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/seq_file.h>
23 #include <linux/netfilter_arp.h>
24 #include <linux/netfilter/x_tables.h>
25 #include <linux/netfilter_ipv4/ip_tables.h>
26 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
27 #include <net/netfilter/nf_conntrack.h>
28 #include <net/net_namespace.h>
29 #include <net/checksum.h>
31 #define CLUSTERIP_VERSION "0.8"
33 MODULE_LICENSE("GPL");
34 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
35 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
37 struct clusterip_config {
38 struct list_head list; /* list of all configs */
39 atomic_t refcount; /* reference count */
40 atomic_t entries; /* number of entries/rules
43 __be32 clusterip; /* the IP address */
44 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
45 struct net_device *dev; /* device */
46 u_int16_t num_total_nodes; /* total number of nodes */
47 unsigned long local_nodes; /* node number array */
50 struct proc_dir_entry *pde; /* proc dir entry */
52 enum clusterip_hashmode hash_mode; /* which hashing mode */
53 u_int32_t hash_initval; /* hash initialization */
56 static LIST_HEAD(clusterip_configs);
58 /* clusterip_lock protects the clusterip_configs list */
59 static DEFINE_RWLOCK(clusterip_lock);
62 static const struct file_operations clusterip_proc_fops;
63 static struct proc_dir_entry *clusterip_procdir;
67 clusterip_config_get(struct clusterip_config *c)
69 atomic_inc(&c->refcount);
73 clusterip_config_put(struct clusterip_config *c)
75 if (atomic_dec_and_test(&c->refcount))
79 /* increase the count of entries(rules) using/referencing this config */
81 clusterip_config_entry_get(struct clusterip_config *c)
83 atomic_inc(&c->entries);
86 /* decrease the count of entries using/referencing this config. If last
87 * entry(rule) is removed, remove the config from lists, but don't free it
88 * yet, since proc-files could still be holding references */
90 clusterip_config_entry_put(struct clusterip_config *c)
92 if (atomic_dec_and_test(&c->entries)) {
93 write_lock_bh(&clusterip_lock);
95 write_unlock_bh(&clusterip_lock);
97 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
100 /* In case anyone still accesses the file, the open/close
101 * functions are also incrementing the refcount on their own,
102 * so it's safe to remove the entry even if it's in use. */
103 #ifdef CONFIG_PROC_FS
104 remove_proc_entry(c->pde->name, c->pde->parent);
109 static struct clusterip_config *
110 __clusterip_config_find(__be32 clusterip)
112 struct clusterip_config *c;
114 list_for_each_entry(c, &clusterip_configs, list) {
115 if (c->clusterip == clusterip)
122 static inline struct clusterip_config *
123 clusterip_config_find_get(__be32 clusterip, int entry)
125 struct clusterip_config *c;
127 read_lock_bh(&clusterip_lock);
128 c = __clusterip_config_find(clusterip);
130 read_unlock_bh(&clusterip_lock);
133 atomic_inc(&c->refcount);
135 atomic_inc(&c->entries);
136 read_unlock_bh(&clusterip_lock);
142 clusterip_config_init_nodelist(struct clusterip_config *c,
143 const struct ipt_clusterip_tgt_info *i)
147 for (n = 0; n < i->num_local_nodes; n++)
148 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
151 static struct clusterip_config *
152 clusterip_config_init(struct ipt_clusterip_tgt_info *i, __be32 ip,
153 struct net_device *dev)
155 struct clusterip_config *c;
157 c = kzalloc(sizeof(*c), GFP_ATOMIC);
163 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
164 c->num_total_nodes = i->num_total_nodes;
165 clusterip_config_init_nodelist(c, i);
166 c->hash_mode = i->hash_mode;
167 c->hash_initval = i->hash_initval;
168 atomic_set(&c->refcount, 1);
169 atomic_set(&c->entries, 1);
171 #ifdef CONFIG_PROC_FS
175 /* create proc dir entry */
176 sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
177 c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR,
184 c->pde->proc_fops = &clusterip_proc_fops;
188 write_lock_bh(&clusterip_lock);
189 list_add(&c->list, &clusterip_configs);
190 write_unlock_bh(&clusterip_lock);
195 #ifdef CONFIG_PROC_FS
197 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
201 nodenum > c->num_total_nodes)
204 /* check if we already have this number in our bitfield */
205 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
212 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
215 nodenum > c->num_total_nodes)
218 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
225 static inline u_int32_t
226 clusterip_hashfn(const struct sk_buff *skb,
227 const struct clusterip_config *config)
229 const struct iphdr *iph = ip_hdr(skb);
230 unsigned long hashval;
231 u_int16_t sport, dport;
232 const u_int16_t *ports;
234 switch (iph->protocol) {
237 case IPPROTO_UDPLITE:
241 ports = (const void *)iph+iph->ihl*4;
247 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
252 switch (config->hash_mode) {
253 case CLUSTERIP_HASHMODE_SIP:
254 hashval = jhash_1word(ntohl(iph->saddr),
255 config->hash_initval);
257 case CLUSTERIP_HASHMODE_SIP_SPT:
258 hashval = jhash_2words(ntohl(iph->saddr), sport,
259 config->hash_initval);
261 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
262 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
263 config->hash_initval);
266 /* to make gcc happy */
268 /* This cannot happen, unless the check function wasn't called
269 * at rule load time */
270 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
