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
152 struct ip_vs_conn
*cp
, struct ip_vs_iphdr
*iph
)
155 unsigned int tcphoff
= iph
->len
;
157 int payload_csum
= 0;
159 #ifdef CONFIG_IP_VS_IPV6
160 if (cp
->af
== AF_INET6
&& iph
->fragoffs
)
163 oldlen
= skb
->len
- tcphoff
;
165 /* csum_check requires unshared skb */
166 if (!skb_make_writable(skb
, tcphoff
+sizeof(*tcph
)))
169 if (unlikely(cp
->app
!= NULL
)) {
172 /* Some checks before mangling */
173 if (pp
->csum_check
&& !pp
->csum_check(cp
->af
, skb
, pp
))
176 /* Call application helper if needed */
177 if (!(ret
= ip_vs_app_pkt_out(cp
, skb
)))
179 /* ret=2: csum update is needed after payload mangling */
181 oldlen
= skb
->len
- tcphoff
;
186 tcph
= (void *)skb_network_header(skb
) + tcphoff
;
187 tcph
->source
= cp
->vport
;
189 /* Adjust TCP checksums */
190 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
191 tcp_partial_csum_update(cp
->af
, tcph
, &cp
->daddr
, &cp
->vaddr
,
193 htons(skb
->len
- tcphoff
));
194 } else if (!payload_csum
) {
195 /* Only port and addr are changed, do fast csum update */
196 tcp_fast_csum_update(cp
->af
, tcph
, &cp
->daddr
, &cp
->vaddr
,
197 cp
->dport
, cp
->vport
);
198 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
199 skb
->ip_summed
= (cp
->app
&& pp
->csum_check
) ?
200 CHECKSUM_UNNECESSARY
: CHECKSUM_NONE
;
202 /* full checksum calculation */
204 skb
->csum
= skb_checksum(skb
, tcphoff
, skb
->len
- tcphoff
, 0);
205 #ifdef CONFIG_IP_VS_IPV6
206 if (cp
->af
== AF_INET6
)
207 tcph
->check
= csum_ipv6_magic(&cp
->vaddr
.in6
,
210 cp
->protocol
, skb
->csum
);
213 tcph
->check
= csum_tcpudp_magic(cp
->vaddr
.ip
,
218 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
220 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
221 pp
->name
, tcph
->check
,
222 (char*)&(tcph
->check
) - (char*)tcph
);
229 tcp_dnat_handler(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
230 struct ip_vs_conn
*cp
, struct ip_vs_iphdr
*iph
)
233 unsigned int tcphoff
= iph
->len
;
235 int payload_csum
= 0;
237 #ifdef CONFIG_IP_VS_IPV6
238 if (cp
->af
== AF_INET6
&& iph
->fragoffs
)
241 oldlen
= skb
->len
- tcphoff
;
243 /* csum_check requires unshared skb */
244 if (!skb_make_writable(skb
, tcphoff
+sizeof(*tcph
)))
247 if (unlikely(cp
->app
!= NULL
)) {
250 /* Some checks before mangling */
251 if (pp
->csum_check
&& !pp
->csum_check(cp
->af
, skb
, pp
))
255 * Attempt ip_vs_app call.
256 * It will fix ip_vs_conn and iph ack_seq stuff
258 if (!(ret
= ip_vs_app_pkt_in(cp
, skb
)))
260 /* ret=2: csum update is needed after payload mangling */
262 oldlen
= skb
->len
- tcphoff
;
267 tcph
= (void *)skb_network_header(skb
) + tcphoff
;
268 tcph
->dest
= cp
->dport
;
271 * Adjust TCP checksums
273 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
274 tcp_partial_csum_update(cp
->af
, tcph
, &cp
->vaddr
, &cp
->daddr
,
276 htons(skb
->len
- tcphoff
));
277 } else if (!payload_csum
) {
278 /* Only port and addr are changed, do fast csum update */
279 tcp_fast_csum_update(cp
->af
, tcph
, &cp
->vaddr
, &cp
->daddr
,
280 cp
->vport
, cp
->dport
);
281 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
282 skb
->ip_summed
= (cp
->app
&& pp
->csum_check
) ?
