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/inet.h>
31 #include <linux/net.h>
32 #include <linux/kernel.h>
33 #include <linux/module.h>
34 #include <linux/vmalloc.h>
35 #include <linux/proc_fs.h> /* for proc_net_* */
36 #include <linux/slab.h>
37 #include <linux/seq_file.h>
38 #include <linux/jhash.h>
39 #include <linux/random.h>
41 #include <net/net_namespace.h>
42 #include <net/ip_vs.h>
45 #ifndef CONFIG_IP_VS_TAB_BITS
46 #define CONFIG_IP_VS_TAB_BITS 12
50 * Connection hash size. Default is what was selected at compile time.
52 static int ip_vs_conn_tab_bits
= CONFIG_IP_VS_TAB_BITS
;
53 module_param_named(conn_tab_bits
, ip_vs_conn_tab_bits
, int, 0444);
54 MODULE_PARM_DESC(conn_tab_bits
, "Set connections' hash size");
56 /* size and mask values */
57 int ip_vs_conn_tab_size __read_mostly
;
58 static int ip_vs_conn_tab_mask __read_mostly
;
61 * Connection hash table: for input and output packets lookups of IPVS
63 static struct hlist_head
*ip_vs_conn_tab __read_mostly
;
65 /* SLAB cache for IPVS connections */
66 static struct kmem_cache
*ip_vs_conn_cachep __read_mostly
;
68 /* counter for no client port connections */
69 static atomic_t ip_vs_conn_no_cport_cnt
= ATOMIC_INIT(0);
71 /* random value for IPVS connection hash */
72 static unsigned int ip_vs_conn_rnd __read_mostly
;
75 * Fine locking granularity for big connection hash table
77 #define CT_LOCKARRAY_BITS 5
78 #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
79 #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
81 /* We need an addrstrlen that works with or without v6 */
82 #ifdef CONFIG_IP_VS_IPV6
83 #define IP_VS_ADDRSTRLEN INET6_ADDRSTRLEN
85 #define IP_VS_ADDRSTRLEN (8+1)
88 struct ip_vs_aligned_lock
91 } __attribute__((__aligned__(SMP_CACHE_BYTES
)));
93 /* lock array for conn table */
94 static struct ip_vs_aligned_lock
95 __ip_vs_conntbl_lock_array
[CT_LOCKARRAY_SIZE
] __cacheline_aligned
;
97 static inline void ct_write_lock_bh(unsigned int key
)
99 spin_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
102 static inline void ct_write_unlock_bh(unsigned int key
)
104 spin_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
109 * Returns hash value for IPVS connection entry
111 static unsigned int ip_vs_conn_hashkey(struct net
*net
, int af
, unsigned int proto
,
112 const union nf_inet_addr
*addr
,
115 #ifdef CONFIG_IP_VS_IPV6
117 return (jhash_3words(jhash(addr
, 16, ip_vs_conn_rnd
),
118 (__force u32
)port
, proto
, ip_vs_conn_rnd
) ^
119 ((size_t)net
>>8)) & ip_vs_conn_tab_mask
;
121 return (jhash_3words((__force u32
)addr
->ip
, (__force u32
)port
, proto
,
123 ((size_t)net
>>8)) & ip_vs_conn_tab_mask
;
126 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param
*p
,
129 const union nf_inet_addr
*addr
;
132 if (p
->pe_data
&& p
->pe
->hashkey_raw
)
133 return p
->pe
->hashkey_raw(p
, ip_vs_conn_rnd
, inverse
) &
136 if (likely(!inverse
)) {
144 return ip_vs_conn_hashkey(p
->net
, p
->af
, p
->protocol
, addr
, port
);
147 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn
*cp
)
149 struct ip_vs_conn_param p
;
151 ip_vs_conn_fill_param(ip_vs_conn_net(cp
), cp
->af
, cp
->protocol
,
152 &cp
->caddr
, cp
->cport
, NULL
, 0, &p
);
156 p
.pe_data
= cp
->pe_data
;
157 p
.pe_data_len
= cp
->pe_data_len
;
160 return ip_vs_conn_hashkey_param(&p
, false);
164 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
165 * returns bool success.
167 static inline int ip_vs_conn_hash(struct ip_vs_conn
*cp
)
172 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
175 /* Hash by protocol, client address and port */
176 hash
= ip_vs_conn_hashkey_conn(cp
);
178 ct_write_lock_bh(hash
);
179 spin_lock(&cp
->lock
);
181 if (!(cp
->flags
& IP_VS_CONN_F_HASHED
)) {
182 cp
->flags
|= IP_VS_CONN_F_HASHED
;
183 atomic_inc(&cp
->refcnt
);
184 hlist_add_head_rcu(&cp
->c_list
, &ip_vs_conn_tab
[hash
]);
187 pr_err("%s(): request for already hashed, called from %pF\n",
188 __func__
, __builtin_return_address(0));
192 spin_unlock(&cp
->lock
);
193 ct_write_unlock_bh(hash
);
200 * UNhashes ip_vs_conn from ip_vs_conn_tab.
201 * returns bool success. Caller should hold conn reference.
