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/seq_file.h>
23 #include <linux/netfilter_arp.h>
24 #include <linux/netfilter/x_tables.h>
25 #include <linux/netfilter_ipv4/ip_tables.h>
26 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
27 #include <net/netfilter/nf_conntrack.h>
28 #include <net/net_namespace.h>
29 #include <net/checksum.h>
31 #define CLUSTERIP_VERSION "0.8"
33 MODULE_LICENSE("GPL");
34 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
35 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
37 struct clusterip_config
{
38 struct list_head list
; /* list of all configs */
39 atomic_t refcount
; /* reference count */
40 atomic_t entries
; /* number of entries/rules
43 __be32 clusterip
; /* the IP address */
44 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
45 struct net_device
*dev
; /* device */
46 u_int16_t num_total_nodes
; /* total number of nodes */
47 unsigned long local_nodes
; /* node number array */
50 struct proc_dir_entry
*pde
; /* proc dir entry */
52 enum clusterip_hashmode hash_mode
; /* which hashing mode */
53 u_int32_t hash_initval
; /* hash initialization */
56 static LIST_HEAD(clusterip_configs
);
58 /* clusterip_lock protects the clusterip_configs list */
59 static DEFINE_RWLOCK(clusterip_lock
);
62 static const struct file_operations clusterip_proc_fops
;
63 static struct proc_dir_entry
*clusterip_procdir
;
67 clusterip_config_get(struct clusterip_config
*c
)
69 atomic_inc(&c
->refcount
);
73 clusterip_config_put(struct clusterip_config
*c
)
75 if (atomic_dec_and_test(&c
->refcount
))
79 /* decrease the count of entries using/referencing this config. If last
80 * entry(rule) is removed, remove the config from lists, but don't free it
81 * yet, since proc-files could still be holding references */
83 clusterip_config_entry_put(struct clusterip_config
*c
)
85 if (atomic_dec_and_test(&c
->entries
)) {
86 write_lock_bh(&clusterip_lock
);
88 write_unlock_bh(&clusterip_lock
);
90 dev_mc_delete(c
->dev
, c
->clustermac
, ETH_ALEN
, 0);
93 /* In case anyone still accesses the file, the open/close
94 * functions are also incrementing the refcount on their own,
95 * so it's safe to remove the entry even if it's in use. */
97 remove_proc_entry(c
->pde
->name
, c
->pde
->parent
);
102 static struct clusterip_config
*
103 __clusterip_config_find(__be32 clusterip
)
105 struct clusterip_config
*c
;
107 list_for_each_entry(c
, &clusterip_configs
, list
) {
108 if (c
->clusterip
== clusterip
)
115 static inline struct clusterip_config
*
116 clusterip_config_find_get(__be32 clusterip
, int entry
)
118 struct clusterip_config
*c
;
120 read_lock_bh(&clusterip_lock
);
121 c
= __clusterip_config_find(clusterip
);
123 read_unlock_bh(&clusterip_lock
);
126 atomic_inc(&c
->refcount
);
128 atomic_inc(&c
->entries
);
129 read_unlock_bh(&clusterip_lock
);
135 clusterip_config_init_nodelist(struct clusterip_config
*c
,
136 const struct ipt_clusterip_tgt_info
*i
)
140 for (n
= 0; n
< i
->num_local_nodes
; n
++)
141 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
144 static struct clusterip_config
*
145 clusterip_config_init(struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
146 struct net_device
*dev
)
148 struct clusterip_config
*c
;
150 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
156 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
157 c
->num_total_nodes
= i
->num_total_nodes
;
158 clusterip_config_init_nodelist(c
, i
);
159 c
->hash_mode
= i
->hash_mode
;
160 c
->hash_initval
= i
->hash_initval
;
161 atomic_set(&c
->refcount
, 1);
162 atomic_set(&c
->entries
, 1);
164 #ifdef CONFIG_PROC_FS
168 /* create proc dir entry */
169 sprintf(buffer
, "%u.%u.%u.%u", NIPQUAD(ip
));
170 c
->pde
= create_proc_entry(buffer
, S_IWUSR
|S_IRUSR
,
177 c
->pde
->proc_fops
= &clusterip_proc_fops
;
181 write_lock_bh(&clusterip_lock
);
182 list_add(&c
->list
, &clusterip_configs
);
183 write_unlock_bh(&clusterip_lock
);
188 #ifdef CONFIG_PROC_FS
190 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
194 nodenum
> c
->num_total_nodes
)
197 /* check if we already have this number in our bitfield */
