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>
24 #include <linux/netfilter_arp.h>
25 #include <linux/netfilter/x_tables.h>
26 #include <linux/netfilter_ipv4/ip_tables.h>
27 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
28 #include <net/netfilter/nf_conntrack.h>
29 #include <net/checksum.h>
31 #define CLUSTERIP_VERSION "0.8"
33 #define DEBUG_CLUSTERIP
35 #ifdef DEBUG_CLUSTERIP
41 MODULE_LICENSE("GPL");
42 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
43 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
45 struct clusterip_config
{
46 struct list_head list
; /* list of all configs */
47 atomic_t refcount
; /* reference count */
48 atomic_t entries
; /* number of entries/rules
51 __be32 clusterip
; /* the IP address */
52 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
53 struct net_device
*dev
; /* device */
54 u_int16_t num_total_nodes
; /* total number of nodes */
55 unsigned long local_nodes
; /* node number array */
58 struct proc_dir_entry
*pde
; /* proc dir entry */
60 enum clusterip_hashmode hash_mode
; /* which hashing mode */
61 u_int32_t hash_initval
; /* hash initialization */
64 static LIST_HEAD(clusterip_configs
);
66 /* clusterip_lock protects the clusterip_configs list */
67 static DEFINE_RWLOCK(clusterip_lock
);
70 static const struct file_operations clusterip_proc_fops
;
71 static struct proc_dir_entry
*clusterip_procdir
;
75 clusterip_config_get(struct clusterip_config
*c
)
77 atomic_inc(&c
->refcount
);
81 clusterip_config_put(struct clusterip_config
*c
)
83 if (atomic_dec_and_test(&c
->refcount
))
87 /* increase the count of entries(rules) using/referencing this config */
89 clusterip_config_entry_get(struct clusterip_config
*c
)
91 atomic_inc(&c
->entries
);
94 /* decrease the count of entries using/referencing this config. If last
95 * entry(rule) is removed, remove the config from lists, but don't free it
96 * yet, since proc-files could still be holding references */
98 clusterip_config_entry_put(struct clusterip_config
*c
)
100 if (atomic_dec_and_test(&c
->entries
)) {
101 write_lock_bh(&clusterip_lock
);
103 write_unlock_bh(&clusterip_lock
);
105 dev_mc_delete(c
->dev
, c
->clustermac
, ETH_ALEN
, 0);
108 /* In case anyone still accesses the file, the open/close
109 * functions are also incrementing the refcount on their own,
110 * so it's safe to remove the entry even if it's in use. */
111 #ifdef CONFIG_PROC_FS
112 remove_proc_entry(c
->pde
->name
, c
->pde
->parent
);
117 static struct clusterip_config
*
118 __clusterip_config_find(__be32 clusterip
)
120 struct list_head
*pos
;
122 list_for_each(pos
, &clusterip_configs
) {
123 struct clusterip_config
*c
= list_entry(pos
,
124 struct clusterip_config
, list
);
125 if (c
->clusterip
== clusterip
) {
133 static inline struct clusterip_config
*
134 clusterip_config_find_get(__be32 clusterip
, int entry
)
136 struct clusterip_config
*c
;
138 read_lock_bh(&clusterip_lock
);
139 c
= __clusterip_config_find(clusterip
);
141 read_unlock_bh(&clusterip_lock
);
144 atomic_inc(&c
->refcount
);
146 atomic_inc(&c
->entries
);
147 read_unlock_bh(&clusterip_lock
);
153 clusterip_config_init_nodelist(struct clusterip_config
*c
,
154 const struct ipt_clusterip_tgt_info
*i
)
158 for (n
= 0; n
< i
->num_local_nodes
; n
++) {
159 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
163 static struct clusterip_config
*
164 clusterip_config_init(struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
165 struct net_device
*dev
)
167 struct clusterip_config
*c
;
169 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
175 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
176 c
->num_total_nodes
= i
->num_total_nodes
;
177 clusterip_config_init_nodelist(c
, i
);
178 c
->hash_mode
= i
->hash_mode
;
179 c
->hash_initval
= i
->hash_initval
;
180 atomic_set(&c
->refcount
, 1);
181 atomic_set(&c
->entries
, 1);
183 #ifdef CONFIG_PROC_FS
187 /* create proc dir entry */
188 sprintf(buffer
, "%u.%u.%u.%u", NIPQUAD(ip
));
189 c
->pde
= create_proc_entry(buffer
, S_IWUSR
|S_IRUSR
,
196 c
->pde
->proc_fops
