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
);
107 static void ip_vs_conn_expire(unsigned long data
);
110 * Returns hash value for IPVS connection entry
112 static unsigned int ip_vs_conn_hashkey(struct netns_ipvs
*ipvs
, int af
, unsigned int proto
,
113 const union nf_inet_addr
*addr
,
116 #ifdef CONFIG_IP_VS_IPV6
118 return (jhash_3words(jhash(addr
, 16, ip_vs_conn_rnd
),
119 (__force u32
)port
, proto
, ip_vs_conn_rnd
) ^
120 ((size_t)ipvs
>>8)) & ip_vs_conn_tab_mask
;
122 return (jhash_3words((__force u32
)addr
->ip
, (__force u32
)port
, proto
,
124 ((size_t)ipvs
>>8)) & ip_vs_conn_tab_mask
;
127 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param
*p
,
130 const union nf_inet_addr
*addr
;
133 if (p
->pe_data
&& p
->pe
->hashkey_raw
)
134 return p
->pe
->hashkey_raw(p
, ip_vs_conn_rnd
, inverse
) &
137 if (likely(!inverse
)) {
145 return ip_vs_conn_hashkey(p
->ipvs
, p
->af
, p
->protocol
, addr
, port
);
148 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn
*cp
)
150 struct ip_vs_conn_param p
;
152 ip_vs_conn_fill_param(cp
->ipvs
, cp
->af
, cp
->protocol
,
153 &cp
->caddr
, cp
->cport
, NULL
, 0, &p
);
157 p
.pe_data
= cp
->pe_data
;
158 p
.pe_data_len
= cp
->pe_data_len
;
161 return ip_vs_conn_hashkey_param(&p
, false);
165 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
166 * returns bool success.
168 static inline int ip_vs_conn_hash(struct ip_vs_conn
*cp
)
173 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
176 /* Hash by protocol, client address and port */
177 hash
= ip_vs_conn_hashkey_conn(cp
);
179 ct_write_lock_bh(hash
);
180 spin_lock(&cp
->lock
);
182 if (!(cp
->flags
& IP_VS_CONN_F_HASHED
)) {
183 cp
->flags
|= IP_VS_CONN_F_HASHED
;
184 atomic_inc(&cp
->refcnt
);
185 hlist_add_head_rcu(&cp
->c_list
, &ip_vs_conn_tab
[hash
]);
188 pr_err("%s(): request for already hashed, called from %pF\n",
189 __func__
, __builtin_return_address(0));
193 spin_unlock(&cp
->lock
);
194 ct_write_unlock_bh(hash
);
201 * UNhashes ip_vs_conn from ip_vs_conn_tab.
202 * returns bool success. Caller should hold conn reference.
204 static inline int ip_vs_conn_unhash(struct ip_vs_conn
*cp
)
209 /* unhash it and decrease its reference counter */
210 hash
= ip_vs_conn_hashkey_conn(cp
);
212 ct_write_lock_bh(hash
);
213 spin_lock(&cp
->lock
);
215 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
216 hlist_del_rcu(&cp
->c_list
);
217 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
218 atomic_dec(&cp
->refcnt
);
223 spin_unlock(&cp
->lock
);
224 ct_write_unlock_bh(hash
);
229 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
230 * returns bool success.
232 static inline bool ip_vs_conn_unlink(struct ip_vs_conn
*cp
)
237 hash
= ip_vs_conn_hashkey_conn(cp
);
239 ct_write_lock_bh(hash
);
240 spin_lock(&cp
->lock
);
242 if (cp
->flags
& IP_VS_CONN_F_HASHED
) {
244 /* Decrease refcnt and unlink conn only if we are last user */
245 if (atomic_cmpxchg(&cp
->refcnt
, 1, 0) == 1) {
246 hlist_del_rcu(&cp
->c_list
);
247 cp
->flags
&= ~IP_VS_CONN_F_HASHED
;
251 ret
= atomic_read(&cp
->refcnt
) ? false : true;
253 spin_unlock(&cp
->lock
);
254 ct_write_unlock_bh(hash
);
261 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
262 * Called for pkts coming from OUTside-to-INside.
