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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/module.h>
14 #include <linux/proc_fs.h>
15 #include <linux/jhash.h>
16 #include <linux/bitops.h>
17 #include <linux/skbuff.h>
18 #include <linux/slab.h>
20 #include <linux/tcp.h>
21 #include <linux/udp.h>
22 #include <linux/icmp.h>
23 #include <linux/if_arp.h>
24 #include <linux/seq_file.h>
25 #include <linux/netfilter_arp.h>
26 #include <linux/netfilter/x_tables.h>
27 #include <linux/netfilter_ipv4/ip_tables.h>
28 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/net_namespace.h>
31 #include <net/checksum.h>
34 #define CLUSTERIP_VERSION "0.8"
36 MODULE_LICENSE("GPL");
37 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
38 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
40 struct clusterip_config {
41 struct list_head list; /* list of all configs */
42 atomic_t refcount; /* reference count */
43 atomic_t entries; /* number of entries/rules
46 __be32 clusterip; /* the IP address */
47 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
48 struct net_device *dev; /* device */
49 u_int16_t num_total_nodes; /* total number of nodes */
50 unsigned long local_nodes; /* node number array */
53 struct proc_dir_entry *pde; /* proc dir entry */
55 enum clusterip_hashmode hash_mode; /* which hashing mode */
56 u_int32_t hash_initval; /* hash initialization */
60 static LIST_HEAD(clusterip_configs);
62 /* clusterip_lock protects the clusterip_configs list */
63 static DEFINE_SPINLOCK(clusterip_lock);
66 static const struct file_operations clusterip_proc_fops;
67 static struct proc_dir_entry *clusterip_procdir;
71 clusterip_config_get(struct clusterip_config *c)
73 atomic_inc(&c->refcount);
77 static void clusterip_config_rcu_free(struct rcu_head *head)
79 kfree(container_of(head, struct clusterip_config, rcu));
83 clusterip_config_put(struct clusterip_config *c)
85 if (atomic_dec_and_test(&c->refcount))
86 call_rcu_bh(&c->rcu, clusterip_config_rcu_free);
89 /* decrease the count of entries using/referencing this config. If last
90 * entry(rule) is removed, remove the config from lists, but don't free it
91 * yet, since proc-files could still be holding references */
93 clusterip_config_entry_put(struct clusterip_config *c)
96 if (atomic_dec_and_lock(&c->entries, &clusterip_lock)) {
97 list_del_rcu(&c->list);
98 spin_unlock(&clusterip_lock);
101 dev_mc_del(c->dev, c->clustermac);
104 /* In case anyone still accesses the file, the open/close
105 * functions are also incrementing the refcount on their own,
106 * so it's safe to remove the entry even if it's in use. */
107 #ifdef CONFIG_PROC_FS
108 remove_proc_entry(c->pde->name, c->pde->parent);
115 static struct clusterip_config *
116 __clusterip_config_find(__be32 clusterip)
118 struct clusterip_config *c;
120 list_for_each_entry_rcu(c, &clusterip_configs, list) {
121 if (c->clusterip == clusterip)
128 static inline struct clusterip_config *
129 clusterip_config_find_get(__be32 clusterip, int entry)
131 struct clusterip_config *c;
134 c = __clusterip_config_find(clusterip);
136 if (unlikely(!atomic_inc_not_zero(&c->refcount)))
139 if (unlikely(!atomic_inc_not_zero(&c->entries))) {
140 clusterip_config_put(c);
145 rcu_read_unlock_bh();
151 clusterip_config_init_nodelist(struct clusterip_config *c,
152 const struct ipt_clusterip_tgt_info *i)
156 for (n = 0; n < i->num_local_nodes; n++)
157 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
160 static struct clusterip_config *
161 clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip,
162 struct net_device *dev)
164 struct clusterip_config *c;
166 c = kzalloc(sizeof(*c), GFP_ATOMIC);
172 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
173 c->num_total_nodes = i->num_total_nodes;
174 clusterip_config_init_nodelist(c, i);
175 c->hash_mode = i->hash_mode;
176 c->hash_initval = i->hash_initval;
177 atomic_set(&c->refcount, 1);
178 atomic_set(&c->entries, 1);
180 #ifdef CONFIG_PROC_FS
184 /* create proc dir entry */
185 sprintf(buffer, "%pI4", &ip);
186 c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR,
188 &clusterip_proc_fops, c);
196 spin_lock_bh(&clusterip_lock);
197 list_add_rcu(&c->list, &clusterip_configs);
198 spin_unlock_bh(&clusterip_lock);
203 #ifdef CONFIG_PROC_FS
205 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
209 nodenum > c->num_total_nodes)
212 /* check if we already have this number in our bitfield */
213 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
220 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
223 nodenum > c->num_total_nodes)
226 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
233 static inline u_int32_t
234 clusterip_hashfn(const struct sk_buff *skb,
235 const struct clusterip_config *config)
237 const struct iphdr *iph = ip_hdr(skb);
238 unsigned long hashval;
239 u_int16_t sport = 0, dport = 0;
242 poff = proto_ports_offset(iph->protocol);
244 const u_int16_t *ports;
247 ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports);
254 pr_info("unknown protocol %u\n", iph->protocol);
257 switch (config->hash_mode) {
