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;
405 const union nf_inet_addr *saddr;
409 * Check for "full" addressed entries
411 hash = ip_vs_conn_hashkey_param(p, true);
415 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
416 if (p->vport != cp->cport)
419 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
427 if (p->cport == sport && cp->af == p->af &&
428 ip_vs_addr_equal(p->af, p->vaddr, &cp->caddr) &&
429 ip_vs_addr_equal(p->af, p->caddr, saddr) &&
430 p->protocol == cp->protocol &&
431 cp->ipvs == p->ipvs) {
432 if (!__ip_vs_conn_get(cp))
442 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
443 ip_vs_proto_name(p->protocol),
444 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
445 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
446 ret ? "hit" : "not hit");
452 ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af,
453 const struct sk_buff *skb,
454 const struct ip_vs_iphdr *iph)
456 struct ip_vs_conn_param p;
458 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, &p))
461 return ip_vs_conn_out_get(&p);
463 EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
466 * Put back the conn and restart its timer with its timeout
468 static void __ip_vs_conn_put_timer(struct ip_vs_conn *cp)
470 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
472 mod_timer(&cp->timer, jiffies+t);
474 __ip_vs_conn_put(cp);
477 void ip_vs_conn_put(struct ip_vs_conn *cp)
479 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) &&
480 (refcount_read(&cp->refcnt) == 1) &&
481 !timer_pending(&cp->timer))
482 /* expire connection immediately */
483 ip_vs_conn_expire(&cp->timer);
485 __ip_vs_conn_put_timer(cp);
489 * Fill a no_client_port connection with a client port number
491 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
493 if (ip_vs_conn_unhash(cp)) {
494 spin_lock_bh(&cp->lock);
495 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
496 atomic_dec(&ip_vs_conn_no_cport_cnt);
497 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
500 spin_unlock_bh(&cp->lock);
502 /* hash on new dport */
509 * Bind a connection entry with the corresponding packet_xmit.
510 * Called by ip_vs_conn_new.
512 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
514 switch (IP_VS_FWD_METHOD(cp)) {
515 case IP_VS_CONN_F_MASQ:
516 cp->packet_xmit = ip_vs_nat_xmit;
519 case IP_VS_CONN_F_TUNNEL:
520 #ifdef CONFIG_IP_VS_IPV6
521 if (cp->daf == AF_INET6)
522 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
525 cp->packet_xmit = ip_vs_tunnel_xmit;
528 case IP_VS_CONN_F_DROUTE:
529 cp->packet_xmit = ip_vs_dr_xmit;
532 case IP_VS_CONN_F_LOCALNODE:
533 cp->packet_xmit = ip_vs_null_xmit;
536 case IP_VS_CONN_F_BYPASS:
537 cp->packet_xmit = ip_vs_bypass_xmit;
542 #ifdef CONFIG_IP_VS_IPV6
543 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
545 switch (IP_VS_FWD_METHOD(cp)) {
546 case IP_VS_CONN_F_MASQ:
547 cp->packet_xmit = ip_vs_nat_xmit_v6;
550 case IP_VS_CONN_F_TUNNEL:
551 if (cp->daf == AF_INET6)
552 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
554 cp->packet_xmit = ip_vs_tunnel_xmit;
557 case IP_VS_CONN_F_DROUTE:
558 cp->packet_xmit = ip_vs_dr_xmit_v6;
561 case IP_VS_CONN_F_LOCALNODE:
562 cp->packet_xmit = ip_vs_null_xmit;
565 case IP_VS_CONN_F_BYPASS:
566 cp->packet_xmit = ip_vs_bypass_xmit_v6;
573 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
575 return atomic_read(&dest->activeconns)
576 + atomic_read(&dest->inactconns);
580 * Bind a connection entry with a virtual service destination
581 * Called just after a new connection entry is created.