198 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
205 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
208 nodenum
> c
->num_total_nodes
)
211 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
218 static inline u_int32_t
219 clusterip_hashfn(const struct sk_buff
*skb
,
220 const struct clusterip_config
*config
)
222 const struct iphdr
*iph
= ip_hdr(skb
);
223 unsigned long hashval
;
224 u_int16_t sport
, dport
;
225 const u_int16_t
*ports
;
227 switch (iph
->protocol
) {
230 case IPPROTO_UDPLITE
:
234 ports
= (const void *)iph
+iph
->ihl
*4;
240 printk(KERN_NOTICE
"CLUSTERIP: unknown protocol `%u'\n",
245 switch (config
->hash_mode
) {
246 case CLUSTERIP_HASHMODE_SIP
:
247 hashval
= jhash_1word(ntohl(iph
->saddr
),
248 config
->hash_initval
);
250 case CLUSTERIP_HASHMODE_SIP_SPT
:
251 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
252 config
->hash_initval
);
254 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
255 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
256 config
->hash_initval
);
259 /* to make gcc happy */
261 /* This cannot happen, unless the check function wasn't called
262 * at rule load time */
263 printk("CLUSTERIP: unknown mode `%u'\n", config
->hash_mode
);
268 /* node numbers are 1..n, not 0..n */
269 return (((u64
)hashval
* config
->num_total_nodes
) >> 32) + 1;
273 clusterip_responsible(const struct clusterip_config
*config
, u_int32_t hash
)
275 return test_bit(hash
- 1, &config
->local_nodes
);
278 /***********************************************************************
280 ***********************************************************************/
283 clusterip_tg(struct sk_buff
*skb
, const struct net_device
*in
,
284 const struct net_device
*out
, unsigned int hooknum
,
285 const struct xt_target
*target
, const void *targinfo
)
287 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
289 enum ip_conntrack_info ctinfo
;
292 /* don't need to clusterip_config_get() here, since refcount
293 * is only decremented by destroy() - and ip_tables guarantees
294 * that the ->target() function isn't called after ->destroy() */
296 ct
= nf_ct_get(skb
, &ctinfo
);
298 printk(KERN_ERR
"CLUSTERIP: no conntrack!\n");
299 /* FIXME: need to drop invalid ones, since replies
300 * to outgoing connections of other nodes will be
301 * marked as INVALID */
305 /* special case: ICMP error handling. conntrack distinguishes between
306 * error messages (RELATED) and information requests (see below) */
307 if (ip_hdr(skb
)->protocol
== IPPROTO_ICMP
308 && (ctinfo
== IP_CT_RELATED
309 || ctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
))
312 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
313 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
314 * on, which all have an ID field [relevant for hashing]. */
316 hash
= clusterip_hashfn(skb
, cipinfo
->config
);
323 case IP_CT_RELATED
+IP_CT_IS_REPLY
:
324 /* FIXME: we don't handle expectations at the
325 * moment. they can arrive on a different node than
326 * the master connection (e.g. FTP passive mode) */
327 case IP_CT_ESTABLISHED
:
328 case IP_CT_ESTABLISHED
+IP_CT_IS_REPLY
:
335 DUMP_TUPLE(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
337 pr_debug("hash=%u ct_hash=%u ", hash
, ct
->mark
);
338 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
339 pr_debug("not responsible\n");
342 pr_debug("responsible\n");
344 /* despite being received via linklayer multicast, this is
345 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
346 skb
->pkt_type
= PACKET_HOST
;
352 clusterip_tg_check(const char *tablename
, const void *e_void
,
353 const struct xt_target
*target
, void *targinfo
,
354 unsigned int hook_mask
)
356 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
357 const struct ipt_entry
*e
= e_void
;
359 struct clusterip_config
*config
;
361 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
362 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
363 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
364 printk(KERN_WARNING
"CLUSTERIP: unknown mode `%u'\n",
369 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff)
370 || e
