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/checksum.h>
33 #define CLUSTERIP_VERSION "0.8"
35 MODULE_LICENSE("GPL");
36 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
37 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
39 struct clusterip_config
{
40 struct list_head list
; /* list of all configs */
41 atomic_t refcount
; /* reference count */
42 atomic_t entries
; /* number of entries/rules
45 __be32 clusterip
; /* the IP address */
46 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
47 struct net_device
*dev
; /* device */
48 u_int16_t num_total_nodes
; /* total number of nodes */
49 unsigned long local_nodes
; /* node number array */
52 struct proc_dir_entry
*pde
; /* proc dir entry */
54 enum clusterip_hashmode hash_mode
; /* which hashing mode */
55 u_int32_t hash_initval
; /* hash initialization */
59 static LIST_HEAD(clusterip_configs
);
61 /* clusterip_lock protects the clusterip_configs list */
62 static DEFINE_SPINLOCK(clusterip_lock
);
65 static const struct file_operations clusterip_proc_fops
;
66 static struct proc_dir_entry
*clusterip_procdir
;
70 clusterip_config_get(struct clusterip_config
*c
)
72 atomic_inc(&c
->refcount
);
76 static void clusterip_config_rcu_free(struct rcu_head
*head
)
78 kfree(container_of(head
, struct clusterip_config
, rcu
));
82 clusterip_config_put(struct clusterip_config
*c
)
84 if (atomic_dec_and_test(&c
->refcount
))
85 call_rcu_bh(&c
->rcu
, clusterip_config_rcu_free
);
88 /* decrease the count of entries using/referencing this config. If last
89 * entry(rule) is removed, remove the config from lists, but don't free it
90 * yet, since proc-files could still be holding references */
92 clusterip_config_entry_put(struct clusterip_config
*c
)
95 if (atomic_dec_and_lock(&c
->entries
, &clusterip_lock
)) {
96 list_del_rcu(&c
->list
);
97 spin_unlock(&clusterip_lock
);
100 dev_mc_del(c
->dev
, c
->clustermac
);
103 /* In case anyone still accesses the file, the open/close
104 * functions are also incrementing the refcount on their own,
105 * so it's safe to remove the entry even if it's in use. */
106 #ifdef CONFIG_PROC_FS
107 remove_proc_entry(c
->pde
->name
, c
->pde
->parent
);
114 static struct clusterip_config
*
115 __clusterip_config_find(__be32 clusterip
)
117 struct clusterip_config
*c
;
119 list_for_each_entry_rcu(c
, &clusterip_configs
, list
) {
120 if (c
->clusterip
== clusterip
)
127 static inline struct clusterip_config
*
128 clusterip_config_find_get(__be32 clusterip
, int entry
)
130 struct clusterip_config
*c
;
133 c
= __clusterip_config_find(clusterip
);
135 if (unlikely(!atomic_inc_not_zero(&c
->refcount
)))
138 atomic_inc(&c
->entries
);
140 rcu_read_unlock_bh();
146 clusterip_config_init_nodelist(struct clusterip_config
*c
,
147 const struct ipt_clusterip_tgt_info
*i
)
151 for (n
= 0; n
< i
->num_local_nodes
; n
++)
152 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
155 static struct clusterip_config
*
156 clusterip_config_init(const struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
157 struct net_device
*dev
)
159 struct clusterip_config
*c
;
161 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
167 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
168 c
->num_total_nodes
= i
->num_total_nodes
;
169 clusterip_config_init_nodelist(c
, i
);
170 c
->hash_mode
= i
->hash_mode
;
171 c
->hash_initval
= i
->hash_initval
;
172 atomic_set(&c
->refcount
, 1);
173 atomic_set(&c
->entries
, 1);
175 #ifdef CONFIG_PROC_FS
179 /* create proc dir entry */
180 sprintf(buffer
, "%pI4", &ip
);
181 c
->pde
= proc_create_data(buffer
, S_IWUSR
|S_IRUSR
,
183 &clusterip_proc_fops
, c
);
191 spin_lock_bh(&clusterip_lock
);
192 list_add_rcu(&c
->list
, &clusterip_configs
);
193 spin_unlock_bh(&clusterip_lock
);
198 #ifdef CONFIG_PROC_FS