263 * p->caddr, p->cport: pkt source address (foreign host)
264 * p->vaddr, p->vport: pkt dest address (load balancer)
266 static inline struct ip_vs_conn
*
267 __ip_vs_conn_in_get(const struct ip_vs_conn_param
*p
)
270 struct ip_vs_conn
*cp
;
272 hash
= ip_vs_conn_hashkey_param(p
, false);
276 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
277 if (p
->cport
== cp
->cport
&& p
->vport
== cp
->vport
&&
279 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->caddr
) &&
280 ip_vs_addr_equal(p
->af
, p
->vaddr
, &cp
->vaddr
) &&
281 ((!p
->cport
) ^ (!(cp
->flags
& IP_VS_CONN_F_NO_CPORT
))) &&
282 p
->protocol
== cp
->protocol
&&
283 cp
->ipvs
== p
->ipvs
) {
284 if (!__ip_vs_conn_get(cp
))
297 struct ip_vs_conn
*ip_vs_conn_in_get(const struct ip_vs_conn_param
*p
)
299 struct ip_vs_conn
*cp
;
301 cp
= __ip_vs_conn_in_get(p
);
302 if (!cp
&& atomic_read(&ip_vs_conn_no_cport_cnt
)) {
303 struct ip_vs_conn_param cport_zero_p
= *p
;
304 cport_zero_p
.cport
= 0;
305 cp
= __ip_vs_conn_in_get(&cport_zero_p
);
308 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
309 ip_vs_proto_name(p
->protocol
),
310 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
311 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
312 cp
? "hit" : "not hit");
318 ip_vs_conn_fill_param_proto(struct netns_ipvs
*ipvs
,
319 int af
, const struct sk_buff
*skb
,
320 const struct ip_vs_iphdr
*iph
,
321 struct ip_vs_conn_param
*p
)
323 __be16 _ports
[2], *pptr
;
325 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
329 if (likely(!ip_vs_iph_inverse(iph
)))
330 ip_vs_conn_fill_param(ipvs
, af
, iph
->protocol
, &iph
->saddr
,
331 pptr
[0], &iph
->daddr
, pptr
[1], p
);
333 ip_vs_conn_fill_param(ipvs
, af
, iph
->protocol
, &iph
->daddr
,
334 pptr
[1], &iph
->saddr
, pptr
[0], p
);
339 ip_vs_conn_in_get_proto(struct netns_ipvs
*ipvs
, int af
,
340 const struct sk_buff
*skb
,
341 const struct ip_vs_iphdr
*iph
)
343 struct ip_vs_conn_param p
;
345 if (ip_vs_conn_fill_param_proto(ipvs
, af
, skb
, iph
, &p
))
348 return ip_vs_conn_in_get(&p
);
350 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto
);
352 /* Get reference to connection template */
353 struct ip_vs_conn
*ip_vs_ct_in_get(const struct ip_vs_conn_param
*p
)
356 struct ip_vs_conn
*cp
;
358 hash
= ip_vs_conn_hashkey_param(p
, false);
362 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
363 if (unlikely(p
->pe_data
&& p
->pe
->ct_match
)) {
364 if (cp
->ipvs
!= p
->ipvs
)
366 if (p
->pe
== cp
->pe
&& p
->pe
->ct_match(p
, cp
)) {
367 if (__ip_vs_conn_get(cp
))
373 if (cp
->af
== p
->af
&&
374 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->caddr
) &&
375 /* protocol should only be IPPROTO_IP if
376 * p->vaddr is a fwmark */
377 ip_vs_addr_equal(p
->protocol
== IPPROTO_IP
? AF_UNSPEC
:
378 p
->af
, p
->vaddr
, &cp
->vaddr
) &&
379 p
->vport
== cp
->vport
&& p
->cport
== cp
->cport
&&
380 cp
->flags
& IP_VS_CONN_F_TEMPLATE
&&
381 p
->protocol
== cp
->protocol
&&
382 cp
->ipvs
== p
->ipvs
) {
383 if (__ip_vs_conn_get(cp
))
392 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
393 ip_vs_proto_name(p
->protocol
),
394 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
395 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
396 cp
? "hit" : "not hit");
401 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
402 * Called for pkts coming from inside-to-OUTside.
403 * p->caddr, p->cport: pkt source address (inside host)
404 * p->vaddr, p->vport: pkt dest address (foreign host) */
405 struct ip_vs_conn
*ip_vs_conn_out_get(const struct ip_vs_conn_param
*p
)
408 struct ip_vs_conn
*cp
, *ret
=NULL
;
411 * Check for "full" addressed entries
413 hash
= ip_vs_conn_hashkey_param(p
, true);
417 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
418 if (p
->vport
== cp
->cport
&& p
->cport
== cp
->dport
&&
420 ip_vs_addr_equal(p
->af
, p
->vaddr
, &cp
->caddr
) &&
421 ip_vs_addr_equal(p
->af
, p
->caddr
, &cp
->daddr
) &&
422 p
->protocol
== cp
->protocol
&&
423 cp
->ipvs
== p
->ipvs
) {
424 if (!__ip_vs_conn_get(cp
))
434 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
435 ip_vs_proto_name(p
->protocol
),
436 IP_VS_DBG_ADDR(p
->af
, p
->caddr
), ntohs(p
->cport
),
437 IP_VS_DBG_ADDR(p
->af
, p
->vaddr
), ntohs(p
->vport
),
438 ret
? "hit" : "not hit");
444 ip_vs_conn_out_get_proto(struct netns_ipvs
*ipvs
, int af
,
445 const struct sk_buff
*skb
,
446 const struct ip_vs_iphdr
*iph
)
448 struct ip_vs_conn_param p
;
450 if (ip_vs_conn_fill_param_proto(ipvs
, af
, skb
, iph
, &p
))
453 return ip_vs_conn_out_get(&p
);
455 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto
);
457 static void __ip_vs_conn_put_notimer(struct ip_vs_conn
*cp
)
459 __ip_vs_conn_put(cp
);
460 ip_vs_conn_expire((unsigned long)cp
);
464 * Put back the conn and restart its timer with its timeout
466 static void __ip_vs_conn_put_timer(struct ip_vs_conn
*cp
)
468 unsigned long t
= (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
) ?
