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 * Version: $Id: ip_vs_conn.c,v 1.31 2003/04/18 09:03:16 wensong Exp $
10 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
11 * Peter Kese <peter.kese@ijs.si>
12 * Julian Anastasov <ja@ssi.bg>
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License
16 * as published by the Free Software Foundation; either version
17 * 2 of the License, or (at your option) any later version.
19 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
20 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
21 * and others. Many code here is taken from IP MASQ code of kernel 2.2.
27 #include <linux/kernel.h>
28 #include <linux/vmalloc.h>
29 #include <linux/proc_fs.h> /* for proc_net_* */
30 #include <linux/seq_file.h>
31 #include <linux/jhash.h>
32 #include <linux/random.h>
34 #include <net/ip_vs.h>
38 * Connection hash table: for input and output packets lookups of IPVS
40 static struct list_head
*ip_vs_conn_tab
;
42 /* SLAB cache for IPVS connections */
43 static kmem_cache_t
*ip_vs_conn_cachep __read_mostly
;
45 /* counter for current IPVS connections */
46 static atomic_t ip_vs_conn_count
= ATOMIC_INIT(0);
48 /* counter for no client port connections */
49 static atomic_t ip_vs_conn_no_cport_cnt
= ATOMIC_INIT(0);
51 /* random value for IPVS connection hash */
52 static unsigned int ip_vs_conn_rnd
;
55 * Fine locking granularity for big connection hash table
57 #define CT_LOCKARRAY_BITS 4
58 #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
59 #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
61 struct ip_vs_aligned_lock
64 } __attribute__((__aligned__(SMP_CACHE_BYTES
)));
66 /* lock array for conn table */
67 static struct ip_vs_aligned_lock
68 __ip_vs_conntbl_lock_array
[CT_LOCKARRAY_SIZE
] __cacheline_aligned
;
70 static inline void ct_read_lock(unsigned key
)
72 read_lock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
75 static inline void ct_read_unlock(unsigned key
)
77 read_unlock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
80 static inline void ct_write_lock(unsigned key
)
82 write_lock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
85 static inline void ct_write_unlock(unsigned key
)
87 write_unlock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
90 static inline void ct_read_lock_bh(unsigned key
)
92 read_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
95 static inline void ct_read_unlock_bh(unsigned key
)
97 read_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
100 static inline void ct_write_lock_bh(unsigned key
)
102 write_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
105 static inline void ct_write_unlock_bh(unsigned key
)
107 write_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
112 * Returns hash value for IPVS connection entry
114 static unsigned int ip_vs_conn_hashkey(unsigned proto
, __u32 addr
, __u16 port
)
116 return jhash_3words(addr
, port
, proto
, ip_vs_conn_rnd
)
117 & IP_VS_CONN_TAB_MASK
;
122 * Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
123 * returns bool success.
125 static inline int ip_vs_conn_hash(struct ip_vs_conn
*cp
)
130 /* Hash by protocol, client address and port */
131 hash
= ip_vs_conn_hashkey(cp
->protocol
, cp
->caddr
, cp
->cport
);
135 if (!(cp
->flags
& IP_VS_CONN_F_HASHED
)) {
136 list_add(&cp
->c_list
, &ip_vs_conn_tab
[hash
]);
137 cp
->flags
|= IP_VS_CONN_F_HASHED
;
138 atomic_inc(&cp
->refcnt
);
141 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
142 "called from %p\n", __builtin_return_address(0));
146 ct_write_unlock(hash
);
153 * UNhashes ip_vs_conn from ip_vs_conn_tab.
154 * returns bool success.
156 static inline int ip_vs_conn_unhash(struct ip_vs_conn
*cp
)
161 /* unhash it and decrease its reference counter */
162 hash
= ip_vs_conn_hashkey(cp
->protocol
, cp
->caddr
, cp
->cport
);
166 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
167 list_del(&cp
->c_list
);
168 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
169 atomic_dec(&cp
->refcnt
);
174 ct_write_unlock(hash
);
181 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
182 * Called for pkts coming from OUTside-to-INside.
