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_read_lock(unsigned int key)
91 read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
94 static inline void ct_read_unlock(unsigned int key)
96 read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
99 static inline void ct_write_lock(unsigned int key)
101 write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
104 static inline void ct_write_unlock(unsigned int key)
106 write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
109 static inline void ct_read_lock_bh(unsigned int key)
111 read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
114 static inline void ct_read_unlock_bh(unsigned int key)
116 read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
119 static inline void ct_write_lock_bh(unsigned int key)
121 write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
124 static inline void ct_write_unlock_bh(unsigned int key)
126 write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
131 * Returns hash value for IPVS connection entry
133 static unsigned int ip_vs_conn_hashkey(struct net *net, int af, unsigned int proto,
134 const union nf_inet_addr *addr,
137 #ifdef CONFIG_IP_VS_IPV6
139 return (jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
140 (__force u32)port, proto, ip_vs_conn_rnd) ^
141 ((size_t)net>>8)) & ip_vs_conn_tab_mask;
143 return (jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
145 ((size_t)net>>8)) & ip_vs_conn_tab_mask;
148 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
151 const union nf_inet_addr *addr;
154 if (p->pe_data && p->pe->hashkey_raw)
155 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
158 if (likely(!inverse)) {
166 return ip_vs_conn_hashkey(p->net, p->af, p->protocol, addr, port);
169 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
171 struct ip_vs_conn_param p;
173 ip_vs_conn_fill_param(ip_vs_conn_net(cp), cp->af, cp->protocol,
174 &cp->caddr, cp->cport, NULL, 0, &p);
178 p.pe_data = cp->pe_data;
179 p.pe_data_len = cp->pe_data_len;
182 return ip_vs_conn_hashkey_param(&p, false);
186 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
187 * returns bool success.
189 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
194 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
197 /* Hash by protocol, client address and port */
198 hash = ip_vs_conn_hashkey_conn(cp);
201 spin_lock(&cp->lock);
203 if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
204 hlist_add_head(&cp->c_list, &ip_vs_conn_tab[hash]);
205 cp->flags |= IP_VS_CONN_F_HASHED;
206 atomic_inc(&cp->refcnt);
209 pr_err("%s(): request for already hashed, called from %pF\n",
210 __func__, __builtin_return_address(0));
214 spin_unlock(&cp->lock);
215 ct_write_unlock(hash);
222 * UNhashes ip_vs_conn from ip_vs_conn_tab.
223 * returns bool success.
225 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
230 /* unhash it and decrease its reference counter */
231 hash = ip_vs_conn_hashkey_conn(cp);
234 spin_lock(&cp->lock);
236 if (cp->flags & IP_VS_CONN_F_HASHED) {
237 hlist_del(&cp->c_list);
238 cp->flags &= ~IP_VS_CONN_F_HASHED;
239 atomic_dec(&cp->refcnt);
244 spin_unlock(&cp->lock);
245 ct_write_unlock(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;
262 struct hlist_node *n;
264 hash = ip_vs_conn_hashkey_param(p, false);
268 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
269 if (cp->af == p->af &&
270 p->cport == cp->cport && p->vport == cp->vport &&
271 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
272 ip_vs_addr_equal(p->af, p->vaddr, &cp->vaddr) &&
273 ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
274 p->protocol == cp->protocol &&
275 ip_vs_conn_net_eq(cp, p->net)) {
277 atomic_inc(&cp->refcnt);
278 ct_read_unlock(hash);
283 ct_read_unlock(hash);
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;
347 struct hlist_node *n;
349 hash = ip_vs_conn_hashkey_param(p, false);
353 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
354 if (!ip_vs_conn_net_eq(cp, p->net))
356 if (p->pe_data && p->pe->ct_match) {
357 if (p->pe == cp->pe && p->pe->ct_match(p, cp))
362 if (cp->af == p->af &&
363 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
364 /* protocol should only be IPPROTO_IP if
365 * p->vaddr is a fwmark */
366 ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
367 p->af, p->vaddr, &cp->vaddr) &&
368 p->cport == cp->cport && p->vport == cp->vport &&
369 cp->flags & IP_VS_CONN_F_TEMPLATE &&
370 p->protocol == cp->protocol)
377 atomic_inc(&cp->refcnt);
378 ct_read_unlock(hash);
380 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
381 ip_vs_proto_name(p->protocol),
382 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
383 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
384 cp ? "hit" : "not hit");
389 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
390 * Called for pkts coming from inside-to-OUTside.