= &clusterip_proc_fops
;
200 write_lock_bh(&clusterip_lock
);
201 list_add(&c
->list
, &clusterip_configs
);
202 write_unlock_bh(&clusterip_lock
);
207 #ifdef CONFIG_PROC_FS
209 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
213 nodenum
> c
->num_total_nodes
)
216 /* check if we already have this number in our bitfield */
217 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
224 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
227 nodenum
> c
->num_total_nodes
)
230 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
237 static inline u_int32_t
238 clusterip_hashfn(struct sk_buff
*skb
, struct clusterip_config
*config
)
240 struct iphdr
*iph
= ip_hdr(skb
);
241 unsigned long hashval
;
242 u_int16_t sport
, dport
;
245 switch (iph
->protocol
) {
248 case IPPROTO_UDPLITE
:
252 ports
= (void *)iph
+iph
->ihl
*4;
257 if (net_ratelimit()) {
258 printk(KERN_NOTICE
"CLUSTERIP: unknown protocol `%u'\n",
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 printk("CLUSTERIP: unknown mode `%u'\n", config
->hash_mode
);
287 /* node numbers are 1..n, not 0..n */
288 return ((hashval
% config
->num_total_nodes
)+1);
292 clusterip_responsible(struct clusterip_config
*config
, u_int32_t hash
)
294 return test_bit(hash
- 1, &config
->local_nodes
);
297 /***********************************************************************
299 ***********************************************************************/
302 target(struct sk_buff
**pskb
,
303 const struct net_device
*in
,
304 const struct net_device
*out
,
305 unsigned int hooknum
,
306 const struct xt_target
*target
,
307 const void *targinfo
)
309 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
311 enum ip_conntrack_info ctinfo
;
314 /* don't need to clusterip_config_get() here, since refcount
315 * is only decremented by destroy() - and ip_tables guarantees
316 * that the ->target() function isn't called after ->destroy() */
318 ct
= nf_ct_get(*pskb
, &ctinfo
);
320 printk(KERN_ERR
"CLUSTERIP: no conntrack!\n");
321 /* FIXME: need to drop invalid ones, since replies
322 * to outgoing connections of other nodes will be
323 * marked as INVALID */
327 /* special case: ICMP error handling. conntrack distinguishes between
328 * error messages (RELATED) and information requests (see below) */
329 if (ip_hdr(*pskb
)->protocol
== IPPROTO_ICMP
330 && (ctinfo
== IP_CT_RELATED
331 || ctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
))
334 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
335 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
336 * on, which all have an ID field [relevant for hashing]. */
338 hash
= clusterip_hashfn(*pskb
, cipinfo
->config
);
345 case IP_CT_RELATED
+IP_CT_IS_REPLY
:
346 /* FIXME: we don't handle expectations at the
347 * moment. they can arrive on a different node than
348 * the master connection (e.g. FTP passive mode) */
349 case IP_CT_ESTABLISHED
:
350 case IP_CT_ESTABLISHED
+IP_CT_IS_REPLY
:
356 #ifdef DEBUG_CLUSTERP
357 DUMP_TUPLE(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
359 DEBUGP("hash=%u ct_hash=%u ", hash
, ct
->mark
);
360 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
361 DEBUGP("not responsible\n");
364 DEBUGP("responsible\n");
366 /* despite being received via linklayer multicast, this is
367 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
368 (*pskb
)->pkt_type
= PACKET_HOST
;
374 checkentry(const char *tablename
,
376 const struct xt_target
*target
,
378 unsigned int hook_mask
)
380 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
381 const struct ipt_entry
*e
= e_void
;
383 struct clusterip_config
*config
;
385 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
386 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
387 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
388 printk(KERN_WARNING
"CLUSTERIP: unknown mode `%u'\n",
393 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff)
394 || e
->ip
.dst
.s_addr
== 0) {
395 printk(KERN_ERR
"CLUSTERIP: Please specify destination IP\n");
399 /* FIXME: further sanity checks */