203 static inline int ip_vs_conn_unhash(struct ip_vs_conn
*cp
)
208 /* unhash it and decrease its reference counter */
209 hash
= ip_vs_conn_hashkey_conn(cp
);
211 ct_write_lock_bh(hash
);
212 spin_lock(&cp
->lock
);
214 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
215 hlist_del_rcu(&cp
->c_list
);
216 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
217 atomic_dec(&cp
->refcnt
);
222 spin_unlock(&cp
->lock
);
223 ct_write_unlock_bh(hash
);
228 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
229 * returns bool success.
231 static inline bool ip_vs_conn_unlink(struct ip_vs_conn
*cp
)
236 hash
= ip_vs_conn_hashkey_conn(cp
);
238 ct_write_lock_bh(hash
);
239 spin_lock(&cp
->lock
);
241 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
243 /* Decrease refcnt and unlink conn only if we are last user */
244 if (atomic_cmpxchg(&cp
->refcnt
, 1, 0) == 1) {
245 hlist_del_rcu(&cp
->c_list
);
246 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
250 ret
= atomic_read(&cp
->refcnt
) ? false : true;
252 spin_unlock(&cp
->lock
);
253 ct_write_unlock_bh(hash
);
260 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
261 * Called for pkts coming from OUTside-to-INside.
262 * p->caddr, p->cport: pkt source address (foreign host)
263 * p->vaddr, p->vport: pkt dest address (load balancer)
265 static inline struct ip_vs_conn
*
266 __ip_vs_conn_in_get(const struct ip_vs_conn_param
*p
)
269 struct ip_vs_conn
*cp
;
271 hash
= ip_vs_conn_hashkey_param(p
, false);
275 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
276 if (p
->cport
== cp
->cport
&& p
->vport
== cp
->vport
&&
278 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->caddr
) &&
279 ip_vs_addr_equal(p
->af
, p
->vaddr
, &cp
->vaddr
) &&
280 ((!p
->cport
) ^ (!(cp
->flags
& IP_VS_CONN_F_NO_CPORT
))) &&
281 p
->protocol
== cp
->protocol
&&
282 ip_vs_conn_net_eq(cp
, p
->net
)) {
283 if (!__ip_vs_conn_get(cp
))
296 struct ip_vs_conn
*ip_vs_conn_in_get(const struct ip_vs_conn_param
*p
)
298 struct ip_vs_conn
*cp
;
300 cp
= __ip_vs_conn_in_get(p
);
301 if (!cp
&& atomic_read(&ip_vs_conn_no_cport_cnt
)) {
302 struct ip_vs_conn_param cport_zero_p
= *p
;
303 cport_zero_p
.cport
= 0;
304 cp
= __ip_vs_conn_in_get(&cport_zero_p
);
307 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
308 ip_vs_proto_name(p
->protocol
),
309 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
310 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
311 cp
? "hit" : "not hit");
317 ip_vs_conn_fill_param_proto(int af
, const struct sk_buff
*skb
,
318 const struct ip_vs_iphdr
*iph
,
319 int inverse
, struct ip_vs_conn_param
*p
)
321 __be16 _ports
[2], *pptr
;
322 struct net
*net
= skb_net(skb
);
324 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
328 if (likely(!inverse
))
329 ip_vs_conn_fill_param(net
, af
, iph
->protocol
, &iph
->saddr
,
330 pptr
[0], &iph
->daddr
, pptr
[1], p
);
332 ip_vs_conn_fill_param(net
, af
, iph
->protocol
, &iph
->daddr
,
333 pptr
[1], &iph
->saddr
, pptr
[0], p
);
338 ip_vs_conn_in_get_proto(int af
, const struct sk_buff
*skb
,
339 const struct ip_vs_iphdr
*iph
, int inverse
)
341 struct ip_vs_conn_param p
;
343 if (ip_vs_conn_fill_param_proto(af
, skb
, iph
, inverse
, &p
))
346 return ip_vs_conn_in_get(&p
);
348 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto
);
350 /* Get reference to connection template */
351 struct ip_vs_conn
*ip_vs_ct_in_get(const struct ip_vs_conn_param
*p
)
354 struct ip_vs_conn
*cp
;
356 hash
= ip_vs_conn_hashkey_param(p
, false);
360 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
361 if (unlikely(p
->pe_data
&& p
->pe
->ct_match
)) {
362 if (!ip_vs_conn_net_eq(cp
, p
->net
))
364 if (p
->pe
== cp
->pe
&& p
->pe
->ct_match(p
, cp
)) {
365 if (__ip_vs_conn_get(cp
))
371 if (cp
->af
== p
->af
&&
372 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->caddr
) &&
373 /* protocol should only be IPPROTO_IP if
374 * p->vaddr is a fwmark */
375 ip_vs_addr_equal(p
->protocol
== IPPROTO_IP
? AF_UNSPEC
:
376 p
->af
, p
->vaddr
, &cp
->vaddr
) &&
377 p
->vport
== cp
->vport
&& p
->cport
== cp
->cport
&&
378 cp
->flags
& IP_VS_CONN_F_TEMPLATE
&&
379 p
->protocol
== cp
->protocol
&&
380 ip_vs_conn_net_eq(cp
, p
->net
)) {
381 if (__ip_vs_conn_get(cp
))
390 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
391 ip_vs_proto_name(p
->protocol
),
392 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
393 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
394 cp
? "hit" : "not hit");
399 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
400 * Called for pkts coming from inside-to-OUTside.