584 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
586 unsigned int conn_flags;
589 /* if dest is NULL, then return directly */
593 /* Increase the refcnt counter of the dest */
594 ip_vs_dest_hold(dest);
596 conn_flags = atomic_read(&dest->conn_flags);
597 if (cp->protocol != IPPROTO_UDP)
598 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
600 /* Bind with the destination and its corresponding transmitter */
601 if (flags & IP_VS_CONN_F_SYNC) {
602 /* if the connection is not template and is created
603 * by sync, preserve the activity flag.
605 if (!(flags & IP_VS_CONN_F_TEMPLATE))
606 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
607 /* connections inherit forwarding method from dest */
608 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
614 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
615 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
617 ip_vs_proto_name(cp->protocol),
618 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
619 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
620 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
621 ip_vs_fwd_tag(cp), cp->state,
622 cp->flags, refcount_read(&cp->refcnt),
623 refcount_read(&dest->refcnt));
625 /* Update the connection counters */
626 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
627 /* It is a normal connection, so modify the counters
628 * according to the flags, later the protocol can
629 * update them on state change
631 if (!(flags & IP_VS_CONN_F_INACTIVE))
632 atomic_inc(&dest->activeconns);
634 atomic_inc(&dest->inactconns);
636 /* It is a persistent connection/template, so increase
637 the persistent connection counter */
638 atomic_inc(&dest->persistconns);
641 if (dest->u_threshold != 0 &&
642 ip_vs_dest_totalconns(dest) >= dest->u_threshold)
643 dest->flags |= IP_VS_DEST_F_OVERLOAD;
648 * Check if there is a destination for the connection, if so
649 * bind the connection to the destination.
651 void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
653 struct ip_vs_dest *dest;
657 /* This function is only invoked by the synchronization code. We do
658 * not currently support heterogeneous pools with synchronization,
659 * so we can make the assumption that the svc_af is the same as the
662 dest = ip_vs_find_dest(cp->ipvs, cp->af, cp->af, &cp->daddr,
663 cp->dport, &cp->vaddr, cp->vport,
664 cp->protocol, cp->fwmark, cp->flags);
666 struct ip_vs_proto_data *pd;
668 spin_lock_bh(&cp->lock);
670 spin_unlock_bh(&cp->lock);
675 /* Applications work depending on the forwarding method
676 * but better to reassign them always when binding dest */
678 ip_vs_unbind_app(cp);
680 ip_vs_bind_dest(cp, dest);
681 spin_unlock_bh(&cp->lock);
683 /* Update its packet transmitter */
684 cp->packet_xmit = NULL;
685 #ifdef CONFIG_IP_VS_IPV6
686 if (cp->af == AF_INET6)
687 ip_vs_bind_xmit_v6(cp);
692 pd = ip_vs_proto_data_get(cp->ipvs, cp->protocol);
693 if (pd && atomic_read(&pd->appcnt))
694 ip_vs_bind_app(cp, pd->pp);
701 * Unbind a connection entry with its VS destination
702 * Called by the ip_vs_conn_expire function.
704 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
706 struct ip_vs_dest *dest = cp->dest;
711 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
712 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
714 ip_vs_proto_name(cp->protocol),
715 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
716 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
717 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
718 ip_vs_fwd_tag(cp), cp->state,
719 cp->flags, refcount_read(&cp->refcnt),
720 refcount_read(&dest->refcnt));
722 /* Update the connection counters */
723 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
724 /* It is a normal connection, so decrease the inactconns
725 or activeconns counter */
726 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
727 atomic_dec(&dest->inactconns);
729 atomic_dec(&dest->activeconns);
732 /* It is a persistent connection/template, so decrease
733 the persistent connection counter */
734 atomic_dec(&dest->persistconns);
737 if (dest->l_threshold != 0) {
738 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
739 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
740 } else if (dest->u_threshold != 0) {
741 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
742 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
744 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
745 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
748 ip_vs_dest_put(dest);
751 static int expire_quiescent_template(struct netns_ipvs *ipvs,
752 struct ip_vs_dest *dest)
755 return ipvs->sysctl_expire_quiescent_template &&
756 (atomic_read(&dest->weight) == 0);
763 * Checking if the destination of a connection template is available.
