2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the Netfilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
19 * and others. Many code here is taken from IP MASQ code of kernel 2.2.
25 #include <linux/interrupt.h>
27 #include <linux/net.h>
28 #include <linux/kernel.h>
29 #include <linux/module.h>
30 #include <linux/vmalloc.h>
31 #include <linux/proc_fs.h> /* for proc_net_* */
32 #include <linux/seq_file.h>
33 #include <linux/jhash.h>
34 #include <linux/random.h>
36 #include <net/net_namespace.h>
37 #include <net/ip_vs.h>
41 * Connection hash table: for input and output packets lookups of IPVS
43 static struct list_head
*ip_vs_conn_tab
;
45 /* SLAB cache for IPVS connections */
46 static struct kmem_cache
*ip_vs_conn_cachep __read_mostly
;
48 /* counter for current IPVS connections */
49 static atomic_t ip_vs_conn_count
= ATOMIC_INIT(0);
51 /* counter for no client port connections */
52 static atomic_t ip_vs_conn_no_cport_cnt
= ATOMIC_INIT(0);
54 /* random value for IPVS connection hash */
55 static unsigned int ip_vs_conn_rnd
;
58 * Fine locking granularity for big connection hash table
60 #define CT_LOCKARRAY_BITS 4
61 #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
62 #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
64 struct ip_vs_aligned_lock
67 } __attribute__((__aligned__(SMP_CACHE_BYTES
)));
69 /* lock array for conn table */
70 static struct ip_vs_aligned_lock
71 __ip_vs_conntbl_lock_array
[CT_LOCKARRAY_SIZE
] __cacheline_aligned
;
73 static inline void ct_read_lock(unsigned key
)
75 read_lock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
78 static inline void ct_read_unlock(unsigned key
)
80 read_unlock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
83 static inline void ct_write_lock(unsigned key
)
85 write_lock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
88 static inline void ct_write_unlock(unsigned key
)
90 write_unlock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
93 static inline void ct_read_lock_bh(unsigned key
)
95 read_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
98 static inline void ct_read_unlock_bh(unsigned key
)
100 read_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
103 static inline void ct_write_lock_bh(unsigned key
)
105 write_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
108 static inline void ct_write_unlock_bh(unsigned key
)
110 write_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
115 * Returns hash value for IPVS connection entry
117 static unsigned int ip_vs_conn_hashkey(int af
, unsigned proto
,
118 const union nf_inet_addr
*addr
,
121 #ifdef CONFIG_IP_VS_IPV6
123 return jhash_3words(jhash(addr
, 16, ip_vs_conn_rnd
),
124 (__force u32
)port
, proto
, ip_vs_conn_rnd
)
125 & IP_VS_CONN_TAB_MASK
;
127 return jhash_3words((__force u32
)addr
->ip
, (__force u32
)port
, proto
,
129 & IP_VS_CONN_TAB_MASK
;
134 * Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
135 * returns bool success.
137 static inline int ip_vs_conn_hash(struct ip_vs_conn
*cp
)
142 /* Hash by protocol, client address and port */
143 hash
= ip_vs_conn_hashkey(cp
->af
, cp
->protocol
, &cp
->caddr
, cp
->cport
);
147 if (!(cp
->flags
& IP_VS_CONN_F_HASHED
)) {
148 list_add(&cp
->c_list
, &ip_vs_conn_tab
[hash
]);
149 cp
->flags
|= IP_VS_CONN_F_HASHED
;
150 atomic_inc(&cp
->refcnt
);
153 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
154 "called from %p\n", __builtin_return_address(0));
158 ct_write_unlock(hash
);
165 * UNhashes ip_vs_conn from ip_vs_conn_tab.
166 * returns bool success.
168 static inline int ip_vs_conn_unhash(struct ip_vs_conn
*cp
)
173 /* unhash it and decrease its reference counter */
174 hash
= ip_vs_conn_hashkey(cp
->af
, cp
->protocol
, &cp
->caddr
, cp
->cport
);
178 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
179 list_del(&cp
->c_list
);
180 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
181 atomic_dec(&cp
->refcnt
);
186 ct_write_unlock(hash
);
193 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
194 * Called for pkts coming from OUTside-to-INside.
