1 /* Cluster IP hashmark target
2 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
3 * based on ideas of Fabio Olive Leite <olive@unixforge.org>
5 * Development of this code funded by SuSE Linux AG, http://www.suse.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/module.h>
14 #include <linux/proc_fs.h>
15 #include <linux/jhash.h>
16 #include <linux/bitops.h>
17 #include <linux/skbuff.h>
18 #include <linux/slab.h>
20 #include <linux/tcp.h>
21 #include <linux/udp.h>
22 #include <linux/icmp.h>
23 #include <linux/if_arp.h>
24 #include <linux/seq_file.h>
25 #include <linux/netfilter_arp.h>
26 #include <linux/netfilter/x_tables.h>
27 #include <linux/netfilter_ipv4/ip_tables.h>
28 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/net_namespace.h>
31 #include <net/netns/generic.h>
32 #include <net/checksum.h>
35 #define CLUSTERIP_VERSION "0.8"
37 MODULE_LICENSE("GPL");
38 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
39 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
41 struct clusterip_config
{
42 struct list_head list
; /* list of all configs */
43 atomic_t refcount
; /* reference count */
44 atomic_t entries
; /* number of entries/rules
47 __be32 clusterip
; /* the IP address */
48 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
49 struct net_device
*dev
; /* device */
50 u_int16_t num_total_nodes
; /* total number of nodes */
51 unsigned long local_nodes
; /* node number array */
54 struct proc_dir_entry
*pde
; /* proc dir entry */
56 enum clusterip_hashmode hash_mode
; /* which hashing mode */
57 u_int32_t hash_initval
; /* hash initialization */
62 static const struct file_operations clusterip_proc_fops
;
65 static unsigned int clusterip_net_id __read_mostly
;
67 struct clusterip_net
{
68 struct list_head configs
;
69 /* lock protects the configs list */
73 struct proc_dir_entry
*procdir
;
78 clusterip_config_get(struct clusterip_config
*c
)
80 atomic_inc(&c
->refcount
);
84 static void clusterip_config_rcu_free(struct rcu_head
*head
)
86 kfree(container_of(head
, struct clusterip_config
, rcu
));
90 clusterip_config_put(struct clusterip_config
*c
)
92 if (atomic_dec_and_test(&c
->refcount
))
93 call_rcu_bh(&c
->rcu
, clusterip_config_rcu_free
);
96 /* decrease the count of entries using/referencing this config. If last
97 * entry(rule) is removed, remove the config from lists, but don't free it
98 * yet, since proc-files could still be holding references */
100 clusterip_config_entry_put(struct clusterip_config
*c
)
102 struct net
*net
= dev_net(c
->dev
);
103 struct clusterip_net
*cn
= net_generic(net
, clusterip_net_id
);
106 if (atomic_dec_and_lock(&c
->entries
, &cn
->lock
)) {
107 list_del_rcu(&c
->list
);
108 spin_unlock(&cn
->lock
);
111 dev_mc_del(c
->dev
, c
->clustermac
);
114 /* In case anyone still accesses the file, the open/close
115 * functions are also incrementing the refcount on their own,
116 * so it's safe to remove the entry even if it's in use. */
117 #ifdef CONFIG_PROC_FS
125 static struct clusterip_config
*
126 __clusterip_config_find(struct net
*net
, __be32 clusterip
)
128 struct clusterip_config
*c
;
129 struct clusterip_net
*cn
= net_generic(net
, clusterip_net_id
);
131 list_for_each_entry_rcu(c
, &cn
->configs
, list
) {
132 if (c
->clusterip
== clusterip
)
139 static inline struct clusterip_config
*
140 clusterip_config_find_get(struct net
*net
, __be32 clusterip
, int entry
)
142 struct clusterip_config
*c
;
145 c
= __clusterip_config_find(net
, clusterip
);
147 if (unlikely(!atomic_inc_not_zero(&c
->refcount
)))
150 atomic_inc(&c
->entries
);
152 rcu_read_unlock_bh();
158 clusterip_config_init_nodelist(struct clusterip_config
*c
,
159 const struct ipt_clusterip_tgt_info
*i
)
163 for (n
= 0; n
< i
->num_local_nodes
; n
++)