391 * p->caddr, p->cport: pkt source address (inside host)
392 * p->vaddr, p->vport: pkt dest address (foreign host) */
393 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
396 struct ip_vs_conn *cp, *ret=NULL;
397 struct hlist_node *n;
400 * Check for "full" addressed entries
402 hash = ip_vs_conn_hashkey_param(p, true);
406 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
407 if (cp->af == p->af &&
408 p->vport == cp->cport && p->cport == cp->dport &&
409 ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
410 ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
411 p->protocol == cp->protocol &&
412 ip_vs_conn_net_eq(cp, p->net)) {
414 atomic_inc(&cp->refcnt);
420 ct_read_unlock(hash);
422 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
423 ip_vs_proto_name(p->protocol),
424 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
425 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
426 ret ? "hit" : "not hit");
432 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
433 const struct ip_vs_iphdr *iph, int inverse)
435 struct ip_vs_conn_param p;
437 if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
440 return ip_vs_conn_out_get(&p);
442 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
445 * Put back the conn and restart its timer with its timeout
447 void ip_vs_conn_put(struct ip_vs_conn *cp)
449 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
451 mod_timer(&cp->timer, jiffies+t);
453 __ip_vs_conn_put(cp);
458 * Fill a no_client_port connection with a client port number
460 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
462 if (ip_vs_conn_unhash(cp)) {
463 spin_lock(&cp->lock);
464 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
465 atomic_dec(&ip_vs_conn_no_cport_cnt);
466 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
469 spin_unlock(&cp->lock);
471 /* hash on new dport */
478 * Bind a connection entry with the corresponding packet_xmit.
479 * Called by ip_vs_conn_new.
481 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
483 switch (IP_VS_FWD_METHOD(cp)) {
484 case IP_VS_CONN_F_MASQ:
485 cp->packet_xmit = ip_vs_nat_xmit;
488 case IP_VS_CONN_F_TUNNEL:
489 cp->packet_xmit = ip_vs_tunnel_xmit;
492 case IP_VS_CONN_F_DROUTE:
493 cp->packet_xmit = ip_vs_dr_xmit;
496 case IP_VS_CONN_F_LOCALNODE:
497 cp->packet_xmit = ip_vs_null_xmit;
500 case IP_VS_CONN_F_BYPASS:
501 cp->packet_xmit = ip_vs_bypass_xmit;
506 #ifdef CONFIG_IP_VS_IPV6
507 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
509 switch (IP_VS_FWD_METHOD(cp)) {
510 case IP_VS_CONN_F_MASQ:
511 cp->packet_xmit = ip_vs_nat_xmit_v6;
514 case IP_VS_CONN_F_TUNNEL:
515 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
518 case IP_VS_CONN_F_DROUTE:
519 cp->packet_xmit = ip_vs_dr_xmit_v6;
522 case IP_VS_CONN_F_LOCALNODE:
523 cp->packet_xmit = ip_vs_null_xmit;
526 case IP_VS_CONN_F_BYPASS:
527 cp->packet_xmit = ip_vs_bypass_xmit_v6;
534 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
536 return atomic_read(&dest->activeconns)
537 + atomic_read(&dest->inactconns);
541 * Bind a connection entry with a virtual service destination
542 * Called just after a new connection entry is created.
