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_ipv4/ip_tables.h>
30 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
31 #include <net/netfilter/nf_conntrack_compat.h>
33 #define CLUSTERIP_VERSION "0.8"
35 #define DEBUG_CLUSTERIP
37 #ifdef DEBUG_CLUSTERIP
43 MODULE_LICENSE("GPL");
44 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
45 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
47 struct clusterip_config
{
48 struct list_head list
; /* list of all configs */
49 atomic_t refcount
; /* reference count */
50 atomic_t entries
; /* number of entries/rules
53 __be32 clusterip
; /* the IP address */
54 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
55 struct net_device
*dev
; /* device */
56 u_int16_t num_total_nodes
; /* total number of nodes */
57 unsigned long local_nodes
; /* node number array */
60 struct proc_dir_entry
*pde
; /* proc dir entry */
62 enum clusterip_hashmode hash_mode
; /* which hashing mode */
63 u_int32_t hash_initval
; /* hash initialization */
66 static LIST_HEAD(clusterip_configs
);
68 /* clusterip_lock protects the clusterip_configs list */
69 static DEFINE_RWLOCK(clusterip_lock
);
72 static struct file_operations clusterip_proc_fops
;
73 static struct proc_dir_entry
*clusterip_procdir
;
77 clusterip_config_get(struct clusterip_config
*c
)
79 atomic_inc(&c
->refcount
);
83 clusterip_config_put(struct clusterip_config
*c
)
85 if (atomic_dec_and_test(&c
->refcount
))
89 /* increase the count of entries(rules) using/referencing this config */
91 clusterip_config_entry_get(struct clusterip_config
*c
)
93 atomic_inc(&c
->entries
);
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 if (atomic_dec_and_test(&c
->entries
)) {
103 write_lock_bh(&clusterip_lock
);
105 write_unlock_bh(&clusterip_lock
);
107 dev_mc_delete(c
->dev
, c
->clustermac
, ETH_ALEN
, 0);
110 /* In case anyone still accesses the file, the open/close
111 * functions are also incrementing the refcount on their own,
112 * so it's safe to remove the entry even if it's in use. */
113 #ifdef CONFIG_PROC_FS
114 remove_proc_entry(c
->pde
->name
, c
->pde
->parent
);
119 static struct clusterip_config
*
120 __clusterip_config_find(__be32 clusterip
)
122 struct list_head
*pos
;
124 list_for_each(pos
, &clusterip_configs
) {
125 struct clusterip_config
*c
= list_entry(pos
,
126 struct clusterip_config
, list
);
127 if (c
->clusterip
== clusterip
) {
135 static inline struct clusterip_config
*
136 clusterip_config_find_get(__be32 clusterip
, int entry
)
138 struct clusterip_config
*c
;
140 read_lock_bh(&clusterip_lock
);
141 c
= __clusterip_config_find(clusterip
);
143 read_unlock_bh(&clusterip_lock
);
146 atomic_inc(&c
->refcount
);
148 atomic_inc(&c
->entries
);
149 read_unlock_bh(&clusterip_lock
);
155 clusterip_config_init_nodelist(struct clusterip_config
*c
,
156 const struct ipt_clusterip_tgt_info
*i
)
160 for (n
= 0; n
< i
->num_local_nodes
; n
++) {
161 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
165 static struct clusterip_config
*
166 clusterip_config_init(struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
167 struct net_device
*dev
)
169 struct clusterip_config
*c
;
171 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
177 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
178 c
->num_total_nodes
= i
->num_total_nodes
;
179 clusterip_config_init_nodelist(c
, i
);
180 c
->hash_mode
= i
->hash_mode
;
181 c
->hash_initval
= i
->hash_initval
;
182 atomic_set(&c
->refcount
, 1);
183 atomic_set(&c
->entries
, 1);
185 #ifdef CONFIG_PROC_FS
189 /* create proc dir entry */
190 sprintf(buffer
, "%u.%u.%u.%u", NIPQUAD(ip
));
191 c
->pde
= create_proc_entry(buffer
, S_IWUSR
|S_IRUSR
,
198 c
->pde
->proc_fops
= &clusterip_proc_fops
;
202 write_lock_bh(&clusterip_lock
);
203 list_add(&c
->list
, &clusterip_configs
);
204 write_unlock_bh(&clusterip_lock
);
209 #ifdef CONFIG_PROC_FS
211 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
215 nodenum
> c
->num_total_nodes
)
218 /* check if we already have this number in our bitfield */
219 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
226 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
229 nodenum
> c
->num_total_nodes
)
232 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
