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;
263 hash = ip_vs_conn_hashkey_param(p, false);
267 hlist_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
268 if (cp->af == p->af &&
269 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)) {
276 atomic_inc(&cp->refcnt);
277 ct_read_unlock(hash);
282 ct_read_unlock(hash);
287 struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
289 struct ip_vs_conn *cp;
291 cp = __ip_vs_conn_in_get(p);
292 if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) {
293 struct ip_vs_conn_param cport_zero_p = *p;
294 cport_zero_p.cport = 0;
295 cp = __ip_vs_conn_in_get(&cport_zero_p);
298 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
299 ip_vs_proto_name(p->protocol),
300 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
301 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
302 cp ? "hit" : "not hit");
308 ip_vs_conn_fill_param_proto(int af, const struct sk_buff *skb,
309 const struct ip_vs_iphdr *iph,
310 int inverse, struct ip_vs_conn_param *p)
312 __be16 _ports[2], *pptr;
313 struct net *net = skb_net(skb);
315 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
319 if (likely(!inverse))
320 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->saddr,
321 pptr[0], &iph->daddr, pptr[1], p);
323 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->daddr,
324 pptr[1], &iph->saddr, pptr[0], p);
329 ip_vs_conn_in_get_proto(int af, const struct sk_buff *skb,
330 const struct ip_vs_iphdr *iph, int inverse)
332 struct ip_vs_conn_param p;
334 if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
337 return ip_vs_conn_in_get(&p);
339 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
341 /* Get reference to connection template */
342 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
345 struct ip_vs_conn *cp;
347 hash = ip_vs_conn_hashkey_param(p, false);
351 hlist_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
352 if (!ip_vs_conn_net_eq(cp, p->net))
354 if (p->pe_data && p->pe->ct_match) {
355 if (p->pe == cp->pe && p->pe->ct_match(p, cp))
360 if (cp->af == p->af &&
361 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
362 /* protocol should only be IPPROTO_IP if
363 * p->vaddr is a fwmark */
364 ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
365 p->af, p->vaddr, &cp->vaddr) &&
366 p->cport == cp->cport && p->vport == cp->vport &&
367 cp->flags & IP_VS_CONN_F_TEMPLATE &&
368 p->protocol == cp->protocol)
375 atomic_inc(&cp->refcnt);
376 ct_read_unlock(hash);
378 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
379 ip_vs_proto_name(p->protocol),
380 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
381 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
382 cp ? "hit" : "not hit");
387 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
388 * Called for pkts coming from inside-to-OUTside.
389 * p->caddr, p->cport: pkt source address (inside host)
390 * p->vaddr, p->vport: pkt dest address (foreign host) */
391 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
394 struct ip_vs_conn *cp, *ret=NULL;
397 * Check for "full" addressed entries
399 hash = ip_vs_conn_hashkey_param(p, true);
403 hlist_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
404 if (cp->af == p->af &&
405 p->vport == cp->cport && p->cport == cp->dport &&
406 ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
407 ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
408 p->protocol == cp->protocol &&
409 ip_vs_conn_net_eq(cp, p->net)) {
411 atomic_inc(&cp->refcnt);
417 ct_read_unlock(hash);
419 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
420 ip_vs_proto_name(p->protocol),
421 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
422 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
423 ret ? "hit" : "not hit");
429 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
430 const struct ip_vs_iphdr *iph, int inverse)
432 struct ip_vs_conn_param p;
434 if (ip_vs_conn_fill_param_proto(af, skb, iph, inverse, &p))
437 return ip_vs_conn_out_get(&p);
439 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
442 * Put back the conn and restart its timer with its timeout
444 void ip_vs_conn_put(struct ip_vs_conn *cp)
446 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
448 mod_timer(&cp->timer, jiffies+t);
450 __ip_vs_conn_put(cp);
455 * Fill a no_client_port connection with a client port number
457 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
459 if (ip_vs_conn_unhash(cp)) {
460 spin_lock(&cp->lock);
461 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
462 atomic_dec(&ip_vs_conn_no_cport_cnt);
463 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
466 spin_unlock(&cp->lock);
468 /* hash on new dport */
475 * Bind a connection entry with the corresponding packet_xmit.