283 CHECKSUM_UNNECESSARY
: CHECKSUM_NONE
;
285 /* full checksum calculation */
287 skb
->csum
= skb_checksum(skb
, tcphoff
, skb
->len
- tcphoff
, 0);
288 #ifdef CONFIG_IP_VS_IPV6
289 if (cp
->af
== AF_INET6
)
290 tcph
->check
= csum_ipv6_magic(&cp
->caddr
.in6
,
293 cp
->protocol
, skb
->csum
);
296 tcph
->check
= csum_tcpudp_magic(cp
->caddr
.ip
,
301 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
308 tcp_csum_check(int af
, struct sk_buff
*skb
, struct ip_vs_protocol
*pp
)
310 unsigned int tcphoff
;
312 #ifdef CONFIG_IP_VS_IPV6
314 tcphoff
= sizeof(struct ipv6hdr
);
317 tcphoff
= ip_hdrlen(skb
);
319 switch (skb
->ip_summed
) {
321 skb
->csum
= skb_checksum(skb
, tcphoff
, skb
->len
- tcphoff
, 0);
322 case CHECKSUM_COMPLETE
:
323 #ifdef CONFIG_IP_VS_IPV6
324 if (af
== AF_INET6
) {
325 if (csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
326 &ipv6_hdr(skb
)->daddr
,
328 ipv6_hdr(skb
)->nexthdr
,
330 IP_VS_DBG_RL_PKT(0, af
, pp
, skb
, 0,
331 "Failed checksum for");
336 if (csum_tcpudp_magic(ip_hdr(skb
)->saddr
,
339 ip_hdr(skb
)->protocol
,
341 IP_VS_DBG_RL_PKT(0, af
, pp
, skb
, 0,
342 "Failed checksum for");
347 /* No need to checksum. */
355 #define TCP_DIR_INPUT 0
356 #define TCP_DIR_OUTPUT 4
357 #define TCP_DIR_INPUT_ONLY 8
359 static const int tcp_state_off
[IP_VS_DIR_LAST
] = {
360 [IP_VS_DIR_INPUT
] = TCP_DIR_INPUT
,
361 [IP_VS_DIR_OUTPUT
] = TCP_DIR_OUTPUT
,
362 [IP_VS_DIR_INPUT_ONLY
] = TCP_DIR_INPUT_ONLY
,
366 * Timeout table[state]
368 static const int tcp_timeouts
[IP_VS_TCP_S_LAST
+1] = {
369 [IP_VS_TCP_S_NONE
] = 2*HZ
,
370 [IP_VS_TCP_S_ESTABLISHED
] = 15*60*HZ
,
371 [IP_VS_TCP_S_SYN_SENT
] = 2*60*HZ
,
372 [IP_VS_TCP_S_SYN_RECV
] = 1*60*HZ
,
373 [IP_VS_TCP_S_FIN_WAIT
] = 2*60*HZ
,
374 [IP_VS_TCP_S_TIME_WAIT
] = 2*60*HZ
,
375 [IP_VS_TCP_S_CLOSE
] = 10*HZ
,
376 [IP_VS_TCP_S_CLOSE_WAIT
] = 60*HZ
,
377 [IP_VS_TCP_S_LAST_ACK
] = 30*HZ
,
378 [IP_VS_TCP_S_LISTEN
] = 2*60*HZ
,
379 [IP_VS_TCP_S_SYNACK
] = 120*HZ
,
380 [IP_VS_TCP_S_LAST
] = 2*HZ
,
383 static const char *const tcp_state_name_table
[IP_VS_TCP_S_LAST
+1] = {
384 [IP_VS_TCP_S_NONE
] = "NONE",
385 [IP_VS_TCP_S_ESTABLISHED
] = "ESTABLISHED",
386 [IP_VS_TCP_S_SYN_SENT
] = "SYN_SENT",
387 [IP_VS_TCP_S_SYN_RECV
] = "SYN_RECV",
388 [IP_VS_TCP_S_FIN_WAIT
] = "FIN_WAIT",
389 [IP_VS_TCP_S_TIME_WAIT
] = "TIME_WAIT",
390 [IP_VS_TCP_S_CLOSE
] = "CLOSE",
391 [IP_VS_TCP_S_CLOSE_WAIT
] = "CLOSE_WAIT",
392 [IP_VS_TCP_S_LAST_ACK
] = "LAST_ACK",
393 [IP_VS_TCP_S_LISTEN
] = "LISTEN",
394 [IP_VS_TCP_S_SYNACK
] = "SYNACK",
395 [IP_VS_TCP_S_LAST
] = "BUG!",
398 static const bool tcp_state_active_table
[IP_VS_TCP_S_LAST
] = {
399 [IP_VS_TCP_S_NONE
] = false,
400 [IP_VS_TCP_S_ESTABLISHED
] = true,
401 [IP_VS_TCP_S_SYN_SENT
] = true,
402 [IP_VS_TCP_S_SYN_RECV
] = true,
403 [IP_VS_TCP_S_FIN_WAIT
] = false,
404 [IP_VS_TCP_S_TIME_WAIT
] = false,
405 [IP_VS_TCP_S_CLOSE
] = false,
406 [IP_VS_TCP_S_CLOSE_WAIT
] = false,
407 [IP_VS_TCP_S_LAST_ACK
] = false,
408 [IP_VS_TCP_S_LISTEN
] = false,
409 [IP_VS_TCP_S_SYNACK
] = true,
412 #define sNO IP_VS_TCP_S_NONE
413 #define sES IP_VS_TCP_S_ESTABLISHED
414 #define sSS IP_VS_TCP_S_SYN_SENT
415 #define sSR IP_VS_TCP_S_SYN_RECV
416 #define sFW IP_VS_TCP_S_FIN_WAIT
417 #define sTW IP_VS_TCP_S_TIME_WAIT
418 #define sCL IP_VS_TCP_S_CLOSE