764 * If available, return 1, otherwise invalidate this connection
765 * template and return 0.
767 int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest)
769 struct ip_vs_dest *dest = ct->dest;
770 struct netns_ipvs *ipvs = ct->ipvs;
773 * Checking the dest server status.
775 if ((dest == NULL) ||
776 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
777 expire_quiescent_template(ipvs, dest) ||
778 (cdest && (dest != cdest))) {
779 IP_VS_DBG_BUF(9, "check_template: dest not available for "
780 "protocol %s s:%s:%d v:%s:%d "
782 ip_vs_proto_name(ct->protocol),
783 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
785 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
787 IP_VS_DBG_ADDR(ct->daf, &ct->daddr),
791 * Invalidate the connection template
793 if (ct->vport != htons(0xffff)) {
794 if (ip_vs_conn_unhash(ct)) {
795 ct->dport = htons(0xffff);
796 ct->vport = htons(0xffff);
803 * Simply decrease the refcnt of the template,
804 * don't restart its timer.
806 __ip_vs_conn_put(ct);
812 static void ip_vs_conn_rcu_free(struct rcu_head *head)
814 struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
817 ip_vs_pe_put(cp->pe);
819 kmem_cache_free(ip_vs_conn_cachep, cp);
822 /* Try to delete connection while not holding reference */
823 static void ip_vs_conn_del(struct ip_vs_conn *cp)
825 if (del_timer(&cp->timer)) {
826 /* Drop cp->control chain too */
829 ip_vs_conn_expire(&cp->timer);
833 /* Try to delete connection while holding reference */
834 static void ip_vs_conn_del_put(struct ip_vs_conn *cp)
836 if (del_timer(&cp->timer)) {
837 /* Drop cp->control chain too */
840 __ip_vs_conn_put(cp);
841 ip_vs_conn_expire(&cp->timer);
843 __ip_vs_conn_put(cp);
847 static void ip_vs_conn_expire(struct timer_list *t)
849 struct ip_vs_conn *cp = from_timer(cp, t, timer);
850 struct netns_ipvs *ipvs = cp->ipvs;
853 * do I control anybody?
855 if (atomic_read(&cp->n_control))
858 /* Unlink conn if not referenced anymore */
859 if (likely(ip_vs_conn_unlink(cp))) {
860 struct ip_vs_conn *ct = cp->control;
862 /* delete the timer if it is activated by other users */
863 del_timer(&cp->timer);
865 /* does anybody control me? */
867 bool has_ref = !cp->timeout && __ip_vs_conn_get(ct);
869 ip_vs_control_del(cp);
870 /* Drop CTL or non-assured TPL if not used anymore */
871 if (has_ref && !atomic_read(&ct->n_control) &&
872 (!(ct->flags & IP_VS_CONN_F_TEMPLATE) ||
873 !(ct->state & IP_VS_CTPL_S_ASSURED))) {
874 IP_VS_DBG(4, "drop controlling connection\n");
875 ip_vs_conn_del_put(ct);
876 } else if (has_ref) {
877 __ip_vs_conn_put(ct);
881 if ((cp->flags & IP_VS_CONN_F_NFCT) &&
882 !(cp->flags & IP_VS_CONN_F_ONE_PACKET)) {
883 /* Do not access conntracks during subsys cleanup
884 * because nf_conntrack_find_get can not be used after
885 * conntrack cleanup for the net.