476 * Called by ip_vs_conn_new.
478 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
480 switch (IP_VS_FWD_METHOD(cp)) {
481 case IP_VS_CONN_F_MASQ:
482 cp->packet_xmit = ip_vs_nat_xmit;
485 case IP_VS_CONN_F_TUNNEL:
486 cp->packet_xmit = ip_vs_tunnel_xmit;
489 case IP_VS_CONN_F_DROUTE:
490 cp->packet_xmit = ip_vs_dr_xmit;
493 case IP_VS_CONN_F_LOCALNODE:
494 cp->packet_xmit = ip_vs_null_xmit;
497 case IP_VS_CONN_F_BYPASS:
498 cp->packet_xmit = ip_vs_bypass_xmit;
503 #ifdef CONFIG_IP_VS_IPV6
504 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
506 switch (IP_VS_FWD_METHOD(cp)) {
507 case IP_VS_CONN_F_MASQ:
508 cp->packet_xmit = ip_vs_nat_xmit_v6;
511 case IP_VS_CONN_F_TUNNEL:
512 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
515 case IP_VS_CONN_F_DROUTE:
516 cp->packet_xmit = ip_vs_dr_xmit_v6;
519 case IP_VS_CONN_F_LOCALNODE:
520 cp->packet_xmit = ip_vs_null_xmit;
523 case IP_VS_CONN_F_BYPASS:
524 cp->packet_xmit = ip_vs_bypass_xmit_v6;
531 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
533 return atomic_read(&dest->activeconns)
534 + atomic_read(&dest->inactconns);
538 * Bind a connection entry with a virtual service destination
539 * Called just after a new connection entry is created.
542 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
544 unsigned int conn_flags;
547 /* if dest is NULL, then return directly */
551 /* Increase the refcnt counter of the dest */
552 atomic_inc(&dest->refcnt);
554 conn_flags = atomic_read(&dest->conn_flags);
555 if (cp->protocol != IPPROTO_UDP)
556 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
558 /* Bind with the destination and its corresponding transmitter */
559 if (flags & IP_VS_CONN_F_SYNC) {
560 /* if the connection is not template and is created
561 * by sync, preserve the activity flag.
563 if (!(flags & IP_VS_CONN_F_TEMPLATE))
564 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
565 /* connections inherit forwarding method from dest */
566 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
572 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
573 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
575 ip_vs_proto_name(cp->protocol),
576 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
577 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
578 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
579 ip_vs_fwd_tag(cp), cp->state,
580 cp->flags, atomic_read(&cp->refcnt),
581 atomic_read(&dest->refcnt));
583 /* Update the connection counters */
584 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
585 /* It is a normal connection, so modify the counters
586 * according to the flags, later the protocol can
587 * update them on state change
589 if (!(flags & IP_VS_CONN_F_INACTIVE))
590 atomic_inc(&dest->activeconns);
592 atomic_inc(&dest->inactconns);
594 /* It is a persistent connection/template, so increase
595 the persistent connection counter */
596 atomic_inc(&dest->persistconns);
599 if (dest->u_threshold != 0 &&
600 ip_vs_dest_totalconns(dest) >= dest->u_threshold)
601 dest->flags |= IP_VS_DEST_F_OVERLOAD;
606 * Check if there is a destination for the connection, if so
607 * bind the connection to the destination.
609 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
611 struct ip_vs_dest *dest;
613 dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, &cp->daddr,
614 cp->dport, &cp->vaddr, cp->vport,
615 cp->protocol, cp->fwmark, cp->flags);
617 struct ip_vs_proto_data *pd;
619 spin_lock(&cp->lock);
621 spin_unlock(&cp->lock);
625 /* Applications work depending on the forwarding method
626 * but better to reassign them always when binding dest */
628 ip_vs_unbind_app(cp);
630 ip_vs_bind_dest(cp, dest);
631 spin_unlock(&cp->lock);
633 /* Update its packet transmitter */
634 cp->packet_xmit = NULL;
635 #ifdef CONFIG_IP_VS_IPV6
636 if (cp->af == AF_INET6)
637 ip_vs_bind_xmit_v6(cp);
642 pd = ip_vs_proto_data_get(ip_vs_conn_net(cp), cp->protocol);
643 if (pd && atomic_read(&pd->appcnt))
644 ip_vs_bind_app(cp, pd->pp);
651 * Unbind a connection entry with its VS destination
652 * Called by the ip_vs_conn_expire function.
