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 key)
91 read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
94 static inline void ct_read_unlock(unsigned key)
96 read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
99 static inline void ct_write_lock(unsigned key)
101 write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
104 static inline void ct_write_unlock(unsigned key)
106 write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
109 static inline void ct_read_lock_bh(unsigned key)
111 read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
114 static inline void ct_read_unlock_bh(unsigned key)
116 read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
119 static inline void ct_write_lock_bh(unsigned key)
121 write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
124 static inline void ct_write_unlock_bh(unsigned 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 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 unsigned int proto_off, int inverse,
312 struct ip_vs_conn_param *p)
314 __be16 _ports[2], *pptr;
315 struct net *net = skb_net(skb);
317 pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
321 if (likely(!inverse))
322 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->saddr,
323 pptr[0], &iph->daddr, pptr[1], p);
325 ip_vs_conn_fill_param(net, af, iph->protocol, &iph->daddr,
326 pptr[1], &iph->saddr, pptr[0], p);
331 ip_vs_conn_in_get_proto(int af, const struct sk_buff *skb,
332 const struct ip_vs_iphdr *iph,
333 unsigned int proto_off, int inverse)
335 struct ip_vs_conn_param p;
337 if (ip_vs_conn_fill_param_proto(af, skb, iph, proto_off, inverse, &p))
340 return ip_vs_conn_in_get(&p);
342 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
344 /* Get reference to connection template */
345 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
348 struct ip_vs_conn *cp;
349 struct hlist_node *n;
351 hash = ip_vs_conn_hashkey_param(p, false);
355 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
356 if (!ip_vs_conn_net_eq(cp, p->net))
358 if (p->pe_data && p->pe->ct_match) {
359 if (p->pe == cp->pe && p->pe->ct_match(p, cp))
364 if (cp->af == p->af &&
365 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
366 /* protocol should only be IPPROTO_IP if
367 * p->vaddr is a fwmark */
368 ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
369 p->af, p->vaddr, &cp->vaddr) &&
370 p->cport == cp->cport && p->vport == cp->vport &&
371 cp->flags & IP_VS_CONN_F_TEMPLATE &&
372 p->protocol == cp->protocol)
379 atomic_inc(&cp->refcnt);
380 ct_read_unlock(hash);
382 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
383 ip_vs_proto_name(p->protocol),
384 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
385 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
386 cp ? "hit" : "not hit");
391 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
392 * Called for pkts coming from inside-to-OUTside.
393 * p->caddr, p->cport: pkt source address (inside host)
394 * p->vaddr, p->vport: pkt dest address (foreign host) */
395 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
398 struct ip_vs_conn *cp, *ret=NULL;
399 struct hlist_node *n;
402 * Check for "full" addressed entries
404 hash = ip_vs_conn_hashkey_param(p, true);
408 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
409 if (cp->af == p->af &&
410 p->vport == cp->cport && p->cport == cp->dport &&
411 ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
412 ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
413 p->protocol == cp->protocol &&
414 ip_vs_conn_net_eq(cp, p->net)) {
416 atomic_inc(&cp->refcnt);
422 ct_read_unlock(hash);
424 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
425 ip_vs_proto_name(p->protocol),
426 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
427 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
428 ret ? "hit" : "not hit");
434 ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
435 const struct ip_vs_iphdr *iph,
436 unsigned int proto_off, int inverse)
438 struct ip_vs_conn_param p;
440 if (ip_vs_conn_fill_param_proto(af, skb, iph, proto_off, inverse, &p))
443 return ip_vs_conn_out_get(&p);
445 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
448 * Put back the conn and restart its timer with its timeout
450 void ip_vs_conn_put(struct ip_vs_conn *cp)
452 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
454 mod_timer(&cp->timer, jiffies+t);
456 __ip_vs_conn_put(cp);
461 * Fill a no_client_port connection with a client port number
463 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
465 if (ip_vs_conn_unhash(cp)) {
466 spin_lock(&cp->lock);
467 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
468 atomic_dec(&ip_vs_conn_no_cport_cnt);
469 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
472 spin_unlock(&cp->lock);
474 /* hash on new dport */
481 * Bind a connection entry with the corresponding packet_xmit.
