2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the Netfilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
19 * and others. Many code here is taken from IP MASQ code of kernel 2.2.
25 #define KMSG_COMPONENT "IPVS"
26 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
28 #include <linux/interrupt.h>
30 #include <linux/inet.h>
31 #include <linux/net.h>
32 #include <linux/kernel.h>
33 #include <linux/module.h>
34 #include <linux/vmalloc.h>
35 #include <linux/proc_fs.h> /* for proc_net_* */
36 #include <linux/slab.h>
37 #include <linux/seq_file.h>
38 #include <linux/jhash.h>
39 #include <linux/random.h>
41 #include <net/net_namespace.h>
42 #include <net/ip_vs.h>
45 #ifndef CONFIG_IP_VS_TAB_BITS
46 #define CONFIG_IP_VS_TAB_BITS 12
50 * Connection hash size. Default is what was selected at compile time.
52 static int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
53 module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
54 MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
56 /* size and mask values */
57 int ip_vs_conn_tab_size __read_mostly;
58 static int ip_vs_conn_tab_mask __read_mostly;
61 * Connection hash table: for input and output packets lookups of IPVS
63 static struct hlist_head *ip_vs_conn_tab __read_mostly;
65 /* SLAB cache for IPVS connections */
66 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
68 /* counter for no client port connections */
69 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
71 /* random value for IPVS connection hash */
72 static unsigned int ip_vs_conn_rnd __read_mostly;
75 * Fine locking granularity for big connection hash table
77 #define CT_LOCKARRAY_BITS 5
78 #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
79 #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
81 /* We need an addrstrlen that works with or without v6 */
82 #ifdef CONFIG_IP_VS_IPV6
83 #define IP_VS_ADDRSTRLEN INET6_ADDRSTRLEN
85 #define IP_VS_ADDRSTRLEN (8+1)
88 struct ip_vs_aligned_lock
91 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
93 /* lock array for conn table */
94 static struct ip_vs_aligned_lock
95 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
97 static inline void ct_write_lock_bh(unsigned int key)
99 spin_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
102 static inline void ct_write_unlock_bh(unsigned int key)
104 spin_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
107 static void ip_vs_conn_expire(struct timer_list *t);
110 * Returns hash value for IPVS connection entry
112 static unsigned int ip_vs_conn_hashkey(struct netns_ipvs *ipvs, int af, unsigned int proto,
113 const union nf_inet_addr *addr,
116 #ifdef CONFIG_IP_VS_IPV6
118 return (jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
119 (__force u32)port, proto, ip_vs_conn_rnd) ^
120 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
122 return (jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
124 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
127 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
130 const union nf_inet_addr *addr;
133 if (p->pe_data && p->pe->hashkey_raw)
134 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
137 if (likely(!inverse)) {
145 return ip_vs_conn_hashkey(p->ipvs, p->af, p->protocol, addr, port);
148 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
150 struct ip_vs_conn_param p;
152 ip_vs_conn_fill_param(cp->ipvs, cp->af, cp->protocol,
153 &cp->caddr, cp->cport, NULL, 0, &p);
157 p.pe_data = cp->pe_data;
158 p.pe_data_len = cp->pe_data_len;
161 return ip_vs_conn_hashkey_param(&p, false);
165 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
166 * returns bool success.
168 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
173 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
176 /* Hash by protocol, client address and port */
177 hash = ip_vs_conn_hashkey_conn(cp);
179 ct_write_lock_bh(hash);
180 spin_lock(&cp->lock);
182 if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
183 cp->flags |= IP_VS_CONN_F_HASHED;
184 refcount_inc(&cp->refcnt);
185 hlist_add_head_rcu(&cp->c_list, &ip_vs_conn_tab[hash]);
188 pr_err("%s(): request for already hashed, called from %pS\n",
189 __func__, __builtin_return_address(0));
193 spin_unlock(&cp->lock);
194 ct_write_unlock_bh(hash);
201 * UNhashes ip_vs_conn from ip_vs_conn_tab.
202 * returns bool success. Caller should hold conn reference.
