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 if (unlikely(!refcount_inc_not_zero(&c->entries))) {
159 clusterip_config_put(c);
164 rcu_read_unlock_bh();
170 clusterip_config_init_nodelist(struct clusterip_config *c,
171 const struct ipt_clusterip_tgt_info *i)
175 for (n = 0; n < i->num_local_nodes; n++)
176 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
180 clusterip_netdev_event(struct notifier_block *this, unsigned long event,
183 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
184 struct clusterip_config *c;
186 c = container_of(this, struct clusterip_config, notifier);
188 case NETDEV_REGISTER:
189 if (!strcmp(dev->name, c->ifname)) {
190 c->ifindex = dev->ifindex;
191 dev_mc_add(dev, c->clustermac);
194 case NETDEV_UNREGISTER:
195 if (dev->ifindex == c->ifindex) {
196 dev_mc_del(dev, c->clustermac);
200 case NETDEV_CHANGENAME:
201 if (!strcmp(dev->name, c->ifname)) {
202 c->ifindex = dev->ifindex;
203 dev_mc_add(dev, c->clustermac);
204 } else if (dev->ifindex == c->ifindex) {
205 dev_mc_del(dev, c->clustermac);
214 static struct clusterip_config *
215 clusterip_config_init(struct net *net, const struct ipt_clusterip_tgt_info *i,
216 __be32 ip, const char *iniface)
218 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
219 struct clusterip_config *c;
222 c = kzalloc(sizeof(*c), GFP_ATOMIC);
224 return ERR_PTR(-ENOMEM);
226 strcpy(c->ifname, iniface);
229 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
230 c->num_total_nodes = i->num_total_nodes;
231 clusterip_config_init_nodelist(c, i);
232 c->hash_mode = i->hash_mode;
233 c->hash_initval = i->hash_initval;
234 refcount_set(&c->refcount, 1);
236 spin_lock_bh(&cn->lock);
237 if (__clusterip_config_find(net, ip)) {
238 spin_unlock_bh(&cn->lock);
241 return ERR_PTR(-EBUSY);
244 list_add_rcu(&c->list, &cn->configs);
245 spin_unlock_bh(&cn->lock);
247 #ifdef CONFIG_PROC_FS
251 /* create proc dir entry */
252 sprintf(buffer, "%pI4", &ip);
253 c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR,
255 &clusterip_proc_fops, c);
263 c->notifier.notifier_call = clusterip_netdev_event;
264 err = register_netdevice_notifier(&c->notifier);
266 refcount_set(&c->entries, 1);
270 #ifdef CONFIG_PROC_FS
274 spin_lock_bh(&cn->lock);
275 list_del_rcu(&c->list);
276 spin_unlock_bh(&cn->lock);
277 clusterip_config_put(c);
282 #ifdef CONFIG_PROC_FS
284 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
288 nodenum > c->num_total_nodes)
291 /* check if we already have this number in our bitfield */
292 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
299 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
302 nodenum > c->num_total_nodes)
305 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
312 static inline u_int32_t
313 clusterip_hashfn(const struct sk_buff *skb,
314 const struct clusterip_config *config)
316 const struct iphdr *iph = ip_hdr(skb);
317 unsigned long hashval;
318 u_int16_t sport = 0, dport = 0;
321 poff = proto_ports_offset(iph->protocol);
323 const u_int16_t *ports;
326 ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports);
332 net_info_ratelimited("unknown protocol %u\n", iph->protocol);
335 switch (config->hash_mode) {
336 case CLUSTERIP_HASHMODE_SIP:
337 hashval = jhash_1word(ntohl(iph->saddr),
338 config->hash_initval);
340 case CLUSTERIP_HASHMODE_SIP_SPT:
341 hashval = jhash_2words(ntohl(iph->saddr), sport,
342 config->hash_initval);
344 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
345 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
346 config->hash_initval);
349 /* to make gcc happy */
351 /* This cannot happen, unless the check function wasn't called
352 * at rule load time */
353 pr_info("unknown mode %u\n", config->hash_mode);
358 /* node numbers are 1..n, not 0..n */
359 return reciprocal_scale(hashval, config->num_total_nodes) + 1;
363 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
365 return test_bit(hash - 1, &config->local_nodes);
368 /***********************************************************************
370 ***********************************************************************/
373 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
375 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
377 enum ip_conntrack_info ctinfo;
380 /* don't need to clusterip_config_get() here, since refcount
381 * is only decremented by destroy() - and ip_tables guarantees
382 * that the ->target() function isn't called after ->destroy() */
384 ct = nf_ct_get(skb, &ctinfo);
388 /* special case: ICMP error handling. conntrack distinguishes between
389 * error messages (RELATED) and information requests (see below) */
390 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
391 (ctinfo == IP_CT_RELATED ||
392 ctinfo == IP_CT_RELATED_REPLY))
395 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
396 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
397 * on, which all have an ID field [relevant for hashing]. */
399 hash = clusterip_hashfn(skb, cipinfo->config);
406 case IP_CT_RELATED_REPLY:
407 /* FIXME: we don't handle expectations at the moment.