545 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
547 unsigned int conn_flags;
550 /* if dest is NULL, then return directly */
554 /* Increase the refcnt counter of the dest */
555 atomic_inc(&dest->refcnt);
557 conn_flags = atomic_read(&dest->conn_flags);
558 if (cp->protocol != IPPROTO_UDP)
559 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
561 /* Bind with the destination and its corresponding transmitter */
562 if (flags & IP_VS_CONN_F_SYNC) {
563 /* if the connection is not template and is created
564 * by sync, preserve the activity flag.
566 if (!(flags & IP_VS_CONN_F_TEMPLATE))
567 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
568 /* connections inherit forwarding method from dest */
569 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
575 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
576 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
578 ip_vs_proto_name(cp->protocol),
579 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
580 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
581 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
582 ip_vs_fwd_tag(cp), cp->state,
583 cp->flags, atomic_read(&cp->refcnt),
584 atomic_read(&dest->refcnt));
586 /* Update the connection counters */
587 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
588 /* It is a normal connection, so modify the counters
589 * according to the flags, later the protocol can
590 * update them on state change
592 if (!(flags & IP_VS_CONN_F_INACTIVE))
593 atomic_inc(&dest->activeconns);
595 atomic_inc(&dest->inactconns);
597 /* It is a persistent connection/template, so increase
598 the persistent connection counter */
599 atomic_inc(&dest->persistconns);
602 if (dest->u_threshold != 0 &&
603 ip_vs_dest_totalconns(dest) >= dest->u_threshold)
604 dest->flags |= IP_VS_DEST_F_OVERLOAD;
609 * Check if there is a destination for the connection, if so
610 * bind the connection to the destination.
612 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
614 struct ip_vs_dest *dest;
616 dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, &cp->daddr,
617 cp->dport, &cp->vaddr, cp->vport,
618 cp->protocol, cp->fwmark, cp->flags);
620 struct ip_vs_proto_data *pd;
622 spin_lock(&cp->lock);
624 spin_unlock(&cp->lock);
628 /* Applications work depending on the forwarding method
629 * but better to reassign them always when binding dest */
631 ip_vs_unbind_app(cp);
633 ip_vs_bind_dest(cp, dest);
634 spin_unlock(&cp->lock);
636 /* Update its packet transmitter */
637 cp->packet_xmit = NULL;
638 #ifdef CONFIG_IP_VS_IPV6
639 if (cp->af == AF_INET6)
640 ip_vs_bind_xmit_v6(cp);
645 pd = ip_vs_proto_data_get(ip_vs_conn_net(cp), cp->protocol);
646 if (pd && atomic_read(&pd->appcnt))
647 ip_vs_bind_app(cp, pd->pp);
654 * Unbind a connection entry with its VS destination
655 * Called by the ip_vs_conn_expire function.
657 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
659 struct ip_vs_dest *dest = cp->dest;
664 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
665 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
667 ip_vs_proto_name(cp->protocol),
668 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
669 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
670 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
671 ip_vs_fwd_tag(cp), cp->state,
672 cp->flags, atomic_read(&cp->refcnt),
673 atomic_read(&dest->refcnt));
675 /* Update the connection counters */
676 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
677 /* It is a normal connection, so decrease the inactconns
678 or activeconns counter */
679 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
680 atomic_dec(&dest->inactconns);
682 atomic_dec(&dest->activeconns);
685 /* It is a persistent connection/template, so decrease
686 the persistent connection counter */
687 atomic_dec(&dest->persistconns);
690 if (dest->l_threshold != 0) {
691 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
692 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
693 } else if (dest->u_threshold != 0) {
694 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
695 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
697 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
698 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
702 * Simply decrease the refcnt of the dest, because the
703 * dest will be either in service's destination list
706 atomic_dec(&dest->refcnt);
709 static int expire_quiescent_template(struct netns_ipvs *ipvs,
710 struct ip_vs_dest *dest)
713 return ipvs->sysctl_expire_quiescent_template &&
714 (atomic_read(&dest->weight) == 0);
721 * Checking if the destination of a connection template is available.
722 * If available, return 1, otherwise invalidate this connection
723 * template and return 0.
