1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * IPVS An implementation of the IP virtual server support for the
4 * LINUX operating system. IPVS is now implemented as a module
5 * over the Netfilter framework. IPVS can be used to build a
6 * high-performance and highly available server based on a
9 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
10 * Peter Kese <peter.kese@ijs.si>
11 * Julian Anastasov <ja@ssi.bg>
13 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
14 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
15 * and others. Many code here is taken from IP MASQ code of kernel 2.2.
20 #define KMSG_COMPONENT "IPVS"
21 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
23 #include <linux/interrupt.h>
25 #include <linux/inet.h>
26 #include <linux/net.h>
27 #include <linux/kernel.h>
28 #include <linux/module.h>
29 #include <linux/vmalloc.h>
30 #include <linux/proc_fs.h> /* for proc_net_* */
31 #include <linux/slab.h>
32 #include <linux/seq_file.h>
33 #include <linux/jhash.h>
34 #include <linux/random.h>
36 #include <net/net_namespace.h>
37 #include <net/ip_vs.h>
40 #ifndef CONFIG_IP_VS_TAB_BITS
41 #define CONFIG_IP_VS_TAB_BITS 12
45 * Connection hash size. Default is what was selected at compile time.
47 static int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
48 module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
49 MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
51 /* size and mask values */
52 int ip_vs_conn_tab_size __read_mostly;
53 static int ip_vs_conn_tab_mask __read_mostly;
56 * Connection hash table: for input and output packets lookups of IPVS
58 static struct hlist_head *ip_vs_conn_tab __read_mostly;
60 /* SLAB cache for IPVS connections */
61 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
63 /* counter for no client port connections */
64 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
66 /* random value for IPVS connection hash */
67 static unsigned int ip_vs_conn_rnd __read_mostly;
70 * Fine locking granularity for big connection hash table
72 #define CT_LOCKARRAY_BITS 5
73 #define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
74 #define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
76 /* We need an addrstrlen that works with or without v6 */
77 #ifdef CONFIG_IP_VS_IPV6
78 #define IP_VS_ADDRSTRLEN INET6_ADDRSTRLEN
80 #define IP_VS_ADDRSTRLEN (8+1)
83 struct ip_vs_aligned_lock
86 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
88 /* lock array for conn table */
89 static struct ip_vs_aligned_lock
90 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
92 static inline void ct_write_lock_bh(unsigned int key)
94 spin_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
97 static inline void ct_write_unlock_bh(unsigned int key)
99 spin_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
102 static void ip_vs_conn_expire(struct timer_list *t);
105 * Returns hash value for IPVS connection entry
107 static unsigned int ip_vs_conn_hashkey(struct netns_ipvs *ipvs, int af, unsigned int proto,
108 const union nf_inet_addr *addr,
111 #ifdef CONFIG_IP_VS_IPV6
113 return (jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
114 (__force u32)port, proto, ip_vs_conn_rnd) ^
115 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
117 return (jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
119 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
122 static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
125 const union nf_inet_addr *addr;
128 if (p->pe_data && p->pe->hashkey_raw)
129 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
132 if (likely(!inverse)) {
140 return ip_vs_conn_hashkey(p->ipvs, p->af, p->protocol, addr, port);
143 static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
145 struct ip_vs_conn_param p;
147 ip_vs_conn_fill_param(cp->ipvs, cp->af, cp->protocol,
148 &cp->caddr, cp->cport, NULL, 0, &p);
152 p.pe_data = cp->pe_data;
153 p.pe_data_len = cp->pe_data_len;
156 return ip_vs_conn_hashkey_param(&p, false);
160 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
161 * returns bool success.
163 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
168 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
171 /* Hash by protocol, client address and port */
172 hash = ip_vs_conn_hashkey_conn(cp);
174 ct_write_lock_bh(hash);
175 spin_lock(&cp->lock);
177 if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
178 cp->flags |= IP_VS_CONN_F_HASHED;
179 refcount_inc(&cp->refcnt);
180 hlist_add_head_rcu(&cp->c_list, &ip_vs_conn_tab[hash]);
183 pr_err("%s(): request for already hashed, called from %pS\n",
184 __func__, __builtin_return_address(0));
188 spin_unlock(&cp->lock);
189 ct_write_unlock_bh(hash);
196 * UNhashes ip_vs_conn from ip_vs_conn_tab.
197 * returns bool success. Caller should hold conn reference.
