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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/module.h>
14 #include <linux/proc_fs.h>
15 #include <linux/jhash.h>
16 #include <linux/bitops.h>
17 #include <linux/skbuff.h>
18 #include <linux/slab.h>
20 #include <linux/tcp.h>
21 #include <linux/udp.h>
22 #include <linux/icmp.h>
23 #include <linux/if_arp.h>
24 #include <linux/seq_file.h>
25 #include <linux/netfilter_arp.h>
26 #include <linux/netfilter/x_tables.h>
27 #include <linux/netfilter_ipv4/ip_tables.h>
28 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/net_namespace.h>
31 #include <net/checksum.h>
33 #define CLUSTERIP_VERSION "0.8"
35 MODULE_LICENSE("GPL");
36 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
37 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
39 struct clusterip_config
{
40 struct list_head list
; /* list of all configs */
41 atomic_t refcount
; /* reference count */
42 atomic_t entries
; /* number of entries/rules
45 __be32 clusterip
; /* the IP address */
46 u_int8_t clustermac
[ETH_ALEN
]; /* the MAC address */
47 struct net_device
*dev
; /* device */
48 u_int16_t num_total_nodes
; /* total number of nodes */
49 unsigned long local_nodes
; /* node number array */
52 struct proc_dir_entry
*pde
; /* proc dir entry */
54 enum clusterip_hashmode hash_mode
; /* which hashing mode */
55 u_int32_t hash_initval
; /* hash initialization */
58 static LIST_HEAD(clusterip_configs
);
60 /* clusterip_lock protects the clusterip_configs list */
61 static DEFINE_RWLOCK(clusterip_lock
);
64 static const struct file_operations clusterip_proc_fops
;
65 static struct proc_dir_entry
*clusterip_procdir
;
69 clusterip_config_get(struct clusterip_config
*c
)
71 atomic_inc(&c
->refcount
);
75 clusterip_config_put(struct clusterip_config
*c
)
77 if (atomic_dec_and_test(&c
->refcount
))
81 /* decrease the count of entries using/referencing this config. If last
82 * entry(rule) is removed, remove the config from lists, but don't free it
83 * yet, since proc-files could still be holding references */
85 clusterip_config_entry_put(struct clusterip_config
*c
)
87 write_lock_bh(&clusterip_lock
);
88 if (atomic_dec_and_test(&c
->entries
)) {
90 write_unlock_bh(&clusterip_lock
);
92 dev_mc_del(c
->dev
, c
->clustermac
);
95 /* In case anyone still accesses the file, the open/close
96 * functions are also incrementing the refcount on their own,
97 * so it's safe to remove the entry even if it's in use. */
99 remove_proc_entry(c
->pde
->name
, c
->pde
->parent
);
103 write_unlock_bh(&clusterip_lock
);
106 static struct clusterip_config
*
107 __clusterip_config_find(__be32 clusterip
)
109 struct clusterip_config
*c
;
111 list_for_each_entry(c
, &clusterip_configs
, list
) {
112 if (c
->clusterip
== clusterip
)
119 static inline struct clusterip_config
*
120 clusterip_config_find_get(__be32 clusterip
, int entry
)
122 struct clusterip_config
*c
;
124 read_lock_bh(&clusterip_lock
);
125 c
= __clusterip_config_find(clusterip
);
127 read_unlock_bh(&clusterip_lock
);
130 atomic_inc(&c
->refcount
);
132 atomic_inc(&c
->entries
);
133 read_unlock_bh(&clusterip_lock
);
139 clusterip_config_init_nodelist(struct clusterip_config
*c
,
140 const struct ipt_clusterip_tgt_info
*i
)
144 for (n
= 0; n
< i
->num_local_nodes
; n
++)
145 set_bit(i
->local_nodes
[n
] - 1, &c
->local_nodes
);
148 static struct clusterip_config
*
149 clusterip_config_init(const struct ipt_clusterip_tgt_info
*i
, __be32 ip
,
150 struct net_device
*dev
)
152 struct clusterip_config
*c
;
154 c
= kzalloc(sizeof(*c
), GFP_ATOMIC
);
160 memcpy(&c
->clustermac
, &i
->clustermac
, ETH_ALEN
);
161 c
->num_total_nodes
= i
->num_total_nodes
;