195 * s_addr, s_port: pkt source address (foreign host)
196 * d_addr, d_port: pkt dest address (load balancer)
198 static inline struct ip_vs_conn
*__ip_vs_conn_in_get
199 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
200 const union nf_inet_addr
*d_addr
, __be16 d_port
)
203 struct ip_vs_conn
*cp
;
205 hash
= ip_vs_conn_hashkey(af
, protocol
, s_addr
, s_port
);
209 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
211 ip_vs_addr_equal(af
, s_addr
, &cp
->caddr
) &&
212 ip_vs_addr_equal(af
, d_addr
, &cp
->vaddr
) &&
213 s_port
== cp
->cport
&& d_port
== cp
->vport
&&
214 ((!s_port
) ^ (!(cp
->flags
& IP_VS_CONN_F_NO_CPORT
))) &&
215 protocol
== cp
->protocol
) {
217 atomic_inc(&cp
->refcnt
);
218 ct_read_unlock(hash
);
223 ct_read_unlock(hash
);
228 struct ip_vs_conn
*ip_vs_conn_in_get
229 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
230 const union nf_inet_addr
*d_addr
, __be16 d_port
)
232 struct ip_vs_conn
*cp
;
234 cp
= __ip_vs_conn_in_get(af
, protocol
, s_addr
, s_port
, d_addr
, d_port
);
235 if (!cp
&& atomic_read(&ip_vs_conn_no_cport_cnt
))
236 cp
= __ip_vs_conn_in_get(af
, protocol
, s_addr
, 0, d_addr
,
239 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
240 ip_vs_proto_name(protocol
),
241 IP_VS_DBG_ADDR(af
, s_addr
), ntohs(s_port
),
242 IP_VS_DBG_ADDR(af
, d_addr
), ntohs(d_port
),
243 cp
? "hit" : "not hit");
248 /* Get reference to connection template */
249 struct ip_vs_conn
*ip_vs_ct_in_get
250 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
251 const union nf_inet_addr
*d_addr
, __be16 d_port
)
254 struct ip_vs_conn
*cp
;
256 hash
= ip_vs_conn_hashkey(af
, protocol
, s_addr
, s_port
);
260 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
262 ip_vs_addr_equal(af
, s_addr
, &cp
->caddr
) &&
263 ip_vs_addr_equal(af
, d_addr
, &cp
->vaddr
) &&
264 s_port
== cp
->cport
&& d_port
== cp
->vport
&&
265 cp
->flags
& IP_VS_CONN_F_TEMPLATE
&&
266 protocol
== cp
->protocol
) {
268 atomic_inc(&cp
->refcnt
);
275 ct_read_unlock(hash
);
277 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
278 ip_vs_proto_name(protocol
),
279 IP_VS_DBG_ADDR(af
, s_addr
), ntohs(s_port
),
280 IP_VS_DBG_ADDR(af
, d_addr
), ntohs(d_port
),
281 cp
? "hit" : "not hit");
287 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
288 * Called for pkts coming from inside-to-OUTside.
289 * s_addr, s_port: pkt source address (inside host)
290 * d_addr, d_port: pkt dest address (foreign host)
292 struct ip_vs_conn
*ip_vs_conn_out_get
293 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
294 const union nf_inet_addr
*d_addr
, __be16 d_port
)
297 struct ip_vs_conn
*cp
, *ret
=NULL
;
300 * Check for "full" addressed entries
302 hash
= ip_vs_conn_hashkey(af
, protocol
, d_addr
, d_port
);
306 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
308 ip_vs_addr_equal(af
, d_addr
, &cp
->caddr
) &&
309 ip_vs_addr_equal(af
, s_addr
, &cp
->daddr
) &&
310 d_port
== cp
->cport
&& s_port
== cp
->dport
&&
311 protocol
== cp
->protocol
) {
313 atomic_inc(&cp
->refcnt
);
319 ct_read_unlock(hash
);
321 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
322 ip_vs_proto_name(protocol
),
323 IP_VS_DBG_ADDR(af
, s_addr
), ntohs(s_port
),
324 IP_VS_DBG_ADDR(af
, d_addr
), ntohs(d_port
),
325 ret
? "hit" : "not hit");
332 * Put back the conn and restart its timer with its timeout
334 void ip_vs_conn_put(struct ip_vs_conn
*cp
)
336 /* reset it expire in its timeout */
337 mod_timer(&cp
->timer
, jiffies
+cp
->timeout
);
339 __ip_vs_conn_put(cp
);
344 * Fill a no_client_port connection with a client port number
346 void ip_vs_conn_fill_cport(struct ip_vs_conn
*cp
, __be16 cport
)
348 if (ip_vs_conn_unhash(cp
)) {
349 spin_lock(&cp
->lock
);
350 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
) {
351 atomic_dec(&ip_vs_conn_no_cport_cnt
);
352 cp
->flags
&= ~IP_VS_CONN_F_NO_CPORT
;
355 spin_unlock(&cp
->lock
);
357 /* hash on new dport */
364 * Bind a connection entry with the corresponding packet_xmit.