258 case CLUSTERIP_HASHMODE_SIP:
259 hashval = jhash_1word(ntohl(iph->saddr),
260 config->hash_initval);
262 case CLUSTERIP_HASHMODE_SIP_SPT:
263 hashval = jhash_2words(ntohl(iph->saddr), sport,
264 config->hash_initval);
266 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
267 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
268 config->hash_initval);
271 /* to make gcc happy */
273 /* This cannot happen, unless the check function wasn't called
274 * at rule load time */
275 pr_info("unknown mode %u\n", config->hash_mode);
280 /* node numbers are 1..n, not 0..n */
281 return (((u64)hashval * config->num_total_nodes) >> 32) + 1;
285 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
287 return test_bit(hash - 1, &config->local_nodes);
290 /***********************************************************************
292 ***********************************************************************/
295 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
297 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
299 enum ip_conntrack_info ctinfo;
302 /* don't need to clusterip_config_get() here, since refcount
303 * is only decremented by destroy() - and ip_tables guarantees
304 * that the ->target() function isn't called after ->destroy() */
306 ct = nf_ct_get(skb, &ctinfo);
310 /* special case: ICMP error handling. conntrack distinguishes between
311 * error messages (RELATED) and information requests (see below) */
312 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
313 (ctinfo == IP_CT_RELATED ||
314 ctinfo == IP_CT_RELATED_REPLY))
317 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
318 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
319 * on, which all have an ID field [relevant for hashing]. */
321 hash = clusterip_hashfn(skb, cipinfo->config);
328 case IP_CT_RELATED_REPLY:
329 /* FIXME: we don't handle expectations at the moment.
330 * They can arrive on a different node than
331 * the master connection (e.g. FTP passive mode) */
332 case IP_CT_ESTABLISHED:
333 case IP_CT_ESTABLISHED_REPLY:
335 default: /* Prevent gcc warnings */
340 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
342 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
343 if (!clusterip_responsible(cipinfo->config, hash)) {
344 pr_debug("not responsible\n");
347 pr_debug("responsible\n");
349 /* despite being received via linklayer multicast, this is
350 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
351 skb->pkt_type = PACKET_HOST;
356 static int clusterip_tg_check(const struct xt_tgchk_param *par)
358 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
359 const struct ipt_entry *e = par->entryinfo;
360 struct clusterip_config *config;
363 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
364 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
365 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
366 pr_info("unknown mode %u\n", cipinfo->hash_mode);
370 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
371 e->ip.dst.s_addr == 0) {
372 pr_info("Please specify destination IP\n");
375 if (cipinfo->num_local_nodes > ARRAY_SIZE(cipinfo->local_nodes)) {
376 pr_info("bad num_local_nodes %u\n", cipinfo->num_local_nodes);
379 for (i = 0; i < cipinfo->num_local_nodes; i++) {
380 if (cipinfo->local_nodes[i] - 1 >=
381 sizeof(config->local_nodes) * 8) {
382 pr_info("bad local_nodes[%d] %u\n",
383 i, cipinfo->local_nodes[i]);
388 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
390 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
391 pr_info("no config found for %pI4, need 'new'\n",
395 struct net_device *dev;
397 if (e->ip.iniface[0] == '\0') {
398 pr_info("Please specify an interface name\n");
402 dev = dev_get_by_name(&init_net, e->ip.iniface);
404 pr_info("no such interface %s\n",
409 config = clusterip_config_init(cipinfo,
410 e->ip.dst.s_addr, dev);
415 dev_mc_add(config->dev, config->clustermac);
418 cipinfo->config = config;
420 ret = nf_ct_l3proto_try_module_get(par->family);
422 pr_info("cannot load conntrack support for proto=%u\n",
427 /* drop reference count of cluster config when rule is deleted */
428 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
430 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
432 /* if no more entries are referencing the config, remove it
433 * from the list and destroy the proc entry */
434 clusterip_config_entry_put(cipinfo->config);
436 clusterip_config_put(cipinfo->config);
438 nf_ct_l3proto_module_put(par->family);
442 struct compat_ipt_clusterip_tgt_info
445 u_int8_t clustermac[6];
446 u_int16_t num_total_nodes;
447 u_int16_t num_local_nodes;
448 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
450 u_int32_t hash_initval;
451 compat_uptr_t config;
453 #endif /* CONFIG_COMPAT */
455 static struct xt_target clusterip_tg_reg __read_mostly = {
457 .family = NFPROTO_IPV4,
458 .target = clusterip_tg,
459 .checkentry = clusterip_tg_check,
460 .destroy = clusterip_tg_destroy,
461 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
463 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
464 #endif /* CONFIG_COMPAT */
