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/config.h>
14 #include <linux/proc_fs.h>
15 #include <linux/jhash.h>
16 #include <linux/bitops.h>
17 #include <linux/skbuff.h>
19 #include <linux/tcp.h>
20 #include <linux/udp.h>
21 #include <linux/icmp.h>
22 #include <linux/if_arp.h>
23 #include <linux/proc_fs.h>
24 #include <linux/seq_file.h>
26 #include <net/checksum.h>
28 #include <linux/netfilter_arp.h>
30 #include <linux/netfilter_ipv4/ip_tables.h>
31 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
32 #include <net/netfilter/nf_conntrack_compat.h>
34 #define CLUSTERIP_VERSION "0.8"
36 #define DEBUG_CLUSTERIP
38 #ifdef DEBUG_CLUSTERIP
44 #define ASSERT_READ_LOCK(x)
46 MODULE_LICENSE("GPL");
47 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
48 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
50 struct clusterip_config {
51 struct list_head list; /* list of all configs */
52 atomic_t refcount; /* reference count */
53 atomic_t entries; /* number of entries/rules
56 u_int32_t clusterip; /* the IP address */
57 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
58 struct net_device *dev; /* device */
59 u_int16_t num_total_nodes; /* total number of nodes */
60 unsigned long local_nodes; /* node number array */
63 struct proc_dir_entry *pde; /* proc dir entry */
65 enum clusterip_hashmode hash_mode; /* which hashing mode */
66 u_int32_t hash_initval; /* hash initialization */
69 static LIST_HEAD(clusterip_configs);
71 /* clusterip_lock protects the clusterip_configs list */
72 static DEFINE_RWLOCK(clusterip_lock);
75 static struct file_operations clusterip_proc_fops;
76 static struct proc_dir_entry *clusterip_procdir;
80 clusterip_config_get(struct clusterip_config *c)
82 atomic_inc(&c->refcount);
86 clusterip_config_put(struct clusterip_config *c)
88 if (atomic_dec_and_test(&c->refcount))
92 /* increase the count of entries(rules) using/referencing this config */
94 clusterip_config_entry_get(struct clusterip_config *c)
96 atomic_inc(&c->entries);
99 /* decrease the count of entries using/referencing this config. If last
100 * entry(rule) is removed, remove the config from lists, but don't free it
101 * yet, since proc-files could still be holding references */
103 clusterip_config_entry_put(struct clusterip_config *c)
105 if (atomic_dec_and_test(&c->entries)) {
106 write_lock_bh(&clusterip_lock);
108 write_unlock_bh(&clusterip_lock);
110 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
113 /* In case anyone still accesses the file, the open/close
114 * functions are also incrementing the refcount on their own,
115 * so it's safe to remove the entry even if it's in use. */
116 #ifdef CONFIG_PROC_FS
117 remove_proc_entry(c->pde->name, c->pde->parent);
122 static struct clusterip_config *
123 __clusterip_config_find(u_int32_t clusterip)
125 struct list_head *pos;
127 ASSERT_READ_LOCK(&clusterip_lock);
128 list_for_each(pos, &clusterip_configs) {
129 struct clusterip_config *c = list_entry(pos,
130 struct clusterip_config, list);
131 if (c->clusterip == clusterip) {
139 static inline struct clusterip_config *
140 clusterip_config_find_get(u_int32_t clusterip, int entry)
142 struct clusterip_config *c;
144 read_lock_bh(&clusterip_lock);
145 c = __clusterip_config_find(clusterip);
147 read_unlock_bh(&clusterip_lock);
150 atomic_inc(&c->refcount);
152 atomic_inc(&c->entries);
153 read_unlock_bh(&clusterip_lock);
159 clusterip_config_init_nodelist(struct clusterip_config *c,
160 const struct ipt_clusterip_tgt_info *i)
164 for (n = 0; n < i->num_local_nodes; n++) {
165 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
169 static struct clusterip_config *
170 clusterip_config_init(struct ipt_clusterip_tgt_info *i, u_int32_t ip,
171 struct net_device *dev)
173 struct clusterip_config *c;
176 c = kmalloc(sizeof(*c), GFP_ATOMIC);
180 memset(c, 0, sizeof(*c));
183 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
184 c->num_total_nodes = i->num_total_nodes;
185 clusterip_config_init_nodelist(c, i);
186 c->hash_mode = i->hash_mode;
187 c->hash_initval = i->hash_initval;
188 atomic_set(&c->refcount, 1);
189 atomic_set(&c->entries, 1);
191 #ifdef CONFIG_PROC_FS
192 /* create proc dir entry */
193 sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
194 c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR, clusterip_procdir);
199 c->pde->proc_fops = &clusterip_proc_fops;
203 write_lock_bh(&clusterip_lock);
204 list_add(&c->list, &clusterip_configs);
205 write_unlock_bh(&clusterip_lock);
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))
238 static inline u_int32_t
239 clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config)
241 struct iphdr *iph = skb->nh.iph;
242 unsigned long hashval;
243 u_int16_t sport, dport;
246 switch (iph->protocol) {
252 ports = (void *)iph+iph->ihl*4;