->ip
.dst
.s_addr
== 0) {
371 printk(KERN_ERR
"CLUSTERIP: Please specify destination IP\n");
375 /* FIXME: further sanity checks */
377 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
379 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
380 printk(KERN_WARNING
"CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e
->ip
.dst
.s_addr
));
383 struct net_device
*dev
;
385 if (e
->ip
.iniface
[0] == '\0') {
386 printk(KERN_WARNING
"CLUSTERIP: Please specify an interface name\n");
390 dev
= dev_get_by_name(&init_net
, e
->ip
.iniface
);
392 printk(KERN_WARNING
"CLUSTERIP: no such interface %s\n", e
->ip
.iniface
);
396 config
= clusterip_config_init(cipinfo
,
397 e
->ip
.dst
.s_addr
, dev
);
399 printk(KERN_WARNING
"CLUSTERIP: cannot allocate config\n");
403 dev_mc_add(config
->dev
,config
->clustermac
, ETH_ALEN
, 0);
406 cipinfo
->config
= config
;
408 if (nf_ct_l3proto_try_module_get(target
->family
) < 0) {
409 printk(KERN_WARNING
"can't load conntrack support for "
410 "proto=%u\n", target
->family
);
417 /* drop reference count of cluster config when rule is deleted */
418 static void clusterip_tg_destroy(const struct xt_target
*target
, void *targinfo
)
420 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
422 /* if no more entries are referencing the config, remove it
423 * from the list and destroy the proc entry */
424 clusterip_config_entry_put(cipinfo
->config
);
426 clusterip_config_put(cipinfo
->config
);
428 nf_ct_l3proto_module_put(target
->family
);
432 struct compat_ipt_clusterip_tgt_info
435 u_int8_t clustermac
[6];
436 u_int16_t num_total_nodes
;
437 u_int16_t num_local_nodes
;
438 u_int16_t local_nodes
[CLUSTERIP_MAX_NODES
];
440 u_int32_t hash_initval
;
441 compat_uptr_t config
;
443 #endif /* CONFIG_COMPAT */
445 static struct xt_target clusterip_tg_reg __read_mostly
= {
448 .target
= clusterip_tg
,
449 .checkentry
= clusterip_tg_check
,
450 .destroy
= clusterip_tg_destroy
,
451 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
453 .compatsize
= sizeof(struct compat_ipt_clusterip_tgt_info
),
454 #endif /* CONFIG_COMPAT */
459 /***********************************************************************
461 ***********************************************************************/
463 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
465 u_int8_t src_hw
[ETH_ALEN
];
467 u_int8_t dst_hw
[ETH_ALEN
];
469 } __attribute__ ((packed
));
472 static void arp_print(struct arp_payload
*payload
)
474 #define HBUFFERLEN 30
475 char hbuffer
[HBUFFERLEN
];
477 const char hexbuf
[]= "0123456789abcdef";
479 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
480 hbuffer
[k
++]=hexbuf
[(payload
->src_hw
[j
]>>4)&15];
481 hbuffer
[k
++]=hexbuf
[payload
->src_hw
[j
]&15];
486 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
487 NIPQUAD(payload
->src_ip
), hbuffer
,
488 NIPQUAD(payload
->dst_ip
));
493 arp_mangle(unsigned int hook
,
495 const struct net_device
*in
,
496 const struct net_device
*out
,
497 int (*okfn
)(struct sk_buff
*))
499 struct arphdr
*arp
= arp_hdr(skb
);
500 struct arp_payload
*payload
;
501 struct clusterip_config
*c
;
503 /* we don't care about non-ethernet and non-ipv4 ARP */
504 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
)
505 || arp
->ar_pro
!= htons(ETH_P_IP
)
506 || arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
509 /* we only want to mangle arp requests and replies */
510 if (arp
->ar_op
!= htons(ARPOP_REPLY
)
511 && arp
->ar_op
!= htons(ARPOP_REQUEST
))
514 payload
= (void *)(arp
+1);
516 /* if there is no clusterip configuration for the arp reply's
517 * source ip, we don't want to mangle it */
518 c
= clusterip_config_find_get(payload
->src_ip
, 0);
522 /* normally the linux kernel always replies to arp queries of
523 * addresses on different interfacs. However, in the CLUSTERIP case
524 * this wouldn't work, since we didn't subscribe the mcast group on
525 * other interfaces */
527 pr_debug("CLUSTERIP: not mangling arp reply on different "
528 "interface: cip'%s'-skb'%s'\n",
529 c
->dev
->name
, out
->name
);
530 clusterip_config_put(c
);
534 /* mangle reply hardware address */