725 int ip_vs_check_template(struct ip_vs_conn *ct)
727 struct ip_vs_dest *dest = ct->dest;
728 struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
731 * Checking the dest server status.
733 if ((dest == NULL) ||
734 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
735 expire_quiescent_template(ipvs, dest)) {
736 IP_VS_DBG_BUF(9, "check_template: dest not available for "
737 "protocol %s s:%s:%d v:%s:%d "
739 ip_vs_proto_name(ct->protocol),
740 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
742 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
744 IP_VS_DBG_ADDR(ct->af, &ct->daddr),
748 * Invalidate the connection template
750 if (ct->vport != htons(0xffff)) {
751 if (ip_vs_conn_unhash(ct)) {
752 ct->dport = htons(0xffff);
753 ct->vport = htons(0xffff);
760 * Simply decrease the refcnt of the template,
761 * don't restart its timer.
763 atomic_dec(&ct->refcnt);
769 static void ip_vs_conn_expire(unsigned long data)
771 struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
772 struct net *net = ip_vs_conn_net(cp);
773 struct netns_ipvs *ipvs = net_ipvs(net);
780 atomic_inc(&cp->refcnt);
783 * do I control anybody?
785 if (atomic_read(&cp->n_control))
789 * unhash it if it is hashed in the conn table
791 if (!ip_vs_conn_unhash(cp) && !(cp->flags & IP_VS_CONN_F_ONE_PACKET))
795 * refcnt==1 implies I'm the only one referrer
797 if (likely(atomic_read(&cp->refcnt) == 1)) {
798 /* delete the timer if it is activated by other users */
799 del_timer(&cp->timer);
801 /* does anybody control me? */
803 ip_vs_control_del(cp);
805 if (cp->flags & IP_VS_CONN_F_NFCT) {
806 ip_vs_conn_drop_conntrack(cp);
807 /* Do not access conntracks during subsys cleanup
808 * because nf_conntrack_find_get can not be used after
809 * conntrack cleanup for the net.
813 ip_vs_conn_drop_conntrack(cp);
816 ip_vs_pe_put(cp->pe);
818 if (unlikely(cp->app != NULL))
819 ip_vs_unbind_app(cp);
820 ip_vs_unbind_dest(cp);
821 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
822 atomic_dec(&ip_vs_conn_no_cport_cnt);
823 atomic_dec(&ipvs->conn_count);
825 kmem_cache_free(ip_vs_conn_cachep, cp);
829 /* hash it back to the table */
833 IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
834 atomic_read(&cp->refcnt)-1,
835 atomic_read(&cp->n_control));
837 if (ipvs->sync_state & IP_VS_STATE_MASTER)
838 ip_vs_sync_conn(net, cp, sysctl_sync_threshold(ipvs));
844 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
846 if (del_timer(&cp->timer))
847 mod_timer(&cp->timer, jiffies);
852 * Create a new connection entry and hash it into the ip_vs_conn_tab
855 ip_vs_conn_new(const struct ip_vs_conn_param *p,
856 const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
857 struct ip_vs_dest *dest, __u32 fwmark)
859 struct ip_vs_conn *cp;
860 struct netns_ipvs *ipvs = net_ipvs(p->net);
861 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->net,
864 cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
866 IP_VS_ERR_RL("%s(): no memory\n", __func__);
870 INIT_HLIST_NODE(&cp->c_list);
871 setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
872 ip_vs_conn_net_set(cp, p->net);
874 cp->protocol = p->protocol;
875 ip_vs_addr_copy(p->af, &cp->caddr, p->caddr);
876 cp->cport = p->cport;
877 ip_vs_addr_copy(p->af, &cp->vaddr, p->vaddr);
878 cp->vport = p->vport;
879 /* proto should only be IPPROTO_IP if d_addr is a fwmark */
880 ip_vs_addr_copy(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
885 if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
888 cp->pe_data = p->pe_data;
889 cp->pe_data_len = p->pe_data_len;
891 spin_lock_init(&cp->lock);
894 * Set the entry is referenced by the current thread before hashing
895 * it in the table, so that other thread run ip_vs_random_dropentry
896 * but cannot drop this entry.