199 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
204 /* unhash it and decrease its reference counter */
205 hash = ip_vs_conn_hashkey_conn(cp);
207 ct_write_lock_bh(hash);
208 spin_lock(&cp->lock);
210 if (cp->flags & IP_VS_CONN_F_HASHED) {
211 hlist_del_rcu(&cp->c_list);
212 cp->flags &= ~IP_VS_CONN_F_HASHED;
213 refcount_dec(&cp->refcnt);
218 spin_unlock(&cp->lock);
219 ct_write_unlock_bh(hash);
224 /* Try to unlink ip_vs_conn from ip_vs_conn_tab.
225 * returns bool success.
227 static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
232 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
233 return refcount_dec_if_one(&cp->refcnt);
235 hash = ip_vs_conn_hashkey_conn(cp);
237 ct_write_lock_bh(hash);
238 spin_lock(&cp->lock);
240 if (cp->flags & IP_VS_CONN_F_HASHED) {
241 /* Decrease refcnt and unlink conn only if we are last user */
242 if (refcount_dec_if_one(&cp->refcnt)) {
243 hlist_del_rcu(&cp->c_list);
244 cp->flags &= ~IP_VS_CONN_F_HASHED;
249 spin_unlock(&cp->lock);
250 ct_write_unlock_bh(hash);
257 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
258 * Called for pkts coming from OUTside-to-INside.
259 * p->caddr, p->cport: pkt source address (foreign host)
260 * p->vaddr, p->vport: pkt dest address (load balancer)
262 static inline struct ip_vs_conn *
263 __ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
266 struct ip_vs_conn *cp;
268 hash = ip_vs_conn_hashkey_param(p, false);
272 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
273 if (p->cport == cp->cport && p->vport == cp->vport &&
275 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
276 ip_vs_addr_equal(p->af, p->vaddr, &cp->vaddr) &&
277 ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
278 p->protocol == cp->protocol &&
279 cp->ipvs == p->ipvs) {
280 if (!__ip_vs_conn_get(cp))
293 struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
295 struct ip_vs_conn *cp;
297 cp = __ip_vs_conn_in_get(p);
298 if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt)) {
299 struct ip_vs_conn_param cport_zero_p = *p;
300 cport_zero_p.cport = 0;
301 cp = __ip_vs_conn_in_get(&cport_zero_p);
304 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
305 ip_vs_proto_name(p->protocol),
306 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
307 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
308 cp ? "hit" : "not hit");
314 ip_vs_conn_fill_param_proto(struct netns_ipvs *ipvs,
315 int af, const struct sk_buff *skb,
316 const struct ip_vs_iphdr *iph,
317 struct ip_vs_conn_param *p)
319 __be16 _ports[2], *pptr;
321 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
325 if (likely(!ip_vs_iph_inverse(iph)))
326 ip_vs_conn_fill_param(ipvs, af, iph->protocol, &iph->saddr,
327 pptr[0], &iph->daddr, pptr[1], p);
329 ip_vs_conn_fill_param(ipvs, af, iph->protocol, &iph->daddr,
330 pptr[1], &iph->saddr, pptr[0], p);
335 ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af,
336 const struct sk_buff *skb,
337 const struct ip_vs_iphdr *iph)
339 struct ip_vs_conn_param p;
341 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, &p))
344 return ip_vs_conn_in_get(&p);
346 EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
348 /* Get reference to connection template */
349 struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
352 struct ip_vs_conn *cp;
354 hash = ip_vs_conn_hashkey_param(p, false);
358 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
359 if (unlikely(p->pe_data && p->pe->ct_match)) {
360 if (cp->ipvs != p->ipvs)
362 if (p->pe == cp->pe && p->pe->ct_match(p, cp)) {
363 if (__ip_vs_conn_get(cp))
369 if (cp->af == p->af &&
370 ip_vs_addr_equal(p->af, p->caddr, &cp->caddr) &&
371 /* protocol should only be IPPROTO_IP if
372 * p->vaddr is a fwmark */
373 ip_vs_addr_equal(p->protocol == IPPROTO_IP ? AF_UNSPEC :
374 p->af, p->vaddr, &cp->vaddr) &&
375 p->vport == cp->vport && p->cport == cp->cport &&
376 cp->flags & IP_VS_CONN_F_TEMPLATE &&
377 p->protocol == cp->protocol &&
378 cp->ipvs == p->ipvs) {
379 if (__ip_vs_conn_get(cp))
388 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
389 ip_vs_proto_name(p->protocol),
390 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
391 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
392 cp ? "hit" : "not hit");
397 /* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
398 * Called for pkts coming from inside-to-OUTside.