204 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
209 /* unhash it and decrease its reference counter */
210 hash = ip_vs_conn_hashkey_conn(cp);
212 ct_write_lock_bh(hash);
213 spin_lock(&cp->lock);
215 if (cp->flags & IP_VS_CONN_F_HASHED) {
216 hlist_del_rcu(&cp->c_list);
217 cp->flags &= ~IP_VS_CONN_F_HASHED;
218 refcount_dec(&cp->refcnt);
223 spin_unlock(&cp->lock);
224 ct_write_unlock_bh(hash);
229 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
230 * returns bool success.
232 static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
237 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
238 return refcount_dec_if_one(&cp->refcnt);
240 hash = ip_vs_conn_hashkey_conn(cp);
242 ct_write_lock_bh(hash);
243 spin_lock(&cp->lock);
245 if (cp->flags & IP_VS_CONN_F_HASHED) {
246 /* Decrease refcnt and unlink conn only if we are last user */
247 if (refcount_dec_if_one(&cp->refcnt)) {
248 hlist_del_rcu(&cp->c_list);
249 cp->flags &= ~IP_VS_CONN_F_HASHED;
254 spin_unlock(&cp->lock);
255 ct_write_unlock_bh(hash);
262 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
263 * Called for pkts coming from OUTside-to-INside.
264 * p->caddr, p->cport: pkt source address (foreign host)
265 * p->vaddr, p->vport: pkt dest address (load balancer)
267 static inline struct ip_vs_conn *
268 __ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
271 struct ip_vs_conn *cp;
273 hash = ip_vs_conn_hashkey_param(p, false);
277 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
278 if (p->cport == cp->cport && p->vport == cp->vport &&
280 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
281 ip_vs_addr_equal(p->af, p->vaddr, &cp->vaddr) &&
282 ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
283 p->protocol == cp->protocol &&
284 cp->ipvs == p->ipvs) {
285 if (!__ip_vs_conn_get(cp))
298 struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
300 struct ip_vs_conn *cp;
302 cp = __ip_vs_conn_in_get(p);
303 if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) {
304 struct ip_vs_conn_param cport_zero_p = *p;
305 cport_zero_p.cport = 0;
306 cp = __ip_vs_conn_in_get(&cport_zero_p);
309 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
310 ip_vs_proto_name(p->protocol),
311 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
312 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
313 cp ? "hit" : "not hit");
319 ip_vs_conn_fill_param_proto(struct netns_ipvs *ipvs,
320 int af, const struct sk_buff *skb,
321 const struct ip_vs_iphdr *iph,
322 struct ip_vs_conn_param *p)
324 __be16 _ports[2], *pptr;
326 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
330 if (likely(!ip_vs_iph_inverse(iph)))
331 ip_vs_conn_fill_param(ipvs, af, iph->protocol, &iph->saddr,
332 pptr[0], &iph->daddr, pptr[1], p);
334 ip_vs_conn_fill_param(ipvs, af, iph->protocol, &iph->daddr,
335 pptr[1], &iph->saddr, pptr[0], p);
340 ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af,
341 const struct sk_buff *skb,
342 const struct ip_vs_iphdr *iph)
344 struct ip_vs_conn_param p;
346 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, &p))
349 return ip_vs_conn_in_get(&p);
351 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
353 /* Get reference to connection template */
354 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
357 struct ip_vs_conn *cp;
359 hash = ip_vs_conn_hashkey_param(p, false);
363 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
364 if (unlikely(p->pe_data && p->pe->ct_match)) {
365 if (cp->ipvs != p->ipvs)
367 if (p->pe == cp->pe && p->pe->ct_match(p, cp)) {
368 if (__ip_vs_conn_get(cp))
374 if (cp->af == p->af &&
375 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
376 /* protocol should only be IPPROTO_IP if
377 * p->vaddr is a fwmark */
378 ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
379 p->af, p->vaddr, &cp->vaddr) &&
380 p->vport == cp->vport && p->cport == cp->cport &&
381 cp->flags & IP_VS_CONN_F_TEMPLATE &&
382 p->protocol == cp->protocol &&
383 cp->ipvs == p->ipvs) {
384 if (__ip_vs_conn_get(cp))
393 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
394 ip_vs_proto_name(p->protocol),
395 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
396 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
397 cp ? "hit" : "not hit");
402 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
403 * Called for pkts coming from inside-to-OUTside.
