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 /* decrease the count of entries using/referencing this config. If last
80 * entry(rule) is removed, remove the config from lists, but don't free it
81 * yet, since proc-files could still be holding references */
83 clusterip_config_entry_put(struct clusterip_config *c)
85 write_lock_bh(&clusterip_lock);
86 if (atomic_dec_and_test(&c->entries)) {
88 write_unlock_bh(&clusterip_lock);
90 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
93 /* In case anyone still accesses the file, the open/close
94 * functions are also incrementing the refcount on their own,
95 * so it's safe to remove the entry even if it's in use. */
97 remove_proc_entry(c->pde->name, c->pde->parent);
101 write_unlock_bh(&clusterip_lock);
104 static struct clusterip_config *
105 __clusterip_config_find(__be32 clusterip)
107 struct clusterip_config *c;
109 list_for_each_entry(c, &clusterip_configs, list) {
110 if (c->clusterip == clusterip)
117 static inline struct clusterip_config *
118 clusterip_config_find_get(__be32 clusterip, int entry)
120 struct clusterip_config *c;
122 read_lock_bh(&clusterip_lock);
123 c = __clusterip_config_find(clusterip);
125 read_unlock_bh(&clusterip_lock);
128 atomic_inc(&c->refcount);
130 atomic_inc(&c->entries);
131 read_unlock_bh(&clusterip_lock);
137 clusterip_config_init_nodelist(struct clusterip_config *c,
138 const struct ipt_clusterip_tgt_info *i)
142 for (n = 0; n < i->num_local_nodes; n++)
143 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
146 static struct clusterip_config *
147 clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip,
148 struct net_device *dev)
150 struct clusterip_config *c;
152 c = kzalloc(sizeof(*c), GFP_ATOMIC);
158 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
159 c->num_total_nodes = i->num_total_nodes;
160 clusterip_config_init_nodelist(c, i);
161 c->hash_mode = i->hash_mode;
162 c->hash_initval = i->hash_initval;
163 atomic_set(&c->refcount, 1);
164 atomic_set(&c->entries, 1);
166 #ifdef CONFIG_PROC_FS
170 /* create proc dir entry */
171 sprintf(buffer, "%pI4", &ip);
172 c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR,
174 &clusterip_proc_fops, c);
182 write_lock_bh(&clusterip_lock);
183 list_add(&c->list, &clusterip_configs);
184 write_unlock_bh(&clusterip_lock);
189 #ifdef CONFIG_PROC_FS
191 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
195 nodenum > c->num_total_nodes)
198 /* check if we already have this number in our bitfield */
199 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
206 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
209 nodenum > c->num_total_nodes)
212 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
219 static inline u_int32_t
220 clusterip_hashfn(const struct sk_buff *skb,
221 const struct clusterip_config *config)
223 const struct iphdr *iph = ip_hdr(skb);
224 unsigned long hashval;
225 u_int16_t sport, dport;
226 const u_int16_t *ports;
228 switch (iph->protocol) {
231 case IPPROTO_UDPLITE:
235 ports = (const void *)iph+iph->ihl*4;
241 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
246 switch (config->hash_mode) {
247 case CLUSTERIP_HASHMODE_SIP:
248 hashval = jhash_1word(ntohl(iph->saddr),
249 config->hash_initval);
251 case CLUSTERIP_HASHMODE_SIP_SPT:
252 hashval = jhash_2words(ntohl(iph->saddr), sport,
253 config->hash_initval);
255 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
256 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
257 config->hash_initval);
260 /* to make gcc happy */
262 /* This cannot happen, unless the check function wasn't called
263 * at rule load time */
264 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
269 /* node numbers are 1..n, not 0..n */
270 return (((u64)hashval * config->num_total_nodes) >> 32) + 1;
274 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
276 return test_bit(hash - 1, &config->local_nodes);
279 /***********************************************************************
281 ***********************************************************************/
284 clusterip_tg(struct sk_buff *skb, const struct xt_target_param *par)
286 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
288 enum ip_conntrack_info ctinfo;
291 /* don't need to clusterip_config_get() here, since refcount
292 * is only decremented by destroy() - and ip_tables guarantees