399 * p->caddr, p->cport: pkt source address (inside host)
400 * p->vaddr, p->vport: pkt dest address (foreign host) */
401 struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
404 struct ip_vs_conn *cp, *ret=NULL;
407 * Check for "full" addressed entries
409 hash = ip_vs_conn_hashkey_param(p, true);
413 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
414 if (p->vport == cp->cport && p->cport == cp->dport &&
416 ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
417 ip_vs_addr_equal(p->af, p->caddr, &cp->daddr) &&
418 p->protocol == cp->protocol &&
419 cp->ipvs == p->ipvs) {
420 if (!__ip_vs_conn_get(cp))
430 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
431 ip_vs_proto_name(p->protocol),
432 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
433 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
434 ret ? "hit" : "not hit");
440 ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af,
441 const struct sk_buff *skb,
442 const struct ip_vs_iphdr *iph)
444 struct ip_vs_conn_param p;
446 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, &p))
449 return ip_vs_conn_out_get(&p);
451 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
454 * Put back the conn and restart its timer with its timeout
456 static void __ip_vs_conn_put_timer(struct ip_vs_conn *cp)
458 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
460 mod_timer(&cp->timer, jiffies+t);
462 __ip_vs_conn_put(cp);
465 void ip_vs_conn_put(struct ip_vs_conn *cp)
467 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) &&
468 (refcount_read(&cp->refcnt) == 1) &&
469 !timer_pending(&cp->timer))
470 /* expire connection immediately */
471 ip_vs_conn_expire(&cp->timer);
473 __ip_vs_conn_put_timer(cp);
477 * Fill a no_client_port connection with a client port number
479 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
481 if (ip_vs_conn_unhash(cp)) {
482 spin_lock_bh(&cp->lock);
483 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
484 atomic_dec(&ip_vs_conn_no_cport_cnt);
485 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
488 spin_unlock_bh(&cp->lock);
490 /* hash on new dport */
497 * Bind a connection entry with the corresponding packet_xmit.
498 * Called by ip_vs_conn_new.
500 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
502 switch (IP_VS_FWD_METHOD(cp)) {
503 case IP_VS_CONN_F_MASQ:
504 cp->packet_xmit = ip_vs_nat_xmit;
507 case IP_VS_CONN_F_TUNNEL:
508 #ifdef CONFIG_IP_VS_IPV6
509 if (cp->daf == AF_INET6)
510 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
513 cp->packet_xmit = ip_vs_tunnel_xmit;
516 case IP_VS_CONN_F_DROUTE:
517 cp->packet_xmit = ip_vs_dr_xmit;
520 case IP_VS_CONN_F_LOCALNODE:
521 cp->packet_xmit = ip_vs_null_xmit;
524 case IP_VS_CONN_F_BYPASS:
525 cp->packet_xmit = ip_vs_bypass_xmit;
530 #ifdef CONFIG_IP_VS_IPV6
531 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
533 switch (IP_VS_FWD_METHOD(cp)) {
534 case IP_VS_CONN_F_MASQ:
535 cp->packet_xmit = ip_vs_nat_xmit_v6;
538 case IP_VS_CONN_F_TUNNEL:
539 if (cp->daf == AF_INET6)
540 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
542 cp->packet_xmit = ip_vs_tunnel_xmit;
545 case IP_VS_CONN_F_DROUTE:
546 cp->packet_xmit = ip_vs_dr_xmit_v6;
549 case IP_VS_CONN_F_LOCALNODE:
550 cp->packet_xmit = ip_vs_null_xmit;
553 case IP_VS_CONN_F_BYPASS:
554 cp->packet_xmit = ip_vs_bypass_xmit_v6;
561 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
563 return atomic_read(&dest->activeconns)
564 + atomic_read(&dest->inactconns);
568 * Bind a connection entry with a virtual service destination
569 * Called just after a new connection entry is created.
