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/proc_fs.h>
23 #include <linux/seq_file.h>
24 #include <linux/netfilter_arp.h>
25 #include <linux/netfilter/x_tables.h>
26 #include <linux/netfilter_ipv4/ip_tables.h>
27 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
28 #include <net/netfilter/nf_conntrack.h>
29 #include <net/checksum.h>
31 #define CLUSTERIP_VERSION "0.8"
33 #define DEBUG_CLUSTERIP
35 #ifdef DEBUG_CLUSTERIP
41 MODULE_LICENSE("GPL");
42 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
43 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
45 struct clusterip_config {
46 struct list_head list; /* list of all configs */
47 atomic_t refcount; /* reference count */
48 atomic_t entries; /* number of entries/rules
51 __be32 clusterip; /* the IP address */
52 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
53 struct net_device *dev; /* device */
54 u_int16_t num_total_nodes; /* total number of nodes */
55 unsigned long local_nodes; /* node number array */
58 struct proc_dir_entry *pde; /* proc dir entry */
60 enum clusterip_hashmode hash_mode; /* which hashing mode */
61 u_int32_t hash_initval; /* hash initialization */
64 static LIST_HEAD(clusterip_configs);
66 /* clusterip_lock protects the clusterip_configs list */
67 static DEFINE_RWLOCK(clusterip_lock);
70 static const struct file_operations clusterip_proc_fops;
71 static struct proc_dir_entry *clusterip_procdir;
75 clusterip_config_get(struct clusterip_config *c)
77 atomic_inc(&c->refcount);
81 clusterip_config_put(struct clusterip_config *c)
83 if (atomic_dec_and_test(&c->refcount))
87 /* increase the count of entries(rules) using/referencing this config */
89 clusterip_config_entry_get(struct clusterip_config *c)
91 atomic_inc(&c->entries);
94 /* decrease the count of entries using/referencing this config. If last
95 * entry(rule) is removed, remove the config from lists, but don't free it
96 * yet, since proc-files could still be holding references */
98 clusterip_config_entry_put(struct clusterip_config *c)
100 if (atomic_dec_and_test(&c->entries)) {
101 write_lock_bh(&clusterip_lock);
103 write_unlock_bh(&clusterip_lock);
105 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
108 /* In case anyone still accesses the file, the open/close
109 * functions are also incrementing the refcount on their own,
110 * so it's safe to remove the entry even if it's in use. */
111 #ifdef CONFIG_PROC_FS
112 remove_proc_entry(c->pde->name, c->pde->parent);
117 static struct clusterip_config *
118 __clusterip_config_find(__be32 clusterip)
120 struct list_head *pos;
122 list_for_each(pos, &clusterip_configs) {
123 struct clusterip_config *c = list_entry(pos,
124 struct clusterip_config, list);
125 if (c->clusterip == clusterip)
132 static inline struct clusterip_config *
133 clusterip_config_find_get(__be32 clusterip, int entry)
135 struct clusterip_config *c;
137 read_lock_bh(&clusterip_lock);
138 c = __clusterip_config_find(clusterip);
140 read_unlock_bh(&clusterip_lock);
143 atomic_inc(&c->refcount);
145 atomic_inc(&c->entries);
146 read_unlock_bh(&clusterip_lock);
152 clusterip_config_init_nodelist(struct clusterip_config *c,
153 const struct ipt_clusterip_tgt_info *i)
157 for (n = 0; n < i->num_local_nodes; n++)
158 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
161 static struct clusterip_config *
162 clusterip_config_init(struct ipt_clusterip_tgt_info *i, __be32 ip,
163 struct net_device *dev)
165 struct clusterip_config *c;
167 c = kzalloc(sizeof(*c), GFP_ATOMIC);
173 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
174 c->num_total_nodes = i->num_total_nodes;
175 clusterip_config_init_nodelist(c, i);
176 c->hash_mode = i->hash_mode;
177 c->hash_initval = i->hash_initval;
178 atomic_set(&c->refcount, 1);
179 atomic_set(&c->entries, 1);
181 #ifdef CONFIG_PROC_FS
185 /* create proc dir entry */
186 sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
187 c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR,
194 c->pde->proc_fops = &clusterip_proc_fops;
198 write_lock_bh(&clusterip_lock);
199 list_add(&c->list, &clusterip_configs);
