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 #define KMSG_COMPONENT "IPVS"
26 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
28 #include <linux/interrupt.h>
30 #include <linux/net.h>
31 #include <linux/kernel.h>
32 #include <linux/module.h>
33 #include <linux/vmalloc.h>
34 #include <linux/proc_fs.h> /* for proc_net_* */
35 #include <linux/seq_file.h>
36 #include <linux/jhash.h>
37 #include <linux/random.h>
39 #include <net/net_namespace.h>
40 #include <net/ip_vs.h>
44 * Connection hash table: for input and output packets lookups of IPVS
46 static struct list_head
*ip_vs_conn_tab
;
48 /* SLAB cache for IPVS connections */
49 static struct kmem_cache
*ip_vs_conn_cachep __read_mostly
;
51 /* counter for current IPVS connections */
52 static atomic_t ip_vs_conn_count
= ATOMIC_INIT(0);
54 /* counter for no client port connections */
55 static atomic_t ip_vs_conn_no_cport_cnt
= ATOMIC_INIT(0);
57 /* random value for IPVS connection hash */
58 static unsigned int ip_vs_conn_rnd
;
61 * Fine locking granularity for big connection hash table
63 #define CT_LOCKARRAY_BITS 4
64 #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
65 #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
67 struct ip_vs_aligned_lock
70 } __attribute__((__aligned__(SMP_CACHE_BYTES
)));
72 /* lock array for conn table */
73 static struct ip_vs_aligned_lock
74 __ip_vs_conntbl_lock_array
[CT_LOCKARRAY_SIZE
] __cacheline_aligned
;
76 static inline void ct_read_lock(unsigned key
)
78 read_lock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
81 static inline void ct_read_unlock(unsigned key
)
83 read_unlock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
86 static inline void ct_write_lock(unsigned key
)
88 write_lock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
91 static inline void ct_write_unlock(unsigned key
)
93 write_unlock(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
96 static inline void ct_read_lock_bh(unsigned key
)
98 read_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
101 static inline void ct_read_unlock_bh(unsigned key
)
103 read_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
106 static inline void ct_write_lock_bh(unsigned key
)
108 write_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
111 static inline void ct_write_unlock_bh(unsigned key
)
113 write_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
118 * Returns hash value for IPVS connection entry
120 static unsigned int ip_vs_conn_hashkey(int af
, unsigned proto
,
121 const union nf_inet_addr
*addr
,
124 #ifdef CONFIG_IP_VS_IPV6
126 return jhash_3words(jhash(addr
, 16, ip_vs_conn_rnd
),
127 (__force u32
)port
, proto
, ip_vs_conn_rnd
)
128 & IP_VS_CONN_TAB_MASK
;
130 return jhash_3words((__force u32
)addr
->ip
, (__force u32
)port
, proto
,
132 & IP_VS_CONN_TAB_MASK
;
137 * Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
138 * returns bool success.
140 static inline int ip_vs_conn_hash(struct ip_vs_conn
*cp
)
145 /* Hash by protocol, client address and port */
146 hash
= ip_vs_conn_hashkey(cp
->af
, cp
->protocol
, &cp
->caddr
, cp
->cport
);
149 spin_lock(&cp
->lock
);
151 if (!(cp
->flags
& IP_VS_CONN_F_HASHED
)) {
152 list_add(&cp
->c_list
, &ip_vs_conn_tab
[hash
]);
153 cp
->flags
|= IP_VS_CONN_F_HASHED
;
154 atomic_inc(&cp
->refcnt
);
157 pr_err("%s(): request for already hashed, called from %pF\n",
158 __func__
, __builtin_return_address(0));
162 spin_unlock(&cp
->lock
);
163 ct_write_unlock(hash
);
170 * UNhashes ip_vs_conn from ip_vs_conn_tab.
171 * returns bool success.