183 * s_addr, s_port: pkt source address (foreign host)
184 * d_addr, d_port: pkt dest address (load balancer)
186 static inline struct ip_vs_conn
*__ip_vs_conn_in_get
187 (int protocol
, __u32 s_addr
, __u16 s_port
, __u32 d_addr
, __u16 d_port
)
190 struct ip_vs_conn
*cp
;
192 hash
= ip_vs_conn_hashkey(protocol
, s_addr
, s_port
);
196 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
197 if (s_addr
==cp
->caddr
&& s_port
==cp
->cport
&&
198 d_port
==cp
->vport
&& d_addr
==cp
->vaddr
&&
199 ((!s_port
) ^ (!(cp
->flags
& IP_VS_CONN_F_NO_CPORT
))) &&
200 protocol
==cp
->protocol
) {
202 atomic_inc(&cp
->refcnt
);
203 ct_read_unlock(hash
);
208 ct_read_unlock(hash
);
213 struct ip_vs_conn
*ip_vs_conn_in_get
214 (int protocol
, __u32 s_addr
, __u16 s_port
, __u32 d_addr
, __u16 d_port
)
216 struct ip_vs_conn
*cp
;
218 cp
= __ip_vs_conn_in_get(protocol
, s_addr
, s_port
, d_addr
, d_port
);
219 if (!cp
&& atomic_read(&ip_vs_conn_no_cport_cnt
))
220 cp
= __ip_vs_conn_in_get(protocol
, s_addr
, 0, d_addr
, d_port
);
222 IP_VS_DBG(7, "lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
223 ip_vs_proto_name(protocol
),
224 NIPQUAD(s_addr
), ntohs(s_port
),
225 NIPQUAD(d_addr
), ntohs(d_port
),
231 /* Get reference to connection template */
232 struct ip_vs_conn
*ip_vs_ct_in_get
233 (int protocol
, __u32 s_addr
, __u16 s_port
, __u32 d_addr
, __u16 d_port
)
236 struct ip_vs_conn
*cp
;
238 hash
= ip_vs_conn_hashkey(protocol
, s_addr
, s_port
);
242 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
243 if (s_addr
==cp
->caddr
&& s_port
==cp
->cport
&&
244 d_port
==cp
->vport
&& d_addr
==cp
->vaddr
&&
245 cp
->flags
& IP_VS_CONN_F_TEMPLATE
&&
246 protocol
==cp
->protocol
) {
248 atomic_inc(&cp
->refcnt
);
255 ct_read_unlock(hash
);
257 IP_VS_DBG(7, "template lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
258 ip_vs_proto_name(protocol
),
259 NIPQUAD(s_addr
), ntohs(s_port
),
260 NIPQUAD(d_addr
), ntohs(d_port
),
267 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
268 * Called for pkts coming from inside-to-OUTside.
269 * s_addr, s_port: pkt source address (inside host)
270 * d_addr, d_port: pkt dest address (foreign host)
272 struct ip_vs_conn
*ip_vs_conn_out_get
273 (int protocol
, __u32 s_addr
, __u16 s_port
, __u32 d_addr
, __u16 d_port
)
276 struct ip_vs_conn
*cp
, *ret
=NULL
;
279 * Check for "full" addressed entries
281 hash
= ip_vs_conn_hashkey(protocol
, d_addr
, d_port
);
285 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
286 if (d_addr
== cp
->caddr
&& d_port
== cp
->cport
&&
287 s_port
== cp
->dport
&& s_addr
== cp
->daddr
&&
288 protocol
== cp
->protocol
) {
290 atomic_inc(&cp
->refcnt
);
296 ct_read_unlock(hash
);
298 IP_VS_DBG(7, "lookup/out %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
299 ip_vs_proto_name(protocol
),
300 NIPQUAD(s_addr
), ntohs(s_port
),
301 NIPQUAD(d_addr
), ntohs(d_port
),
302 ret
?"hit":"not hit");
309 * Put back the conn and restart its timer with its timeout
311 void ip_vs_conn_put(struct ip_vs_conn
*cp
)
313 /* reset it expire in its timeout */
314 mod_timer(&cp
->timer
, jiffies
+cp
->timeout
);
316 __ip_vs_conn_put(cp
);
321 * Fill a no_client_port connection with a client port number
323 void ip_vs_conn_fill_cport(struct ip_vs_conn
*cp
, __u16 cport
)
325 if (ip_vs_conn_unhash(cp
)) {
326 spin_lock(&cp
->lock
);
327 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
) {
328 atomic_dec(&ip_vs_conn_no_cport_cnt
);
329 cp
->flags
&= ~IP_VS_CONN_F_NO_CPORT
;
332 spin_unlock(&cp
->lock
);
334 /* hash on new dport */
341 * Bind a connection entry with the corresponding packet_xmit.