535 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
538 pr_debug(KERN_DEBUG
"CLUSTERIP mangled arp reply: ");
542 clusterip_config_put(c
);
547 static struct nf_hook_ops cip_arp_ops __read_mostly
= {
550 .hooknum
= NF_ARP_OUT
,
554 /***********************************************************************
556 ***********************************************************************/
558 #ifdef CONFIG_PROC_FS
560 struct clusterip_seq_position
{
561 unsigned int pos
; /* position */
562 unsigned int weight
; /* number of bits set == size */
563 unsigned int bit
; /* current bit */
564 unsigned long val
; /* current value */
567 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
569 struct proc_dir_entry
*pde
= s
->private;
570 struct clusterip_config
*c
= pde
->data
;
572 u_int32_t local_nodes
;
573 struct clusterip_seq_position
*idx
;
575 /* FIXME: possible race */
576 local_nodes
= c
->local_nodes
;
577 weight
= hweight32(local_nodes
);
581 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
583 return ERR_PTR(-ENOMEM
);
586 idx
->weight
= weight
;
587 idx
->bit
= ffs(local_nodes
);
588 idx
->val
= local_nodes
;
589 clear_bit(idx
->bit
- 1, &idx
->val
);
594 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
596 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
599 if (*pos
>= idx
->weight
) {
603 idx
->bit
= ffs(idx
->val
);
604 clear_bit(idx
->bit
- 1, &idx
->val
);
608 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
613 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
615 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
620 seq_printf(s
, "%u", idx
->bit
);
622 if (idx
->pos
== idx
->weight
- 1)
628 static const struct seq_operations clusterip_seq_ops
= {
629 .start
= clusterip_seq_start
,
630 .next
= clusterip_seq_next
,
631 .stop
= clusterip_seq_stop
,
632 .show
= clusterip_seq_show
,
635 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
637 int ret
= seq_open(file
, &clusterip_seq_ops
);
640 struct seq_file
*sf
= file
->private_data
;
641 struct proc_dir_entry
*pde
= PDE(inode
);
642 struct clusterip_config
*c
= pde
->data
;
646 clusterip_config_get(c
);
652 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
654 struct proc_dir_entry
*pde
= PDE(inode
);
655 struct clusterip_config
*c
= pde
->data
;
658 ret
= seq_release(inode
, file
);
661 clusterip_config_put(c
);
666 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
667 size_t size
, loff_t
*ofs
)
669 #define PROC_WRITELEN 10
670 char buffer
[PROC_WRITELEN
+1];
671 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
672 struct clusterip_config
*c
= pde
->data
;
673 unsigned long nodenum
;
675 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
678 if (*buffer
== '+') {
679 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
680 if (clusterip_add_node(c
, nodenum
))
682 } else if (*buffer
== '-') {
683 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
684 if (clusterip_del_node(c
, nodenum
))
692 static const struct file_operations clusterip_proc_fops
= {
693 .owner
= THIS_MODULE
,
694 .open
= clusterip_proc_open
,
696 .write
= clusterip_proc_write
,
698 .release
= clusterip_proc_release
,
701 #endif /* CONFIG_PROC_FS */
703 static int __init
clusterip_tg_init(void)
707 ret
= xt_register_target(&clusterip_tg_reg
);
711 ret
= nf_register_hook(&cip_arp_ops
);
715 #ifdef CONFIG_PROC_FS
716 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", init_net
.proc_net
);
717 if (!clusterip_procdir
) {
718 printk(KERN_ERR
"CLUSTERIP: Unable to proc dir entry\n");
722 #endif /* CONFIG_PROC_FS */
724 printk(KERN_NOTICE
"ClusterIP Version %s loaded successfully\n",
728 #ifdef CONFIG_PROC_FS
730 nf_unregister_hook(&cip_arp_ops
);
731 #endif /* CONFIG_PROC_FS */
733 xt_unregister_target(&clusterip_tg_reg
);
737 static void __exit
clusterip_tg_exit(void)
739 printk(KERN_NOTICE
"ClusterIP Version %s unloading\n",
741 #ifdef CONFIG_PROC_FS
742 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
744 nf_unregister_hook(&cip_arp_ops
);
745 xt_unregister_target(&clusterip_tg_reg
);
748 module_init(clusterip_tg_init
);
749 module_exit(clusterip_tg_exit
);