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 #define ASSERT_READ_LOCK(x)
45 MODULE_LICENSE("GPL");
46 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
47 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
49 struct clusterip_config
{
50 struct list_head list
; /* list of all configs */
51 atomic_t refcount
; /* reference count */
52 atomic_t entries
; /* number of entries/rules
55 u_int32_t clusterip
; /* the IP address */
56 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
57 struct net_device
*dev
; /* device */
58 u_int16_t num_total_nodes
; /* total number of nodes */
59 unsigned long local_nodes
; /* node number array */
62 struct proc_dir_entry
*pde
; /* proc dir entry */
64 enum clusterip_hashmode hash_mode
; /* which hashing mode */
65 u_int32_t hash_initval
; /* hash initialization */
68 static LIST_HEAD(clusterip_configs
);
70 /* clusterip_lock protects the clusterip_configs list */
71 static DEFINE_RWLOCK(clusterip_lock
);
74 static struct file_operations clusterip_proc_fops
;
75 static struct proc_dir_entry
*clusterip_procdir
;
79 clusterip_config_get(struct clusterip_config
*c
)
81 atomic_inc(&c
->refcount
);
85 clusterip_config_put(struct clusterip_config
*c
)
87 if (atomic_dec_and_test(&c
->refcount
))
91 /* increase the count of entries(rules) using/referencing this config */
93 clusterip_config_entry_get(struct clusterip_config
*c
)
95 atomic_inc(&c
->entries
);
98 /* decrease the count of entries using/referencing this config. If last
99 * entry(rule) is removed, remove the config from lists, but don't free it
100 * yet, since proc-files could still be holding references */
102 clusterip_config_entry_put(struct clusterip_config
*c
)
104 if (atomic_dec_and_test(&c
->entries
)) {
105 write_lock_bh(&clusterip_lock
);
107 write_unlock_bh(&clusterip_lock
);
109 dev_mc_delete(c
->dev
, c
->clustermac
, ETH_ALEN
, 0);
112 /* In case anyone still accesses the file, the open/close
113 * functions are also incrementing the refcount on their own,
114 * so it's safe to remove the entry even if it's in use. */
115 #ifdef CONFIG_PROC_FS
116 remove_proc_entry(c
->pde
->name
, c
->pde
->parent
);
121 static struct clusterip_config
*
122 __clusterip_config_find(u_int32_t clusterip
)
124 struct list_head
*pos
;
126 ASSERT_READ_LOCK(&clusterip_lock
);
127 list_for_each(pos
, &clusterip_configs
) {
128 struct clusterip_config
*c
= list_entry(pos
,
129 struct clusterip_config
, list
);
130 if (c
->clusterip
== clusterip
) {
138 static inline struct clusterip_config
*
139 clusterip_config_find_get(u_int32_t clusterip
, int entry
)
141 struct clusterip_config
*c
;
143 read_lock_bh(&clusterip_lock
);
144 c
= __clusterip_config_find(clusterip
);
146 read_unlock_bh(&clusterip_lock
);
149 atomic_inc(&c
->refcount
);
151 atomic_inc(&c
->entries
);
152 read_unlock_bh(&clusterip_lock
);
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
);
168 static struct clusterip_config
*
169 clusterip_config_init(struct ipt_clusterip_tgt_info
*i
, u_int32_t ip
,
170 struct net_device
*dev
)
172 struct clusterip_config
*c
;
175 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
181 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
182 c
->num_total_nodes
= i
->num_total_nodes
;
183 clusterip_config_init_nodelist(c
, i
);
184 c
->hash_mode
= i
->hash_mode
;
185 c
->hash_initval
= i
->hash_initval
;
186 atomic_set(&c
->refcount
, 1);
187 atomic_set(&c
->entries
, 1);
189 #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
, clusterip_procdir
);
197 c
->pde
->proc_fops
= &clusterip_proc_fops
;
201 write_lock_bh(&clusterip_lock
);
202 list_add(&c
->list
, &clusterip_configs
);
203 write_unlock_bh(&clusterip_lock
);
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
))
236 static inline u_int32_t
237 clusterip_hashfn(struct sk_buff
*skb
, struct clusterip_config
*config
)
239 struct iphdr
*iph
= skb
->nh
.iph
;
240 unsigned long hashval
;
241 u_int16_t sport