342 * Called by ip_vs_conn_new.
344 static inline void ip_vs_bind_xmit(struct ip_vs_conn
*cp
)
346 switch (IP_VS_FWD_METHOD(cp
)) {
347 case IP_VS_CONN_F_MASQ
:
348 cp
->packet_xmit
= ip_vs_nat_xmit
;
351 case IP_VS_CONN_F_TUNNEL
:
352 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
355 case IP_VS_CONN_F_DROUTE
:
356 cp
->packet_xmit
= ip_vs_dr_xmit
;
359 case IP_VS_CONN_F_LOCALNODE
:
360 cp
->packet_xmit
= ip_vs_null_xmit
;
363 case IP_VS_CONN_F_BYPASS
:
364 cp
->packet_xmit
= ip_vs_bypass_xmit
;
370 static inline int ip_vs_dest_totalconns(struct ip_vs_dest
*dest
)
372 return atomic_read(&dest
->activeconns
)
373 + atomic_read(&dest
->inactconns
);
377 * Bind a connection entry with a virtual service destination
378 * Called just after a new connection entry is created.
381 ip_vs_bind_dest(struct ip_vs_conn
*cp
, struct ip_vs_dest
*dest
)
383 /* if dest is NULL, then return directly */
387 /* Increase the refcnt counter of the dest */
388 atomic_inc(&dest
->refcnt
);
390 /* Bind with the destination and its corresponding transmitter */
391 cp
->flags
|= atomic_read(&dest
->conn_flags
);
394 IP_VS_DBG(9, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
395 "d:%u.%u.%u.%u:%d fwd:%c s:%u flg:%X cnt:%d destcnt:%d\n",
396 ip_vs_proto_name(cp
->protocol
),
397 NIPQUAD(cp
->caddr
), ntohs(cp
->cport
),
398 NIPQUAD(cp
->vaddr
), ntohs(cp
->vport
),
399 NIPQUAD(cp
->daddr
), ntohs(cp
->dport
),
400 ip_vs_fwd_tag(cp
), cp
->state
,
401 cp
->flags
, atomic_read(&cp
->refcnt
),
402 atomic_read(&dest
->refcnt
));
404 /* Update the connection counters */
405 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
406 /* It is a normal connection, so increase the inactive
407 connection counter because it is in TCP SYNRECV
408 state (inactive) or other protocol inacive state */
409 atomic_inc(&dest
->inactconns
);
411 /* It is a persistent connection/template, so increase
412 the peristent connection counter */
413 atomic_inc(&dest
->persistconns
);
416 if (dest
->u_threshold
!= 0 &&
417 ip_vs_dest_totalconns(dest
) >= dest
->u_threshold
)
418 dest
->flags
|= IP_VS_DEST_F_OVERLOAD
;
423 * Unbind a connection entry with its VS destination
424 * Called by the ip_vs_conn_expire function.
426 static inline void ip_vs_unbind_dest(struct ip_vs_conn
*cp
)
428 struct ip_vs_dest
*dest
= cp
->dest
;
433 IP_VS_DBG(9, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
434 "d:%u.%u.%u.%u:%d fwd:%c s:%u flg:%X cnt:%d destcnt:%d\n",
435 ip_vs_proto_name(cp
->protocol
),
436 NIPQUAD(cp
->caddr
), ntohs(cp
->cport
),
437 NIPQUAD(cp
->vaddr
), ntohs(cp
->vport
),
438 NIPQUAD(cp
->daddr
), ntohs(cp
->dport
),
439 ip_vs_fwd_tag(cp
), cp
->state
,
440 cp
->flags
, atomic_read(&cp
->refcnt
),
441 atomic_read(&dest
->refcnt
));
443 /* Update the connection counters */
444 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
445 /* It is a normal connection, so decrease the inactconns
446 or activeconns counter */
447 if (cp
->flags
& IP_VS_CONN_F_INACTIVE
) {
448 atomic_dec(&dest
->inactconns
);
450 atomic_dec(&dest
->activeconns
);
453 /* It is a persistent connection/template, so decrease
454 the peristent connection counter */
455 atomic_dec(&dest
->persistconns
);
458 if (dest
->l_threshold
!= 0) {
459 if (ip_vs_dest_totalconns(dest
) < dest
->l_threshold
)
460 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
461 } else if (dest
->u_threshold
!= 0) {
462 if (ip_vs_dest_totalconns(dest
) * 4 < dest
->u_threshold
* 3)
463 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
465 if (dest
->flags
& IP_VS_DEST_F_OVERLOAD
)
466 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
470 * Simply decrease the refcnt of the dest, because the
471 * dest will be either in service's destination list
474 atomic_dec(&dest
->refcnt
);
479 * Checking if the destination of a connection template is available.