200 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
204 nodenum
> c
->num_total_nodes
)
207 /* check if we already have this number in our bitfield */
208 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
215 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
218 nodenum
> c
->num_total_nodes
)
221 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
228 static inline u_int32_t
229 clusterip_hashfn(const struct sk_buff
*skb
,
230 const struct clusterip_config
*config
)
232 const struct iphdr
*iph
= ip_hdr(skb
);
233 unsigned long hashval
;
234 u_int16_t sport
, dport
;
235 const u_int16_t
*ports
;
237 switch (iph
->protocol
) {
240 case IPPROTO_UDPLITE
:
244 ports
= (const void *)iph
+iph
->ihl
*4;
250 pr_info("unknown protocol %u\n", iph
->protocol
);
254 switch (config
->hash_mode
) {
255 case CLUSTERIP_HASHMODE_SIP
:
256 hashval
= jhash_1word(ntohl(iph
->saddr
),
257 config
->hash_initval
);
259 case CLUSTERIP_HASHMODE_SIP_SPT
:
260 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
261 config
->hash_initval
);
263 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
264 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
265 config
->hash_initval
);
268 /* to make gcc happy */
270 /* This cannot happen, unless the check function wasn't called
271 * at rule load time */
272 pr_info("unknown mode %u\n", config
->hash_mode
);
277 /* node numbers are 1..n, not 0..n */
278 return (((u64
)hashval
* config
->num_total_nodes
) >> 32) + 1;
282 clusterip_responsible(const struct clusterip_config
*config
, u_int32_t hash
)
284 return test_bit(hash
- 1, &config
->local_nodes
);
287 /***********************************************************************
289 ***********************************************************************/
292 clusterip_tg(struct sk_buff
*skb
, const struct xt_action_param
*par
)
294 const struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
296 enum ip_conntrack_info ctinfo
;
299 /* don't need to clusterip_config_get() here, since refcount
300 * is only decremented by destroy() - and ip_tables guarantees
301 * that the ->target() function isn't called after ->destroy() */
303 ct
= nf_ct_get(skb
, &ctinfo
);
305 pr_info("no conntrack!\n");
309 /* special case: ICMP error handling. conntrack distinguishes between
310 * error messages (RELATED) and information requests (see below) */
311 if (ip_hdr(skb
)->protocol
== IPPROTO_ICMP
&&
312 (ctinfo
== IP_CT_RELATED
||
313 ctinfo
== IP_CT_RELATED
+ IP_CT_IS_REPLY
))
316 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
317 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
318 * on, which all have an ID field [relevant for hashing]. */
320 hash
= clusterip_hashfn(skb
, cipinfo
->config
);
327 case IP_CT_RELATED
+IP_CT_IS_REPLY
:
328 case IP_CT_ESTABLISHED
:
329 case IP_CT_ESTABLISHED
+IP_CT_IS_REPLY
:
336 nf_ct_dump_tuple_ip(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
338 pr_debug("hash=%u ct_hash=%u ", hash
, ct
->mark
);
339 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
340 pr_debug("not responsible\n");
343 pr_debug("responsible\n");
345 /* despite being received via linklayer multicast, this is
346 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
347 skb
->pkt_type
= PACKET_HOST
;
352 static int clusterip_tg_check(const struct xt_tgchk_param
*par
)
354 struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
355 const struct ipt_entry
*e
= par
->entryinfo
;
356 struct clusterip_config
*config
;
359 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
360 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
361 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
362 pr_info("unknown mode %u\n", cipinfo
->hash_mode
);
366 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff) ||
367 e
->ip
.dst
.s_addr
== 0) {
368 pr_info("Please specify destination IP\n");