404 * p->caddr, p->cport: pkt source address (inside host)
405 * p->vaddr, p->vport: pkt dest address (foreign host) */
406 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
409 struct ip_vs_conn *cp, *ret=NULL;
412 * Check for "full" addressed entries
414 hash = ip_vs_conn_hashkey_param(p, true);
418 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
419 if (p->vport == cp->cport && p->cport == cp->dport &&
421 ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
422 ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
423 p->protocol == cp->protocol &&
424 cp->ipvs == p->ipvs) {
425 if (!__ip_vs_conn_get(cp))
435 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
436 ip_vs_proto_name(p->protocol),
437 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
438 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
439 ret ? "hit" : "not hit");
445 ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af,
446 const struct sk_buff *skb,
447 const struct ip_vs_iphdr *iph)
449 struct ip_vs_conn_param p;
451 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, &p))
454 return ip_vs_conn_out_get(&p);
456 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
459 * Put back the conn and restart its timer with its timeout
461 static void __ip_vs_conn_put_timer(struct ip_vs_conn *cp)
463 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
465 mod_timer(&cp->timer, jiffies+t);
467 __ip_vs_conn_put(cp);
470 void ip_vs_conn_put(struct ip_vs_conn *cp)
472 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) &&
473 (refcount_read(&cp->refcnt) == 1) &&
474 !timer_pending(&cp->timer))
475 /* expire connection immediately */
476 ip_vs_conn_expire(&cp->timer);
478 __ip_vs_conn_put_timer(cp);
482 * Fill a no_client_port connection with a client port number
484 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
486 if (ip_vs_conn_unhash(cp)) {
487 spin_lock_bh(&cp->lock);
488 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
489 atomic_dec(&ip_vs_conn_no_cport_cnt);
490 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
493 spin_unlock_bh(&cp->lock);
495 /* hash on new dport */
502 * Bind a connection entry with the corresponding packet_xmit.
503 * Called by ip_vs_conn_new.
505 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
507 switch (IP_VS_FWD_METHOD(cp)) {
508 case IP_VS_CONN_F_MASQ:
509 cp->packet_xmit = ip_vs_nat_xmit;
512 case IP_VS_CONN_F_TUNNEL:
513 #ifdef CONFIG_IP_VS_IPV6
514 if (cp->daf == AF_INET6)
515 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
518 cp->packet_xmit = ip_vs_tunnel_xmit;
521 case IP_VS_CONN_F_DROUTE:
522 cp->packet_xmit = ip_vs_dr_xmit;
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;
535 #ifdef CONFIG_IP_VS_IPV6
536 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
538 switch (IP_VS_FWD_METHOD(cp)) {
539 case IP_VS_CONN_F_MASQ:
540 cp->packet_xmit = ip_vs_nat_xmit_v6;
543 case IP_VS_CONN_F_TUNNEL:
544 if (cp->daf == AF_INET6)
545 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
547 cp->packet_xmit = ip_vs_tunnel_xmit;
550 case IP_VS_CONN_F_DROUTE:
551 cp->packet_xmit = ip_vs_dr_xmit_v6;
554 case IP_VS_CONN_F_LOCALNODE:
555 cp->packet_xmit = ip_vs_null_xmit;
558 case IP_VS_CONN_F_BYPASS:
559 cp->packet_xmit = ip_vs_bypass_xmit_v6;
566 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
568 return atomic_read(&dest->activeconns)
569 + atomic_read(&dest->inactconns);
573 * Bind a connection entry with a virtual service destination
574 * Called just after a new connection entry is created.
577 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
579 unsigned int conn_flags;
582 /* if dest is NULL, then return directly */
586 /* Increase the refcnt counter of the dest */
587 ip_vs_dest_hold(dest);
589 conn_flags = atomic_read(&dest->conn_flags);
590 if (cp->protocol != IPPROTO_UDP)
591 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
593 /* Bind with the destination and its corresponding transmitter */
594 if (flags & IP_VS_CONN_F_SYNC) {
595 /* if the connection is not template and is created
596 * by sync, preserve the activity flag.