889 ip_vs_conn_drop_conntrack(cp);
892 if (unlikely(cp->app != NULL))
893 ip_vs_unbind_app(cp);
894 ip_vs_unbind_dest(cp);
895 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
896 atomic_dec(&ip_vs_conn_no_cport_cnt);
897 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
898 ip_vs_conn_rcu_free(&cp->rcu_head);
900 call_rcu(&cp->rcu_head, ip_vs_conn_rcu_free);
901 atomic_dec(&ipvs->conn_count);
906 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
907 refcount_read(&cp->refcnt),
908 atomic_read(&cp->n_control));
910 refcount_inc(&cp->refcnt);
913 if (ipvs->sync_state & IP_VS_STATE_MASTER)
914 ip_vs_sync_conn(ipvs, cp, sysctl_sync_threshold(ipvs));
916 __ip_vs_conn_put_timer(cp);
919 /* Modify timer, so that it expires as soon as possible.
920 * Can be called without reference only if under RCU lock.
921 * We can have such chain of conns linked with ->control: DATA->CTL->TPL
922 * - DATA (eg. FTP) and TPL (persistence) can be present depending on setup
923 * - cp->timeout=0 indicates all conns from chain should be dropped but
924 * TPL is not dropped if in assured state
926 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
928 /* Using mod_timer_pending will ensure the timer is not
929 * modified after the final del_timer in ip_vs_conn_expire.
931 if (timer_pending(&cp->timer) &&
932 time_after(cp->timer.expires, jiffies))
933 mod_timer_pending(&cp->timer, jiffies);
938 * Create a new connection entry and hash it into the ip_vs_conn_tab
941 ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
942 const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
943 struct ip_vs_dest *dest, __u32 fwmark)
945 struct ip_vs_conn *cp;
946 struct netns_ipvs *ipvs = p->ipvs;
947 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(p->ipvs,
950 cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
952 IP_VS_ERR_RL("%s(): no memory\n", __func__);
956 INIT_HLIST_NODE(&cp->c_list);
957 timer_setup(&cp->timer, ip_vs_conn_expire, 0);
961 cp->protocol = p->protocol;
962 ip_vs_addr_set(p->af, &cp->caddr, p->caddr);
963 cp->cport = p->cport;
964 /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
965 ip_vs_addr_set(p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
966 &cp->vaddr, p->vaddr);
967 cp->vport = p->vport;
968 ip_vs_addr_set(cp->daf, &cp->daddr, daddr);
972 if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
975 cp->pe_data = p->pe_data;
976 cp->pe_data_len = p->pe_data_len;
982 spin_lock_init(&cp->lock);
985 * Set the entry is referenced by the current thread before hashing
986 * it in the table, so that other thread run ip_vs_random_dropentry
987 * but cannot drop this entry.
989 refcount_set(&cp->refcnt, 1);
992 atomic_set(&cp->n_control, 0);
993 atomic_set(&cp->in_pkts, 0);
995 cp->packet_xmit = NULL;
998 /* reset struct ip_vs_seq */
999 cp->in_seq.delta = 0;
1000 cp->out_seq.delta = 0;
1002 atomic_inc(&ipvs->conn_count);
1003 if (flags & IP_VS_CONN_F_NO_CPORT)
1004 atomic_inc(&ip_vs_conn_no_cport_cnt);
1006 /* Bind the connection with a destination server */
1008 ip_vs_bind_dest(cp, dest);
1010 /* Set its state and timeout */
1014 cp->sync_endtime = jiffies & ~3UL;
1016 /* Bind its packet transmitter */
1017 #ifdef CONFIG_IP_VS_IPV6
1018 if (p->af == AF_INET6)
1019 ip_vs_bind_xmit_v6(cp);
1022 ip_vs_bind_xmit(cp);
1024 if (unlikely(pd && atomic_read(&pd->appcnt)))
1025 ip_vs_bind_app(cp, pd->pp);
1028 * Allow conntrack to be preserved. By default, conntrack
1029 * is created and destroyed for every packet.
1030 * Sometimes keeping conntrack can be useful for
1031 * IP_VS_CONN_F_ONE_PACKET too.