173 static inline int ip_vs_conn_unhash(struct ip_vs_conn
*cp
)
178 /* unhash it and decrease its reference counter */
179 hash
= ip_vs_conn_hashkey(cp
->af
, cp
->protocol
, &cp
->caddr
, cp
->cport
);
182 spin_lock(&cp
->lock
);
184 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
185 list_del(&cp
->c_list
);
186 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
187 atomic_dec(&cp
->refcnt
);
192 spin_unlock(&cp
->lock
);
193 ct_write_unlock(hash
);
200 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
201 * Called for pkts coming from OUTside-to-INside.
202 * s_addr, s_port: pkt source address (foreign host)
203 * d_addr, d_port: pkt dest address (load balancer)
205 static inline struct ip_vs_conn
*__ip_vs_conn_in_get
206 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
207 const union nf_inet_addr
*d_addr
, __be16 d_port
)
210 struct ip_vs_conn
*cp
;
212 hash
= ip_vs_conn_hashkey(af
, protocol
, s_addr
, s_port
);
216 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
218 ip_vs_addr_equal(af
, s_addr
, &cp
->caddr
) &&
219 ip_vs_addr_equal(af
, d_addr
, &cp
->vaddr
) &&
220 s_port
== cp
->cport
&& d_port
== cp
->vport
&&
221 ((!s_port
) ^ (!(cp
->flags
& IP_VS_CONN_F_NO_CPORT
))) &&
222 protocol
== cp
->protocol
) {
224 atomic_inc(&cp
->refcnt
);
225 ct_read_unlock(hash
);
230 ct_read_unlock(hash
);
235 struct ip_vs_conn
*ip_vs_conn_in_get
236 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
237 const union nf_inet_addr
*d_addr
, __be16 d_port
)
239 struct ip_vs_conn
*cp
;
241 cp
= __ip_vs_conn_in_get(af
, protocol
, s_addr
, s_port
, d_addr
, d_port
);
242 if (!cp
&& atomic_read(&ip_vs_conn_no_cport_cnt
))
243 cp
= __ip_vs_conn_in_get(af
, protocol
, s_addr
, 0, d_addr
,
246 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
247 ip_vs_proto_name(protocol
),
248 IP_VS_DBG_ADDR(af
, s_addr
), ntohs(s_port
),
249 IP_VS_DBG_ADDR(af
, d_addr
), ntohs(d_port
),
250 cp
? "hit" : "not hit");
255 /* Get reference to connection template */
256 struct ip_vs_conn
*ip_vs_ct_in_get
257 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
258 const union nf_inet_addr
*d_addr
, __be16 d_port
)
261 struct ip_vs_conn
*cp
;
263 hash
= ip_vs_conn_hashkey(af
, protocol
, s_addr
, s_port
);
267 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
269 ip_vs_addr_equal(af
, s_addr
, &cp
->caddr
) &&
270 /* protocol should only be IPPROTO_IP if
271 * d_addr is a fwmark */
272 ip_vs_addr_equal(protocol
== IPPROTO_IP
? AF_UNSPEC
: af
,
273 d_addr
, &cp
->vaddr
) &&
274 s_port
== cp
->cport
&& d_port
== cp
->vport
&&
275 cp
->flags
& IP_VS_CONN_F_TEMPLATE
&&
276 protocol
== cp
->protocol
) {
278 atomic_inc(&cp
->refcnt
);
285 ct_read_unlock(hash
);
287 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
288 ip_vs_proto_name(protocol
),
289 IP_VS_DBG_ADDR(af
, s_addr
), ntohs(s_port
),
290 IP_VS_DBG_ADDR(af
, d_addr
), ntohs(d_port
),
291 cp
? "hit" : "not hit");
297 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
298 * Called for pkts coming from inside-to-OUTside.