598 if (!(flags & IP_VS_CONN_F_TEMPLATE))
599 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
600 /* connections inherit forwarding method from dest */
601 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
607 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
608 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
610 ip_vs_proto_name(cp->protocol),
611 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
612 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
613 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
614 ip_vs_fwd_tag(cp), cp->state,
615 cp->flags, refcount_read(&cp->refcnt),
616 refcount_read(&dest->refcnt));
618 /* Update the connection counters */
619 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
620 /* It is a normal connection, so modify the counters
621 * according to the flags, later the protocol can
622 * update them on state change
624 if (!(flags & IP_VS_CONN_F_INACTIVE))
625 atomic_inc(&dest->activeconns);
627 atomic_inc(&dest->inactconns);
629 /* It is a persistent connection/template, so increase
630 the persistent connection counter */
631 atomic_inc(&dest->persistconns);
634 if (dest->u_threshold != 0 &&
635 ip_vs_dest_totalconns(dest) >= dest->u_threshold)
636 dest->flags |= IP_VS_DEST_F_OVERLOAD;
641 * Check if there is a destination for the connection, if so
642 * bind the connection to the destination.
644 void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
646 struct ip_vs_dest *dest;
650 /* This function is only invoked by the synchronization code. We do
651 * not currently support heterogeneous pools with synchronization,
652 * so we can make the assumption that the svc_af is the same as the
655 dest = ip_vs_find_dest(cp->ipvs, cp->af, cp->af, &cp->daddr,
656 cp->dport, &cp->vaddr, cp->vport,
657 cp->protocol, cp->fwmark, cp->flags);
659 struct ip_vs_proto_data *pd;
661 spin_lock_bh(&cp->lock);
663 spin_unlock_bh(&cp->lock);
668 /* Applications work depending on the forwarding method
669 * but better to reassign them always when binding dest */
671 ip_vs_unbind_app(cp);
673 ip_vs_bind_dest(cp, dest);
674 spin_unlock_bh(&cp->lock);
676 /* Update its packet transmitter */
677 cp->packet_xmit = NULL;
678 #ifdef CONFIG_IP_VS_IPV6
679 if (cp->af == AF_INET6)
680 ip_vs_bind_xmit_v6(cp);
685 pd = ip_vs_proto_data_get(cp->ipvs, cp->protocol);
686 if (pd && atomic_read(&pd->appcnt))
687 ip_vs_bind_app(cp, pd->pp);
694 * Unbind a connection entry with its VS destination
695 * Called by the ip_vs_conn_expire function.
697 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
699 struct ip_vs_dest *dest = cp->dest;
704 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
705 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
707 ip_vs_proto_name(cp->protocol),
708 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
709 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
710 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
711 ip_vs_fwd_tag(cp), cp->state,
712 cp->flags, refcount_read(&cp->refcnt),
713 refcount_read(&dest->refcnt));
715 /* Update the connection counters */
716 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
717 /* It is a normal connection, so decrease the inactconns
718 or activeconns counter */
719 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
720 atomic_dec(&dest->inactconns);
722 atomic_dec(&dest->activeconns);
725 /* It is a persistent connection/template, so decrease
726 the persistent connection counter */
727 atomic_dec(&dest->persistconns);
730 if (dest->l_threshold != 0) {
731 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
732 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
733 } else if (dest->u_threshold != 0) {
734 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
735 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
737 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
738 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
741 ip_vs_dest_put(dest);
744 static int expire_quiescent_template(struct netns_ipvs *ipvs,
745 struct ip_vs_dest *dest)
748 return ipvs->sysctl_expire_quiescent_template &&
749 (atomic_read(&dest->weight) == 0);
756 * Checking if the destination of a connection template is available.
757 * If available, return 1, otherwise invalidate this connection
758 * template and return 0.
760 int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest)
762 struct ip_vs_dest *dest = ct->dest;
763 struct netns_ipvs *ipvs = ct->ipvs;
766 * Checking the dest server status.
768 if ((dest == NULL) ||
769 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
770 expire_quiescent_template(ipvs, dest) ||
771 (cdest && (dest != cdest))) {
772 IP_VS_DBG_BUF(9, "check_template: dest not available for "
773 "protocol %s s:%s:%d v:%s:%d "
775 ip_vs_proto_name(ct->protocol),
776 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
778 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
780 IP_VS_DBG_ADDR(ct->daf, &ct->daddr),
784 * Invalidate the connection template
786 if (ct->vport != htons(0xffff)) {
787 if (ip_vs_conn_unhash(ct)) {
788 ct->dport = htons(0xffff);
789 ct->vport = htons(0xffff);
796 * Simply decrease the refcnt of the template,
797 * don't restart its timer.
