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 /* In case anyone still accesses the file, the open/close
111 * functions are also incrementing the refcount on their own,
112 * so it's safe to remove the entry even if it's in use. */
113 #ifdef CONFIG_PROC_FS
117 list_del_rcu(&c->list);
118 spin_unlock(&cn->lock);
121 unregister_netdevice_notifier(&c->notifier);
128 static struct clusterip_config *
129 __clusterip_config_find(struct net *net, __be32 clusterip)
131 struct clusterip_config *c;
132 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
134 list_for_each_entry_rcu(c, &cn->configs, list) {
135 if (c->clusterip == clusterip)
142 static inline struct clusterip_config *
143 clusterip_config_find_get(struct net *net, __be32 clusterip, int entry)
145 struct clusterip_config *c;
148 c = __clusterip_config_find(net, clusterip);
150 #ifdef CONFIG_PROC_FS
155 if (unlikely(!refcount_inc_not_zero(&c->refcount)))
158 refcount_inc(&c->entries);
160 rcu_read_unlock_bh();
166 clusterip_config_init_nodelist(struct clusterip_config *c,
167 const struct ipt_clusterip_tgt_info *i)
171 for (n = 0; n < i->num_local_nodes; n++)
172 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
176 clusterip_netdev_event(struct notifier_block *this, unsigned long event,
179 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
180 struct clusterip_config *c;
182 c = container_of(this, struct clusterip_config, notifier);
184 case NETDEV_REGISTER:
185 if (!strcmp(dev->name, c->ifname)) {
186 c->ifindex = dev->ifindex;
187 dev_mc_add(dev, c->clustermac);
190 case NETDEV_UNREGISTER:
191 if (dev->ifindex == c->ifindex) {
192 dev_mc_del(dev, c->clustermac);
196 case NETDEV_CHANGENAME:
197 if (!strcmp(dev->name, c->ifname)) {
198 c->ifindex = dev->ifindex;
199 dev_mc_add(dev, c->clustermac);
200 } else if (dev->ifindex == c->ifindex) {
201 dev_mc_del(dev, c->clustermac);
210 static struct clusterip_config *
211 clusterip_config_init(struct net *net, const struct ipt_clusterip_tgt_info *i,
212 __be32 ip, const char *iniface)
214 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
215 struct clusterip_config *c;
218 c = kzalloc(sizeof(*c), GFP_ATOMIC);
220 return ERR_PTR(-ENOMEM);
222 strcpy(c->ifname, iniface);
225 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
226 c->num_total_nodes = i->num_total_nodes;
227 clusterip_config_init_nodelist(c, i);
228 c->hash_mode = i->hash_mode;
229 c->hash_initval = i->hash_initval;
230 refcount_set(&c->refcount, 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);
262 refcount_set(&c->entries, 1);
266 #ifdef CONFIG_PROC_FS
270 spin_lock_bh(&cn->lock);
271 list_del_rcu(&c->list);
272 spin_unlock_bh(&cn->lock);
273 clusterip_config_put(c);
278 #ifdef CONFIG_PROC_FS
280 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
284 nodenum > c->num_total_nodes)
287 /* check if we already have this number in our bitfield */
288 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
295 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
298 nodenum > c->num_total_nodes)
301 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
308 static inline u_int32_t
309 clusterip_hashfn(const struct sk_buff *skb,
310 const struct clusterip_config *config)
312 const struct iphdr *iph = ip_hdr(skb);
313 unsigned long hashval;
314 u_int16_t sport = 0, dport = 0;
317 poff = proto_ports_offset(iph->protocol);
319 const u_int16_t *ports;
322 ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports);
328 net_info_ratelimited("unknown protocol %u\n", iph->protocol);
331 switch (config->hash_mode) {
332 case CLUSTERIP_HASHMODE_SIP:
333 hashval = jhash_1word(ntohl(iph->saddr),
334 config->hash_initval);
336 case CLUSTERIP_HASHMODE_SIP_SPT:
337 hashval = jhash_2words(ntohl(iph->saddr), sport,
338 config->hash_initval);
340 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
341 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
342 config->hash_initval);
345 /* to make gcc happy */
347 /* This cannot happen, unless the check function wasn't called
348 * at rule load time */
349 pr_info("unknown mode %u\n", config->hash_mode);
354 /* node numbers are 1..n, not 0..n */
355 return reciprocal_scale(hashval, config->num_total_nodes) + 1;
359 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
361 return test_bit(hash - 1, &config->local_nodes);
364 /***********************************************************************
366 ***********************************************************************/
369 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
371 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
373 enum ip_conntrack_info ctinfo;
376 /* don't need to clusterip_config_get() here, since refcount
377 * is only decremented by destroy() - and ip_tables guarantees
378 * that the ->target() function isn't called after ->destroy() */
380 ct = nf_ct_get(skb, &ctinfo);
384 /* special case: ICMP error handling. conntrack distinguishes between
385 * error messages (RELATED) and information requests (see below) */
386 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
387 (ctinfo == IP_CT_RELATED ||
388 ctinfo == IP_CT_RELATED_REPLY))
391 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
392 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
393 * on, which all have an ID field [relevant for hashing]. */
395 hash = clusterip_hashfn(skb, cipinfo->config);
402 case IP_CT_RELATED_REPLY:
403 /* FIXME: we don't handle expectations at the moment.