572 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
574 unsigned int conn_flags;
577 /* if dest is NULL, then return directly */
581 /* Increase the refcnt counter of the dest */
582 ip_vs_dest_hold(dest);
584 conn_flags = atomic_read(&dest->conn_flags);
585 if (cp->protocol != IPPROTO_UDP)
586 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
588 /* Bind with the destination and its corresponding transmitter */
589 if (flags & IP_VS_CONN_F_SYNC) {
590 /* if the connection is not template and is created
591 * by sync, preserve the activity flag.
593 if (!(flags & IP_VS_CONN_F_TEMPLATE))
594 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
595 /* connections inherit forwarding method from dest */
596 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
602 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
603 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
605 ip_vs_proto_name(cp->protocol),
606 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
607 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
608 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
609 ip_vs_fwd_tag(cp), cp->state,
610 cp->flags, refcount_read(&cp->refcnt),
611 refcount_read(&dest->refcnt));
613 /* Update the connection counters */
614 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
615 /* It is a normal connection, so modify the counters
616 * according to the flags, later the protocol can
617 * update them on state change
619 if (!(flags & IP_VS_CONN_F_INACTIVE))
620 atomic_inc(&dest->activeconns);
622 atomic_inc(&dest->inactconns);
624 /* It is a persistent connection/template, so increase
625 the persistent connection counter */
626 atomic_inc(&dest->persistconns);
629 if (dest->u_threshold != 0 &&
630 ip_vs_dest_totalconns(dest) >= dest->u_threshold)
631 dest->flags |= IP_VS_DEST_F_OVERLOAD;
636 * Check if there is a destination for the connection, if so
637 * bind the connection to the destination.
639 void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
641 struct ip_vs_dest *dest;
645 /* This function is only invoked by the synchronization code. We do
646 * not currently support heterogeneous pools with synchronization,
647 * so we can make the assumption that the svc_af is the same as the
650 dest = ip_vs_find_dest(cp->ipvs, cp->af, cp->af, &cp->daddr,
651 cp->dport, &cp->vaddr, cp->vport,
652 cp->protocol, cp->fwmark, cp->flags);
654 struct ip_vs_proto_data *pd;
656 spin_lock_bh(&cp->lock);
658 spin_unlock_bh(&cp->lock);
663 /* Applications work depending on the forwarding method
664 * but better to reassign them always when binding dest */
666 ip_vs_unbind_app(cp);
668 ip_vs_bind_dest(cp, dest);
669 spin_unlock_bh(&cp->lock);
671 /* Update its packet transmitter */
672 cp->packet_xmit = NULL;
673 #ifdef CONFIG_IP_VS_IPV6
674 if (cp->af == AF_INET6)
675 ip_vs_bind_xmit_v6(cp);
680 pd = ip_vs_proto_data_get(cp->ipvs, cp->protocol);
681 if (pd && atomic_read(&pd->appcnt))
682 ip_vs_bind_app(cp, pd->pp);
689 * Unbind a connection entry with its VS destination
690 * Called by the ip_vs_conn_expire function.
692 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
694 struct ip_vs_dest *dest = cp->dest;
699 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
700 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
702 ip_vs_proto_name(cp->protocol),
703 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
704 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
705 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
706 ip_vs_fwd_tag(cp), cp->state,
707 cp->flags, refcount_read(&cp->refcnt),
708 refcount_read(&dest->refcnt));
710 /* Update the connection counters */
711 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
712 /* It is a normal connection, so decrease the inactconns
713 or activeconns counter */
714 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
715 atomic_dec(&dest->inactconns);
717 atomic_dec(&dest->activeconns);
720 /* It is a persistent connection/template, so decrease
721 the persistent connection counter */
722 atomic_dec(&dest->persistconns);
725 if (dest->l_threshold != 0) {
726 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
727 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
728 } else if (dest->u_threshold != 0) {
729 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
730 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
732 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
733 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
736 ip_vs_dest_put(dest);
739 static int expire_quiescent_template(struct netns_ipvs *ipvs,
740 struct ip_vs_dest *dest)
743 return ipvs->sysctl_expire_quiescent_template &&
744 (atomic_read(&dest->weight) == 0);
751 * Checking if the destination of a connection template is available.
752 * If available, return 1, otherwise invalidate this connection
753 * template and return 0.
755 int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest)
757 struct ip_vs_dest *dest = ct->dest;
758 struct netns_ipvs *ipvs = ct->ipvs;
761 * Checking the dest server status.