470 mod_timer(&cp
->timer
, jiffies
+t
);
472 __ip_vs_conn_put(cp
);
475 void ip_vs_conn_put(struct ip_vs_conn
*cp
)
477 if ((cp
->flags
& IP_VS_CONN_F_ONE_PACKET
) &&
478 (atomic_read(&cp
->refcnt
) == 1) &&
479 !timer_pending(&cp
->timer
))
480 /* expire connection immediately */
481 __ip_vs_conn_put_notimer(cp
);
483 __ip_vs_conn_put_timer(cp
);
487 * Fill a no_client_port connection with a client port number
489 void ip_vs_conn_fill_cport(struct ip_vs_conn
*cp
, __be16 cport
)
491 if (ip_vs_conn_unhash(cp
)) {
492 spin_lock_bh(&cp
->lock
);
493 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
) {
494 atomic_dec(&ip_vs_conn_no_cport_cnt
);
495 cp
->flags
&= ~IP_VS_CONN_F_NO_CPORT
;
498 spin_unlock_bh(&cp
->lock
);
500 /* hash on new dport */
507 * Bind a connection entry with the corresponding packet_xmit.
508 * Called by ip_vs_conn_new.
510 static inline void ip_vs_bind_xmit(struct ip_vs_conn
*cp
)
512 switch (IP_VS_FWD_METHOD(cp
)) {
513 case IP_VS_CONN_F_MASQ
:
514 cp
->packet_xmit
= ip_vs_nat_xmit
;
517 case IP_VS_CONN_F_TUNNEL
:
518 #ifdef CONFIG_IP_VS_IPV6
519 if (cp
->daf
== AF_INET6
)
520 cp
->packet_xmit
= ip_vs_tunnel_xmit_v6
;
523 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
526 case IP_VS_CONN_F_DROUTE
:
527 cp
->packet_xmit
= ip_vs_dr_xmit
;
530 case IP_VS_CONN_F_LOCALNODE
:
531 cp
->packet_xmit
= ip_vs_null_xmit
;
534 case IP_VS_CONN_F_BYPASS
:
535 cp
->packet_xmit
= ip_vs_bypass_xmit
;
540 #ifdef CONFIG_IP_VS_IPV6
541 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn
*cp
)
543 switch (IP_VS_FWD_METHOD(cp
)) {
544 case IP_VS_CONN_F_MASQ
:
545 cp
->packet_xmit
= ip_vs_nat_xmit_v6
;
548 case IP_VS_CONN_F_TUNNEL
:
549 if (cp
->daf
== AF_INET6
)
550 cp
->packet_xmit
= ip_vs_tunnel_xmit_v6
;
552 cp
->packet_xmit
= ip_vs_tunnel_xmit
;
555 case IP_VS_CONN_F_DROUTE
:
556 cp
->packet_xmit
= ip_vs_dr_xmit_v6
;
559 case IP_VS_CONN_F_LOCALNODE
:
560 cp
->packet_xmit
= ip_vs_null_xmit
;
563 case IP_VS_CONN_F_BYPASS
:
564 cp
->packet_xmit
= ip_vs_bypass_xmit_v6
;
571 static inline int ip_vs_dest_totalconns(struct ip_vs_dest
*dest
)
573 return atomic_read(&dest
->activeconns
)
574 + atomic_read(&dest
->inactconns
);
578 * Bind a connection entry with a virtual service destination
579 * Called just after a new connection entry is created.
582 ip_vs_bind_dest(struct ip_vs_conn
*cp
, struct ip_vs_dest
*dest
)
584 unsigned int conn_flags
;
587 /* if dest is NULL, then return directly */
591 /* Increase the refcnt counter of the dest */
592 ip_vs_dest_hold(dest
);
594 conn_flags
= atomic_read(&dest
->conn_flags
);
595 if (cp
->protocol
!= IPPROTO_UDP
)
596 conn_flags
&= ~IP_VS_CONN_F_ONE_PACKET
;
598 /* Bind with the destination and its corresponding transmitter */
599 if (flags
& IP_VS_CONN_F_SYNC
) {
600 /* if the connection is not template and is created
601 * by sync, preserve the activity flag.
603 if (!(flags
& IP_VS_CONN_F_TEMPLATE
))
604 conn_flags
&= ~IP_VS_CONN_F_INACTIVE
;
605 /* connections inherit forwarding method from dest */
606 flags
&= ~(IP_VS_CONN_F_FWD_MASK
| IP_VS_CONN_F_NOOUTPUT
);
612 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
613 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
615 ip_vs_proto_name(cp
->protocol
),
616 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
617 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
618 IP_VS_DBG_ADDR(cp
->daf
, &cp
->daddr
), ntohs(cp
->dport
),
619 ip_vs_fwd_tag(cp
), cp
->state
,
620 cp
->flags
, atomic_read(&cp
->refcnt
),
621 atomic_read(&dest
->refcnt
));
623 /* Update the connection counters */
624 if (!(flags
& IP_VS_CONN_F_TEMPLATE
)) {
625 /* It is a normal connection, so modify the counters
626 * according to the flags, later the protocol can
627 * update them on state change
629 if (!(flags
& IP_VS_CONN_F_INACTIVE
))
630 atomic_inc(&dest
->activeconns
);
632 atomic_inc(&dest
->inactconns
);
634 /* It is a persistent connection/template, so increase
635 the persistent connection counter */
636 atomic_inc(&dest
->persistconns
);
639 if (dest
->u_threshold
!= 0 &&
640 ip_vs_dest_totalconns(dest
) >= dest
->u_threshold
)
641 dest
->flags
|= IP_VS_DEST_F_OVERLOAD
;
646 * Check if there is a destination for the connection, if so
647 * bind the connection to the destination.