239 static inline u_int32_t
240 clusterip_hashfn(struct sk_buff
*skb
, struct clusterip_config
*config
)
242 struct iphdr
*iph
= skb
->nh
.iph
;
243 unsigned long hashval
;
244 u_int16_t sport
, dport
;
247 switch (iph
->protocol
) {
253 ports
= (void *)iph
+iph
->ihl
*4;
258 if (net_ratelimit()) {
259 printk(KERN_NOTICE
"CLUSTERIP: unknown protocol `%u'\n",
265 switch (config
->hash_mode
) {
266 case CLUSTERIP_HASHMODE_SIP
:
267 hashval
= jhash_1word(ntohl(iph
->saddr
),
268 config
->hash_initval
);
270 case CLUSTERIP_HASHMODE_SIP_SPT
:
271 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
272 config
->hash_initval
);
274 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
275 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
276 config
->hash_initval
);
279 /* to make gcc happy */
281 /* This cannot happen, unless the check function wasn't called
282 * at rule load time */
283 printk("CLUSTERIP: unknown mode `%u'\n", config
->hash_mode
);
288 /* node numbers are 1..n, not 0..n */
289 return ((hashval
% config
->num_total_nodes
)+1);
293 clusterip_responsible(struct clusterip_config
*config
, u_int32_t hash
)
295 return test_bit(hash
- 1, &config
->local_nodes
);
298 /***********************************************************************
300 ***********************************************************************/
303 target(struct sk_buff
**pskb
,
304 const struct net_device
*in
,
305 const struct net_device
*out
,
306 unsigned int hooknum
,
307 const struct xt_target
*target
,
308 const void *targinfo
)
310 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
311 enum ip_conntrack_info ctinfo
;
312 u_int32_t
*mark
, hash
;
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 mark
= nf_ct_get_mark((*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 ((*pskb
)->nh
.iph
->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
, *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");
412 clusterip_config_entry_get(cipinfo
->config
);
414 /* Case B: This is a new rule referring to an existing
415 * clusterip config. */
416 cipinfo
->config
= config
;
417 clusterip_config_entry_get(cipinfo
->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 ipt_target clusterip_tgt
= {
476 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
477 .checkentry
= checkentry
,
483 /***********************************************************************
485 ***********************************************************************/
487 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
489 u_int8_t src_hw
[ETH_ALEN
];
491 u_int8_t dst_hw
[ETH_ALEN
];
493 } __attribute__ ((packed
));
495 #ifdef CLUSTERIP_DEBUG
496 static void arp_print(struct arp_payload
*payload
)
498 #define HBUFFERLEN 30
499 char hbuffer
[HBUFFERLEN
];
501 const char hexbuf
[]= "0123456789abcdef";
503 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
504 hbuffer
[k
++]=hexbuf
[(payload
->src_hw
[j
]>>4)&15];
505 hbuffer
[k
++]=hexbuf
[payload
->src_hw
[j
]&15];
510 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
511 NIPQUAD(payload
->src_ip
), hbuffer
,
512 NIPQUAD(payload
->dst_ip
));
517 arp_mangle(unsigned int hook
,
518 struct sk_buff
**pskb
,
519 const struct net_device
*in
,
520 const struct net_device
*out
,
521 int (*okfn
)(struct sk_buff
*))
523 struct arphdr
*arp
= (*pskb
)->nh
.arph
;
524 struct arp_payload
*payload
;
525 struct clusterip_config
*c
;
527 /* we don't care about non-ethernet and non-ipv4 ARP */
528 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
)
529 || arp
->ar_pro
!= htons(ETH_P_IP
)
530 || arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
533 /* we only want to mangle arp requests and replies */
534 if (arp
->ar_op
!= htons(ARPOP_REPLY
)
535 && arp
->ar_op
!= htons(ARPOP_REQUEST
))
538 payload
= (void *)(arp
+1);
540 /* if there is no clusterip configuration for the arp reply's
541 * source ip, we don't want to mangle it */
542 c
= clusterip_config_find_get(payload
->src_ip
, 0);
546 /* normally the linux kernel always replies to arp queries of
547 * addresses on different interfacs. However, in the CLUSTERIP case
548 * this wouldn't work, since we didn't subscribe the mcast group on
549 * other interfaces */
551 DEBUGP("CLUSTERIP: not mangling arp reply on different "