164 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
167 static struct clusterip_config
*
168 clusterip_config_init(const struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
169 struct net_device
*dev
)
171 struct clusterip_config
*c
;
172 struct clusterip_net
*cn
= net_generic(dev_net(dev
), clusterip_net_id
);
174 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
180 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
181 c
->num_total_nodes
= i
->num_total_nodes
;
182 clusterip_config_init_nodelist(c
, i
);
183 c
->hash_mode
= i
->hash_mode
;
184 c
->hash_initval
= i
->hash_initval
;
185 atomic_set(&c
->refcount
, 1);
186 atomic_set(&c
->entries
, 1);
188 #ifdef CONFIG_PROC_FS
192 /* create proc dir entry */
193 sprintf(buffer
, "%pI4", &ip
);
194 c
->pde
= proc_create_data(buffer
, S_IWUSR
|S_IRUSR
,
196 &clusterip_proc_fops
, c
);
204 spin_lock_bh(&cn
->lock
);
205 list_add_rcu(&c
->list
, &cn
->configs
);
206 spin_unlock_bh(&cn
->lock
);
211 #ifdef CONFIG_PROC_FS
213 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
217 nodenum
> c
->num_total_nodes
)
220 /* check if we already have this number in our bitfield */
221 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
228 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
231 nodenum
> c
->num_total_nodes
)
234 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
241 static inline u_int32_t
242 clusterip_hashfn(const struct sk_buff
*skb
,
243 const struct clusterip_config
*config
)
245 const struct iphdr
*iph
= ip_hdr(skb
);
246 unsigned long hashval
;
247 u_int16_t sport
= 0, dport
= 0;
250 poff
= proto_ports_offset(iph
->protocol
);
252 const u_int16_t
*ports
;
255 ports
= skb_header_pointer(skb
, iph
->ihl
* 4 + poff
, 4, _ports
);
261 net_info_ratelimited("unknown protocol %u\n", iph
->protocol
);
264 switch (config
->hash_mode
) {
265 case CLUSTERIP_HASHMODE_SIP
:
266 hashval
= jhash_1word(ntohl(iph
->saddr
),
267 config
->hash_initval
);
269 case CLUSTERIP_HASHMODE_SIP_SPT
:
270 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
271 config
->hash_initval
);
273 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
274 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
275 config
->hash_initval
);
278 /* to make gcc happy */
280 /* This cannot happen, unless the check function wasn't called
281 * at rule load time */
282 pr_info("unknown mode %u\n", config
->hash_mode
);
287 /* node numbers are 1..n, not 0..n */
288 return reciprocal_scale(hashval
, config
->num_total_nodes
) + 1;
292 clusterip_responsible(const struct clusterip_config
*config
, u_int32_t hash
)
294 return test_bit(hash
- 1, &config
->local_nodes
);
297 /***********************************************************************
299 ***********************************************************************/
302 clusterip_tg(struct sk_buff
*skb
, const struct xt_action_param
*par
)
304 const struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
306 enum ip_conntrack_info ctinfo
;
309 /* don't need to clusterip_config_get() here, since refcount
310 * is only decremented by destroy() - and ip_tables guarantees
311 * that the ->target() function isn't called after ->destroy() */
313 ct
= nf_ct_get(skb
, &ctinfo
);
317 /* special case: ICMP error handling. conntrack distinguishes between
318 * error messages (RELATED) and information requests (see below) */
319 if (ip_hdr(skb
)->protocol
== IPPROTO_ICMP
&&
320 (ctinfo
== IP_CT_RELATED
||
321 ctinfo
== IP_CT_RELATED_REPLY
))
324 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
325 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
326 * on, which all have an ID field [relevant for hashing]. */
328 hash
= clusterip_hashfn(skb
, cipinfo
->config
);
335 case IP_CT_RELATED_REPLY
:
336 /* FIXME: we don't handle expectations at the moment.
