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 #define ASSERT_READ_LOCK(x)
45 MODULE_LICENSE("GPL");
46 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
47 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
49 struct clusterip_config {
50 struct list_head list; /* list of all configs */
51 atomic_t refcount; /* reference count */
52 atomic_t entries; /* number of entries/rules
55 u_int32_t clusterip; /* the IP address */
56 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
57 struct net_device *dev; /* device */
58 u_int16_t num_total_nodes; /* total number of nodes */
59 unsigned long local_nodes; /* node number array */
62 struct proc_dir_entry *pde; /* proc dir entry */
64 enum clusterip_hashmode hash_mode; /* which hashing mode */
65 u_int32_t hash_initval; /* hash initialization */
68 static LIST_HEAD(clusterip_configs);
70 /* clusterip_lock protects the clusterip_configs list */
71 static DEFINE_RWLOCK(clusterip_lock);
74 static struct file_operations clusterip_proc_fops;
75 static struct proc_dir_entry *clusterip_procdir;
79 clusterip_config_get(struct clusterip_config *c)
81 atomic_inc(&c->refcount);
85 clusterip_config_put(struct clusterip_config *c)
87 if (atomic_dec_and_test(&c->refcount))
91 /* increase the count of entries(rules) using/referencing this config */
93 clusterip_config_entry_get(struct clusterip_config *c)
95 atomic_inc(&c->entries);
98 /* decrease the count of entries using/referencing this config. If last
99 * entry(rule) is removed, remove the config from lists, but don't free it
100 * yet, since proc-files could still be holding references */
102 clusterip_config_entry_put(struct clusterip_config *c)
104 if (atomic_dec_and_test(&c->entries)) {
105 write_lock_bh(&clusterip_lock);
107 write_unlock_bh(&clusterip_lock);
109 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
112 /* In case anyone still accesses the file, the open/close
113 * functions are also incrementing the refcount on their own,
114 * so it's safe to remove the entry even if it's in use. */
115 #ifdef CONFIG_PROC_FS
116 remove_proc_entry(c->pde->name, c->pde->parent);
121 static struct clusterip_config *
122 __clusterip_config_find(u_int32_t clusterip)
124 struct list_head *pos;
126 ASSERT_READ_LOCK(&clusterip_lock);
127 list_for_each(pos, &clusterip_configs) {
128 struct clusterip_config *c = list_entry(pos,
129 struct clusterip_config, list);
130 if (c->clusterip == clusterip) {
138 static inline struct clusterip_config *
139 clusterip_config_find_get(u_int32_t clusterip, int entry)
141 struct clusterip_config *c;
143 read_lock_bh(&clusterip_lock);
144 c = __clusterip_config_find(clusterip);
146 read_unlock_bh(&clusterip_lock);
149 atomic_inc(&c->refcount);
151 atomic_inc(&c->entries);
152 read_unlock_bh(&clusterip_lock);
158 clusterip_config_init_nodelist(struct clusterip_config *c,
159 const struct ipt_clusterip_tgt_info *i)
163 for (n = 0; n < i->num_local_nodes; n++) {
164 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
168 static struct clusterip_config *
169 clusterip_config_init(struct ipt_clusterip_tgt_info *i, u_int32_t ip,
170 struct net_device *dev)
172 struct clusterip_config *c;
175 c = kzalloc(sizeof(*c), GFP_ATOMIC);
181 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
182 c->num_total_nodes = i->num_total_nodes;
183 clusterip_config_init_nodelist(c, i);
184 c->hash_mode = i->hash_mode;
185 c->hash_initval = i->hash_initval;
186 atomic_set(&c->refcount, 1);
187 atomic_set(&c->entries, 1);
189 #ifdef CONFIG_PROC_FS
190 /* create proc dir entry */
191 sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
192 c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR, clusterip_procdir);
197 c->pde->proc_fops = &clusterip_proc_fops;
201 write_lock_bh(&clusterip_lock);
202 list_add(&c->list, &clusterip_configs);
203 write_unlock_bh(&clusterip_lock);
209 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
213 nodenum > c->num_total_nodes)
216 /* check if we already have this number in our bitfield */
217 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
224 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
227 nodenum > c->num_total_nodes)
230 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
236 static inline u_int32_t
237 clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config)
239 struct iphdr *iph = skb->nh.iph;
240 unsigned long hashval;
241 u_int16_t sport, dport;
244 switch (iph->protocol) {
