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/slab.h>
36 #include <linux/seq_file.h>
37 #include <linux/jhash.h>
38 #include <linux/random.h>
40 #include <net/net_namespace.h>
41 #include <net/ip_vs.h>
44 #ifndef CONFIG_IP_VS_TAB_BITS
45 #define CONFIG_IP_VS_TAB_BITS 12
49 * Connection hash size. Default is what was selected at compile time.
51 static int ip_vs_conn_tab_bits
= CONFIG_IP_VS_TAB_BITS
;
52 module_param_named(conn_tab_bits
, ip_vs_conn_tab_bits
, int, 0444);
53 MODULE_PARM_DESC(conn_tab_bits
, "Set connections' hash size");
55 /* size and mask values */
56 int ip_vs_conn_tab_size __read_mostly
;
57 static int ip_vs_conn_tab_mask __read_mostly
;
60 * Connection hash table: for input and output packets lookups of IPVS
62 static struct hlist_head
*ip_vs_conn_tab __read_mostly
;
64 /* SLAB cache for IPVS connections */
65 static struct kmem_cache
*ip_vs_conn_cachep __read_mostly
;
67 /* counter for no client port connections */
68 static atomic_t ip_vs_conn_no_cport_cnt
= ATOMIC_INIT(0);
70 /* random value for IPVS connection hash */
71 static unsigned int ip_vs_conn_rnd __read_mostly
;
74 * Fine locking granularity for big connection hash table
76 #define CT_LOCKARRAY_BITS 5
77 #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
78 #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
80 struct ip_vs_aligned_lock
83 } __attribute__((__aligned__(SMP_CACHE_BYTES
)));
85 /* lock array for conn table */
86 static struct ip_vs_aligned_lock
87 __ip_vs_conntbl_lock_array
[CT_LOCKARRAY_SIZE
] __cacheline_aligned
;
89 static inline void ct_write_lock_bh(unsigned int key
)
91 spin_lock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
94 static inline void ct_write_unlock_bh(unsigned int key
)
96 spin_unlock_bh(&__ip_vs_conntbl_lock_array
[key
&CT_LOCKARRAY_MASK
].l
);
101 * Returns hash value for IPVS connection entry
103 static unsigned int ip_vs_conn_hashkey(struct net
*net
, int af
, unsigned int proto
,
104 const union nf_inet_addr
*addr
,
107 #ifdef CONFIG_IP_VS_IPV6
109 return (jhash_3words(jhash(addr
, 16, ip_vs_conn_rnd
),
110 (__force u32
)port
, proto
, ip_vs_conn_rnd
) ^
111 ((size_t)net
>>8)) & ip_vs_conn_tab_mask
;
113 return (jhash_3words((__force u32
)addr
->ip
, (__force u32
)port
, proto
,
115 ((size_t)net
>>8)) & ip_vs_conn_tab_mask
;
118 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param
*p
,
121 const union nf_inet_addr
*addr
;
124 if (p
->pe_data
&& p
->pe
->hashkey_raw
)
125 return p
->pe
->hashkey_raw(p
, ip_vs_conn_rnd
, inverse
) &
128 if (likely(!inverse
)) {
136 return ip_vs_conn_hashkey(p
->net
, p
->af
, p
->protocol
, addr
, port
);
139 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn
*cp
)
141 struct ip_vs_conn_param p
;
143 ip_vs_conn_fill_param(ip_vs_conn_net(cp
), cp
->af
, cp
->protocol
,
144 &cp
->caddr
, cp
->cport
, NULL
, 0, &p
);
148 p
.pe_data
= cp
->pe_data
;
149 p
.pe_data_len
= cp
->pe_data_len
;
152 return ip_vs_conn_hashkey_param(&p
, false);
156 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
157 * returns bool success.
159 static inline int ip_vs_conn_hash(struct ip_vs_conn
*cp
)
164 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
167 /* Hash by protocol, client address and port */
168 hash
= ip_vs_conn_hashkey_conn(cp
);
170 ct_write_lock_bh(hash
);
171 spin_lock(&cp
->lock
);
173 if (!(cp
->flags
& IP_VS_CONN_F_HASHED
)) {
174 cp
->flags
|= IP_VS_CONN_F_HASHED
;
175 atomic_inc(&cp
->refcnt
);
176 hlist_add_head_rcu(&cp
->c_list
, &ip_vs_conn_tab
[hash
]);
179 pr_err("%s(): request for already hashed, called from %pF\n",
180 __func__
, __builtin_return_address(0));
184 spin_unlock(&cp
->lock
);
185 ct_write_unlock_bh(hash
);
192 * UNhashes ip_vs_conn from ip_vs_conn_tab.
193 * returns bool success. Caller should hold conn reference.
195 static inline int ip_vs_conn_unhash(struct ip_vs_conn
*cp
)
200 /* unhash it and decrease its reference counter */
201 hash
= ip_vs_conn_hashkey_conn(cp
);
203 ct_write_lock_bh(hash
);
204 spin_lock(&cp
->lock
);
206 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
207 hlist_del_rcu(&cp
->c_list
);
208 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
209 atomic_dec(&cp
->refcnt
);
214 spin_unlock(&cp
->lock
);
215 ct_write_unlock_bh(hash
);
220 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
221 * returns bool success.