482 * Called by ip_vs_conn_new.
484 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
486 switch (IP_VS_FWD_METHOD(cp)) {
487 case IP_VS_CONN_F_MASQ:
488 cp->packet_xmit = ip_vs_nat_xmit;
491 case IP_VS_CONN_F_TUNNEL:
492 cp->packet_xmit = ip_vs_tunnel_xmit;
495 case IP_VS_CONN_F_DROUTE:
496 cp->packet_xmit = ip_vs_dr_xmit;
499 case IP_VS_CONN_F_LOCALNODE:
500 cp->packet_xmit = ip_vs_null_xmit;
503 case IP_VS_CONN_F_BYPASS:
504 cp->packet_xmit = ip_vs_bypass_xmit;
509 #ifdef CONFIG_IP_VS_IPV6
510 static inline void ip_vs_bind_xmit_v6(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_v6;
517 case IP_VS_CONN_F_TUNNEL:
518 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
521 case IP_VS_CONN_F_DROUTE:
522 cp->packet_xmit = ip_vs_dr_xmit_v6;
525 case IP_VS_CONN_F_LOCALNODE:
526 cp->packet_xmit = ip_vs_null_xmit;
529 case IP_VS_CONN_F_BYPASS:
530 cp->packet_xmit = ip_vs_bypass_xmit_v6;
537 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
539 return atomic_read(&dest->activeconns)
540 + atomic_read(&dest->inactconns);
544 * Bind a connection entry with a virtual service destination
545 * Called just after a new connection entry is created.
548 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
550 unsigned int conn_flags;
552 /* if dest is NULL, then return directly */
556 /* Increase the refcnt counter of the dest */
557 atomic_inc(&dest->refcnt);
559 conn_flags = atomic_read(&dest->conn_flags);
560 if (cp->protocol != IPPROTO_UDP)
561 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
562 /* Bind with the destination and its corresponding transmitter */
563 if (cp->flags & IP_VS_CONN_F_SYNC) {
564 /* if the connection is not template and is created
565 * by sync, preserve the activity flag.
567 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE))
568 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
569 /* connections inherit forwarding method from dest */
570 cp->flags &= ~IP_VS_CONN_F_FWD_MASK;
572 cp->flags |= conn_flags;
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 (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
588 /* It is a normal connection, so increase the inactive
589 connection counter because it is in TCP SYNRECV
590 state (inactive) or other protocol inacive state */
591 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
592 (!(cp->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 if ((cp) && (!cp->dest)) {
617 dest = ip_vs_find_dest(ip_vs_conn_net(cp), cp->af, &cp->daddr,
618 cp->dport, &cp->vaddr, cp->vport,
619 cp->protocol, cp->fwmark, cp->flags);
620 ip_vs_bind_dest(cp, dest);
628 * Unbind a connection entry with its VS destination
629 * Called by the ip_vs_conn_expire function.
631 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
633 struct ip_vs_dest *dest = cp->dest;
638 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
639 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
641 ip_vs_proto_name(cp->protocol),
642 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
643 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
644 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
645 ip_vs_fwd_tag(cp), cp->state,
646 cp->flags, atomic_read(&cp->refcnt),
647 atomic_read(&dest->refcnt));
649 /* Update the connection counters */
650 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
651 /* It is a normal connection, so decrease the inactconns
652 or activeconns counter */
653 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
654 atomic_dec(&dest->inactconns);
656 atomic_dec(&dest->activeconns);
659 /* It is a persistent connection/template, so decrease
660 the persistent connection counter */
661 atomic_dec(&dest->persistconns);
664 if (dest->l_threshold != 0) {
665 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
666 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
667 } else if (dest->u_threshold != 0) {
668 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
669 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
671 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
672 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
676 * Simply decrease the refcnt of the dest, because the
677 * dest will be either in service's destination list
680 atomic_dec(&dest->refcnt);
683 static int expire_quiescent_template(struct netns_ipvs *ipvs,
684 struct ip_vs_dest *dest)
687 return ipvs->sysctl_expire_quiescent_template &&
688 (atomic_read(&dest->weight) == 0);
695 * Checking if the destination of a connection template is available.