, dport
;
244 switch (iph
->protocol
) {
250 ports
= (void *)iph
+iph
->ihl
*4;
255 if (net_ratelimit()) {
256 printk(KERN_NOTICE
"CLUSTERIP: unknown protocol `%u'\n",
262 switch (config
->hash_mode
) {
263 case CLUSTERIP_HASHMODE_SIP
:
264 hashval
= jhash_1word(ntohl(iph
->saddr
),
265 config
->hash_initval
);
267 case CLUSTERIP_HASHMODE_SIP_SPT
:
268 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
269 config
->hash_initval
);
271 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
272 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
273 config
->hash_initval
);
276 /* to make gcc happy */
278 /* This cannot happen, unless the check function wasn't called
279 * at rule load time */
280 printk("CLUSTERIP: unknown mode `%u'\n", config
->hash_mode
);
285 /* node numbers are 1..n, not 0..n */
286 return ((hashval
% config
->num_total_nodes
)+1);
290 clusterip_responsible(struct clusterip_config
*config
, u_int32_t hash
)
292 return test_bit(hash
- 1, &config
->local_nodes
);
295 /***********************************************************************
297 ***********************************************************************/
300 target(struct sk_buff
**pskb
,
301 const struct net_device
*in
,
302 const struct net_device
*out
,
303 unsigned int hooknum
,
304 const struct xt_target
*target
,
305 const void *targinfo
)
307 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
308 enum ip_conntrack_info ctinfo
;
309 u_int32_t
*mark
, hash
;
311 /* don't need to clusterip_config_get() here, since refcount
312 * is only decremented by destroy() - and ip_tables guarantees
313 * that the ->target() function isn't called after ->destroy() */
315 mark
= nf_ct_get_mark((*pskb
), &ctinfo
);
317 printk(KERN_ERR
"CLUSTERIP: no conntrack!\n");
318 /* FIXME: need to drop invalid ones, since replies
319 * to outgoing connections of other nodes will be
320 * marked as INVALID */
324 /* special case: ICMP error handling. conntrack distinguishes between
325 * error messages (RELATED) and information requests (see below) */
326 if ((*pskb
)->nh
.iph
->protocol
== IPPROTO_ICMP
327 && (ctinfo
== IP_CT_RELATED
328 || ctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
))
331 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
332 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
333 * on, which all have an ID field [relevant for hashing]. */
335 hash
= clusterip_hashfn(*pskb
, cipinfo
->config
);
342 case IP_CT_RELATED
+IP_CT_IS_REPLY
:
343 /* FIXME: we don't handle expectations at the
344 * moment. they can arrive on a different node than
345 * the master connection (e.g. FTP passive mode) */
346 case IP_CT_ESTABLISHED
:
347 case IP_CT_ESTABLISHED
+IP_CT_IS_REPLY
:
353 #ifdef DEBUG_CLUSTERP
354 DUMP_TUPLE(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
356 DEBUGP("hash=%u ct_hash=%u ", hash
, *mark
);
357 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
358 DEBUGP("not responsible\n");
361 DEBUGP("responsible\n");
363 /* despite being received via linklayer multicast, this is
364 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
365 (*pskb
)->pkt_type
= PACKET_HOST
;
371 checkentry(const char *tablename
,
373 const struct xt_target
*target
,
375 unsigned int hook_mask
)
377 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
378 const struct ipt_entry
*e
= e_void
;
380 struct clusterip_config
*config
;
382 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
383 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
384 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
385 printk(KERN_WARNING
"CLUSTERIP: unknown mode `%u'\n",
390 if (e
->ip
.dmsk
.s_addr
!= 0xffffffff
391 || e
->ip
.dst
.s_addr
== 0) {
392 printk(KERN_ERR
"CLUSTERIP: Please specify destination IP\n");
396 /* FIXME: further sanity checks */
398 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
400 if (cipinfo
->config
!= NULL
) {
401 /* Case A: This is an entry that gets reloaded, since