373 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
375 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
376 pr_info("no config found for %pI4, need 'new'\n",
380 struct net_device
*dev
;
382 if (e
->ip
.iniface
[0] == '\0') {
383 pr_info("Please specify an interface name\n");
387 dev
= dev_get_by_name(&init_net
, e
->ip
.iniface
);
389 pr_info("no such interface %s\n",
394 config
= clusterip_config_init(cipinfo
,
395 e
->ip
.dst
.s_addr
, dev
);
397 pr_info("cannot allocate config\n");
401 dev_mc_add(config
->dev
, config
->clustermac
);
404 cipinfo
->config
= config
;
406 ret
= nf_ct_l3proto_try_module_get(par
->family
);
408 pr_info("cannot load conntrack support for proto=%u\n",
413 /* drop reference count of cluster config when rule is deleted */
414 static void clusterip_tg_destroy(const struct xt_tgdtor_param
*par
)
416 const struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
418 /* if no more entries are referencing the config, remove it
419 * from the list and destroy the proc entry */
420 clusterip_config_entry_put(cipinfo
->config
);
422 clusterip_config_put(cipinfo
->config
);
424 nf_ct_l3proto_module_put(par
->family
);
428 struct compat_ipt_clusterip_tgt_info
431 u_int8_t clustermac
[6];
432 u_int16_t num_total_nodes
;
433 u_int16_t num_local_nodes
;
434 u_int16_t local_nodes
[CLUSTERIP_MAX_NODES
];
436 u_int32_t hash_initval
;
437 compat_uptr_t config
;
439 #endif /* CONFIG_COMPAT */
441 static struct xt_target clusterip_tg_reg __read_mostly
= {
443 .family
= NFPROTO_IPV4
,
444 .target
= clusterip_tg
,
445 .checkentry
= clusterip_tg_check
,
446 .destroy
= clusterip_tg_destroy
,
447 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
449 .compatsize
= sizeof(struct compat_ipt_clusterip_tgt_info
),
450 #endif /* CONFIG_COMPAT */
455 /***********************************************************************
457 ***********************************************************************/
459 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
461 u_int8_t src_hw
[ETH_ALEN
];
463 u_int8_t dst_hw
[ETH_ALEN
];
468 static void arp_print(struct arp_payload
*payload
)
470 #define HBUFFERLEN 30
471 char hbuffer
[HBUFFERLEN
];
474 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
475 hbuffer
[k
++] = hex_asc_hi(payload
->src_hw
[j
]);
476 hbuffer
[k
++] = hex_asc_lo(payload
->src_hw
[j
]);
481 pr_debug("src %pI4@%s, dst %pI4\n",
482 &payload
->src_ip
, hbuffer
, &payload
->dst_ip
);
487 arp_mangle(unsigned int hook
,
489 const struct net_device
*in
,
490 const struct net_device
*out
,
491 int (*okfn
)(struct sk_buff
*))
493 struct arphdr
*arp
= arp_hdr(skb
);
494 struct arp_payload
*payload
;
495 struct clusterip_config
*c
;
497 /* we don't care about non-ethernet and non-ipv4 ARP */
498 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
499 arp
->ar_pro
!= htons(ETH_P_IP
) ||
500 arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
503 /* we only want to mangle arp requests and replies */
504 if (arp
->ar_op
!= htons(ARPOP_REPLY
) &&
505 arp
->ar_op
!= htons(ARPOP_REQUEST
))
508 payload
= (void *)(arp
+1);
510 /* if there is no clusterip configuration for the arp reply's
511 * source ip, we don't want to mangle it */
512 c
= clusterip_config_find_get(payload
->src_ip
, 0);
516 /* normally the linux kernel always replies to arp queries of
517 * addresses on different interfacs. However, in the CLUSTERIP case
518 * this wouldn't work, since we didn't subscribe the mcast group on
519 * other interfaces */
521 pr_debug("not mangling arp reply on different "
522 "interface: cip'%s'-skb'%s'\n",
523 c
->dev
->name
, out
->name
);
524 clusterip_config_put(c
);
528 /* mangle reply hardware address */
529 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
532 pr_debug("mangled arp reply: ");
536 clusterip_config_put(c