898 atomic_set(&cp->refcnt, 1);
900 atomic_set(&cp->n_control, 0);
901 atomic_set(&cp->in_pkts, 0);
903 atomic_inc(&ipvs->conn_count);
904 if (flags & IP_VS_CONN_F_NO_CPORT)
905 atomic_inc(&ip_vs_conn_no_cport_cnt);
907 /* Bind the connection with a destination server */
908 ip_vs_bind_dest(cp, dest);
910 /* Set its state and timeout */
913 cp->sync_endtime = jiffies & ~3UL;
915 /* Bind its packet transmitter */
916 #ifdef CONFIG_IP_VS_IPV6
917 if (p->af == AF_INET6)
918 ip_vs_bind_xmit_v6(cp);
923 if (unlikely(pd && atomic_read(&pd->appcnt)))
924 ip_vs_bind_app(cp, pd->pp);
927 * Allow conntrack to be preserved. By default, conntrack
928 * is created and destroyed for every packet.
929 * Sometimes keeping conntrack can be useful for
930 * IP_VS_CONN_F_ONE_PACKET too.
933 if (ip_vs_conntrack_enabled(ipvs))
934 cp->flags |= IP_VS_CONN_F_NFCT;
936 /* Hash it in the ip_vs_conn_tab finally */
943 * /proc/net/ip_vs_conn entries
945 #ifdef CONFIG_PROC_FS
946 struct ip_vs_iter_state {
947 struct seq_net_private p;
948 struct hlist_head *l;
951 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
954 struct ip_vs_conn *cp;
955 struct ip_vs_iter_state *iter = seq->private;
956 struct hlist_node *n;
958 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
959 ct_read_lock_bh(idx);
960 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[idx], c_list) {
962 iter->l = &ip_vs_conn_tab[idx];
966 ct_read_unlock_bh(idx);
972 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
974 struct ip_vs_iter_state *iter = seq->private;
977 return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
980 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
982 struct ip_vs_conn *cp = v;
983 struct ip_vs_iter_state *iter = seq->private;
984 struct hlist_node *e;
985 struct hlist_head *l = iter->l;
989 if (v == SEQ_START_TOKEN)
990 return ip_vs_conn_array(seq, 0);
992 /* more on same hash chain? */
993 if ((e = cp->c_list.next))
994 return hlist_entry(e, struct ip_vs_conn, c_list);
996 idx = l - ip_vs_conn_tab;
997 ct_read_unlock_bh(idx);
999 while (++idx < ip_vs_conn_tab_size) {
1000 ct_read_lock_bh(idx);
1001 hlist_for_each_entry(cp, e, &ip_vs_conn_tab[idx], c_list) {
1002 iter->l = &ip_vs_conn_tab[idx];
1005 ct_read_unlock_bh(idx);
1011 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1013 struct ip_vs_iter_state *iter = seq->private;
1014 struct hlist_head *l = iter->l;
1017 ct_read_unlock_bh(l - ip_vs_conn_tab);
1020 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1023 if (v == SEQ_START_TOKEN)
1025 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1027 const struct ip_vs_conn *cp = v;
1028 struct net *net = seq_file_net(seq);
1029 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1032 if (!ip_vs_conn_net_eq(cp, net))
1036 len = strlen(cp->pe->name);
1037 memcpy(pe_data + 1, cp->pe->name, len);
1038 pe_data[len + 1] = ' ';
1040 len += cp->pe->show_pe_data(cp, pe_data + len);
1042 pe_data[len] = '\0';
1044 #ifdef CONFIG_IP_VS_IPV6
1045 if (cp->af == AF_INET6)
1046 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1047 "%pI6 %04X %-11s %7lu%s\n",
1048 ip_vs_proto_name(cp->protocol),
1049 &cp->caddr.in6, ntohs(cp->cport),
1050 &cp->vaddr.in6, ntohs(cp->vport),
1051 &cp->daddr.in6, ntohs(cp->dport),
1052 ip_vs_state_name(cp->protocol, cp->state),
1053 (cp->timer.expires-jiffies)/HZ, pe_data);
1057 "%-3s %08X %04X %08X %04X"
1058 " %08X %04X %-11s %7lu%s\n",
1059 ip_vs_proto_name(cp->protocol),
1060 ntohl(cp->caddr.ip), ntohs(cp->cport),
1061 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1062 ntohl(cp->daddr.ip), ntohs(cp->dport),
