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
= proc_create(buffer
, S_IWUSR
|S_IRUSR
,
171 clusterip_procdir
, &clusterip_proc_fops
);
180 write_lock_bh(&clusterip_lock
);
181 list_add(&c
->list
, &clusterip_configs
);
182 write_unlock_bh(&clusterip_lock
);
187 #ifdef CONFIG_PROC_FS
189 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
193 nodenum
> c
->num_total_nodes
)
196 /* check if we already have this number in our bitfield */
197 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
204 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
207 nodenum
> c
->num_total_nodes
)
210 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
217 static inline u_int32_t
218 clusterip_hashfn(const struct sk_buff
*skb
,
219 const struct clusterip_config
*config
)
221 const struct iphdr
*iph
= ip_hdr(skb
);
222 unsigned long hashval
;
223 u_int16_t sport
, dport
;
224 const u_int16_t
*ports
;
226 switch (iph
->protocol
) {
229 case IPPROTO_UDPLITE
:
233 ports
= (const void *)iph
+iph
->ihl
*4;
239 printk(KERN_NOTICE
"CLUSTERIP: unknown protocol `%u'\n",
244 switch (config
->hash_mode
) {
245 case CLUSTERIP_HASHMODE_SIP
:
246 hashval
= jhash_1word(ntohl(iph
->saddr
),
247 config
->hash_initval
);
249 case CLUSTERIP_HASHMODE_SIP_SPT
:
250 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
251 config
->hash_initval
);
253 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
254 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
255 config
->hash_initval
);
258 /* to make gcc happy */
260 /* This cannot happen, unless the check function wasn't called
261 * at rule load time */
262 printk("CLUSTERIP: unknown mode `%u'\n", config
->hash_mode
);
267 /* node numbers are 1..n, not 0..n */
268 return (((u64
)hashval
* config
->num_total_nodes
) >> 32) + 1;
272 clusterip_responsible(const struct clusterip_config
*config
, u_int32_t hash
)
274 return test_bit(hash
- 1, &config
->local_nodes
);
277 /***********************************************************************
279 ***********************************************************************/
282 clusterip_tg(struct sk_buff
*skb
, const struct net_device
*in
,
283 const struct net_device
*out
, unsigned int hooknum
,
284 const struct xt_target
*target
, const void *targinfo
)
286 const struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
288 enum ip_conntrack_info ctinfo
;
291 /* don't need to clusterip_config_get() here, since refcount
292 * is only decremented by destroy() - and ip_tables guarantees
293 * that the ->target() function isn't called after ->destroy() */
295 ct
= nf_ct_get(skb
, &ctinfo
);
297 printk(KERN_ERR
"CLUSTERIP: no conntrack!\n");
298 /* FIXME: need to drop invalid ones, since replies
299 * to outgoing connections of other nodes will be
300 * marked as INVALID */
304 /* special case: ICMP error handling. conntrack distinguishes between
305 * error messages (RELATED) and information requests (see below) */
306 if (ip_hdr(skb
)->protocol
== IPPROTO_ICMP
307 && (ctinfo
== IP_CT_RELATED
308 || ctinfo
== IP_CT_RELATED
+IP_CT_IS_REPLY
))
311 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
312 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
313 * on, which all have an ID field [relevant for hashing]. */
315 hash
= clusterip_hashfn(skb
, cipinfo
->config
);
322 case IP_CT_RELATED
+IP_CT_IS_REPLY
:
323 /* FIXME: we don't handle expectations at the
324 * moment. they can arrive on a different node than
325 * the master connection (e.g. FTP passive mode) */
326 case IP_CT_ESTABLISHED
:
327 case IP_CT_ESTABLISHED
+IP_CT_IS_REPLY
:
334 DUMP_TUPLE(&ct
->tuplehash
[IP_CT_DIR_ORIGINAL
].tuple
);
336 pr_debug("hash=%u ct_hash=%u ", hash
, ct
->mark
);
337 if (!clusterip_responsible(cipinfo
->config
, hash
)) {
338 pr_debug("not responsible\n");
341 pr_debug("responsible\n");
343 /* despite being received via linklayer multicast, this is
344 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
345 skb
->pkt_type
= PACKET_HOST
;
351 clusterip_tg_check(const char *tablename
, const void *e_void
,
352 const struct xt_target
*target
, void *targinfo
,
353 unsigned int hook_mask
)
355 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
356 const struct ipt_entry
*e
= e_void
;
358 struct clusterip_config
*config
;