763 if ((dest == NULL) ||
764 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
765 expire_quiescent_template(ipvs, dest) ||
766 (cdest && (dest != cdest))) {
767 IP_VS_DBG_BUF(9, "check_template: dest not available for "
768 "protocol %s s:%s:%d v:%s:%d "
770 ip_vs_proto_name(ct->protocol),
771 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
773 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
775 IP_VS_DBG_ADDR(ct->daf, &ct->daddr),
779 * Invalidate the connection template
781 if (ct->vport != htons(0xffff)) {
782 if (ip_vs_conn_unhash(ct)) {
783 ct->dport = htons(0xffff);
784 ct->vport = htons(0xffff);
791 * Simply decrease the refcnt of the template,
792 * don't restart its timer.
794 __ip_vs_conn_put(ct);
800 static void ip_vs_conn_rcu_free(struct rcu_head *head)
802 struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
805 ip_vs_pe_put(cp->pe);
807 kmem_cache_free(ip_vs_conn_cachep, cp);
810 static void ip_vs_conn_expire(struct timer_list *t)
812 struct ip_vs_conn *cp = from_timer(cp, t, timer);
813 struct netns_ipvs *ipvs = cp->ipvs;
816 * do I control anybody?
818 if (atomic_read(&cp->n_control))
821 /* Unlink conn if not referenced anymore */
822 if (likely(ip_vs_conn_unlink(cp))) {
823 struct ip_vs_conn *ct = cp->control;
825 /* delete the timer if it is activated by other users */
826 del_timer(&cp->timer);
828 /* does anybody control me? */
830 ip_vs_control_del(cp);
831 /* Drop CTL or non-assured TPL if not used anymore */
832 if (!cp->timeout && !atomic_read(&ct->n_control) &&
833 (!(ct->flags & IP_VS_CONN_F_TEMPLATE) ||
834 !(ct->state & IP_VS_CTPL_S_ASSURED))) {
835 IP_VS_DBG(4, "drop controlling connection\n");
837 ip_vs_conn_expire_now(ct);
841 if ((cp->flags & IP_VS_CONN_F_NFCT) &&
842 !(cp->flags & IP_VS_CONN_F_ONE_PACKET)) {
843 /* Do not access conntracks during subsys cleanup
844 * because nf_conntrack_find_get can not be used after
845 * conntrack cleanup for the net.
849 ip_vs_conn_drop_conntrack(cp);
852 if (unlikely(cp->app != NULL))
853 ip_vs_unbind_app(cp);
854 ip_vs_unbind_dest(cp);
855 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
856 atomic_dec(&ip_vs_conn_no_cport_cnt);
857 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
858 ip_vs_conn_rcu_free(&cp->rcu_head);
860 call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
861 atomic_dec(&ipvs->conn_count);
866 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
867 refcount_read(&cp->refcnt),
868 atomic_read(&cp->n_control));
870 refcount_inc(&cp->refcnt);
873 if (ipvs->sync_state & IP_VS_STATE_MASTER)
874 ip_vs_sync_conn(ipvs, cp, sysctl_sync_threshold(ipvs));
876 __ip_vs_conn_put_timer(cp);
879 /* Modify timer, so that it expires as soon as possible.
880 * Can be called without reference only if under RCU lock.
881 * We can have such chain of conns linked with ->control: DATA->CTL->TPL
882 * - DATA (eg. FTP) and TPL (persistence) can be present depending on setup
883 * - cp->timeout=0 indicates all conns from chain should be dropped but
884 * TPL is not dropped if in assured state
886 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
888 /* Using mod_timer_pending will ensure the timer is not
889 * modified after the final del_timer in ip_vs_conn_expire.
891 if (timer_pending(&cp->timer) &&
892 time_after(cp->timer.expires, jiffies))
893 mod_timer_pending(&cp->timer, jiffies);
898 * Create a new connection entry and hash it into the ip_vs_conn_tab
901 ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
902 const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
903 struct ip_vs_dest *dest, __u32 fwmark)
905 struct ip_vs_conn *cp;
906 struct netns_ipvs *ipvs = p->ipvs;
907 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->ipvs,
910 cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
912 IP_VS_ERR_RL("%s(): no memory\n", __func__);
916 INIT_HLIST_NODE(&cp->c_list);
917 timer_setup(&cp->timer, ip_vs_conn_expire, 0);
921 cp->protocol = p->protocol;
922 ip_vs_addr_set(p->af, &cp->caddr, p->caddr);
923 cp->cport = p->cport;
924 /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
925 ip_vs_addr_set(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
926 &cp->vaddr, p->vaddr);
927 cp->vport = p->vport;
928 ip_vs_addr_set(cp->daf, &cp->daddr, daddr);
932 if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
935 cp->pe_data = p->pe_data;
936 cp->pe_data_len = p->pe_data_len;
942 spin_lock_init(&cp->lock);
945 * Set the entry is referenced by the current thread before hashing
946 * it in the table, so that other thread run ip_vs_random_dropentry
947 * but cannot drop this entry.
