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 #include <linux/module.h>
13 #include <linux/proc_fs.h>
14 #include <linux/jhash.h>
15 #include <linux/bitops.h>
16 #include <linux/skbuff.h>
18 #include <linux/tcp.h>
19 #include <linux/udp.h>
20 #include <linux/icmp.h>
21 #include <linux/if_arp.h>
22 #include <linux/proc_fs.h>
23 #include <linux/seq_file.h>
25 #include <net/checksum.h>
27 #include <linux/netfilter_arp.h>
29 #include <linux/netfilter/x_tables.h>
30 #include <linux/netfilter_ipv4/ip_tables.h>
31 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
32 #include <net/netfilter/nf_conntrack_compat.h>
34 #define CLUSTERIP_VERSION "0.8"
36 #define DEBUG_CLUSTERIP
38 #ifdef DEBUG_CLUSTERIP
44 MODULE_LICENSE("GPL");
45 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
46 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
48 struct clusterip_config
{
49 struct list_head list
; /* list of all configs */
50 atomic_t refcount
; /* reference count */
51 atomic_t entries
; /* number of entries/rules
54 __be32 clusterip
; /* the IP address */
55 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
56 struct net_device
*dev
; /* device */
57 u_int16_t num_total_nodes
; /* total number of nodes */
58 unsigned long local_nodes
; /* node number array */
61 struct proc_dir_entry
*pde
; /* proc dir entry */
63 enum clusterip_hashmode hash_mode
; /* which hashing mode */
64 u_int32_t hash_initval
; /* hash initialization */
67 static LIST_HEAD(clusterip_configs
);
69 /* clusterip_lock protects the clusterip_configs list */
70 static DEFINE_RWLOCK(clusterip_lock
);
73 static const struct file_operations clusterip_proc_fops
;
74 static struct proc_dir_entry
*clusterip_procdir
;
78 clusterip_config_get(struct clusterip_config
*c
)
80 atomic_inc(&c
->refcount
);
84 clusterip_config_put(struct clusterip_config
*c
)
86 if (atomic_dec_and_test(&c
->refcount
))
90 /* increase the count of entries(rules) using/referencing this config */
92 clusterip_config_entry_get(struct clusterip_config
*c
)
94 atomic_inc(&c
->entries
);
97 /* decrease the count of entries using/referencing this config. If last
98 * entry(rule) is removed, remove the config from lists, but don't free it
99 * yet, since proc-files could still be holding references */
101 clusterip_config_entry_put(struct clusterip_config
*c
)
103 if (atomic_dec_and_test(&c
->entries
)) {
104 write_lock_bh(&clusterip_lock
);
106 write_unlock_bh(&clusterip_lock
);
108 dev_mc_delete(c
->dev
, c
->clustermac
, ETH_ALEN
, 0);
111 /* In case anyone still accesses the file, the open/close
112 * functions are also incrementing the refcount on their own,
113 * so it's safe to remove the entry even if it's in use. */
114 #ifdef CONFIG_PROC_FS
115 remove_proc_entry(c
->pde
->name
, c
->pde
->parent
);
120 static struct clusterip_config
*
121 __clusterip_config_find(__be32 clusterip
)
123 struct list_head
*pos
;
125 list_for_each(pos
, &clusterip_configs
) {
126 struct clusterip_config
*c
= list_entry(pos
,
127 struct clusterip_config
, list
);
128 if (c
->clusterip
== clusterip
) {
136 static inline struct clusterip_config
*
137 clusterip_config_find_get(__be32 clusterip
, int entry
)
139 struct clusterip_config
*c
;
141 read_lock_bh(&clusterip_lock
);
142 c
= __clusterip_config_find(clusterip
);
144 read_unlock_bh(&clusterip_lock
);
147 atomic_inc(&c
->refcount
);
149 atomic_inc(&c
->entries
);
150 read_unlock_bh(&clusterip_lock
);
156 clusterip_config_init_nodelist(struct clusterip_config
*c
,
157 const struct ipt_clusterip_tgt_info
*i
)
161 for (n
= 0; n
< i
->num_local_nodes
; n
++) {
162 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
166 static struct clusterip_config
*
167 clusterip_config_init(struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
168 struct net_device
*dev
)
170 struct clusterip_config
*c
;
172 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
178 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
179 c
->num_total_nodes
= i
->num_total_nodes
;
180 clusterip_config_init_nodelist(c
, i
);
181 c
->hash_mode
= i
->hash_mode
;
182 c
->hash_initval
= i
->hash_initval
;
183 atomic_set(&c
->refcount
, 1);
184 atomic_set(&c
->entries
, 1);
186 #ifdef CONFIG_PROC_FS
190 /* create proc dir entry */
191 sprintf(buffer
, "%u.%u.%u.%u", NIPQUAD(ip