401 * p->caddr, p->cport: pkt source address (inside host)
402 * p->vaddr, p->vport: pkt dest address (foreign host) */
403 struct ip_vs_conn
*ip_vs_conn_out_get(const struct ip_vs_conn_param
*p
)
406 struct ip_vs_conn
*cp
, *ret
=NULL
;
409 * Check for "full" addressed entries
411 hash
= ip_vs_conn_hashkey_param(p
, true);
415 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
416 if (p
->vport
== cp
->cport
&& p
->cport
== cp
->dport
&&
418 ip_vs_addr_equal(p
->af
, p
->vaddr
, &cp
->caddr
) &&
419 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->daddr
) &&
420 p
->protocol
== cp
->protocol
&&
421 ip_vs_conn_net_eq(cp
, p
->net
)) {
422 if (!__ip_vs_conn_get(cp
))
432 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
433 ip_vs_proto_name(p
->protocol
),
434 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
435 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
436 ret
? "hit" : "not hit");
442 ip_vs_conn_out_get_proto(int af
, const struct sk_buff
*skb
,
443 const struct ip_vs_iphdr
*iph
, int inverse
)
445 struct ip_vs_conn_param p
;
447 if (ip_vs_conn_fill_param_proto(af
, skb
, iph
, inverse
, &p
))
450 return ip_vs_conn_out_get(&p
);
452 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto
);
455 * Put back the conn and restart its timer with its timeout
457 void ip_vs_conn_put(struct ip_vs_conn
*cp
)
459 unsigned long t
= (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
) ?
461 mod_timer(&cp
->timer
, jiffies
+t
);
463 __ip_vs_conn_put(cp
);
468 * Fill a no_client_port connection with a client port number
470 void ip_vs_conn_fill_cport(struct ip_vs_conn
*cp
, __be16 cport
)
472 if (ip_vs_conn_unhash(cp
)) {
473 spin_lock_bh(&cp
->lock
);
474 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
) {
475 atomic_dec(&ip_vs_conn_no_cport_cnt
);
476 cp
->flags
&= ~IP_VS_CONN_F_NO_CPORT
;
479 spin_unlock_bh(&cp
->lock
);
481 /* hash on new dport */
488 * Bind a connection entry with the corresponding packet_xmit.
489 * Called by ip_vs_conn_new.
491 static inline void ip_vs_bind_xmit(struct ip_vs_conn
*cp
)
493 switch (IP_VS_FWD_METHOD(cp
)) {
494 case IP_VS_CONN_F_MASQ
:
495 cp
->packet_xmit
= ip_vs_nat_xmit
;
498 case IP_VS_CONN_F_TUNNEL
:
499 #ifdef CONFIG_IP_VS_IPV6
500 if (cp
->daf
== AF_INET6
)
501 cp
->packet_xmit
= ip_vs_tunnel_xmit_v6
;
504 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
507 case IP_VS_CONN_F_DROUTE
:
508 cp
->packet_xmit
= ip_vs_dr_xmit
;
511 case IP_VS_CONN_F_LOCALNODE
:
512 cp
->packet_xmit
= ip_vs_null_xmit
;
515 case IP_VS_CONN_F_BYPASS
:
516 cp
->packet_xmit
= ip_vs_bypass_xmit
;
521 #ifdef CONFIG_IP_VS_IPV6
522 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn
*cp
)
524 switch (IP_VS_FWD_METHOD(cp
)) {
525 case IP_VS_CONN_F_MASQ
:
526 cp
->packet_xmit
= ip_vs_nat_xmit_v6
;
529 case IP_VS_CONN_F_TUNNEL
:
530 if (cp
->daf
== AF_INET6
)
531 cp
->packet_xmit
= ip_vs_tunnel_xmit_v6
;
533 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
536 case IP_VS_CONN_F_DROUTE
:
537 cp
->packet_xmit
= ip_vs_dr_xmit_v6
;
540 case IP_VS_CONN_F_LOCALNODE
:
541 cp
->packet_xmit
= ip_vs_null_xmit
;
544 case IP_VS_CONN_F_BYPASS
:
545 cp
->packet_xmit
= ip_vs_bypass_xmit_v6
;
552 static inline int ip_vs_dest_totalconns(struct ip_vs_dest
*dest
)
554 return atomic_read(&dest
->activeconns
)
555 + atomic_read(&dest
->inactconns
);
559 * Bind a connection entry with a virtual service destination
560 * Called just after a new connection entry is created.
563 ip_vs_bind_dest(struct ip_vs_conn
*cp
, struct ip_vs_dest
*dest
)
565 unsigned int conn_flags
;
568 /* if dest is NULL, then return directly */
572 /* Increase the refcnt counter of the dest */
573 ip_vs_dest_hold(dest
);
575 conn_flags
= atomic_read(&dest
->conn_flags
);
576 if (cp
->protocol
!= IPPROTO_UDP
)
577 conn_flags
&= ~IP_VS_CONN_F_ONE_PACKET
;
579 /* Bind with the destination and its corresponding transmitter */
580 if (flags
& IP_VS_CONN_F_SYNC
) {
581 /* if the connection is not template and is created
582 * by sync, preserve the activity flag.