419 #define sCW IP_VS_TCP_S_CLOSE_WAIT
420 #define sLA IP_VS_TCP_S_LAST_ACK
421 #define sLI IP_VS_TCP_S_LISTEN
422 #define sSA IP_VS_TCP_S_SYNACK
424 struct tcp_states_t
{
425 int next_state
[IP_VS_TCP_S_LAST
];
428 static const char * tcp_state_name(int state
)
430 if (state
>= IP_VS_TCP_S_LAST
)
432 return tcp_state_name_table
[state
] ? tcp_state_name_table
[state
] : "?";
435 static bool tcp_state_active(int state
)
437 if (state
>= IP_VS_TCP_S_LAST
)
439 return tcp_state_active_table
[state
];
442 static struct tcp_states_t tcp_states
[] = {
444 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
445 /*syn*/ {{sSR
, sES
, sES
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
}},
446 /*fin*/ {{sCL
, sCW
, sSS
, sTW
, sTW
, sTW
, sCL
, sCW
, sLA
, sLI
, sTW
}},
447 /*ack*/ {{sES
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sES
}},
448 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sSR
}},
451 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
452 /*syn*/ {{sSS
, sES
, sSS
, sSR
, sSS
, sSS
, sSS
, sSS
, sSS
, sLI
, sSR
}},
453 /*fin*/ {{sTW
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sTW
, sLA
, sLI
, sTW
}},
454 /*ack*/ {{sES
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sLA
, sES
, sES
}},
455 /*rst*/ {{sCL
, sCL
, sSS
, sCL
, sCL
, sTW
, sCL
, sCL
, sCL
, sCL
, sCL
}},
458 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
459 /*syn*/ {{sSR
, sES
, sES
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
}},
460 /*fin*/ {{sCL
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sCW
, sLA
, sLI
, sTW
}},
461 /*ack*/ {{sES
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sES
}},
462 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sCL
}},
465 static struct tcp_states_t tcp_states_dos
[] = {
467 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
468 /*syn*/ {{sSR
, sES
, sES
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSR
, sSA
}},
469 /*fin*/ {{sCL
, sCW
, sSS
, sTW
, sTW
, sTW
, sCL
, sCW
, sLA
, sLI
, sSA
}},
470 /*ack*/ {{sES
, sES
, sSS
, sSR
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sSA
}},
471 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sCL
}},
474 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
475 /*syn*/ {{sSS
, sES
, sSS
, sSA
, sSS
, sSS
, sSS
, sSS
, sSS
, sLI
, sSA
}},
476 /*fin*/ {{sTW
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sTW
, sLA
, sLI
, sTW
}},
477 /*ack*/ {{sES
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sLA
, sES
, sES
}},
478 /*rst*/ {{sCL
, sCL
, sSS
, sCL
, sCL
, sTW
, sCL
, sCL
, sCL
, sCL
, sCL
}},
481 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
482 /*syn*/ {{sSA
, sES
, sES
, sSR
, sSA
, sSA
, sSA
, sSA
, sSA
, sSA
, sSA
}},
483 /*fin*/ {{sCL
, sFW
, sSS
, sTW
, sFW
, sTW
, sCL
, sCW
, sLA
, sLI
, sTW
}},
484 /*ack*/ {{sES
, sES
, sSS
, sES
, sFW
, sTW
, sCL
, sCW
, sCL
, sLI
, sES
}},
485 /*rst*/ {{sCL
, sCL
, sCL
, sSR
, sCL
, sCL
, sCL
, sCL
, sLA
, sLI
, sCL
}},
488 static void tcp_timeout_change(struct ip_vs_proto_data
*pd
, int flags
)
490 int on
= (flags
& 1); /* secure_tcp */
493 ** FIXME: change secure_tcp to independent sysctl var
494 ** or make it per-service or per-app because it is valid
495 ** for most if not for all of the applications. Something
496 ** like "capabilities" (flags) for each object.