));
192 c
->pde
= create_proc_entry(buffer
, S_IWUSR
|S_IRUSR
,
199 c
->pde
->proc_fops
= &clusterip_proc_fops
;
203 write_lock_bh(&clusterip_lock
);
204 list_add(&c
->list
, &clusterip_configs
);
205 write_unlock_bh(&clusterip_lock
);
210 #ifdef CONFIG_PROC_FS
212 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
216 nodenum
> c
->num_total_nodes
)
219 /* check if we already have this number in our bitfield */
220 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
227 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
230 nodenum
> c
->num_total_nodes
)
233 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
240 static inline u_int32_t
241 clusterip_hashfn(struct sk_buff
*skb
, struct clusterip_config
*config
)
243 struct iphdr
*iph
= ip_hdr(skb
);
244 unsigned long hashval
;
245 u_int16_t sport
, dport
;
248 switch (iph
->protocol
) {
251 case IPPROTO_UDPLITE
:
255 ports
= (void *)iph
+iph
->ihl
*4;
260 if (net_ratelimit()) {
261 printk(KERN_NOTICE
"CLUSTERIP: unknown protocol `%u'\n",
267 switch (config
->hash_mode
) {
268 case CLUSTERIP_HASHMODE_SIP
:
269 hashval
= jhash_1word(ntohl(iph
->saddr
),
270 config
->hash_initval
);
272 case CLUSTERIP_HASHMODE_SIP_SPT
:
273 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
274 config
->hash_initval
);
276 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
277 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
278 config
->hash_initval
);
281 /* to make gcc happy */
283 /* This cannot happen, unless the check function wasn't called
284 * at rule load time */
285 printk("CLUSTERIP: unknown mode `%u'\n", config
->hash_mode
);
290 /* node numbers are 1..n, not 0..n */
291 return ((hashval
% config
->num_total_nodes
)+1);
295 clusterip_responsible(struct clusterip_config
*config
, u_int32_t hash
)
297 return test_bit(hash
- 1, &config
->local_nodes
);
300 /***********************************************************************
302 ***********************************************************************/
305 target(struct sk_buff
**pskb
,
306 const struct net_device
*in
,
307 const struct net_device
*out
,
308 unsigned int hooknum
,
309 const struct xt_target
*target
,
310 const void *targinfo
)
312 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
313 enum ip_conntrack_info ctinfo
;
314 u_int32_t
*mark
, hash
;
316 /* don't need to clusterip_config_get() here, since refcount
317 * is only decremented by destroy() - and ip_tables guarantees
318 * that the ->target() function isn't called after ->destroy() */
320 mark
= nf_ct_get_mark((*pskb
), &ctinfo
);
322 printk(KERN_ERR
"CLUSTERIP: no conntrack!\n");
323 /* FIXME: need to drop invalid ones, since replies
324 * to outgoing connections of other nodes will be
325 * marked as INVALID */
329 /* special case: ICMP error handling. conntrack distinguishes between
330 * error messages (RELATED) and information requests (see below) */
331 if (ip_hdr(*pskb
)->protocol
== IPPROTO_ICMP
332 && (ctinfo
== IP_CT_RELATED
333 || ctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
))
336 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
337 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
338 * on, which all have an ID field [relevant for hashing]. */
340 hash
= clusterip_hashfn(*pskb
, cipinfo
->config
);
347 case IP_CT_RELATED
+IP_CT_IS_REPLY
:
348 /* FIXME: we don't handle expectations at the
349 * moment. they can arrive on a different node than
350 * the master connection (e.g. FTP passive mode) */
351 case IP_CT_ESTABLISHED
:
352 case IP_CT_ESTABLISHED
+IP_CT_IS_REPLY
:
358 #ifdef DEBUG_CLUSTERP
359 DUMP_TUPLE(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
361 DEBUGP("hash=%u ct_hash=%u ", hash
, *mark
);
362 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
363 DEBUGP("not responsible\n");
366 DEBUGP("responsible\n");
368 /* despite being received via linklayer multicast, this is
369 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
370 (*pskb
)->pkt_type
= PACKET_HOST
;
376 checkentry(const char *tablename
,
378 const struct xt_target
*target
,
380 unsigned int hook_mask
)
382 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
383 const struct ipt_entry
*e
= e_void
;
385 struct clusterip_config
*config
;
387 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