162 clusterip_config_init_nodelist(c
, i
);
163 c
->hash_mode
= i
->hash_mode
;
164 c
->hash_initval
= i
->hash_initval
;
165 atomic_set(&c
->refcount
, 1);
166 atomic_set(&c
->entries
, 1);
168 #ifdef CONFIG_PROC_FS
172 /* create proc dir entry */
173 sprintf(buffer
, "%pI4", &ip
);
174 c
->pde
= proc_create_data(buffer
, S_IWUSR
|S_IRUSR
,
176 &clusterip_proc_fops
, c
);
184 write_lock_bh(&clusterip_lock
);
185 list_add(&c
->list
, &clusterip_configs
);
186 write_unlock_bh(&clusterip_lock
);
191 #ifdef CONFIG_PROC_FS
193 clusterip_add_node(struct clusterip_config
*c
, u_int16_t nodenum
)
197 nodenum
> c
->num_total_nodes
)
200 /* check if we already have this number in our bitfield */
201 if (test_and_set_bit(nodenum
- 1, &c
->local_nodes
))
208 clusterip_del_node(struct clusterip_config
*c
, u_int16_t nodenum
)
211 nodenum
> c
->num_total_nodes
)
214 if (test_and_clear_bit(nodenum
- 1, &c
->local_nodes
))
221 static inline u_int32_t
222 clusterip_hashfn(const struct sk_buff
*skb
,
223 const struct clusterip_config
*config
)
225 const struct iphdr
*iph
= ip_hdr(skb
);
226 unsigned long hashval
;
227 u_int16_t sport
, dport
;
228 const u_int16_t
*ports
;
230 switch (iph
->protocol
) {
233 case IPPROTO_UDPLITE
:
237 ports
= (const void *)iph
+iph
->ihl
*4;
243 pr_info("unknown protocol %u\n", iph
->protocol
);
247 switch (config
->hash_mode
) {
248 case CLUSTERIP_HASHMODE_SIP
:
249 hashval
= jhash_1word(ntohl(iph
->saddr
),
250 config
->hash_initval
);
252 case CLUSTERIP_HASHMODE_SIP_SPT
:
253 hashval
= jhash_2words(ntohl(iph
->saddr
), sport
,
254 config
->hash_initval
);
256 case CLUSTERIP_HASHMODE_SIP_SPT_DPT
:
257 hashval
= jhash_3words(ntohl(iph
->saddr
), sport
, dport
,
258 config
->hash_initval
);
261 /* to make gcc happy */
263 /* This cannot happen, unless the check function wasn't called
264 * at rule load time */
265 pr_info("unknown mode %u\n", config
->hash_mode
);
270 /* node numbers are 1..n, not 0..n */
271 return (((u64
)hashval
* config
->num_total_nodes
) >> 32) + 1;
275 clusterip_responsible(const struct clusterip_config
*config
, u_int32_t hash
)
277 return test_bit(hash
- 1, &config
->local_nodes
);
280 /***********************************************************************
282 ***********************************************************************/
285 clusterip_tg(struct sk_buff
*skb
, const struct xt_action_param
*par
)
287 const struct ipt_clusterip_tgt_info
*cipinfo
= par
->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 pr_info("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 nf_ct_dump_tuple_ip(&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
;
351 static int clusterip_tg_check(const struct xt_tgchk_param
*par
)
353 struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
354 const struct ipt_entry
*e
= par
->entryinfo
;
355 struct clusterip_config
*config
;
358 if (cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP
&&
359 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT
&&
360 cipinfo
->hash_mode
!= CLUSTERIP_HASHMODE_SIP_SPT_DPT
) {
361 pr_info("unknown mode %u\n", cipinfo
->hash_mode
);
365 if (e
->ip
.dmsk
.s_addr
!= htonl(0xffffffff) ||
366 e
->ip
.dst
.s_addr
== 0) {
367 pr_info("Please specify destination IP\n");
371 /* FIXME: further sanity checks */
373 config
= clusterip_config_find_get(e
->ip
.dst
.s_addr
, 1);
375 if (!(cipinfo
->flags
& CLUSTERIP_FLAG_NEW
)) {
376 pr_info("no config found for %pI4, need 'new'\n",
380 struct net_device
*dev
;
382 if (e
->ip
.iniface
[0] == '\0') {
383 pr_info("Please specify an interface name\n");
387 dev
= dev_get_by_name(&init_net
, e
->ip
.iniface
);
389 pr_info("no such interface %s\n",
394 config