498 pd
->tcp_state_table
= (on
? tcp_states_dos
: tcp_states
);
501 static inline int tcp_state_idx(struct tcphdr
*th
)
515 set_tcp_state(struct ip_vs_proto_data
*pd
, struct ip_vs_conn
*cp
,
516 int direction
, struct tcphdr
*th
)
519 int new_state
= IP_VS_TCP_S_CLOSE
;
520 int state_off
= tcp_state_off
[direction
];
523 * Update state offset to INPUT_ONLY if necessary
524 * or delete NO_OUTPUT flag if output packet detected
526 if (cp
->flags
& IP_VS_CONN_F_NOOUTPUT
) {
527 if (state_off
== TCP_DIR_OUTPUT
)
528 cp
->flags
&= ~IP_VS_CONN_F_NOOUTPUT
;
530 state_off
= TCP_DIR_INPUT_ONLY
;
533 if ((state_idx
= tcp_state_idx(th
)) < 0) {
534 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx
);
539 pd
->tcp_state_table
[state_off
+state_idx
].next_state
[cp
->state
];
542 if (new_state
!= cp
->state
) {
543 struct ip_vs_dest
*dest
= cp
->dest
;
545 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->"
546 "%s:%d state: %s->%s conn->refcnt:%d\n",
548 ((state_off
== TCP_DIR_OUTPUT
) ?
549 "output " : "input "),
554 IP_VS_DBG_ADDR(cp
->daf
, &cp
->daddr
),
556 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
),
558 tcp_state_name(cp
->state
),
559 tcp_state_name(new_state
),
560 refcount_read(&cp
->refcnt
));
563 if (!(cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
564 !tcp_state_active(new_state
)) {
565 atomic_dec(&dest
->activeconns
);
566 atomic_inc(&dest
->inactconns
);
567 cp
->flags
|= IP_VS_CONN_F_INACTIVE
;
568 } else if ((cp
->flags
& IP_VS_CONN_F_INACTIVE
) &&
569 tcp_state_active(new_state
)) {
570 atomic_inc(&dest
->activeconns
);
571 atomic_dec(&dest
->inactconns
);
572 cp
->flags
&= ~IP_VS_CONN_F_INACTIVE
;
578 cp
->timeout
= pd
->timeout_table
[cp
->state
= new_state
];
579 else /* What to do ? */
580 cp
->timeout
= tcp_timeouts
[cp
->state
= new_state
];
584 * Handle state transitions
587 tcp_state_transition(struct ip_vs_conn
*cp
, int direction
,
588 const struct sk_buff
*skb
,
589 struct ip_vs_proto_data
*pd
)
591 struct tcphdr _tcph
, *th
;
593 #ifdef CONFIG_IP_VS_IPV6
594 int ihl
= cp
->af
== AF_INET
? ip_hdrlen(skb
) : sizeof(struct ipv6hdr
);
596 int ihl
= ip_hdrlen(skb
);
599 th
= skb_header_pointer(skb
, ihl
, sizeof(_tcph
), &_tcph
);
603 spin_lock_bh(&cp
->lock
);
604 set_tcp_state(pd
, cp
, direction
, th
);
605 spin_unlock_bh(&cp
->lock
);
608 static inline __u16
tcp_app_hashkey(__be16 port
)
610 return (((__force u16
)port
>> TCP_APP_TAB_BITS
) ^ (__force u16
)port
)
615 static int tcp_register_app(struct netns_ipvs
*ipvs
, struct ip_vs_app
*inc
)
619 __be16 port
= inc
->port
;
621 struct ip_vs_proto_data
*pd
= ip_vs_proto_data_get(ipvs
, IPPROTO_TCP
);
623 hash
= tcp_app_hashkey(port
);
625 list_for_each_entry(i
, &ipvs
->tcp_apps
[hash
], p_list
) {
626 if (i
->port
== port
) {
631 list_add_rcu(&inc