584 if (!(flags
& IP_VS_CONN_F_TEMPLATE
))
585 conn_flags
&= ~IP_VS_CONN_F_INACTIVE
;
586 /* connections inherit forwarding method from dest */
587 flags
&= ~(IP_VS_CONN_F_FWD_MASK
| IP_VS_CONN_F_NOOUTPUT
);
593 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
594 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
596 ip_vs_proto_name(cp
->protocol
),
597 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
598 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
599 IP_VS_DBG_ADDR(cp
->daf
, &cp
->daddr
), ntohs(cp
->dport
),
600 ip_vs_fwd_tag(cp
), cp
->state
,
601 cp
->flags
, atomic_read(&cp
->refcnt
),
602 atomic_read(&dest
->refcnt
));
604 /* Update the connection counters */
605 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
606 /* It is a normal connection, so modify the counters
607 * according to the flags, later the protocol can
608 * update them on state change
610 if (!(flags
& IP_VS_CONN_F_INACTIVE
))
611 atomic_inc(&dest
->activeconns
);
613 atomic_inc(&dest
->inactconns
);
615 /* It is a persistent connection/template, so increase
616 the persistent connection counter */
617 atomic_inc(&dest
->persistconns
);
620 if (dest
->u_threshold
!= 0 &&
621 ip_vs_dest_totalconns(dest
) >= dest
->u_threshold
)
622 dest
->flags
|= IP_VS_DEST_F_OVERLOAD
;
627 * Check if there is a destination for the connection, if so
628 * bind the connection to the destination.
630 void ip_vs_try_bind_dest(struct ip_vs_conn
*cp
)
632 struct ip_vs_dest
*dest
;
636 /* This function is only invoked by the synchronization code. We do
637 * not currently support heterogeneous pools with synchronization,
638 * so we can make the assumption that the svc_af is the same as the
641 dest
= ip_vs_find_dest(ip_vs_conn_net(cp
), cp
->af
, cp
->af
, &cp
->daddr
,
642 cp
->dport
, &cp
->vaddr
, cp
->vport
,
643 cp
->protocol
, cp
->fwmark
, cp
->flags
);
645 struct ip_vs_proto_data
*pd
;
647 spin_lock_bh(&cp
->lock
);
649 spin_unlock_bh(&cp
->lock
);
654 /* Applications work depending on the forwarding method
655 * but better to reassign them always when binding dest */
657 ip_vs_unbind_app(cp
);
659 ip_vs_bind_dest(cp
, dest
);
660 spin_unlock_bh(&cp
->lock
);
662 /* Update its packet transmitter */
663 cp
->packet_xmit
= NULL
;
664 #ifdef CONFIG_IP_VS_IPV6
665 if (cp
->af
== AF_INET6
)
666 ip_vs_bind_xmit_v6(cp
);
671 pd
= ip_vs_proto_data_get(ip_vs_conn_net(cp
), cp
->protocol
);
672 if (pd
&& atomic_read(&pd
->appcnt
))
673 ip_vs_bind_app(cp
, pd
->pp
);
680 * Unbind a connection entry with its VS destination
681 * Called by the ip_vs_conn_expire function.
683 static inline void ip_vs_unbind_dest(struct ip_vs_conn
*cp
)
685 struct ip_vs_dest
*dest
= cp
->dest
;
690 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
691 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
693 ip_vs_proto_name(cp
->protocol
),
694 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
695 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
696 IP_VS_DBG_ADDR(cp
->daf
, &cp
->daddr
), ntohs(cp
->dport
),
697 ip_vs_fwd_tag(cp
), cp
->state
,
698 cp
->flags
, atomic_read(&cp
->refcnt
),
699 atomic_read(&dest
->refcnt
));
701 /* Update the connection counters */
702 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
703 /* It is a normal connection, so decrease the inactconns
704 or activeconns counter */
705 if (cp
->flags
& IP_VS_CONN_F_INACTIVE
) {
706 atomic_dec(&dest
->inactconns
);
708 atomic_dec(&dest
->activeconns
);
711 /* It is a persistent connection/template, so decrease
712 the persistent connection counter */
713 atomic_dec(&dest
->persistconns
);
716 if (dest
->l_threshold
!= 0) {
717 if (ip_vs_dest_totalconns(dest
) < dest
->l_threshold
)
718 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
719 } else if (dest
->u_threshold
!= 0) {
720 if (ip_vs_dest_totalconns(dest
) * 4 < dest
->u_threshold
* 3)
721 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
723 if (dest
->flags
& IP_VS_DEST_F_OVERLOAD
)
724 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
727 ip_vs_dest_put(dest
);
730 static int expire_quiescent_template(struct netns_ipvs
*ipvs
,
731 struct ip_vs_dest
*dest
)
734 return ipvs
->sysctl_expire_quiescent_template
&&
735 (atomic_read(&dest
->weight
) == 0);
742 * Checking if the destination of a connection template is available.
743 * If available, return 1, otherwise invalidate this connection
744 * template and return 0.