552 "interface: cip'%s'-skb'%s'\n", c
->dev
->name
, out
->name
);
553 clusterip_config_put(c
);
557 /* mangle reply hardware address */
558 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
560 #ifdef CLUSTERIP_DEBUG
561 DEBUGP(KERN_DEBUG
"CLUSTERIP mangled arp reply: ");
565 clusterip_config_put(c
);
570 static struct nf_hook_ops cip_arp_ops
= {
573 .hooknum
= NF_ARP_OUT
,
577 /***********************************************************************
579 ***********************************************************************/
581 #ifdef CONFIG_PROC_FS
583 struct clusterip_seq_position
{
584 unsigned int pos
; /* position */
585 unsigned int weight
; /* number of bits set == size */
586 unsigned int bit
; /* current bit */
587 unsigned long val
; /* current value */
590 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
592 struct proc_dir_entry
*pde
= s
->private;
593 struct clusterip_config
*c
= pde
->data
;
595 u_int32_t local_nodes
;
596 struct clusterip_seq_position
*idx
;
598 /* FIXME: possible race */
599 local_nodes
= c
->local_nodes
;
600 weight
= hweight32(local_nodes
);
604 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
606 return ERR_PTR(-ENOMEM
);
609 idx
->weight
= weight
;
610 idx
->bit
= ffs(local_nodes
);
611 idx
->val
= local_nodes
;
612 clear_bit(idx
->bit
- 1, &idx
->val
);
617 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
619 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
622 if (*pos
>= idx
->weight
) {
626 idx
->bit
= ffs(idx
->val
);
627 clear_bit(idx
->bit
- 1, &idx
->val
);
631 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
636 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
638 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
643 seq_printf(s
, "%u", idx
->bit
);
645 if (idx
->pos
== idx
->weight
- 1)
651 static struct seq_operations clusterip_seq_ops
= {
652 .start
= clusterip_seq_start
,
653 .next
= clusterip_seq_next
,
654 .stop
= clusterip_seq_stop
,
655 .show
= clusterip_seq_show
,
658 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
660 int ret
= seq_open(file
, &clusterip_seq_ops
);
663 struct seq_file
*sf
= file
->private_data
;
664 struct proc_dir_entry
*pde
= PDE(inode
);
665 struct clusterip_config
*c
= pde
->data
;
669 clusterip_config_get(c
);
675 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
677 struct proc_dir_entry
*pde
= PDE(inode
);
678 struct clusterip_config
*c
= pde
->data
;
681 ret
= seq_release(inode
, file
);
684 clusterip_config_put(c
);
689 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
690 size_t size
, loff_t
*ofs
)
692 #define PROC_WRITELEN 10
693 char buffer
[PROC_WRITELEN
+1];
694 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
695 struct clusterip_config
*c
= pde
->data
;
696 unsigned long nodenum
;
698 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
701 if (*buffer
== '+') {
702 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
703 if (clusterip_add_node(c
, nodenum
))
705 } else if (*buffer
== '-') {
706 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
707 if (clusterip_del_node(c
, nodenum
))
715 static struct file_operations clusterip_proc_fops
= {
716 .owner
= THIS_MODULE
,
717 .open
= clusterip_proc_open
,
719 .write
= clusterip_proc_write
,
721 .release
= clusterip_proc_release
,
724 #endif /* CONFIG_PROC_FS */
726 static int __init
ipt_clusterip_init(void)
730 ret
= ipt_register_target(&clusterip_tgt
);
734 ret
= nf_register_hook(&cip_arp_ops
);
738 #ifdef CONFIG_PROC_FS
739 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", proc_net
);
740 if (!clusterip_procdir
) {
741 printk(KERN_ERR
"CLUSTERIP: Unable to proc dir entry\n");
745 #endif /* CONFIG_PROC_FS */
747 printk(KERN_NOTICE
"ClusterIP Version %s loaded successfully\n",
751 #ifdef CONFIG_PROC_FS
753 nf_unregister_hook(&cip_arp_ops
);
754 #endif /* CONFIG_PROC_FS */
756 ipt_unregister_target(&clusterip_tgt
);
760 static void __exit
ipt_clusterip_fini(void)
762 printk(KERN_NOTICE
"ClusterIP Version %s unloading\n",
764 #ifdef CONFIG_PROC_FS
765 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
767 nf_unregister_hook(&cip_arp_ops
);
768 ipt_unregister_target(&clusterip_tgt
);
771 module_init(ipt_clusterip_init
);
772 module_exit(ipt_clusterip_fini
);