299 * s_addr, s_port: pkt source address (inside host)
300 * d_addr, d_port: pkt dest address (foreign host)
302 struct ip_vs_conn
*ip_vs_conn_out_get
303 (int af
, int protocol
, const union nf_inet_addr
*s_addr
, __be16 s_port
,
304 const union nf_inet_addr
*d_addr
, __be16 d_port
)
307 struct ip_vs_conn
*cp
, *ret
=NULL
;
310 * Check for "full" addressed entries
312 hash
= ip_vs_conn_hashkey(af
, protocol
, d_addr
, d_port
);
316 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
318 ip_vs_addr_equal(af
, d_addr
, &cp
->caddr
) &&
319 ip_vs_addr_equal(af
, s_addr
, &cp
->daddr
) &&
320 d_port
== cp
->cport
&& s_port
== cp
->dport
&&
321 protocol
== cp
->protocol
) {
323 atomic_inc(&cp
->refcnt
);
329 ct_read_unlock(hash
);
331 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
332 ip_vs_proto_name(protocol
),
333 IP_VS_DBG_ADDR(af
, s_addr
), ntohs(s_port
),
334 IP_VS_DBG_ADDR(af
, d_addr
), ntohs(d_port
),
335 ret
? "hit" : "not hit");
342 * Put back the conn and restart its timer with its timeout
344 void ip_vs_conn_put(struct ip_vs_conn
*cp
)
346 /* reset it expire in its timeout */
347 mod_timer(&cp
->timer
, jiffies
+cp
->timeout
);
349 __ip_vs_conn_put(cp
);
354 * Fill a no_client_port connection with a client port number
356 void ip_vs_conn_fill_cport(struct ip_vs_conn
*cp
, __be16 cport
)
358 if (ip_vs_conn_unhash(cp
)) {
359 spin_lock(&cp
->lock
);
360 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
) {
361 atomic_dec(&ip_vs_conn_no_cport_cnt
);
362 cp
->flags
&= ~IP_VS_CONN_F_NO_CPORT
;
365 spin_unlock(&cp
->lock
);
367 /* hash on new dport */
374 * Bind a connection entry with the corresponding packet_xmit.
375 * Called by ip_vs_conn_new.
377 static inline void ip_vs_bind_xmit(struct ip_vs_conn
*cp
)
379 switch (IP_VS_FWD_METHOD(cp
)) {
380 case IP_VS_CONN_F_MASQ
:
381 cp
->packet_xmit
= ip_vs_nat_xmit
;
384 case IP_VS_CONN_F_TUNNEL
:
385 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
388 case IP_VS_CONN_F_DROUTE
:
389 cp
->packet_xmit
= ip_vs_dr_xmit
;
392 case IP_VS_CONN_F_LOCALNODE
:
393 cp
->packet_xmit
= ip_vs_null_xmit
;
396 case IP_VS_CONN_F_BYPASS
:
397 cp
->packet_xmit
= ip_vs_bypass_xmit
;
402 #ifdef CONFIG_IP_VS_IPV6
403 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn
*cp
)
405 switch (IP_VS_FWD_METHOD(cp
)) {
406 case IP_VS_CONN_F_MASQ
:
407 cp
->packet_xmit
= ip_vs_nat_xmit_v6
;
410 case IP_VS_CONN_F_TUNNEL
:
411 cp
->packet_xmit
= ip_vs_tunnel_xmit_v6
;
414 case IP_VS_CONN_F_DROUTE
:
415 cp
->packet_xmit
= ip_vs_dr_xmit_v6
;
418 case IP_VS_CONN_F_LOCALNODE
:
419 cp
->packet_xmit
= ip_vs_null_xmit
;
422 case IP_VS_CONN_F_BYPASS
:
423 cp
->packet_xmit
= ip_vs_bypass_xmit_v6
;
430 static inline int ip_vs_dest_totalconns(struct ip_vs_dest
*dest
)
432 return atomic_read(&dest
->activeconns
)
433 + atomic_read(&dest
->inactconns
);
437 * Bind a connection entry with a virtual service destination
438 * Called just after a new connection entry is created.
