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>
17 #include <linux/slab.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/seq_file.h>
24 #include <linux/netfilter_arp.h>
25 #include <linux/netfilter/x_tables.h>
26 #include <linux/netfilter_ipv4/ip_tables.h>
27 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
28 #include <net/netfilter/nf_conntrack.h>
29 #include <net/net_namespace.h>
30 #include <net/checksum.h>
32 #define CLUSTERIP_VERSION "0.8"
34 MODULE_LICENSE("GPL");
35 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
36 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
38 struct clusterip_config {
39 struct list_head list; /* list of all configs */
40 atomic_t refcount; /* reference count */
41 atomic_t entries; /* number of entries/rules
44 __be32 clusterip; /* the IP address */
45 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
46 struct net_device *dev; /* device */
47 u_int16_t num_total_nodes; /* total number of nodes */
48 unsigned long local_nodes; /* node number array */
51 struct proc_dir_entry *pde; /* proc dir entry */
53 enum clusterip_hashmode hash_mode; /* which hashing mode */
54 u_int32_t hash_initval; /* hash initialization */
57 static LIST_HEAD(clusterip_configs);
59 /* clusterip_lock protects the clusterip_configs list */
60 static DEFINE_RWLOCK(clusterip_lock);
63 static const struct file_operations clusterip_proc_fops;
64 static struct proc_dir_entry *clusterip_procdir;
68 clusterip_config_get(struct clusterip_config *c)
70 atomic_inc(&c->refcount);
74 clusterip_config_put(struct clusterip_config *c)
76 if (atomic_dec_and_test(&c->refcount))
80 /* decrease the count of entries using/referencing this config. If last
81 * entry(rule) is removed, remove the config from lists, but don't free it
82 * yet, since proc-files could still be holding references */
84 clusterip_config_entry_put(struct clusterip_config *c)
86 write_lock_bh(&clusterip_lock);
87 if (atomic_dec_and_test(&c->entries)) {
89 write_unlock_bh(&clusterip_lock);
91 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
94 /* In case anyone still accesses the file, the open/close
95 * functions are also incrementing the refcount on their own,
96 * so it's safe to remove the entry even if it's in use. */
98 remove_proc_entry(c->pde->name, c->pde->parent);
102 write_unlock_bh(&clusterip_lock);
105 static struct clusterip_config *
106 __clusterip_config_find(__be32 clusterip)
108 struct clusterip_config *c;
110 list_for_each_entry(c, &clusterip_configs, list) {
111 if (c->clusterip == clusterip)
118 static inline struct clusterip_config *
119 clusterip_config_find_get(__be32 clusterip, int entry)
121 struct clusterip_config *c;
123 read_lock_bh(&clusterip_lock);
124 c = __clusterip_config_find(clusterip);
126 read_unlock_bh(&clusterip_lock);
129 atomic_inc(&c->refcount);
131 atomic_inc(&c->entries);
132 read_unlock_bh(&clusterip_lock);
138 clusterip_config_init_nodelist(struct clusterip_config *c,
139 const struct ipt_clusterip_tgt_info *i)
143 for (n = 0; n < i->num_local_nodes; n++)
144 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
147 static struct clusterip_config *
148 clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip,
149 struct net_device *dev)
151 struct clusterip_config *c;
153 c = kzalloc(sizeof(*c), GFP_ATOMIC);
159 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
160 c->num_total_nodes = i->num_total_nodes;
161 clusterip_config_init_nodelist(c, i);
162 c->hash_mode = i->hash_mode;
163 c->hash_initval = i->hash_initval;
164 atomic_set(&c->refcount, 1);
165 atomic_set(&c->entries, 1);
167 #ifdef CONFIG_PROC_FS
171 /* create proc dir entry */
172 sprintf(buffer, "%pI4", &ip);
173 c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR,
175 &clusterip_proc_fops, c);
183 write_lock_bh(&clusterip_lock);
184 list_add(&c->list, &clusterip_configs);
185 write_unlock_bh(&clusterip_lock);
190 #ifdef CONFIG_PROC_FS
192 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
196 nodenum > c->num_total_nodes)
199 /* check if we already have this number in our bitfield */
200 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
207 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
210 nodenum > c->num_total_nodes)
213 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
220 static inline u_int32_t
221 clusterip_hashfn(const struct sk_buff *skb,
222 const struct clusterip_config *config)
224 const struct iphdr *iph = ip_hdr(skb);
225 unsigned long hashval;
226 u_int16_t sport, dport;
227 const u_int16_t *ports;
229 switch (iph->protocol) {
232 case IPPROTO_UDPLITE:
236 ports = (const void *)iph+iph->ihl*4;
242 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
247 switch (config->hash_mode) {
248 case CLUSTERIP_HASHMODE_SIP:
249 hashval = jhash_1word(ntohl(iph->saddr),
250 config->hash_initval);
252 case CLUSTERIP_HASHMODE_SIP_SPT:
253 hashval = jhash_2words(ntohl(iph->saddr), sport,
254 config->hash_initval);
256 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
257 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
258 config->hash_initval);
261 /* to make gcc happy */
263 /* This cannot happen, unless the check function wasn't called
264 * at rule load time */
265 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