404 * They can arrive on a different node than
405 * the master connection (e.g. FTP passive mode) */
406 case IP_CT_ESTABLISHED:
407 case IP_CT_ESTABLISHED_REPLY:
409 default: /* Prevent gcc warnings */
414 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
416 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
417 if (!clusterip_responsible(cipinfo->config, hash)) {
418 pr_debug("not responsible\n");
421 pr_debug("responsible\n");
423 /* despite being received via linklayer multicast, this is
424 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
425 skb->pkt_type = PACKET_HOST;
430 static int clusterip_tg_check(const struct xt_tgchk_param *par)
432 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
433 const struct ipt_entry *e = par->entryinfo;
434 struct clusterip_config *config;
437 if (par->nft_compat) {
438 pr_err("cannot use CLUSTERIP target from nftables compat\n");
442 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
443 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
444 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
445 pr_info("unknown mode %u\n", cipinfo->hash_mode);
449 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
450 e->ip.dst.s_addr == 0) {
451 pr_info("Please specify destination IP\n");
454 if (cipinfo->num_local_nodes > ARRAY_SIZE(cipinfo->local_nodes)) {
455 pr_info("bad num_local_nodes %u\n", cipinfo->num_local_nodes);
458 for (i = 0; i < cipinfo->num_local_nodes; i++) {
459 if (cipinfo->local_nodes[i] - 1 >=
460 sizeof(config->local_nodes) * 8) {
461 pr_info("bad local_nodes[%d] %u\n",
462 i, cipinfo->local_nodes[i]);
467 config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1);
469 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
470 pr_info("no config found for %pI4, need 'new'\n",
474 struct net_device *dev;
476 if (e->ip.iniface[0] == '\0') {
477 pr_info("Please specify an interface name\n");
481 dev = dev_get_by_name(par->net, e->ip.iniface);
483 pr_info("no such interface %s\n",
489 config = clusterip_config_init(par->net, cipinfo,
493 return PTR_ERR(config);
495 } else if (memcmp(&config->clustermac, &cipinfo->clustermac, ETH_ALEN))
498 ret = nf_ct_netns_get(par->net, par->family);
500 pr_info("cannot load conntrack support for proto=%u\n",
502 clusterip_config_entry_put(par->net, config);
503 clusterip_config_put(config);
507 if (!par->net->xt.clusterip_deprecated_warning) {
508 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
509 "use xt_cluster instead\n");
510 par->net->xt.clusterip_deprecated_warning = true;
513 cipinfo->config = config;
517 /* drop reference count of cluster config when rule is deleted */
518 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
520 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
522 /* if no more entries are referencing the config, remove it
523 * from the list and destroy the proc entry */
524 clusterip_config_entry_put(par->net, cipinfo->config);
526 clusterip_config_put(cipinfo->config);
528 nf_ct_netns_put(par->net, par->family);
532 struct compat_ipt_clusterip_tgt_info
535 u_int8_t clustermac[6];
536 u_int16_t num_total_nodes;
537 u_int16_t num_local_nodes;
538 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
540 u_int32_t hash_initval;
541 compat_uptr_t config;
543 #endif /* CONFIG_COMPAT */
545 static struct xt_target clusterip_tg_reg __read_mostly = {
547 .family = NFPROTO_IPV4,
548 .target = clusterip_tg,
549 .checkentry = clusterip_tg_check,
550 .destroy = clusterip_tg_destroy,
551 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
552 .usersize = offsetof(struct ipt_clusterip_tgt_info, config),
554 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
555 #endif /* CONFIG_COMPAT */
560 /***********************************************************************
562 ***********************************************************************/
564 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
566 u_int8_t src_hw[ETH_ALEN];
568 u_int8_t dst_hw[ETH_ALEN];
573 static void arp_print(struct arp_payload *payload)
575 #define HBUFFERLEN 30
576 char hbuffer[HBUFFERLEN];
579 for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) {
580 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
581 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
586 pr_debug("src %pI4@%s, dst %pI4\n",
587 &payload->src_ip, hbuffer, &payload->dst_ip);
592 arp_mangle(void *priv,
594 const struct nf_hook_state *state)
596 struct arphdr *arp = arp_hdr(skb);
597 struct arp_payload *payload;
598 struct clusterip_config *c;
599 struct net *net = state->net;
601 /* we don't care about non-ethernet and non-ipv4 ARP */
602 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
603 arp->ar_pro != htons(ETH_P_IP) ||
604 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
607 /* we only want to mangle arp requests and replies */
608 if (arp->ar_op != htons(ARPOP_REPLY) &&
609 arp->ar_op != htons(ARPOP_REQUEST))
612 payload = (void *)(arp+1);