654 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
656 struct ip_vs_dest *dest = cp->dest;
661 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
662 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
664 ip_vs_proto_name(cp->protocol),
665 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
666 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
667 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
668 ip_vs_fwd_tag(cp), cp->state,
669 cp->flags, atomic_read(&cp->refcnt),
670 atomic_read(&dest->refcnt));
672 /* Update the connection counters */
673 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
674 /* It is a normal connection, so decrease the inactconns
675 or activeconns counter */
676 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
677 atomic_dec(&dest->inactconns);
679 atomic_dec(&dest->activeconns);
682 /* It is a persistent connection/template, so decrease
683 the persistent connection counter */
684 atomic_dec(&dest->persistconns);
687 if (dest->l_threshold != 0) {
688 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
689 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
690 } else if (dest->u_threshold != 0) {
691 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
692 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
694 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
695 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
699 * Simply decrease the refcnt of the dest, because the
700 * dest will be either in service's destination list
703 atomic_dec(&dest->refcnt);
706 static int expire_quiescent_template(struct netns_ipvs *ipvs,
707 struct ip_vs_dest *dest)
710 return ipvs->sysctl_expire_quiescent_template &&
711 (atomic_read(&dest->weight) == 0);
718 * Checking if the destination of a connection template is available.
719 * If available, return 1, otherwise invalidate this connection
720 * template and return 0.
722 int ip_vs_check_template(struct ip_vs_conn *ct)
724 struct ip_vs_dest *dest = ct->dest;
725 struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
728 * Checking the dest server status.
730 if ((dest == NULL) ||
731 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
732 expire_quiescent_template(ipvs, dest)) {
733 IP_VS_DBG_BUF(9, "check_template: dest not available for "
734 "protocol %s s:%s:%d v:%s:%d "
736 ip_vs_proto_name(ct->protocol),
737 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
739 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
741 IP_VS_DBG_ADDR(ct->af, &ct->daddr),
745 * Invalidate the connection template
747 if (ct->vport != htons(0xffff)) {
748 if (ip_vs_conn_unhash(ct)) {
749 ct->dport = htons(0xffff);
750 ct->vport = htons(0xffff);
757 * Simply decrease the refcnt of the template,
758 * don't restart its timer.
760 atomic_dec(&ct->refcnt);
766 static void ip_vs_conn_expire(unsigned long data)
768 struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
769 struct net *net = ip_vs_conn_net(cp);
770 struct netns_ipvs *ipvs = net_ipvs(net);
777 atomic_inc(&cp->refcnt);
780 * do I control anybody?
782 if (atomic_read(&cp->n_control))
786 * unhash it if it is hashed in the conn table
788 if (!ip_vs_conn_unhash(cp) && !(cp->flags & IP_VS_CONN_F_ONE_PACKET))
792 * refcnt==1 implies I'm the only one referrer
794 if (likely(atomic_read(&cp->refcnt) == 1)) {
795 /* delete the timer if it is activated by other users */
796 del_timer(&cp->timer);
798 /* does anybody control me? */
800 ip_vs_control_del(cp);
802 if (cp->flags & IP_VS_CONN_F_NFCT) {
803 ip_vs_conn_drop_conntrack(cp);
804 /* Do not access conntracks during subsys cleanup
805 * because nf_conntrack_find_get can not be used after
806 * conntrack cleanup for the net.