441 ip_vs_bind_dest(struct ip_vs_conn
*cp
, struct ip_vs_dest
*dest
)
443 /* if dest is NULL, then return directly */
447 /* Increase the refcnt counter of the dest */
448 atomic_inc(&dest
->refcnt
);
450 /* Bind with the destination and its corresponding transmitter */
451 if ((cp
->flags
& IP_VS_CONN_F_SYNC
) &&
452 (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)))
453 /* if the connection is not template and is created
454 * by sync, preserve the activity flag.
456 cp
->flags
|= atomic_read(&dest
->conn_flags
) &
457 (~IP_VS_CONN_F_INACTIVE
);
459 cp
->flags
|= atomic_read(&dest
->conn_flags
);
462 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
463 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
465 ip_vs_proto_name(cp
->protocol
),
466 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
467 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
468 IP_VS_DBG_ADDR(cp
->af
, &cp
->daddr
), ntohs(cp
->dport
),
469 ip_vs_fwd_tag(cp
), cp
->state
,
470 cp
->flags
, atomic_read(&cp
->refcnt
),
471 atomic_read(&dest
->refcnt
));
473 /* Update the connection counters */
474 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
475 /* It is a normal connection, so increase the inactive
476 connection counter because it is in TCP SYNRECV
477 state (inactive) or other protocol inacive state */
478 if ((cp
->flags
& IP_VS_CONN_F_SYNC
) &&
479 (!(cp
->flags
& IP_VS_CONN_F_INACTIVE
)))
480 atomic_inc(&dest
->activeconns
);
482 atomic_inc(&dest
->inactconns
);
484 /* It is a persistent connection/template, so increase
485 the peristent connection counter */
486 atomic_inc(&dest
->persistconns
);
489 if (dest
->u_threshold
!= 0 &&
490 ip_vs_dest_totalconns(dest
) >= dest
->u_threshold
)
491 dest
->flags
|= IP_VS_DEST_F_OVERLOAD
;
496 * Check if there is a destination for the connection, if so
497 * bind the connection to the destination.
499 struct ip_vs_dest
*ip_vs_try_bind_dest(struct ip_vs_conn
*cp
)
501 struct ip_vs_dest
*dest
;
503 if ((cp
) && (!cp
->dest
)) {
504 dest
= ip_vs_find_dest(cp
->af
, &cp
->daddr
, cp
->dport
,
505 &cp
->vaddr
, cp
->vport
,
507 ip_vs_bind_dest(cp
, dest
);
515 * Unbind a connection entry with its VS destination
516 * Called by the ip_vs_conn_expire function.
518 static inline void ip_vs_unbind_dest(struct ip_vs_conn
*cp
)
520 struct ip_vs_dest
*dest
= cp
->dest
;
525 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
526 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
528 ip_vs_proto_name(cp
->protocol
),
529 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
530 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
531 IP_VS_DBG_ADDR(cp
->af
, &cp
->daddr
), ntohs(cp
->dport
),
532 ip_vs_fwd_tag(cp
), cp
->state
,
533 cp
->flags
, atomic_read(&cp
->refcnt
),
534 atomic_read(&dest
->refcnt
));
536 /* Update the connection counters */
537 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
538 /* It is a normal connection, so decrease the inactconns
539 or activeconns counter */
540 if (cp
->flags
& IP_VS_CONN_F_INACTIVE
) {
541 atomic_dec(&dest
->inactconns
);
543 atomic_dec(&dest
->activeconns
);
546 /* It is a persistent connection/template, so decrease
547 the peristent connection counter */
548 atomic_dec(&dest
->persistconns
);
551 if (dest
->l_threshold
!= 0) {
552 if (ip_vs_dest_totalconns(dest
) < dest
->l_threshold
)
553 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
554 } else if (dest
->u_threshold
!= 0) {
555 if (ip_vs_dest_totalconns(dest
) * 4 < dest
->u_threshold
* 3)
556 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
558 if (dest
->flags
& IP_VS_DEST_F_OVERLOAD
)
559 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
563 * Simply decrease the refcnt of the dest, because the
564 * dest will be either in service's destination list
567 atomic_dec(&dest
->refcnt
);
572 * Checking if the destination of a connection template is available.
