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 */
60 struct net *net; /* netns for pernet list */
61 char ifname[IFNAMSIZ]; /* device ifname */
65 static const struct file_operations clusterip_proc_fops;
68 struct clusterip_net {
69 struct list_head configs;
70 /* lock protects the configs list */
74 struct proc_dir_entry *procdir;
78 static unsigned int clusterip_net_id __read_mostly;
79 static inline struct clusterip_net *clusterip_pernet(struct net *net)
81 return net_generic(net, clusterip_net_id);
85 clusterip_config_get(struct clusterip_config *c)
87 refcount_inc(&c->refcount);
90 static void clusterip_config_rcu_free(struct rcu_head *head)
92 struct clusterip_config *config;
93 struct net_device *dev;
95 config = container_of(head, struct clusterip_config, rcu);
96 dev = dev_get_by_name(config->net, config->ifname);
98 dev_mc_del(dev, config->clustermac);
105 clusterip_config_put(struct clusterip_config *c)
107 if (refcount_dec_and_test(&c->refcount))
108 call_rcu(&c->rcu, clusterip_config_rcu_free);
111 /* decrease the count of entries using/referencing this config. If last
112 * entry(rule) is removed, remove the config from lists, but don't free it
113 * yet, since proc-files could still be holding references */
115 clusterip_config_entry_put(struct clusterip_config *c)
117 struct clusterip_net *cn = clusterip_pernet(c->net);
120 if (refcount_dec_and_lock(&c->entries, &cn->lock)) {
121 /* In case anyone still accesses the file, the open/close
122 * functions are also incrementing the refcount on their own,
123 * so it's safe to remove the entry even if it's in use. */
124 #ifdef CONFIG_PROC_FS
128 list_del_rcu(&c->list);
129 spin_unlock(&cn->lock);
137 static struct clusterip_config *
138 __clusterip_config_find(struct net *net, __be32 clusterip)
140 struct clusterip_config *c;
141 struct clusterip_net *cn = clusterip_pernet(net);
143 list_for_each_entry_rcu(c, &cn->configs, list) {
144 if (c->clusterip == clusterip)
151 static inline struct clusterip_config *
152 clusterip_config_find_get(struct net *net, __be32 clusterip, int entry)
154 struct clusterip_config *c;
157 c = __clusterip_config_find(net, clusterip);
159 #ifdef CONFIG_PROC_FS
164 if (unlikely(!refcount_inc_not_zero(&c->refcount)))
167 if (unlikely(!refcount_inc_not_zero(&c->entries))) {
168 clusterip_config_put(c);
173 rcu_read_unlock_bh();
179 clusterip_config_init_nodelist(struct clusterip_config *c,
180 const struct ipt_clusterip_tgt_info *i)
184 for (n = 0; n < i->num_local_nodes; n++)
185 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
189 clusterip_netdev_event(struct notifier_block *this, unsigned long event,
192 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
193 struct net *net = dev_net(dev);
194 struct clusterip_net *cn = clusterip_pernet(net);
195 struct clusterip_config *c;
197 spin_lock_bh(&cn->lock);
198 list_for_each_entry_rcu(c, &cn->configs, list) {
200 case NETDEV_REGISTER:
201 if (!strcmp(dev->name, c->ifname)) {
202 c->ifindex = dev->ifindex;
203 dev_mc_add(dev, c->clustermac);
206 case NETDEV_UNREGISTER:
207 if (dev->ifindex == c->ifindex) {
208 dev_mc_del(dev, c->clustermac);
212 case NETDEV_CHANGENAME:
213 if (!strcmp(dev->name, c->ifname)) {
214 c->ifindex = dev->ifindex;
215 dev_mc_add(dev, c->clustermac);
216 } else if (dev->ifindex == c->ifindex) {
217 dev_mc_del(dev, c->clustermac);
223 spin_unlock_bh(&cn->lock);
228 static struct clusterip_config *
229 clusterip_config_init(struct net *net, const struct ipt_clusterip_tgt_info *i,
230 __be32 ip, const char *iniface)
232 struct clusterip_net *cn = clusterip_pernet(net);
233 struct clusterip_config *c;
234 struct net_device *dev;
237 if (iniface[0] == '\0') {
238 pr_info("Please specify an interface name\n");
239 return ERR_PTR(-EINVAL);