799 __ip_vs_conn_put(ct);
805 static void ip_vs_conn_rcu_free(struct rcu_head *head)
807 struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
810 ip_vs_pe_put(cp->pe);
812 kmem_cache_free(ip_vs_conn_cachep, cp);
815 static void ip_vs_conn_expire(struct timer_list *t)
817 struct ip_vs_conn *cp = from_timer(cp, t, timer);
818 struct netns_ipvs *ipvs = cp->ipvs;
821 * do I control anybody?
823 if (atomic_read(&cp->n_control))
826 /* Unlink conn if not referenced anymore */
827 if (likely(ip_vs_conn_unlink(cp))) {
828 struct ip_vs_conn *ct = cp->control;
830 /* delete the timer if it is activated by other users */
831 del_timer(&cp->timer);
833 /* does anybody control me? */
835 ip_vs_control_del(cp);
836 /* Drop CTL or non-assured TPL if not used anymore */
837 if (!cp->timeout && !atomic_read(&ct->n_control) &&
838 (!(ct->flags & IP_VS_CONN_F_TEMPLATE) ||
839 !(ct->state & IP_VS_CTPL_S_ASSURED))) {
840 IP_VS_DBG(4, "drop controlling connection\n");
842 ip_vs_conn_expire_now(ct);
846 if ((cp->flags & IP_VS_CONN_F_NFCT) &&
847 !(cp->flags & IP_VS_CONN_F_ONE_PACKET)) {
848 /* Do not access conntracks during subsys cleanup
849 * because nf_conntrack_find_get can not be used after
850 * conntrack cleanup for the net.
854 ip_vs_conn_drop_conntrack(cp);
857 if (unlikely(cp->app != NULL))
858 ip_vs_unbind_app(cp);
859 ip_vs_unbind_dest(cp);
860 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
861 atomic_dec(&ip_vs_conn_no_cport_cnt);
862 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
863 ip_vs_conn_rcu_free(&cp->rcu_head);
865 call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
866 atomic_dec(&ipvs->conn_count);
871 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
872 refcount_read(&cp->refcnt),
873 atomic_read(&cp->n_control));
875 refcount_inc(&cp->refcnt);
878 if (ipvs->sync_state & IP_VS_STATE_MASTER)
879 ip_vs_sync_conn(ipvs, cp, sysctl_sync_threshold(ipvs));
881 __ip_vs_conn_put_timer(cp);
884 /* Modify timer, so that it expires as soon as possible.
885 * Can be called without reference only if under RCU lock.
886 * We can have such chain of conns linked with ->control: DATA->CTL->TPL
887 * - DATA (eg. FTP) and TPL (persistence) can be present depending on setup
888 * - cp->timeout=0 indicates all conns from chain should be dropped but
889 * TPL is not dropped if in assured state
891 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
893 /* Using mod_timer_pending will ensure the timer is not
894 * modified after the final del_timer in ip_vs_conn_expire.
896 if (timer_pending(&cp->timer) &&
897 time_after(cp->timer.expires, jiffies))
898 mod_timer_pending(&cp->timer, jiffies);
903 * Create a new connection entry and hash it into the ip_vs_conn_tab
906 ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
907 const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
908 struct ip_vs_dest *dest, __u32 fwmark)
910 struct ip_vs_conn *cp;
911 struct netns_ipvs *ipvs = p->ipvs;
912 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->ipvs,
915 cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
917 IP_VS_ERR_RL("%s(): no memory\n", __func__);
921 INIT_HLIST_NODE(&cp->c_list);
922 timer_setup(&cp->timer, ip_vs_conn_expire, 0);
926 cp->protocol = p->protocol;
927 ip_vs_addr_set(p->af, &cp->caddr, p->caddr);
928 cp->cport = p->cport;
929 /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
930 ip_vs_addr_set(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
931 &cp->vaddr, p->vaddr);
932 cp->vport = p->vport;
933 ip_vs_addr_set(cp->daf, &cp->daddr, daddr);
937 if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
940 cp->pe_data = p->pe_data;
941 cp->pe_data_len = p->pe_data_len;
947 spin_lock_init(&cp->lock);
950 * Set the entry is referenced by the current thread before hashing
951 * it in the table, so that other thread run ip_vs_random_dropentry
952 * but cannot drop this entry.