337 * They can arrive on a different node than
338 * the master connection (e.g. FTP passive mode) */
339 case IP_CT_ESTABLISHED
:
340 case IP_CT_ESTABLISHED_REPLY
:
342 default: /* Prevent gcc warnings */
347 nf_ct_dump_tuple_ip(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
349 pr_debug("hash=%u ct_hash=%u ", hash
, ct
->mark
);
350 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
351 pr_debug("not responsible\n");
354 pr_debug("responsible\n");
356 /* despite being received via linklayer multicast, this is
357 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
358 skb
->pkt_type
= PACKET_HOST
;
363 static int clusterip_tg_check(const struct xt_tgchk_param
*par
)
365 struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
366 const struct ipt_entry
*e
= par
->entryinfo
;
367 struct clusterip_config
*config
;
370 if (par
->nft_compat
) {
371 pr_err("cannot use CLUSTERIP target from nftables compat\n");
375 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
376 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
377 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
378 pr_info("unknown mode %u\n", cipinfo
->hash_mode
);
382 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff) ||
383 e
->ip
.dst
.s_addr
== 0) {
384 pr_info("Please specify destination IP\n");
388 /* FIXME: further sanity checks */
390 config
= clusterip_config_find_get(par
->net
, e
->ip
.dst
.s_addr
, 1);
392 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
393 pr_info("no config found for %pI4, need 'new'\n",
397 struct net_device
*dev
;
399 if (e
->ip
.iniface
[0] == '\0') {
400 pr_info("Please specify an interface name\n");
404 dev
= dev_get_by_name(par
->net
, e
->ip
.iniface
);
406 pr_info("no such interface %s\n",
411 config
= clusterip_config_init(cipinfo
,
412 e
->ip
.dst
.s_addr
, dev
);
417 dev_mc_add(config
->dev
, config
->clustermac
);
420 cipinfo
->config
= config
;
422 ret
= nf_ct_netns_get(par
->net
, par
->family
);
424 pr_info("cannot load conntrack support for proto=%u\n",
427 if (!par
->net
->xt
.clusterip_deprecated_warning
) {
428 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
429 "use xt_cluster instead\n");
430 par
->net
->xt
.clusterip_deprecated_warning
= true;
436 /* drop reference count of cluster config when rule is deleted */
437 static void clusterip_tg_destroy(const struct xt_tgdtor_param
*par
)
439 const struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
441 /* if no more entries are referencing the config, remove it
442 * from the list and destroy the proc entry */
443 clusterip_config_entry_put(cipinfo
->config
);
445 clusterip_config_put(cipinfo
->config
);
447 nf_ct_netns_get(par
->net
, par
->family
);
451 struct compat_ipt_clusterip_tgt_info
454 u_int8_t clustermac
[6];
455 u_int16_t num_total_nodes
;
456 u_int16_t num_local_nodes
;
457 u_int16_t local_nodes
[CLUSTERIP_MAX_NODES
];
459 u_int32_t hash_initval
;
460 compat_uptr_t config
;
462 #endif /* CONFIG_COMPAT */
464 static struct xt_target clusterip_tg_reg __read_mostly
= {
466 .family
= NFPROTO_IPV4
,
467 .target
= clusterip_tg
,
468 .checkentry
= clusterip_tg_check
,
469 .destroy
= clusterip_tg_destroy
,
470 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
472 .compatsize
= sizeof(struct compat_ipt_clusterip_tgt_info
),
473 #endif /* CONFIG_COMPAT */
478 /***********************************************************************
480 ***********************************************************************/
482 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
484 u_int8_t src_hw
[ETH_ALEN
];
486 u_int8_t dst_hw
[ETH_ALEN
];
491 static void arp_print(struct arp_payload
*payload
)
493 #define HBUFFERLEN 30
494 char hbuffer
[HBUFFERLEN
];
497 for (k
= 0, j
= 0; k
< HBUFFERLEN
- 3 && j
< ETH_ALEN
; j
++) {
498 hbuffer
[k
++] = hex_asc_hi(payload
->src_hw
[j
]);
499 hbuffer
[k
++] = hex_asc_lo(payload
->src_hw
[j
]);
504 pr_debug("src %pI4@%s, dst %pI4\n",
505 &payload
->src_ip
, hbuffer
, &payload
->dst_ip
);
510 arp_mangle(void *priv
,
512 const struct nf_hook_state
*state
)
514 struct arphdr
*arp
= arp_hdr(skb
);
515 struct arp_payload
*payload
;
516 struct clusterip_config
*c
;
517 struct net
*net
= state
->net
;
519 /* we don't care about non-ethernet and non-ipv4 ARP */
520 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
521 arp
->ar_pro
!= htons(ETH_P_IP
) ||
522 arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
525 /* we only want to mangle arp requests and replies */
526 if (arp
->ar_op
!= htons(ARPOP_REPLY
) &&
527 arp
->ar_op
!= htons(ARPOP_REQUEST
))
530 payload
= (void *)(arp
+1);
532 /* if there is no clusterip configuration for the arp reply's
533 * source ip, we don't want to mangle it */
534 c
= clusterip_config_find_get(net
, payload
->src_ip
, 0);
538 /* normally the linux kernel always replies to arp queries of