223 static inline bool ip_vs_conn_unlink(struct ip_vs_conn
*cp
)
228 hash
= ip_vs_conn_hashkey_conn(cp
);
230 ct_write_lock_bh(hash
);
231 spin_lock(&cp
->lock
);
233 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
235 /* Decrease refcnt and unlink conn only if we are last user */
236 if (atomic_cmpxchg(&cp
->refcnt
, 1, 0) == 1) {
237 hlist_del_rcu(&cp
->c_list
);
238 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
242 ret
= atomic_read(&cp
->refcnt
) ? false : true;
244 spin_unlock(&cp
->lock
);
245 ct_write_unlock_bh(hash
);
252 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
253 * Called for pkts coming from OUTside-to-INside.
254 * p->caddr, p->cport: pkt source address (foreign host)
255 * p->vaddr, p->vport: pkt dest address (load balancer)
257 static inline struct ip_vs_conn
*
258 __ip_vs_conn_in_get(const struct ip_vs_conn_param
*p
)
261 struct ip_vs_conn
*cp
;
263 hash
= ip_vs_conn_hashkey_param(p
, false);
267 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
268 if (p
->cport
== cp
->cport
&& p
->vport
== cp
->vport
&&
270 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->caddr
) &&
271 ip_vs_addr_equal(p
->af
, p
->vaddr
, &cp
->vaddr
) &&
272 ((!p
->cport
) ^ (!(cp
->flags
& IP_VS_CONN_F_NO_CPORT
))) &&
273 p
->protocol
== cp
->protocol
&&
274 ip_vs_conn_net_eq(cp
, p
->net
)) {
275 if (!__ip_vs_conn_get(cp
))
288 struct ip_vs_conn
*ip_vs_conn_in_get(const struct ip_vs_conn_param
*p
)
290 struct ip_vs_conn
*cp
;
292 cp
= __ip_vs_conn_in_get(p
);
293 if (!cp
&& atomic_read(&ip_vs_conn_no_cport_cnt
)) {
294 struct ip_vs_conn_param cport_zero_p
= *p
;
295 cport_zero_p
.cport
= 0;
296 cp
= __ip_vs_conn_in_get(&cport_zero_p
);
299 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
300 ip_vs_proto_name(p
->protocol
),
301 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
302 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
303 cp
? "hit" : "not hit");
309 ip_vs_conn_fill_param_proto(int af
, const struct sk_buff
*skb
,
310 const struct ip_vs_iphdr
*iph
,
311 int inverse
, struct ip_vs_conn_param
*p
)
313 __be16 _ports
[2], *pptr
;
314 struct net
*net
= skb_net(skb
);
316 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
320 if (likely(!inverse
))
321 ip_vs_conn_fill_param(net
, af
, iph
->protocol
, &iph
->saddr
,
322 pptr
[0], &iph
->daddr
, pptr
[1], p
);
324 ip_vs_conn_fill_param(net
, af
, iph
->protocol
, &iph
->daddr
,
325 pptr
[1], &iph
->saddr
, pptr
[0], p
);
330 ip_vs_conn_in_get_proto(int af
, const struct sk_buff
*skb
,
331 const struct ip_vs_iphdr
*iph
, int inverse
)
333 struct ip_vs_conn_param p
;
335 if (ip_vs_conn_fill_param_proto(af
, skb
, iph
, inverse
, &p
))
338 return ip_vs_conn_in_get(&p
);
340 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto
);
342 /* Get reference to connection template */
343 struct ip_vs_conn
*ip_vs_ct_in_get(const struct ip_vs_conn_param
*p
)
346 struct ip_vs_conn
*cp
;
348 hash
= ip_vs_conn_hashkey_param(p
, false);
352 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
353 if (unlikely(p
->pe_data
&& p
->pe
->ct_match
)) {
354 if (!ip_vs_conn_net_eq(cp
, p
->net
))
356 if (p
->pe
== cp
->pe
&& p
->pe
->ct_match(p
, cp
)) {
357 if (__ip_vs_conn_get(cp
))
363 if (cp
->af
== p
->af
&&
364 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->caddr
) &&
365 /* protocol should only be IPPROTO_IP if
366 * p->vaddr is a fwmark */
367 ip_vs_addr_equal(p
->protocol
== IPPROTO_IP
? AF_UNSPEC
:
368 p
->af
, p
->vaddr
, &cp
->vaddr
) &&
369 p
->vport
== cp
->vport
&& p
->cport
== cp
->cport
&&
370 cp
->flags
& IP_VS_CONN_F_TEMPLATE
&&
371 p
->protocol
== cp
->protocol
&&
372 ip_vs_conn_net_eq(cp
, p
->net
)) {
373 if (__ip_vs_conn_get(cp
))
382 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
383 ip_vs_proto_name(p
->protocol
),
384 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
385 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
386 cp
? "hit" : "not hit");
391 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
392 * Called for pkts coming from inside-to-OUTside.