293 * that the ->target() function isn't called after ->destroy() */
295 ct = nf_ct_get(skb, &ctinfo);
297 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
298 /* FIXME: need to drop invalid ones, since replies
299 * to outgoing connections of other nodes will be
300 * marked as INVALID */
304 /* special case: ICMP error handling. conntrack distinguishes between
305 * error messages (RELATED) and information requests (see below) */
306 if (ip_hdr(skb)->protocol == IPPROTO_ICMP
307 && (ctinfo == IP_CT_RELATED
308 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
311 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
312 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
313 * on, which all have an ID field [relevant for hashing]. */
315 hash = clusterip_hashfn(skb, cipinfo->config);
322 case IP_CT_RELATED+IP_CT_IS_REPLY:
323 /* FIXME: we don't handle expectations at the
324 * moment. they can arrive on a different node than
325 * the master connection (e.g. FTP passive mode) */
326 case IP_CT_ESTABLISHED:
327 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
334 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
336 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
337 if (!clusterip_responsible(cipinfo->config, hash)) {
338 pr_debug("not responsible\n");
341 pr_debug("responsible\n");
343 /* despite being received via linklayer multicast, this is
344 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
345 skb->pkt_type = PACKET_HOST;
350 static bool clusterip_tg_check(const struct xt_tgchk_param *par)
352 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
353 const struct ipt_entry *e = par->entryinfo;
355 struct clusterip_config *config;
357 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
358 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
359 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
360 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
365 if (e->ip.dmsk.s_addr != htonl(0xffffffff)
366 || e->ip.dst.s_addr == 0) {
367 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
371 /* FIXME: further sanity checks */
373 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
375 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
376 printk(KERN_WARNING "CLUSTERIP: no config found for %pI4, need 'new'\n", &e->ip.dst.s_addr);
379 struct net_device *dev;
381 if (e->ip.iniface[0] == '\0') {
382 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
386 dev = dev_get_by_name(&init_net, e->ip.iniface);
388 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
392 config = clusterip_config_init(cipinfo,
393 e->ip.dst.s_addr, dev);
395 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
399 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
402 cipinfo->config = config;
404 if (nf_ct_l3proto_try_module_get(par->target->family) < 0) {
405 printk(KERN_WARNING "can't load conntrack support for "
406 "proto=%u\n", par->target->family);
413 /* drop reference count of cluster config when rule is deleted */
414 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
416 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
418 /* if no more entries are referencing the config, remove it
419 * from the list and destroy the proc entry */
420 clusterip_config_entry_put(cipinfo->config);
422 clusterip_config_put(cipinfo->config);
424 nf_ct_l3proto_module_put(par->target->family);
428 struct compat_ipt_clusterip_tgt_info
431 u_int8_t clustermac[6];
432 u_int16_t num_total_nodes;
433 u_int16_t num_local_nodes;
434 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
436 u_int32_t hash_initval;
437 compat_uptr_t config;
439 #endif /* CONFIG_COMPAT */
441 static struct xt_target clusterip_tg_reg __read_mostly = {
443 .family = NFPROTO_IPV4,
444 .target = clusterip_tg,
445 .checkentry = clusterip_tg_check,
446 .destroy = clusterip_tg_destroy,
447 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
449 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
450 #endif /* CONFIG_COMPAT */
455 /***********************************************************************
457 ***********************************************************************/
459 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
461 u_int8_t src_hw[ETH_ALEN];
463 u_int8_t dst_hw[ETH_ALEN];
465 } __attribute__ ((packed));
468 static void arp_print(struct arp_payload *payload)
470 #define HBUFFERLEN 30
471 char hbuffer[HBUFFERLEN];
