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/refcount.h>
26 #include <linux/netfilter_arp.h>
27 #include <linux/netfilter/x_tables.h>
28 #include <linux/netfilter_ipv4/ip_tables.h>
29 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
30 #include <net/netfilter/nf_conntrack.h>
31 #include <net/net_namespace.h>
32 #include <net/netns/generic.h>
33 #include <net/checksum.h>
36 #define CLUSTERIP_VERSION "0.8"
38 MODULE_LICENSE("GPL");
39 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
40 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
42 struct clusterip_config {
43 struct list_head list; /* list of all configs */
44 refcount_t refcount; /* reference count */
45 refcount_t entries; /* number of entries/rules
48 __be32 clusterip; /* the IP address */
49 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
50 int ifindex; /* device ifindex */
51 u_int16_t num_total_nodes; /* total number of nodes */
52 unsigned long local_nodes; /* node number array */
55 struct proc_dir_entry *pde; /* proc dir entry */
57 enum clusterip_hashmode hash_mode; /* which hashing mode */
58 u_int32_t hash_initval; /* hash initialization */
61 char ifname[IFNAMSIZ]; /* device ifname */
62 struct notifier_block notifier; /* refresh c->ifindex in it */
66 static const struct file_operations clusterip_proc_fops;
69 static unsigned int clusterip_net_id __read_mostly;
71 struct clusterip_net {
72 struct list_head configs;
73 /* lock protects the configs list */
77 struct proc_dir_entry *procdir;
82 clusterip_config_get(struct clusterip_config *c)
84 refcount_inc(&c->refcount);
88 static void clusterip_config_rcu_free(struct rcu_head *head)
90 kfree(container_of(head, struct clusterip_config, rcu));
94 clusterip_config_put(struct clusterip_config *c)
96 if (refcount_dec_and_test(&c->refcount))
97 call_rcu_bh(&c->rcu, clusterip_config_rcu_free);
100 /* decrease the count of entries using/referencing this config. If last
101 * entry(rule) is removed, remove the config from lists, but don't free it
102 * yet, since proc-files could still be holding references */
104 clusterip_config_entry_put(struct net *net, struct clusterip_config *c)
106 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
109 if (refcount_dec_and_lock(&c->entries, &cn->lock)) {
110 list_del_rcu(&c->list);
111 spin_unlock(&cn->lock);
114 unregister_netdevice_notifier(&c->notifier);
116 /* In case anyone still accesses the file, the open/close
117 * functions are also incrementing the refcount on their own,
118 * so it's safe to remove the entry even if it's in use. */
119 #ifdef CONFIG_PROC_FS
127 static struct clusterip_config *
128 __clusterip_config_find(struct net *net, __be32 clusterip)
130 struct clusterip_config *c;
131 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
133 list_for_each_entry_rcu(c, &cn->configs, list) {
134 if (c->clusterip == clusterip)
141 static inline struct clusterip_config *
142 clusterip_config_find_get(struct net *net, __be32 clusterip, int entry)
144 struct clusterip_config *c;
147 c = __clusterip_config_find(net, clusterip);
149 #ifdef CONFIG_PROC_FS
154 if (unlikely(!refcount_inc_not_zero(&c->refcount)))
157 refcount_inc(&c->entries);
159 rcu_read_unlock_bh();
165 clusterip_config_init_nodelist(struct clusterip_config *c,
166 const struct ipt_clusterip_tgt_info *i)
170 for (n = 0; n < i->num_local_nodes; n++)
171 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
175 clusterip_netdev_event(struct notifier_block *this, unsigned long event,
178 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
179 struct clusterip_config *c;
181 c = container_of(this, struct clusterip_config, notifier);
183 case NETDEV_REGISTER:
184 if (!strcmp(dev->name, c->ifname)) {
185 c->ifindex = dev->ifindex;
186 dev_mc_add(dev, c->clustermac);
189 case NETDEV_UNREGISTER:
190 if (dev->ifindex == c->ifindex) {
191 dev_mc_del(dev, c->clustermac);
195 case NETDEV_CHANGENAME:
196 if (!strcmp(dev->name, c->ifname)) {
197 c->ifindex = dev->ifindex;
198 dev_mc_add(dev, c->clustermac);