954 refcount_set(&cp->refcnt, 1);
957 atomic_set(&cp->n_control, 0);
958 atomic_set(&cp->in_pkts, 0);
960 cp->packet_xmit = NULL;
963 /* reset struct ip_vs_seq */
964 cp->in_seq.delta = 0;
965 cp->out_seq.delta = 0;
967 atomic_inc(&ipvs->conn_count);
968 if (flags & IP_VS_CONN_F_NO_CPORT)
969 atomic_inc(&ip_vs_conn_no_cport_cnt);
971 /* Bind the connection with a destination server */
973 ip_vs_bind_dest(cp, dest);
975 /* Set its state and timeout */
979 cp->sync_endtime = jiffies & ~3UL;
981 /* Bind its packet transmitter */
982 #ifdef CONFIG_IP_VS_IPV6
983 if (p->af == AF_INET6)
984 ip_vs_bind_xmit_v6(cp);
989 if (unlikely(pd && atomic_read(&pd->appcnt)))
990 ip_vs_bind_app(cp, pd->pp);
993 * Allow conntrack to be preserved. By default, conntrack
994 * is created and destroyed for every packet.
995 * Sometimes keeping conntrack can be useful for
996 * IP_VS_CONN_F_ONE_PACKET too.
999 if (ip_vs_conntrack_enabled(ipvs))
1000 cp->flags |= IP_VS_CONN_F_NFCT;
1002 /* Hash it in the ip_vs_conn_tab finally */
1003 ip_vs_conn_hash(cp);
1009 * /proc/net/ip_vs_conn entries
1011 #ifdef CONFIG_PROC_FS
1012 struct ip_vs_iter_state {
1013 struct seq_net_private p;
1014 struct hlist_head *l;
1017 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
1020 struct ip_vs_conn *cp;
1021 struct ip_vs_iter_state *iter = seq->private;
1023 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1024 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1025 /* __ip_vs_conn_get() is not needed by
1026 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
1029 iter->l = &ip_vs_conn_tab[idx];
1039 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
1042 struct ip_vs_iter_state *iter = seq->private;
1046 return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
1049 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1051 struct ip_vs_conn *cp = v;
1052 struct ip_vs_iter_state *iter = seq->private;
1053 struct hlist_node *e;
1054 struct hlist_head *l = iter->l;
1058 if (v == SEQ_START_TOKEN)
1059 return ip_vs_conn_array(seq, 0);
1061 /* more on same hash chain? */
1062 e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1064 return hlist_entry(e, struct ip_vs_conn, c_list);
1066 idx = l - ip_vs_conn_tab;
1067 while (++idx < ip_vs_conn_tab_size) {
1068 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1069 iter->l = &ip_vs_conn_tab[idx];
1078 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1084 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1087 if (v == SEQ_START_TOKEN)
1089 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1091 const struct ip_vs_conn *cp = v;
1092 struct net *net = seq_file_net(seq);
1093 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1095 char dbuf[IP_VS_ADDRSTRLEN];
1097 if (!net_eq(cp->ipvs->net, net))
1101 len = strlen(cp->pe->name);
1102 memcpy(pe_data + 1, cp->pe->name, len);
1103 pe_data[len + 1] = ' ';
1105 len += cp->pe->show_pe_data(cp, pe_data + len);
1107 pe_data[len] = '\0';
1109 #ifdef CONFIG_IP_VS_IPV6
1110 if (cp->daf == AF_INET6)
1111 snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1114 snprintf(dbuf, sizeof(dbuf), "%08X",
1115 ntohl(cp->daddr.ip));
1117 #ifdef CONFIG_IP_VS_IPV6
1118 if (cp->af == AF_INET6)
1119 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1120 "%s %04X %-11s %7u%s\n",
1121 ip_vs_proto_name(cp->protocol),
1122 &cp->caddr.in6, ntohs(cp->cport),
1123 &cp->vaddr.in6, ntohs(cp->vport),
1124 dbuf, ntohs(cp->dport),
1125 ip_vs_state_name(cp),
1126 jiffies_delta_to_msecs(cp->timer.expires -
1132 "%-3s %08X %04X %08X %04X"
1133 " %s %04X %-11s %7u%s\n",
1134 ip_vs_proto_name(cp->protocol),
1135 ntohl(cp->caddr.ip), ntohs(cp->cport),
1136 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1137 dbuf, ntohs(cp->dport),