810 ip_vs_conn_drop_conntrack(cp);
813 ip_vs_pe_put(cp->pe);
815 if (unlikely(cp->app != NULL))
816 ip_vs_unbind_app(cp);
817 ip_vs_unbind_dest(cp);
818 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
819 atomic_dec(&ip_vs_conn_no_cport_cnt);
820 atomic_dec(&ipvs->conn_count);
822 kmem_cache_free(ip_vs_conn_cachep, cp);
826 /* hash it back to the table */
830 IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
831 atomic_read(&cp->refcnt)-1,
832 atomic_read(&cp->n_control));
834 if (ipvs->sync_state & IP_VS_STATE_MASTER)
835 ip_vs_sync_conn(net, cp, sysctl_sync_threshold(ipvs));
841 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
843 if (del_timer(&cp->timer))
844 mod_timer(&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_zalloc(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_copy(p->af, &cp->caddr, p->caddr);
873 cp->cport = p->cport;
874 ip_vs_addr_copy(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_copy(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;
888 spin_lock_init(&cp->lock);
891 * Set the entry is referenced by the current thread before hashing
892 * it in the table, so that other thread run ip_vs_random_dropentry
893 * but cannot drop this entry.
895 atomic_set(&cp->refcnt, 1);
897 atomic_set(&cp->n_control, 0);
898 atomic_set(&cp->in_pkts, 0);
900 atomic_inc(&ipvs->conn_count);
901 if (flags & IP_VS_CONN_F_NO_CPORT)
902 atomic_inc(&ip_vs_conn_no_cport_cnt);
904 /* Bind the connection with a destination server */
905 ip_vs_bind_dest(cp, dest);
907 /* Set its state and timeout */
910 cp->sync_endtime = jiffies & ~3UL;
912 /* Bind its packet transmitter */
913 #ifdef CONFIG_IP_VS_IPV6
914 if (p->af == AF_INET6)
915 ip_vs_bind_xmit_v6(cp);
920 if (unlikely(pd && atomic_read(&pd->appcnt)))
921 ip_vs_bind_app(cp, pd->pp);
924 * Allow conntrack to be preserved. By default, conntrack
925 * is created and destroyed for every packet.
926 * Sometimes keeping conntrack can be useful for
927 * IP_VS_CONN_F_ONE_PACKET too.
930 if (ip_vs_conntrack_enabled(ipvs))
931 cp->flags |= IP_VS_CONN_F_NFCT;
933 /* Hash it in the ip_vs_conn_tab finally */
940 * /proc/net/ip_vs_conn entries
942 #ifdef CONFIG_PROC_FS
943 struct ip_vs_iter_state {
944 struct seq_net_private p;
945 struct hlist_head *l;
948 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
951 struct ip_vs_conn *cp;
952 struct ip_vs_iter_state *iter = seq->private;
954 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
955 ct_read_lock_bh(idx);
956 hlist_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
958 iter->l = &ip_vs_conn_tab[idx];
962 ct_read_unlock_bh(idx);
968 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
970 struct ip_vs_iter_state *iter = seq->private;
973 return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
976 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
978 struct ip_vs_conn *cp = v;
979 struct ip_vs_iter_state *iter = seq->private;
980 struct hlist_head *l = iter->l;
984 if (v == SEQ_START_TOKEN)
985 return ip_vs_conn_array(seq, 0);
987 /* more on same hash chain? */
989 return hlist_entry(cp->c_list.next, struct ip_vs_conn, c_list);
991 idx = l - ip_vs_conn_tab;
992 ct_read_unlock_bh(idx);
994 while (++idx < ip_vs_conn_tab_size) {
995 ct_read_lock_bh(idx);
996 hlist_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
997 iter->l = &ip_vs_conn_tab[idx];
1000 ct_read_unlock_bh(idx);
1006 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1008 struct ip_vs_iter_state *iter = seq->private;
1009 struct hlist_head *l = iter->l;
1012 ct_read_unlock_bh(l - ip_vs_conn_tab);