200 write_unlock_bh(&clusterip_lock);
205 #ifdef CONFIG_PROC_FS
207 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
211 nodenum > c->num_total_nodes)
214 /* check if we already have this number in our bitfield */
215 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
222 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
225 nodenum > c->num_total_nodes)
228 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
235 static inline u_int32_t
236 clusterip_hashfn(const struct sk_buff *skb,
237 const struct clusterip_config *config)
239 const struct iphdr *iph = ip_hdr(skb);
240 unsigned long hashval;
241 u_int16_t sport, dport;
242 const u_int16_t *ports;
244 switch (iph->protocol) {
247 case IPPROTO_UDPLITE:
251 ports = (const void *)iph+iph->ihl*4;
257 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
262 switch (config->hash_mode) {
263 case CLUSTERIP_HASHMODE_SIP:
264 hashval = jhash_1word(ntohl(iph->saddr),
265 config->hash_initval);
267 case CLUSTERIP_HASHMODE_SIP_SPT:
268 hashval = jhash_2words(ntohl(iph->saddr), sport,
269 config->hash_initval);
271 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
272 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
273 config->hash_initval);
276 /* to make gcc happy */
278 /* This cannot happen, unless the check function wasn't called
279 * at rule load time */
280 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
285 /* node numbers are 1..n, not 0..n */
286 return (hashval % config->num_total_nodes) + 1;
290 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
292 return test_bit(hash - 1, &config->local_nodes);
295 /***********************************************************************
297 ***********************************************************************/
300 target(struct sk_buff **pskb,
301 const struct net_device *in,
302 const struct net_device *out,
303 unsigned int hooknum,
304 const struct xt_target *target,
305 const void *targinfo)
307 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
309 enum ip_conntrack_info ctinfo;
312 /* don't need to clusterip_config_get() here, since refcount
313 * is only decremented by destroy() - and ip_tables guarantees
314 * that the ->target() function isn't called after ->destroy() */
316 ct = nf_ct_get(*pskb, &ctinfo);
318 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
319 /* FIXME: need to drop invalid ones, since replies
320 * to outgoing connections of other nodes will be
321 * marked as INVALID */
325 /* special case: ICMP error handling. conntrack distinguishes between
326 * error messages (RELATED) and information requests (see below) */
327 if (ip_hdr(*pskb)->protocol == IPPROTO_ICMP
328 && (ctinfo == IP_CT_RELATED
329 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
332 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
333 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
334 * on, which all have an ID field [relevant for hashing]. */
336 hash = clusterip_hashfn(*pskb, cipinfo->config);
343 case IP_CT_RELATED+IP_CT_IS_REPLY:
344 /* FIXME: we don't handle expectations at the
345 * moment. they can arrive on a different node than
346 * the master connection (e.g. FTP passive mode) */
347 case IP_CT_ESTABLISHED:
348 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
354 #ifdef DEBUG_CLUSTERP
355 DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
357 DEBUGP("hash=%u ct_hash=%u ", hash, ct->mark);
358 if (!clusterip_responsible(cipinfo->config, hash)) {
359 DEBUGP("not responsible\n");
362 DEBUGP("responsible\n");
364 /* despite being received via linklayer multicast, this is
365 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
366 (*pskb)->pkt_type = PACKET_HOST;
372 checkentry(const char *tablename,
374 const struct xt_target *target,
376 unsigned int hook_mask)
378 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
379 const struct ipt_entry *e = e_void;
381 struct clusterip_config *config;
383 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
384 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
385 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
