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 <linux/netfilter_ipv4/ip_conntrack.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;
248 switch (iph->protocol) {
250 th = (void *)iph+iph->ihl*4;
251 sport = ntohs(th->source);
252 dport = ntohs(th->dest);
255 uh = (void *)iph+iph->ihl*4;
256 sport = ntohs(uh->source);
257 dport = ntohs(uh->dest);
260 ih = (void *)iph+iph->ihl*4;
261 sport = ntohs(ih->un.echo.id);
262 dport = (ih->type<<8)|ih->code;
265 if (net_ratelimit()) {
266 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
272 switch (config->hash_mode) {
273 case CLUSTERIP_HASHMODE_SIP:
274 hashval = jhash_1word(ntohl(iph->saddr),
275 config->hash_initval);
277 case CLUSTERIP_HASHMODE_SIP_SPT:
278 hashval = jhash_2words(ntohl(iph->saddr), sport,
279 config->hash_initval);
281 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
282 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
283 config->hash_initval);
286 /* to make gcc happy */
288 /* This cannot happen, unless the check function wasn't called
289 * at rule load time */
290 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
295 /* node numbers are 1..n, not 0..n */
296 return ((hashval % config->num_total_nodes)+1);
300 clusterip_responsible(struct clusterip_config *config, u_int32_t hash)
302 return test_bit(hash - 1, &config->local_nodes);
305 /***********************************************************************
307 ***********************************************************************/
310 target(struct sk_buff **pskb,
311 const struct net_device *in,
312 const struct net_device *out,
313 unsigned int hooknum,
314 const void *targinfo,
317 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
318 enum ip_conntrack_info ctinfo;
319 struct ip_conntrack *ct = ip_conntrack_get((*pskb), &ctinfo);
322 /* don't need to clusterip_config_get() here, since refcount
323 * is only decremented by destroy() - and ip_tables guarantees
324 * that the ->target() function isn't called after ->destroy() */
327 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
328 /* FIXME: need to drop invalid ones, since replies
329 * to outgoing connections of other nodes will be
330 * marked as INVALID */
334 /* special case: ICMP error handling. conntrack distinguishes between
335 * error messages (RELATED) and information requests (see below) */
336 if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP
337 && (ctinfo == IP_CT_RELATED
338 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
341 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
342 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
343 * on, which all have an ID field [relevant for hashing]. */
345 hash = clusterip_hashfn(*pskb, cipinfo->config);
352 case IP_CT_RELATED+IP_CT_IS_REPLY:
353 /* FIXME: we don't handle expectations at the
354 * moment. they can arrive on a different node than
355 * the master connection (e.g. FTP passive mode) */
356 case IP_CT_ESTABLISHED:
357 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
363 #ifdef DEBUG_CLUSTERP
364 DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
366 DEBUGP("hash=%u ct_hash=%u ", hash, ct->mark);
367 if (!clusterip_responsible(cipinfo->config, hash)) {
368 DEBUGP("not responsible\n");
371 DEBUGP("responsible\n");
373 /* despite being received via linklayer multicast, this is
374 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
375 (*pskb)->pkt_type = PACKET_HOST;
381 checkentry(const char *tablename,
382 const struct ipt_entry *e,
384 unsigned int targinfosize,
385 unsigned int hook_mask)
387 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
389 struct clusterip_config *config;
391 if (targinfosize != IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info))) {
392 printk(KERN_WARNING "CLUSTERIP: targinfosize %u != %Zu\n",
394 IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info)));
398 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
399 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
400 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
401 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
406 if (e->ip.dmsk.s_addr != 0xffffffff
407 || e->ip.dst.s_addr == 0) {
408 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
412 /* FIXME: further sanity checks */
414 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
416 if (cipinfo->config != NULL) {
417 /* Case A: This is an entry that gets reloaded, since
418 * it still has a cipinfo->config pointer. Simply
419 * increase the entry refcount and return */
420 if (cipinfo->config != config) {
421 printk(KERN_ERR "CLUSTERIP: Reloaded entry "
422 "has invalid config pointer!\n");
425 clusterip_config_entry_get(cipinfo->config);
427 /* Case B: This is a new rule referring to an existing
428 * clusterip config. */
429 cipinfo->config = config;
430 clusterip_config_entry_get(cipinfo->config);
433 /* Case C: This is a completely new clusterip config */
434 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
435 printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
438 struct net_device *dev;
440 if (e->ip.iniface[0] == '\0') {
441 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
445 dev = dev_get_by_name(e->ip.iniface);
447 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
451 config = clusterip_config_init(cipinfo,
452 e->ip.dst.s_addr, dev);
454 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
458 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
460 cipinfo->config = config;
466 /* drop reference count of cluster config when rule is deleted */
467 static void destroy(void *matchinfo, unsigned int matchinfosize)
469 struct ipt_clusterip_tgt_info *cipinfo = matchinfo;
471 /* if no more entries are referencing the config, remove it
472 * from the list and destroy the proc entry */
473 clusterip_config_entry_put(cipinfo->config);
475 clusterip_config_put(cipinfo->config);
478 static struct ipt_target clusterip_tgt = {