649 void ip_vs_try_bind_dest(struct ip_vs_conn
*cp
)
651 struct ip_vs_dest
*dest
;
655 /* This function is only invoked by the synchronization code. We do
656 * not currently support heterogeneous pools with synchronization,
657 * so we can make the assumption that the svc_af is the same as the
660 dest
= ip_vs_find_dest(cp
->ipvs
, cp
->af
, cp
->af
, &cp
->daddr
,
661 cp
->dport
, &cp
->vaddr
, cp
->vport
,
662 cp
->protocol
, cp
->fwmark
, cp
->flags
);
664 struct ip_vs_proto_data
*pd
;
666 spin_lock_bh(&cp
->lock
);
668 spin_unlock_bh(&cp
->lock
);
673 /* Applications work depending on the forwarding method
674 * but better to reassign them always when binding dest */
676 ip_vs_unbind_app(cp
);
678 ip_vs_bind_dest(cp
, dest
);
679 spin_unlock_bh(&cp
->lock
);
681 /* Update its packet transmitter */
682 cp
->packet_xmit
= NULL
;
683 #ifdef CONFIG_IP_VS_IPV6
684 if (cp
->af
== AF_INET6
)
685 ip_vs_bind_xmit_v6(cp
);
690 pd
= ip_vs_proto_data_get(cp
->ipvs
, cp
->protocol
);
691 if (pd
&& atomic_read(&pd
->appcnt
))
692 ip_vs_bind_app(cp
, pd
->pp
);
699 * Unbind a connection entry with its VS destination
700 * Called by the ip_vs_conn_expire function.
702 static inline void ip_vs_unbind_dest(struct ip_vs_conn
*cp
)
704 struct ip_vs_dest
*dest
= cp
->dest
;
709 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
710 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
712 ip_vs_proto_name(cp
->protocol
),
713 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
714 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
715 IP_VS_DBG_ADDR(cp
->daf
, &cp
->daddr
), ntohs(cp
->dport
),
716 ip_vs_fwd_tag(cp
), cp
->state
,
717 cp
->flags
, atomic_read(&cp
->refcnt
),
718 atomic_read(&dest
->refcnt
));
720 /* Update the connection counters */
721 if (!(cp
->flags
& IP_VS_CONN_F_TEMPLATE
)) {
722 /* It is a normal connection, so decrease the inactconns
723 or activeconns counter */
724 if (cp
->flags
& IP_VS_CONN_F_INACTIVE
) {
725 atomic_dec(&dest
->inactconns
);
727 atomic_dec(&dest
->activeconns
);
730 /* It is a persistent connection/template, so decrease
731 the persistent connection counter */
732 atomic_dec(&dest
->persistconns
);
735 if (dest
->l_threshold
!= 0) {
736 if (ip_vs_dest_totalconns(dest
) < dest
->l_threshold
)
737 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
738 } else if (dest
->u_threshold
!= 0) {
739 if (ip_vs_dest_totalconns(dest
) * 4 < dest
->u_threshold
* 3)
740 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
742 if (dest
->flags
& IP_VS_DEST_F_OVERLOAD
)
743 dest
->flags
&= ~IP_VS_DEST_F_OVERLOAD
;
746 ip_vs_dest_put(dest
);
749 static int expire_quiescent_template(struct netns_ipvs
*ipvs
,
750 struct ip_vs_dest
*dest
)
753 return ipvs
->sysctl_expire_quiescent_template
&&
754 (atomic_read(&dest
->weight
) == 0);
761 * Checking if the destination of a connection template is available.
762 * If available, return 1, otherwise invalidate this connection
763 * template and return 0.
765 int ip_vs_check_template(struct ip_vs_conn
*ct
, struct ip_vs_dest
*cdest
)
767 struct ip_vs_dest
*dest
= ct
->dest
;
768 struct netns_ipvs
*ipvs
= ct
->ipvs
;
771 * Checking the dest server status.
773 if ((dest
== NULL
) ||
774 !(dest
->flags
& IP_VS_DEST_F_AVAILABLE
) ||
775 expire_quiescent_template(ipvs
, dest
) ||
776 (cdest
&& (dest
!= cdest
))) {
777 IP_VS_DBG_BUF(9, "check_template: dest not available for "
778 "protocol %s s:%s:%d v:%s:%d "
780 ip_vs_proto_name(ct
->protocol
),
781 IP_VS_DBG_ADDR(ct
->af
, &ct
->caddr
),
783 IP_VS_DBG_ADDR(ct
->af
, &ct
->vaddr
),
785 IP_VS_DBG_ADDR(ct
->daf
, &ct
->daddr
),
789 * Invalidate the connection template
791 if (ct
->vport
!= htons(0xffff)) {
792 if (ip_vs_conn_unhash(ct
)) {
793 ct
->dport
= htons(0xffff);
794 ct
->vport
= htons(0xffff);
801 * Simply decrease the refcnt of the template,
802 * don't restart its timer.
