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>
25 #include <net/checksum.h>
27 #include <linux/netfilter_arp.h>
29 #include <linux/netfilter_ipv4/ip_tables.h>
30 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
31 #include <net/netfilter/nf_conntrack_compat.h>
33 #define CLUSTERIP_VERSION "0.8"
35 #define DEBUG_CLUSTERIP
37 #ifdef DEBUG_CLUSTERIP
43 MODULE_LICENSE("GPL");
44 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
45 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
47 struct clusterip_config {
48 struct list_head list; /* list of all configs */
49 atomic_t refcount; /* reference count */
50 atomic_t entries; /* number of entries/rules
53 __be32 clusterip; /* the IP address */
54 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
55 struct net_device *dev; /* device */
56 u_int16_t num_total_nodes; /* total number of nodes */
57 unsigned long local_nodes; /* node number array */
60 struct proc_dir_entry *pde; /* proc dir entry */
62 enum clusterip_hashmode hash_mode; /* which hashing mode */
63 u_int32_t hash_initval; /* hash initialization */
66 static LIST_HEAD(clusterip_configs);
68 /* clusterip_lock protects the clusterip_configs list */
69 static DEFINE_RWLOCK(clusterip_lock);
72 static struct file_operations clusterip_proc_fops;
73 static struct proc_dir_entry *clusterip_procdir;
77 clusterip_config_get(struct clusterip_config *c)
79 atomic_inc(&c->refcount);
83 clusterip_config_put(struct clusterip_config *c)
85 if (atomic_dec_and_test(&c->refcount))
89 /* increase the count of entries(rules) using/referencing this config */
91 clusterip_config_entry_get(struct clusterip_config *c)
93 atomic_inc(&c->entries);
96 /* decrease the count of entries using/referencing this config. If last
97 * entry(rule) is removed, remove the config from lists, but don't free it
98 * yet, since proc-files could still be holding references */
100 clusterip_config_entry_put(struct clusterip_config *c)
102 if (atomic_dec_and_test(&c->entries)) {
103 write_lock_bh(&clusterip_lock);
105 write_unlock_bh(&clusterip_lock);
107 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
110 /* In case anyone still accesses the file, the open/close
111 * functions are also incrementing the refcount on their own,
112 * so it's safe to remove the entry even if it's in use. */
113 #ifdef CONFIG_PROC_FS
114 remove_proc_entry(c->pde->name, c->pde->parent);
119 static struct clusterip_config *
120 __clusterip_config_find(__be32 clusterip)
122 struct list_head *pos;
124 list_for_each(pos, &clusterip_configs) {
125 struct clusterip_config *c = list_entry(pos,
126 struct clusterip_config, list);
127 if (c->clusterip == clusterip) {
135 static inline struct clusterip_config *
136 clusterip_config_find_get(__be32 clusterip, int entry)
138 struct clusterip_config *c;
140 read_lock_bh(&clusterip_lock);
141 c = __clusterip_config_find(clusterip);
143 read_unlock_bh(&clusterip_lock);
146 atomic_inc(&c->refcount);
148 atomic_inc(&c->entries);
149 read_unlock_bh(&clusterip_lock);
155 clusterip_config_init_nodelist(struct clusterip_config *c,
156 const struct ipt_clusterip_tgt_info *i)
160 for (n = 0; n < i->num_local_nodes; n++) {
161 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
165 static struct clusterip_config *
166 clusterip_config_init(struct ipt_clusterip_tgt_info *i, __be32 ip,
167 struct net_device *dev)
169 struct clusterip_config *c;
171 c = kzalloc(sizeof(*c), GFP_ATOMIC);
177 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
178 c->num_total_nodes = i->num_total_nodes;
179 clusterip_config_init_nodelist(c, i);
180 c->hash_mode = i->hash_mode;
181 c->hash_initval = i->hash_initval;
182 atomic_set(&c->refcount, 1);
183 atomic_set(&c->entries, 1);
185 #ifdef CONFIG_PROC_FS
189 /* create proc dir entry */
190 sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
191 c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR,
198 c->pde->proc_fops = &clusterip_proc_fops;
202 write_lock_bh(&clusterip_lock);
203 list_add(&c->list, &clusterip_configs);
204 write_unlock_bh(&clusterip_lock);
209 #ifdef CONFIG_PROC_FS