1034 if (ip_vs_conntrack_enabled(ipvs))
1035 cp->flags |= IP_VS_CONN_F_NFCT;
1037 /* Hash it in the ip_vs_conn_tab finally */
1038 ip_vs_conn_hash(cp);
1044 * /proc/net/ip_vs_conn entries
1046 #ifdef CONFIG_PROC_FS
1047 struct ip_vs_iter_state {
1048 struct seq_net_private p;
1049 struct hlist_head *l;
1052 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
1055 struct ip_vs_conn *cp;
1056 struct ip_vs_iter_state *iter = seq->private;
1058 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1059 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1060 /* __ip_vs_conn_get() is not needed by
1061 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
1064 iter->l = &ip_vs_conn_tab[idx];
1074 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
1077 struct ip_vs_iter_state *iter = seq->private;
1081 return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
1084 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1086 struct ip_vs_conn *cp = v;
1087 struct ip_vs_iter_state *iter = seq->private;
1088 struct hlist_node *e;
1089 struct hlist_head *l = iter->l;
1093 if (v == SEQ_START_TOKEN)
1094 return ip_vs_conn_array(seq, 0);
1096 /* more on same hash chain? */
1097 e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1099 return hlist_entry(e, struct ip_vs_conn, c_list);
1101 idx = l - ip_vs_conn_tab;
1102 while (++idx < ip_vs_conn_tab_size) {
1103 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1104 iter->l = &ip_vs_conn_tab[idx];
1113 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1119 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1122 if (v == SEQ_START_TOKEN)
1124 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1126 const struct ip_vs_conn *cp = v;
1127 struct net *net = seq_file_net(seq);
1128 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1130 char dbuf[IP_VS_ADDRSTRLEN];
1132 if (!net_eq(cp->ipvs->net, net))
1136 len = strlen(cp->pe->name);
1137 memcpy(pe_data + 1, cp->pe->name, len);
1138 pe_data[len + 1] = ' ';
1140 len += cp->pe->show_pe_data(cp, pe_data + len);
1142 pe_data[len] = '\0';
1144 #ifdef CONFIG_IP_VS_IPV6
1145 if (cp->daf == AF_INET6)
1146 snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1149 snprintf(dbuf, sizeof(dbuf), "%08X",
1150 ntohl(cp->daddr.ip));
1152 #ifdef CONFIG_IP_VS_IPV6
1153 if (cp->af == AF_INET6)
1154 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1155 "%s %04X %-11s %7u%s\n",
1156 ip_vs_proto_name(cp->protocol),
1157 &cp->caddr.in6, ntohs(cp->cport),
1158 &cp->vaddr.in6, ntohs(cp->vport),
1159 dbuf, ntohs(cp->dport),
1160 ip_vs_state_name(cp),
1161 jiffies_delta_to_msecs(cp->timer.expires -
1167 "%-3s %08X %04X %08X %04X"
1168 " %s %04X %-11s %7u%s\n",
1169 ip_vs_proto_name(cp->protocol),
1170 ntohl(cp->caddr.ip), ntohs(cp->cport),
1171 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1172 dbuf, ntohs(cp->dport),
1173 ip_vs_state_name(cp),
1174 jiffies_delta_to_msecs(cp->timer.expires -
1181 static const struct seq_operations ip_vs_conn_seq_ops = {
1182 .start = ip_vs_conn_seq_start,
1183 .next = ip_vs_conn_seq_next,
1184 .stop = ip_vs_conn_seq_stop,
1185 .show = ip_vs_conn_seq_show,
1188 static const char *ip_vs_origin_name(unsigned int flags)
1190 if (flags & IP_VS_CONN_F_SYNC)
1196 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1198 char dbuf[IP_VS_ADDRSTRLEN];
1200 if (v == SEQ_START_TOKEN)
1202 "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1204 const struct ip_vs_conn *cp = v;
1205 struct net *net = seq_file_net(seq);