199 } else if (dev->ifindex == c->ifindex) {
200 dev_mc_del(dev, c->clustermac);
209 static struct clusterip_config *
210 clusterip_config_init(struct net *net, const struct ipt_clusterip_tgt_info *i,
211 __be32 ip, const char *iniface)
213 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
214 struct clusterip_config *c;
217 c = kzalloc(sizeof(*c), GFP_ATOMIC);
219 return ERR_PTR(-ENOMEM);
221 strcpy(c->ifname, iniface);
224 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
225 c->num_total_nodes = i->num_total_nodes;
226 clusterip_config_init_nodelist(c, i);
227 c->hash_mode = i->hash_mode;
228 c->hash_initval = i->hash_initval;
229 refcount_set(&c->refcount, 1);
230 refcount_set(&c->entries, 1);
232 spin_lock_bh(&cn->lock);
233 if (__clusterip_config_find(net, ip)) {
234 spin_unlock_bh(&cn->lock);
237 return ERR_PTR(-EBUSY);
240 list_add_rcu(&c->list, &cn->configs);
241 spin_unlock_bh(&cn->lock);
243 #ifdef CONFIG_PROC_FS
247 /* create proc dir entry */
248 sprintf(buffer, "%pI4", &ip);
249 c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR,
251 &clusterip_proc_fops, c);
259 c->notifier.notifier_call = clusterip_netdev_event;
260 err = register_netdevice_notifier(&c->notifier);
264 #ifdef CONFIG_PROC_FS
268 spin_lock_bh(&cn->lock);
269 list_del_rcu(&c->list);
270 spin_unlock_bh(&cn->lock);
276 #ifdef CONFIG_PROC_FS
278 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
282 nodenum > c->num_total_nodes)
285 /* check if we already have this number in our bitfield */
286 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
293 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
296 nodenum > c->num_total_nodes)
299 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
306 static inline u_int32_t
307 clusterip_hashfn(const struct sk_buff *skb,
308 const struct clusterip_config *config)
310 const struct iphdr *iph = ip_hdr(skb);
311 unsigned long hashval;
312 u_int16_t sport = 0, dport = 0;
315 poff = proto_ports_offset(iph->protocol);
317 const u_int16_t *ports;
320 ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports);
326 net_info_ratelimited("unknown protocol %u\n", iph->protocol);
329 switch (config->hash_mode) {
330 case CLUSTERIP_HASHMODE_SIP:
331 hashval = jhash_1word(ntohl(iph->saddr),
332 config->hash_initval);
334 case CLUSTERIP_HASHMODE_SIP_SPT:
335 hashval = jhash_2words(ntohl(iph->saddr), sport,
336 config->hash_initval);
338 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
339 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
340 config->hash_initval);
343 /* to make gcc happy */
345 /* This cannot happen, unless the check function wasn't called
346 * at rule load time */
347 pr_info("unknown mode %u\n", config->hash_mode);
352 /* node numbers are 1..n, not 0..n */
353 return reciprocal_scale(hashval, config->num_total_nodes) + 1;
357 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
359 return test_bit(hash - 1, &config->local_nodes);
362 /***********************************************************************
364 ***********************************************************************/
367 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
369 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
371 enum ip_conntrack_info ctinfo;
374 /* don't need to clusterip_config_get() here, since refcount
375 * is only decremented by destroy() - and ip_tables guarantees
376 * that the ->target() function isn't called after ->destroy() */
378 ct = nf_ct_get(skb, &ctinfo);
382 /* special case: ICMP error handling. conntrack distinguishes between
383 * error messages (RELATED) and information requests (see below) */
384 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
385 (ctinfo == IP_CT_RELATED ||
386 ctinfo == IP_CT_RELATED_REPLY))
389 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
390 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
391 * on, which all have an ID field [relevant for hashing]. */
393 hash = clusterip_hashfn(skb, cipinfo->config);
400 case IP_CT_RELATED_REPLY:
401 /* FIXME: we don't handle expectations at the moment.