211 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
215 nodenum > c->num_total_nodes)
218 /* check if we already have this number in our bitfield */
219 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
226 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
229 nodenum > c->num_total_nodes)
232 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
239 static inline u_int32_t
240 clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config)
242 struct iphdr *iph = skb->nh.iph;
243 unsigned long hashval;
244 u_int16_t sport, dport;
247 switch (iph->protocol) {
250 case IPPROTO_UDPLITE:
254 ports = (void *)iph+iph->ihl*4;
259 if (net_ratelimit()) {
260 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
266 switch (config->hash_mode) {
267 case CLUSTERIP_HASHMODE_SIP:
268 hashval = jhash_1word(ntohl(iph->saddr),
269 config->hash_initval);
271 case CLUSTERIP_HASHMODE_SIP_SPT:
272 hashval = jhash_2words(ntohl(iph->saddr), sport,
273 config->hash_initval);
275 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
276 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
277 config->hash_initval);
280 /* to make gcc happy */
282 /* This cannot happen, unless the check function wasn't called
283 * at rule load time */
284 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
289 /* node numbers are 1..n, not 0..n */
290 return ((hashval % config->num_total_nodes)+1);
294 clusterip_responsible(struct clusterip_config *config, u_int32_t hash)
296 return test_bit(hash - 1, &config->local_nodes);
299 /***********************************************************************
301 ***********************************************************************/
304 target(struct sk_buff **pskb,
305 const struct net_device *in,
306 const struct net_device *out,
307 unsigned int hooknum,
308 const struct xt_target *target,
309 const void *targinfo)
311 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
312 enum ip_conntrack_info ctinfo;
313 u_int32_t *mark, hash;
315 /* don't need to clusterip_config_get() here, since refcount
316 * is only decremented by destroy() - and ip_tables guarantees
317 * that the ->target() function isn't called after ->destroy() */
319 mark = nf_ct_get_mark((*pskb), &ctinfo);
321 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
322 /* FIXME: need to drop invalid ones, since replies
323 * to outgoing connections of other nodes will be
324 * marked as INVALID */
328 /* special case: ICMP error handling. conntrack distinguishes between
329 * error messages (RELATED) and information requests (see below) */
330 if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP
331 && (ctinfo == IP_CT_RELATED
332 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
335 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
336 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
337 * on, which all have an ID field [relevant for hashing]. */
339 hash = clusterip_hashfn(*pskb, cipinfo->config);
346 case IP_CT_RELATED+IP_CT_IS_REPLY:
347 /* FIXME: we don't handle expectations at the
348 * moment. they can arrive on a different node than
349 * the master connection (e.g. FTP passive mode) */
350 case IP_CT_ESTABLISHED:
351 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
357 #ifdef DEBUG_CLUSTERP
358 DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
360 DEBUGP("hash=%u ct_hash=%u ", hash, *mark);
361 if (!clusterip_responsible(cipinfo->config, hash)) {
362 DEBUGP("not responsible\n");
365 DEBUGP("responsible\n");
367 /* despite being received via linklayer multicast, this is
368 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
369 (*pskb)->pkt_type = PACKET_HOST;
375 checkentry(const char *tablename,
377 const struct xt_target *target,
379 unsigned int hook_mask)
381 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
382 const struct ipt_entry *e = e_void;
384 struct clusterip_config *config;
386 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
387 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
388 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
389 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