1015 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1018 if (v == SEQ_START_TOKEN)
1020 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1022 const struct ip_vs_conn *cp = v;
1023 struct net *net = seq_file_net(seq);
1024 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1027 if (!ip_vs_conn_net_eq(cp, net))
1031 len = strlen(cp->pe->name);
1032 memcpy(pe_data + 1, cp->pe->name, len);
1033 pe_data[len + 1] = ' ';
1035 len += cp->pe->show_pe_data(cp, pe_data + len);
1037 pe_data[len] = '\0';
1039 #ifdef CONFIG_IP_VS_IPV6
1040 if (cp->af == AF_INET6)
1041 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1042 "%pI6 %04X %-11s %7lu%s\n",
1043 ip_vs_proto_name(cp->protocol),
1044 &cp->caddr.in6, ntohs(cp->cport),
1045 &cp->vaddr.in6, ntohs(cp->vport),
1046 &cp->daddr.in6, ntohs(cp->dport),
1047 ip_vs_state_name(cp->protocol, cp->state),
1048 (cp->timer.expires-jiffies)/HZ, pe_data);
1052 "%-3s %08X %04X %08X %04X"
1053 " %08X %04X %-11s %7lu%s\n",
1054 ip_vs_proto_name(cp->protocol),
1055 ntohl(cp->caddr.ip), ntohs(cp->cport),
1056 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1057 ntohl(cp->daddr.ip), ntohs(cp->dport),
1058 ip_vs_state_name(cp->protocol, cp->state),
1059 (cp->timer.expires-jiffies)/HZ, pe_data);
1064 static const struct seq_operations ip_vs_conn_seq_ops = {
1065 .start = ip_vs_conn_seq_start,
1066 .next = ip_vs_conn_seq_next,
1067 .stop = ip_vs_conn_seq_stop,
1068 .show = ip_vs_conn_seq_show,
1071 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1073 return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1074 sizeof(struct ip_vs_iter_state));
1077 static const struct file_operations ip_vs_conn_fops = {
1078 .owner = THIS_MODULE,
1079 .open = ip_vs_conn_open,
1081 .llseek = seq_lseek,
1082 .release = seq_release_net,
1085 static const char *ip_vs_origin_name(unsigned int flags)
1087 if (flags & IP_VS_CONN_F_SYNC)
1093 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1096 if (v == SEQ_START_TOKEN)
1098 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1100 const struct ip_vs_conn *cp = v;
1101 struct net *net = seq_file_net(seq);
1103 if (!ip_vs_conn_net_eq(cp, net))
1106 #ifdef CONFIG_IP_VS_IPV6
1107 if (cp->af == AF_INET6)
1108 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1109 ip_vs_proto_name(cp->protocol),
1110 &cp->caddr.in6, ntohs(cp->cport),
1111 &cp->vaddr.in6, ntohs(cp->vport),
1112 &cp->daddr.in6, ntohs(cp->dport),
1113 ip_vs_state_name(cp->protocol, cp->state),
1114 ip_vs_origin_name(cp->flags),
1115 (cp->timer.expires-jiffies)/HZ);
1119 "%-3s %08X %04X %08X %04X "
1120 "%08X %04X %-11s %-6s %7lu\n",
1121 ip_vs_proto_name(cp->protocol),
1122 ntohl(cp->caddr.ip), ntohs(cp->cport),
1123 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1124 ntohl(cp->daddr.ip), ntohs(cp->dport),
1125 ip_vs_state_name(cp->protocol, cp->state),
1126 ip_vs_origin_name(cp->flags),
1127 (cp->timer.expires-jiffies)/HZ);
1132 static const struct seq_operations ip_vs_conn_sync_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_sync_seq_show,
1139 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1141 return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1142 sizeof(struct ip_vs_iter_state));
1145 static const struct file_operations ip_vs_conn_sync_fops = {
1146 .owner = THIS_MODULE,
1147 .open = ip_vs_conn_sync_open,
1149 .llseek = seq_lseek,
1150 .release = seq_release_net,
1157 * Randomly drop connection entries before running out of memory
1159 static inline int todrop_entry(struct ip_vs_conn *cp)
1162 * The drop rate array needs tuning for real environments.