402 * They can arrive on a different node than
403 * the master connection (e.g. FTP passive mode) */
404 case IP_CT_ESTABLISHED:
405 case IP_CT_ESTABLISHED_REPLY:
407 default: /* Prevent gcc warnings */
412 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
414 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
415 if (!clusterip_responsible(cipinfo->config, hash)) {
416 pr_debug("not responsible\n");
419 pr_debug("responsible\n");
421 /* despite being received via linklayer multicast, this is
422 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
423 skb->pkt_type = PACKET_HOST;
428 static int clusterip_tg_check(const struct xt_tgchk_param *par)
430 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
431 const struct ipt_entry *e = par->entryinfo;
432 struct clusterip_config *config;
435 if (par->nft_compat) {
436 pr_err("cannot use CLUSTERIP target from nftables compat\n");
440 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
441 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
442 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
443 pr_info("unknown mode %u\n", cipinfo->hash_mode);
447 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
448 e->ip.dst.s_addr == 0) {
449 pr_info("Please specify destination IP\n");
453 /* FIXME: further sanity checks */
455 config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1);
457 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
458 pr_info("no config found for %pI4, need 'new'\n",
462 struct net_device *dev;
464 if (e->ip.iniface[0] == '\0') {
465 pr_info("Please specify an interface name\n");
469 dev = dev_get_by_name(par->net, e->ip.iniface);
471 pr_info("no such interface %s\n",
477 config = clusterip_config_init(par->net, cipinfo,
481 return PTR_ERR(config);
484 cipinfo->config = config;
486 ret = nf_ct_netns_get(par->net, par->family);
488 pr_info("cannot load conntrack support for proto=%u\n",
491 if (!par->net->xt.clusterip_deprecated_warning) {
492 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
493 "use xt_cluster instead\n");
494 par->net->xt.clusterip_deprecated_warning = true;
500 /* drop reference count of cluster config when rule is deleted */
501 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
503 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
505 /* if no more entries are referencing the config, remove it
506 * from the list and destroy the proc entry */
507 clusterip_config_entry_put(par->net, cipinfo->config);
509 clusterip_config_put(cipinfo->config);
511 nf_ct_netns_put(par->net, par->family);
515 struct compat_ipt_clusterip_tgt_info
518 u_int8_t clustermac[6];
519 u_int16_t num_total_nodes;
520 u_int16_t num_local_nodes;
521 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
523 u_int32_t hash_initval;
524 compat_uptr_t config;
526 #endif /* CONFIG_COMPAT */
528 static struct xt_target clusterip_tg_reg __read_mostly = {
530 .family = NFPROTO_IPV4,
531 .target = clusterip_tg,
532 .checkentry = clusterip_tg_check,
533 .destroy = clusterip_tg_destroy,
534 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
535 .usersize = offsetof(struct ipt_clusterip_tgt_info, config),
537 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
538 #endif /* CONFIG_COMPAT */
543 /***********************************************************************
545 ***********************************************************************/
547 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
549 u_int8_t src_hw[ETH_ALEN];
551 u_int8_t dst_hw[ETH_ALEN];
556 static void arp_print(struct arp_payload *payload)
558 #define HBUFFERLEN 30
559 char hbuffer[HBUFFERLEN];
562 for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) {
563 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
564 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
569 pr_debug("src %pI4@%s, dst %pI4\n",
570 &payload->src_ip, hbuffer, &payload->dst_ip);
575 arp_mangle(void *priv,
577 const struct nf_hook_state *state)
579 struct arphdr *arp = arp_hdr(skb);
580 struct arp_payload *payload;
581 struct clusterip_config *c;
582 struct net *net = state->net;
584 /* we don't care about non-ethernet and non-ipv4 ARP */
585 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
586 arp->ar_pro != htons(ETH_P_IP) ||
587 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
590 /* we only want to mangle arp requests and replies */
591 if (arp->ar_op != htons(ARPOP_REPLY) &&
592 arp->ar_op != htons(ARPOP_REQUEST))
595 payload = (void *)(arp+1);
597 /* if there is no clusterip configuration for the arp reply's
598 * source ip, we don't want to mangle it */
599 c = clusterip_config_find_get(net, payload->src_ip, 0);
603 /* normally the linux kernel always replies to arp queries of