1138 ip_vs_state_name(cp),
1139 jiffies_delta_to_msecs(cp->timer.expires -
1146 static const struct seq_operations ip_vs_conn_seq_ops = {
1147 .start = ip_vs_conn_seq_start,
1148 .next = ip_vs_conn_seq_next,
1149 .stop = ip_vs_conn_seq_stop,
1150 .show = ip_vs_conn_seq_show,
1153 static const char *ip_vs_origin_name(unsigned int flags)
1155 if (flags & IP_VS_CONN_F_SYNC)
1161 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1163 char dbuf[IP_VS_ADDRSTRLEN];
1165 if (v == SEQ_START_TOKEN)
1167 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1169 const struct ip_vs_conn *cp = v;
1170 struct net *net = seq_file_net(seq);
1172 if (!net_eq(cp->ipvs->net, net))
1175 #ifdef CONFIG_IP_VS_IPV6
1176 if (cp->daf == AF_INET6)
1177 snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1180 snprintf(dbuf, sizeof(dbuf), "%08X",
1181 ntohl(cp->daddr.ip));
1183 #ifdef CONFIG_IP_VS_IPV6
1184 if (cp->af == AF_INET6)
1185 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1186 "%s %04X %-11s %-6s %7u\n",
1187 ip_vs_proto_name(cp->protocol),
1188 &cp->caddr.in6, ntohs(cp->cport),
1189 &cp->vaddr.in6, ntohs(cp->vport),
1190 dbuf, ntohs(cp->dport),
1191 ip_vs_state_name(cp),
1192 ip_vs_origin_name(cp->flags),
1193 jiffies_delta_to_msecs(cp->timer.expires -
1198 "%-3s %08X %04X %08X %04X "
1199 "%s %04X %-11s %-6s %7u\n",
1200 ip_vs_proto_name(cp->protocol),
1201 ntohl(cp->caddr.ip), ntohs(cp->cport),
1202 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1203 dbuf, ntohs(cp->dport),
1204 ip_vs_state_name(cp),
1205 ip_vs_origin_name(cp->flags),
1206 jiffies_delta_to_msecs(cp->timer.expires -
1212 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1213 .start = ip_vs_conn_seq_start,
1214 .next = ip_vs_conn_seq_next,
1215 .stop = ip_vs_conn_seq_stop,
1216 .show = ip_vs_conn_sync_seq_show,
1221 /* Randomly drop connection entries before running out of memory
1222 * Can be used for DATA and CTL conns. For TPL conns there are exceptions:
1223 * - traffic for services in OPS mode increases ct->in_pkts, so it is supported
1224 * - traffic for services not in OPS mode does not increase ct->in_pkts in
1225 * all cases, so it is not supported
1227 static inline int todrop_entry(struct ip_vs_conn *cp)
1230 * The drop rate array needs tuning for real environments.
1231 * Called from timer bh only => no locking
1233 static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1234 static char todrop_counter[9] = {0};
1237 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1238 This will leave enough time for normal connection to get
1240 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1243 /* Don't drop the entry if its number of incoming packets is not
1244 located in [0, 8] */
1245 i = atomic_read(&cp->in_pkts);
1246 if (i > 8 || i < 0) return 0;
1248 if (!todrop_rate[i]) return 0;
1249 if (--todrop_counter[i] > 0) return 0;
1251 todrop_counter[i] = todrop_rate[i];
1255 static inline bool ip_vs_conn_ops_mode(struct ip_vs_conn *cp)
1257 struct ip_vs_service *svc;
1261 svc = rcu_dereference(cp->dest->svc);
1262 return svc && (svc->flags & IP_VS_SVC_F_ONEPACKET);
1265 /* Called from keventd and must protect itself from softirqs */
1266 void ip_vs_random_dropentry(struct netns_ipvs *ipvs)
1269 struct ip_vs_conn *cp;
1273 * Randomly scan 1/32 of the whole table every second
1275 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1276 unsigned int hash = prandom_u32() & ip_vs_conn_tab_mask;
1278 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1279 if (cp->ipvs != ipvs)
1281 if (atomic_read(&cp->n_control))
1283 if (cp->flags & IP_VS_CONN_F_TEMPLATE) {
1284 /* connection template of OPS */
1285 if (ip_vs_conn_ops_mode(cp))