573 * If available, return 1, otherwise invalidate this connection
574 * template and return 0.
576 int ip_vs_check_template(struct ip_vs_conn
*ct
)
578 struct ip_vs_dest
*dest
= ct
->dest
;
581 * Checking the dest server status.
583 if ((dest
== NULL
) ||
584 !(dest
->flags
& IP_VS_DEST_F_AVAILABLE
) ||
585 (sysctl_ip_vs_expire_quiescent_template
&&
586 (atomic_read(&dest
->weight
) == 0))) {
587 IP_VS_DBG_BUF(9, "check_template: dest not available for "
588 "protocol %s s:%s:%d v:%s:%d "
590 ip_vs_proto_name(ct
->protocol
),
591 IP_VS_DBG_ADDR(ct
->af
, &ct
->caddr
),
593 IP_VS_DBG_ADDR(ct
->af
, &ct
->vaddr
),
595 IP_VS_DBG_ADDR(ct
->af
, &ct
->daddr
),
599 * Invalidate the connection template
601 if (ct
->vport
!= htons(0xffff)) {
602 if (ip_vs_conn_unhash(ct
)) {
603 ct
->dport
= htons(0xffff);
604 ct
->vport
= htons(0xffff);
611 * Simply decrease the refcnt of the template,
612 * don't restart its timer.
614 atomic_dec(&ct
->refcnt
);
620 static void ip_vs_conn_expire(unsigned long data
)
622 struct ip_vs_conn
*cp
= (struct ip_vs_conn
*)data
;
629 atomic_inc(&cp
->refcnt
);
632 * do I control anybody?
634 if (atomic_read(&cp
->n_control
))
638 * unhash it if it is hashed in the conn table
640 if (!ip_vs_conn_unhash(cp
))
644 * refcnt==1 implies I'm the only one referrer
646 if (likely(atomic_read(&cp
->refcnt
) == 1)) {
647 /* delete the timer if it is activated by other users */
648 if (timer_pending(&cp
->timer
))
649 del_timer(&cp
->timer
);
651 /* does anybody control me? */
653 ip_vs_control_del(cp
);
655 if (unlikely(cp
->app
!= NULL
))
656 ip_vs_unbind_app(cp
);
657 ip_vs_unbind_dest(cp
);
658 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
)
659 atomic_dec(&ip_vs_conn_no_cport_cnt
);
660 atomic_dec(&ip_vs_conn_count
);
662 kmem_cache_free(ip_vs_conn_cachep
, cp
);
666 /* hash it back to the table */
670 IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
671 atomic_read(&cp
->refcnt
)-1,
672 atomic_read(&cp
->n_control
));
678 void ip_vs_conn_expire_now(struct ip_vs_conn
*cp
)
680 if (del_timer(&cp
->timer
))
681 mod_timer(&cp
->timer
, jiffies
);
686 * Create a new connection entry and hash it into the ip_vs_conn_tab
689 ip_vs_conn_new(int af
, int proto
, const union nf_inet_addr
*caddr
, __be16 cport
,
690 const union nf_inet_addr
*vaddr
, __be16 vport
,
691 const union nf_inet_addr
*daddr
, __be16 dport
, unsigned flags
,
692 struct ip_vs_dest
*dest
)
694 struct ip_vs_conn
*cp
;
695 struct ip_vs_protocol
*pp
= ip_vs_proto_get(proto
);
697 cp
= kmem_cache_zalloc(ip_vs_conn_cachep
, GFP_ATOMIC
);
699 IP_VS_ERR_RL("%s(): no memory\n", __func__
);
703 INIT_LIST_HEAD(&cp
->c_list
);
704 setup_timer(&cp
->timer
, ip_vs_conn_expire
, (unsigned long)cp
);
706 cp
->protocol
= proto
;
707 ip_vs_addr_copy(af
, &cp
->caddr
, caddr
);
709 ip_vs_addr_copy(af
, &cp
->vaddr
, vaddr
);
711 /* proto should only be IPPROTO_IP if d_addr is a fwmark */
712 ip_vs_addr_copy(proto
== IPPROTO_IP
? AF_UNSPEC
: af
,
716 spin_lock_init(&cp
->lock
);
719 * Set the entry is referenced by the current thread before hashing
720 * it in the table, so that other thread run ip_vs_random_dropentry
721 * but cannot drop this entry.