393 * p->caddr, p->cport: pkt source address (inside host)
394 * p->vaddr, p->vport: pkt dest address (foreign host) */
395 struct ip_vs_conn
*ip_vs_conn_out_get(const struct ip_vs_conn_param
*p
)
398 struct ip_vs_conn
*cp
, *ret
=NULL
;
401 * Check for "full" addressed entries
403 hash
= ip_vs_conn_hashkey_param(p
, true);
407 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
408 if (p
->vport
== cp
->cport
&& p
->cport
== cp
->dport
&&
410 ip_vs_addr_equal(p
->af
, p
->vaddr
, &cp
->caddr
) &&
411 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->daddr
) &&
412 p
->protocol
== cp
->protocol
&&
413 ip_vs_conn_net_eq(cp
, p
->net
)) {
414 if (!__ip_vs_conn_get(cp
))
424 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
425 ip_vs_proto_name(p
->protocol
),
426 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
427 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
428 ret
? "hit" : "not hit");
434 ip_vs_conn_out_get_proto(int af
, const struct sk_buff
*skb
,
435 const struct ip_vs_iphdr
*iph
, int inverse
)
437 struct ip_vs_conn_param p
;
439 if (ip_vs_conn_fill_param_proto(af
, skb
, iph
, inverse
, &p
))
442 return ip_vs_conn_out_get(&p
);
444 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto
);
447 * Put back the conn and restart its timer with its timeout
449 void ip_vs_conn_put(struct ip_vs_conn
*cp
)
451 unsigned long t
= (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
) ?
453 mod_timer(&cp
->timer
, jiffies
+t
);
455 __ip_vs_conn_put(cp
);
460 * Fill a no_client_port connection with a client port number
462 void ip_vs_conn_fill_cport(struct ip_vs_conn
*cp
, __be16 cport
)
464 if (ip_vs_conn_unhash(cp
)) {
465 spin_lock_bh(&cp
->lock
);
466 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
) {
467 atomic_dec(&ip_vs_conn_no_cport_cnt
);
468 cp
->flags
&= ~IP_VS_CONN_F_NO_CPORT
;
471 spin_unlock_bh(&cp
->lock
);
473 /* hash on new dport */
480 * Bind a connection entry with the corresponding packet_xmit.
481 * Called by ip_vs_conn_new.
483 static inline void ip_vs_bind_xmit(struct ip_vs_conn
*cp
)
485 switch (IP_VS_FWD_METHOD(cp
)) {
486 case IP_VS_CONN_F_MASQ
:
487 cp
->packet_xmit
= ip_vs_nat_xmit
;
490 case IP_VS_CONN_F_TUNNEL
:
491 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
494 case IP_VS_CONN_F_DROUTE
:
495 cp
->packet_xmit
= ip_vs_dr_xmit
;
498 case IP_VS_CONN_F_LOCALNODE
:
499 cp
->packet_xmit
= ip_vs_null_xmit
;
502 case IP_VS_CONN_F_BYPASS
:
503 cp
->packet_xmit
= ip_vs_bypass_xmit
;
508 #ifdef CONFIG_IP_VS_IPV6
509 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn
*cp
)
511 switch (IP_VS_FWD_METHOD(cp
)) {
512 case IP_VS_CONN_F_MASQ
:
513 cp
->packet_xmit
= ip_vs_nat_xmit_v6
;
516 case IP_VS_CONN_F_TUNNEL
:
517 cp
->packet_xmit
= ip_vs_tunnel_xmit_v6
;
520 case IP_VS_CONN_F_DROUTE
:
521 cp
->packet_xmit
= ip_vs_dr_xmit_v6
;
524 case IP_VS_CONN_F_LOCALNODE
:
525 cp
->packet_xmit
= ip_vs_null_xmit
;
528 case IP_VS_CONN_F_BYPASS
:
529 cp
->packet_xmit
= ip_vs_bypass_xmit_v6
;
536 static inline int ip_vs_dest_totalconns(struct ip_vs_dest
*dest
)
538 return atomic_read(&dest
->activeconns
)
539 + atomic_read(&dest
->inactconns
);
543 * Bind a connection entry with a virtual service destination
544 * Called just after a new connection entry is created.
547 ip_vs_bind_dest(struct ip_vs_conn
*cp
, struct ip_vs_dest
*dest
)
549 unsigned int conn_flags
;
552 /* if dest is NULL, then return directly */
556 /* Increase the refcnt counter of the dest */
557 ip_vs_dest_hold(dest
);
559 conn_flags
= atomic_read(&dest
->conn_flags
);
560 if (cp
->protocol
!= IPPROTO_UDP
)
561 conn_flags
&= ~IP_VS_CONN_F_ONE_PACKET
;
563 /* Bind with the destination and its corresponding transmitter */
564 if (flags
& IP_VS_CONN_F_SYNC
) {
565 /* if the connection is not template and is created
566 * by sync, preserve the activity flag.