746 int ip_vs_check_template(struct ip_vs_conn
*ct
)
748 struct ip_vs_dest
*dest
= ct
->dest
;
749 struct netns_ipvs
*ipvs
= net_ipvs(ip_vs_conn_net(ct
));
752 * Checking the dest server status.
754 if ((dest
== NULL
) ||
755 !(dest
->flags
& IP_VS_DEST_F_AVAILABLE
) ||
756 expire_quiescent_template(ipvs
, dest
)) {
757 IP_VS_DBG_BUF(9, "check_template: dest not available for "
758 "protocol %s s:%s:%d v:%s:%d "
760 ip_vs_proto_name(ct
->protocol
),
761 IP_VS_DBG_ADDR(ct
->af
, &ct
->caddr
),
763 IP_VS_DBG_ADDR(ct
->af
, &ct
->vaddr
),
765 IP_VS_DBG_ADDR(ct
->daf
, &ct
->daddr
),
769 * Invalidate the connection template
771 if (ct
->vport
!= htons(0xffff)) {
772 if (ip_vs_conn_unhash(ct
)) {
773 ct
->dport
= htons(0xffff);
774 ct
->vport
= htons(0xffff);
781 * Simply decrease the refcnt of the template,
782 * don't restart its timer.
784 __ip_vs_conn_put(ct
);
790 static void ip_vs_conn_rcu_free(struct rcu_head
*head
)
792 struct ip_vs_conn
*cp
= container_of(head
, struct ip_vs_conn
,
795 ip_vs_pe_put(cp
->pe
);
797 kmem_cache_free(ip_vs_conn_cachep
, cp
);
800 static void ip_vs_conn_expire(unsigned long data
)
802 struct ip_vs_conn
*cp
= (struct ip_vs_conn
*)data
;
803 struct net
*net
= ip_vs_conn_net(cp
);
804 struct netns_ipvs
*ipvs
= net_ipvs(net
);
807 * do I control anybody?
809 if (atomic_read(&cp
->n_control
))
812 /* Unlink conn if not referenced anymore */
813 if (likely(ip_vs_conn_unlink(cp
))) {
814 /* delete the timer if it is activated by other users */
815 del_timer(&cp
->timer
);
817 /* does anybody control me? */
819 ip_vs_control_del(cp
);
821 if (cp
->flags
& IP_VS_CONN_F_NFCT
) {
822 /* Do not access conntracks during subsys cleanup
823 * because nf_conntrack_find_get can not be used after
824 * conntrack cleanup for the net.
828 ip_vs_conn_drop_conntrack(cp
);
831 if (unlikely(cp
->app
!= NULL
))
832 ip_vs_unbind_app(cp
);
833 ip_vs_unbind_dest(cp
);
834 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
)
835 atomic_dec(&ip_vs_conn_no_cport_cnt
);
836 call_rcu(&cp
->rcu_head
, ip_vs_conn_rcu_free
);
837 atomic_dec(&ipvs
->conn_count
);
842 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
843 atomic_read(&cp
->refcnt
),
844 atomic_read(&cp
->n_control
));
846 atomic_inc(&cp
->refcnt
);
849 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
850 ip_vs_sync_conn(net
, cp
, sysctl_sync_threshold(ipvs
));
855 /* Modify timer, so that it expires as soon as possible.
856 * Can be called without reference only if under RCU lock.
858 void ip_vs_conn_expire_now(struct ip_vs_conn
*cp
)
860 /* Using mod_timer_pending will ensure the timer is not
861 * modified after the final del_timer in ip_vs_conn_expire.
863 if (timer_pending(&cp
->timer
) &&
864 time_after(cp
->timer
.expires
, jiffies
))
865 mod_timer_pending(&cp
->timer
, jiffies
);
870 * Create a new connection entry and hash it into the ip_vs_conn_tab
873 ip_vs_conn_new(const struct ip_vs_conn_param
*p
, int dest_af
,
874 const union nf_inet_addr
*daddr
, __be16 dport
, unsigned int flags
,
875 struct ip_vs_dest
*dest
, __u32 fwmark
)
877 struct ip_vs_conn
*cp
;
878 struct netns_ipvs
*ipvs
= net_ipvs(p
->net
);
879 struct ip_vs_proto_data
*pd
= ip_vs_proto_data_get(p
->net
,
882 cp
= kmem_cache_alloc(ip_vs_conn_cachep
, GFP_ATOMIC
);
884 IP_VS_ERR_RL("%s(): no memory\n", __func__
);
888 INIT_HLIST_NODE(&cp
->c_list
);
889 setup_timer(&cp
->timer
, ip_vs_conn_expire
, (unsigned long)cp
);
890 ip_vs_conn_net_set(cp
, p
->net
);
893 cp
->protocol
= p
->protocol
;
894 ip_vs_addr_set(p
->af
, &cp
->caddr
, p
->caddr
);
895 cp
->cport
= p
->cport
;
896 /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
897 ip_vs_addr_set(p
->protocol
== IPPROTO_IP
? AF_UNSPEC
: p
->af
,
898 &cp
->vaddr
, p
->vaddr
);
899 cp
->vport
= p
->vport
;
900 ip_vs_addr_set(cp
->daf
, &cp
->daddr
, daddr
);
904 if (flags
& IP_VS_CONN_F_TEMPLATE
&& p
->pe
) {
907 cp
->pe_data
= p
->pe_data
;
908 cp
->pe_data_len
= p
->pe_data_len
;
914 spin_lock_init(&cp
->lock
);
917 * Set the entry is referenced by the current thread before hashing
918 * it in the table, so that other thread run ip_vs_random_dropentry
919 * but cannot drop this entry.