723 atomic_set(&cp
->refcnt
, 1);
725 atomic_set(&cp
->n_control
, 0);
726 atomic_set(&cp
->in_pkts
, 0);
728 atomic_inc(&ip_vs_conn_count
);
729 if (flags
& IP_VS_CONN_F_NO_CPORT
)
730 atomic_inc(&ip_vs_conn_no_cport_cnt
);
732 /* Bind the connection with a destination server */
733 ip_vs_bind_dest(cp
, dest
);
735 /* Set its state and timeout */
739 /* Bind its packet transmitter */
740 #ifdef CONFIG_IP_VS_IPV6
742 ip_vs_bind_xmit_v6(cp
);
747 if (unlikely(pp
&& atomic_read(&pp
->appcnt
)))
748 ip_vs_bind_app(cp
, pp
);
750 /* Hash it in the ip_vs_conn_tab finally */
758 * /proc/net/ip_vs_conn entries
760 #ifdef CONFIG_PROC_FS
762 static void *ip_vs_conn_array(struct seq_file
*seq
, loff_t pos
)
765 struct ip_vs_conn
*cp
;
767 for(idx
= 0; idx
< IP_VS_CONN_TAB_SIZE
; idx
++) {
768 ct_read_lock_bh(idx
);
769 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
771 seq
->private = &ip_vs_conn_tab
[idx
];
775 ct_read_unlock_bh(idx
);
781 static void *ip_vs_conn_seq_start(struct seq_file
*seq
, loff_t
*pos
)
784 return *pos
? ip_vs_conn_array(seq
, *pos
- 1) :SEQ_START_TOKEN
;
787 static void *ip_vs_conn_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
789 struct ip_vs_conn
*cp
= v
;
790 struct list_head
*e
, *l
= seq
->private;
794 if (v
== SEQ_START_TOKEN
)
795 return ip_vs_conn_array(seq
, 0);
797 /* more on same hash chain? */
798 if ((e
= cp
->c_list
.next
) != l
)
799 return list_entry(e
, struct ip_vs_conn
, c_list
);
801 idx
= l
- ip_vs_conn_tab
;
802 ct_read_unlock_bh(idx
);
804 while (++idx
< IP_VS_CONN_TAB_SIZE
) {
805 ct_read_lock_bh(idx
);
806 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
807 seq
->private = &ip_vs_conn_tab
[idx
];
810 ct_read_unlock_bh(idx
);
816 static void ip_vs_conn_seq_stop(struct seq_file
*seq
, void *v
)
818 struct list_head
*l
= seq
->private;
821 ct_read_unlock_bh(l
- ip_vs_conn_tab
);
824 static int ip_vs_conn_seq_show(struct seq_file
*seq
, void *v
)
827 if (v
== SEQ_START_TOKEN
)
829 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires\n");
831 const struct ip_vs_conn
*cp
= v
;
833 #ifdef CONFIG_IP_VS_IPV6
834 if (cp
->af
== AF_INET6
)
835 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %7lu\n",
836 ip_vs_proto_name(cp
->protocol
),
837 &cp
->caddr
.in6
, ntohs(cp
->cport
),
838 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
839 &cp
->daddr
.in6
, ntohs(cp
->dport
),
840 ip_vs_state_name(cp
->protocol
, cp
->state
),
841 (cp
->timer
.expires
-jiffies
)/HZ
);
845 "%-3s %08X %04X %08X %04X"
846 " %08X %04X %-11s %7lu\n",
847 ip_vs_proto_name(cp
->protocol
),
848 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
849 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
850 ntohl(cp
->daddr
.ip
), ntohs(cp
->dport
),
851 ip_vs_state_name(cp
->protocol
, cp
->state
),
852 (cp
->timer
.expires
-jiffies
)/HZ
);
857 static const struct seq_operations ip_vs_conn_seq_ops
= {
858 .start
= ip_vs_conn_seq_start
,
859 .next
= ip_vs_conn_seq_next
,
860 .stop
= ip_vs_conn_seq_stop
,
861 .show
= ip_vs_conn_seq_show
,