257 if (net_ratelimit()) {
258 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
264 switch (config->hash_mode) {
265 case CLUSTERIP_HASHMODE_SIP:
266 hashval = jhash_1word(ntohl(iph->saddr),
267 config->hash_initval);
269 case CLUSTERIP_HASHMODE_SIP_SPT:
270 hashval = jhash_2words(ntohl(iph->saddr), sport,
271 config->hash_initval);
273 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
274 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
275 config->hash_initval);
278 /* to make gcc happy */
280 /* This cannot happen, unless the check function wasn't called
281 * at rule load time */
282 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
287 /* node numbers are 1..n, not 0..n */
288 return ((hashval % config->num_total_nodes)+1);
292 clusterip_responsible(struct clusterip_config *config, u_int32_t hash)
294 return test_bit(hash - 1, &config->local_nodes);
297 /***********************************************************************
299 ***********************************************************************/
302 target(struct sk_buff **pskb,
303 const struct net_device *in,
304 const struct net_device *out,
305 unsigned int hooknum,
306 const struct xt_target *target,
307 const void *targinfo,
310 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
311 enum ip_conntrack_info ctinfo;
312 u_int32_t *mark, hash;
314 /* don't need to clusterip_config_get() here, since refcount
315 * is only decremented by destroy() - and ip_tables guarantees
316 * that the ->target() function isn't called after ->destroy() */
318 mark = nf_ct_get_mark((*pskb), &ctinfo);
320 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
321 /* FIXME: need to drop invalid ones, since replies
322 * to outgoing connections of other nodes will be
323 * marked as INVALID */
327 /* special case: ICMP error handling. conntrack distinguishes between
328 * error messages (RELATED) and information requests (see below) */
329 if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP
330 && (ctinfo == IP_CT_RELATED
331 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
334 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
335 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
336 * on, which all have an ID field [relevant for hashing]. */
338 hash = clusterip_hashfn(*pskb, cipinfo->config);
345 case IP_CT_RELATED+IP_CT_IS_REPLY:
346 /* FIXME: we don't handle expectations at the
347 * moment. they can arrive on a different node than
348 * the master connection (e.g. FTP passive mode) */
349 case IP_CT_ESTABLISHED:
350 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
356 #ifdef DEBUG_CLUSTERP
357 DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
359 DEBUGP("hash=%u ct_hash=%u ", hash, *mark);
360 if (!clusterip_responsible(cipinfo->config, hash)) {
361 DEBUGP("not responsible\n");
364 DEBUGP("responsible\n");
366 /* despite being received via linklayer multicast, this is
367 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
368 (*pskb)->pkt_type = PACKET_HOST;
374 checkentry(const char *tablename,
376 const struct xt_target *target,
378 unsigned int targinfosize,
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 != 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;
454 /* drop reference count of cluster config when rule is deleted */
455 static void destroy(const struct xt_target *target, void *targinfo,
456 unsigned int targinfosize)
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);
467 static struct ipt_target clusterip_tgt = {
470 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
471 .checkentry = checkentry,
477 /***********************************************************************
479 ***********************************************************************/
481 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
483 u_int8_t src_hw[ETH_ALEN];
485 u_int8_t dst_hw[ETH_ALEN];
487 } __attribute__ ((packed));
489 #ifdef CLUSTERIP_DEBUG
490 static void arp_print(struct arp_payload *payload)
492 #define HBUFFERLEN 30
493 char hbuffer[HBUFFERLEN];
495 const char hexbuf[]= "0123456789abcdef";
497 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
498 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
499 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
504 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
505 NIPQUAD(payload->src_ip), hbuffer,
506 NIPQUAD(payload->dst_ip));
511 arp_mangle(unsigned int hook,
512 struct sk_buff **pskb,
513 const struct net_device *in,
514 const struct net_device *out,
515 int (*okfn)(struct sk_buff *))
517 struct arphdr *arp = (*pskb)->nh.arph;
518 struct arp_payload *payload;
519 struct clusterip_config *c;
521 /* we don't care about non-ethernet and non-ipv4 ARP */
522 if (arp->ar_hrd != htons(ARPHRD_ETHER)
523 || arp->ar_pro != htons(ETH_P_IP)