= clusterip_config_init(cipinfo
,
395 e
->ip
.dst
.s_addr
, dev
);
397 pr_info("cannot allocate config\n");
401 dev_mc_add(config
->dev
, config
->clustermac
);
404 cipinfo
->config
= config
;
406 ret
= nf_ct_l3proto_try_module_get(par
->family
);
408 pr_info("cannot load conntrack support for proto=%u\n",
413 /* drop reference count of cluster config when rule is deleted */
414 static void clusterip_tg_destroy(const struct xt_tgdtor_param
*par
)
416 const struct ipt_clusterip_tgt_info
*cipinfo
= par
->targinfo
;
418 /* if no more entries are referencing the config, remove it
419 * from the list and destroy the proc entry */
420 clusterip_config_entry_put(cipinfo
->config
);
422 clusterip_config_put(cipinfo
->config
);
424 nf_ct_l3proto_module_put(par
->family
);
428 struct compat_ipt_clusterip_tgt_info
431 u_int8_t clustermac
[6];
432 u_int16_t num_total_nodes
;
433 u_int16_t num_local_nodes
;
434 u_int16_t local_nodes
[CLUSTERIP_MAX_NODES
];
436 u_int32_t hash_initval
;
437 compat_uptr_t config
;
439 #endif /* CONFIG_COMPAT */
441 static struct xt_target clusterip_tg_reg __read_mostly
= {
443 .family
= NFPROTO_IPV4
,
444 .target
= clusterip_tg
,
445 .checkentry
= clusterip_tg_check
,
446 .destroy
= clusterip_tg_destroy
,
447 .targetsize
= sizeof(struct ipt_clusterip_tgt_info
),
449 .compatsize
= sizeof(struct compat_ipt_clusterip_tgt_info
),
450 #endif /* CONFIG_COMPAT */
455 /***********************************************************************
457 ***********************************************************************/
459 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
461 u_int8_t src_hw
[ETH_ALEN
];
463 u_int8_t dst_hw
[ETH_ALEN
];
465 } __attribute__ ((packed
));
468 static void arp_print(struct arp_payload
*payload
)
470 #define HBUFFERLEN 30
471 char hbuffer
[HBUFFERLEN
];
474 for (k
=0, j
=0; k
< HBUFFERLEN
-3 && j
< ETH_ALEN
; j
++) {
475 hbuffer
[k
++] = hex_asc_hi(payload
->src_hw
[j
]);
476 hbuffer
[k
++] = hex_asc_lo(payload
->src_hw
[j
]);
481 pr_debug("src %pI4@%s, dst %pI4\n",
482 &payload
->src_ip
, hbuffer
, &payload
->dst_ip
);
487 arp_mangle(unsigned int hook
,
489 const struct net_device
*in
,
490 const struct net_device
*out
,
491 int (*okfn
)(struct sk_buff
*))
493 struct arphdr
*arp
= arp_hdr(skb
);
494 struct arp_payload
*payload
;
495 struct clusterip_config
*c
;
497 /* we don't care about non-ethernet and non-ipv4 ARP */
498 if (arp
->ar_hrd
!= htons(ARPHRD_ETHER
) ||
499 arp
->ar_pro
!= htons(ETH_P_IP
) ||
500 arp
->ar_pln
!= 4 || arp
->ar_hln
!= ETH_ALEN
)
503 /* we only want to mangle arp requests and replies */
504 if (arp
->ar_op
!= htons(ARPOP_REPLY
) &&
505 arp
->ar_op
!= htons(ARPOP_REQUEST
))
508 payload
= (void *)(arp
+1);
510 /* if there is no clusterip configuration for the arp reply's
511 * source ip, we don't want to mangle it */
512 c
= clusterip_config_find_get(payload
->src_ip
, 0);
516 /* normally the linux kernel always replies to arp queries of
517 * addresses on different interfacs. However, in the CLUSTERIP case
518 * this wouldn't work, since we didn't subscribe the mcast group on
519 * other interfaces */
521 pr_debug("not mangling arp reply on different "
522 "interface: cip'%s'-skb'%s'\n",
523 c
->dev
->name
, out
->name
);
524 clusterip_config_put(c
);
528 /* mangle reply hardware address */
529 memcpy(payload
->src_hw
, c
->clustermac
, arp
->ar_hln
);
532 pr_debug("mangled arp reply: ");
536 clusterip_config_put(c
);
541 static struct nf_hook_ops cip_arp_ops __read_mostly
= {
544 .hooknum
= NF_ARP_OUT
,
548 /***********************************************************************
550 ***********************************************************************/