402 * it still has a cipinfo->config pointer. Simply
403 * increase the entry refcount and return */
404 if (cipinfo
->config
!= config
) {
405 printk(KERN_ERR
"CLUSTERIP: Reloaded entry "
406 "has invalid config pointer!\n");
409 clusterip_config_entry_get(cipinfo
->config
);
411 /* Case B: This is a new rule referring to an existing
412 * clusterip config. */
413 cipinfo
->config
= config
;
414 clusterip_config_entry_get(cipinfo
->config
);
417 /* Case C: This is a completely new clusterip config */
418 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
419 printk(KERN_WARNING
"CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e
->ip
.dst
.s_addr
));
422 struct net_device
*dev
;
424 if (e
->ip
.iniface
[0] == '\0') {
425 printk(KERN_WARNING
"CLUSTERIP: Please specify an interface name\n");
429 dev
= dev_get_by_name(e
->ip
.iniface
);
431 printk(KERN_WARNING
"CLUSTERIP: no such interface %s\n", e
->ip
.iniface
);
435 config
= clusterip_config_init(cipinfo
,
436 e
->ip
.dst
.s_addr
, dev
);
438 printk(KERN_WARNING
"CLUSTERIP: cannot allocate config\n");
442 dev_mc_add(config
->dev
,config
->clustermac
, ETH_ALEN
, 0);
444 cipinfo
->config
= config
;
450 /* drop reference count of cluster config when rule is deleted */
451 static void destroy(const struct xt_target
*target
, void *targinfo
)
453 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
455 /* if no more entries are referencing the config, remove it
456 * from the list and destroy the proc entry */
457 clusterip_config_entry_put(cipinfo
->config
);
459 clusterip_config_put(cipinfo
->config
);
462 static struct ipt_target clusterip_tgt
= {
465 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
466 .checkentry
= checkentry
,
472 /***********************************************************************
474 ***********************************************************************/
476 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
478 u_int8_t src_hw
[ETH_ALEN
];
480 u_int8_t dst_hw
[ETH_ALEN
];
482 } __attribute__ ((packed
));
484 #ifdef CLUSTERIP_DEBUG
485 static void arp_print(struct arp_payload
*payload
)
487 #define HBUFFERLEN 30
488 char hbuffer
[HBUFFERLEN
];
490 const char hexbuf
[]= "0123456789abcdef";
492 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
493 hbuffer
[k
++]=hexbuf
[(payload
->src_hw
[j
]>>4)&15];
494 hbuffer
[k
++]=hexbuf
[payload
->src_hw
[j
]&15];
499 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
500 NIPQUAD(payload
->src_ip
), hbuffer
,
501 NIPQUAD(payload
->dst_ip
));
506 arp_mangle(unsigned int hook
,
507 struct sk_buff
**pskb
,
508 const struct net_device
*in
,
509 const struct net_device
*out
,
510 int (*okfn
)(struct sk_buff
*))
512 struct arphdr
*arp
= (*pskb
)->nh
.arph
;
513 struct arp_payload
*payload
;
514 struct clusterip_config
*c
;
516 /* we don't care about non-ethernet and non-ipv4 ARP */
517 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
)
518 || arp
->ar_pro
!= htons(ETH_P_IP
)
519 || arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
522 /* we only want to mangle arp requests and replies */
523 if (arp
->ar_op
!= htons(ARPOP_REPLY
)
524 && arp
->ar_op
!= htons(ARPOP_REQUEST
))
527 payload
= (void *)(arp
+1);
529 /* if there is no clusterip configuration for the arp reply's
530 * source ip, we don't want to mangle it */
531 c
= clusterip_config_find_get(payload
->src_ip
, 0);
535 /* normally the linux kernel always replies to arp queries of
536 * addresses on different interfacs. However, in the CLUSTERIP case
537 * this wouldn't work, since we didn't subscribe the mcast group on
538 * other interfaces */
540 DEBUGP("CLUSTERIP: not mangling arp reply on different "
541 "interface: cip'%s'-skb'%s'\n", c
->dev
->name
, out
->name
);
542 clusterip_config_put(c
);
546 /* mangle reply hardware address */
547 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
549 #ifdef CLUSTERIP_DEBUG
550 DEBUGP(KERN_DEBUG