270 /* node numbers are 1..n, not 0..n */
271 return (((u64)hashval * config->num_total_nodes) >> 32) + 1;
275 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
277 return test_bit(hash - 1, &config->local_nodes);
280 /***********************************************************************
282 ***********************************************************************/
285 clusterip_tg(struct sk_buff *skb, const struct xt_target_param *par)
287 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
289 enum ip_conntrack_info ctinfo;
292 /* don't need to clusterip_config_get() here, since refcount
293 * is only decremented by destroy() - and ip_tables guarantees
294 * that the ->target() function isn't called after ->destroy() */
296 ct = nf_ct_get(skb, &ctinfo);
298 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
299 /* FIXME: need to drop invalid ones, since replies
300 * to outgoing connections of other nodes will be
301 * marked as INVALID */
305 /* special case: ICMP error handling. conntrack distinguishes between
306 * error messages (RELATED) and information requests (see below) */
307 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
308 (ctinfo == IP_CT_RELATED ||
309 ctinfo == IP_CT_RELATED + IP_CT_IS_REPLY))
312 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
313 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
314 * on, which all have an ID field [relevant for hashing]. */
316 hash = clusterip_hashfn(skb, cipinfo->config);
323 case IP_CT_RELATED+IP_CT_IS_REPLY:
324 /* FIXME: we don't handle expectations at the
325 * moment. they can arrive on a different node than
326 * the master connection (e.g. FTP passive mode) */
327 case IP_CT_ESTABLISHED:
328 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
335 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
337 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
338 if (!clusterip_responsible(cipinfo->config, hash)) {
339 pr_debug("not responsible\n");
342 pr_debug("responsible\n");
344 /* despite being received via linklayer multicast, this is
345 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
346 skb->pkt_type = PACKET_HOST;
351 static bool clusterip_tg_check(const struct xt_tgchk_param *par)
353 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
354 const struct ipt_entry *e = par->entryinfo;
356 struct clusterip_config *config;
358 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
359 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
360 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
361 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
366 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
367 e->ip.dst.s_addr == 0) {
368 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
372 /* FIXME: further sanity checks */
374 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
376 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
377 printk(KERN_WARNING "CLUSTERIP: no config found for %pI4, need 'new'\n", &e->ip.dst.s_addr);
380 struct net_device *dev;
382 if (e->ip.iniface[0] == '\0') {
383 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
387 dev = dev_get_by_name(&init_net, e->ip.iniface);
389 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
393 config = clusterip_config_init(cipinfo,
394 e->ip.dst.s_addr, dev);
396 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
400 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
403 cipinfo->config = config;
405 if (nf_ct_l3proto_try_module_get(par->target->family) < 0) {
406 printk(KERN_WARNING "can't load conntrack support for "
407 "proto=%u\n", par->target->family);
414 /* drop reference count of cluster config when rule is deleted */
415 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
417 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
419 /* if no more entries are referencing the config, remove it
420 * from the list and destroy the proc entry */
421 clusterip_config_entry_put(cipinfo->config);
423 clusterip_config_put(cipinfo->config);
425 nf_ct_l3proto_module_put(par->target->family);
429 struct compat_ipt_clusterip_tgt_info
432 u_int8_t clustermac[6];
433 u_int16_t num_total_nodes;
434 u_int16_t num_local_nodes;
435 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
437 u_int32_t hash_initval;
438 compat_uptr_t config;
440 #endif /* CONFIG_COMPAT */
442 static struct xt_target clusterip_tg_reg __read_mostly = {
444 .family = NFPROTO_IPV4,
445 .target = clusterip_tg,
446 .checkentry = clusterip_tg_check,
447 .destroy = clusterip_tg_destroy,
448 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
450 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
451 #endif /* CONFIG_COMPAT */
456 /***********************************************************************
458 ***********************************************************************/
460 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
462 u_int8_t src_hw[ETH_ALEN];
464 u_int8_t dst_hw[ETH_ALEN];
466 } __attribute__ ((packed));
469 static void arp_print(struct arp_payload *payload)
471 #define HBUFFERLEN 30
472 char hbuffer[HBUFFERLEN];
475 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
476 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
