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 write_lock_bh(&clusterip_lock
);
86 if (atomic_dec_and_test(&c
->entries
)) {
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
);
101 write_unlock_bh(&clusterip_lock
);
104 static struct clusterip_config
*
105 __clusterip_config_find(__be32 clusterip
)
107 struct clusterip_config
*c
;
109 list_for_each_entry(c
, &clusterip_configs
, list
) {
110 if (c
->clusterip
== clusterip
)
117 static inline struct clusterip_config
*
118 clusterip_config_find_get(__be32 clusterip
, int entry
)
120 struct clusterip_config
*c
;
122 read_lock_bh(&clusterip_lock
);
123 c
= __clusterip_config_find(clusterip
);
125 read_unlock_bh(&clusterip_lock
);
128 atomic_inc(&c
->refcount
);
130 atomic_inc(&c
->entries
);
131 read_unlock_bh(&clusterip_lock
);
137 clusterip_config_init_nodelist(struct clusterip_config
*c
,
138 const struct ipt_clusterip_tgt_info
*i
)
142 for (n
= 0; n
< i
->num_local_nodes
; n
++)
143 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
146 static struct clusterip_config
*
147 clusterip_config_init(const struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
148 struct net_device
*dev
)
150 struct clusterip_config
*c
;
152 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
158 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
159 c
->num_total_nodes
= i
->num_total_nodes
;
160 clusterip_config_init_nodelist(c
, i
);
161 c
->hash_mode
= i
->hash_mode
;
162 c
->hash_initval
= i
->hash_initval
;
163 atomic_set(&c
->refcount
, 1);
164 atomic_set(&c
->entries
, 1);
166 #ifdef CONFIG_PROC_FS
170 /* create proc dir entry */
171 sprintf(buffer
, "%u.%u.%u.%u", NIPQUAD(ip
));
172 c
->pde
= proc_create_data(buffer
, S_IWUSR
|S_IRUSR
,
174 &clusterip_proc_fops
, c
);
182 write_lock_bh(&clusterip_lock
);
183 list_add(&c
->list
, &clusterip_configs
);
184 write_unlock_bh(&clusterip_lock
);
189 #ifdef CONFIG_PROC_FS
191 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
195 nodenum
> c
->num_total_nodes
)
198 /* check if we already have this number in our bitfield */
199 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
206 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
209 nodenum
> c
->num_total_nodes
)
212 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
219 static inline u_int32_t
220 clusterip_hashfn(const struct sk_buff
*skb
,
221 const struct clusterip_config
*config
)
223 const struct iphdr
*iph
= ip_hdr(skb
);
224 unsigned long hashval
;
225 u_int16_t sport
, dport
;
226 const u_int16_t
*ports
;
228 switch (iph
->protocol
) {
231 case IPPROTO_UDPLITE
:
235 ports
= (const void *)iph
+iph
->ihl
*4;
241 printk(KERN_NOTICE
"CLUSTERIP: unknown protocol `%u'\n",
246 switch (config
->hash_mode
) {
247 case CLUSTERIP_HASHMODE_SIP
:
248 hashval
= jhash_1word(ntohl(iph
->saddr
),
249 config
->hash_initval
);
251 case CLUSTERIP_HASHMODE_SIP_SPT
:
252 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
253 config
->hash_initval
);
255 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
256 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
257 config
->hash_initval
);
260 /* to make gcc happy */
262 /* This cannot happen, unless the check function wasn't called
263 * at rule load time */
264 printk("CLUSTERIP: unknown mode `%u'\n", config
->hash_mode
);
269 /* node numbers are 1..n, not 0..n */
270 return (((u64
)hashval
* config
->num_total_nodes
) >> 32) + 1;
274 clusterip_responsible(const struct clusterip_config
*config
, u_int32_t hash
)
276 return test_bit(hash
- 1, &config
->local_nodes
);
279 /***********************************************************************
281 ***********************************************************************/
284 clusterip_tg(struct sk_buff
*skb
, const struct net_device
*in
,
285 const struct net_device
*out
, unsigned int hooknum
,
286 const struct xt_target
*target
, const void *targinfo
)
288 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
290 enum ip_conntrack_info ctinfo
;
293 /* don't need to clusterip_config_get() here, since refcount
294 * is only decremented by destroy() - and ip_tables guarantees
295 * that the ->target() function isn't called after ->destroy() */
297 ct
= nf_ct_get(skb
, &ctinfo
);
299 printk(KERN_ERR
"CLUSTERIP: no conntrack!\n");
300 /* FIXME: need to drop invalid ones, since replies
301 * to outgoing connections of other nodes will be
302 * marked as INVALID */
306 /* special case: ICMP error handling. conntrack distinguishes between
307 * error messages (RELATED) and information requests (see below) */
308 if (ip_hdr(skb
)->protocol
== IPPROTO_ICMP
309 && (ctinfo
== IP_CT_RELATED
310 || ctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
))
313 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
314 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
