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("iptables target for CLUSTERIP");
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 /* increase the count of entries(rules) using/referencing this config */
81 clusterip_config_entry_get(struct clusterip_config *c)
83 atomic_inc(&c->entries);
86 /* decrease the count of entries using/referencing this config. If last
87 * entry(rule) is removed, remove the config from lists, but don't free it
88 * yet, since proc-files could still be holding references */
90 clusterip_config_entry_put(struct clusterip_config *c)
92 if (atomic_dec_and_test(&c->entries)) {
93 write_lock_bh(&clusterip_lock);
95 write_unlock_bh(&clusterip_lock);
97 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
100 /* In case anyone still accesses the file, the open/close
101 * functions are also incrementing the refcount on their own,
102 * so it's safe to remove the entry even if it's in use. */
103 #ifdef CONFIG_PROC_FS
104 remove_proc_entry(c->pde->name, c->pde->parent);
109 static struct clusterip_config *
110 __clusterip_config_find(__be32 clusterip)
112 struct list_head *pos;
114 list_for_each(pos, &clusterip_configs) {
115 struct clusterip_config *c = list_entry(pos,
116 struct clusterip_config, list);
117 if (c->clusterip == clusterip)
124 static inline struct clusterip_config *
125 clusterip_config_find_get(__be32 clusterip, int entry)
127 struct clusterip_config *c;
129 read_lock_bh(&clusterip_lock);
130 c = __clusterip_config_find(clusterip);
132 read_unlock_bh(&clusterip_lock);
135 atomic_inc(&c->refcount);
137 atomic_inc(&c->entries);
138 read_unlock_bh(&clusterip_lock);
144 clusterip_config_init_nodelist(struct clusterip_config *c,
145 const struct ipt_clusterip_tgt_info *i)
149 for (n = 0; n < i->num_local_nodes; n++)
150 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
153 static struct clusterip_config *
154 clusterip_config_init(struct ipt_clusterip_tgt_info *i, __be32 ip,
155 struct net_device *dev)
157 struct clusterip_config *c;
159 c = kzalloc(sizeof(*c), GFP_ATOMIC);
165 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
166 c->num_total_nodes = i->num_total_nodes;
167 clusterip_config_init_nodelist(c, i);
168 c->hash_mode = i->hash_mode;
169 c->hash_initval = i->hash_initval;
170 atomic_set(&c->refcount, 1);
171 atomic_set(&c->entries, 1);
173 #ifdef CONFIG_PROC_FS
177 /* create proc dir entry */
178 sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
179 c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR,
186 c->pde->proc_fops = &clusterip_proc_fops;
190 write_lock_bh(&clusterip_lock);
191 list_add(&c->list, &clusterip_configs);
192 write_unlock_bh(&clusterip_lock);
197 #ifdef CONFIG_PROC_FS
199 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
203 nodenum > c->num_total_nodes)
206 /* check if we already have this number in our bitfield */
207 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
214 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
217 nodenum > c->num_total_nodes)
220 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
227 static inline u_int32_t
228 clusterip_hashfn(const struct sk_buff *skb,
229 const struct clusterip_config *config)
231 const struct iphdr *iph = ip_hdr(skb);
232 unsigned long hashval;
233 u_int16_t sport, dport;
234 const u_int16_t *ports;
236 switch (iph->protocol) {
239 case IPPROTO_UDPLITE:
243 ports = (const void *)iph+iph->ihl*4;
249 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
254 switch (config->hash_mode) {
255 case CLUSTERIP_HASHMODE_SIP:
256 hashval = jhash_1word(ntohl(iph->saddr),
257 config->hash_initval);
259 case CLUSTERIP_HASHMODE_SIP_SPT:
260 hashval = jhash_2words(ntohl(iph->saddr), sport,
261 config->hash_initval);
263 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
264 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
265 config->hash_initval);
268 /* to make gcc happy */
270 /* This cannot happen, unless the check function wasn't called
271 * at rule load time */
272 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
277 /* node numbers are 1..n, not 0..n */
278 return (hashval % config->num_total_nodes) + 1;
282 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
284 return test_bit(hash - 1, &config->local_nodes);
287 /***********************************************************************
289 ***********************************************************************/
292 target(struct sk_buff *skb,
293 const struct net_device *in,
294 const struct net_device *out,
295 unsigned int hooknum,
296 const struct xt_target *target,
297 const void *targinfo)
299 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
301 enum ip_conntrack_info ctinfo;
304 /* don't need to clusterip_config_get() here, since refcount
305 * is only decremented by destroy() - and ip_tables guarantees
306 * that the ->target() function isn't called after ->destroy() */