568 if (!(flags
& IP_VS_CONN_F_TEMPLATE
))
569 conn_flags
&= ~IP_VS_CONN_F_INACTIVE
;
570 /* connections inherit forwarding method from dest */
571 flags
&= ~(IP_VS_CONN_F_FWD_MASK
| IP_VS_CONN_F_NOOUTPUT
);
577 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
578 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
580 ip_vs_proto_name(cp
->protocol
),
581 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
582 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
583 IP_VS_DBG_ADDR(cp
->af
, &cp
->daddr
), ntohs(cp
->dport
),
584 ip_vs_fwd_tag(cp
), cp
->state
,
585 cp
->flags
, atomic_read(&cp
->refcnt
),
586 atomic_read(&dest
->refcnt
));
588 /* Update the connection counters */
589 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
590 /* It is a normal connection, so modify the counters
591 * according to the flags, later the protocol can
592 * update them on state change
594 if (!(flags
& IP_VS_CONN_F_INACTIVE
))
595 atomic_inc(&dest
->activeconns
);
597 atomic_inc(&dest
->inactconns
);
599 /* It is a persistent connection/template, so increase
600 the persistent connection counter */
601 atomic_inc(&dest
->persistconns
);
604 if (dest
->u_threshold
!= 0 &&
605 ip_vs_dest_totalconns(dest
) >= dest
->u_threshold
)
606 dest
->flags
|= IP_VS_DEST_F_OVERLOAD
;
611 * Check if there is a destination for the connection, if so
612 * bind the connection to the destination.
614 void ip_vs_try_bind_dest(struct ip_vs_conn
*cp
)
616 struct ip_vs_dest
*dest
;
619 dest
= ip_vs_find_dest(ip_vs_conn_net(cp
), cp
->af
, &cp
->daddr
,
620 cp
->dport
, &cp
->vaddr
, cp
->vport
,
621 cp
->protocol
, cp
->fwmark
, cp
->flags
);
623 struct ip_vs_proto_data
*pd
;
625 spin_lock_bh(&cp
->lock
);
627 spin_unlock_bh(&cp
->lock
);
632 /* Applications work depending on the forwarding method
633 * but better to reassign them always when binding dest */
635 ip_vs_unbind_app(cp
);
637 ip_vs_bind_dest(cp
, dest
);
638 spin_unlock_bh(&cp
->lock
);
640 /* Update its packet transmitter */
641 cp
->packet_xmit
= NULL
;
642 #ifdef CONFIG_IP_VS_IPV6
643 if (cp
->af
== AF_INET6
)
644 ip_vs_bind_xmit_v6(cp
);
649 pd
= ip_vs_proto_data_get(ip_vs_conn_net(cp
), cp
->protocol
);
650 if (pd
&& atomic_read(&pd
->appcnt
))
651 ip_vs_bind_app(cp
, pd
->pp
);
658 * Unbind a connection entry with its VS destination
659 * Called by the ip_vs_conn_expire function.
661 static inline void ip_vs_unbind_dest(struct ip_vs_conn
*cp
)
663 struct ip_vs_dest
*dest
= cp
->dest
;
668 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
669 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
671 ip_vs_proto_name(cp
->protocol
),
672 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
673 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
674 IP_VS_DBG_ADDR(cp
->af
, &cp
->daddr
), ntohs(cp
->dport
),
675 ip_vs_fwd_tag(cp
), cp
->state
,
676 cp
->flags
, atomic_read(&cp
->refcnt
),
677 atomic_read(&dest
->refcnt
));
679 /* Update the connection counters */
680 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
681 /* It is a normal connection, so decrease the inactconns
682 or activeconns counter */
683 if (cp
->flags
& IP_VS_CONN_F_INACTIVE
) {
684 atomic_dec(&dest
->inactconns
);
686 atomic_dec(&dest
->activeconns
);
689 /* It is a persistent connection/template, so decrease
690 the persistent connection counter */
691 atomic_dec(&dest
->persistconns
);
694 if (dest
->l_threshold
!= 0) {
695 if (ip_vs_dest_totalconns(dest
) < dest
->l_threshold
)
696 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
697 } else if (dest
->u_threshold
!= 0) {
698 if (ip_vs_dest_totalconns(dest
) * 4 < dest
->u_threshold
* 3)
699 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
701 if (dest
->flags
& IP_VS_DEST_F_OVERLOAD
)
702 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
705 ip_vs_dest_put(dest
);
708 static int expire_quiescent_template(struct netns_ipvs
*ipvs
,
709 struct ip_vs_dest
*dest
)
712 return ipvs
->sysctl_expire_quiescent_template
&&
713 (atomic_read(&dest
->weight
) == 0);
720 * Checking if the destination of a connection template is available.
721 * If available, return 1, otherwise invalidate this connection
722 * template and return 0.
724 int ip_vs_check_template(struct ip_vs_conn
*ct
)
726 struct ip_vs_dest
*dest
= ct
->dest
;
727 struct netns_ipvs
*ipvs
= net_ipvs(ip_vs_conn_net(ct
));
730 * Checking the dest server status.