365 * Called by ip_vs_conn_new.
367 static inline void ip_vs_bind_xmit(struct ip_vs_conn
*cp
)
369 switch (IP_VS_FWD_METHOD(cp
)) {
370 case IP_VS_CONN_F_MASQ
:
371 cp
->packet_xmit
= ip_vs_nat_xmit
;
374 case IP_VS_CONN_F_TUNNEL
:
375 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
378 case IP_VS_CONN_F_DROUTE
:
379 cp
->packet_xmit
= ip_vs_dr_xmit
;
382 case IP_VS_CONN_F_LOCALNODE
:
383 cp
->packet_xmit
= ip_vs_null_xmit
;
386 case IP_VS_CONN_F_BYPASS
:
387 cp
->packet_xmit
= ip_vs_bypass_xmit
;
393 static inline int ip_vs_dest_totalconns(struct ip_vs_dest
*dest
)
395 return atomic_read(&dest
->activeconns
)
396 + atomic_read(&dest
->inactconns
);
400 * Bind a connection entry with a virtual service destination
401 * Called just after a new connection entry is created.
404 ip_vs_bind_dest(struct ip_vs_conn
*cp
, struct ip_vs_dest
*dest
)
406 /* if dest is NULL, then return directly */
410 /* Increase the refcnt counter of the dest */
411 atomic_inc(&dest
->refcnt
);
413 /* Bind with the destination and its corresponding transmitter */
414 if ((cp
->flags
& IP_VS_CONN_F_SYNC
) &&
415 (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)))
416 /* if the connection is not template and is created
417 * by sync, preserve the activity flag.
419 cp
->flags
|= atomic_read(&dest
->conn_flags
) &
420 (~IP_VS_CONN_F_INACTIVE
);
422 cp
->flags
|= atomic_read(&dest
->conn_flags
);
425 IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
426 "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
428 ip_vs_proto_name(cp
->protocol
),
429 NIPQUAD(cp
->caddr
.ip
), ntohs(cp
->cport
),
430 NIPQUAD(cp
->vaddr
.ip
), ntohs(cp
->vport
),
431 NIPQUAD(cp
->daddr
.ip
), ntohs(cp
->dport
),
432 ip_vs_fwd_tag(cp
), cp
->state
,
433 cp
->flags
, atomic_read(&cp
->refcnt
),
434 atomic_read(&dest
->refcnt
));
436 /* Update the connection counters */
437 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
438 /* It is a normal connection, so increase the inactive
439 connection counter because it is in TCP SYNRECV
440 state (inactive) or other protocol inacive state */
441 if ((cp
->flags
& IP_VS_CONN_F_SYNC
) &&
442 (!(cp
->flags
& IP_VS_CONN_F_INACTIVE
)))
443 atomic_inc(&dest
->activeconns
);
445 atomic_inc(&dest
->inactconns
);
447 /* It is a persistent connection/template, so increase
448 the peristent connection counter */
449 atomic_inc(&dest
->persistconns
);
452 if (dest
->u_threshold
!= 0 &&
453 ip_vs_dest_totalconns(dest
) >= dest
->u_threshold
)
454 dest
->flags
|= IP_VS_DEST_F_OVERLOAD
;
459 * Check if there is a destination for the connection, if so
460 * bind the connection to the destination.