696 * If available, return 1, otherwise invalidate this connection
697 * template and return 0.
699 int ip_vs_check_template(struct ip_vs_conn *ct)
701 struct ip_vs_dest *dest = ct->dest;
702 struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(ct));
705 * Checking the dest server status.
707 if ((dest == NULL) ||
708 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
709 expire_quiescent_template(ipvs, dest)) {
710 IP_VS_DBG_BUF(9, "check_template: dest not available for "
711 "protocol %s s:%s:%d v:%s:%d "
713 ip_vs_proto_name(ct->protocol),
714 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
716 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
718 IP_VS_DBG_ADDR(ct->af, &ct->daddr),
722 * Invalidate the connection template
724 if (ct->vport != htons(0xffff)) {
725 if (ip_vs_conn_unhash(ct)) {
726 ct->dport = htons(0xffff);
727 ct->vport = htons(0xffff);
734 * Simply decrease the refcnt of the template,
735 * don't restart its timer.
737 atomic_dec(&ct->refcnt);
743 static void ip_vs_conn_expire(unsigned long data)
745 struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
746 struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(cp));
753 atomic_inc(&cp->refcnt);
756 * do I control anybody?
758 if (atomic_read(&cp->n_control))
762 * unhash it if it is hashed in the conn table
764 if (!ip_vs_conn_unhash(cp) && !(cp->flags & IP_VS_CONN_F_ONE_PACKET))
768 * refcnt==1 implies I'm the only one referrer
770 if (likely(atomic_read(&cp->refcnt) == 1)) {
771 /* delete the timer if it is activated by other users */
772 if (timer_pending(&cp->timer))
773 del_timer(&cp->timer);
775 /* does anybody control me? */
777 ip_vs_control_del(cp);
779 if (cp->flags & IP_VS_CONN_F_NFCT) {
780 /* Do not access conntracks during subsys cleanup
781 * because nf_conntrack_find_get can not be used after
782 * conntrack cleanup for the net.
786 ip_vs_conn_drop_conntrack(cp);
789 ip_vs_pe_put(cp->pe);
791 if (unlikely(cp->app != NULL))
792 ip_vs_unbind_app(cp);
793 ip_vs_unbind_dest(cp);
794 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
795 atomic_dec(&ip_vs_conn_no_cport_cnt);
796 atomic_dec(&ipvs->conn_count);
798 kmem_cache_free(ip_vs_conn_cachep, cp);
802 /* hash it back to the table */
806 IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
807 atomic_read(&cp->refcnt)-1,
808 atomic_read(&cp->n_control));
814 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
816 if (del_timer(&cp->timer))
817 mod_timer(&cp->timer, jiffies);
822 * Create a new connection entry and hash it into the ip_vs_conn_tab
825 ip_vs_conn_new(const struct ip_vs_conn_param *p,
826 const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
827 struct ip_vs_dest *dest, __u32 fwmark)
829 struct ip_vs_conn *cp;
830 struct netns_ipvs *ipvs = net_ipvs(p->net);
831 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->net,
834 cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
836 IP_VS_ERR_RL("%s(): no memory\n", __func__);
840 INIT_HLIST_NODE(&cp->c_list);
841 setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
842 ip_vs_conn_net_set(cp, p->net);
844 cp->protocol = p->protocol;
845 ip_vs_addr_copy(p->af, &cp->caddr, p->caddr);
846 cp->cport = p->cport;
847 ip_vs_addr_copy(p->af, &cp->vaddr, p->vaddr);
848 cp->vport = p->vport;
849 /* proto should only be IPPROTO_IP if d_addr is a fwmark */
850 ip_vs_addr_copy(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
855 if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
858 cp->pe_data = p->pe_data;
859 cp->pe_data_len = p->pe_data_len;
861 spin_lock_init(&cp->lock);
864 * Set the entry is referenced by the current thread before hashing
865 * it in the table, so that other thread run ip_vs_random_dropentry
866 * but cannot drop this entry.