1207 if (!net_eq(cp->ipvs->net, net))
1210 #ifdef CONFIG_IP_VS_IPV6
1211 if (cp->daf == AF_INET6)
1212 snprintf(dbuf, sizeof(dbuf), "%pI6", &cp->daddr.in6);
1215 snprintf(dbuf, sizeof(dbuf), "%08X",
1216 ntohl(cp->daddr.ip));
1218 #ifdef CONFIG_IP_VS_IPV6
1219 if (cp->af == AF_INET6)
1220 seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X "
1221 "%s %04X %-11s %-6s %7u\n",
1222 ip_vs_proto_name(cp->protocol),
1223 &cp->caddr.in6, ntohs(cp->cport),
1224 &cp->vaddr.in6, ntohs(cp->vport),
1225 dbuf, ntohs(cp->dport),
1226 ip_vs_state_name(cp),
1227 ip_vs_origin_name(cp->flags),
1228 jiffies_delta_to_msecs(cp->timer.expires -
1233 "%-3s %08X %04X %08X %04X "
1234 "%s %04X %-11s %-6s %7u\n",
1235 ip_vs_proto_name(cp->protocol),
1236 ntohl(cp->caddr.ip), ntohs(cp->cport),
1237 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1238 dbuf, ntohs(cp->dport),
1239 ip_vs_state_name(cp),
1240 ip_vs_origin_name(cp->flags),
1241 jiffies_delta_to_msecs(cp->timer.expires -
1247 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1248 .start = ip_vs_conn_seq_start,
1249 .next = ip_vs_conn_seq_next,
1250 .stop = ip_vs_conn_seq_stop,
1251 .show = ip_vs_conn_sync_seq_show,
1256 /* Randomly drop connection entries before running out of memory
1257 * Can be used for DATA and CTL conns. For TPL conns there are exceptions:
1258 * - traffic for services in OPS mode increases ct->in_pkts, so it is supported
1259 * - traffic for services not in OPS mode does not increase ct->in_pkts in
1260 * all cases, so it is not supported
1262 static inline int todrop_entry(struct ip_vs_conn *cp)
1265 * The drop rate array needs tuning for real environments.
1266 * Called from timer bh only => no locking
1268 static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1269 static char todrop_counter[9] = {0};
1272 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1273 This will leave enough time for normal connection to get
1275 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1278 /* Don't drop the entry if its number of incoming packets is not
1279 located in [0, 8] */
1280 i = atomic_read(&cp->in_pkts);
1281 if (i > 8 || i < 0) return 0;
1283 if (!todrop_rate[i]) return 0;
1284 if (--todrop_counter[i] > 0) return 0;
1286 todrop_counter[i] = todrop_rate[i];
1290 static inline bool ip_vs_conn_ops_mode(struct ip_vs_conn *cp)
1292 struct ip_vs_service *svc;
1296 svc = rcu_dereference(cp->dest->svc);
1297 return svc && (svc->flags & IP_VS_SVC_F_ONEPACKET);
1300 /* Called from keventd and must protect itself from softirqs */
1301 void ip_vs_random_dropentry(struct netns_ipvs *ipvs)
1304 struct ip_vs_conn *cp;
1308 * Randomly scan 1/32 of the whole table every second
1310 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1311 unsigned int hash = prandom_u32() & ip_vs_conn_tab_mask;
1313 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1314 if (cp->ipvs != ipvs)
1316 if (atomic_read(&cp->n_control))
1318 if (cp->flags & IP_VS_CONN_F_TEMPLATE) {
1319 /* connection template of OPS */
1320 if (ip_vs_conn_ops_mode(cp))
1322 if (!(cp->state & IP_VS_CTPL_S_ASSURED))
1326 if (cp->protocol == IPPROTO_TCP) {
1328 case IP_VS_TCP_S_SYN_RECV:
1329 case IP_VS_TCP_S_SYNACK:
1332 case IP_VS_TCP_S_ESTABLISHED:
1333 if (todrop_entry(cp))
1340 } else if (cp->protocol == IPPROTO_SCTP) {
1341 switch (cp->state) {
1342 case IP_VS_SCTP_S_INIT1:
1343 case IP_VS_SCTP_S_INIT:
1345 case IP_VS_SCTP_S_ESTABLISHED:
1346 if (todrop_entry(cp))