474 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
475 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
476 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
481 printk("src %pI4@%s, dst %pI4\n",
482 &payload->src_ip, hbuffer, &payload->dst_ip);
487 arp_mangle(unsigned int hook,
489 const struct net_device *in,
490 const struct net_device *out,
491 int (*okfn)(struct sk_buff *))
493 struct arphdr *arp = arp_hdr(skb);
494 struct arp_payload *payload;
495 struct clusterip_config *c;
497 /* we don't care about non-ethernet and non-ipv4 ARP */
498 if (arp->ar_hrd != htons(ARPHRD_ETHER)
499 || arp->ar_pro != htons(ETH_P_IP)
500 || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
503 /* we only want to mangle arp requests and replies */
504 if (arp->ar_op != htons(ARPOP_REPLY)
505 && arp->ar_op != htons(ARPOP_REQUEST))
508 payload = (void *)(arp+1);
510 /* if there is no clusterip configuration for the arp reply's
511 * source ip, we don't want to mangle it */
512 c = clusterip_config_find_get(payload->src_ip, 0);
516 /* normally the linux kernel always replies to arp queries of
517 * addresses on different interfacs. However, in the CLUSTERIP case
518 * this wouldn't work, since we didn't subscribe the mcast group on
519 * other interfaces */
521 pr_debug("CLUSTERIP: not mangling arp reply on different "
522 "interface: cip'%s'-skb'%s'\n",
523 c->dev->name, out->name);
524 clusterip_config_put(c);
528 /* mangle reply hardware address */
529 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
532 pr_debug(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
536 clusterip_config_put(c);
541 static struct nf_hook_ops cip_arp_ops __read_mostly = {
544 .hooknum = NF_ARP_OUT,
548 /***********************************************************************
550 ***********************************************************************/
552 #ifdef CONFIG_PROC_FS
554 struct clusterip_seq_position {
555 unsigned int pos; /* position */
556 unsigned int weight; /* number of bits set == size */
557 unsigned int bit; /* current bit */
558 unsigned long val; /* current value */
561 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
563 const struct proc_dir_entry *pde = s->private;
564 struct clusterip_config *c = pde->data;
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 proc_dir_entry *pde = PDE(inode);
636 struct clusterip_config *c = pde->data;
640 clusterip_config_get(c);
646 static int clusterip_proc_release(struct inode *inode, struct file *file)
648 struct proc_dir_entry *pde = PDE(inode);
649 struct clusterip_config *c = pde->data;
652 ret = seq_release(inode, file);
655 clusterip_config_put(c);
660 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
661 size_t size, loff_t *ofs)
663 #define PROC_WRITELEN 10
664 char buffer[PROC_WRITELEN+1];
665 const struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
666 struct clusterip_config *c = pde->data;
667 unsigned long nodenum;
669 if (copy_from_user(buffer, input, PROC_WRITELEN))
672 if (*buffer == '+') {
673 nodenum = simple_strtoul(buffer+1, NULL, 10);
674 if (clusterip_add_node(c, nodenum))
676 } else if (*buffer == '-') {
677 nodenum = simple_strtoul(buffer+1, NULL,10);
678 if (clusterip_del_node(c, nodenum))
686 static const struct file_operations clusterip_proc_fops = {
687 .owner = THIS_MODULE,
688 .open = clusterip_proc_open,
690 .write = clusterip_proc_write,
692 .release = clusterip_proc_release,
695 #endif /* CONFIG_PROC_FS */
697 static int __init clusterip_tg_init(void)
701 ret = xt_register_target(&clusterip_tg_reg);
705 ret = nf_register_hook(&cip_arp_ops);
709 #ifdef CONFIG_PROC_FS
710 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", init_net.proc_net);
711 if (!clusterip_procdir) {
712 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
716 #endif /* CONFIG_PROC_FS */
718 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
722 #ifdef CONFIG_PROC_FS
724 nf_unregister_hook(&cip_arp_ops);
725 #endif /* CONFIG_PROC_FS */
727 xt_unregister_target(&clusterip_tg_reg);
731 static void __exit clusterip_tg_exit(void)
733 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
735 #ifdef CONFIG_PROC_FS
736 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
738 nf_unregister_hook(&cip_arp_ops);
739 xt_unregister_target(&clusterip_tg_reg);
742 module_init(clusterip_tg_init);
743 module_exit(clusterip_tg_exit);