732 if ((dest
== NULL
) ||
733 !(dest
->flags
& IP_VS_DEST_F_AVAILABLE
) ||
734 expire_quiescent_template(ipvs
, dest
)) {
735 IP_VS_DBG_BUF(9, "check_template: dest not available for "
736 "protocol %s s:%s:%d v:%s:%d "
738 ip_vs_proto_name(ct
->protocol
),
739 IP_VS_DBG_ADDR(ct
->af
, &ct
->caddr
),
741 IP_VS_DBG_ADDR(ct
->af
, &ct
->vaddr
),
743 IP_VS_DBG_ADDR(ct
->af
, &ct
->daddr
),
747 * Invalidate the connection template
749 if (ct
->vport
!= htons(0xffff)) {
750 if (ip_vs_conn_unhash(ct
)) {
751 ct
->dport
= htons(0xffff);
752 ct
->vport
= htons(0xffff);
759 * Simply decrease the refcnt of the template,
760 * don't restart its timer.
762 __ip_vs_conn_put(ct
);
768 static void ip_vs_conn_rcu_free(struct rcu_head
*head
)
770 struct ip_vs_conn
*cp
= container_of(head
, struct ip_vs_conn
,
773 ip_vs_pe_put(cp
->pe
);
775 kmem_cache_free(ip_vs_conn_cachep
, cp
);
778 static void ip_vs_conn_expire(unsigned long data
)
780 struct ip_vs_conn
*cp
= (struct ip_vs_conn
*)data
;
781 struct net
*net
= ip_vs_conn_net(cp
);
782 struct netns_ipvs
*ipvs
= net_ipvs(net
);
785 * do I control anybody?
787 if (atomic_read(&cp
->n_control
))
790 /* Unlink conn if not referenced anymore */
791 if (likely(ip_vs_conn_unlink(cp
))) {
792 /* delete the timer if it is activated by other users */
793 del_timer(&cp
->timer
);
795 /* does anybody control me? */
797 ip_vs_control_del(cp
);
799 if (cp
->flags
& IP_VS_CONN_F_NFCT
) {
800 ip_vs_conn_drop_conntrack(cp
);
801 /* Do not access conntracks during subsys cleanup
802 * because nf_conntrack_find_get can not be used after
803 * conntrack cleanup for the net.
807 ip_vs_conn_drop_conntrack(cp
);
810 if (unlikely(cp
->app
!= NULL
))
811 ip_vs_unbind_app(cp
);
812 ip_vs_unbind_dest(cp
);
813 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
)
814 atomic_dec(&ip_vs_conn_no_cport_cnt
);
815 call_rcu(&cp
->rcu_head
, ip_vs_conn_rcu_free
);
816 atomic_dec(&ipvs
->conn_count
);
821 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
822 atomic_read(&cp
->refcnt
),
823 atomic_read(&cp
->n_control
));
825 atomic_inc(&cp
->refcnt
);
828 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
829 ip_vs_sync_conn(net
, cp
, sysctl_sync_threshold(ipvs
));
834 /* Modify timer, so that it expires as soon as possible.
835 * Can be called without reference only if under RCU lock.
837 void ip_vs_conn_expire_now(struct ip_vs_conn
*cp
)
839 /* Using mod_timer_pending will ensure the timer is not
840 * modified after the final del_timer in ip_vs_conn_expire.
842 if (timer_pending(&cp
->timer
) &&
843 time_after(cp
->timer
.expires
, jiffies
))
844 mod_timer_pending(&cp
->timer
, jiffies
);
849 * Create a new connection entry and hash it into the ip_vs_conn_tab
852 ip_vs_conn_new(const struct ip_vs_conn_param
*p
,
853 const union nf_inet_addr
*daddr
, __be16 dport
, unsigned int flags
,
854 struct ip_vs_dest
*dest
, __u32 fwmark
)
856 struct ip_vs_conn
*cp
;
857 struct netns_ipvs
*ipvs
= net_ipvs(p
->net
);
858 struct ip_vs_proto_data
*pd
= ip_vs_proto_data_get(p
->net
,
861 cp
= kmem_cache_alloc(ip_vs_conn_cachep
, GFP_ATOMIC
);
863 IP_VS_ERR_RL("%s(): no memory\n", __func__
);
867 INIT_HLIST_NODE(&cp
->c_list
);
868 setup_timer(&cp
->timer
, ip_vs_conn_expire
, (unsigned long)cp
);
869 ip_vs_conn_net_set(cp
, p
->net
);
871 cp
->protocol
= p
->protocol
;
872 ip_vs_addr_set(p
->af
, &cp
->caddr
, p
->caddr
);
873 cp
->cport
= p
->cport
;
874 ip_vs_addr_set(p
->af
, &cp
->vaddr
, p
->vaddr
);
875 cp
->vport
= p
->vport
;
876 /* proto should only be IPPROTO_IP if d_addr is a fwmark */
877 ip_vs_addr_set(p
->protocol
== IPPROTO_IP
? AF_UNSPEC
: p
->af
,
882 if (flags
& IP_VS_CONN_F_TEMPLATE
&& p
->pe
) {
885 cp
->pe_data
= p
->pe_data
;
886 cp
->pe_data_len
= p
->pe_data_len
;
892 spin_lock_init(&cp
->lock
);
895 * Set the entry is referenced by the current thread before hashing
896 * it in the table, so that other thread run ip_vs_random_dropentry
897 * but cannot drop this entry.