614 /* if there is no clusterip configuration for the arp reply's
615 * source ip, we don't want to mangle it */
616 c = clusterip_config_find_get(net, payload->src_ip, 0);
620 /* normally the linux kernel always replies to arp queries of
621 * addresses on different interfacs. However, in the CLUSTERIP case
622 * this wouldn't work, since we didn't subscribe the mcast group on
623 * other interfaces */
624 if (c->ifindex != state->out->ifindex) {
625 pr_debug("not mangling arp reply on different interface: cip'%d'-skb'%d'\n",
626 c->ifindex, state->out->ifindex);
627 clusterip_config_put(c);
631 /* mangle reply hardware address */
632 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
635 pr_debug("mangled arp reply: ");
639 clusterip_config_put(c);
644 static const struct nf_hook_ops cip_arp_ops = {
647 .hooknum = NF_ARP_OUT,
651 /***********************************************************************
653 ***********************************************************************/
655 #ifdef CONFIG_PROC_FS
657 struct clusterip_seq_position {
658 unsigned int pos; /* position */
659 unsigned int weight; /* number of bits set == size */
660 unsigned int bit; /* current bit */
661 unsigned long val; /* current value */
664 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
666 struct clusterip_config *c = s->private;
668 u_int32_t local_nodes;
669 struct clusterip_seq_position *idx;
671 /* FIXME: possible race */
672 local_nodes = c->local_nodes;
673 weight = hweight32(local_nodes);
677 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
679 return ERR_PTR(-ENOMEM);
682 idx->weight = weight;
683 idx->bit = ffs(local_nodes);
684 idx->val = local_nodes;
685 clear_bit(idx->bit - 1, &idx->val);
690 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
692 struct clusterip_seq_position *idx = v;
695 if (*pos >= idx->weight) {
699 idx->bit = ffs(idx->val);
700 clear_bit(idx->bit - 1, &idx->val);
704 static void clusterip_seq_stop(struct seq_file *s, void *v)
710 static int clusterip_seq_show(struct seq_file *s, void *v)
712 struct clusterip_seq_position *idx = v;
717 seq_printf(s, "%u", idx->bit);
719 if (idx->pos == idx->weight - 1)
725 static const struct seq_operations clusterip_seq_ops = {
726 .start = clusterip_seq_start,
727 .next = clusterip_seq_next,
728 .stop = clusterip_seq_stop,
729 .show = clusterip_seq_show,
732 static int clusterip_proc_open(struct inode *inode, struct file *file)
734 int ret = seq_open(file, &clusterip_seq_ops);
737 struct seq_file *sf = file->private_data;
738 struct clusterip_config *c = PDE_DATA(inode);
742 clusterip_config_get(c);
748 static int clusterip_proc_release(struct inode *inode, struct file *file)
750 struct clusterip_config *c = PDE_DATA(inode);
753 ret = seq_release(inode, file);
756 clusterip_config_put(c);
761 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
762 size_t size, loff_t *ofs)
764 struct clusterip_config *c = PDE_DATA(file_inode(file));
765 #define PROC_WRITELEN 10
766 char buffer[PROC_WRITELEN+1];
767 unsigned long nodenum;
770 if (size > PROC_WRITELEN)
772 if (copy_from_user(buffer, input, size))
776 if (*buffer == '+') {
777 rc = kstrtoul(buffer+1, 10, &nodenum);
780 if (clusterip_add_node(c, nodenum))
782 } else if (*buffer == '-') {
783 rc = kstrtoul(buffer+1, 10, &nodenum);
786 if (clusterip_del_node(c, nodenum))
794 static const struct file_operations clusterip_proc_fops = {
795 .owner = THIS_MODULE,
796 .open = clusterip_proc_open,
798 .write = clusterip_proc_write,
800 .release = clusterip_proc_release,
803 #endif /* CONFIG_PROC_FS */
805 static int clusterip_net_init(struct net *net)
807 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
810 INIT_LIST_HEAD(&cn->configs);
812 spin_lock_init(&cn->lock);
814 ret = nf_register_net_hook(net, &cip_arp_ops);
818 #ifdef CONFIG_PROC_FS
819 cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net);
821 nf_unregister_net_hook(net, &cip_arp_ops);
822 pr_err("Unable to proc dir entry\n");
825 #endif /* CONFIG_PROC_FS */
830 static void clusterip_net_exit(struct net *net)
832 #ifdef CONFIG_PROC_FS
833 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
834 proc_remove(cn->procdir);
837 nf_unregister_net_hook(net, &cip_arp_ops);
840 static struct pernet_operations clusterip_net_ops = {
841 .init = clusterip_net_init,
842 .exit = clusterip_net_exit,
843 .id = &clusterip_net_id,
844 .size = sizeof(struct clusterip_net),
847 static int __init clusterip_tg_init(void)
851 ret = register_pernet_subsys(&clusterip_net_ops);
855 ret = xt_register_target(&clusterip_tg_reg);
859 pr_info("ClusterIP Version %s loaded successfully\n",
865 unregister_pernet_subsys(&clusterip_net_ops);
869 static void __exit clusterip_tg_exit(void)
871 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
873 xt_unregister_target(&clusterip_tg_reg);
874 unregister_pernet_subsys(&clusterip_net_ops);
876 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
880 module_init(clusterip_tg_init);
881 module_exit(clusterip_tg_exit);