408 * They can arrive on a different node than
409 * the master connection (e.g. FTP passive mode) */
410 case IP_CT_ESTABLISHED:
411 case IP_CT_ESTABLISHED_REPLY:
413 default: /* Prevent gcc warnings */
418 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
420 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
421 if (!clusterip_responsible(cipinfo->config, hash)) {
422 pr_debug("not responsible\n");
425 pr_debug("responsible\n");
427 /* despite being received via linklayer multicast, this is
428 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
429 skb->pkt_type = PACKET_HOST;
434 static int clusterip_tg_check(const struct xt_tgchk_param *par)
436 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
437 const struct ipt_entry *e = par->entryinfo;
438 struct clusterip_config *config;
441 if (par->nft_compat) {
442 pr_err("cannot use CLUSTERIP target from nftables compat\n");
446 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
447 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
448 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
449 pr_info("unknown mode %u\n", cipinfo->hash_mode);
453 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
454 e->ip.dst.s_addr == 0) {
455 pr_info("Please specify destination IP\n");
458 if (cipinfo->num_local_nodes > ARRAY_SIZE(cipinfo->local_nodes)) {
459 pr_info("bad num_local_nodes %u\n", cipinfo->num_local_nodes);
462 for (i = 0; i < cipinfo->num_local_nodes; i++) {
463 if (cipinfo->local_nodes[i] - 1 >=
464 sizeof(config->local_nodes) * 8) {
465 pr_info("bad local_nodes[%d] %u\n",
466 i, cipinfo->local_nodes[i]);
471 config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1);
473 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
474 pr_info("no config found for %pI4, need 'new'\n",
478 struct net_device *dev;
480 if (e->ip.iniface[0] == '\0') {
481 pr_info("Please specify an interface name\n");
485 dev = dev_get_by_name(par->net, e->ip.iniface);
487 pr_info("no such interface %s\n",
493 config = clusterip_config_init(par->net, cipinfo,
497 return PTR_ERR(config);
501 ret = nf_ct_netns_get(par->net, par->family);
503 pr_info("cannot load conntrack support for proto=%u\n",
505 clusterip_config_entry_put(par->net, config);
506 clusterip_config_put(config);
510 if (!par->net->xt.clusterip_deprecated_warning) {
511 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
512 "use xt_cluster instead\n");
513 par->net->xt.clusterip_deprecated_warning = true;
516 cipinfo->config = config;
520 /* drop reference count of cluster config when rule is deleted */
521 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
523 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
525 /* if no more entries are referencing the config, remove it
526 * from the list and destroy the proc entry */
527 clusterip_config_entry_put(par->net, cipinfo->config);
529 clusterip_config_put(cipinfo->config);
531 nf_ct_netns_put(par->net, par->family);
535 struct compat_ipt_clusterip_tgt_info
538 u_int8_t clustermac[6];
539 u_int16_t num_total_nodes;
540 u_int16_t num_local_nodes;
541 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
543 u_int32_t hash_initval;
544 compat_uptr_t config;
546 #endif /* CONFIG_COMPAT */
548 static struct xt_target clusterip_tg_reg __read_mostly = {
550 .family = NFPROTO_IPV4,
551 .target = clusterip_tg,
552 .checkentry = clusterip_tg_check,
553 .destroy = clusterip_tg_destroy,
554 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
555 .usersize = offsetof(struct ipt_clusterip_tgt_info, config),
557 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
558 #endif /* CONFIG_COMPAT */
563 /***********************************************************************
565 ***********************************************************************/
567 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
569 u_int8_t src_hw[ETH_ALEN];
571 u_int8_t dst_hw[ETH_ALEN];
576 static void arp_print(struct arp_payload *payload)
578 #define HBUFFERLEN 30
579 char hbuffer[HBUFFERLEN];
582 for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) {
583 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
584 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
589 pr_debug("src %pI4@%s, dst %pI4\n",
590 &payload->src_ip, hbuffer, &payload->dst_ip);
595 arp_mangle(void *priv,
597 const struct nf_hook_state *state)
599 struct arphdr *arp = arp_hdr(skb);
600 struct arp_payload *payload;
601 struct clusterip_config *c;
602 struct net *net = state->net;
604 /* we don't care about non-ethernet and non-ipv4 ARP */
605 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
606 arp->ar_pro != htons(ETH_P_IP) ||
607 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
610 /* we only want to mangle arp requests and replies */
611 if (arp->ar_op != htons(ARPOP_REPLY) &&
612 arp->ar_op != htons(ARPOP_REQUEST))
615 payload = (void *)(arp+1);
617 /* if there is no clusterip configuration for the arp reply's