242 c = kzalloc(sizeof(*c), GFP_ATOMIC);
244 return ERR_PTR(-ENOMEM);
246 dev = dev_get_by_name(net, iniface);
248 pr_info("no such interface %s\n", iniface);
250 return ERR_PTR(-ENOENT);
252 c->ifindex = dev->ifindex;
253 strcpy(c->ifname, dev->name);
254 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
255 dev_mc_add(dev, c->clustermac);
259 c->num_total_nodes = i->num_total_nodes;
260 clusterip_config_init_nodelist(c, i);
261 c->hash_mode = i->hash_mode;
262 c->hash_initval = i->hash_initval;
264 refcount_set(&c->refcount, 1);
266 spin_lock_bh(&cn->lock);
267 if (__clusterip_config_find(net, ip)) {
272 list_add_rcu(&c->list, &cn->configs);
273 spin_unlock_bh(&cn->lock);
275 #ifdef CONFIG_PROC_FS
279 /* create proc dir entry */
280 sprintf(buffer, "%pI4", &ip);
281 c->pde = proc_create_data(buffer, 0600,
283 &clusterip_proc_fops, c);
291 refcount_set(&c->entries, 1);
294 #ifdef CONFIG_PROC_FS
297 spin_lock_bh(&cn->lock);
298 list_del_rcu(&c->list);
300 spin_unlock_bh(&cn->lock);
301 clusterip_config_put(c);
305 #ifdef CONFIG_PROC_FS
307 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
311 nodenum > c->num_total_nodes)
314 /* check if we already have this number in our bitfield */
315 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
322 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
325 nodenum > c->num_total_nodes)
328 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
335 static inline u_int32_t
336 clusterip_hashfn(const struct sk_buff *skb,
337 const struct clusterip_config *config)
339 const struct iphdr *iph = ip_hdr(skb);
340 unsigned long hashval;
341 u_int16_t sport = 0, dport = 0;
344 poff = proto_ports_offset(iph->protocol);
346 const u_int16_t *ports;
349 ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports);
355 net_info_ratelimited("unknown protocol %u\n", iph->protocol);
358 switch (config->hash_mode) {
359 case CLUSTERIP_HASHMODE_SIP:
360 hashval = jhash_1word(ntohl(iph->saddr),
361 config->hash_initval);
363 case CLUSTERIP_HASHMODE_SIP_SPT:
364 hashval = jhash_2words(ntohl(iph->saddr), sport,
365 config->hash_initval);
367 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
368 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
369 config->hash_initval);
372 /* to make gcc happy */
374 /* This cannot happen, unless the check function wasn't called
375 * at rule load time */
376 pr_info("unknown mode %u\n", config->hash_mode);
381 /* node numbers are 1..n, not 0..n */
382 return reciprocal_scale(hashval, config->num_total_nodes) + 1;
386 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
388 return test_bit(hash - 1, &config->local_nodes);
391 /***********************************************************************
393 ***********************************************************************/
396 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
398 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
400 enum ip_conntrack_info ctinfo;
403 /* don't need to clusterip_config_get() here, since refcount
404 * is only decremented by destroy() - and ip_tables guarantees
405 * that the ->target() function isn't called after ->destroy() */
407 ct = nf_ct_get(skb, &ctinfo);
411 /* special case: ICMP error handling. conntrack distinguishes between
412 * error messages (RELATED) and information requests (see below) */
413 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
414 (ctinfo == IP_CT_RELATED ||
415 ctinfo == IP_CT_RELATED_REPLY))
418 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
419 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
420 * on, which all have an ID field [relevant for hashing]. */
422 hash = clusterip_hashfn(skb, cipinfo->config);
429 case IP_CT_RELATED_REPLY:
430 /* FIXME: we don't handle expectations at the moment.