804 __ip_vs_conn_put(ct
);
810 static void ip_vs_conn_rcu_free(struct rcu_head
*head
)
812 struct ip_vs_conn
*cp
= container_of(head
, struct ip_vs_conn
,
815 ip_vs_pe_put(cp
->pe
);
817 kmem_cache_free(ip_vs_conn_cachep
, cp
);
820 static void ip_vs_conn_expire(unsigned long data
)
822 struct ip_vs_conn
*cp
= (struct ip_vs_conn
*)data
;
823 struct netns_ipvs
*ipvs
= cp
->ipvs
;
826 * do I control anybody?
828 if (atomic_read(&cp
->n_control
))
831 /* Unlink conn if not referenced anymore */
832 if (likely(ip_vs_conn_unlink(cp
))) {
833 /* delete the timer if it is activated by other users */
834 del_timer(&cp
->timer
);
836 /* does anybody control me? */
838 ip_vs_control_del(cp
);
840 if ((cp
->flags
& IP_VS_CONN_F_NFCT
) &&
841 !(cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)) {
842 /* Do not access conntracks during subsys cleanup
843 * because nf_conntrack_find_get can not be used after
844 * conntrack cleanup for the net.
848 ip_vs_conn_drop_conntrack(cp
);
851 if (unlikely(cp
->app
!= NULL
))
852 ip_vs_unbind_app(cp
);
853 ip_vs_unbind_dest(cp
);
854 if (cp
->flags
& IP_VS_CONN_F_NO_CPORT
)
855 atomic_dec(&ip_vs_conn_no_cport_cnt
);
856 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
857 ip_vs_conn_rcu_free(&cp
->rcu_head
);
859 call_rcu(&cp
->rcu_head
, ip_vs_conn_rcu_free
);
860 atomic_dec(&ipvs
->conn_count
);
865 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
866 atomic_read(&cp
->refcnt
),
867 atomic_read(&cp
->n_control
));
869 atomic_inc(&cp
->refcnt
);
872 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
873 ip_vs_sync_conn(ipvs
, cp
, sysctl_sync_threshold(ipvs
));
875 __ip_vs_conn_put_timer(cp
);
878 /* Modify timer, so that it expires as soon as possible.
879 * Can be called without reference only if under RCU lock.
881 void ip_vs_conn_expire_now(struct ip_vs_conn
*cp
)
883 /* Using mod_timer_pending will ensure the timer is not
884 * modified after the final del_timer in ip_vs_conn_expire.
886 if (timer_pending(&cp
->timer
) &&
887 time_after(cp
->timer
.expires
, jiffies
))
888 mod_timer_pending(&cp
->timer
, jiffies
);
893 * Create a new connection entry and hash it into the ip_vs_conn_tab
896 ip_vs_conn_new(const struct ip_vs_conn_param
*p
, int dest_af
,
897 const union nf_inet_addr
*daddr
, __be16 dport
, unsigned int flags
,
898 struct ip_vs_dest
*dest
, __u32 fwmark
)
900 struct ip_vs_conn
*cp
;
901 struct netns_ipvs
*ipvs
= p
->ipvs
;
902 struct ip_vs_proto_data
*pd
= ip_vs_proto_data_get(p
->ipvs
,
905 cp
= kmem_cache_alloc(ip_vs_conn_cachep
, GFP_ATOMIC
);
907 IP_VS_ERR_RL("%s(): no memory\n", __func__
);
911 INIT_HLIST_NODE(&cp
->c_list
);
912 setup_timer(&cp
->timer
, ip_vs_conn_expire
, (unsigned long)cp
);
916 cp
->protocol
= p
->protocol
;
917 ip_vs_addr_set(p
->af
, &cp
->caddr
, p
->caddr
);
918 cp
->cport
= p
->cport
;
919 /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
920 ip_vs_addr_set(p
->protocol
== IPPROTO_IP
? AF_UNSPEC
: p
->af
,
921 &cp
->vaddr
, p
->vaddr
);
922 cp
->vport
= p
->vport
;
923 ip_vs_addr_set(cp
->daf
, &cp
->daddr
, daddr
);
927 if (flags
& IP_VS_CONN_F_TEMPLATE
&& p
->pe
) {
930 cp
->pe_data
= p
->pe_data
;
931 cp
->pe_data_len
= p
->pe_data_len
;
937 spin_lock_init(&cp
->lock
);
940 * Set the entry is referenced by the current thread before hashing
941 * it in the table, so that other thread run ip_vs_random_dropentry
942 * but cannot drop this entry.
944 atomic_set(&cp
->refcnt
, 1);
947 atomic_set(&cp
->n_control
, 0);
948 atomic_set(&cp
->in_pkts
, 0);
950 cp
->packet_xmit
= NULL
;
953 /* reset struct ip_vs_seq */
954 cp
->in_seq
.delta
= 0;
955 cp
->out_seq
.delta
= 0;
957 atomic_inc(&ipvs
->conn_count
);
958 if (flags
& IP_VS_CONN_F_NO_CPORT
)
959 atomic_inc(&ip_vs_conn_no_cport_cnt
);
961 /* Bind the connection with a destination server */
963 ip_vs_bind_dest(cp
, dest
);
965 /* Set its state and timeout */
969 cp
->sync_endtime
= jiffies
& ~3UL;
971 /* Bind its packet transmitter */
972 #ifdef CONFIG_IP_VS_IPV6
973 if (p
->af
== AF_INET6
)
974 ip_vs_bind_xmit_v6(cp
);
979 if (unlikely(pd
&& atomic_read(&pd
->appcnt
)))
980 ip_vs_bind_app(cp
, pd
->pp
);
983 * Allow conntrack to be preserved. By default, conntrack
984 * is created and destroyed for every packet.