386 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
391 if (e->ip.dmsk.s_addr != htonl(0xffffffff)
392 || e->ip.dst.s_addr == 0) {
393 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
397 /* FIXME: further sanity checks */
399 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
401 if (cipinfo->config != NULL) {
402 /* Case A: This is an entry that gets reloaded, since
403 * it still has a cipinfo->config pointer. Simply
404 * increase the entry refcount and return */
405 if (cipinfo->config != config) {
406 printk(KERN_ERR "CLUSTERIP: Reloaded entry "
407 "has invalid config pointer!\n");
411 /* Case B: This is a new rule referring to an existing
412 * clusterip config. */
413 cipinfo->config = config;
416 /* Case C: This is a completely new clusterip config */
417 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
418 printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
421 struct net_device *dev;
423 if (e->ip.iniface[0] == '\0') {
424 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
428 dev = dev_get_by_name(e->ip.iniface);
430 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
434 config = clusterip_config_init(cipinfo,
435 e->ip.dst.s_addr, dev);
437 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
441 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
443 cipinfo->config = config;
446 if (nf_ct_l3proto_try_module_get(target->family) < 0) {
447 printk(KERN_WARNING "can't load conntrack support for "
448 "proto=%d\n", target->family);
455 /* drop reference count of cluster config when rule is deleted */
456 static void destroy(const struct xt_target *target, void *targinfo)
458 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
460 /* if no more entries are referencing the config, remove it
461 * from the list and destroy the proc entry */
462 clusterip_config_entry_put(cipinfo->config);
464 clusterip_config_put(cipinfo->config);
466 nf_ct_l3proto_module_put(target->family);
469 static struct xt_target clusterip_tgt = {
473 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
474 .checkentry = checkentry,
480 /***********************************************************************
482 ***********************************************************************/
484 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
486 u_int8_t src_hw[ETH_ALEN];
488 u_int8_t dst_hw[ETH_ALEN];
490 } __attribute__ ((packed));
492 #ifdef CLUSTERIP_DEBUG
493 static void arp_print(struct arp_payload *payload)
495 #define HBUFFERLEN 30
496 char hbuffer[HBUFFERLEN];
498 const char hexbuf[]= "0123456789abcdef";
500 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
501 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
502 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
507 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
508 NIPQUAD(payload->src_ip), hbuffer,
509 NIPQUAD(payload->dst_ip));
514 arp_mangle(unsigned int hook,
515 struct sk_buff **pskb,
516 const struct net_device *in,
517 const struct net_device *out,
518 int (*okfn)(struct sk_buff *))
520 struct arphdr *arp = arp_hdr(*pskb);
521 struct arp_payload *payload;
522 struct clusterip_config *c;
524 /* we don't care about non-ethernet and non-ipv4 ARP */
525 if (arp->ar_hrd != htons(ARPHRD_ETHER)
526 || arp->ar_pro != htons(ETH_P_IP)
527 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
530 /* we only want to mangle arp requests and replies */
531 if (arp->ar_op != htons(ARPOP_REPLY)
532 && arp->ar_op != htons(ARPOP_REQUEST))
535 payload = (void *)(arp+1);
537 /* if there is no clusterip configuration for the arp reply's
538 * source ip, we don't want to mangle it */
539 c = clusterip_config_find_get(payload->src_ip, 0);
543 /* normally the linux kernel always replies to arp queries of
544 * addresses on different interfacs. However, in the CLUSTERIP case
545 * this wouldn't work, since we didn't subscribe the mcast group on
546 * other interfaces */
548 DEBUGP("CLUSTERIP: not mangling arp reply on different "
549 "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name);