949 refcount_set(&cp->refcnt, 1);
952 atomic_set(&cp->n_control, 0);
953 atomic_set(&cp->in_pkts, 0);
955 cp->packet_xmit = NULL;
958 /* reset struct ip_vs_seq */
959 cp->in_seq.delta = 0;
960 cp->out_seq.delta = 0;
962 atomic_inc(&ipvs->conn_count);
963 if (flags & IP_VS_CONN_F_NO_CPORT)
964 atomic_inc(&ip_vs_conn_no_cport_cnt);
966 /* Bind the connection with a destination server */
968 ip_vs_bind_dest(cp, dest);
970 /* Set its state and timeout */
974 cp->sync_endtime = jiffies & ~3UL;
976 /* Bind its packet transmitter */
977 #ifdef CONFIG_IP_VS_IPV6
978 if (p->af == AF_INET6)
979 ip_vs_bind_xmit_v6(cp);
984 if (unlikely(pd && atomic_read(&pd->appcnt)))
985 ip_vs_bind_app(cp, pd->pp);
988 * Allow conntrack to be preserved. By default, conntrack
989 * is created and destroyed for every packet.
990 * Sometimes keeping conntrack can be useful for
991 * IP_VS_CONN_F_ONE_PACKET too.
994 if (ip_vs_conntrack_enabled(ipvs))
995 cp->flags |= IP_VS_CONN_F_NFCT;
997 /* Hash it in the ip_vs_conn_tab finally */
1004 * /proc/net/ip_vs_conn entries
1006 #ifdef CONFIG_PROC_FS
1007 struct ip_vs_iter_state {
1008 struct seq_net_private p;
1009 struct hlist_head *l;
1012 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
1015 struct ip_vs_conn *cp;
1016 struct ip_vs_iter_state *iter = seq->private;
1018 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1019 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1020 /* __ip_vs_conn_get() is not needed by
1021 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
1024 iter->l = &ip_vs_conn_tab[idx];
1034 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
1037 struct ip_vs_iter_state *iter = seq->private;
1041 return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
1044 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1046 struct ip_vs_conn *cp = v;
1047 struct ip_vs_iter_state *iter = seq->private;
1048 struct hlist_node *e;
1049 struct hlist_head *l = iter->l;
1053 if (v == SEQ_START_TOKEN)
1054 return ip_vs_conn_array(seq, 0);
1056 /* more on same hash chain? */
1057 e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1059 return hlist_entry(e, struct ip_vs_conn, c_list);
1061 idx = l - ip_vs_conn_tab;
1062 while (++idx < ip_vs_conn_tab_size) {
1063 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1064 iter->l = &ip_vs_conn_tab[idx];
1073 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1079 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1082 if (v == SEQ_START_TOKEN)
1084 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1086 const struct ip_vs_conn *cp = v;
1087 struct net *net = seq_file_net(seq);
1088 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1090 char dbuf[IP_VS_ADDRSTRLEN];
1092 if (!net_eq(cp->ipvs->net, net))
1096 len = strlen(cp->pe->name);
1097 memcpy(pe_data + 1, cp->pe->name, len);
1098 pe_data[len + 1] = ' ';
1100 len += cp->pe->show_pe_data(cp, pe_data + len);
1102 pe_data[len] = '\0';
1104 #ifdef CONFIG_IP_VS_IPV6
1105 if (cp->daf == AF_INET6)