477 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
482 printk("src %pI4@%s, dst %pI4\n",
483 &payload->src_ip, hbuffer, &payload->dst_ip);
488 arp_mangle(unsigned int hook,
490 const struct net_device *in,
491 const struct net_device *out,
492 int (*okfn)(struct sk_buff *))
494 struct arphdr *arp = arp_hdr(skb);
495 struct arp_payload *payload;
496 struct clusterip_config *c;
498 /* we don't care about non-ethernet and non-ipv4 ARP */
499 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
500 arp->ar_pro != htons(ETH_P_IP) ||
501 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
504 /* we only want to mangle arp requests and replies */
505 if (arp->ar_op != htons(ARPOP_REPLY) &&
506 arp->ar_op != htons(ARPOP_REQUEST))
509 payload = (void *)(arp+1);
511 /* if there is no clusterip configuration for the arp reply's
512 * source ip, we don't want to mangle it */
513 c = clusterip_config_find_get(payload->src_ip, 0);
517 /* normally the linux kernel always replies to arp queries of
518 * addresses on different interfacs. However, in the CLUSTERIP case
519 * this wouldn't work, since we didn't subscribe the mcast group on
520 * other interfaces */
522 pr_debug("CLUSTERIP: not mangling arp reply on different "
523 "interface: cip'%s'-skb'%s'\n",
524 c->dev->name, out->name);
525 clusterip_config_put(c);
529 /* mangle reply hardware address */
530 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
533 pr_debug(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
537 clusterip_config_put(c);
542 static struct nf_hook_ops cip_arp_ops __read_mostly = {
545 .hooknum = NF_ARP_OUT,
549 /***********************************************************************
551 ***********************************************************************/
553 #ifdef CONFIG_PROC_FS
555 struct clusterip_seq_position {
556 unsigned int pos; /* position */
557 unsigned int weight; /* number of bits set == size */
558 unsigned int bit; /* current bit */
559 unsigned long val; /* current value */
562 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
564 struct clusterip_config *c = s->private;
566 u_int32_t local_nodes;
567 struct clusterip_seq_position *idx;
569 /* FIXME: possible race */
570 local_nodes = c->local_nodes;
571 weight = hweight32(local_nodes);
575 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
577 return ERR_PTR(-ENOMEM);
580 idx->weight = weight;
581 idx->bit = ffs(local_nodes);
582 idx->val = local_nodes;
583 clear_bit(idx->bit - 1, &idx->val);
588 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
590 struct clusterip_seq_position *idx = v;
593 if (*pos >= idx->weight) {
597 idx->bit = ffs(idx->val);
598 clear_bit(idx->bit - 1, &idx->val);
602 static void clusterip_seq_stop(struct seq_file *s, void *v)
607 static int clusterip_seq_show(struct seq_file *s, void *v)
609 struct clusterip_seq_position *idx = v;
614 seq_printf(s, "%u", idx->bit);
616 if (idx->pos == idx->weight - 1)
622 static const struct seq_operations clusterip_seq_ops = {
623 .start = clusterip_seq_start,
624 .next = clusterip_seq_next,
625 .stop = clusterip_seq_stop,
626 .show = clusterip_seq_show,
629 static int clusterip_proc_open(struct inode *inode, struct file *file)
631 int ret = seq_open(file, &clusterip_seq_ops);
634 struct seq_file *sf = file->private_data;
635 struct clusterip_config *c = PDE(inode)->data;
639 clusterip_config_get(c);
645 static int clusterip_proc_release(struct inode *inode, struct file *file)
647 struct clusterip_config *c = PDE(inode)->data;
650 ret = seq_release(inode, file);
653 clusterip_config_put(c);
658 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
659 size_t size, loff_t *ofs)
661 struct clusterip_config *c = PDE(file->f_path.dentry->d_inode)->data;
662 #define PROC_WRITELEN 10
663 char buffer[PROC_WRITELEN+1];
664 unsigned long nodenum;
666 if (copy_from_user(buffer, input, PROC_WRITELEN))
669 if (*buffer == '+') {
670 nodenum = simple_strtoul(buffer+1, NULL, 10);
671 if (clusterip_add_node(c, nodenum))
673 } else if (*buffer == '-') {
674 nodenum = simple_strtoul(buffer+1, NULL,10);
675 if (clusterip_del_node(c, nodenum))
683 static const struct file_operations clusterip_proc_fops = {
684 .owner = THIS_MODULE,
685 .open = clusterip_proc_open,
687 .write = clusterip_proc_write,
689 .release = clusterip_proc_release,
692 #endif /* CONFIG_PROC_FS */
694 static int __init clusterip_tg_init(void)
698 ret = xt_register_target(&clusterip_tg_reg);
702 ret = nf_register_hook(&cip_arp_ops);
706 #ifdef CONFIG_PROC_FS
707 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", init_net.proc_net);
708 if (!clusterip_procdir) {
709 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
713 #endif /* CONFIG_PROC_FS */
715 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
719 #ifdef CONFIG_PROC_FS
721 nf_unregister_hook(&cip_arp_ops);
722 #endif /* CONFIG_PROC_FS */
724 xt_unregister_target(&clusterip_tg_reg);
728 static void __exit clusterip_tg_exit(void)
730 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
732 #ifdef CONFIG_PROC_FS
733 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
735 nf_unregister_hook(&cip_arp_ops);
736 xt_unregister_target(&clusterip_tg_reg);
739 module_init(clusterip_tg_init);
740 module_exit(clusterip_tg_exit);