1063 ip_vs_state_name(cp->protocol, cp->state),
1064 (cp->timer.expires-jiffies)/HZ, pe_data);
1069 static const struct seq_operations ip_vs_conn_seq_ops = {
1070 .start = ip_vs_conn_seq_start,
1071 .next = ip_vs_conn_seq_next,
1072 .stop = ip_vs_conn_seq_stop,
1073 .show = ip_vs_conn_seq_show,
1076 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1078 return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1079 sizeof(struct ip_vs_iter_state));
1082 static const struct file_operations ip_vs_conn_fops = {
1083 .owner = THIS_MODULE,
1084 .open = ip_vs_conn_open,
1086 .llseek = seq_lseek,
1087 .release = seq_release_net,
1090 static const char *ip_vs_origin_name(unsigned int flags)
1092 if (flags & IP_VS_CONN_F_SYNC)
1098 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1101 if (v == SEQ_START_TOKEN)
1103 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1105 const struct ip_vs_conn *cp = v;
1106 struct net *net = seq_file_net(seq);
1108 if (!ip_vs_conn_net_eq(cp, net))
1111 #ifdef CONFIG_IP_VS_IPV6
1112 if (cp->af == AF_INET6)
1113 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1114 ip_vs_proto_name(cp->protocol),
1115 &cp->caddr.in6, ntohs(cp->cport),
1116 &cp->vaddr.in6, ntohs(cp->vport),
1117 &cp->daddr.in6, ntohs(cp->dport),
1118 ip_vs_state_name(cp->protocol, cp->state),
1119 ip_vs_origin_name(cp->flags),
1120 (cp->timer.expires-jiffies)/HZ);
1124 "%-3s %08X %04X %08X %04X "
1125 "%08X %04X %-11s %-6s %7lu\n",
1126 ip_vs_proto_name(cp->protocol),
1127 ntohl(cp->caddr.ip), ntohs(cp->cport),
1128 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1129 ntohl(cp->daddr.ip), ntohs(cp->dport),
1130 ip_vs_state_name(cp->protocol, cp->state),
1131 ip_vs_origin_name(cp->flags),
1132 (cp->timer.expires-jiffies)/HZ);
1137 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1138 .start = ip_vs_conn_seq_start,
1139 .next = ip_vs_conn_seq_next,
1140 .stop = ip_vs_conn_seq_stop,
1141 .show = ip_vs_conn_sync_seq_show,
1144 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1146 return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1147 sizeof(struct ip_vs_iter_state));
1150 static const struct file_operations ip_vs_conn_sync_fops = {
1151 .owner = THIS_MODULE,
1152 .open = ip_vs_conn_sync_open,
1154 .llseek = seq_lseek,
1155 .release = seq_release_net,
1162 * Randomly drop connection entries before running out of memory
1164 static inline int todrop_entry(struct ip_vs_conn *cp)
1167 * The drop rate array needs tuning for real environments.
1168 * Called from timer bh only => no locking
1170 static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1171 static char todrop_counter[9] = {0};
1174 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1175 This will leave enough time for normal connection to get
1177 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1180 /* Don't drop the entry if its number of incoming packets is not
1181 located in [0, 8] */
1182 i = atomic_read(&cp->in_pkts);
1183 if (i > 8 || i < 0) return 0;
1185 if (!todrop_rate[i]) return 0;
1186 if (--todrop_counter[i] > 0) return 0;
1188 todrop_counter[i] = todrop_rate[i];
1192 /* Called from keventd and must protect itself from softirqs */
1193 void ip_vs_random_dropentry(struct net *net)
1196 struct ip_vs_conn *cp;
1199 * Randomly scan 1/32 of the whole table every second
1201 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1202 unsigned int hash = net_random() & ip_vs_conn_tab_mask;
1203 struct hlist_node *n;
1206 * Lock is actually needed in this loop.