388 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
389 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
390 printk(KERN_WARNING
"CLUSTERIP: unknown mode `%u'\n",
395 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff)
396 || e
->ip
.dst
.s_addr
== 0) {
397 printk(KERN_ERR
"CLUSTERIP: Please specify destination IP\n");
401 /* FIXME: further sanity checks */
403 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
405 if (cipinfo
->config
!= NULL
) {
406 /* Case A: This is an entry that gets reloaded, since
407 * it still has a cipinfo->config pointer. Simply
408 * increase the entry refcount and return */
409 if (cipinfo
->config
!= config
) {
410 printk(KERN_ERR
"CLUSTERIP: Reloaded entry "
411 "has invalid config pointer!\n");
415 /* Case B: This is a new rule referring to an existing
416 * clusterip config. */
417 cipinfo
->config
= config
;
420 /* Case C: This is a completely new clusterip config */
421 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
422 printk(KERN_WARNING
"CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e
->ip
.dst
.s_addr
));
425 struct net_device
*dev
;
427 if (e
->ip
.iniface
[0] == '\0') {
428 printk(KERN_WARNING
"CLUSTERIP: Please specify an interface name\n");
432 dev
= dev_get_by_name(e
->ip
.iniface
);
434 printk(KERN_WARNING
"CLUSTERIP: no such interface %s\n", e
->ip
.iniface
);
438 config
= clusterip_config_init(cipinfo
,
439 e
->ip
.dst
.s_addr
, dev
);
441 printk(KERN_WARNING
"CLUSTERIP: cannot allocate config\n");
445 dev_mc_add(config
->dev
,config
->clustermac
, ETH_ALEN
, 0);
447 cipinfo
->config
= config
;
450 if (nf_ct_l3proto_try_module_get(target
->family
) < 0) {
451 printk(KERN_WARNING
"can't load conntrack support for "
452 "proto=%d\n", target
->family
);
459 /* drop reference count of cluster config when rule is deleted */
460 static void destroy(const struct xt_target
*target
, void *targinfo
)
462 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
464 /* if no more entries are referencing the config, remove it
465 * from the list and destroy the proc entry */
466 clusterip_config_entry_put(cipinfo
->config
);
468 clusterip_config_put(cipinfo
->config
);
470 nf_ct_l3proto_module_put(target
->family
);
473 static struct xt_target clusterip_tgt
= {
477 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
478 .checkentry
= checkentry
,
484 /***********************************************************************
486 ***********************************************************************/
488 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
490 u_int8_t src_hw
[ETH_ALEN
];
492 u_int8_t dst_hw
[ETH_ALEN
];
494 } __attribute__ ((packed
));
496 #ifdef CLUSTERIP_DEBUG
497 static void arp_print(struct arp_payload
*payload
)
499 #define HBUFFERLEN 30
500 char hbuffer
[HBUFFERLEN
];
502 const char hexbuf
[]= "0123456789abcdef";
504 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
505 hbuffer
[k
++]=hexbuf
[(payload
->src_hw
[j
]>>4)&15];
506 hbuffer
[k
++]=hexbuf
[payload
->src_hw
[j
]&15];
511 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
512 NIPQUAD(payload
->src_ip
), hbuffer
,
513 NIPQUAD(payload
->dst_ip
));
518 arp_mangle(unsigned int hook
,
519 struct sk_buff
**pskb
,
520 const struct net_device
*in
,
521 const struct net_device
*out
,
522 int (*okfn
)(struct sk_buff
*))
524 struct arphdr
*arp
= (*pskb
)->nh
.arph
;
525 struct arp_payload
*payload
;
526 struct clusterip_config
*c
;
528 /* we don't care about non-ethernet and non-ipv4 ARP */
529 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
)
530 || arp
->ar_pro
!= htons(ETH_P_IP
)
531 || arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
534 /* we only want to mangle arp requests and replies */
535 if (arp
->ar_op
!= htons(ARPOP_REPLY
)
536 && arp
->ar_op
!= htons(ARPOP_REQUEST
))
539 payload
= (void *)(arp
+1);
541 /* if there is no clusterip configuration for the arp reply's
542 * source ip, we don't want to mangle it */
543 c
= clusterip_config_find_get(payload
->src_ip
, 0);
547 /* normally the linux kernel always replies to arp queries of
548 * addresses on different interfacs. However, in the CLUSTERIP case
549 * this wouldn't work, since we didn't subscribe the mcast group on
550 * other interfaces */