868 atomic_set(&cp->refcnt, 1);
870 atomic_set(&cp->n_control, 0);
871 atomic_set(&cp->in_pkts, 0);
873 atomic_inc(&ipvs->conn_count);
874 if (flags & IP_VS_CONN_F_NO_CPORT)
875 atomic_inc(&ip_vs_conn_no_cport_cnt);
877 /* Bind the connection with a destination server */
878 ip_vs_bind_dest(cp, dest);
880 /* Set its state and timeout */
884 /* Bind its packet transmitter */
885 #ifdef CONFIG_IP_VS_IPV6
886 if (p->af == AF_INET6)
887 ip_vs_bind_xmit_v6(cp);
892 if (unlikely(pd && atomic_read(&pd->appcnt)))
893 ip_vs_bind_app(cp, pd->pp);
896 * Allow conntrack to be preserved. By default, conntrack
897 * is created and destroyed for every packet.
898 * Sometimes keeping conntrack can be useful for
899 * IP_VS_CONN_F_ONE_PACKET too.
902 if (ip_vs_conntrack_enabled(ipvs))
903 cp->flags |= IP_VS_CONN_F_NFCT;
905 /* Hash it in the ip_vs_conn_tab finally */
912 * /proc/net/ip_vs_conn entries
914 #ifdef CONFIG_PROC_FS
915 struct ip_vs_iter_state {
916 struct seq_net_private p;
917 struct hlist_head *l;
920 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
923 struct ip_vs_conn *cp;
924 struct ip_vs_iter_state *iter = seq->private;
925 struct hlist_node *n;
927 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
928 ct_read_lock_bh(idx);
929 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[idx], c_list) {
931 iter->l = &ip_vs_conn_tab[idx];
935 ct_read_unlock_bh(idx);
941 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
943 struct ip_vs_iter_state *iter = seq->private;
946 return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
949 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
951 struct ip_vs_conn *cp = v;
952 struct ip_vs_iter_state *iter = seq->private;
953 struct hlist_node *e;
954 struct hlist_head *l = iter->l;
958 if (v == SEQ_START_TOKEN)
959 return ip_vs_conn_array(seq, 0);
961 /* more on same hash chain? */
962 if ((e = cp->c_list.next))
963 return hlist_entry(e, struct ip_vs_conn, c_list);
965 idx = l - ip_vs_conn_tab;
966 ct_read_unlock_bh(idx);
968 while (++idx < ip_vs_conn_tab_size) {
969 ct_read_lock_bh(idx);
970 hlist_for_each_entry(cp, e, &ip_vs_conn_tab[idx], c_list) {
971 iter->l = &ip_vs_conn_tab[idx];
974 ct_read_unlock_bh(idx);
980 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
982 struct ip_vs_iter_state *iter = seq->private;
983 struct hlist_head *l = iter->l;
986 ct_read_unlock_bh(l - ip_vs_conn_tab);
989 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
992 if (v == SEQ_START_TOKEN)
994 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
996 const struct ip_vs_conn *cp = v;
997 struct net *net = seq_file_net(seq);
998 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1001 if (!ip_vs_conn_net_eq(cp, net))
1005 len = strlen(cp->pe->name);
1006 memcpy(pe_data + 1, cp->pe->name, len);
1007 pe_data[len + 1] = ' ';
1009 len += cp->pe->show_pe_data(cp, pe_data + len);
1011 pe_data[len] = '\0';
1013 #ifdef CONFIG_IP_VS_IPV6
1014 if (cp->af == AF_INET6)
1015 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1016 "%pI6 %04X %-11s %7lu%s\n",