394 if (e->ip.dmsk.s_addr != htonl(0xffffffff)
395 || e->ip.dst.s_addr == 0) {
396 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
400 /* FIXME: further sanity checks */
402 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
404 if (cipinfo->config != NULL) {
405 /* Case A: This is an entry that gets reloaded, since
406 * it still has a cipinfo->config pointer. Simply
407 * increase the entry refcount and return */
408 if (cipinfo->config != config) {
409 printk(KERN_ERR "CLUSTERIP: Reloaded entry "
410 "has invalid config pointer!\n");
413 clusterip_config_entry_get(cipinfo->config);
415 /* Case B: This is a new rule referring to an existing
416 * clusterip config. */
417 cipinfo->config = config;
418 clusterip_config_entry_get(cipinfo->config);
421 /* Case C: This is a completely new clusterip config */
422 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
423 printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
426 struct net_device *dev;
428 if (e->ip.iniface[0] == '\0') {
429 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
433 dev = dev_get_by_name(e->ip.iniface);
435 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
439 config = clusterip_config_init(cipinfo,
440 e->ip.dst.s_addr, dev);
442 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
446 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
448 cipinfo->config = config;
451 if (nf_ct_l3proto_try_module_get(target->family) < 0) {
452 printk(KERN_WARNING "can't load conntrack support for "
453 "proto=%d\n", target->family);
460 /* drop reference count of cluster config when rule is deleted */
461 static void destroy(const struct xt_target *target, void *targinfo)
463 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
465 /* if no more entries are referencing the config, remove it
466 * from the list and destroy the proc entry */
467 clusterip_config_entry_put(cipinfo->config);
469 clusterip_config_put(cipinfo->config);
471 nf_ct_l3proto_module_put(target->family);
474 static struct ipt_target clusterip_tgt = {
477 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
478 .checkentry = checkentry,
484 /***********************************************************************
486 ***********************************************************************/
488 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
490 u_int8_t src_hw[ETH_ALEN];
492 u_int8_t dst_hw[ETH_ALEN];
494 } __attribute__ ((packed));
496 #ifdef CLUSTERIP_DEBUG
497 static void arp_print(struct arp_payload *payload)
499 #define HBUFFERLEN 30
500 char hbuffer[HBUFFERLEN];
502 const char hexbuf[]= "0123456789abcdef";
504 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
505 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
506 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
511 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
512 NIPQUAD(payload->src_ip), hbuffer,
513 NIPQUAD(payload->dst_ip));
518 arp_mangle(unsigned int hook,
519 struct sk_buff **pskb,
520 const struct net_device *in,
521 const struct net_device *out,
522 int (*okfn)(struct sk_buff *))
524 struct arphdr *arp = (*pskb)->nh.arph;
525 struct arp_payload *payload;
526 struct clusterip_config *c;
528 /* we don't care about non-ethernet and non-ipv4 ARP */
529 if (arp->ar_hrd != htons(ARPHRD_ETHER)
530 || arp->ar_pro != htons(ETH_P_IP)
531 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
534 /* we only want to mangle arp requests and replies */
535 if (arp->ar_op != htons(ARPOP_REPLY)
536 && arp->ar_op != htons(ARPOP_REQUEST))
539 payload = (void *)(arp+1);
541 /* if there is no clusterip configuration for the arp reply's
542 * source ip, we don't want to mangle it */
543 c = clusterip_config_find_get(payload->src_ip, 0);
547 /* normally the linux kernel always replies to arp queries of
548 * addresses on different interfacs. However, in the CLUSTERIP case
549 * this wouldn't work, since we didn't subscribe the mcast group on
550 * other interfaces */
552 DEBUGP("CLUSTERIP: not mangling arp reply on different "
553 "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name);