552 DEBUGP("CLUSTERIP: not mangling arp reply on different "
553 "interface: cip'%s'-skb'%s'\n", c
->dev
->name
, out
->name
);
554 clusterip_config_put(c
);
558 /* mangle reply hardware address */
559 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
561 #ifdef CLUSTERIP_DEBUG
562 DEBUGP(KERN_DEBUG
"CLUSTERIP mangled arp reply: ");
566 clusterip_config_put(c
);
571 static struct nf_hook_ops cip_arp_ops
= {
574 .hooknum
= NF_ARP_OUT
,
578 /***********************************************************************
580 ***********************************************************************/
582 #ifdef CONFIG_PROC_FS
584 struct clusterip_seq_position
{
585 unsigned int pos
; /* position */
586 unsigned int weight
; /* number of bits set == size */
587 unsigned int bit
; /* current bit */
588 unsigned long val
; /* current value */
591 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
593 struct proc_dir_entry
*pde
= s
->private;
594 struct clusterip_config
*c
= pde
->data
;
596 u_int32_t local_nodes
;
597 struct clusterip_seq_position
*idx
;
599 /* FIXME: possible race */
600 local_nodes
= c
->local_nodes
;
601 weight
= hweight32(local_nodes
);
605 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
607 return ERR_PTR(-ENOMEM
);
610 idx
->weight
= weight
;
611 idx
->bit
= ffs(local_nodes
);
612 idx
->val
= local_nodes
;
613 clear_bit(idx
->bit
- 1, &idx
->val
);
618 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
620 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
623 if (*pos
>= idx
->weight
) {
627 idx
->bit
= ffs(idx
->val
);
628 clear_bit(idx
->bit
- 1, &idx
->val
);
632 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
637 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
639 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
644 seq_printf(s
, "%u", idx
->bit
);
646 if (idx
->pos
== idx
->weight
- 1)
652 static struct seq_operations clusterip_seq_ops
= {
653 .start
= clusterip_seq_start
,
654 .next
= clusterip_seq_next
,
655 .stop
= clusterip_seq_stop
,
656 .show
= clusterip_seq_show
,
659 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
661 int ret
= seq_open(file
, &clusterip_seq_ops
);
664 struct seq_file
*sf
= file
->private_data
;
665 struct proc_dir_entry
*pde
= PDE(inode
);
666 struct clusterip_config
*c
= pde
->data
;
670 clusterip_config_get(c
);
676 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
678 struct proc_dir_entry
*pde
= PDE(inode
);
679 struct clusterip_config
*c
= pde
->data
;
682 ret
= seq_release(inode
, file
);
685 clusterip_config_put(c
);
690 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
691 size_t size
, loff_t
*ofs
)
693 #define PROC_WRITELEN 10
694 char buffer
[PROC_WRITELEN
+1];
695 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
696 struct clusterip_config
*c
= pde
->data
;
697 unsigned long nodenum
;
699 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
702 if (*buffer
== '+') {
703 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
704 if (clusterip_add_node(c
, nodenum
))
706 } else if (*buffer
== '-') {
707 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
708 if (clusterip_del_node(c
, nodenum
))
716 static const struct file_operations clusterip_proc_fops
= {
717 .owner
= THIS_MODULE
,
718 .open
= clusterip_proc_open
,
720 .write
= clusterip_proc_write
,
722 .release
= clusterip_proc_release
,
725 #endif /* CONFIG_PROC_FS */
727 static int __init
ipt_clusterip_init(void)
731 ret
= xt_register_target(&clusterip_tgt
);
735 ret
= nf_register_hook(&cip_arp_ops
);
739 #ifdef CONFIG_PROC_FS
740 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", proc_net
);
741 if (!clusterip_procdir
) {
742 printk(KERN_ERR
"CLUSTERIP: Unable to proc dir entry\n");
746 #endif /* CONFIG_PROC_FS */
748 printk(KERN_NOTICE
"ClusterIP Version %s loaded successfully\n",
752 #ifdef CONFIG_PROC_FS
754 nf_unregister_hook(&cip_arp_ops
);
755 #endif /* CONFIG_PROC_FS */
757 xt_unregister_target(&clusterip_tgt
);
761 static void __exit
ipt_clusterip_fini(void)
763 printk(KERN_NOTICE
"ClusterIP Version %s unloading\n",
765 #ifdef CONFIG_PROC_FS
766 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
768 nf_unregister_hook(&cip_arp_ops
);
769 xt_unregister_target(&clusterip_tgt
);
772 module_init(ipt_clusterip_init
);
773 module_exit(ipt_clusterip_fini
);