1354 if (!todrop_entry(cp))
1359 IP_VS_DBG(4, "drop connection\n");
1369 * Flush all the connection entries in the ip_vs_conn_tab
1371 static void ip_vs_conn_flush(struct netns_ipvs *ipvs)
1374 struct ip_vs_conn *cp, *cp_c;
1378 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1380 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1381 if (cp->ipvs != ipvs)
1383 if (atomic_read(&cp->n_control))
1386 IP_VS_DBG(4, "del connection\n");
1388 if (cp_c && !atomic_read(&cp_c->n_control)) {
1389 IP_VS_DBG(4, "del controlling connection\n");
1390 ip_vs_conn_del(cp_c);
1397 /* the counter may be not NULL, because maybe some conn entries
1398 are run by slow timer handler or unhashed but still referred */
1399 if (atomic_read(&ipvs->conn_count) != 0) {
1405 #ifdef CONFIG_SYSCTL
1406 void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
1409 struct ip_vs_conn *cp, *cp_c;
1410 struct ip_vs_dest *dest;
1413 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1414 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1415 if (cp->ipvs != ipvs)
1419 if (!dest || (dest->flags & IP_VS_DEST_F_AVAILABLE))
1422 if (atomic_read(&cp->n_control))
1426 IP_VS_DBG(4, "del connection\n");
1428 if (cp_c && !atomic_read(&cp_c->n_control)) {
1429 IP_VS_DBG(4, "del controlling connection\n");
1430 ip_vs_conn_del(cp_c);
1435 /* netns clean up started, abort delayed work */
1444 * per netns init and exit
1446 int __net_init ip_vs_conn_net_init(struct netns_ipvs *ipvs)
1448 atomic_set(&ipvs->conn_count, 0);
1450 proc_create_net("ip_vs_conn", 0, ipvs->net->proc_net,
1451 &ip_vs_conn_seq_ops, sizeof(struct ip_vs_iter_state));
1452 proc_create_net("ip_vs_conn_sync", 0, ipvs->net->proc_net,
1453 &ip_vs_conn_sync_seq_ops,
1454 sizeof(struct ip_vs_iter_state));
1458 void __net_exit ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs)
1460 /* flush all the connection entries first */
1461 ip_vs_conn_flush(ipvs);
1462 remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1463 remove_proc_entry("ip_vs_conn_sync", ipvs->net->proc_net);
1466 int __init ip_vs_conn_init(void)
1470 /* Compute size and mask */
1471 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1472 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1475 * Allocate the connection hash table and initialize its list heads
1477 ip_vs_conn_tab = vmalloc(array_size(ip_vs_conn_tab_size,
1478 sizeof(*ip_vs_conn_tab)));
1479 if (!ip_vs_conn_tab)
1482 /* Allocate ip_vs_conn slab cache */
1483 ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1484 sizeof(struct ip_vs_conn), 0,
1485 SLAB_HWCACHE_ALIGN, NULL);
1486 if (!ip_vs_conn_cachep) {
1487 vfree(ip_vs_conn_tab);
1491 pr_info("Connection hash table configured "
1492 "(size=%d, memory=%ldKbytes)\n",
1493 ip_vs_conn_tab_size,
1494 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1495 IP_VS_DBG(0, "Each connection entry needs %zd bytes at least\n",
1496 sizeof(struct ip_vs_conn));
1498 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1499 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1501 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1502 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1505 /* calculate the random value for connection hash */
1506 get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1511 void ip_vs_conn_cleanup(void)
1513 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1515 /* Release the empty cache */
1516 kmem_cache_destroy(ip_vs_conn_cachep);
1517 vfree(ip_vs_conn_tab);