539 * addresses on different interfacs. However, in the CLUSTERIP case
540 * this wouldn't work, since we didn't subscribe the mcast group on
541 * other interfaces */
542 if (c
->dev
!= state
->out
) {
543 pr_debug("not mangling arp reply on different "
544 "interface: cip'%s'-skb'%s'\n",
545 c
->dev
->name
, state
->out
->name
);
546 clusterip_config_put(c
);
550 /* mangle reply hardware address */
551 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
554 pr_debug("mangled arp reply: ");
558 clusterip_config_put(c
);
563 static struct nf_hook_ops cip_arp_ops __read_mostly
= {
566 .hooknum
= NF_ARP_OUT
,
570 /***********************************************************************
572 ***********************************************************************/
574 #ifdef CONFIG_PROC_FS
576 struct clusterip_seq_position
{
577 unsigned int pos
; /* position */
578 unsigned int weight
; /* number of bits set == size */
579 unsigned int bit
; /* current bit */
580 unsigned long val
; /* current value */
583 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
585 struct clusterip_config
*c
= s
->private;
587 u_int32_t local_nodes
;
588 struct clusterip_seq_position
*idx
;
590 /* FIXME: possible race */
591 local_nodes
= c
->local_nodes
;
592 weight
= hweight32(local_nodes
);
596 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
598 return ERR_PTR(-ENOMEM
);
601 idx
->weight
= weight
;
602 idx
->bit
= ffs(local_nodes
);
603 idx
->val
= local_nodes
;
604 clear_bit(idx
->bit
- 1, &idx
->val
);
609 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
611 struct clusterip_seq_position
*idx
= v
;
614 if (*pos
>= idx
->weight
) {
618 idx
->bit
= ffs(idx
->val
);
619 clear_bit(idx
->bit
- 1, &idx
->val
);
623 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
629 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
631 struct clusterip_seq_position
*idx
= v
;
636 seq_printf(s
, "%u", idx
->bit
);
638 if (idx
->pos
== idx
->weight
- 1)
644 static const struct seq_operations clusterip_seq_ops
= {
645 .start
= clusterip_seq_start
,
646 .next
= clusterip_seq_next
,
647 .stop
= clusterip_seq_stop
,
648 .show
= clusterip_seq_show
,
651 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
653 int ret
= seq_open(file
, &clusterip_seq_ops
);
656 struct seq_file
*sf
= file
->private_data
;
657 struct clusterip_config
*c
= PDE_DATA(inode
);
661 clusterip_config_get(c
);
667 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
669 struct clusterip_config
*c
= PDE_DATA(inode
);
672 ret
= seq_release(inode
, file
);
675 clusterip_config_put(c
);
680 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
681 size_t size
, loff_t
*ofs
)
683 struct clusterip_config
*c
= PDE_DATA(file_inode(file
));
684 #define PROC_WRITELEN 10
685 char buffer
[PROC_WRITELEN
+1];
686 unsigned long nodenum
;
689 if (size
> PROC_WRITELEN
)
691 if (copy_from_user(buffer
, input
, size
))
695 if (*buffer
== '+') {
696 rc
= kstrtoul(buffer
+1, 10, &nodenum
);
699 if (clusterip_add_node(c
, nodenum
))
701 } else if (*buffer
== '-') {
702 rc
= kstrtoul(buffer
+1, 10, &nodenum
);
705 if (clusterip_del_node(c
, nodenum
))
713 static const struct file_operations clusterip_proc_fops
= {
714 .owner
= THIS_MODULE
,
715 .open
= clusterip_proc_open
,
717 .write
= clusterip_proc_write
,
719 .release
= clusterip_proc_release
,
722 #endif /* CONFIG_PROC_FS */
724 static int clusterip_net_init(struct net
*net
)
726 struct clusterip_net
*cn
= net_generic(net
, clusterip_net_id
);
728 INIT_LIST_HEAD(&cn
->configs
);
730 spin_lock_init(&cn
->lock
);
732 #ifdef CONFIG_PROC_FS
733 cn
->procdir
= proc_mkdir("ipt_CLUSTERIP", net
->proc_net
);
735 pr_err("Unable to proc dir entry\n");
738 #endif /* CONFIG_PROC_FS */
743 static void clusterip_net_exit(struct net
*net
)
745 #ifdef CONFIG_PROC_FS
746 struct clusterip_net
*cn
= net_generic(net
, clusterip_net_id
);
747 proc_remove(cn
->procdir
);
751 static struct pernet_operations clusterip_net_ops
= {
752 .init
= clusterip_net_init
,
753 .exit
= clusterip_net_exit
,
754 .id
= &clusterip_net_id
,
755 .size
= sizeof(struct clusterip_net
),
758 static int __init
clusterip_tg_init(void)
762 ret
= register_pernet_subsys(&clusterip_net_ops
);
766 ret
= xt_register_target(&clusterip_tg_reg
);
770 ret
= nf_register_hook(&cip_arp_ops
);
774 pr_info("ClusterIP Version %s loaded successfully\n",
780 xt_unregister_target(&clusterip_tg_reg
);
782 unregister_pernet_subsys(&clusterip_net_ops
);
786 static void __exit
clusterip_tg_exit(void)
788 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION
);
790 nf_unregister_hook(&cip_arp_ops
);
791 xt_unregister_target(&clusterip_tg_reg
);
792 unregister_pernet_subsys(&clusterip_net_ops
);
794 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
798 module_init(clusterip_tg_init
);
799 module_exit(clusterip_tg_exit
);