604 * addresses on different interfacs. However, in the CLUSTERIP case
605 * this wouldn't work, since we didn't subscribe the mcast group on
606 * other interfaces */
607 if (c->ifindex != state->out->ifindex) {
608 pr_debug("not mangling arp reply on different interface: cip'%d'-skb'%d'\n",
609 c->ifindex, state->out->ifindex);
610 clusterip_config_put(c);
614 /* mangle reply hardware address */
615 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
618 pr_debug("mangled arp reply: ");
622 clusterip_config_put(c);
627 static struct nf_hook_ops cip_arp_ops __read_mostly = {
630 .hooknum = NF_ARP_OUT,
634 /***********************************************************************
636 ***********************************************************************/
638 #ifdef CONFIG_PROC_FS
640 struct clusterip_seq_position {
641 unsigned int pos; /* position */
642 unsigned int weight; /* number of bits set == size */
643 unsigned int bit; /* current bit */
644 unsigned long val; /* current value */
647 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
649 struct clusterip_config *c = s->private;
651 u_int32_t local_nodes;
652 struct clusterip_seq_position *idx;
654 /* FIXME: possible race */
655 local_nodes = c->local_nodes;
656 weight = hweight32(local_nodes);
660 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
662 return ERR_PTR(-ENOMEM);
665 idx->weight = weight;
666 idx->bit = ffs(local_nodes);
667 idx->val = local_nodes;
668 clear_bit(idx->bit - 1, &idx->val);
673 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
675 struct clusterip_seq_position *idx = v;
678 if (*pos >= idx->weight) {
682 idx->bit = ffs(idx->val);
683 clear_bit(idx->bit - 1, &idx->val);
687 static void clusterip_seq_stop(struct seq_file *s, void *v)
693 static int clusterip_seq_show(struct seq_file *s, void *v)
695 struct clusterip_seq_position *idx = v;
700 seq_printf(s, "%u", idx->bit);
702 if (idx->pos == idx->weight - 1)
708 static const struct seq_operations clusterip_seq_ops = {
709 .start = clusterip_seq_start,
710 .next = clusterip_seq_next,
711 .stop = clusterip_seq_stop,
712 .show = clusterip_seq_show,
715 static int clusterip_proc_open(struct inode *inode, struct file *file)
717 int ret = seq_open(file, &clusterip_seq_ops);
720 struct seq_file *sf = file->private_data;
721 struct clusterip_config *c = PDE_DATA(inode);
725 clusterip_config_get(c);
731 static int clusterip_proc_release(struct inode *inode, struct file *file)
733 struct clusterip_config *c = PDE_DATA(inode);
736 ret = seq_release(inode, file);
739 clusterip_config_put(c);
744 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
745 size_t size, loff_t *ofs)
747 struct clusterip_config *c = PDE_DATA(file_inode(file));
748 #define PROC_WRITELEN 10
749 char buffer[PROC_WRITELEN+1];
750 unsigned long nodenum;
753 if (size > PROC_WRITELEN)
755 if (copy_from_user(buffer, input, size))
759 if (*buffer == '+') {
760 rc = kstrtoul(buffer+1, 10, &nodenum);
763 if (clusterip_add_node(c, nodenum))
765 } else if (*buffer == '-') {
766 rc = kstrtoul(buffer+1, 10, &nodenum);
769 if (clusterip_del_node(c, nodenum))
777 static const struct file_operations clusterip_proc_fops = {
778 .owner = THIS_MODULE,
779 .open = clusterip_proc_open,
781 .write = clusterip_proc_write,
783 .release = clusterip_proc_release,
786 #endif /* CONFIG_PROC_FS */
788 static int clusterip_net_init(struct net *net)
790 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
793 INIT_LIST_HEAD(&cn->configs);
795 spin_lock_init(&cn->lock);
797 ret = nf_register_net_hook(net, &cip_arp_ops);
801 #ifdef CONFIG_PROC_FS
802 cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net);
804 nf_unregister_net_hook(net, &cip_arp_ops);
805 pr_err("Unable to proc dir entry\n");
808 #endif /* CONFIG_PROC_FS */
813 static void clusterip_net_exit(struct net *net)
815 #ifdef CONFIG_PROC_FS
816 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
817 proc_remove(cn->procdir);
819 nf_unregister_net_hook(net, &cip_arp_ops);
822 static struct pernet_operations clusterip_net_ops = {
823 .init = clusterip_net_init,
824 .exit = clusterip_net_exit,
825 .id = &clusterip_net_id,
826 .size = sizeof(struct clusterip_net),
829 static int __init clusterip_tg_init(void)
833 ret = register_pernet_subsys(&clusterip_net_ops);
837 ret = xt_register_target(&clusterip_tg_reg);
841 pr_info("ClusterIP Version %s loaded successfully\n",
847 unregister_pernet_subsys(&clusterip_net_ops);
851 static void __exit clusterip_tg_exit(void)
853 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
855 xt_unregister_target(&clusterip_tg_reg);
856 unregister_pernet_subsys(&clusterip_net_ops);
858 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
862 module_init(clusterip_tg_init);
863 module_exit(clusterip_tg_exit);