1163 * Called from timer bh only => no locking
1165 static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1166 static char todrop_counter[9] = {0};
1169 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1170 This will leave enough time for normal connection to get
1172 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1175 /* Don't drop the entry if its number of incoming packets is not
1176 located in [0, 8] */
1177 i = atomic_read(&cp->in_pkts);
1178 if (i > 8 || i < 0) return 0;
1180 if (!todrop_rate[i]) return 0;
1181 if (--todrop_counter[i] > 0) return 0;
1183 todrop_counter[i] = todrop_rate[i];
1187 /* Called from keventd and must protect itself from softirqs */
1188 void ip_vs_random_dropentry(struct net *net)
1191 struct ip_vs_conn *cp;
1194 * Randomly scan 1/32 of the whole table every second
1196 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1197 unsigned int hash = net_random() & ip_vs_conn_tab_mask;
1200 * Lock is actually needed in this loop.
1202 ct_write_lock_bh(hash);
1204 hlist_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
1205 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1206 /* connection template */
1208 if (!ip_vs_conn_net_eq(cp, net))
1210 if (cp->protocol == IPPROTO_TCP) {
1212 case IP_VS_TCP_S_SYN_RECV:
1213 case IP_VS_TCP_S_SYNACK:
1216 case IP_VS_TCP_S_ESTABLISHED:
1217 if (todrop_entry(cp))
1225 if (!todrop_entry(cp))
1229 IP_VS_DBG(4, "del connection\n");
1230 ip_vs_conn_expire_now(cp);
1232 IP_VS_DBG(4, "del conn template\n");
1233 ip_vs_conn_expire_now(cp->control);
1236 ct_write_unlock_bh(hash);
1242 * Flush all the connection entries in the ip_vs_conn_tab
1244 static void ip_vs_conn_flush(struct net *net)
1247 struct ip_vs_conn *cp;
1248 struct netns_ipvs *ipvs = net_ipvs(net);
1251 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1253 * Lock is actually needed in this loop.
1255 ct_write_lock_bh(idx);
1257 hlist_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1258 if (!ip_vs_conn_net_eq(cp, net))
1260 IP_VS_DBG(4, "del connection\n");
1261 ip_vs_conn_expire_now(cp);
1263 IP_VS_DBG(4, "del conn template\n");
1264 ip_vs_conn_expire_now(cp->control);
1267 ct_write_unlock_bh(idx);
1270 /* the counter may be not NULL, because maybe some conn entries
1271 are run by slow timer handler or unhashed but still referred */
1272 if (atomic_read(&ipvs->conn_count) != 0) {
1278 * per netns init and exit
1280 int __net_init ip_vs_conn_net_init(struct net *net)
1282 struct netns_ipvs *ipvs = net_ipvs(net);
1284 atomic_set(&ipvs->conn_count, 0);
1286 proc_create("ip_vs_conn", 0, net->proc_net, &ip_vs_conn_fops);
1287 proc_create("ip_vs_conn_sync", 0, net->proc_net, &ip_vs_conn_sync_fops);
1291 void __net_exit ip_vs_conn_net_cleanup(struct net *net)
1293 /* flush all the connection entries first */
1294 ip_vs_conn_flush(net);
1295 remove_proc_entry("ip_vs_conn", net->proc_net);
1296 remove_proc_entry("ip_vs_conn_sync", net->proc_net);
1299 int __init ip_vs_conn_init(void)
1303 /* Compute size and mask */
1304 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1305 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1308 * Allocate the connection hash table and initialize its list heads
1310 ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1311 if (!ip_vs_conn_tab)
1314 /* Allocate ip_vs_conn slab cache */
1315 ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1316 sizeof(struct ip_vs_conn), 0,
1317 SLAB_HWCACHE_ALIGN, NULL);
1318 if (!ip_vs_conn_cachep) {
1319 vfree(ip_vs_conn_tab);
1323 pr_info("Connection hash table configured "
1324 "(size=%d, memory=%ldKbytes)\n",
1325 ip_vs_conn_tab_size,
1326 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1327 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1328 sizeof(struct ip_vs_conn));
1330 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1331 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1333 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1334 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1337 /* calculate the random value for connection hash */
1338 get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1343 void ip_vs_conn_cleanup(void)
1345 /* Release the empty cache */
1346 kmem_cache_destroy(ip_vs_conn_cachep);
1347 vfree(ip_vs_conn_tab);