308 ct = nf_ct_get(skb, &ctinfo);
310 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
311 /* FIXME: need to drop invalid ones, since replies
312 * to outgoing connections of other nodes will be
313 * marked as INVALID */
317 /* special case: ICMP error handling. conntrack distinguishes between
318 * error messages (RELATED) and information requests (see below) */
319 if (ip_hdr(skb)->protocol == IPPROTO_ICMP
320 && (ctinfo == IP_CT_RELATED
321 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
324 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
325 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
326 * on, which all have an ID field [relevant for hashing]. */
328 hash = clusterip_hashfn(skb, cipinfo->config);
335 case IP_CT_RELATED+IP_CT_IS_REPLY:
336 /* FIXME: we don't handle expectations at the
337 * moment. they can arrive on a different node than
338 * the master connection (e.g. FTP passive mode) */
339 case IP_CT_ESTABLISHED:
340 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
347 DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
349 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
350 if (!clusterip_responsible(cipinfo->config, hash)) {
351 pr_debug("not responsible\n");
354 pr_debug("responsible\n");
356 /* despite being received via linklayer multicast, this is
357 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
358 skb->pkt_type = PACKET_HOST;
364 checkentry(const char *tablename,
366 const struct xt_target *target,
368 unsigned int hook_mask)
370 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
371 const struct ipt_entry *e = e_void;
373 struct clusterip_config *config;
375 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
376 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
377 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
378 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
383 if (e->ip.dmsk.s_addr != htonl(0xffffffff)
384 || e->ip.dst.s_addr == 0) {
385 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
389 /* FIXME: further sanity checks */
391 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
393 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
394 printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
397 struct net_device *dev;
399 if (e->ip.iniface[0] == '\0') {
400 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
404 dev = dev_get_by_name(&init_net, e->ip.iniface);
406 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
410 config = clusterip_config_init(cipinfo,
411 e->ip.dst.s_addr, dev);
413 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
417 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
420 cipinfo->config = config;
422 if (nf_ct_l3proto_try_module_get(target->family) < 0) {
423 printk(KERN_WARNING "can't load conntrack support for "
424 "proto=%d\n", target->family);
431 /* drop reference count of cluster config when rule is deleted */
432 static void destroy(const struct xt_target *target, void *targinfo)
434 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
436 /* if no more entries are referencing the config, remove it
437 * from the list and destroy the proc entry */
438 clusterip_config_entry_put(cipinfo->config);
440 clusterip_config_put(cipinfo->config);
442 nf_ct_l3proto_module_put(target->family);
446 struct compat_ipt_clusterip_tgt_info
449 u_int8_t clustermac[6];
450 u_int16_t num_total_nodes;
451 u_int16_t num_local_nodes;
452 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
454 u_int32_t hash_initval;
455 compat_uptr_t config;
457 #endif /* CONFIG_COMPAT */
459 static struct xt_target clusterip_tgt __read_mostly = {
463 .checkentry = checkentry,
465 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
467 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
468 #endif /* CONFIG_COMPAT */
473 /***********************************************************************
475 ***********************************************************************/
477 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
479 u_int8_t src_hw[ETH_ALEN];
481 u_int8_t dst_hw[ETH_ALEN];
483 } __attribute__ ((packed));
486 static void arp_print(struct arp_payload *payload)
488 #define HBUFFERLEN 30
489 char hbuffer[HBUFFERLEN];
491 const char hexbuf[]= "0123456789abcdef";
493 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
494 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
495 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
500 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
501 NIPQUAD(payload->src_ip), hbuffer,
502 NIPQUAD(payload->dst_ip));
507 arp_mangle(unsigned int hook,
509 const struct net_device *in,
510 const struct net_device *out,
511 int (*okfn)(struct sk_buff *))
513 struct arphdr *arp = arp_hdr(skb);
514 struct arp_payload *payload;
515 struct clusterip_config *c;
517 /* we don't care about non-ethernet and non-ipv4 ARP */
518 if (arp->ar_hrd != htons(ARPHRD_ETHER)