401 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
403 if (cipinfo
->config
!= NULL
) {
404 /* Case A: This is an entry that gets reloaded, since
405 * it still has a cipinfo->config pointer. Simply
406 * increase the entry refcount and return */
407 if (cipinfo
->config
!= config
) {
408 printk(KERN_ERR
"CLUSTERIP: Reloaded entry "
409 "has invalid config pointer!\n");
413 /* Case B: This is a new rule referring to an existing
414 * clusterip config. */
415 cipinfo
->config
= config
;
418 /* Case C: This is a completely new clusterip config */
419 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
420 printk(KERN_WARNING
"CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e
->ip
.dst
.s_addr
));
423 struct net_device
*dev
;
425 if (e
->ip
.iniface
[0] == '\0') {
426 printk(KERN_WARNING
"CLUSTERIP: Please specify an interface name\n");
430 dev
= dev_get_by_name(e
->ip
.iniface
);
432 printk(KERN_WARNING
"CLUSTERIP: no such interface %s\n", e
->ip
.iniface
);
436 config
= clusterip_config_init(cipinfo
,
437 e
->ip
.dst
.s_addr
, dev
);
439 printk(KERN_WARNING
"CLUSTERIP: cannot allocate config\n");
443 dev_mc_add(config
->dev
,config
->clustermac
, ETH_ALEN
, 0);
445 cipinfo
->config
= config
;
448 if (nf_ct_l3proto_try_module_get(target
->family
) < 0) {
449 printk(KERN_WARNING
"can't load conntrack support for "
450 "proto=%d\n", target
->family
);
457 /* drop reference count of cluster config when rule is deleted */
458 static void destroy(const struct xt_target
*target
, void *targinfo
)
460 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
462 /* if no more entries are referencing the config, remove it
463 * from the list and destroy the proc entry */
464 clusterip_config_entry_put(cipinfo
->config
);
466 clusterip_config_put(cipinfo
->config
);
468 nf_ct_l3proto_module_put(target
->family
);
471 static struct xt_target clusterip_tgt
= {
475 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
476 .checkentry
= checkentry
,
482 /***********************************************************************
484 ***********************************************************************/
486 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
488 u_int8_t src_hw
[ETH_ALEN
];
490 u_int8_t dst_hw
[ETH_ALEN
];
492 } __attribute__ ((packed
));
494 #ifdef CLUSTERIP_DEBUG
495 static void arp_print(struct arp_payload
*payload
)
497 #define HBUFFERLEN 30
498 char hbuffer
[HBUFFERLEN
];
500 const char hexbuf
[]= "0123456789abcdef";
502 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
503 hbuffer
[k
++]=hexbuf
[(payload
->src_hw
[j
]>>4)&15];
504 hbuffer
[k
++]=hexbuf
[payload
->src_hw
[j
]&15];
509 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
510 NIPQUAD(payload
->src_ip
), hbuffer
,
511 NIPQUAD(payload
->dst_ip
));
516 arp_mangle(unsigned int hook
,
517 struct sk_buff
**pskb
,
518 const struct net_device
*in
,
519 const struct net_device
*out
,
520 int (*okfn
)(struct sk_buff
*))
522 struct arphdr
*arp
= arp_hdr(*pskb
);
523 struct arp_payload
*payload
;
524 struct clusterip_config
*c
;
526 /* we don't care about non-ethernet and non-ipv4 ARP */
527 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
)
528 || arp
->ar_pro
!= htons(ETH_P_IP
)
529 || arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
532 /* we only want to mangle arp requests and replies */
533 if (arp
->ar_op
!= htons(ARPOP_REPLY
)
534 && arp
->ar_op
!= htons(ARPOP_REQUEST
))
537 payload
= (void *)(arp
+1);
539 /* if there is no clusterip configuration for the arp reply's
540 * source ip, we don't want to mangle it */
541 c
= clusterip_config_find_get(payload
->src_ip
, 0);
545 /* normally the linux kernel always replies to arp queries of
546 * addresses on different interfacs. However, in the CLUSTERIP case
547 * this wouldn't work, since we didn't subscribe the mcast group on
548 * other interfaces */
550 DEBUGP("CLUSTERIP: not mangling arp reply on different "
551 "interface: cip'%s'-skb'%s'\n", c
->dev
->name
, out
->name
);
552 clusterip_config_put(c
);
556 /* mangle reply hardware address */
557 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