275 /* node numbers are 1..n, not 0..n */
276 return (((u64)hashval * config->num_total_nodes) >> 32) + 1;
280 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
282 return test_bit(hash - 1, &config->local_nodes);
285 /***********************************************************************
287 ***********************************************************************/
290 clusterip_tg(struct sk_buff *skb, const struct net_device *in,
291 const struct net_device *out, unsigned int hooknum,
292 const struct xt_target *target, const void *targinfo)
294 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
296 enum ip_conntrack_info ctinfo;
299 /* don't need to clusterip_config_get() here, since refcount
300 * is only decremented by destroy() - and ip_tables guarantees
301 * that the ->target() function isn't called after ->destroy() */
303 ct = nf_ct_get(skb, &ctinfo);
305 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
306 /* FIXME: need to drop invalid ones, since replies
307 * to outgoing connections of other nodes will be
308 * marked as INVALID */
312 /* special case: ICMP error handling. conntrack distinguishes between
313 * error messages (RELATED) and information requests (see below) */
314 if (ip_hdr(skb)->protocol == IPPROTO_ICMP
315 && (ctinfo == IP_CT_RELATED
316 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
319 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
320 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
321 * on, which all have an ID field [relevant for hashing]. */
323 hash = clusterip_hashfn(skb, cipinfo->config);
330 case IP_CT_RELATED+IP_CT_IS_REPLY:
331 /* FIXME: we don't handle expectations at the
332 * moment. they can arrive on a different node than
333 * the master connection (e.g. FTP passive mode) */
334 case IP_CT_ESTABLISHED:
335 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
342 DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
344 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
345 if (!clusterip_responsible(cipinfo->config, hash)) {
346 pr_debug("not responsible\n");
349 pr_debug("responsible\n");
351 /* despite being received via linklayer multicast, this is
352 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
353 skb->pkt_type = PACKET_HOST;
359 clusterip_tg_check(const char *tablename, const void *e_void,
360 const struct xt_target *target, void *targinfo,
361 unsigned int hook_mask)
363 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
364 const struct ipt_entry *e = e_void;
366 struct clusterip_config *config;
368 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
369 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
370 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
371 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
376 if (e->ip.dmsk.s_addr != htonl(0xffffffff)
377 || e->ip.dst.s_addr == 0) {
378 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
382 /* FIXME: further sanity checks */
384 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
386 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
387 printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
390 struct net_device *dev;
392 if (e->ip.iniface[0] == '\0') {
393 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
397 dev = dev_get_by_name(&init_net, e->ip.iniface);
399 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
403 config = clusterip_config_init(cipinfo,
404 e->ip.dst.s_addr, dev);
406 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
410 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
413 cipinfo->config = config;
415 if (nf_ct_l3proto_try_module_get(target->family) < 0) {
416 printk(KERN_WARNING "can't load conntrack support for "
417 "proto=%u\n", target->family);
424 /* drop reference count of cluster config when rule is deleted */
425 static void clusterip_tg_destroy(const struct xt_target *target, void *targinfo)
427 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
429 /* if no more entries are referencing the config, remove it
430 * from the list and destroy the proc entry */
431 clusterip_config_entry_put(cipinfo->config);
433 clusterip_config_put(cipinfo->config);
435 nf_ct_l3proto_module_put(target->family);
439 struct compat_ipt_clusterip_tgt_info
442 u_int8_t clustermac[6];
443 u_int16_t num_total_nodes;
444 u_int16_t num_local_nodes;
445 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
447 u_int32_t hash_initval;
448 compat_uptr_t config;
450 #endif /* CONFIG_COMPAT */
452 static struct xt_target clusterip_tg_reg __read_mostly = {
455 .target = clusterip_tg,
456 .checkentry = clusterip_tg_check,
457 .destroy = clusterip_tg_destroy,
458 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
460 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
461 #endif /* CONFIG_COMPAT */
466 /***********************************************************************
468 ***********************************************************************/
470 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
472 u_int8_t src_hw[ETH_ALEN];
474 u_int8_t dst_hw[ETH_ALEN];
476 } __attribute__ ((packed));
479 static void arp_print(struct arp_payload *payload)
481 #define HBUFFERLEN 30
482 char hbuffer[HBUFFERLEN];
484 const char hexbuf[]= "0123456789abcdef";
486 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
487 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
488 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