250 ports = (void *)iph+iph->ihl*4;
255 if (net_ratelimit()) {
256 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(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,
308 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
309 enum ip_conntrack_info ctinfo;
310 u_int32_t *mark, hash;
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 mark = nf_ct_get_mark((*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 ((*pskb)->nh.iph->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, *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 targinfosize,
377 unsigned int hook_mask)
379 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
380 const struct ipt_entry *e = e_void;
382 struct clusterip_config *config;
384 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
385 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
386 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
387 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
392 if (e->ip.dmsk.s_addr != 0xffffffff
393 || e->ip.dst.s_addr == 0) {
394 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
398 /* FIXME: further sanity checks */
400 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
402 if (cipinfo->config != NULL) {
403 /* Case A: This is an entry that gets reloaded, since
404 * it still has a cipinfo->config pointer. Simply
405 * increase the entry refcount and return */
406 if (cipinfo->config != config) {
407 printk(KERN_ERR "CLUSTERIP: Reloaded entry "
408 "has invalid config pointer!\n");
411 clusterip_config_entry_get(cipinfo->config);
413 /* Case B: This is a new rule referring to an existing
414 * clusterip config. */
415 cipinfo->config = config;
416 clusterip_config_entry_get(cipinfo->config);
419 /* Case C: This is a completely new clusterip config */
420 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
421 printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
424 struct net_device *dev;
426 if (e->ip.iniface[0] == '\0') {
427 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
431 dev = dev_get_by_name(e->ip.iniface);
433 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
437 config = clusterip_config_init(cipinfo,
438 e->ip.dst.s_addr, dev);
440 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
444 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
446 cipinfo->config = config;
452 /* drop reference count of cluster config when rule is deleted */
453 static void destroy(const struct xt_target *target, void *targinfo,
454 unsigned int targinfosize)
456 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
458 /* if no more entries are referencing the config, remove it
459 * from the list and destroy the proc entry */
460 clusterip_config_entry_put(cipinfo->config);
462 clusterip_config_put(cipinfo->config);
465 static struct ipt_target clusterip_tgt = {
468 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
469 .checkentry = checkentry,
475 /***********************************************************************
477 ***********************************************************************/
479 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
481 u_int8_t src_hw[ETH_ALEN];
483 u_int8_t dst_hw[ETH_ALEN];
485 } __attribute__ ((packed));
487 #ifdef CLUSTERIP_DEBUG
488 static void arp_print(struct arp_payload *payload)
490 #define HBUFFERLEN 30
491 char hbuffer[HBUFFERLEN];
493 const char hexbuf[]= "0123456789abcdef";
495 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
496 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
497 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
502 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
503 NIPQUAD(payload->src_ip), hbuffer,
504 NIPQUAD(payload->dst_ip));
509 arp_mangle(unsigned int hook,
510 struct sk_buff **pskb,
511 const struct net_device *in,
512 const struct net_device *out,
513 int (*okfn)(struct sk_buff *))
515 struct arphdr *arp = (*pskb)->nh.arph;
516 struct arp_payload *payload;
517 struct clusterip_config *c;
519 /* we don't care about non-ethernet and non-ipv4 ARP */
520 if (arp->ar_hrd != htons(ARPHRD_ETHER)
521 || arp->ar_pro != htons(ETH_P_IP)
522 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
525 /* we only want to mangle arp requests and replies */
526 if (arp->ar_op != htons(ARPOP_REPLY)
527 && arp->ar_op != htons(ARPOP_REQUEST))
530 payload = (void *)(arp+1);
532 /* if there is no clusterip configuration for the arp reply's
533 * source ip, we don't want to mangle it */