462 struct ip_vs_dest
*ip_vs_try_bind_dest(struct ip_vs_conn
*cp
)
464 struct ip_vs_dest
*dest
;
466 if ((cp
) && (!cp
->dest
)) {
467 dest
= ip_vs_find_dest(cp
->daddr
.ip
, cp
->dport
,
468 cp
->vaddr
.ip
, cp
->vport
, cp
->protocol
);
469 ip_vs_bind_dest(cp
, dest
);
477 * Unbind a connection entry with its VS destination
478 * Called by the ip_vs_conn_expire function.
480 static inline void ip_vs_unbind_dest(struct ip_vs_conn
*cp
)
482 struct ip_vs_dest
*dest
= cp
->dest
;
487 IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
488 "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
490 ip_vs_proto_name(cp
->protocol
),
491 NIPQUAD(cp
->caddr
.ip
), ntohs(cp
->cport
),
492 NIPQUAD(cp
->vaddr
.ip
), ntohs(cp
->vport
),
493 NIPQUAD(cp
->daddr
.ip
), ntohs(cp
->dport
),
494 ip_vs_fwd_tag(cp
), cp
->state
,
495 cp
->flags
, atomic_read(&cp
->refcnt
),
496 atomic_read(&dest
->refcnt
));
498 /* Update the connection counters */
499 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
500 /* It is a normal connection, so decrease the inactconns
501 or activeconns counter */
502 if (cp
->flags
& IP_VS_CONN_F_INACTIVE
) {
503 atomic_dec(&dest
->inactconns
);
505 atomic_dec(&dest
->activeconns
);
508 /* It is a persistent connection/template, so decrease
509 the peristent connection counter */
510 atomic_dec(&dest
->persistconns
);
513 if (dest
->l_threshold
!= 0) {
514 if (ip_vs_dest_totalconns(dest
) < dest
->l_threshold
)
515 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
516 } else if (dest
->u_threshold
!= 0) {
517 if (ip_vs_dest_totalconns(dest
) * 4 < dest
->u_threshold
* 3)
518 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
520 if (dest
->flags
& IP_VS_DEST_F_OVERLOAD
)
521 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
525 * Simply decrease the refcnt of the dest, because the
526 * dest will be either in service's destination list
529 atomic_dec(&dest
->refcnt
);
534 * Checking if the destination of a connection template is available.
535 * If available, return 1, otherwise invalidate this connection
536 * template and return 0.
538 int ip_vs_check_template(struct ip_vs_conn
*ct
)
540 struct ip_vs_dest
*dest
= ct
->dest
;
543 * Checking the dest server status.
545 if ((dest
== NULL
) ||
546 !(dest
->flags
& IP_VS_DEST_F_AVAILABLE
) ||
547 (sysctl_ip_vs_expire_quiescent_template
&&
548 (atomic_read(&dest
->weight
) == 0))) {
549 IP_VS_DBG(9, "check_template: dest not available for "
550 "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
551 "-> d:%u.%u.%u.%u:%d\n",
552 ip_vs_proto_name(ct
->protocol
),
553 NIPQUAD(ct
->caddr
.ip
), ntohs(ct
->cport
),
554 NIPQUAD(ct
->vaddr
.ip
), ntohs(ct
->vport
),
555 NIPQUAD(ct
->daddr
.ip
), ntohs(ct
->dport
));
558 * Invalidate the connection template
560 if (ct
->vport
!= htons(0xffff)) {
561 if (ip_vs_conn_unhash(ct
)) {
562 ct
->dport
= htons(0xffff);
563 ct
->vport
= htons(0xffff);
570 * Simply decrease the refcnt of the template,
571 * don't restart its timer.
573 atomic_dec(&ct
->refcnt
);
579 static void ip_vs_conn_expire(unsigned long data
)
581 struct ip_vs_conn
*cp
= (struct ip_vs_conn
*)data
;
588 atomic_inc(&cp
->refcnt
);
591 * do I control anybody?