985 * Sometimes keeping conntrack can be useful for
986 * IP_VS_CONN_F_ONE_PACKET too.
989 if (ip_vs_conntrack_enabled(ipvs
))
990 cp
->flags
|= IP_VS_CONN_F_NFCT
;
992 /* Hash it in the ip_vs_conn_tab finally */
999 * /proc/net/ip_vs_conn entries
1001 #ifdef CONFIG_PROC_FS
1002 struct ip_vs_iter_state
{
1003 struct seq_net_private p
;
1004 struct hlist_head
*l
;
1007 static void *ip_vs_conn_array(struct seq_file
*seq
, loff_t pos
)
1010 struct ip_vs_conn
*cp
;
1011 struct ip_vs_iter_state
*iter
= seq
->private;
1013 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++) {
1014 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1015 /* __ip_vs_conn_get() is not needed by
1016 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
1019 iter
->l
= &ip_vs_conn_tab
[idx
];
1029 static void *ip_vs_conn_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1032 struct ip_vs_iter_state
*iter
= seq
->private;
1036 return *pos
? ip_vs_conn_array(seq
, *pos
- 1) :SEQ_START_TOKEN
;
1039 static void *ip_vs_conn_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1041 struct ip_vs_conn
*cp
= v
;
1042 struct ip_vs_iter_state
*iter
= seq
->private;
1043 struct hlist_node
*e
;
1044 struct hlist_head
*l
= iter
->l
;
1048 if (v
== SEQ_START_TOKEN
)
1049 return ip_vs_conn_array(seq
, 0);
1051 /* more on same hash chain? */
1052 e
= rcu_dereference(hlist_next_rcu(&cp
->c_list
));
1054 return hlist_entry(e
, struct ip_vs_conn
, c_list
);
1056 idx
= l
- ip_vs_conn_tab
;
1057 while (++idx
< ip_vs_conn_tab_size
) {
1058 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1059 iter
->l
= &ip_vs_conn_tab
[idx
];
1068 static void ip_vs_conn_seq_stop(struct seq_file
*seq
, void *v
)
1074 static int ip_vs_conn_seq_show(struct seq_file
*seq
, void *v
)
1077 if (v
== SEQ_START_TOKEN
)
1079 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1081 const struct ip_vs_conn
*cp
= v
;
1082 struct net
*net
= seq_file_net(seq
);
1083 char pe_data
[IP_VS_PENAME_MAXLEN
+ IP_VS_PEDATA_MAXLEN
+ 3];
1085 char dbuf
[IP_VS_ADDRSTRLEN
];
1087 if (!net_eq(cp
->ipvs
->net
, net
))
1091 len
= strlen(cp
->pe
->name
);
1092 memcpy(pe_data
+ 1, cp
->pe
->name
, len
);
1093 pe_data
[len
+ 1] = ' ';
1095 len
+= cp
->pe
->show_pe_data(cp
, pe_data
+ len
);
1097 pe_data
[len
] = '\0';
1099 #ifdef CONFIG_IP_VS_IPV6
1100 if (cp
->daf
== AF_INET6
)
1101 snprintf(dbuf
, sizeof(dbuf
), "%pI6", &cp
->daddr
.in6
);
1104 snprintf(dbuf
, sizeof(dbuf
), "%08X",
1105 ntohl(cp
->daddr
.ip
));
1107 #ifdef CONFIG_IP_VS_IPV6
1108 if (cp
->af
== AF_INET6
)
1109 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X "
1110 "%s %04X %-11s %7lu%s\n",
1111 ip_vs_proto_name(cp
->protocol
),
1112 &cp
->caddr
.in6
, ntohs(cp
->cport
),
1113 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
1114 dbuf
, ntohs(cp
->dport
),
1115 ip_vs_state_name(cp
->protocol
, cp
->state
),
1116 (cp
->timer
.expires
-jiffies
)/HZ
, pe_data
);
1120 "%-3s %08X %04X %08X %04X"
1121 " %s %04X %-11s %7lu%s\n",
1122 ip_vs_proto_name(cp
->protocol
),
1123 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
1124 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
1125 dbuf
, ntohs(cp
->dport
),
1126 ip_vs_state_name(cp
->protocol
, cp
->state
),
1127 (cp
->timer
.expires
-jiffies
)/HZ
, pe_data
);
1132 static const struct seq_operations ip_vs_conn_seq_ops
= {
1133 .start
= ip_vs_conn_seq_start
,
1134 .next
= ip_vs_conn_seq_next
,
1135 .stop
= ip_vs_conn_seq_stop
,
1136 .show
= ip_vs_conn_seq_show
,
1139 static int ip_vs_conn_open(struct inode
*inode
, struct file
*file
)
1141 return seq_open_net(inode
, file
, &ip_vs_conn_seq_ops
,
1142 sizeof(struct ip_vs_iter_state
));