481 .checkentry = &checkentry,
487 /***********************************************************************
489 ***********************************************************************/
491 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
493 u_int8_t src_hw[ETH_ALEN];
495 u_int8_t dst_hw[ETH_ALEN];
497 } __attribute__ ((packed));
499 #ifdef CLUSTERIP_DEBUG
500 static void arp_print(struct arp_payload *payload)
502 #define HBUFFERLEN 30
503 char hbuffer[HBUFFERLEN];
505 const char hexbuf[]= "0123456789abcdef";
507 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
508 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
509 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
514 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
515 NIPQUAD(payload->src_ip), hbuffer,
516 NIPQUAD(payload->dst_ip));
521 arp_mangle(unsigned int hook,
522 struct sk_buff **pskb,
523 const struct net_device *in,
524 const struct net_device *out,
525 int (*okfn)(struct sk_buff *))
527 struct arphdr *arp = (*pskb)->nh.arph;
528 struct arp_payload *payload;
529 struct clusterip_config *c;
531 /* we don't care about non-ethernet and non-ipv4 ARP */
532 if (arp->ar_hrd != htons(ARPHRD_ETHER)
533 || arp->ar_pro != htons(ETH_P_IP)
534 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
537 /* we only want to mangle arp requests and replies */
538 if (arp->ar_op != htons(ARPOP_REPLY)
539 && arp->ar_op != htons(ARPOP_REQUEST))
542 payload = (void *)(arp+1);
544 /* if there is no clusterip configuration for the arp reply's
545 * source ip, we don't want to mangle it */
546 c = clusterip_config_find_get(payload->src_ip, 0);
550 /* normally the linux kernel always replies to arp queries of
551 * addresses on different interfacs. However, in the CLUSTERIP case
552 * this wouldn't work, since we didn't subscribe the mcast group on
553 * other interfaces */
555 DEBUGP("CLUSTERIP: not mangling arp reply on different "
556 "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name);
557 clusterip_config_put(c);
561 /* mangle reply hardware address */
562 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
564 #ifdef CLUSTERIP_DEBUG
565 DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
569 clusterip_config_put(c);
574 static struct nf_hook_ops cip_arp_ops = {
577 .hooknum = NF_ARP_OUT,
581 /***********************************************************************
583 ***********************************************************************/
585 #ifdef CONFIG_PROC_FS
587 struct clusterip_seq_position {
588 unsigned int pos; /* position */
589 unsigned int weight; /* number of bits set == size */
590 unsigned int bit; /* current bit */
591 unsigned long val; /* current value */
594 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
596 struct proc_dir_entry *pde = s->private;
597 struct clusterip_config *c = pde->data;
599 u_int32_t local_nodes;
600 struct clusterip_seq_position *idx;
602 /* FIXME: possible race */
603 local_nodes = c->local_nodes;
604 weight = hweight32(local_nodes);
608 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
610 return ERR_PTR(-ENOMEM);
613 idx->weight = weight;
614 idx->bit = ffs(local_nodes);
615 idx->val = local_nodes;
616 clear_bit(idx->bit - 1, &idx->val);
621 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
623 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
626 if (*pos >= idx->weight) {
630 idx->bit = ffs(idx->val);
631 clear_bit(idx->bit - 1, &idx->val);
635 static void clusterip_seq_stop(struct seq_file *s, void *v)
640 static int clusterip_seq_show(struct seq_file *s, void *v)
642 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
647 seq_printf(s, "%u", idx->bit);
649 if (idx->pos == idx->weight - 1)
655 static struct seq_operations clusterip_seq_ops = {
656 .start = clusterip_seq_start,
657 .next = clusterip_seq_next,
658 .stop = clusterip_seq_stop,
659 .show = clusterip_seq_show,
662 static int clusterip_proc_open(struct inode *inode, struct file *file)
664 int ret = seq_open(file, &clusterip_seq_ops);
667 struct seq_file *sf = file->private_data;
668 struct proc_dir_entry *pde = PDE(inode);
669 struct clusterip_config *c = pde->data;
673 clusterip_config_get(c);
679 static int clusterip_proc_release(struct inode *inode, struct file *file)
681 struct proc_dir_entry *pde = PDE(inode);
682 struct clusterip_config *c = pde->data;
685 ret = seq_release(inode, file);
688 clusterip_config_put(c);
693 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
694 size_t size, loff_t *ofs)
696 #define PROC_WRITELEN 10
697 char buffer[PROC_WRITELEN+1];
698 struct proc_dir_entry *pde = PDE(file->f_dentry->d_inode);
699 struct clusterip_config *c = pde->data;
700 unsigned long nodenum;
702 if (copy_from_user(buffer, input, PROC_WRITELEN))
705 if (*buffer == '+') {
706 nodenum = simple_strtoul(buffer+1, NULL, 10);
707 if (clusterip_add_node(c, nodenum))
709 } else if (*buffer == '-') {
710 nodenum = simple_strtoul(buffer+1, NULL,10);
711 if (clusterip_del_node(c, nodenum))
719 static struct file_operations clusterip_proc_fops = {
720 .owner = THIS_MODULE,
721 .open = clusterip_proc_open,
723 .write = clusterip_proc_write,
725 .release = clusterip_proc_release,
728 #endif /* CONFIG_PROC_FS */
730 static int init_or_cleanup(int fini)
737 if (ipt_register_target(&clusterip_tgt)) {
742 if (nf_register_hook(&cip_arp_ops) < 0) {
747 #ifdef CONFIG_PROC_FS
748 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net);
749 if (!clusterip_procdir) {
750 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
754 #endif /* CONFIG_PROC_FS */
756 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
762 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
764 #ifdef CONFIG_PROC_FS
765 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
768 nf_unregister_hook(&cip_arp_ops);
770 ipt_unregister_target(&clusterip_tgt);
775 static int __init init(void)
777 return init_or_cleanup(0);
780 static void __exit fini(void)