431 * They can arrive on a different node than
432 * the master connection (e.g. FTP passive mode) */
433 case IP_CT_ESTABLISHED:
434 case IP_CT_ESTABLISHED_REPLY:
436 default: /* Prevent gcc warnings */
441 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
443 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
444 if (!clusterip_responsible(cipinfo->config, hash)) {
445 pr_debug("not responsible\n");
448 pr_debug("responsible\n");
450 /* despite being received via linklayer multicast, this is
451 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
452 skb->pkt_type = PACKET_HOST;
457 static int clusterip_tg_check(const struct xt_tgchk_param *par)
459 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
460 const struct ipt_entry *e = par->entryinfo;
461 struct clusterip_config *config;
464 if (par->nft_compat) {
465 pr_err("cannot use CLUSTERIP target from nftables compat\n");
469 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
470 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
471 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
472 pr_info("unknown mode %u\n", cipinfo->hash_mode);
476 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
477 e->ip.dst.s_addr == 0) {
478 pr_info("Please specify destination IP\n");
481 if (cipinfo->num_local_nodes > ARRAY_SIZE(cipinfo->local_nodes)) {
482 pr_info("bad num_local_nodes %u\n", cipinfo->num_local_nodes);
485 for (i = 0; i < cipinfo->num_local_nodes; i++) {
486 if (cipinfo->local_nodes[i] - 1 >=
487 sizeof(config->local_nodes) * 8) {
488 pr_info("bad local_nodes[%d] %u\n",
489 i, cipinfo->local_nodes[i]);
494 config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1);
496 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
497 pr_info("no config found for %pI4, need 'new'\n",
501 config = clusterip_config_init(par->net, cipinfo,
505 return PTR_ERR(config);
509 ret = nf_ct_netns_get(par->net, par->family);
511 pr_info("cannot load conntrack support for proto=%u\n",
513 clusterip_config_entry_put(config);
514 clusterip_config_put(config);
518 if (!par->net->xt.clusterip_deprecated_warning) {
519 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
520 "use xt_cluster instead\n");
521 par->net->xt.clusterip_deprecated_warning = true;
524 cipinfo->config = config;
528 /* drop reference count of cluster config when rule is deleted */
529 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
531 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
533 /* if no more entries are referencing the config, remove it
534 * from the list and destroy the proc entry */
535 clusterip_config_entry_put(cipinfo->config);
537 clusterip_config_put(cipinfo->config);
539 nf_ct_netns_put(par->net, par->family);
543 struct compat_ipt_clusterip_tgt_info
546 u_int8_t clustermac[6];
547 u_int16_t num_total_nodes;
548 u_int16_t num_local_nodes;
549 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
551 u_int32_t hash_initval;
552 compat_uptr_t config;
554 #endif /* CONFIG_COMPAT */
556 static struct xt_target clusterip_tg_reg __read_mostly = {
558 .family = NFPROTO_IPV4,
559 .target = clusterip_tg,
560 .checkentry = clusterip_tg_check,
561 .destroy = clusterip_tg_destroy,
562 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
563 .usersize = offsetof(struct ipt_clusterip_tgt_info, config),
565 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
566 #endif /* CONFIG_COMPAT */
571 /***********************************************************************
573 ***********************************************************************/
575 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
577 u_int8_t src_hw[ETH_ALEN];
579 u_int8_t dst_hw[ETH_ALEN];
584 static void arp_print(struct arp_payload *payload)
586 #define HBUFFERLEN 30
587 char hbuffer[HBUFFERLEN];
590 for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) {
591 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
592 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
597 pr_debug("src %pI4@%s, dst %pI4\n",
598 &payload->src_ip, hbuffer, &payload->dst_ip);
603 arp_mangle(void *priv,
605 const struct nf_hook_state *state)
607 struct arphdr *arp = arp_hdr(skb);
608 struct arp_payload *payload;
609 struct clusterip_config *c;
610 struct net *net = state->net;
612 /* we don't care about non-ethernet and non-ipv4 ARP */
613 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
614 arp->ar_pro != htons(ETH_P_IP) ||
615 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
618 /* we only want to mangle arp requests and replies */
619 if (arp->ar_op != htons(ARPOP_REPLY) &&
620 arp->ar_op != htons(ARPOP_REQUEST))
623 payload = (void *)(arp+1);
625 /* if there is no clusterip configuration for the arp reply's
626 * source ip, we don't want to mangle it */
627 c = clusterip_config_find_get(net, payload->src_ip, 0);
631 /* normally the linux kernel always replies to arp queries of
632 * addresses on different interfacs. However, in the CLUSTERIP case