480 * If available, return 1, otherwise invalidate this connection
481 * template and return 0.
483 int ip_vs_check_template(struct ip_vs_conn
*ct
)
485 struct ip_vs_dest
*dest
= ct
->dest
;
488 * Checking the dest server status.
490 if ((dest
== NULL
) ||
491 !(dest
->flags
& IP_VS_DEST_F_AVAILABLE
) ||
492 (sysctl_ip_vs_expire_quiescent_template
&&
493 (atomic_read(&dest
->weight
) == 0))) {
494 IP_VS_DBG(9, "check_template: dest not available for "
495 "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
496 "-> d:%u.%u.%u.%u:%d\n",
497 ip_vs_proto_name(ct
->protocol
),
498 NIPQUAD(ct
->caddr
), ntohs(ct
->cport
),
499 NIPQUAD(ct
->vaddr
), ntohs(ct
->vport
),
500 NIPQUAD(ct
->daddr
), ntohs(ct
->dport
));
503 * Invalidate the connection template
505 if (ct
->vport
!= 65535) {
506 if (ip_vs_conn_unhash(ct
)) {
515 * Simply decrease the refcnt of the template,
516 * don't restart its timer.
518 atomic_dec(&ct
->refcnt
);
524 static void ip_vs_conn_expire(unsigned long data
)
526 struct ip_vs_conn
*cp
= (struct ip_vs_conn
*)data
;
533 atomic_inc(&cp
->refcnt
);
536 * do I control anybody?
538 if (atomic_read(&cp
->n_control
))
542 * unhash it if it is hashed in the conn table
544 if (!ip_vs_conn_unhash(cp
))
548 * refcnt==1 implies I'm the only one referrer
550 if (likely(atomic_read(&cp
->refcnt
) == 1)) {
551 /* delete the timer if it is activated by other users */
552 if (timer_pending(&cp
->timer
))
553 del_timer(&cp
->timer
);
555 /* does anybody control me? */
557 ip_vs_control_del(cp
);
559 if (unlikely(cp
->app
!= NULL
))
560 ip_vs_unbind_app(cp
);
561 ip_vs_unbind_dest(cp
);
562 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
)
563 atomic_dec(&ip_vs_conn_no_cport_cnt
);
564 atomic_dec(&ip_vs_conn_count
);
566 kmem_cache_free(ip_vs_conn_cachep
, cp
);
570 /* hash it back to the table */
574 IP_VS_DBG(7, "delayed: refcnt-1=%d conn.n_control=%d\n",
575 atomic_read(&cp
->refcnt
)-1,
576 atomic_read(&cp
->n_control
));
582 void ip_vs_conn_expire_now(struct ip_vs_conn
*cp
)
584 if (del_timer(&cp
->timer
))
585 mod_timer(&cp
->timer
, jiffies
);
590 * Create a new connection entry and hash it into the ip_vs_conn_tab
593 ip_vs_conn_new(int proto
, __u32 caddr
, __u16 cport
, __u32 vaddr
, __u16 vport
,
594 __u32 daddr
, __u16 dport
, unsigned flags
,
595 struct ip_vs_dest
*dest
)
597 struct ip_vs_conn
*cp
;
598 struct ip_vs_protocol
*pp
= ip_vs_proto_get(proto
);
600 cp
= kmem_cache_alloc(ip_vs_conn_cachep
, GFP_ATOMIC
);
602 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
606 memset(cp
, 0, sizeof(*cp
));
607 INIT_LIST_HEAD(&cp
->c_list
);
608 init_timer(&cp
->timer
);
609 cp
->timer
.data
= (unsigned long)cp
;
610 cp
->timer
.function
= ip_vs_conn_expire
;
611 cp
->protocol
= proto
;
619 spin_lock_init(&cp
->lock
);
622 * Set the entry is referenced by the current thread before hashing
623 * it in the table, so that other thread run ip_vs_random_dropentry
624 * but cannot drop this entry.