554 clusterip_config_put(c);
558 /* mangle reply hardware address */
559 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
561 #ifdef CLUSTERIP_DEBUG
562 DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
566 clusterip_config_put(c);
571 static struct nf_hook_ops cip_arp_ops = {
574 .hooknum = NF_ARP_OUT,
578 /***********************************************************************
580 ***********************************************************************/
582 #ifdef CONFIG_PROC_FS
584 struct clusterip_seq_position {
585 unsigned int pos; /* position */
586 unsigned int weight; /* number of bits set == size */
587 unsigned int bit; /* current bit */
588 unsigned long val; /* current value */
591 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
593 struct proc_dir_entry *pde = s->private;
594 struct clusterip_config *c = pde->data;
596 u_int32_t local_nodes;
597 struct clusterip_seq_position *idx;
599 /* FIXME: possible race */
600 local_nodes = c->local_nodes;
601 weight = hweight32(local_nodes);
605 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
607 return ERR_PTR(-ENOMEM);
610 idx->weight = weight;
611 idx->bit = ffs(local_nodes);
612 idx->val = local_nodes;
613 clear_bit(idx->bit - 1, &idx->val);
618 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
620 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
623 if (*pos >= idx->weight) {
627 idx->bit = ffs(idx->val);
628 clear_bit(idx->bit - 1, &idx->val);
632 static void clusterip_seq_stop(struct seq_file *s, void *v)
637 static int clusterip_seq_show(struct seq_file *s, void *v)
639 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
644 seq_printf(s, "%u", idx->bit);
646 if (idx->pos == idx->weight - 1)
652 static struct seq_operations clusterip_seq_ops = {
653 .start = clusterip_seq_start,
654 .next = clusterip_seq_next,
655 .stop = clusterip_seq_stop,
656 .show = clusterip_seq_show,
659 static int clusterip_proc_open(struct inode *inode, struct file *file)
661 int ret = seq_open(file, &clusterip_seq_ops);
664 struct seq_file *sf = file->private_data;
665 struct proc_dir_entry *pde = PDE(inode);
666 struct clusterip_config *c = pde->data;
670 clusterip_config_get(c);
676 static int clusterip_proc_release(struct inode *inode, struct file *file)
678 struct proc_dir_entry *pde = PDE(inode);
679 struct clusterip_config *c = pde->data;
682 ret = seq_release(inode, file);
685 clusterip_config_put(c);
690 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
691 size_t size, loff_t *ofs)
693 #define PROC_WRITELEN 10
694 char buffer[PROC_WRITELEN+1];
695 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
696 struct clusterip_config *c = pde->data;
697 unsigned long nodenum;
699 if (copy_from_user(buffer, input, PROC_WRITELEN))
702 if (*buffer == '+') {
703 nodenum = simple_strtoul(buffer+1, NULL, 10);
704 if (clusterip_add_node(c, nodenum))
706 } else if (*buffer == '-') {
707 nodenum = simple_strtoul(buffer+1, NULL,10);
708 if (clusterip_del_node(c, nodenum))
716 static struct file_operations clusterip_proc_fops = {
717 .owner = THIS_MODULE,
718 .open = clusterip_proc_open,
720 .write = clusterip_proc_write,
722 .release = clusterip_proc_release,
725 #endif /* CONFIG_PROC_FS */
727 static int __init ipt_clusterip_init(void)
731 ret = ipt_register_target(&clusterip_tgt);
735 ret = nf_register_hook(&cip_arp_ops);
739 #ifdef CONFIG_PROC_FS
740 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net);
741 if (!clusterip_procdir) {
742 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
746 #endif /* CONFIG_PROC_FS */
748 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
752 #ifdef CONFIG_PROC_FS
754 nf_unregister_hook(&cip_arp_ops);
755 #endif /* CONFIG_PROC_FS */
757 ipt_unregister_target(&clusterip_tgt);
761 static void __exit ipt_clusterip_fini(void)
763 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
765 #ifdef CONFIG_PROC_FS
766 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
768 nf_unregister_hook(&cip_arp_ops);
769 ipt_unregister_target(&clusterip_tgt);
772 module_init(ipt_clusterip_init);
773 module_exit(ipt_clusterip_fini);