552 #ifdef CONFIG_PROC_FS
554 struct clusterip_seq_position
{
555 unsigned int pos
; /* position */
556 unsigned int weight
; /* number of bits set == size */
557 unsigned int bit
; /* current bit */
558 unsigned long val
; /* current value */
561 static void *clusterip_seq_start(struct seq_file
*s
, loff_t
*pos
)
563 struct clusterip_config
*c
= s
->private;
565 u_int32_t local_nodes
;
566 struct clusterip_seq_position
*idx
;
568 /* FIXME: possible race */
569 local_nodes
= c
->local_nodes
;
570 weight
= hweight32(local_nodes
);
574 idx
= kmalloc(sizeof(struct clusterip_seq_position
), GFP_KERNEL
);
576 return ERR_PTR(-ENOMEM
);
579 idx
->weight
= weight
;
580 idx
->bit
= ffs(local_nodes
);
581 idx
->val
= local_nodes
;
582 clear_bit(idx
->bit
- 1, &idx
->val
);
587 static void *clusterip_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
589 struct clusterip_seq_position
*idx
= v
;
592 if (*pos
>= idx
->weight
) {
596 idx
->bit
= ffs(idx
->val
);
597 clear_bit(idx
->bit
- 1, &idx
->val
);
601 static void clusterip_seq_stop(struct seq_file
*s
, void *v
)
607 static int clusterip_seq_show(struct seq_file
*s
, void *v
)
609 struct clusterip_seq_position
*idx
= v
;
614 seq_printf(s
, "%u", idx
->bit
);
616 if (idx
->pos
== idx
->weight
- 1)
622 static const struct seq_operations clusterip_seq_ops
= {
623 .start
= clusterip_seq_start
,
624 .next
= clusterip_seq_next
,
625 .stop
= clusterip_seq_stop
,
626 .show
= clusterip_seq_show
,
629 static int clusterip_proc_open(struct inode
*inode
, struct file
*file
)
631 int ret
= seq_open(file
, &clusterip_seq_ops
);
634 struct seq_file
*sf
= file
->private_data
;
635 struct clusterip_config
*c
= PDE(inode
)->data
;
639 clusterip_config_get(c
);
645 static int clusterip_proc_release(struct inode
*inode
, struct file
*file
)
647 struct clusterip_config
*c
= PDE(inode
)->data
;
650 ret
= seq_release(inode
, file
);
653 clusterip_config_put(c
);
658 static ssize_t
clusterip_proc_write(struct file
*file
, const char __user
*input
,
659 size_t size
, loff_t
*ofs
)
661 struct clusterip_config
*c
= PDE(file
->f_path
.dentry
->d_inode
)->data
;
662 #define PROC_WRITELEN 10
663 char buffer
[PROC_WRITELEN
+1];
664 unsigned long nodenum
;
666 if (copy_from_user(buffer
, input
, PROC_WRITELEN
))
669 if (*buffer
== '+') {
670 nodenum
= simple_strtoul(buffer
+1, NULL
, 10);
671 if (clusterip_add_node(c
, nodenum
))
673 } else if (*buffer
== '-') {
674 nodenum
= simple_strtoul(buffer
+1, NULL
,10);
675 if (clusterip_del_node(c
, nodenum
))
683 static const struct file_operations clusterip_proc_fops
= {
684 .owner
= THIS_MODULE
,
685 .open
= clusterip_proc_open
,
687 .write
= clusterip_proc_write
,
689 .release
= clusterip_proc_release
,
692 #endif /* CONFIG_PROC_FS */
694 static int __init
clusterip_tg_init(void)
698 ret
= xt_register_target(&clusterip_tg_reg
);
702 ret
= nf_register_hook(&cip_arp_ops
);
706 #ifdef CONFIG_PROC_FS
707 clusterip_procdir
= proc_mkdir("ipt_CLUSTERIP", init_net
.proc_net
);
708 if (!clusterip_procdir
) {
709 pr_err("Unable to proc dir entry\n");
713 #endif /* CONFIG_PROC_FS */
715 pr_info("ClusterIP Version %s loaded successfully\n",
719 #ifdef CONFIG_PROC_FS
721 nf_unregister_hook(&cip_arp_ops
);
722 #endif /* CONFIG_PROC_FS */
724 xt_unregister_target(&clusterip_tg_reg
);
728 static void __exit
clusterip_tg_exit(void)
730 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION
);
731 #ifdef CONFIG_PROC_FS
732 remove_proc_entry(clusterip_procdir
->name
, clusterip_procdir
->parent
);
734 nf_unregister_hook(&cip_arp_ops
);
735 xt_unregister_target(&clusterip_tg_reg
);
738 module_init(clusterip_tg_init
);
739 module_exit(clusterip_tg_exit
);