626 atomic_set(&cp
->refcnt
, 1);
628 atomic_set(&cp
->n_control
, 0);
629 atomic_set(&cp
->in_pkts
, 0);
631 atomic_inc(&ip_vs_conn_count
);
632 if (flags
& IP_VS_CONN_F_NO_CPORT
)
633 atomic_inc(&ip_vs_conn_no_cport_cnt
);
635 /* Bind the connection with a destination server */
636 ip_vs_bind_dest(cp
, dest
);
638 /* Set its state and timeout */
642 /* Bind its packet transmitter */
645 if (unlikely(pp
&& atomic_read(&pp
->appcnt
)))
646 ip_vs_bind_app(cp
, pp
);
648 /* Hash it in the ip_vs_conn_tab finally */
656 * /proc/net/ip_vs_conn entries
658 #ifdef CONFIG_PROC_FS
660 static void *ip_vs_conn_array(struct seq_file
*seq
, loff_t pos
)
663 struct ip_vs_conn
*cp
;
665 for(idx
= 0; idx
< IP_VS_CONN_TAB_SIZE
; idx
++) {
666 ct_read_lock_bh(idx
);
667 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
669 seq
->private = &ip_vs_conn_tab
[idx
];
673 ct_read_unlock_bh(idx
);
679 static void *ip_vs_conn_seq_start(struct seq_file
*seq
, loff_t
*pos
)
682 return *pos
? ip_vs_conn_array(seq
, *pos
- 1) :SEQ_START_TOKEN
;
685 static void *ip_vs_conn_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
687 struct ip_vs_conn
*cp
= v
;
688 struct list_head
*e
, *l
= seq
->private;
692 if (v
== SEQ_START_TOKEN
)
693 return ip_vs_conn_array(seq
, 0);
695 /* more on same hash chain? */
696 if ((e
= cp
->c_list
.next
) != l
)
697 return list_entry(e
, struct ip_vs_conn
, c_list
);
699 idx
= l
- ip_vs_conn_tab
;
700 ct_read_unlock_bh(idx
);
702 while (++idx
< IP_VS_CONN_TAB_SIZE
) {
703 ct_read_lock_bh(idx
);
704 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
705 seq
->private = &ip_vs_conn_tab
[idx
];
708 ct_read_unlock_bh(idx
);
714 static void ip_vs_conn_seq_stop(struct seq_file
*seq
, void *v
)
716 struct list_head
*l
= seq
->private;
719 ct_read_unlock_bh(l
- ip_vs_conn_tab
);
722 static int ip_vs_conn_seq_show(struct seq_file
*seq
, void *v
)
725 if (v
== SEQ_START_TOKEN
)
727 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires\n");
729 const struct ip_vs_conn
*cp
= v
;
732 "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n",
733 ip_vs_proto_name(cp
->protocol
),
734 ntohl(cp
->caddr
), ntohs(cp
->cport
),
735 ntohl(cp
->vaddr
), ntohs(cp
->vport
),
736 ntohl(cp
->daddr
), ntohs(cp
->dport
),
737 ip_vs_state_name(cp
->protocol
, cp
->state
),
738 (cp
->timer
.expires
-jiffies
)/HZ
);
743 static struct seq_operations ip_vs_conn_seq_ops
= {
744 .start
= ip_vs_conn_seq_start
,
745 .next
= ip_vs_conn_seq_next
,
746 .stop
= ip_vs_conn_seq_stop
,
747 .show
= ip_vs_conn_seq_show
,
750 static int ip_vs_conn_open(struct inode
*inode
, struct file
*file
)
752 return seq_open(file
, &ip_vs_conn_seq_ops
);
755 static struct file_operations ip_vs_conn_fops
= {
756 .owner
= THIS_MODULE
,
757 .open
= ip_vs_conn_open
,
760 .release
= seq_release
,
766 * Randomly drop connection entries before running out of memory
768 static inline int todrop_entry(struct ip_vs_conn
*cp
)
771 * The drop rate array needs tuning for real environments.