864 static int ip_vs_conn_open(struct inode
*inode
, struct file
*file
)
866 return seq_open(file
, &ip_vs_conn_seq_ops
);
869 static const struct file_operations ip_vs_conn_fops
= {
870 .owner
= THIS_MODULE
,
871 .open
= ip_vs_conn_open
,
874 .release
= seq_release
,
877 static const char *ip_vs_origin_name(unsigned flags
)
879 if (flags
& IP_VS_CONN_F_SYNC
)
885 static int ip_vs_conn_sync_seq_show(struct seq_file
*seq
, void *v
)
888 if (v
== SEQ_START_TOKEN
)
890 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
892 const struct ip_vs_conn
*cp
= v
;
894 #ifdef CONFIG_IP_VS_IPV6
895 if (cp
->af
== AF_INET6
)
896 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
897 ip_vs_proto_name(cp
->protocol
),
898 &cp
->caddr
.in6
, ntohs(cp
->cport
),
899 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
900 &cp
->daddr
.in6
, ntohs(cp
->dport
),
901 ip_vs_state_name(cp
->protocol
, cp
->state
),
902 ip_vs_origin_name(cp
->flags
),
903 (cp
->timer
.expires
-jiffies
)/HZ
);
907 "%-3s %08X %04X %08X %04X "
908 "%08X %04X %-11s %-6s %7lu\n",
909 ip_vs_proto_name(cp
->protocol
),
910 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
911 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
912 ntohl(cp
->daddr
.ip
), ntohs(cp
->dport
),
913 ip_vs_state_name(cp
->protocol
, cp
->state
),
914 ip_vs_origin_name(cp
->flags
),
915 (cp
->timer
.expires
-jiffies
)/HZ
);
920 static const struct seq_operations ip_vs_conn_sync_seq_ops
= {
921 .start
= ip_vs_conn_seq_start
,
922 .next
= ip_vs_conn_seq_next
,
923 .stop
= ip_vs_conn_seq_stop
,
924 .show
= ip_vs_conn_sync_seq_show
,
927 static int ip_vs_conn_sync_open(struct inode
*inode
, struct file
*file
)
929 return seq_open(file
, &ip_vs_conn_sync_seq_ops
);
932 static const struct file_operations ip_vs_conn_sync_fops
= {
933 .owner
= THIS_MODULE
,
934 .open
= ip_vs_conn_sync_open
,
937 .release
= seq_release
,
944 * Randomly drop connection entries before running out of memory
946 static inline int todrop_entry(struct ip_vs_conn
*cp
)
949 * The drop rate array needs tuning for real environments.
950 * Called from timer bh only => no locking
952 static const char todrop_rate
[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
953 static char todrop_counter
[9] = {0};
956 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
957 This will leave enough time for normal connection to get
959 if (time_before(cp
->timeout
+ jiffies
, cp
->timer
.expires
+ 60*HZ
))
962 /* Don't drop the entry if its number of incoming packets is not
964 i
= atomic_read(&cp
->in_pkts
);
965 if (i
> 8 || i
< 0) return 0;
967 if (!todrop_rate
[i
]) return 0;
968 if (--todrop_counter
[i
] > 0) return 0;
970 todrop_counter
[i
] = todrop_rate
[i
];
974 /* Called from keventd and must protect itself from softirqs */
975 void ip_vs_random_dropentry(void)
978 struct ip_vs_conn
*cp
;
981 * Randomly scan 1/32 of the whole table every second
983 for (idx
= 0; idx
< (IP_VS_CONN_TAB_SIZE
>>5); idx
++) {
984 unsigned hash
= net_random() & IP_VS_CONN_TAB_MASK
;
987 * Lock is actually needed in this loop.