921 atomic_set(&cp
->refcnt
, 1);
924 atomic_set(&cp
->n_control
, 0);
925 atomic_set(&cp
->in_pkts
, 0);
927 cp
->packet_xmit
= NULL
;
930 /* reset struct ip_vs_seq */
931 cp
->in_seq
.delta
= 0;
932 cp
->out_seq
.delta
= 0;
934 atomic_inc(&ipvs
->conn_count
);
935 if (flags
& IP_VS_CONN_F_NO_CPORT
)
936 atomic_inc(&ip_vs_conn_no_cport_cnt
);
938 /* Bind the connection with a destination server */
940 ip_vs_bind_dest(cp
, dest
);
942 /* Set its state and timeout */
946 cp
->sync_endtime
= jiffies
& ~3UL;
948 /* Bind its packet transmitter */
949 #ifdef CONFIG_IP_VS_IPV6
950 if (p
->af
== AF_INET6
)
951 ip_vs_bind_xmit_v6(cp
);
956 if (unlikely(pd
&& atomic_read(&pd
->appcnt
)))
957 ip_vs_bind_app(cp
, pd
->pp
);
960 * Allow conntrack to be preserved. By default, conntrack
961 * is created and destroyed for every packet.
962 * Sometimes keeping conntrack can be useful for
963 * IP_VS_CONN_F_ONE_PACKET too.
966 if (ip_vs_conntrack_enabled(ipvs
))
967 cp
->flags
|= IP_VS_CONN_F_NFCT
;
969 /* Hash it in the ip_vs_conn_tab finally */
976 * /proc/net/ip_vs_conn entries
978 #ifdef CONFIG_PROC_FS
979 struct ip_vs_iter_state
{
980 struct seq_net_private p
;
981 struct hlist_head
*l
;
984 static void *ip_vs_conn_array(struct seq_file
*seq
, loff_t pos
)
987 struct ip_vs_conn
*cp
;
988 struct ip_vs_iter_state
*iter
= seq
->private;
990 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++) {
991 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
992 /* __ip_vs_conn_get() is not needed by
993 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
996 iter
->l
= &ip_vs_conn_tab
[idx
];
1006 static void *ip_vs_conn_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1009 struct ip_vs_iter_state
*iter
= seq
->private;
1013 return *pos
? ip_vs_conn_array(seq
, *pos
- 1) :SEQ_START_TOKEN
;
1016 static void *ip_vs_conn_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1018 struct ip_vs_conn
*cp
= v
;
1019 struct ip_vs_iter_state
*iter
= seq
->private;
1020 struct hlist_node
*e
;
1021 struct hlist_head
*l
= iter
->l
;
1025 if (v
== SEQ_START_TOKEN
)
1026 return ip_vs_conn_array(seq
, 0);
1028 /* more on same hash chain? */
1029 e
= rcu_dereference(hlist_next_rcu(&cp
->c_list
));
1031 return hlist_entry(e
, struct ip_vs_conn
, c_list
);
1033 idx
= l
- ip_vs_conn_tab
;
1034 while (++idx
< ip_vs_conn_tab_size
) {
1035 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1036 iter
->l
= &ip_vs_conn_tab
[idx
];
1045 static void ip_vs_conn_seq_stop(struct seq_file
*seq
, void *v
)
1051 static int ip_vs_conn_seq_show(struct seq_file
*seq
, void *v
)
1054 if (v
== SEQ_START_TOKEN
)
1056 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1058 const struct ip_vs_conn
*cp
= v
;
1059 struct net
*net
= seq_file_net(seq
);
1060 char pe_data
[IP_VS_PENAME_MAXLEN
+ IP_VS_PEDATA_MAXLEN
+ 3];
1062 char dbuf
[IP_VS_ADDRSTRLEN
];
1064 if (!ip_vs_conn_net_eq(cp
, net
))
1068 len
= strlen(cp
->pe
->name
);
1069 memcpy(pe_data
+ 1, cp
->pe
->name
, len
);
1070 pe_data
[len
+ 1] = ' ';
1072 len
+= cp
->pe
->show_pe_data(cp
, pe_data
+ len
);
1074 pe_data
[len
] = '\0';
1076 #ifdef CONFIG_IP_VS_IPV6
1077 if (cp
->daf
== AF_INET6
)
1078 snprintf(dbuf
, sizeof(dbuf
), "%pI6", &cp
->daddr
.in6
);
1081 snprintf(dbuf
, sizeof(dbuf
), "%08X",
1082 ntohl(cp
->daddr
.ip
));
1084 #ifdef CONFIG_IP_VS_IPV6
1085 if (cp
->af
== AF_INET6
)
1086 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X "
1087 "%s %04X %-11s %7lu%s\n",