1017 ip_vs_proto_name(cp->protocol),
1018 &cp->caddr.in6, ntohs(cp->cport),
1019 &cp->vaddr.in6, ntohs(cp->vport),
1020 &cp->daddr.in6, ntohs(cp->dport),
1021 ip_vs_state_name(cp->protocol, cp->state),
1022 (cp->timer.expires-jiffies)/HZ, pe_data);
1026 "%-3s %08X %04X %08X %04X"
1027 " %08X %04X %-11s %7lu%s\n",
1028 ip_vs_proto_name(cp->protocol),
1029 ntohl(cp->caddr.ip), ntohs(cp->cport),
1030 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1031 ntohl(cp->daddr.ip), ntohs(cp->dport),
1032 ip_vs_state_name(cp->protocol, cp->state),
1033 (cp->timer.expires-jiffies)/HZ, pe_data);
1038 static const struct seq_operations ip_vs_conn_seq_ops = {
1039 .start = ip_vs_conn_seq_start,
1040 .next = ip_vs_conn_seq_next,
1041 .stop = ip_vs_conn_seq_stop,
1042 .show = ip_vs_conn_seq_show,
1045 static int ip_vs_conn_open(struct inode *inode, struct file *file)
1047 return seq_open_net(inode, file, &ip_vs_conn_seq_ops,
1048 sizeof(struct ip_vs_iter_state));
1051 static const struct file_operations ip_vs_conn_fops = {
1052 .owner = THIS_MODULE,
1053 .open = ip_vs_conn_open,
1055 .llseek = seq_lseek,
1056 .release = seq_release_net,
1059 static const char *ip_vs_origin_name(unsigned flags)
1061 if (flags & IP_VS_CONN_F_SYNC)
1067 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1070 if (v == SEQ_START_TOKEN)
1072 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1074 const struct ip_vs_conn *cp = v;
1075 struct net *net = seq_file_net(seq);
1077 if (!ip_vs_conn_net_eq(cp, net))
1080 #ifdef CONFIG_IP_VS_IPV6
1081 if (cp->af == AF_INET6)
1082 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
1083 ip_vs_proto_name(cp->protocol),
1084 &cp->caddr.in6, ntohs(cp->cport),
1085 &cp->vaddr.in6, ntohs(cp->vport),
1086 &cp->daddr.in6, ntohs(cp->dport),
1087 ip_vs_state_name(cp->protocol, cp->state),
1088 ip_vs_origin_name(cp->flags),
1089 (cp->timer.expires-jiffies)/HZ);
1093 "%-3s %08X %04X %08X %04X "
1094 "%08X %04X %-11s %-6s %7lu\n",
1095 ip_vs_proto_name(cp->protocol),
1096 ntohl(cp->caddr.ip), ntohs(cp->cport),
1097 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1098 ntohl(cp->daddr.ip), ntohs(cp->dport),
1099 ip_vs_state_name(cp->protocol, cp->state),
1100 ip_vs_origin_name(cp->flags),
1101 (cp->timer.expires-jiffies)/HZ);
1106 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1107 .start = ip_vs_conn_seq_start,
1108 .next = ip_vs_conn_seq_next,
1109 .stop = ip_vs_conn_seq_stop,
1110 .show = ip_vs_conn_sync_seq_show,
1113 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
1115 return seq_open_net(inode, file, &ip_vs_conn_sync_seq_ops,
1116 sizeof(struct ip_vs_iter_state));
1119 static const struct file_operations ip_vs_conn_sync_fops = {
1120 .owner = THIS_MODULE,
1121 .open = ip_vs_conn_sync_open,
1123 .llseek = seq_lseek,
1124 .release = seq_release_net,
1131 * Randomly drop connection entries before running out of memory
1133 static inline int todrop_entry(struct ip_vs_conn *cp)
1136 * The drop rate array needs tuning for real environments.