);
541 static struct nf_hook_ops cip_arp_ops __read_mostly
= {
544 .hooknum
= NF_ARP_OUT
,
548 /***********************************************************************
550 ***********************************************************************/
552 #ifdef CONFIG_PROC_FS
554 struct clusterip_seq_position
{
555 unsigned int pos
; /* position */
556 unsigned int weight
; /* number of bits set == size */
557 unsigned int bit
; /* current bit */
558 unsigned long val
; /* current value */
561 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
563 struct clusterip_config
*c
= s
->private;
565 u_int32_t local_nodes
;
566 struct clusterip_seq_position
*idx
;
568 local_nodes
= c
->local_nodes
;
569 weight
= hweight32(local_nodes
);
573 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
575 return ERR_PTR(-ENOMEM
);
578 idx
->weight
= weight
;
579 idx
->bit
= ffs(local_nodes
);
580 idx
->val
= local_nodes
;
581 clear_bit(idx
->bit
- 1, &idx
->val
);
586 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
588 struct clusterip_seq_position
*idx
= v
;
591 if (*pos
>= idx
->weight
) {
595 idx
->bit
= ffs(idx
->val
);
596 clear_bit(idx
->bit
- 1, &idx
->val
);
600 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
606 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
608 struct clusterip_seq_position
*idx
= v
;
613 seq_printf(s
, "%u", idx
->bit
);
615 if (idx
->pos
== idx
->weight
- 1)
621 static const struct seq_operations clusterip_seq_ops
= {
622 .start
= clusterip_seq_start
,
623 .next
= clusterip_seq_next
,
624 .stop
= clusterip_seq_stop
,
625 .show
= clusterip_seq_show
,
628 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
630 int ret
= seq_open(file
, &clusterip_seq_ops
);
633 struct seq_file
*sf
= file
->private_data
;
634 struct clusterip_config
*c
= PDE(inode
)->data
;
638 clusterip_config_get(c
);
644 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
646 struct clusterip_config
*c
= PDE(inode
)->data
;
649 ret
= seq_release(inode
, file
);
652 clusterip_config_put(c
);
657 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
658 size_t size
, loff_t
*ofs
)
660 struct clusterip_config
*c
= PDE(file
->f_path
.dentry
->d_inode
)->data
;
661 #define PROC_WRITELEN 10
662 char buffer
[PROC_WRITELEN
+1];
663 unsigned long nodenum
;
665 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
668 if (*buffer
== '+') {
669 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
670 if (clusterip_add_node(c
, nodenum
))
672 } else if (*buffer
== '-') {
673 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
674 if (clusterip_del_node(c
, nodenum
))
682 static const struct file_operations clusterip_proc_fops
= {
683 .owner
= THIS_MODULE
,
684 .open
= clusterip_proc_open
,
686 .write
= clusterip_proc_write
,
688 .release
= clusterip_proc_release
,
691 #endif /* CONFIG_PROC_FS */
693 static int __init
clusterip_tg_init(void)
697 ret
= xt_register_target(&clusterip_tg_reg
);
701 ret
= nf_register_hook(&cip_arp_ops
);
705 #ifdef CONFIG_PROC_FS
706 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", init_net
.proc_net
);
707 if (!clusterip_procdir
) {
708 pr_err("Unable to proc dir entry\n");
712 #endif /* CONFIG_PROC_FS */
714 pr_info("ClusterIP Version %s loaded successfully\n",
718 #ifdef CONFIG_PROC_FS
720 nf_unregister_hook(&cip_arp_ops
);
721 #endif /* CONFIG_PROC_FS */
723 xt_unregister_target(&clusterip_tg_reg
);
727 static void __exit
clusterip_tg_exit(void)
729 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION
);
730 #ifdef CONFIG_PROC_FS
731 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
733 nf_unregister_hook(&cip_arp_ops
);
734 xt_unregister_target(&clusterip_tg_reg
);
736 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
740 module_init(clusterip_tg_init
);
741 module_exit(clusterip_tg_exit
);