899 atomic_set(&cp
->refcnt
, 1);
902 atomic_set(&cp
->n_control
, 0);
903 atomic_set(&cp
->in_pkts
, 0);
905 cp
->packet_xmit
= NULL
;
908 /* reset struct ip_vs_seq */
909 cp
->in_seq
.delta
= 0;
910 cp
->out_seq
.delta
= 0;
912 atomic_inc(&ipvs
->conn_count
);
913 if (flags
& IP_VS_CONN_F_NO_CPORT
)
914 atomic_inc(&ip_vs_conn_no_cport_cnt
);
916 /* Bind the connection with a destination server */
918 ip_vs_bind_dest(cp
, dest
);
920 /* Set its state and timeout */
924 cp
->sync_endtime
= jiffies
& ~3UL;
926 /* Bind its packet transmitter */
927 #ifdef CONFIG_IP_VS_IPV6
928 if (p
->af
== AF_INET6
)
929 ip_vs_bind_xmit_v6(cp
);
934 if (unlikely(pd
&& atomic_read(&pd
->appcnt
)))
935 ip_vs_bind_app(cp
, pd
->pp
);
938 * Allow conntrack to be preserved. By default, conntrack
939 * is created and destroyed for every packet.
940 * Sometimes keeping conntrack can be useful for
941 * IP_VS_CONN_F_ONE_PACKET too.
944 if (ip_vs_conntrack_enabled(ipvs
))
945 cp
->flags
|= IP_VS_CONN_F_NFCT
;
947 /* Hash it in the ip_vs_conn_tab finally */
954 * /proc/net/ip_vs_conn entries
956 #ifdef CONFIG_PROC_FS
957 struct ip_vs_iter_state
{
958 struct seq_net_private p
;
959 struct hlist_head
*l
;
962 static void *ip_vs_conn_array(struct seq_file
*seq
, loff_t pos
)
965 struct ip_vs_conn
*cp
;
966 struct ip_vs_iter_state
*iter
= seq
->private;
968 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++) {
970 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
971 /* __ip_vs_conn_get() is not needed by
972 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
975 iter
->l
= &ip_vs_conn_tab
[idx
];
985 static void *ip_vs_conn_seq_start(struct seq_file
*seq
, loff_t
*pos
)
987 struct ip_vs_iter_state
*iter
= seq
->private;
990 return *pos
? ip_vs_conn_array(seq
, *pos
- 1) :SEQ_START_TOKEN
;
993 static void *ip_vs_conn_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
995 struct ip_vs_conn
*cp
= v
;
996 struct ip_vs_iter_state
*iter
= seq
->private;
997 struct hlist_node
*e
;
998 struct hlist_head
*l
= iter
->l
;
1002 if (v
== SEQ_START_TOKEN
)
1003 return ip_vs_conn_array(seq
, 0);
1005 /* more on same hash chain? */
1006 e
= rcu_dereference(hlist_next_rcu(&cp
->c_list
));
1008 return hlist_entry(e
, struct ip_vs_conn
, c_list
);
1011 idx
= l
- ip_vs_conn_tab
;
1012 while (++idx
< ip_vs_conn_tab_size
) {
1014 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1015 iter
->l
= &ip_vs_conn_tab
[idx
];
1024 static void ip_vs_conn_seq_stop(struct seq_file
*seq
, void *v
)
1026 struct ip_vs_iter_state
*iter
= seq
->private;
1027 struct hlist_head
*l
= iter
->l
;
1033 static int ip_vs_conn_seq_show(struct seq_file
*seq
, void *v
)
1036 if (v
== SEQ_START_TOKEN
)
1038 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1040 const struct ip_vs_conn
*cp
= v
;
1041 struct net
*net
= seq_file_net(seq
);
1042 char pe_data
[IP_VS_PENAME_MAXLEN
+ IP_VS_PEDATA_MAXLEN
+ 3];
1045 if (!ip_vs_conn_net_eq(cp
, net
))
1049 len
= strlen(cp
->pe
->name
);
1050 memcpy(pe_data
+ 1, cp
->pe
->name
, len
);
1051 pe_data
[len
+ 1] = ' ';
1053 len
+= cp
->pe
->show_pe_data(cp
, pe_data
+ len
);
1055 pe_data
[len
] = '\0';
1057 #ifdef CONFIG_IP_VS_IPV6
1058 if (cp
->af
== AF_INET6
)
1059 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X "
1060 "%pI6 %04X %-11s %7lu%s\n",
1061 ip_vs_proto_name(cp
->protocol
),
1062 &cp
->caddr
.in6
, ntohs(cp
->cport
),
1063 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
1064 &cp
->daddr
.in6
, ntohs(cp
->dport
),
1065 ip_vs_state_name(cp
->protocol
, cp
->state
),
1066 (cp
->timer
.expires
-jiffies
)/HZ
, pe_data
);
1070 "%-3s %08X %04X %08X %04X"
1071 " %08X %04X %-11s %7lu%s\n",
1072 ip_vs_proto_name(cp
->protocol
),
1073 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
1074 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