1137 * Called from timer bh only => no locking
1139 static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1140 static char todrop_counter[9] = {0};
1143 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1144 This will leave enough time for normal connection to get
1146 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1149 /* Don't drop the entry if its number of incoming packets is not
1150 located in [0, 8] */
1151 i = atomic_read(&cp->in_pkts);
1152 if (i > 8 || i < 0) return 0;
1154 if (!todrop_rate[i]) return 0;
1155 if (--todrop_counter[i] > 0) return 0;
1157 todrop_counter[i] = todrop_rate[i];
1161 /* Called from keventd and must protect itself from softirqs */
1162 void ip_vs_random_dropentry(struct net *net)
1165 struct ip_vs_conn *cp;
1168 * Randomly scan 1/32 of the whole table every second
1170 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1171 unsigned hash = net_random() & ip_vs_conn_tab_mask;
1172 struct hlist_node *n;
1175 * Lock is actually needed in this loop.
1177 ct_write_lock_bh(hash);
1179 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[hash], c_list) {
1180 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1181 /* connection template */
1183 if (!ip_vs_conn_net_eq(cp, net))
1185 if (cp->protocol == IPPROTO_TCP) {
1187 case IP_VS_TCP_S_SYN_RECV:
1188 case IP_VS_TCP_S_SYNACK:
1191 case IP_VS_TCP_S_ESTABLISHED:
1192 if (todrop_entry(cp))
1200 if (!todrop_entry(cp))
1204 IP_VS_DBG(4, "del connection\n");
1205 ip_vs_conn_expire_now(cp);
1207 IP_VS_DBG(4, "del conn template\n");
1208 ip_vs_conn_expire_now(cp->control);
1211 ct_write_unlock_bh(hash);
1217 * Flush all the connection entries in the ip_vs_conn_tab
1219 static void ip_vs_conn_flush(struct net *net)
1222 struct ip_vs_conn *cp;
1223 struct netns_ipvs *ipvs = net_ipvs(net);
1226 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1227 struct hlist_node *n;
1230 * Lock is actually needed in this loop.
1232 ct_write_lock_bh(idx);
1234 hlist_for_each_entry(cp, n, &ip_vs_conn_tab[idx], c_list) {
1235 if (!ip_vs_conn_net_eq(cp, net))
1237 IP_VS_DBG(4, "del connection\n");
1238 ip_vs_conn_expire_now(cp);
1240 IP_VS_DBG(4, "del conn template\n");
1241 ip_vs_conn_expire_now(cp->control);
1244 ct_write_unlock_bh(idx);
1247 /* the counter may be not NULL, because maybe some conn entries
1248 are run by slow timer handler or unhashed but still referred */
1249 if (atomic_read(&ipvs->conn_count) != 0) {
1255 * per netns init and exit
1257 int __net_init ip_vs_conn_net_init(struct net *net)
1259 struct netns_ipvs *ipvs = net_ipvs(net);
1261 atomic_set(&ipvs->conn_count, 0);
1263 proc_net_fops_create(net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1264 proc_net_fops_create(net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1268 void __net_exit ip_vs_conn_net_cleanup(struct net *net)
1270 /* flush all the connection entries first */
1271 ip_vs_conn_flush(net);
1272 proc_net_remove(net, "ip_vs_conn");
1273 proc_net_remove(net, "ip_vs_conn_sync");
1276 int __init ip_vs_conn_init(void)
1280 /* Compute size and mask */
1281 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1282 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1285 * Allocate the connection hash table and initialize its list heads
1287 ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size * sizeof(*ip_vs_conn_tab));
1288 if (!ip_vs_conn_tab)
1291 /* Allocate ip_vs_conn slab cache */
1292 ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1293 sizeof(struct ip_vs_conn), 0,
1294 SLAB_HWCACHE_ALIGN, NULL);
1295 if (!ip_vs_conn_cachep) {
1296 vfree(ip_vs_conn_tab);
1300 pr_info("Connection hash table configured "
1301 "(size=%d, memory=%ldKbytes)\n",
1302 ip_vs_conn_tab_size,
1303 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1304 IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1305 sizeof(struct ip_vs_conn));
1307 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1308 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1310 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1311 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1314 /* calculate the random value for connection hash */
1315 get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1320 void ip_vs_conn_cleanup(void)
1322 /* Release the empty cache */
1323 kmem_cache_destroy(ip_vs_conn_cachep);
1324 vfree(ip_vs_conn_tab);