1287 if (!(cp->state & IP_VS_CTPL_S_ASSURED))
1291 if (cp->protocol == IPPROTO_TCP) {
1293 case IP_VS_TCP_S_SYN_RECV:
1294 case IP_VS_TCP_S_SYNACK:
1297 case IP_VS_TCP_S_ESTABLISHED:
1298 if (todrop_entry(cp))
1305 } else if (cp->protocol == IPPROTO_SCTP) {
1306 switch (cp->state) {
1307 case IP_VS_SCTP_S_INIT1:
1308 case IP_VS_SCTP_S_INIT:
1310 case IP_VS_SCTP_S_ESTABLISHED:
1311 if (todrop_entry(cp))
1319 if (!todrop_entry(cp))
1324 IP_VS_DBG(4, "drop connection\n");
1326 ip_vs_conn_expire_now(cp);
1335 * Flush all the connection entries in the ip_vs_conn_tab
1337 static void ip_vs_conn_flush(struct netns_ipvs *ipvs)
1340 struct ip_vs_conn *cp, *cp_c;
1344 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1346 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1347 if (cp->ipvs != ipvs)
1349 /* As timers are expired in LIFO order, restart
1350 * the timer of controlling connection first, so
1351 * that it is expired after us.
1354 /* cp->control is valid only with reference to cp */
1355 if (cp_c && __ip_vs_conn_get(cp)) {
1356 IP_VS_DBG(4, "del controlling connection\n");
1357 ip_vs_conn_expire_now(cp_c);
1358 __ip_vs_conn_put(cp);
1360 IP_VS_DBG(4, "del connection\n");
1361 ip_vs_conn_expire_now(cp);
1367 /* the counter may be not NULL, because maybe some conn entries
1368 are run by slow timer handler or unhashed but still referred */
1369 if (atomic_read(&ipvs->conn_count) != 0) {
1375 * per netns init and exit
1377 int __net_init ip_vs_conn_net_init(struct netns_ipvs *ipvs)
1379 atomic_set(&ipvs->conn_count, 0);
1381 proc_create_net("ip_vs_conn", 0, ipvs->net->proc_net,
1382 &ip_vs_conn_seq_ops, sizeof(struct ip_vs_iter_state));
1383 proc_create_net("ip_vs_conn_sync", 0, ipvs->net->proc_net,
1384 &ip_vs_conn_sync_seq_ops,
1385 sizeof(struct ip_vs_iter_state));
1389 void __net_exit ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs)
1391 /* flush all the connection entries first */
1392 ip_vs_conn_flush(ipvs);
1393 remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1394 remove_proc_entry("ip_vs_conn_sync", ipvs->net->proc_net);
1397 int __init ip_vs_conn_init(void)
1401 /* Compute size and mask */
1402 if (ip_vs_conn_tab_bits < 8 || ip_vs_conn_tab_bits > 20) {
1403 pr_info("conn_tab_bits not in [8, 20]. Using default value\n");
1404 ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
1406 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1407 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1410 * Allocate the connection hash table and initialize its list heads
1412 ip_vs_conn_tab = vmalloc(array_size(ip_vs_conn_tab_size,
1413 sizeof(*ip_vs_conn_tab)));
1414 if (!ip_vs_conn_tab)
1417 /* Allocate ip_vs_conn slab cache */
1418 ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1419 sizeof(struct ip_vs_conn), 0,
1420 SLAB_HWCACHE_ALIGN, NULL);
1421 if (!ip_vs_conn_cachep) {
1422 vfree(ip_vs_conn_tab);
1426 pr_info("Connection hash table configured "
1427 "(size=%d, memory=%ldKbytes)\n",
1428 ip_vs_conn_tab_size,
1429 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1430 IP_VS_DBG(0, "Each connection entry needs %zd bytes at least\n",
1431 sizeof(struct ip_vs_conn));
1433 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1434 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1436 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1437 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1440 /* calculate the random value for connection hash */
1441 get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1446 void ip_vs_conn_cleanup(void)
1448 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1450 /* Release the empty cache */
1451 kmem_cache_destroy(ip_vs_conn_cachep);
1452 vfree(ip_vs_conn_tab);