593 if (atomic_read(&cp
->n_control
))
597 * unhash it if it is hashed in the conn table
599 if (!ip_vs_conn_unhash(cp
))
603 * refcnt==1 implies I'm the only one referrer
605 if (likely(atomic_read(&cp
->refcnt
) == 1)) {
606 /* delete the timer if it is activated by other users */
607 if (timer_pending(&cp
->timer
))
608 del_timer(&cp
->timer
);
610 /* does anybody control me? */
612 ip_vs_control_del(cp
);
614 if (unlikely(cp
->app
!= NULL
))
615 ip_vs_unbind_app(cp
);
616 ip_vs_unbind_dest(cp
);
617 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
)
618 atomic_dec(&ip_vs_conn_no_cport_cnt
);
619 atomic_dec(&ip_vs_conn_count
);
621 kmem_cache_free(ip_vs_conn_cachep
, cp
);
625 /* hash it back to the table */
629 IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
630 atomic_read(&cp
->refcnt
)-1,
631 atomic_read(&cp
->n_control
));
637 void ip_vs_conn_expire_now(struct ip_vs_conn
*cp
)
639 if (del_timer(&cp
->timer
))
640 mod_timer(&cp
->timer
, jiffies
);
645 * Create a new connection entry and hash it into the ip_vs_conn_tab
648 ip_vs_conn_new(int af
, int proto
, const union nf_inet_addr
*caddr
, __be16 cport
,
649 const union nf_inet_addr
*vaddr
, __be16 vport
,
650 const union nf_inet_addr
*daddr
, __be16 dport
, unsigned flags
,
651 struct ip_vs_dest
*dest
)
653 struct ip_vs_conn
*cp
;
654 struct ip_vs_protocol
*pp
= ip_vs_proto_get(proto
);
656 cp
= kmem_cache_zalloc(ip_vs_conn_cachep
, GFP_ATOMIC
);
658 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
662 INIT_LIST_HEAD(&cp
->c_list
);
663 setup_timer(&cp
->timer
, ip_vs_conn_expire
, (unsigned long)cp
);
665 cp
->protocol
= proto
;
666 ip_vs_addr_copy(af
, &cp
->caddr
, caddr
);
668 ip_vs_addr_copy(af
, &cp
->vaddr
, vaddr
);
670 ip_vs_addr_copy(af
, &cp
->daddr
, daddr
);
673 spin_lock_init(&cp
->lock
);
676 * Set the entry is referenced by the current thread before hashing
677 * it in the table, so that other thread run ip_vs_random_dropentry
678 * but cannot drop this entry.
680 atomic_set(&cp
->refcnt
, 1);
682 atomic_set(&cp
->n_control
, 0);
683 atomic_set(&cp
->in_pkts
, 0);
685 atomic_inc(&ip_vs_conn_count
);
686 if (flags
& IP_VS_CONN_F_NO_CPORT
)
687 atomic_inc(&ip_vs_conn_no_cport_cnt
);
689 /* Bind the connection with a destination server */
690 ip_vs_bind_dest(cp
, dest
);
692 /* Set its state and timeout */
696 /* Bind its packet transmitter */
699 if (unlikely(pp
&& atomic_read(&pp
->appcnt
)))
700 ip_vs_bind_app(cp
, pp
);
702 /* Hash it in the ip_vs_conn_tab finally */
710 * /proc/net/ip_vs_conn entries
712 #ifdef CONFIG_PROC_FS
714 static void *ip_vs_conn_array(struct seq_file
*seq
, loff_t pos
)
717 struct ip_vs_conn
*cp
;
719 for(idx
= 0; idx
< IP_VS_CONN_TAB_SIZE
; idx
++) {
720 ct_read_lock_bh(idx
);
721 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
723 seq
->private = &ip_vs_conn_tab
[idx
];
727 ct_read_unlock_bh(idx
);
733 static void *ip_vs_conn_seq_start(struct seq_file
*seq
, loff_t
*pos
)
736 return *pos
? ip_vs_conn_array(seq
, *pos
- 1) :SEQ_START_TOKEN
;
739 static void *ip_vs_conn_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
741 struct ip_vs_conn
*cp
= v
;
742 struct list_head
*e
, *l
= seq
->private;
746 if (v
== SEQ_START_TOKEN
)
747 return ip_vs_conn_array(seq
, 0);
749 /* more on same hash chain? */
750 if ((e
= cp
->c_list
.next
) != l
)
751 return list_entry(e
, struct ip_vs_conn
, c_list
);
753 idx
= l
- ip_vs_conn_tab
;
754 ct_read_unlock_bh(idx
);
756 while (++idx
< IP_VS_CONN_TAB_SIZE
) {
757 ct_read_lock_bh(idx
);
758 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
759 seq
->private = &ip_vs_conn_tab