519 || arp->ar_pro != htons(ETH_P_IP)
520 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
523 /* we only want to mangle arp requests and replies */
524 if (arp->ar_op != htons(ARPOP_REPLY)
525 && arp->ar_op != htons(ARPOP_REQUEST))
528 payload = (void *)(arp+1);
530 /* if there is no clusterip configuration for the arp reply's
531 * source ip, we don't want to mangle it */
532 c = clusterip_config_find_get(payload->src_ip, 0);
536 /* normally the linux kernel always replies to arp queries of
537 * addresses on different interfacs. However, in the CLUSTERIP case
538 * this wouldn't work, since we didn't subscribe the mcast group on
539 * other interfaces */
541 pr_debug("CLUSTERIP: not mangling arp reply on different "
542 "interface: cip'%s'-skb'%s'\n",
543 c->dev->name, out->name);
544 clusterip_config_put(c);
548 /* mangle reply hardware address */
549 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
552 pr_debug(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
556 clusterip_config_put(c);
561 static struct nf_hook_ops cip_arp_ops = {
564 .hooknum = NF_ARP_OUT,
568 /***********************************************************************
570 ***********************************************************************/
572 #ifdef CONFIG_PROC_FS
574 struct clusterip_seq_position {
575 unsigned int pos; /* position */
576 unsigned int weight; /* number of bits set == size */
577 unsigned int bit; /* current bit */
578 unsigned long val; /* current value */
581 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
583 struct proc_dir_entry *pde = s->private;
584 struct clusterip_config *c = pde->data;
586 u_int32_t local_nodes;
587 struct clusterip_seq_position *idx;
589 /* FIXME: possible race */
590 local_nodes = c->local_nodes;
591 weight = hweight32(local_nodes);
595 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
597 return ERR_PTR(-ENOMEM);
600 idx->weight = weight;
601 idx->bit = ffs(local_nodes);
602 idx->val = local_nodes;
603 clear_bit(idx->bit - 1, &idx->val);
608 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
610 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
613 if (*pos >= idx->weight) {
617 idx->bit = ffs(idx->val);
618 clear_bit(idx->bit - 1, &idx->val);
622 static void clusterip_seq_stop(struct seq_file *s, void *v)
627 static int clusterip_seq_show(struct seq_file *s, void *v)
629 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
634 seq_printf(s, "%u", idx->bit);
636 if (idx->pos == idx->weight - 1)
642 static const struct seq_operations clusterip_seq_ops = {
643 .start = clusterip_seq_start,
644 .next = clusterip_seq_next,
645 .stop = clusterip_seq_stop,
646 .show = clusterip_seq_show,
649 static int clusterip_proc_open(struct inode *inode, struct file *file)
651 int ret = seq_open(file, &clusterip_seq_ops);
654 struct seq_file *sf = file->private_data;
655 struct proc_dir_entry *pde = PDE(inode);
656 struct clusterip_config *c = pde->data;
660 clusterip_config_get(c);
666 static int clusterip_proc_release(struct inode *inode, struct file *file)
668 struct proc_dir_entry *pde = PDE(inode);
669 struct clusterip_config *c = pde->data;
672 ret = seq_release(inode, file);
675 clusterip_config_put(c);
680 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
681 size_t size, loff_t *ofs)
683 #define PROC_WRITELEN 10
684 char buffer[PROC_WRITELEN+1];
685 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
686 struct clusterip_config *c = pde->data;
687 unsigned long nodenum;
689 if (copy_from_user(buffer, input, PROC_WRITELEN))
692 if (*buffer == '+') {
693 nodenum = simple_strtoul(buffer+1, NULL, 10);
694 if (clusterip_add_node(c, nodenum))
696 } else if (*buffer == '-') {
697 nodenum = simple_strtoul(buffer+1, NULL,10);
698 if (clusterip_del_node(c, nodenum))
706 static const struct file_operations clusterip_proc_fops = {
707 .owner = THIS_MODULE,
708 .open = clusterip_proc_open,
710 .write = clusterip_proc_write,
712 .release = clusterip_proc_release,
715 #endif /* CONFIG_PROC_FS */
717 static int __init ipt_clusterip_init(void)
721 ret = xt_register_target(&clusterip_tgt);
725 ret = nf_register_hook(&cip_arp_ops);
729 #ifdef CONFIG_PROC_FS
730 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", init_net.proc_net);
731 if (!clusterip_procdir) {
732 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
736 #endif /* CONFIG_PROC_FS */
738 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
742 #ifdef CONFIG_PROC_FS
744 nf_unregister_hook(&cip_arp_ops);
745 #endif /* CONFIG_PROC_FS */
747 xt_unregister_target(&clusterip_tgt);
751 static void __exit ipt_clusterip_fini(void)
753 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
755 #ifdef CONFIG_PROC_FS
756 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
758 nf_unregister_hook(&cip_arp_ops);
759 xt_unregister_target(&clusterip_tgt);
762 module_init(ipt_clusterip_init);
763 module_exit(ipt_clusterip_fini);