315 * on, which all have an ID field [relevant for hashing]. */
317 hash
= clusterip_hashfn(skb
, cipinfo
->config
);
324 case IP_CT_RELATED
+IP_CT_IS_REPLY
:
325 /* FIXME: we don't handle expectations at the
326 * moment. they can arrive on a different node than
327 * the master connection (e.g. FTP passive mode) */
328 case IP_CT_ESTABLISHED
:
329 case IP_CT_ESTABLISHED
+IP_CT_IS_REPLY
:
336 nf_ct_dump_tuple_ip(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
338 pr_debug("hash=%u ct_hash=%u ", hash
, ct
->mark
);
339 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
340 pr_debug("not responsible\n");
343 pr_debug("responsible\n");
345 /* despite being received via linklayer multicast, this is
346 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
347 skb
->pkt_type
= PACKET_HOST
;
353 clusterip_tg_check(const char *tablename
, const void *e_void
,
354 const struct xt_target
*target
, void *targinfo
,
355 unsigned int hook_mask
)
357 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
358 const struct ipt_entry
*e
= e_void
;
360 struct clusterip_config
*config
;
362 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
363 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
364 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
365 printk(KERN_WARNING
"CLUSTERIP: unknown mode `%u'\n",
370 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff)
371 || e
->ip
.dst
.s_addr
== 0) {
372 printk(KERN_ERR
"CLUSTERIP: Please specify destination IP\n");
376 /* FIXME: further sanity checks */
378 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
380 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
381 printk(KERN_WARNING
"CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e
->ip
.dst
.s_addr
));
384 struct net_device
*dev
;
386 if (e
->ip
.iniface
[0] == '\0') {
387 printk(KERN_WARNING
"CLUSTERIP: Please specify an interface name\n");
391 dev
= dev_get_by_name(&init_net
, e
->ip
.iniface
);
393 printk(KERN_WARNING
"CLUSTERIP: no such interface %s\n", e
->ip
.iniface
);
397 config
= clusterip_config_init(cipinfo
,
398 e
->ip
.dst
.s_addr
, dev
);
400 printk(KERN_WARNING
"CLUSTERIP: cannot allocate config\n");
404 dev_mc_add(config
->dev
,config
->clustermac
, ETH_ALEN
, 0);
407 cipinfo
->config
= config
;
409 if (nf_ct_l3proto_try_module_get(target
->family
) < 0) {
410 printk(KERN_WARNING
"can't load conntrack support for "
411 "proto=%u\n", target
->family
);
418 /* drop reference count of cluster config when rule is deleted */
419 static void clusterip_tg_destroy(const struct xt_target
*target
, void *targinfo
)
421 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
423 /* if no more entries are referencing the config, remove it
424 * from the list and destroy the proc entry */
425 clusterip_config_entry_put(cipinfo
->config
);
427 clusterip_config_put(cipinfo
->config
);
429 nf_ct_l3proto_module_put(target
->family
);
433 struct compat_ipt_clusterip_tgt_info
436 u_int8_t clustermac
[6];
437 u_int16_t num_total_nodes
;
438 u_int16_t num_local_nodes
;
439 u_int16_t local_nodes
[CLUSTERIP_MAX_NODES
];
441 u_int32_t hash_initval
;
442 compat_uptr_t config
;
444 #endif /* CONFIG_COMPAT */
446 static struct xt_target clusterip_tg_reg __read_mostly
= {
449 .target
= clusterip_tg
,
450 .checkentry
= clusterip_tg_check
,
451 .destroy
= clusterip_tg_destroy
,
452 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
454 .compatsize
= sizeof(struct compat_ipt_clusterip_tgt_info
),
455 #endif /* CONFIG_COMPAT */
460 /***********************************************************************
462 ***********************************************************************/
464 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
466 u_int8_t src_hw
[ETH_ALEN
];
468 u_int8_t dst_hw
[ETH_ALEN
];
470 } __attribute__ ((packed
));
473 static void arp_print(struct arp_payload
*payload
)
475 #define HBUFFERLEN 30
476 char hbuffer
[HBUFFERLEN
];
478 const char hexbuf
[]= "0123456789abcdef";
480 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
481 hbuffer
[k
++]=hexbuf
[(payload
->src_hw
[j
]>>4)&15];
482 hbuffer
[k
++]=hexbuf
[payload
->src_hw
[j
]&15];
487 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
488 NIPQUAD(payload
->src_ip
), hbuffer
,
489 NIPQUAD(payload
->dst_ip
));
494 arp_mangle(unsigned int hook
,
496 const struct net_device
*in
,
497 const struct net_device
*out
,
498 int (*okfn
)(struct sk_buff
*))
500 struct arphdr
*arp
= arp_hdr(skb
);
501 struct arp_payload
*payload
;
502 struct clusterip_config
*c
;
504 /* we don't care about non-ethernet and non-ipv4 ARP */
505 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
)
506 || arp
->ar_pro
!= htons(ETH_P_IP
)
507 || arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
510 /* we only want to mangle arp requests and replies */
511 if (arp
->ar_op