493 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
494 NIPQUAD(payload->src_ip), hbuffer,
495 NIPQUAD(payload->dst_ip));
500 arp_mangle(unsigned int hook,
502 const struct net_device *in,
503 const struct net_device *out,
504 int (*okfn)(struct sk_buff *))
506 struct arphdr *arp = arp_hdr(skb);
507 struct arp_payload *payload;
508 struct clusterip_config *c;
510 /* we don't care about non-ethernet and non-ipv4 ARP */
511 if (arp->ar_hrd != htons(ARPHRD_ETHER)
512 || arp->ar_pro != htons(ETH_P_IP)
513 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
516 /* we only want to mangle arp requests and replies */
517 if (arp->ar_op != htons(ARPOP_REPLY)
518 && arp->ar_op != htons(ARPOP_REQUEST))
521 payload = (void *)(arp+1);
523 /* if there is no clusterip configuration for the arp reply's
524 * source ip, we don't want to mangle it */
525 c = clusterip_config_find_get(payload->src_ip, 0);
529 /* normally the linux kernel always replies to arp queries of
530 * addresses on different interfacs. However, in the CLUSTERIP case
531 * this wouldn't work, since we didn't subscribe the mcast group on
532 * other interfaces */
534 pr_debug("CLUSTERIP: not mangling arp reply on different "
535 "interface: cip'%s'-skb'%s'\n",
536 c->dev->name, out->name);
537 clusterip_config_put(c);
541 /* mangle reply hardware address */
542 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
545 pr_debug(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
549 clusterip_config_put(c);
554 static struct nf_hook_ops cip_arp_ops __read_mostly = {
557 .hooknum = NF_ARP_OUT,
561 /***********************************************************************
563 ***********************************************************************/
565 #ifdef CONFIG_PROC_FS
567 struct clusterip_seq_position {
568 unsigned int pos; /* position */
569 unsigned int weight; /* number of bits set == size */
570 unsigned int bit; /* current bit */
571 unsigned long val; /* current value */
574 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
576 struct proc_dir_entry *pde = s->private;
577 struct clusterip_config *c = pde->data;
579 u_int32_t local_nodes;
580 struct clusterip_seq_position *idx;
582 /* FIXME: possible race */
583 local_nodes = c->local_nodes;
584 weight = hweight32(local_nodes);
588 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
590 return ERR_PTR(-ENOMEM);
593 idx->weight = weight;
594 idx->bit = ffs(local_nodes);
595 idx->val = local_nodes;
596 clear_bit(idx->bit - 1, &idx->val);
601 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
603 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
606 if (*pos >= idx->weight) {
610 idx->bit = ffs(idx->val);
611 clear_bit(idx->bit - 1, &idx->val);
615 static void clusterip_seq_stop(struct seq_file *s, void *v)
620 static int clusterip_seq_show(struct seq_file *s, void *v)
622 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
627 seq_printf(s, "%u", idx->bit);
629 if (idx->pos == idx->weight - 1)
635 static const struct seq_operations clusterip_seq_ops = {
636 .start = clusterip_seq_start,
637 .next = clusterip_seq_next,
638 .stop = clusterip_seq_stop,
639 .show = clusterip_seq_show,
642 static int clusterip_proc_open(struct inode *inode, struct file *file)
644 int ret = seq_open(file, &clusterip_seq_ops);
647 struct seq_file *sf = file->private_data;
648 struct proc_dir_entry *pde = PDE(inode);
649 struct clusterip_config *c = pde->data;
653 clusterip_config_get(c);
659 static int clusterip_proc_release(struct inode *inode, struct file *file)
661 struct proc_dir_entry *pde = PDE(inode);
662 struct clusterip_config *c = pde->data;
665 ret = seq_release(inode, file);
668 clusterip_config_put(c);
673 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
674 size_t size, loff_t *ofs)
676 #define PROC_WRITELEN 10
677 char buffer[PROC_WRITELEN+1];
678 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
679 struct clusterip_config *c = pde->data;
680 unsigned long nodenum;
682 if (copy_from_user(buffer, input, PROC_WRITELEN))
685 if (*buffer == '+') {
686 nodenum = simple_strtoul(buffer+1, NULL, 10);
687 if (clusterip_add_node(c, nodenum))
689 } else if (*buffer == '-') {
690 nodenum = simple_strtoul(buffer+1, NULL,10);
691 if (clusterip_del_node(c, nodenum))
699 static const struct file_operations clusterip_proc_fops = {
700 .owner = THIS_MODULE,
701 .open = clusterip_proc_open,
703 .write = clusterip_proc_write,
705 .release = clusterip_proc_release,
708 #endif /* CONFIG_PROC_FS */
710 static int __init clusterip_tg_init(void)
714 ret = xt_register_target(&clusterip_tg_reg);
718 ret = nf_register_hook(&cip_arp_ops);
722 #ifdef CONFIG_PROC_FS
723 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", init_net.proc_net);
724 if (!clusterip_procdir) {
725 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
729 #endif /* CONFIG_PROC_FS */
731 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
735 #ifdef CONFIG_PROC_FS
737 nf_unregister_hook(&cip_arp_ops);
738 #endif /* CONFIG_PROC_FS */
740 xt_unregister_target(&clusterip_tg_reg);
744 static void __exit clusterip_tg_exit(void)
746 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
748 #ifdef CONFIG_PROC_FS
749 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
751 nf_unregister_hook(&cip_arp_ops);
752 xt_unregister_target(&clusterip_tg_reg);
755 module_init(clusterip_tg_init);
756 module_exit(clusterip_tg_exit);