469 /***********************************************************************
471 ***********************************************************************/
473 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
475 u_int8_t src_hw[ETH_ALEN];
477 u_int8_t dst_hw[ETH_ALEN];
482 static void arp_print(struct arp_payload *payload)
484 #define HBUFFERLEN 30
485 char hbuffer[HBUFFERLEN];
488 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
489 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
490 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
495 pr_debug("src %pI4@%s, dst %pI4\n",
496 &payload->src_ip, hbuffer, &payload->dst_ip);
501 arp_mangle(unsigned int hook,
503 const struct net_device *in,
504 const struct net_device *out,
505 int (*okfn)(struct sk_buff *))
507 struct arphdr *arp = arp_hdr(skb);
508 struct arp_payload *payload;
509 struct clusterip_config *c;
511 /* we don't care about non-ethernet and non-ipv4 ARP */
512 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
513 arp->ar_pro != htons(ETH_P_IP) ||
514 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
517 /* we only want to mangle arp requests and replies */
518 if (arp->ar_op != htons(ARPOP_REPLY) &&
519 arp->ar_op != htons(ARPOP_REQUEST))
522 payload = (void *)(arp+1);
524 /* if there is no clusterip configuration for the arp reply's
525 * source ip, we don't want to mangle it */
526 c = clusterip_config_find_get(payload->src_ip, 0);
530 /* normally the linux kernel always replies to arp queries of
531 * addresses on different interfacs. However, in the CLUSTERIP case
532 * this wouldn't work, since we didn't subscribe the mcast group on
533 * other interfaces */
535 pr_debug("not mangling arp reply on different "
536 "interface: cip'%s'-skb'%s'\n",
537 c->dev->name, out->name);
538 clusterip_config_put(c);
542 /* mangle reply hardware address */
543 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
546 pr_debug("mangled arp reply: ");
550 clusterip_config_put(c);
555 static struct nf_hook_ops cip_arp_ops __read_mostly = {
558 .hooknum = NF_ARP_OUT,
562 /***********************************************************************
564 ***********************************************************************/
566 #ifdef CONFIG_PROC_FS
568 struct clusterip_seq_position {
569 unsigned int pos; /* position */
570 unsigned int weight; /* number of bits set == size */
571 unsigned int bit; /* current bit */
572 unsigned long val; /* current value */
575 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
577 struct clusterip_config *c = s->private;
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 = 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)
621 static int clusterip_seq_show(struct seq_file *s, void *v)
623 struct clusterip_seq_position *idx = v;
628 seq_printf(s, "%u", idx->bit);
630 if (idx->pos == idx->weight - 1)
636 static const struct seq_operations clusterip_seq_ops = {
637 .start = clusterip_seq_start,
638 .next = clusterip_seq_next,
639 .stop = clusterip_seq_stop,
640 .show = clusterip_seq_show,
643 static int clusterip_proc_open(struct inode *inode, struct file *file)
645 int ret = seq_open(file, &clusterip_seq_ops);
648 struct seq_file *sf = file->private_data;
649 struct clusterip_config *c = PDE(inode)->data;
653 clusterip_config_get(c);
659 static int clusterip_proc_release(struct inode *inode, struct file *file)
661 struct clusterip_config *c = PDE(inode)->data;
664 ret = seq_release(inode, file);
667 clusterip_config_put(c);
672 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
673 size_t size, loff_t *ofs)
675 struct clusterip_config *c = PDE(file->f_path.dentry->d_inode)->data;
676 #define PROC_WRITELEN 10
677 char buffer[PROC_WRITELEN+1];
678 unsigned long nodenum;
680 if (size > PROC_WRITELEN)
682 if (copy_from_user(buffer, input, size))
686 if (*buffer == '+') {
687 nodenum = simple_strtoul(buffer+1, NULL, 10);
688 if (clusterip_add_node(c, nodenum))
690 } else if (*buffer == '-') {
691 nodenum = simple_strtoul(buffer+1, NULL,10);
692 if (clusterip_del_node(c, nodenum))
700 static const struct file_operations clusterip_proc_fops = {
701 .owner = THIS_MODULE,
702 .open = clusterip_proc_open,
704 .write = clusterip_proc_write,
706 .release = clusterip_proc_release,
709 #endif /* CONFIG_PROC_FS */
711 static int __init clusterip_tg_init(void)
715 ret = xt_register_target(&clusterip_tg_reg);
719 ret = nf_register_hook(&cip_arp_ops);
723 #ifdef CONFIG_PROC_FS
724 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", init_net.proc_net);
725 if (!clusterip_procdir) {
726 pr_err("Unable to proc dir entry\n");
730 #endif /* CONFIG_PROC_FS */
732 pr_info("ClusterIP Version %s loaded successfully\n",
736 #ifdef CONFIG_PROC_FS
738 nf_unregister_hook(&cip_arp_ops);
739 #endif /* CONFIG_PROC_FS */
741 xt_unregister_target(&clusterip_tg_reg);
745 static void __exit clusterip_tg_exit(void)
747 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
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);
754 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
758 module_init(clusterip_tg_init);
759 module_exit(clusterip_tg_exit);