1145 static const struct file_operations ip_vs_conn_fops
= {
1146 .owner
= THIS_MODULE
,
1147 .open
= ip_vs_conn_open
,
1149 .llseek
= seq_lseek
,
1150 .release
= seq_release_net
,
1153 static const char *ip_vs_origin_name(unsigned int flags
)
1155 if (flags
& IP_VS_CONN_F_SYNC
)
1161 static int ip_vs_conn_sync_seq_show(struct seq_file
*seq
, void *v
)
1163 char dbuf
[IP_VS_ADDRSTRLEN
];
1165 if (v
== SEQ_START_TOKEN
)
1167 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1169 const struct ip_vs_conn
*cp
= v
;
1170 struct net
*net
= seq_file_net(seq
);
1172 if (!net_eq(cp
->ipvs
->net
, net
))
1175 #ifdef CONFIG_IP_VS_IPV6
1176 if (cp
->daf
== AF_INET6
)
1177 snprintf(dbuf
, sizeof(dbuf
), "%pI6", &cp
->daddr
.in6
);
1180 snprintf(dbuf
, sizeof(dbuf
), "%08X",
1181 ntohl(cp
->daddr
.ip
));
1183 #ifdef CONFIG_IP_VS_IPV6
1184 if (cp
->af
== AF_INET6
)
1185 seq_printf(seq
, "%-3s %pI6 %04X %pI6 %04X "
1186 "%s %04X %-11s %-6s %7lu\n",
1187 ip_vs_proto_name(cp
->protocol
),
1188 &cp
->caddr
.in6
, ntohs(cp
->cport
),
1189 &cp
->vaddr
.in6
, ntohs(cp
->vport
),
1190 dbuf
, ntohs(cp
->dport
),
1191 ip_vs_state_name(cp
->protocol
, cp
->state
),
1192 ip_vs_origin_name(cp
->flags
),
1193 (cp
->timer
.expires
-jiffies
)/HZ
);
1197 "%-3s %08X %04X %08X %04X "
1198 "%s %04X %-11s %-6s %7lu\n",
1199 ip_vs_proto_name(cp
->protocol
),
1200 ntohl(cp
->caddr
.ip
), ntohs(cp
->cport
),
1201 ntohl(cp
->vaddr
.ip
), ntohs(cp
->vport
),
1202 dbuf
, ntohs(cp
->dport
),
1203 ip_vs_state_name(cp
->protocol
, cp
->state
),
1204 ip_vs_origin_name(cp
->flags
),
1205 (cp
->timer
.expires
-jiffies
)/HZ
);
1210 static const struct seq_operations ip_vs_conn_sync_seq_ops
= {
1211 .start
= ip_vs_conn_seq_start
,
1212 .next
= ip_vs_conn_seq_next
,
1213 .stop
= ip_vs_conn_seq_stop
,
1214 .show
= ip_vs_conn_sync_seq_show
,
1217 static int ip_vs_conn_sync_open(struct inode
*inode
, struct file
*file
)
1219 return seq_open_net(inode
, file
, &ip_vs_conn_sync_seq_ops
,
1220 sizeof(struct ip_vs_iter_state
));
1223 static const struct file_operations ip_vs_conn_sync_fops
= {
1224 .owner
= THIS_MODULE
,
1225 .open
= ip_vs_conn_sync_open
,
1227 .llseek
= seq_lseek
,
1228 .release
= seq_release_net
,
1235 * Randomly drop connection entries before running out of memory
1237 static inline int todrop_entry(struct ip_vs_conn
*cp
)
1240 * The drop rate array needs tuning for real environments.
1241 * Called from timer bh only => no locking
1243 static const char todrop_rate
[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1244 static char todrop_counter
[9] = {0};
1247 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1248 This will leave enough time for normal connection to get
1250 if (time_before(cp
->timeout
+ jiffies
, cp
->timer
.expires
+ 60*HZ
))
1253 /* Don't drop the entry if its number of incoming packets is not
1254 located in [0, 8] */
1255 i
= atomic_read(&cp
->in_pkts
);
1256 if (i
> 8 || i
< 0) return 0;
1258 if (!todrop_rate
[i
]) return 0;
1259 if (--todrop_counter
[i
] > 0) return 0;
1261 todrop_counter
[i
] = todrop_rate
[i
];
1265 static inline bool ip_vs_conn_ops_mode(struct ip_vs_conn
*cp
)
1267 struct ip_vs_service
*svc
;
1271 svc
= rcu_dereference(cp
->dest
->svc
);
1272 return svc
&& (svc
->flags
& IP_VS_SVC_F_ONEPACKET
);
1275 /* Called from keventd and must protect itself from softirqs */
1276 void ip_vs_random_dropentry(struct netns_ipvs
*ipvs
)
1279 struct ip_vs_conn
*cp
, *cp_c
;
1283 * Randomly scan 1/32 of the whole table every second
1285 for (idx
= 0; idx
< (ip_vs_conn_tab_size
>>5); idx
++) {
1286 unsigned int hash
= prandom_u32() & ip_vs_conn_tab_mask
;
1288 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[hash