524 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
527 /* we only want to mangle arp requests and replies */
528 if (arp->ar_op != htons(ARPOP_REPLY)
529 && arp->ar_op != htons(ARPOP_REQUEST))
532 payload = (void *)(arp+1);
534 /* if there is no clusterip configuration for the arp reply's
535 * source ip, we don't want to mangle it */
536 c = clusterip_config_find_get(payload->src_ip, 0);
540 /* normally the linux kernel always replies to arp queries of
541 * addresses on different interfacs. However, in the CLUSTERIP case
542 * this wouldn't work, since we didn't subscribe the mcast group on
543 * other interfaces */
545 DEBUGP("CLUSTERIP: not mangling arp reply on different "
546 "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name);
547 clusterip_config_put(c);
551 /* mangle reply hardware address */
552 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
554 #ifdef CLUSTERIP_DEBUG
555 DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
559 clusterip_config_put(c);
564 static struct nf_hook_ops cip_arp_ops = {
567 .hooknum = NF_ARP_OUT,
571 /***********************************************************************
573 ***********************************************************************/
575 #ifdef CONFIG_PROC_FS
577 struct clusterip_seq_position {
578 unsigned int pos; /* position */
579 unsigned int weight; /* number of bits set == size */
580 unsigned int bit; /* current bit */
581 unsigned long val; /* current value */
584 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
586 struct proc_dir_entry *pde = s->private;
587 struct clusterip_config *c = pde->data;
589 u_int32_t local_nodes;
590 struct clusterip_seq_position *idx;
592 /* FIXME: possible race */
593 local_nodes = c->local_nodes;
594 weight = hweight32(local_nodes);
598 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
600 return ERR_PTR(-ENOMEM);
603 idx->weight = weight;
604 idx->bit = ffs(local_nodes);
605 idx->val = local_nodes;
606 clear_bit(idx->bit - 1, &idx->val);
611 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
613 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
616 if (*pos >= idx->weight) {
620 idx->bit = ffs(idx->val);
621 clear_bit(idx->bit - 1, &idx->val);
625 static void clusterip_seq_stop(struct seq_file *s, void *v)
630 static int clusterip_seq_show(struct seq_file *s, void *v)
632 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
637 seq_printf(s, "%u", idx->bit);
639 if (idx->pos == idx->weight - 1)
645 static struct seq_operations clusterip_seq_ops = {
646 .start = clusterip_seq_start,
647 .next = clusterip_seq_next,
648 .stop = clusterip_seq_stop,
649 .show = clusterip_seq_show,
652 static int clusterip_proc_open(struct inode *inode, struct file *file)
654 int ret = seq_open(file, &clusterip_seq_ops);
657 struct seq_file *sf = file->private_data;
658 struct proc_dir_entry *pde = PDE(inode);
659 struct clusterip_config *c = pde->data;
663 clusterip_config_get(c);
669 static int clusterip_proc_release(struct inode *inode, struct file *file)
671 struct proc_dir_entry *pde = PDE(inode);
672 struct clusterip_config *c = pde->data;
675 ret = seq_release(inode, file);
678 clusterip_config_put(c);
683 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
684 size_t size, loff_t *ofs)
686 #define PROC_WRITELEN 10
687 char buffer[PROC_WRITELEN+1];
688 struct proc_dir_entry *pde = PDE(file->f_dentry->d_inode);
689 struct clusterip_config *c = pde->data;
690 unsigned long nodenum;
692 if (copy_from_user(buffer, input, PROC_WRITELEN))
695 if (*buffer == '+') {
696 nodenum = simple_strtoul(buffer+1, NULL, 10);
697 if (clusterip_add_node(c, nodenum))
699 } else if (*buffer == '-') {
700 nodenum = simple_strtoul(buffer+1, NULL,10);
701 if (clusterip_del_node(c, nodenum))
709 static struct file_operations clusterip_proc_fops = {
710 .owner = THIS_MODULE,
711 .open = clusterip_proc_open,
713 .write = clusterip_proc_write,
715 .release = clusterip_proc_release,
718 #endif /* CONFIG_PROC_FS */
720 static int __init ipt_clusterip_init(void)
724 ret = ipt_register_target(&clusterip_tgt);
728 ret = nf_register_hook(&cip_arp_ops);
732 #ifdef CONFIG_PROC_FS
733 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net);
734 if (!clusterip_procdir) {
735 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
739 #endif /* CONFIG_PROC_FS */
741 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
746 nf_unregister_hook(&cip_arp_ops);
748 ipt_unregister_target(&clusterip_tgt);
752 static void __exit ipt_clusterip_fini(void)
754 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
756 #ifdef CONFIG_PROC_FS
757 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
759 nf_unregister_hook(&cip_arp_ops);
760 ipt_unregister_target(&clusterip_tgt);
763 module_init(ipt_clusterip_init);
764 module_exit(ipt_clusterip_fini);