360 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
361 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
362 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
363 printk(KERN_WARNING
"CLUSTERIP: unknown mode `%u'\n",
368 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff)
369 || e
->ip
.dst
.s_addr
== 0) {
370 printk(KERN_ERR
"CLUSTERIP: Please specify destination IP\n");
374 /* FIXME: further sanity checks */
376 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
378 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
379 printk(KERN_WARNING
"CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e
->ip
.dst
.s_addr
));
382 struct net_device
*dev
;
384 if (e
->ip
.iniface
[0] == '\0') {
385 printk(KERN_WARNING
"CLUSTERIP: Please specify an interface name\n");
389 dev
= dev_get_by_name(&init_net
, e
->ip
.iniface
);
391 printk(KERN_WARNING
"CLUSTERIP: no such interface %s\n", e
->ip
.iniface
);
395 config
= clusterip_config_init(cipinfo
,
396 e
->ip
.dst
.s_addr
, dev
);
398 printk(KERN_WARNING
"CLUSTERIP: cannot allocate config\n");
402 dev_mc_add(config
->dev
,config
->clustermac
, ETH_ALEN
, 0);
405 cipinfo
->config
= config
;
407 if (nf_ct_l3proto_try_module_get(target
->family
) < 0) {
408 printk(KERN_WARNING
"can't load conntrack support for "
409 "proto=%u\n", target
->family
);
416 /* drop reference count of cluster config when rule is deleted */
417 static void clusterip_tg_destroy(const struct xt_target
*target
, void *targinfo
)
419 struct ipt_clusterip_tgt_info
*cipinfo
= targinfo
;
421 /* if no more entries are referencing the config, remove it
422 * from the list and destroy the proc entry */
423 clusterip_config_entry_put(cipinfo
->config
);
425 clusterip_config_put(cipinfo
->config
);
427 nf_ct_l3proto_module_put(target
->family
);
431 struct compat_ipt_clusterip_tgt_info
434 u_int8_t clustermac
[6];
435 u_int16_t num_total_nodes
;
436 u_int16_t num_local_nodes
;
437 u_int16_t local_nodes
[CLUSTERIP_MAX_NODES
];
439 u_int32_t hash_initval
;
440 compat_uptr_t config
;
442 #endif /* CONFIG_COMPAT */
444 static struct xt_target clusterip_tg_reg __read_mostly
= {
447 .target
= clusterip_tg
,
448 .checkentry
= clusterip_tg_check
,
449 .destroy
= clusterip_tg_destroy
,
450 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
452 .compatsize
= sizeof(struct compat_ipt_clusterip_tgt_info
),
453 #endif /* CONFIG_COMPAT */
458 /***********************************************************************
460 ***********************************************************************/
462 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
464 u_int8_t src_hw
[ETH_ALEN
];
466 u_int8_t dst_hw
[ETH_ALEN
];
468 } __attribute__ ((packed
));
471 static void arp_print(struct arp_payload
*payload
)
473 #define HBUFFERLEN 30
474 char hbuffer
[HBUFFERLEN
];
476 const char hexbuf
[]= "0123456789abcdef";
478 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
479 hbuffer
[k
++]=hexbuf
[(payload
->src_hw
[j
]>>4)&15];
480 hbuffer
[k
++]=hexbuf
[payload
->src_hw
[j
]&15];
485 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
486 NIPQUAD(payload
->src_ip
), hbuffer
,
487 NIPQUAD(payload
->dst_ip
));
492 arp_mangle(unsigned int hook
,
494 const struct net_device
*in
,
495 const struct net_device
*out
,
496 int (*okfn
)(struct sk_buff
*))
498 struct arphdr
*arp
= arp_hdr(skb
);
499 struct arp_payload
*payload
;
500 struct clusterip_config
*c
;
502 /* we don't care about non-ethernet and non-ipv4 ARP */
503 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
)
504 || arp
->ar_pro
!= htons(ETH_P_IP
)
505 || arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
508 /* we only want to mangle arp requests and replies */
509 if (arp
->ar_op
!= htons(ARPOP_REPLY
)
510 && arp
->ar_op
!= htons(ARPOP_REQUEST
))
513 payload
= (void *)(arp
+1);
515 /* if there is no clusterip configuration for the arp reply's
516 * source ip, we don't want to mangle it */
517 c
= clusterip_config_find_get(payload
->src_ip
, 0);
521 /* normally the linux kernel always replies to arp queries of
522 * addresses on different interfacs. However, in the CLUSTERIP case
523 * this wouldn't work, since we didn't subscribe the mcast group on
524 * other interfaces */
526 pr_debug("CLUSTERIP: not mangling arp reply on different "