559 #ifdef CLUSTERIP_DEBUG
560 DEBUGP(KERN_DEBUG
"CLUSTERIP mangled arp reply: ");
564 clusterip_config_put(c
);
569 static struct nf_hook_ops cip_arp_ops
= {
572 .hooknum
= NF_ARP_OUT
,
576 /***********************************************************************
578 ***********************************************************************/
580 #ifdef CONFIG_PROC_FS
582 struct clusterip_seq_position
{
583 unsigned int pos
; /* position */
584 unsigned int weight
; /* number of bits set == size */
585 unsigned int bit
; /* current bit */
586 unsigned long val
; /* current value */
589 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
591 struct proc_dir_entry
*pde
= s
->private;
592 struct clusterip_config
*c
= pde
->data
;
594 u_int32_t local_nodes
;
595 struct clusterip_seq_position
*idx
;
597 /* FIXME: possible race */
598 local_nodes
= c
->local_nodes
;
599 weight
= hweight32(local_nodes
);
603 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
605 return ERR_PTR(-ENOMEM
);
608 idx
->weight
= weight
;
609 idx
->bit
= ffs(local_nodes
);
610 idx
->val
= local_nodes
;
611 clear_bit(idx
->bit
- 1, &idx
->val
);
616 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
618 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
621 if (*pos
>= idx
->weight
) {
625 idx
->bit
= ffs(idx
->val
);
626 clear_bit(idx
->bit
- 1, &idx
->val
);
630 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
635 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
637 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
642 seq_printf(s
, "%u", idx
->bit
);
644 if (idx
->pos
== idx
->weight
- 1)
650 static struct seq_operations clusterip_seq_ops
= {
651 .start
= clusterip_seq_start
,
652 .next
= clusterip_seq_next
,
653 .stop
= clusterip_seq_stop
,
654 .show
= clusterip_seq_show
,
657 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
659 int ret
= seq_open(file
, &clusterip_seq_ops
);
662 struct seq_file
*sf
= file
->private_data
;
663 struct proc_dir_entry
*pde
= PDE(inode
);
664 struct clusterip_config
*c
= pde
->data
;
668 clusterip_config_get(c
);
674 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
676 struct proc_dir_entry
*pde
= PDE(inode
);
677 struct clusterip_config
*c
= pde
->data
;
680 ret
= seq_release(inode
, file
);
683 clusterip_config_put(c
);
688 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
689 size_t size
, loff_t
*ofs
)
691 #define PROC_WRITELEN 10
692 char buffer
[PROC_WRITELEN
+1];
693 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
694 struct clusterip_config
*c
= pde
->data
;
695 unsigned long nodenum
;
697 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
700 if (*buffer
== '+') {
701 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
702 if (clusterip_add_node(c
, nodenum
))
704 } else if (*buffer
== '-') {
705 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
706 if (clusterip_del_node(c
, nodenum
))
714 static const struct file_operations clusterip_proc_fops
= {
715 .owner
= THIS_MODULE
,
716 .open
= clusterip_proc_open
,
718 .write
= clusterip_proc_write
,
720 .release
= clusterip_proc_release
,
723 #endif /* CONFIG_PROC_FS */
725 static int __init
ipt_clusterip_init(void)
729 ret
= xt_register_target(&clusterip_tgt
);
733 ret
= nf_register_hook(&cip_arp_ops
);
737 #ifdef CONFIG_PROC_FS
738 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", proc_net
);
739 if (!clusterip_procdir
) {
740 printk(KERN_ERR
"CLUSTERIP: Unable to proc dir entry\n");
744 #endif /* CONFIG_PROC_FS */
746 printk(KERN_NOTICE
"ClusterIP Version %s loaded successfully\n",
750 #ifdef CONFIG_PROC_FS
752 nf_unregister_hook(&cip_arp_ops
);
753 #endif /* CONFIG_PROC_FS */
755 xt_unregister_target(&clusterip_tgt
);
759 static void __exit
ipt_clusterip_fini(void)
761 printk(KERN_NOTICE
"ClusterIP Version %s unloading\n",
763 #ifdef CONFIG_PROC_FS
764 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
766 nf_unregister_hook(&cip_arp_ops
);
767 xt_unregister_target(&clusterip_tgt
);
770 module_init(ipt_clusterip_init
);
771 module_exit(ipt_clusterip_fini
);