1075 ntohl(cp
->daddr
.ip
), ntohs(cp
->dport
),
1076 ip_vs_state_name(cp
->protocol
, cp
->state
),
1077 (cp
->timer
.expires
-jiffies
)/HZ
, pe_data
);
1082 static const struct seq_operations ip_vs_conn_seq_ops
= {
1083 .start
= ip_vs_conn_seq_start
,
1084 .next
= ip_vs_conn_seq_next
,
1085 .stop
= ip_vs_conn_seq_stop
,
1086 .show
= ip_vs_conn_seq_show
,
1089 static int ip_vs_conn_open(struct inode
*inode
, struct file
*file
)
1091 return seq_open_net(inode
, file
, &ip_vs_conn_seq_ops
,
1092 sizeof(struct ip_vs_iter_state
));
1095 static const struct file_operations ip_vs_conn_fops
= {
1096 .owner
= THIS_MODULE
,
1097 .open
= ip_vs_conn_open
,
1099 .llseek
= seq_lseek
,
1100 .release
= seq_release_net
,
1103 static const char *ip_vs_origin_name(unsigned int flags
)
1105 if (flags
& IP_VS_CONN_F_SYNC
)
1111 static int ip_vs_conn_sync_seq_show(struct seq_file
*seq
, void *v
)
1114 if (v
== SEQ_START_TOKEN
)
1116 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1118 const struct ip_vs_conn
*cp
= v
;
1119 struct net
*net
= seq_file_net(seq
);
1121 if (!ip_vs_conn_net_eq(cp
, net
))
1124 #ifdef CONFIG_IP_VS_IPV6
1125 if (cp
->af
== AF_INET6
)
1126 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1127 ip_vs_proto_name(cp
->protocol
),
1128 &cp
->caddr
.in6
, ntohs(cp
->cport
),
1129 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
1130 &cp
->daddr
.in6
, ntohs(cp
->dport
),
1131 ip_vs_state_name(cp
->protocol
, cp
->state
),
1132 ip_vs_origin_name(cp
->flags
),
1133 (cp
->timer
.expires
-jiffies
)/HZ
);
1137 "%-3s %08X %04X %08X %04X "
1138 "%08X %04X %-11s %-6s %7lu\n",
1139 ip_vs_proto_name(cp
->protocol
),
1140 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
1141 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
1142 ntohl(cp
->daddr
.ip
), ntohs(cp
->dport
),
1143 ip_vs_state_name(cp
->protocol
, cp
->state
),
1144 ip_vs_origin_name(cp
->flags
),
1145 (cp
->timer
.expires
-jiffies
)/HZ
);
1150 static const struct seq_operations ip_vs_conn_sync_seq_ops
= {
1151 .start
= ip_vs_conn_seq_start
,
1152 .next
= ip_vs_conn_seq_next
,
1153 .stop
= ip_vs_conn_seq_stop
,
1154 .show
= ip_vs_conn_sync_seq_show
,
1157 static int ip_vs_conn_sync_open(struct inode
*inode
, struct file
*file
)
1159 return seq_open_net(inode
, file
, &ip_vs_conn_sync_seq_ops
,
1160 sizeof(struct ip_vs_iter_state
));
1163 static const struct file_operations ip_vs_conn_sync_fops
= {
1164 .owner
= THIS_MODULE
,
1165 .open
= ip_vs_conn_sync_open
,
1167 .llseek
= seq_lseek
,
1168 .release
= seq_release_net
,
1175 * Randomly drop connection entries before running out of memory
1177 static inline int todrop_entry(struct ip_vs_conn
*cp
)
1180 * The drop rate array needs tuning for real environments.
1181 * Called from timer bh only => no locking
1183 static const char todrop_rate
[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1184 static char todrop_counter
[9] = {0};
1187 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1188 This will leave enough time for normal connection to get
1190 if (time_before(cp
->timeout
+ jiffies
, cp
->timer
.expires
+ 60*HZ
))
1193 /* Don't drop the entry if its number of incoming packets is not
1194 located in [0, 8] */
1195 i
= atomic_read(&cp
->in_pkts
);
1196 if (i
> 8 || i
< 0) return 0;
1198 if (!todrop_rate
[i
]) return 0;
1199 if (--todrop_counter
[i
] > 0) return 0;
1201 todrop_counter
[i
] = todrop_rate
[i
];
1205 /* Called from keventd and must protect itself from softirqs */
1206 void ip_vs_random_dropentry(struct net
*net
)
1209 struct ip_vs_conn
*cp
, *cp_c
;
1212 * Randomly scan 1/32 of the whole table every second
1214 for (idx
= 0; idx
< (ip_vs_conn_tab_size
>>5); idx
++) {
1215 unsigned int hash
= net_random() & ip_vs_conn_tab_mask
;
1218 * Lock is actually needed in this loop.