618 * source ip, we don't want to mangle it */
619 c = clusterip_config_find_get(net, payload->src_ip, 0);
623 /* normally the linux kernel always replies to arp queries of
624 * addresses on different interfacs. However, in the CLUSTERIP case
625 * this wouldn't work, since we didn't subscribe the mcast group on
626 * other interfaces */
627 if (c->ifindex != state->out->ifindex) {
628 pr_debug("not mangling arp reply on different interface: cip'%d'-skb'%d'\n",
629 c->ifindex, state->out->ifindex);
630 clusterip_config_put(c);
634 /* mangle reply hardware address */
635 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
638 pr_debug("mangled arp reply: ");
642 clusterip_config_put(c);
647 static const struct nf_hook_ops cip_arp_ops = {
650 .hooknum = NF_ARP_OUT,
654 /***********************************************************************
656 ***********************************************************************/
658 #ifdef CONFIG_PROC_FS
660 struct clusterip_seq_position {
661 unsigned int pos; /* position */
662 unsigned int weight; /* number of bits set == size */
663 unsigned int bit; /* current bit */
664 unsigned long val; /* current value */
667 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
669 struct clusterip_config *c = s->private;
671 u_int32_t local_nodes;
672 struct clusterip_seq_position *idx;
674 /* FIXME: possible race */
675 local_nodes = c->local_nodes;
676 weight = hweight32(local_nodes);
680 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
682 return ERR_PTR(-ENOMEM);
685 idx->weight = weight;
686 idx->bit = ffs(local_nodes);
687 idx->val = local_nodes;
688 clear_bit(idx->bit - 1, &idx->val);
693 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
695 struct clusterip_seq_position *idx = v;
698 if (*pos >= idx->weight) {
702 idx->bit = ffs(idx->val);
703 clear_bit(idx->bit - 1, &idx->val);
707 static void clusterip_seq_stop(struct seq_file *s, void *v)
713 static int clusterip_seq_show(struct seq_file *s, void *v)
715 struct clusterip_seq_position *idx = v;
720 seq_printf(s, "%u", idx->bit);
722 if (idx->pos == idx->weight - 1)
728 static const struct seq_operations clusterip_seq_ops = {
729 .start = clusterip_seq_start,
730 .next = clusterip_seq_next,
731 .stop = clusterip_seq_stop,
732 .show = clusterip_seq_show,
735 static int clusterip_proc_open(struct inode *inode, struct file *file)
737 int ret = seq_open(file, &clusterip_seq_ops);
740 struct seq_file *sf = file->private_data;
741 struct clusterip_config *c = PDE_DATA(inode);
745 clusterip_config_get(c);
751 static int clusterip_proc_release(struct inode *inode, struct file *file)
753 struct clusterip_config *c = PDE_DATA(inode);
756 ret = seq_release(inode, file);
759 clusterip_config_put(c);
764 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
765 size_t size, loff_t *ofs)
767 struct clusterip_config *c = PDE_DATA(file_inode(file));
768 #define PROC_WRITELEN 10
769 char buffer[PROC_WRITELEN+1];
770 unsigned long nodenum;
773 if (size > PROC_WRITELEN)
775 if (copy_from_user(buffer, input, size))
779 if (*buffer == '+') {
780 rc = kstrtoul(buffer+1, 10, &nodenum);
783 if (clusterip_add_node(c, nodenum))
785 } else if (*buffer == '-') {
786 rc = kstrtoul(buffer+1, 10, &nodenum);
789 if (clusterip_del_node(c, nodenum))
797 static const struct file_operations clusterip_proc_fops = {
798 .open = clusterip_proc_open,
800 .write = clusterip_proc_write,
802 .release = clusterip_proc_release,
805 #endif /* CONFIG_PROC_FS */
807 static int clusterip_net_init(struct net *net)
809 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
812 INIT_LIST_HEAD(&cn->configs);
814 spin_lock_init(&cn->lock);
816 ret = nf_register_net_hook(net, &cip_arp_ops);
820 #ifdef CONFIG_PROC_FS
821 cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net);
823 nf_unregister_net_hook(net, &cip_arp_ops);
824 pr_err("Unable to proc dir entry\n");
827 #endif /* CONFIG_PROC_FS */
832 static void clusterip_net_exit(struct net *net)
834 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
835 #ifdef CONFIG_PROC_FS
836 proc_remove(cn->procdir);
839 nf_unregister_net_hook(net, &cip_arp_ops);
840 WARN_ON_ONCE(!list_empty(&cn->configs));
843 static struct pernet_operations clusterip_net_ops = {
844 .init = clusterip_net_init,
845 .exit = clusterip_net_exit,
846 .id = &clusterip_net_id,
847 .size = sizeof(struct clusterip_net),
850 static int __init clusterip_tg_init(void)
854 ret = register_pernet_subsys(&clusterip_net_ops);
858 ret = xt_register_target(&clusterip_tg_reg);
862 pr_info("ClusterIP Version %s loaded successfully\n",
868 unregister_pernet_subsys(&clusterip_net_ops);
872 static void __exit clusterip_tg_exit(void)
874 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
876 xt_unregister_target(&clusterip_tg_reg);
877 unregister_pernet_subsys(&clusterip_net_ops);
879 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
883 module_init(clusterip_tg_init);
884 module_exit(clusterip_tg_exit);