!= htons(ARPOP_REPLY
)
512 && arp
->ar_op
!= htons(ARPOP_REQUEST
))
515 payload
= (void *)(arp
+1);
517 /* if there is no clusterip configuration for the arp reply's
518 * source ip, we don't want to mangle it */
519 c
= clusterip_config_find_get(payload
->src_ip
, 0);
523 /* normally the linux kernel always replies to arp queries of
524 * addresses on different interfacs. However, in the CLUSTERIP case
525 * this wouldn't work, since we didn't subscribe the mcast group on
526 * other interfaces */
528 pr_debug("CLUSTERIP: not mangling arp reply on different "
529 "interface: cip'%s'-skb'%s'\n",
530 c
->dev
->name
, out
->name
);
531 clusterip_config_put(c
);
535 /* mangle reply hardware address */
536 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
539 pr_debug(KERN_DEBUG
"CLUSTERIP mangled arp reply: ");
543 clusterip_config_put(c
);
548 static struct nf_hook_ops cip_arp_ops __read_mostly
= {
551 .hooknum
= NF_ARP_OUT
,
555 /***********************************************************************
557 ***********************************************************************/
559 #ifdef CONFIG_PROC_FS
561 struct clusterip_seq_position
{
562 unsigned int pos
; /* position */
563 unsigned int weight
; /* number of bits set == size */
564 unsigned int bit
; /* current bit */
565 unsigned long val
; /* current value */
568 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
570 const struct proc_dir_entry
*pde
= s
->private;
571 struct clusterip_config
*c
= pde
->data
;
573 u_int32_t local_nodes
;
574 struct clusterip_seq_position
*idx
;
576 /* FIXME: possible race */
577 local_nodes
= c
->local_nodes
;
578 weight
= hweight32(local_nodes
);
582 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
584 return ERR_PTR(-ENOMEM
);
587 idx
->weight
= weight
;
588 idx
->bit
= ffs(local_nodes
);
589 idx
->val
= local_nodes
;
590 clear_bit(idx
->bit
- 1, &idx
->val
);
595 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
597 struct clusterip_seq_position
*idx
= v
;
600 if (*pos
>= idx
->weight
) {
604 idx
->bit
= ffs(idx
->val
);
605 clear_bit(idx
->bit
- 1, &idx
->val
);
609 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
614 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
616 struct clusterip_seq_position
*idx
= v
;
621 seq_printf(s
, "%u", idx
->bit
);
623 if (idx
->pos
== idx
->weight
- 1)
629 static const struct seq_operations clusterip_seq_ops
= {
630 .start
= clusterip_seq_start
,
631 .next
= clusterip_seq_next
,
632 .stop
= clusterip_seq_stop
,
633 .show
= clusterip_seq_show
,
636 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
638 int ret
= seq_open(file
, &clusterip_seq_ops
);
641 struct seq_file
*sf
= file
->private_data
;
642 struct proc_dir_entry
*pde
= PDE(inode
);
643 struct clusterip_config
*c
= pde
->data
;
647 clusterip_config_get(c
);
653 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
655 struct proc_dir_entry
*pde
= PDE(inode
);
656 struct clusterip_config
*c
= pde
->data
;
659 ret
= seq_release(inode
, file
);
662 clusterip_config_put(c
);
667 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
668 size_t size
, loff_t
*ofs
)
670 #define PROC_WRITELEN 10
671 char buffer
[PROC_WRITELEN
+1];
672 const struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
673 struct clusterip_config
*c
= pde
->data
;
674 unsigned long nodenum
;
676 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
679 if (*buffer
== '+') {
680 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
681 if (clusterip_add_node(c
, nodenum
))
683 } else if (*buffer
== '-') {
684 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
685 if (clusterip_del_node(c
, nodenum
))
693 static const struct file_operations clusterip_proc_fops
= {
694 .owner
= THIS_MODULE
,
695 .open
= clusterip_proc_open
,
697 .write
= clusterip_proc_write
,
699 .release
= clusterip_proc_release
,
702 #endif /* CONFIG_PROC_FS */
704 static int __init
clusterip_tg_init(void)
708 ret
= xt_register_target(&clusterip_tg_reg
);
712 ret
= nf_register_hook(&cip_arp_ops
);
716 #ifdef CONFIG_PROC_FS
717 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", init_net
.proc_net
);
718 if (!clusterip_procdir
) {
719 printk(KERN_ERR
"CLUSTERIP: Unable to proc dir entry\n");
723 #endif /* CONFIG_PROC_FS */
725 printk(KERN_NOTICE
"ClusterIP Version %s loaded successfully\n",
729 #ifdef CONFIG_PROC_FS
731 nf_unregister_hook(&cip_arp_ops
);
732 #endif /* CONFIG_PROC_FS */
734 xt_unregister_target(&clusterip_tg_reg
);
738 static void __exit
clusterip_tg_exit(void)
740 printk(KERN_NOTICE
"ClusterIP Version %s unloading\n",
742 #ifdef CONFIG_PROC_FS
743 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
745 nf_unregister_hook(&cip_arp_ops
);
746 xt_unregister_target(&clusterip_tg_reg
);
749 module_init(clusterip_tg_init
);
750 module_exit(clusterip_tg_exit
);