"CLUSTERIP mangled arp reply: ");
554 clusterip_config_put(c
);
559 static struct nf_hook_ops cip_arp_ops
= {
562 .hooknum
= NF_ARP_OUT
,
566 /***********************************************************************
568 ***********************************************************************/
570 #ifdef CONFIG_PROC_FS
572 struct clusterip_seq_position
{
573 unsigned int pos
; /* position */
574 unsigned int weight
; /* number of bits set == size */
575 unsigned int bit
; /* current bit */
576 unsigned long val
; /* current value */
579 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
581 struct proc_dir_entry
*pde
= s
->private;
582 struct clusterip_config
*c
= pde
->data
;
584 u_int32_t local_nodes
;
585 struct clusterip_seq_position
*idx
;
587 /* FIXME: possible race */
588 local_nodes
= c
->local_nodes
;
589 weight
= hweight32(local_nodes
);
593 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
595 return ERR_PTR(-ENOMEM
);
598 idx
->weight
= weight
;
599 idx
->bit
= ffs(local_nodes
);
600 idx
->val
= local_nodes
;
601 clear_bit(idx
->bit
- 1, &idx
->val
);
606 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
608 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
611 if (*pos
>= idx
->weight
) {
615 idx
->bit
= ffs(idx
->val
);
616 clear_bit(idx
->bit
- 1, &idx
->val
);
620 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
625 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
627 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
632 seq_printf(s
, "%u", idx
->bit
);
634 if (idx
->pos
== idx
->weight
- 1)
640 static struct seq_operations clusterip_seq_ops
= {
641 .start
= clusterip_seq_start
,
642 .next
= clusterip_seq_next
,
643 .stop
= clusterip_seq_stop
,
644 .show
= clusterip_seq_show
,
647 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
649 int ret
= seq_open(file
, &clusterip_seq_ops
);
652 struct seq_file
*sf
= file
->private_data
;
653 struct proc_dir_entry
*pde
= PDE(inode
);
654 struct clusterip_config
*c
= pde
->data
;
658 clusterip_config_get(c
);
664 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
666 struct proc_dir_entry
*pde
= PDE(inode
);
667 struct clusterip_config
*c
= pde
->data
;
670 ret
= seq_release(inode
, file
);
673 clusterip_config_put(c
);
678 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
679 size_t size
, loff_t
*ofs
)
681 #define PROC_WRITELEN 10
682 char buffer
[PROC_WRITELEN
+1];
683 struct proc_dir_entry
*pde
= PDE(file
->f_dentry
->d_inode
);
684 struct clusterip_config
*c
= pde
->data
;
685 unsigned long nodenum
;
687 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
690 if (*buffer
== '+') {
691 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
692 if (clusterip_add_node(c
, nodenum
))
694 } else if (*buffer
== '-') {
695 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
696 if (clusterip_del_node(c
, nodenum
))
704 static struct file_operations clusterip_proc_fops
= {
705 .owner
= THIS_MODULE
,
706 .open
= clusterip_proc_open
,
708 .write
= clusterip_proc_write
,
710 .release
= clusterip_proc_release
,
713 #endif /* CONFIG_PROC_FS */
715 static int __init
ipt_clusterip_init(void)
719 ret
= ipt_register_target(&clusterip_tgt
);
723 ret
= nf_register_hook(&cip_arp_ops
);
727 #ifdef CONFIG_PROC_FS
728 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", proc_net
);
729 if (!clusterip_procdir
) {
730 printk(KERN_ERR
"CLUSTERIP: Unable to proc dir entry\n");
734 #endif /* CONFIG_PROC_FS */
736 printk(KERN_NOTICE
"ClusterIP Version %s loaded successfully\n",
741 nf_unregister_hook(&cip_arp_ops
);
743 ipt_unregister_target(&clusterip_tgt
);
747 static void __exit
ipt_clusterip_fini(void)
749 printk(KERN_NOTICE
"ClusterIP Version %s unloading\n",
751 #ifdef CONFIG_PROC_FS
752 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
754 nf_unregister_hook(&cip_arp_ops
);
755 ipt_unregister_target(&clusterip_tgt
);
758 module_init(ipt_clusterip_init
);
759 module_exit(ipt_clusterip_fini
);