[idx
];
762 ct_read_unlock_bh(idx
);
768 static void ip_vs_conn_seq_stop(struct seq_file
*seq
, void *v
)
770 struct list_head
*l
= seq
->private;
773 ct_read_unlock_bh(l
- ip_vs_conn_tab
);
776 static int ip_vs_conn_seq_show(struct seq_file
*seq
, void *v
)
779 if (v
== SEQ_START_TOKEN
)
781 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires\n");
783 const struct ip_vs_conn
*cp
= v
;
786 "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n",
787 ip_vs_proto_name(cp
->protocol
),
788 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
789 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
790 ntohl(cp
->daddr
.ip
), ntohs(cp
->dport
),
791 ip_vs_state_name(cp
->protocol
, cp
->state
),
792 (cp
->timer
.expires
-jiffies
)/HZ
);
797 static const struct seq_operations ip_vs_conn_seq_ops
= {
798 .start
= ip_vs_conn_seq_start
,
799 .next
= ip_vs_conn_seq_next
,
800 .stop
= ip_vs_conn_seq_stop
,
801 .show
= ip_vs_conn_seq_show
,
804 static int ip_vs_conn_open(struct inode
*inode
, struct file
*file
)
806 return seq_open(file
, &ip_vs_conn_seq_ops
);
809 static const struct file_operations ip_vs_conn_fops
= {
810 .owner
= THIS_MODULE
,
811 .open
= ip_vs_conn_open
,
814 .release
= seq_release
,
817 static const char *ip_vs_origin_name(unsigned flags
)
819 if (flags
& IP_VS_CONN_F_SYNC
)
825 static int ip_vs_conn_sync_seq_show(struct seq_file
*seq
, void *v
)
828 if (v
== SEQ_START_TOKEN
)
830 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
832 const struct ip_vs_conn
*cp
= v
;
835 "%-3s %08X %04X %08X %04X %08X %04X %-11s %-6s %7lu\n",
836 ip_vs_proto_name(cp
->protocol
),
837 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
838 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
839 ntohl(cp
->daddr
.ip
), ntohs(cp
->dport
),
840 ip_vs_state_name(cp
->protocol
, cp
->state
),
841 ip_vs_origin_name(cp
->flags
),
842 (cp
->timer
.expires
-jiffies
)/HZ
);
847 static const struct seq_operations ip_vs_conn_sync_seq_ops
= {
848 .start
= ip_vs_conn_seq_start
,
849 .next
= ip_vs_conn_seq_next
,
850 .stop
= ip_vs_conn_seq_stop
,
851 .show
= ip_vs_conn_sync_seq_show
,
854 static int ip_vs_conn_sync_open(struct inode
*inode
, struct file
*file
)
856 return seq_open(file
, &ip_vs_conn_sync_seq_ops
);
859 static const struct file_operations ip_vs_conn_sync_fops
= {
860 .owner
= THIS_MODULE
,
861 .open
= ip_vs_conn_sync_open
,
864 .release
= seq_release
,
871 * Randomly drop connection entries before running out of memory
873 static inline int todrop_entry(struct ip_vs_conn
*cp
)
876 * The drop rate array needs tuning for real environments.
877 * Called from timer bh only => no locking
879 static const char todrop_rate
[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
880 static char todrop_counter
[9] = {0};
883 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
884 This will leave enough time for normal connection to get
886 if (time_before(cp
->timeout
+ jiffies
, cp
->timer
.expires
+ 60*HZ
))
889 /* Don't drop the entry if its number of incoming packets is not
891 i
= atomic_read(&cp
->in_pkts
);
892 if (i
> 8 || i
< 0) return 0;
894 if (!todrop_rate
[i
]) return 0;
895 if (--todrop_counter
[i
] > 0) return 0;
897 todrop_counter
[i
] = todrop_rate
[i
];
901 /* Called from keventd and must protect itself from softirqs */
902 void ip_vs_random_dropentry(void)
905 struct ip_vs_conn
*cp
;
908 * Randomly scan 1/32 of the whole table every second
910 for (idx
= 0; idx
< (IP_VS_CONN_TAB_SIZE
>>5); idx
++) {
911 unsigned hash
= net_random() & IP_VS_CONN_TAB_MASK
;
914 * Lock is actually needed in this loop.