989 ct_write_lock_bh(hash
);
991 list_for_each_entry(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
992 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
993 /* connection template */
996 if (cp
->protocol
== IPPROTO_TCP
) {
998 case IP_VS_TCP_S_SYN_RECV
:
999 case IP_VS_TCP_S_SYNACK
:
1002 case IP_VS_TCP_S_ESTABLISHED
:
1003 if (todrop_entry(cp
))
1011 if (!todrop_entry(cp
))
1015 IP_VS_DBG(4, "del connection\n");
1016 ip_vs_conn_expire_now(cp
);
1018 IP_VS_DBG(4, "del conn template\n");
1019 ip_vs_conn_expire_now(cp
->control
);
1022 ct_write_unlock_bh(hash
);
1028 * Flush all the connection entries in the ip_vs_conn_tab
1030 static void ip_vs_conn_flush(void)
1033 struct ip_vs_conn
*cp
;
1036 for (idx
=0; idx
<IP_VS_CONN_TAB_SIZE
; idx
++) {
1038 * Lock is actually needed in this loop.
1040 ct_write_lock_bh(idx
);
1042 list_for_each_entry(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1044 IP_VS_DBG(4, "del connection\n");
1045 ip_vs_conn_expire_now(cp
);
1047 IP_VS_DBG(4, "del conn template\n");
1048 ip_vs_conn_expire_now(cp
->control
);
1051 ct_write_unlock_bh(idx
);
1054 /* the counter may be not NULL, because maybe some conn entries
1055 are run by slow timer handler or unhashed but still referred */
1056 if (atomic_read(&ip_vs_conn_count
) != 0) {
1063 int __init
ip_vs_conn_init(void)
1068 * Allocate the connection hash table and initialize its list heads
1070 ip_vs_conn_tab
= vmalloc(IP_VS_CONN_TAB_SIZE
*sizeof(struct list_head
));
1071 if (!ip_vs_conn_tab
)
1074 /* Allocate ip_vs_conn slab cache */
1075 ip_vs_conn_cachep
= kmem_cache_create("ip_vs_conn",
1076 sizeof(struct ip_vs_conn
), 0,
1077 SLAB_HWCACHE_ALIGN
, NULL
);
1078 if (!ip_vs_conn_cachep
) {
1079 vfree(ip_vs_conn_tab
);
1083 pr_info("Connection hash table configured "
1084 "(size=%d, memory=%ldKbytes)\n",
1085 IP_VS_CONN_TAB_SIZE
,
1086 (long)(IP_VS_CONN_TAB_SIZE
*sizeof(struct list_head
))/1024);
1087 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1088 sizeof(struct ip_vs_conn
));
1090 for (idx
= 0; idx
< IP_VS_CONN_TAB_SIZE
; idx
++) {
1091 INIT_LIST_HEAD(&ip_vs_conn_tab
[idx
]);
1094 for (idx
= 0; idx
< CT_LOCKARRAY_SIZE
; idx
++) {
1095 rwlock_init(&__ip_vs_conntbl_lock_array
[idx
].l
);
1098 proc_net_fops_create(&init_net
, "ip_vs_conn", 0, &ip_vs_conn_fops
);
1099 proc_net_fops_create(&init_net
, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops
);
1101 /* calculate the random value for connection hash */
1102 get_random_bytes(&ip_vs_conn_rnd
, sizeof(ip_vs_conn_rnd
));
1108 void ip_vs_conn_cleanup(void)
1110 /* flush all the connection entries first */
1113 /* Release the empty cache */
1114 kmem_cache_destroy(ip_vs_conn_cachep
);
1115 proc_net_remove(&init_net
, "ip_vs_conn");
1116 proc_net_remove(&init_net
, "ip_vs_conn_sync");
1117 vfree(ip_vs_conn_tab
);