1208 ct_write_lock_bh(hash);
1210 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
1211 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1212 /* connection template */
1214 if (!ip_vs_conn_net_eq(cp, net))
1216 if (cp->protocol == IPPROTO_TCP) {
1218 case IP_VS_TCP_S_SYN_RECV:
1219 case IP_VS_TCP_S_SYNACK:
1222 case IP_VS_TCP_S_ESTABLISHED:
1223 if (todrop_entry(cp))
1231 if (!todrop_entry(cp))
1235 IP_VS_DBG(4, "del connection\n");
1236 ip_vs_conn_expire_now(cp);
1238 IP_VS_DBG(4, "del conn template\n");
1239 ip_vs_conn_expire_now(cp->control);
1242 ct_write_unlock_bh(hash);
1248 * Flush all the connection entries in the ip_vs_conn_tab
1250 static void ip_vs_conn_flush(struct net *net)
1253 struct ip_vs_conn *cp;
1254 struct netns_ipvs *ipvs = net_ipvs(net);
1257 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1258 struct hlist_node *n;
1261 * Lock is actually needed in this loop.
1263 ct_write_lock_bh(idx);
1265 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[idx], c_list) {
1266 if (!ip_vs_conn_net_eq(cp, net))
1268 IP_VS_DBG(4, "del connection\n");
1269 ip_vs_conn_expire_now(cp);
1271 IP_VS_DBG(4, "del conn template\n");
1272 ip_vs_conn_expire_now(cp->control);
1275 ct_write_unlock_bh(idx);
1278 /* the counter may be not NULL, because maybe some conn entries
1279 are run by slow timer handler or unhashed but still referred */
1280 if (atomic_read(&ipvs->conn_count) != 0) {
1286 * per netns init and exit
1288 int __net_init ip_vs_conn_net_init(struct net *net)
1290 struct netns_ipvs *ipvs = net_ipvs(net);
1292 atomic_set(&ipvs->conn_count, 0);
1294 proc_create("ip_vs_conn", 0, net->proc_net, &ip_vs_conn_fops);
1295 proc_create("ip_vs_conn_sync", 0, net->proc_net, &ip_vs_conn_sync_fops);
1299 void __net_exit ip_vs_conn_net_cleanup(struct net *net)
1301 /* flush all the connection entries first */
1302 ip_vs_conn_flush(net);
1303 remove_proc_entry("ip_vs_conn", net->proc_net);
1304 remove_proc_entry("ip_vs_conn_sync", net->proc_net);
1307 int __init ip_vs_conn_init(void)
1311 /* Compute size and mask */
1312 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1313 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1316 * Allocate the connection hash table and initialize its list heads
1318 ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1319 if (!ip_vs_conn_tab)
1322 /* Allocate ip_vs_conn slab cache */
1323 ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1324 sizeof(struct ip_vs_conn), 0,
1325 SLAB_HWCACHE_ALIGN, NULL);
1326 if (!ip_vs_conn_cachep) {
1327 vfree(ip_vs_conn_tab);
1331 pr_info("Connection hash table configured "
1332 "(size=%d, memory=%ldKbytes)\n",
1333 ip_vs_conn_tab_size,
1334 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1335 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1336 sizeof(struct ip_vs_conn));
1338 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1339 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1341 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1342 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1345 /* calculate the random value for connection hash */
1346 get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1351 void ip_vs_conn_cleanup(void)
1353 /* Release the empty cache */
1354 kmem_cache_destroy(ip_vs_conn_cachep);
1355 vfree(ip_vs_conn_tab);