772 * Called from timer bh only => no locking
774 static const char todrop_rate
[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
775 static char todrop_counter
[9] = {0};
778 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
779 This will leave enough time for normal connection to get
781 if (time_before(cp
->timeout
+ jiffies
, cp
->timer
.expires
+ 60*HZ
))
784 /* Don't drop the entry if its number of incoming packets is not
786 i
= atomic_read(&cp
->in_pkts
);
787 if (i
> 8 || i
< 0) return 0;
789 if (!todrop_rate
[i
]) return 0;
790 if (--todrop_counter
[i
] > 0) return 0;
792 todrop_counter
[i
] = todrop_rate
[i
];
796 /* Called from keventd and must protect itself from softirqs */
797 void ip_vs_random_dropentry(void)
800 struct ip_vs_conn
*cp
;
803 * Randomly scan 1/32 of the whole table every second
805 for (idx
= 0; idx
< (IP_VS_CONN_TAB_SIZE
>>5); idx
++) {
806 unsigned hash
= net_random() & IP_VS_CONN_TAB_MASK
;
809 * Lock is actually needed in this loop.
811 ct_write_lock_bh(hash
);
813 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
814 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
815 /* connection template */
818 if (cp
->protocol
== IPPROTO_TCP
) {
820 case IP_VS_TCP_S_SYN_RECV
:
821 case IP_VS_TCP_S_SYNACK
:
824 case IP_VS_TCP_S_ESTABLISHED
:
825 if (todrop_entry(cp
))
833 if (!todrop_entry(cp
))
837 IP_VS_DBG(4, "del connection\n");
838 ip_vs_conn_expire_now(cp
);
840 IP_VS_DBG(4, "del conn template\n");
841 ip_vs_conn_expire_now(cp
->control
);
844 ct_write_unlock_bh(hash
);
850 * Flush all the connection entries in the ip_vs_conn_tab
852 static void ip_vs_conn_flush(void)
855 struct ip_vs_conn
*cp
;
858 for (idx
=0; idx
<IP_VS_CONN_TAB_SIZE
; idx
++) {
860 * Lock is actually needed in this loop.
862 ct_write_lock_bh(idx
);
864 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
866 IP_VS_DBG(4, "del connection\n");
867 ip_vs_conn_expire_now(cp
);
869 IP_VS_DBG(4, "del conn template\n");
870 ip_vs_conn_expire_now(cp
->control
);
873 ct_write_unlock_bh(idx
);
876 /* the counter may be not NULL, because maybe some conn entries
877 are run by slow timer handler or unhashed but still referred */
878 if (atomic_read(&ip_vs_conn_count
) != 0) {
885 int ip_vs_conn_init(void)
890 * Allocate the connection hash table and initialize its list heads
892 ip_vs_conn_tab
= vmalloc(IP_VS_CONN_TAB_SIZE
*sizeof(struct list_head
));
896 /* Allocate ip_vs_conn slab cache */
897 ip_vs_conn_cachep
= kmem_cache_create("ip_vs_conn",
898 sizeof(struct ip_vs_conn
), 0,
899 SLAB_HWCACHE_ALIGN
, NULL
, NULL
);
900 if (!ip_vs_conn_cachep
) {
901 vfree(ip_vs_conn_tab
);
905 IP_VS_INFO("Connection hash table configured "
906 "(size=%d, memory=%ldKbytes)\n",
908 (long)(IP_VS_CONN_TAB_SIZE
*sizeof(struct list_head
))/1024);
909 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
910 sizeof(struct ip_vs_conn
));
912 for (idx
= 0; idx
< IP_VS_CONN_TAB_SIZE
; idx
++) {
913 INIT_LIST_HEAD(&ip_vs_conn_tab
[idx
]);
916 for (idx
= 0; idx
< CT_LOCKARRAY_SIZE
; idx
++) {
917 rwlock_init(&__ip_vs_conntbl_lock_array
[idx
].l
);
920 proc_net_fops_create("ip_vs_conn", 0, &ip_vs_conn_fops
);
922 /* calculate the random value for connection hash */
923 get_random_bytes(&ip_vs_conn_rnd
, sizeof(ip_vs_conn_rnd
));
929 void ip_vs_conn_cleanup(void)
931 /* flush all the connection entries first */
934 /* Release the empty cache */
935 kmem_cache_destroy(ip_vs_conn_cachep
);
936 proc_net_remove("ip_vs_conn");
937 vfree(ip_vs_conn_tab
);