550 clusterip_config_put(c);
554 /* mangle reply hardware address */
555 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
557 #ifdef CLUSTERIP_DEBUG
558 DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
562 clusterip_config_put(c);
567 static struct nf_hook_ops cip_arp_ops = {
570 .hooknum = NF_ARP_OUT,
574 /***********************************************************************
576 ***********************************************************************/
578 #ifdef CONFIG_PROC_FS
580 struct clusterip_seq_position {
581 unsigned int pos; /* position */
582 unsigned int weight; /* number of bits set == size */
583 unsigned int bit; /* current bit */
584 unsigned long val; /* current value */
587 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
589 struct proc_dir_entry *pde = s->private;
590 struct clusterip_config *c = pde->data;
592 u_int32_t local_nodes;
593 struct clusterip_seq_position *idx;
595 /* FIXME: possible race */
596 local_nodes = c->local_nodes;
597 weight = hweight32(local_nodes);
601 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
603 return ERR_PTR(-ENOMEM);
606 idx->weight = weight;
607 idx->bit = ffs(local_nodes);
608 idx->val = local_nodes;
609 clear_bit(idx->bit - 1, &idx->val);
614 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
616 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
619 if (*pos >= idx->weight) {
623 idx->bit = ffs(idx->val);
624 clear_bit(idx->bit - 1, &idx->val);
628 static void clusterip_seq_stop(struct seq_file *s, void *v)
633 static int clusterip_seq_show(struct seq_file *s, void *v)
635 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
640 seq_printf(s, "%u", idx->bit);
642 if (idx->pos == idx->weight - 1)
648 static struct seq_operations clusterip_seq_ops = {
649 .start = clusterip_seq_start,
650 .next = clusterip_seq_next,
651 .stop = clusterip_seq_stop,
652 .show = clusterip_seq_show,
655 static int clusterip_proc_open(struct inode *inode, struct file *file)
657 int ret = seq_open(file, &clusterip_seq_ops);
660 struct seq_file *sf = file->private_data;
661 struct proc_dir_entry *pde = PDE(inode);
662 struct clusterip_config *c = pde->data;
666 clusterip_config_get(c);
672 static int clusterip_proc_release(struct inode *inode, struct file *file)
674 struct proc_dir_entry *pde = PDE(inode);
675 struct clusterip_config *c = pde->data;
678 ret = seq_release(inode, file);
681 clusterip_config_put(c);
686 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
687 size_t size, loff_t *ofs)
689 #define PROC_WRITELEN 10
690 char buffer[PROC_WRITELEN+1];
691 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
692 struct clusterip_config *c = pde->data;
693 unsigned long nodenum;
695 if (copy_from_user(buffer, input, PROC_WRITELEN))
698 if (*buffer == '+') {
699 nodenum = simple_strtoul(buffer+1, NULL, 10);
700 if (clusterip_add_node(c, nodenum))
702 } else if (*buffer == '-') {
703 nodenum = simple_strtoul(buffer+1, NULL,10);
704 if (clusterip_del_node(c, nodenum))
712 static const struct file_operations clusterip_proc_fops = {
713 .owner = THIS_MODULE,
714 .open = clusterip_proc_open,
716 .write = clusterip_proc_write,
718 .release = clusterip_proc_release,
721 #endif /* CONFIG_PROC_FS */
723 static int __init ipt_clusterip_init(void)
727 ret = xt_register_target(&clusterip_tgt);
731 ret = nf_register_hook(&cip_arp_ops);
735 #ifdef CONFIG_PROC_FS
736 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net);
737 if (!clusterip_procdir) {
738 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
742 #endif /* CONFIG_PROC_FS */
744 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
748 #ifdef CONFIG_PROC_FS
750 nf_unregister_hook(&cip_arp_ops);
751 #endif /* CONFIG_PROC_FS */
753 xt_unregister_target(&clusterip_tgt);
757 static void __exit ipt_clusterip_fini(void)
759 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
761 #ifdef CONFIG_PROC_FS
762 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
764 nf_unregister_hook(&cip_arp_ops);
765 xt_unregister_target(&clusterip_tgt);
768 module_init(ipt_clusterip_init);
769 module_exit(ipt_clusterip_fini);