633 * this wouldn't work, since we didn't subscribe the mcast group on
634 * other interfaces */
635 if (c->ifindex != state->out->ifindex) {
636 pr_debug("not mangling arp reply on different interface: cip'%d'-skb'%d'\n",
637 c->ifindex, state->out->ifindex);
638 clusterip_config_put(c);
642 /* mangle reply hardware address */
643 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
646 pr_debug("mangled arp reply: ");
650 clusterip_config_put(c);
655 static const struct nf_hook_ops cip_arp_ops = {
658 .hooknum = NF_ARP_OUT,
662 /***********************************************************************
664 ***********************************************************************/
666 #ifdef CONFIG_PROC_FS
668 struct clusterip_seq_position {
669 unsigned int pos; /* position */
670 unsigned int weight; /* number of bits set == size */
671 unsigned int bit; /* current bit */
672 unsigned long val; /* current value */
675 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
677 struct clusterip_config *c = s->private;
679 u_int32_t local_nodes;
680 struct clusterip_seq_position *idx;
682 /* FIXME: possible race */
683 local_nodes = c->local_nodes;
684 weight = hweight32(local_nodes);
688 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
690 return ERR_PTR(-ENOMEM);
693 idx->weight = weight;
694 idx->bit = ffs(local_nodes);
695 idx->val = local_nodes;
696 clear_bit(idx->bit - 1, &idx->val);
701 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
703 struct clusterip_seq_position *idx = v;
706 if (*pos >= idx->weight) {
710 idx->bit = ffs(idx->val);
711 clear_bit(idx->bit - 1, &idx->val);
715 static void clusterip_seq_stop(struct seq_file *s, void *v)
721 static int clusterip_seq_show(struct seq_file *s, void *v)
723 struct clusterip_seq_position *idx = v;
728 seq_printf(s, "%u", idx->bit);
730 if (idx->pos == idx->weight - 1)
736 static const struct seq_operations clusterip_seq_ops = {
737 .start = clusterip_seq_start,
738 .next = clusterip_seq_next,
739 .stop = clusterip_seq_stop,
740 .show = clusterip_seq_show,
743 static int clusterip_proc_open(struct inode *inode, struct file *file)
745 int ret = seq_open(file, &clusterip_seq_ops);
748 struct seq_file *sf = file->private_data;
749 struct clusterip_config *c = PDE_DATA(inode);
753 clusterip_config_get(c);
759 static int clusterip_proc_release(struct inode *inode, struct file *file)
761 struct clusterip_config *c = PDE_DATA(inode);
764 ret = seq_release(inode, file);
767 clusterip_config_put(c);
772 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
773 size_t size, loff_t *ofs)
775 struct clusterip_config *c = PDE_DATA(file_inode(file));
776 #define PROC_WRITELEN 10
777 char buffer[PROC_WRITELEN+1];
778 unsigned long nodenum;
781 if (size > PROC_WRITELEN)
783 if (copy_from_user(buffer, input, size))
787 if (*buffer == '+') {
788 rc = kstrtoul(buffer+1, 10, &nodenum);
791 if (clusterip_add_node(c, nodenum))
793 } else if (*buffer == '-') {
794 rc = kstrtoul(buffer+1, 10, &nodenum);
797 if (clusterip_del_node(c, nodenum))
805 static const struct file_operations clusterip_proc_fops = {
806 .open = clusterip_proc_open,
808 .write = clusterip_proc_write,
810 .release = clusterip_proc_release,
813 #endif /* CONFIG_PROC_FS */
815 static int clusterip_net_init(struct net *net)
817 struct clusterip_net *cn = clusterip_pernet(net);
820 INIT_LIST_HEAD(&cn->configs);
822 spin_lock_init(&cn->lock);
824 ret = nf_register_net_hook(net, &cip_arp_ops);
828 #ifdef CONFIG_PROC_FS
829 cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net);
831 nf_unregister_net_hook(net, &cip_arp_ops);
832 pr_err("Unable to proc dir entry\n");
835 #endif /* CONFIG_PROC_FS */
840 static void clusterip_net_exit(struct net *net)
842 struct clusterip_net *cn = clusterip_pernet(net);
843 #ifdef CONFIG_PROC_FS
844 proc_remove(cn->procdir);
847 nf_unregister_net_hook(net, &cip_arp_ops);
850 static struct pernet_operations clusterip_net_ops = {
851 .init = clusterip_net_init,
852 .exit = clusterip_net_exit,
853 .id = &clusterip_net_id,
854 .size = sizeof(struct clusterip_net),
857 struct notifier_block cip_netdev_notifier = {
858 .notifier_call = clusterip_netdev_event
861 static int __init clusterip_tg_init(void)
865 ret = register_pernet_subsys(&clusterip_net_ops);
869 ret = xt_register_target(&clusterip_tg_reg);
873 ret = register_netdevice_notifier(&cip_netdev_notifier);
875 goto unregister_target;
877 pr_info("ClusterIP Version %s loaded successfully\n",
883 xt_unregister_target(&clusterip_tg_reg);
885 unregister_pernet_subsys(&clusterip_net_ops);
889 static void __exit clusterip_tg_exit(void)
891 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
893 unregister_netdevice_notifier(&cip_netdev_notifier);
894 xt_unregister_target(&clusterip_tg_reg);
895 unregister_pernet_subsys(&clusterip_net_ops);
897 /* Wait for completion of call_rcu()'s (clusterip_config_rcu_free) */
901 module_init(clusterip_tg_init);
902 module_exit(clusterip_tg_exit);