1222 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
1223 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
)
1224 /* connection template */
1226 if (!ip_vs_conn_net_eq(cp
, net
))
1228 if (cp
->protocol
== IPPROTO_TCP
) {
1230 case IP_VS_TCP_S_SYN_RECV
:
1231 case IP_VS_TCP_S_SYNACK
:
1234 case IP_VS_TCP_S_ESTABLISHED
:
1235 if (todrop_entry(cp
))
1243 if (!todrop_entry(cp
))
1247 IP_VS_DBG(4, "del connection\n");
1248 ip_vs_conn_expire_now(cp
);
1250 /* cp->control is valid only with reference to cp */
1251 if (cp_c
&& __ip_vs_conn_get(cp
)) {
1252 IP_VS_DBG(4, "del conn template\n");
1253 ip_vs_conn_expire_now(cp_c
);
1254 __ip_vs_conn_put(cp
);
1263 * Flush all the connection entries in the ip_vs_conn_tab
1265 static void ip_vs_conn_flush(struct net
*net
)
1268 struct ip_vs_conn
*cp
, *cp_c
;
1269 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1272 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++) {
1274 * Lock is actually needed in this loop.
1278 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1279 if (!ip_vs_conn_net_eq(cp
, net
))
1281 IP_VS_DBG(4, "del connection\n");
1282 ip_vs_conn_expire_now(cp
);
1284 /* cp->control is valid only with reference to cp */
1285 if (cp_c
&& __ip_vs_conn_get(cp
)) {
1286 IP_VS_DBG(4, "del conn template\n");
1287 ip_vs_conn_expire_now(cp_c
);
1288 __ip_vs_conn_put(cp
);
1294 /* the counter may be not NULL, because maybe some conn entries
1295 are run by slow timer handler or unhashed but still referred */
1296 if (atomic_read(&ipvs
->conn_count
) != 0) {
1302 * per netns init and exit
1304 int __net_init
ip_vs_conn_net_init(struct net
*net
)
1306 struct netns_ipvs
*ipvs
= net_ipvs(net
);
1308 atomic_set(&ipvs
->conn_count
, 0);
1310 proc_create("ip_vs_conn", 0, net
->proc_net
, &ip_vs_conn_fops
);
1311 proc_create("ip_vs_conn_sync", 0, net
->proc_net
, &ip_vs_conn_sync_fops
);
1315 void __net_exit
ip_vs_conn_net_cleanup(struct net
*net
)
1317 /* flush all the connection entries first */
1318 ip_vs_conn_flush(net
);
1319 remove_proc_entry("ip_vs_conn", net
->proc_net
);
1320 remove_proc_entry("ip_vs_conn_sync", net
->proc_net
);
1323 int __init
ip_vs_conn_init(void)
1327 /* Compute size and mask */
1328 ip_vs_conn_tab_size
= 1 << ip_vs_conn_tab_bits
;
1329 ip_vs_conn_tab_mask
= ip_vs_conn_tab_size
- 1;
1332 * Allocate the connection hash table and initialize its list heads
1334 ip_vs_conn_tab
= vmalloc(ip_vs_conn_tab_size
* sizeof(*ip_vs_conn_tab
));
1335 if (!ip_vs_conn_tab
)
1338 /* Allocate ip_vs_conn slab cache */
1339 ip_vs_conn_cachep
= kmem_cache_create("ip_vs_conn",
1340 sizeof(struct ip_vs_conn
), 0,
1341 SLAB_HWCACHE_ALIGN
, NULL
);
1342 if (!ip_vs_conn_cachep
) {
1343 vfree(ip_vs_conn_tab
);
1347 pr_info("Connection hash table configured "
1348 "(size=%d, memory=%ldKbytes)\n",
1349 ip_vs_conn_tab_size
,
1350 (long)(ip_vs_conn_tab_size
*sizeof(struct list_head
))/1024);
1351 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1352 sizeof(struct ip_vs_conn
));
1354 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++)
1355 INIT_HLIST_HEAD(&ip_vs_conn_tab
[idx
]);
1357 for (idx
= 0; idx
< CT_LOCKARRAY_SIZE
; idx
++) {
1358 spin_lock_init(&__ip_vs_conntbl_lock_array
[idx
].l
);
1361 /* calculate the random value for connection hash */
1362 get_random_bytes(&ip_vs_conn_rnd
, sizeof(ip_vs_conn_rnd
));
1367 void ip_vs_conn_cleanup(void)
1369 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1371 /* Release the empty cache */
1372 kmem_cache_destroy(ip_vs_conn_cachep
);
1373 vfree(ip_vs_conn_tab
);