916 ct_write_lock_bh(hash
);
918 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
919 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
920 /* connection template */
923 if (cp
->protocol
== IPPROTO_TCP
) {
925 case IP_VS_TCP_S_SYN_RECV
:
926 case IP_VS_TCP_S_SYNACK
:
929 case IP_VS_TCP_S_ESTABLISHED
:
930 if (todrop_entry(cp
))
938 if (!todrop_entry(cp
))
942 IP_VS_DBG(4, "del connection\n");
943 ip_vs_conn_expire_now(cp
);
945 IP_VS_DBG(4, "del conn template\n");
946 ip_vs_conn_expire_now(cp
->control
);
949 ct_write_unlock_bh(hash
);
955 * Flush all the connection entries in the ip_vs_conn_tab
957 static void ip_vs_conn_flush(void)
960 struct ip_vs_conn
*cp
;
963 for (idx
=0; idx
<IP_VS_CONN_TAB_SIZE
; idx
++) {
965 * Lock is actually needed in this loop.
967 ct_write_lock_bh(idx
);
969 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
971 IP_VS_DBG(4, "del connection\n");
972 ip_vs_conn_expire_now(cp
);
974 IP_VS_DBG(4, "del conn template\n");
975 ip_vs_conn_expire_now(cp
->control
);
978 ct_write_unlock_bh(idx
);
981 /* the counter may be not NULL, because maybe some conn entries
982 are run by slow timer handler or unhashed but still referred */
983 if (atomic_read(&ip_vs_conn_count
) != 0) {
990 int __init
ip_vs_conn_init(void)
995 * Allocate the connection hash table and initialize its list heads
997 ip_vs_conn_tab
= vmalloc(IP_VS_CONN_TAB_SIZE
*sizeof(struct list_head
));
1001 /* Allocate ip_vs_conn slab cache */
1002 ip_vs_conn_cachep
= kmem_cache_create("ip_vs_conn",
1003 sizeof(struct ip_vs_conn
), 0,
1004 SLAB_HWCACHE_ALIGN
, NULL
);
1005 if (!ip_vs_conn_cachep
) {
1006 vfree(ip_vs_conn_tab
);
1010 IP_VS_INFO("Connection hash table configured "
1011 "(size=%d, memory=%ldKbytes)\n",
1012 IP_VS_CONN_TAB_SIZE
,
1013 (long)(IP_VS_CONN_TAB_SIZE
*sizeof(struct list_head
))/1024);
1014 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1015 sizeof(struct ip_vs_conn
));
1017 for (idx
= 0; idx
< IP_VS_CONN_TAB_SIZE
; idx
++) {
1018 INIT_LIST_HEAD(&ip_vs_conn_tab
[idx
]);
1021 for (idx
= 0; idx
< CT_LOCKARRAY_SIZE
; idx
++) {
1022 rwlock_init(&__ip_vs_conntbl_lock_array
[idx
].l
);
1025 proc_net_fops_create(&init_net
, "ip_vs_conn", 0, &ip_vs_conn_fops
);
1026 proc_net_fops_create(&init_net
, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops
);
1028 /* calculate the random value for connection hash */
1029 get_random_bytes(&ip_vs_conn_rnd
, sizeof(ip_vs_conn_rnd
));
1035 void ip_vs_conn_cleanup(void)
1037 /* flush all the connection entries first */
1040 /* Release the empty cache */
1041 kmem_cache_destroy(ip_vs_conn_cachep
);
1042 proc_net_remove(&init_net
, "ip_vs_conn");
1043 proc_net_remove(&init_net
, "ip_vs_conn_sync");
1044 vfree(ip_vs_conn_tab
);