->p_list
, &ipvs
->tcp_apps
[hash
]);
632 atomic_inc(&pd
->appcnt
);
640 tcp_unregister_app(struct netns_ipvs
*ipvs
, struct ip_vs_app
*inc
)
642 struct ip_vs_proto_data
*pd
= ip_vs_proto_data_get(ipvs
, IPPROTO_TCP
);
644 atomic_dec(&pd
->appcnt
);
645 list_del_rcu(&inc
->p_list
);
650 tcp_app_conn_bind(struct ip_vs_conn
*cp
)
652 struct netns_ipvs
*ipvs
= cp
->ipvs
;
654 struct ip_vs_app
*inc
;
657 /* Default binding: bind app only for NAT */
658 if (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)
661 /* Lookup application incarnations and bind the right one */
662 hash
= tcp_app_hashkey(cp
->vport
);
665 list_for_each_entry_rcu(inc
, &ipvs
->tcp_apps
[hash
], p_list
) {
666 if (inc
->port
== cp
->vport
) {
667 if (unlikely(!ip_vs_app_inc_get(inc
)))
671 IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
672 "%s:%u to app %s on port %u\n",
674 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
),
676 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
),
678 inc
->name
, ntohs(inc
->port
));
682 result
= inc
->init_conn(inc
, cp
);
694 * Set LISTEN timeout. (ip_vs_conn_put will setup timer)
696 void ip_vs_tcp_conn_listen(struct ip_vs_conn
*cp
)
698 struct ip_vs_proto_data
*pd
= ip_vs_proto_data_get(cp
->ipvs
, IPPROTO_TCP
);
700 spin_lock_bh(&cp
->lock
);
701 cp
->state
= IP_VS_TCP_S_LISTEN
;
702 cp
->timeout
= (pd
? pd
->timeout_table
[IP_VS_TCP_S_LISTEN
]
703 : tcp_timeouts
[IP_VS_TCP_S_LISTEN
]);
704 spin_unlock_bh(&cp
->lock
);
707 /* ---------------------------------------------
708 * timeouts is netns related now.
709 * ---------------------------------------------
711 static int __ip_vs_tcp_init(struct netns_ipvs
*ipvs
, struct ip_vs_proto_data
*pd
)
713 ip_vs_init_hash_table(ipvs
->tcp_apps
, TCP_APP_TAB_SIZE
);
714 pd
->timeout_table
= ip_vs_create_timeout_table((int *)tcp_timeouts
,
715 sizeof(tcp_timeouts
));
716 if (!pd
->timeout_table
)
718 pd
->tcp_state_table
= tcp_states
;
722 static void __ip_vs_tcp_exit(struct netns_ipvs
*ipvs
, struct ip_vs_proto_data
*pd
)
724 kfree(pd
->timeout_table
);
728 struct ip_vs_protocol ip_vs_protocol_tcp
= {
730 .protocol
= IPPROTO_TCP
,
731 .num_states
= IP_VS_TCP_S_LAST
,
735 .init_netns
= __ip_vs_tcp_init
,
736 .exit_netns
= __ip_vs_tcp_exit
,
737 .register_app
= tcp_register_app
,
738 .unregister_app
= tcp_unregister_app
,
739 .conn_schedule
= tcp_conn_schedule
,
740 .conn_in_get
= ip_vs_conn_in_get_proto
,
741 .conn_out_get
= ip_vs_conn_out_get_proto
,
742 .snat_handler
= tcp_snat_handler
,
743 .dnat_handler
= tcp_dnat_handler
,
744 .csum_check
= tcp_csum_check
,
745 .state_name
= tcp_state_name
,
746 .state_transition
= tcp_state_transition
,
747 .app_conn_bind
= tcp_app_conn_bind
,
748 .debug_packet
= ip_vs_tcpudp_debug_packet
,
749 .timeout_change
= tcp_timeout_change
,