], c_list
) {
1289 if (cp
->ipvs
!= ipvs
)
1291 if (cp
->flags
& IP_VS_CONN_F_TEMPLATE
) {
1292 if (atomic_read(&cp
->n_control
) ||
1293 !ip_vs_conn_ops_mode(cp
))
1296 /* connection template of OPS */
1299 if (cp
->protocol
== IPPROTO_TCP
) {
1301 case IP_VS_TCP_S_SYN_RECV
:
1302 case IP_VS_TCP_S_SYNACK
:
1305 case IP_VS_TCP_S_ESTABLISHED
:
1306 if (todrop_entry(cp
))
1313 } else if (cp
->protocol
== IPPROTO_SCTP
) {
1314 switch (cp
->state
) {
1315 case IP_VS_SCTP_S_INIT1
:
1316 case IP_VS_SCTP_S_INIT
:
1318 case IP_VS_SCTP_S_ESTABLISHED
:
1319 if (todrop_entry(cp
))
1327 if (!todrop_entry(cp
))
1331 IP_VS_DBG(4, "del connection\n");
1332 ip_vs_conn_expire_now(cp
);
1334 /* cp->control is valid only with reference to cp */
1335 if (cp_c
&& __ip_vs_conn_get(cp
)) {
1336 IP_VS_DBG(4, "del conn template\n");
1337 ip_vs_conn_expire_now(cp_c
);
1338 __ip_vs_conn_put(cp
);
1348 * Flush all the connection entries in the ip_vs_conn_tab
1350 static void ip_vs_conn_flush(struct netns_ipvs
*ipvs
)
1353 struct ip_vs_conn
*cp
, *cp_c
;
1357 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++) {
1359 hlist_for_each_entry_rcu(cp
, &ip_vs_conn_tab
[idx
], c_list
) {
1360 if (cp
->ipvs
!= ipvs
)
1362 IP_VS_DBG(4, "del connection\n");
1363 ip_vs_conn_expire_now(cp
);
1365 /* cp->control is valid only with reference to cp */
1366 if (cp_c
&& __ip_vs_conn_get(cp
)) {
1367 IP_VS_DBG(4, "del conn template\n");
1368 ip_vs_conn_expire_now(cp_c
);
1369 __ip_vs_conn_put(cp
);
1376 /* the counter may be not NULL, because maybe some conn entries
1377 are run by slow timer handler or unhashed but still referred */
1378 if (atomic_read(&ipvs
->conn_count
) != 0) {
1384 * per netns init and exit
1386 int __net_init
ip_vs_conn_net_init(struct netns_ipvs
*ipvs
)
1388 atomic_set(&ipvs
->conn_count
, 0);
1390 proc_create("ip_vs_conn", 0, ipvs
->net
->proc_net
, &ip_vs_conn_fops
);
1391 proc_create("ip_vs_conn_sync", 0, ipvs
->net
->proc_net
,
1392 &ip_vs_conn_sync_fops
);
1396 void __net_exit
ip_vs_conn_net_cleanup(struct netns_ipvs
*ipvs
)
1398 /* flush all the connection entries first */
1399 ip_vs_conn_flush(ipvs
);
1400 remove_proc_entry("ip_vs_conn", ipvs
->net
->proc_net
);
1401 remove_proc_entry("ip_vs_conn_sync", ipvs
->net
->proc_net
);
1404 int __init
ip_vs_conn_init(void)
1408 /* Compute size and mask */
1409 ip_vs_conn_tab_size
= 1 << ip_vs_conn_tab_bits
;
1410 ip_vs_conn_tab_mask
= ip_vs_conn_tab_size
- 1;
1413 * Allocate the connection hash table and initialize its list heads
1415 ip_vs_conn_tab
= vmalloc(ip_vs_conn_tab_size
* sizeof(*ip_vs_conn_tab
));
1416 if (!ip_vs_conn_tab
)
1419 /* Allocate ip_vs_conn slab cache */
1420 ip_vs_conn_cachep
= kmem_cache_create("ip_vs_conn",
1421 sizeof(struct ip_vs_conn
), 0,
1422 SLAB_HWCACHE_ALIGN
, NULL
);
1423 if (!ip_vs_conn_cachep
) {
1424 vfree(ip_vs_conn_tab
);
1428 pr_info("Connection hash table configured "
1429 "(size=%d, memory=%ldKbytes)\n",
1430 ip_vs_conn_tab_size
,
1431 (long)(ip_vs_conn_tab_size
*sizeof(struct list_head
))/1024);
1432 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1433 sizeof(struct ip_vs_conn
));
1435 for (idx
= 0; idx
< ip_vs_conn_tab_size
; idx
++)
1436 INIT_HLIST_HEAD(&ip_vs_conn_tab
[idx
]);
1438 for (idx
= 0; idx
< CT_LOCKARRAY_SIZE
; idx
++) {
1439 spin_lock_init(&__ip_vs_conntbl_lock_array
[idx
].l
);
1442 /* calculate the random value for connection hash */
1443 get_random_bytes(&ip_vs_conn_rnd
, sizeof(ip_vs_conn_rnd
));
1448 void ip_vs_conn_cleanup(void)
1450 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1452 /* Release the empty cache */
1453 kmem_cache_destroy(ip_vs_conn_cachep
);
1454 vfree(ip_vs_conn_tab
);