1088 ip_vs_proto_name(cp
->protocol
),
1089 &cp
->caddr
.in6
, ntohs(cp
->cport
),
1090 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
1091 dbuf
, ntohs(cp
->dport
),
1092 ip_vs_state_name(cp
->protocol
, cp
->state
),
1093 (cp
->timer
.expires
-jiffies
)/HZ
, pe_data
);
1097 "%-3s %08X %04X %08X %04X"
1098 " %s %04X %-11s %7lu%s\n",
1099 ip_vs_proto_name(cp
->protocol
),
1100 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
1101 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
1102 dbuf
, ntohs(cp
->dport
),
1103 ip_vs_state_name(cp
->protocol
, cp
->state
),
1104 (cp
->timer
.expires
-jiffies
)/HZ
, pe_data
);
1109 static const struct seq_operations ip_vs_conn_seq_ops
= {
1110 .start
= ip_vs_conn_seq_start
,
1111 .next
= ip_vs_conn_seq_next
,
1112 .stop
= ip_vs_conn_seq_stop
,
1113 .show
= ip_vs_conn_seq_show
,
1116 static int ip_vs_conn_open(struct inode
*inode
, struct file
*file
)
1118 return seq_open_net(inode
, file
, &ip_vs_conn_seq_ops
,
1119 sizeof(struct ip_vs_iter_state
));
1122 static const struct file_operations ip_vs_conn_fops
= {
1123 .owner
= THIS_MODULE
,
1124 .open
= ip_vs_conn_open
,
1126 .llseek
= seq_lseek
,
1127 .release
= seq_release_net
,
1130 static const char *ip_vs_origin_name(unsigned int flags
)
1132 if (flags
& IP_VS_CONN_F_SYNC
)
1138 static int ip_vs_conn_sync_seq_show(struct seq_file
*seq
, void *v
)
1140 char dbuf
[IP_VS_ADDRSTRLEN
];
1142 if (v
== SEQ_START_TOKEN
)
1144 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1146 const struct ip_vs_conn
*cp
= v
;
1147 struct net
*net
= seq_file_net(seq
);
1149 if (!ip_vs_conn_net_eq(cp
, net
))
1152 #ifdef CONFIG_IP_VS_IPV6
1153 if (cp
->daf
== AF_INET6
)
1154 snprintf(dbuf
, sizeof(dbuf
), "%pI6", &cp
->daddr
.in6
);
1157 snprintf(dbuf
, sizeof(dbuf
), "%08X",
1158 ntohl(cp
->daddr
.ip
));
1160 #ifdef CONFIG_IP_VS_IPV6
1161 if (cp
->af
== AF_INET6
)
1162 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X "
1163 "%s %04X %-11s %-6s %7lu\n",
1164 ip_vs_proto_name(cp
->protocol
),
1165 &cp
->caddr
.in6
, ntohs(cp
->cport
),
1166 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
1167 dbuf
, ntohs(cp
->dport
),
1168 ip_vs_state_name(cp
->protocol
, cp
->state
),
1169 ip_vs_origin_name(cp
->flags
),
1170 (cp
->timer
.expires
-jiffies
)/HZ
);
1174 "%-3s %08X %04X %08X %04X "
1175 "%s %04X %-11s %-6s %7lu\n",
1176 ip_vs_proto_name(cp
->protocol
),
1177 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
1178 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
1179 dbuf
, ntohs(cp
->dport
),
1180 ip_vs_state_name(cp
->protocol
, cp
->state
),
1181 ip_vs_origin_name(cp
->flags
),
1182 (cp
->timer
.expires
-jiffies
)/HZ
);
1187 static const struct seq_operations ip_vs_conn_sync_seq_ops
= {
1188 .start
= ip_vs_conn_seq_start
,
1189 .next
= ip_vs_conn_seq_next
,
1190 .stop
= ip_vs_conn_seq_stop
,
1191 .show
= ip_vs_conn_sync_seq_show
,
1194 static int ip_vs_conn_sync_open(struct inode
*inode
, struct file
*file
)
1196 return seq_open_net(inode
, file
, &ip_vs_conn_sync_seq_ops
,
1197 sizeof(struct ip_vs_iter_state
));
1200 static const struct file_operations ip_vs_conn_sync_fops
= {
1201 .owner
= THIS_MODULE
,
1202 .open
= ip_vs_conn_sync_open
,
1204 .llseek
= seq_lseek
,
1205 .release
= seq_release_net
,
1212 * Randomly drop connection entries before running out of memory
1214 static inline int todrop_entry(struct ip_vs_conn
*cp
)
1217 * The drop rate array needs tuning for real environments.