534 c = clusterip_config_find_get(payload->src_ip, 0);
538 /* normally the linux kernel always replies to arp queries of
539 * addresses on different interfacs. However, in the CLUSTERIP case
540 * this wouldn't work, since we didn't subscribe the mcast group on
541 * other interfaces */
543 DEBUGP("CLUSTERIP: not mangling arp reply on different "
544 "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name);
545 clusterip_config_put(c);
549 /* mangle reply hardware address */
550 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
552 #ifdef CLUSTERIP_DEBUG
553 DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
557 clusterip_config_put(c);
562 static struct nf_hook_ops cip_arp_ops = {
565 .hooknum = NF_ARP_OUT,
569 /***********************************************************************
571 ***********************************************************************/
573 #ifdef CONFIG_PROC_FS
575 struct clusterip_seq_position {
576 unsigned int pos; /* position */
577 unsigned int weight; /* number of bits set == size */
578 unsigned int bit; /* current bit */
579 unsigned long val; /* current value */
582 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
584 struct proc_dir_entry *pde = s->private;
585 struct clusterip_config *c = pde->data;
587 u_int32_t local_nodes;
588 struct clusterip_seq_position *idx;
590 /* FIXME: possible race */
591 local_nodes = c->local_nodes;
592 weight = hweight32(local_nodes);
596 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
598 return ERR_PTR(-ENOMEM);
601 idx->weight = weight;
602 idx->bit = ffs(local_nodes);
603 idx->val = local_nodes;
604 clear_bit(idx->bit - 1, &idx->val);
609 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
611 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
614 if (*pos >= idx->weight) {
618 idx->bit = ffs(idx->val);
619 clear_bit(idx->bit - 1, &idx->val);
623 static void clusterip_seq_stop(struct seq_file *s, void *v)
628 static int clusterip_seq_show(struct seq_file *s, void *v)
630 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
635 seq_printf(s, "%u", idx->bit);
637 if (idx->pos == idx->weight - 1)
643 static struct seq_operations clusterip_seq_ops = {
644 .start = clusterip_seq_start,
645 .next = clusterip_seq_next,
646 .stop = clusterip_seq_stop,
647 .show = clusterip_seq_show,
650 static int clusterip_proc_open(struct inode *inode, struct file *file)
652 int ret = seq_open(file, &clusterip_seq_ops);
655 struct seq_file *sf = file->private_data;
656 struct proc_dir_entry *pde = PDE(inode);
657 struct clusterip_config *c = pde->data;
661 clusterip_config_get(c);
667 static int clusterip_proc_release(struct inode *inode, struct file *file)
669 struct proc_dir_entry *pde = PDE(inode);
670 struct clusterip_config *c = pde->data;
673 ret = seq_release(inode, file);
676 clusterip_config_put(c);
681 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
682 size_t size, loff_t *ofs)
684 #define PROC_WRITELEN 10
685 char buffer[PROC_WRITELEN+1];
686 struct proc_dir_entry *pde = PDE(file->f_dentry->d_inode);
687 struct clusterip_config *c = pde->data;
688 unsigned long nodenum;
690 if (copy_from_user(buffer, input, PROC_WRITELEN))
693 if (*buffer == '+') {
694 nodenum = simple_strtoul(buffer+1, NULL, 10);
695 if (clusterip_add_node(c, nodenum))
697 } else if (*buffer == '-') {
698 nodenum = simple_strtoul(buffer+1, NULL,10);
699 if (clusterip_del_node(c, nodenum))
707 static struct file_operations clusterip_proc_fops = {
708 .owner = THIS_MODULE,
709 .open = clusterip_proc_open,
711 .write = clusterip_proc_write,
713 .release = clusterip_proc_release,
716 #endif /* CONFIG_PROC_FS */
718 static int __init ipt_clusterip_init(void)
722 ret = ipt_register_target(&clusterip_tgt);
726 ret = nf_register_hook(&cip_arp_ops);
730 #ifdef CONFIG_PROC_FS
731 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net);
732 if (!clusterip_procdir) {
733 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
737 #endif /* CONFIG_PROC_FS */
739 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
744 nf_unregister_hook(&cip_arp_ops);
746 ipt_unregister_target(&clusterip_tgt);
750 static void __exit ipt_clusterip_fini(void)
752 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
754 #ifdef CONFIG_PROC_FS
755 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
757 nf_unregister_hook(&cip_arp_ops);
758 ipt_unregister_target(&clusterip_tgt);
761 module_init(ipt_clusterip_init);
762 module_exit(ipt_clusterip_fini);