527 "interface: cip'%s'-skb'%s'\n",
528 c
->dev
->name
, out
->name
);
529 clusterip_config_put(c
);
533 /* mangle reply hardware address */
534 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
537 pr_debug(KERN_DEBUG
"CLUSTERIP mangled arp reply: ");
541 clusterip_config_put(c
);
546 static struct nf_hook_ops cip_arp_ops __read_mostly
= {
549 .hooknum
= NF_ARP_OUT
,
553 /***********************************************************************
555 ***********************************************************************/
557 #ifdef CONFIG_PROC_FS
559 struct clusterip_seq_position
{
560 unsigned int pos
; /* position */
561 unsigned int weight
; /* number of bits set == size */
562 unsigned int bit
; /* current bit */
563 unsigned long val
; /* current value */
566 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
568 struct proc_dir_entry
*pde
= s
->private;
569 struct clusterip_config
*c
= pde
->data
;
571 u_int32_t local_nodes
;
572 struct clusterip_seq_position
*idx
;
574 /* FIXME: possible race */
575 local_nodes
= c
->local_nodes
;
576 weight
= hweight32(local_nodes
);
580 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
582 return ERR_PTR(-ENOMEM
);
585 idx
->weight
= weight
;
586 idx
->bit
= ffs(local_nodes
);
587 idx
->val
= local_nodes
;
588 clear_bit(idx
->bit
- 1, &idx
->val
);
593 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
595 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
598 if (*pos
>= idx
->weight
) {
602 idx
->bit
= ffs(idx
->val
);
603 clear_bit(idx
->bit
- 1, &idx
->val
);
607 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
612 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
614 struct clusterip_seq_position
*idx
= (struct clusterip_seq_position
*)v
;
619 seq_printf(s
, "%u", idx
->bit
);
621 if (idx
->pos
== idx
->weight
- 1)
627 static const struct seq_operations clusterip_seq_ops
= {
628 .start
= clusterip_seq_start
,
629 .next
= clusterip_seq_next
,
630 .stop
= clusterip_seq_stop
,
631 .show
= clusterip_seq_show
,
634 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
636 int ret
= seq_open(file
, &clusterip_seq_ops
);
639 struct seq_file
*sf
= file
->private_data
;
640 struct proc_dir_entry
*pde
= PDE(inode
);
641 struct clusterip_config
*c
= pde
->data
;
645 clusterip_config_get(c
);
651 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
653 struct proc_dir_entry
*pde
= PDE(inode
);
654 struct clusterip_config
*c
= pde
->data
;
657 ret
= seq_release(inode
, file
);
660 clusterip_config_put(c
);
665 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
666 size_t size
, loff_t
*ofs
)
668 #define PROC_WRITELEN 10
669 char buffer
[PROC_WRITELEN
+1];
670 struct proc_dir_entry
*pde
= PDE(file
->f_path
.dentry
->d_inode
);
671 struct clusterip_config
*c
= pde
->data
;
672 unsigned long nodenum
;
674 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
677 if (*buffer
== '+') {
678 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
679 if (clusterip_add_node(c
, nodenum
))
681 } else if (*buffer
== '-') {
682 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
683 if (clusterip_del_node(c
, nodenum
))
691 static const struct file_operations clusterip_proc_fops
= {
692 .owner
= THIS_MODULE
,
693 .open
= clusterip_proc_open
,
695 .write
= clusterip_proc_write
,
697 .release
= clusterip_proc_release
,
700 #endif /* CONFIG_PROC_FS */
702 static int __init
clusterip_tg_init(void)
706 ret
= xt_register_target(&clusterip_tg_reg
);
710 ret
= nf_register_hook(&cip_arp_ops
);
714 #ifdef CONFIG_PROC_FS
715 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", init_net
.proc_net
);
716 if (!clusterip_procdir
) {
717 printk(KERN_ERR
"CLUSTERIP: Unable to proc dir entry\n");
721 #endif /* CONFIG_PROC_FS */
723 printk(KERN_NOTICE
"ClusterIP Version %s loaded successfully\n",
727 #ifdef CONFIG_PROC_FS
729 nf_unregister_hook(&cip_arp_ops
);
730 #endif /* CONFIG_PROC_FS */
732 xt_unregister_target(&clusterip_tg_reg
);
736 static void __exit
clusterip_tg_exit(void)
738 printk(KERN_NOTICE
"ClusterIP Version %s unloading\n",
740 #ifdef CONFIG_PROC_FS
741 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
743 nf_unregister_hook(&cip_arp_ops
);
744 xt_unregister_target(&clusterip_tg_reg
);
747 module_init(clusterip_tg_init
);
748 module_exit(clusterip_tg_exit
);