1106 snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1109 snprintf(dbuf, sizeof(dbuf), "%08X",
1110 ntohl(cp->daddr.ip));
1112 #ifdef CONFIG_IP_VS_IPV6
1113 if (cp->af == AF_INET6)
1114 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1115 "%s %04X %-11s %7u%s\n",
1116 ip_vs_proto_name(cp->protocol),
1117 &cp->caddr.in6, ntohs(cp->cport),
1118 &cp->vaddr.in6, ntohs(cp->vport),
1119 dbuf, ntohs(cp->dport),
1120 ip_vs_state_name(cp),
1121 jiffies_delta_to_msecs(cp->timer.expires -
1127 "%-3s %08X %04X %08X %04X"
1128 " %s %04X %-11s %7u%s\n",
1129 ip_vs_proto_name(cp->protocol),
1130 ntohl(cp->caddr.ip), ntohs(cp->cport),
1131 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1132 dbuf, ntohs(cp->dport),
1133 ip_vs_state_name(cp),
1134 jiffies_delta_to_msecs(cp->timer.expires -
1141 static const struct seq_operations ip_vs_conn_seq_ops = {
1142 .start = ip_vs_conn_seq_start,
1143 .next = ip_vs_conn_seq_next,
1144 .stop = ip_vs_conn_seq_stop,
1145 .show = ip_vs_conn_seq_show,
1148 static const char *ip_vs_origin_name(unsigned int flags)
1150 if (flags & IP_VS_CONN_F_SYNC)
1156 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1158 char dbuf[IP_VS_ADDRSTRLEN];
1160 if (v == SEQ_START_TOKEN)
1162 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1164 const struct ip_vs_conn *cp = v;
1165 struct net *net = seq_file_net(seq);
1167 if (!net_eq(cp->ipvs->net, net))
1170 #ifdef CONFIG_IP_VS_IPV6
1171 if (cp->daf == AF_INET6)
1172 snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1175 snprintf(dbuf, sizeof(dbuf), "%08X",
1176 ntohl(cp->daddr.ip));
1178 #ifdef CONFIG_IP_VS_IPV6
1179 if (cp->af == AF_INET6)
1180 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1181 "%s %04X %-11s %-6s %7u\n",
1182 ip_vs_proto_name(cp->protocol),
1183 &cp->caddr.in6, ntohs(cp->cport),
1184 &cp->vaddr.in6, ntohs(cp->vport),
1185 dbuf, ntohs(cp->dport),
1186 ip_vs_state_name(cp),
1187 ip_vs_origin_name(cp->flags),
1188 jiffies_delta_to_msecs(cp->timer.expires -
1193 "%-3s %08X %04X %08X %04X "
1194 "%s %04X %-11s %-6s %7u\n",
1195 ip_vs_proto_name(cp->protocol),
1196 ntohl(cp->caddr.ip), ntohs(cp->cport),
1197 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1198 dbuf, ntohs(cp->dport),
1199 ip_vs_state_name(cp),
1200 ip_vs_origin_name(cp->flags),
1201 jiffies_delta_to_msecs(cp->timer.expires -
1207 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1208 .start = ip_vs_conn_seq_start,
1209 .next = ip_vs_conn_seq_next,
1210 .stop = ip_vs_conn_seq_stop,
1211 .show = ip_vs_conn_sync_seq_show,
1216 /* Randomly drop connection entries before running out of memory
1217 * Can be used for DATA and CTL conns. For TPL conns there are exceptions:
1218 * - traffic for services in OPS mode increases ct->in_pkts, so it is supported
1219 * - traffic for services not in OPS mode does not increase ct->in_pkts in
1220 * all cases, so it is not supported
1222 static inline int todrop_entry(struct ip_vs_conn *cp)
1225 * The drop rate array needs tuning for real environments.