1218 * Called from timer bh only => no locking
1220 static const char todrop_rate
[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1221 static char todrop_counter
[9] = {0};
1224 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1225 This will leave enough time for normal connection to get
1227 if (time_before(cp
->timeout
+ jiffies
, cp
->timer
.expires
+ 60*HZ
))
1230 /* Don't drop the entry if its number of incoming packets is not
1231 located in [0, 8] */
1232 i
= atomic_read(&cp
->in_pkts
);
1233 if (i
> 8 || i
< 0) return 0;
1235 if (!todrop_rate
[i
]) return 0;
1236 if (--todrop_counter
[i
] > 0) return 0;
1238 todrop_counter
[i
] = todrop_rate
[i
];
1242 /* Called from keventd and must protect itself from softirqs */
1243 void ip_vs_random_dropentry(struct net
*net
)
1246 struct ip_vs_conn
*cp
, *cp_c
;
1250 * Randomly scan 1/32 of the whole table every second
1252 for (idx
= 0; idx
< (ip_vs_conn_tab_size
>>5); idx
++) {
1253 unsigned int hash
= prandom_u32() & ip_vs_conn_tab_mask
;
1255 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
1256 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
1257 /* connection template */
1259 if (!ip_vs_conn_net_eq(cp
, net
))
1261 if (cp
->protocol
== IPPROTO_TCP
) {
1263 case IP_VS_TCP_S_SYN_RECV
:
1264 case IP_VS_TCP_S_SYNACK
:
1267 case IP_VS_TCP_S_ESTABLISHED
:
1268 if (todrop_entry(cp
))
1275 } else if (cp
->protocol
== IPPROTO_SCTP
) {
1276 switch (cp
->state
) {
1277 case IP_VS_SCTP_S_INIT1
:
1278 case IP_VS_SCTP_S_INIT
:
1280 case IP_VS_SCTP_S_ESTABLISHED
:
1281 if (todrop_entry(cp
))
1288 if (!todrop_entry(cp
))
1292 IP_VS_DBG(4, "del connection\n");
1293 ip_vs_conn_expire_now(cp
);
1295 /* cp->control is valid only with reference to cp */
1296 if (cp_c
&& __ip_vs_conn_get(cp
)) {
1297 IP_VS_DBG(4, "del conn template\n");
1298 ip_vs_conn_expire_now(cp_c
);
1299 __ip_vs_conn_put(cp
);
1309 * Flush all the connection entries in the ip_vs_conn_tab
1311 static void ip_vs_conn_flush(struct net
*net
)
1314 struct ip_vs_conn
*cp
, *cp_c
;
1315 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1319 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++) {
1321 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1322 if (!ip_vs_conn_net_eq(cp
, net
))
1324 IP_VS_DBG(4, "del connection\n");
1325 ip_vs_conn_expire_now(cp
);
1327 /* cp->control is valid only with reference to cp */
1328 if (cp_c
&& __ip_vs_conn_get(cp
)) {
1329 IP_VS_DBG(4, "del conn template\n");
1330 ip_vs_conn_expire_now(cp_c
);
1331 __ip_vs_conn_put(cp
);
1338 /* the counter may be not NULL, because maybe some conn entries
1339 are run by slow timer handler or unhashed but still referred */
1340 if (atomic_read(&ipvs
->conn_count
) != 0) {
1346 * per netns init and exit
1348 int __net_init
ip_vs_conn_net_init(struct net
*net
)
1350 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1352 atomic_set(&ipvs
->conn_count
, 0);
1354 proc_create("ip_vs_conn", 0, net
->proc_net
, &ip_vs_conn_fops
);
1355 proc_create("ip_vs_conn_sync", 0, net
->proc_net
, &ip_vs_conn_sync_fops
);
1359 void __net_exit
ip_vs_conn_net_cleanup(struct net
*net
)
1361 /* flush all the connection entries first */
1362 ip_vs_conn_flush(net
);
1363 remove_proc_entry("ip_vs_conn", net
->proc_net
);
1364 remove_proc_entry("ip_vs_conn_sync", net
->proc_net
);
1367 int __init
ip_vs_conn_init(void)
1371 /* Compute size and mask */
1372 ip_vs_conn_tab_size
= 1 << ip_vs_conn_tab_bits
;
1373 ip_vs_conn_tab_mask
= ip_vs_conn_tab_size
- 1;
1376 * Allocate the connection hash table and initialize its list heads
1378 ip_vs_conn_tab
= vmalloc(ip_vs_conn_tab_size
* sizeof(*ip_vs_conn_tab
));
1379 if (!ip_vs_conn_tab
)
1382 /* Allocate ip_vs_conn slab cache */
1383 ip_vs_conn_cachep
= kmem_cache_create("ip_vs_conn",
1384 sizeof(struct ip_vs_conn
), 0,
1385 SLAB_HWCACHE_ALIGN
, NULL
);
1386 if (!ip_vs_conn_cachep
) {
1387 vfree(ip_vs_conn_tab
);
1391 pr_info("Connection hash table configured "
1392 "(size=%d, memory=%ldKbytes)\n",
1393 ip_vs_conn_tab_size
,
1394 (long)(ip_vs_conn_tab_size
*sizeof(struct list_head
))/1024);
1395 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1396 sizeof(struct ip_vs_conn
));
1398 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++)
1399 INIT_HLIST_HEAD(&ip_vs_conn_tab
[idx
]);
1401 for (idx
= 0; idx
< CT_LOCKARRAY_SIZE
; idx
++) {
1402 spin_lock_init(&__ip_vs_conntbl_lock_array
[idx
].l
);
1405 /* calculate the random value for connection hash */
1406 get_random_bytes(&ip_vs_conn_rnd
, sizeof(ip_vs_conn_rnd
));
1411 void ip_vs_conn_cleanup(void)
1413 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1415 /* Release the empty cache */
1416 kmem_cache_destroy(ip_vs_conn_cachep
);
1417 vfree(ip_vs_conn_tab
);