1226 * Called from timer bh only => no locking
1228 static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1229 static char todrop_counter[9] = {0};
1232 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1233 This will leave enough time for normal connection to get
1235 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1238 /* Don't drop the entry if its number of incoming packets is not
1239 located in [0, 8] */
1240 i = atomic_read(&cp->in_pkts);
1241 if (i > 8 || i < 0) return 0;
1243 if (!todrop_rate[i]) return 0;
1244 if (--todrop_counter[i] > 0) return 0;
1246 todrop_counter[i] = todrop_rate[i];
1250 static inline bool ip_vs_conn_ops_mode(struct ip_vs_conn *cp)
1252 struct ip_vs_service *svc;
1256 svc = rcu_dereference(cp->dest->svc);
1257 return svc && (svc->flags & IP_VS_SVC_F_ONEPACKET);
1260 /* Called from keventd and must protect itself from softirqs */
1261 void ip_vs_random_dropentry(struct netns_ipvs *ipvs)
1264 struct ip_vs_conn *cp;
1268 * Randomly scan 1/32 of the whole table every second
1270 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1271 unsigned int hash = prandom_u32() & ip_vs_conn_tab_mask;
1273 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1274 if (cp->ipvs != ipvs)
1276 if (atomic_read(&cp->n_control))
1278 if (cp->flags & IP_VS_CONN_F_TEMPLATE) {
1279 /* connection template of OPS */
1280 if (ip_vs_conn_ops_mode(cp))
1282 if (!(cp->state & IP_VS_CTPL_S_ASSURED))
1286 if (cp->protocol == IPPROTO_TCP) {
1288 case IP_VS_TCP_S_SYN_RECV:
1289 case IP_VS_TCP_S_SYNACK:
1292 case IP_VS_TCP_S_ESTABLISHED:
1293 if (todrop_entry(cp))
1300 } else if (cp->protocol == IPPROTO_SCTP) {
1301 switch (cp->state) {
1302 case IP_VS_SCTP_S_INIT1:
1303 case IP_VS_SCTP_S_INIT:
1305 case IP_VS_SCTP_S_ESTABLISHED:
1306 if (todrop_entry(cp))
1314 if (!todrop_entry(cp))
1319 IP_VS_DBG(4, "drop connection\n");
1321 ip_vs_conn_expire_now(cp);
1330 * Flush all the connection entries in the ip_vs_conn_tab
1332 static void ip_vs_conn_flush(struct netns_ipvs *ipvs)
1335 struct ip_vs_conn *cp, *cp_c;
1339 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1341 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1342 if (cp->ipvs != ipvs)
1344 /* As timers are expired in LIFO order, restart
1345 * the timer of controlling connection first, so
1346 * that it is expired after us.
1349 /* cp->control is valid only with reference to cp */
1350 if (cp_c && __ip_vs_conn_get(cp)) {
1351 IP_VS_DBG(4, "del controlling connection\n");
1352 ip_vs_conn_expire_now(cp_c);
1353 __ip_vs_conn_put(cp);
1355 IP_VS_DBG(4, "del connection\n");
1356 ip_vs_conn_expire_now(cp);
1362 /* the counter may be not NULL, because maybe some conn entries
1363 are run by slow timer handler or unhashed but still referred */
1364 if (atomic_read(&ipvs->conn_count) != 0) {
1370 * per netns init and exit
1372 int __net_init ip_vs_conn_net_init(struct netns_ipvs *ipvs)
1374 atomic_set(&ipvs->conn_count, 0);
1376 proc_create_net("ip_vs_conn", 0, ipvs->net->proc_net,
1377 &ip_vs_conn_seq_ops, sizeof(struct ip_vs_iter_state));
1378 proc_create_net("ip_vs_conn_sync", 0, ipvs->net->proc_net,
1379 &ip_vs_conn_sync_seq_ops,
1380 sizeof(struct ip_vs_iter_state));
1384 void __net_exit ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs)
1386 /* flush all the connection entries first */
1387 ip_vs_conn_flush(ipvs);
1388 remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1389 remove_proc_entry("ip_vs_conn_sync", ipvs->net->proc_net);
1392 int __init ip_vs_conn_init(void)
1396 /* Compute size and mask */
1397 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1398 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1401 * Allocate the connection hash table and initialize its list heads
1403 ip_vs_conn_tab = vmalloc(array_size(ip_vs_conn_tab_size,
1404 sizeof(*ip_vs_conn_tab)));
1405 if (!ip_vs_conn_tab)
1408 /* Allocate ip_vs_conn slab cache */
1409 ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1410 sizeof(struct ip_vs_conn), 0,
1411 SLAB_HWCACHE_ALIGN, NULL);
1412 if (!ip_vs_conn_cachep) {
1413 vfree(ip_vs_conn_tab);
1417 pr_info("Connection hash table configured "
1418 "(size=%d, memory=%ldKbytes)\n",
1419 ip_vs_conn_tab_size,
1420 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1421 IP_VS_DBG(0, "Each connection entry needs %zd bytes at least\n",
1422 sizeof(struct ip_vs_conn));
1424 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1425 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1427 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1428 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1431 /* calculate the random value for connection hash */
1432 get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1437 void ip_vs_conn_cleanup(void)
1439 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1441 /* Release the empty cache */
1442 kmem_cache_destroy(ip_vs_conn_cachep);
1443 vfree(ip_vs_conn_tab);