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/netns/generic.h>
32 #include <net/checksum.h>
35 #define CLUSTERIP_VERSION "0.8"
37 MODULE_LICENSE("GPL");
38 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
39 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
41 struct clusterip_config {
42 struct list_head list; /* list of all configs */
43 atomic_t refcount; /* reference count */
44 atomic_t entries; /* number of entries/rules
47 __be32 clusterip; /* the IP address */
48 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
49 struct net_device *dev; /* device */
50 u_int16_t num_total_nodes; /* total number of nodes */
51 unsigned long local_nodes; /* node number array */
54 struct proc_dir_entry *pde; /* proc dir entry */
56 enum clusterip_hashmode hash_mode; /* which hashing mode */
57 u_int32_t hash_initval; /* hash initialization */
62 static const struct file_operations clusterip_proc_fops;
65 static unsigned int clusterip_net_id __read_mostly;
67 struct clusterip_net {
68 struct list_head configs;
69 /* lock protects the configs list */
73 struct proc_dir_entry *procdir;
78 clusterip_config_get(struct clusterip_config *c)
80 atomic_inc(&c->refcount);
84 static void clusterip_config_rcu_free(struct rcu_head *head)
86 kfree(container_of(head, struct clusterip_config, rcu));
90 clusterip_config_put(struct clusterip_config *c)
92 if (atomic_dec_and_test(&c->refcount))
93 call_rcu_bh(&c->rcu, clusterip_config_rcu_free);
96 /* decrease the count of entries using/referencing this config. If last
97 * entry(rule) is removed, remove the config from lists, but don't free it
98 * yet, since proc-files could still be holding references */
100 clusterip_config_entry_put(struct clusterip_config *c)
102 struct net *net = dev_net(c->dev);
103 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
106 if (atomic_dec_and_lock(&c->entries, &cn->lock)) {
107 list_del_rcu(&c->list);
108 spin_unlock(&cn->lock);
111 dev_mc_del(c->dev, c->clustermac);
114 /* In case anyone still accesses the file, the open/close
115 * functions are also incrementing the refcount on their own,
116 * so it's safe to remove the entry even if it's in use. */
117 #ifdef CONFIG_PROC_FS
125 static struct clusterip_config *
126 __clusterip_config_find(struct net *net, __be32 clusterip)
128 struct clusterip_config *c;
129 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
131 list_for_each_entry_rcu(c, &cn->configs, list) {
132 if (c->clusterip == clusterip)
139 static inline struct clusterip_config *
140 clusterip_config_find_get(struct net *net, __be32 clusterip, int entry)
142 struct clusterip_config *c;
145 c = __clusterip_config_find(net, clusterip);
147 if (!c->pde || unlikely(!atomic_inc_not_zero(&c->refcount)))
150 atomic_inc(&c->entries);
152 rcu_read_unlock_bh();
158 clusterip_config_init_nodelist(struct clusterip_config *c,
159 const struct ipt_clusterip_tgt_info *i)
163 for (n = 0; n < i->num_local_nodes; n++)
164 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
167 static struct clusterip_config *
168 clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip,
169 struct net_device *dev)
171 struct net *net = dev_net(dev);
172 struct clusterip_config *c;
173 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
175 c = kzalloc(sizeof(*c), GFP_ATOMIC);
177 return ERR_PTR(-ENOMEM);
181 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
182 c->num_total_nodes = i->num_total_nodes;
183 clusterip_config_init_nodelist(c, i);
184 c->hash_mode = i->hash_mode;
185 c->hash_initval = i->hash_initval;
186 atomic_set(&c->refcount, 1);
187 atomic_set(&c->entries, 1);
189 spin_lock_bh(&cn->lock);
190 if (__clusterip_config_find(net, ip)) {
191 spin_unlock_bh(&cn->lock);
194 return ERR_PTR(-EBUSY);
197 list_add_rcu(&c->list, &cn->configs);
198 spin_unlock_bh(&cn->lock);
200 #ifdef CONFIG_PROC_FS
204 /* create proc dir entry */
205 sprintf(buffer, "%pI4", &ip);
206 c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR,
208 &clusterip_proc_fops, c);
210 spin_lock_bh(&cn->lock);
211 list_del_rcu(&c->list);
212 spin_unlock_bh(&cn->lock);
215 return ERR_PTR(-ENOMEM);
223 #ifdef CONFIG_PROC_FS
225 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
229 nodenum > c->num_total_nodes)
232 /* check if we already have this number in our bitfield */
233 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
240 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
243 nodenum > c->num_total_nodes)
246 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
253 static inline u_int32_t
254 clusterip_hashfn(const struct sk_buff *skb,
255 const struct clusterip_config *config)
257 const struct iphdr *iph = ip_hdr(skb);
258 unsigned long hashval;
259 u_int16_t sport = 0, dport = 0;
262 poff = proto_ports_offset(iph->protocol);
264 const u_int16_t *ports;
267 ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports);
273 net_info_ratelimited("unknown protocol %u\n", iph->protocol);
276 switch (config->hash_mode) {
277 case CLUSTERIP_HASHMODE_SIP:
278 hashval = jhash_1word(ntohl(iph->saddr),
279 config->hash_initval);
281 case CLUSTERIP_HASHMODE_SIP_SPT:
282 hashval = jhash_2words(ntohl(iph->saddr), sport,
283 config->hash_initval);
285 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
286 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
287 config->hash_initval);
290 /* to make gcc happy */
292 /* This cannot happen, unless the check function wasn't called
293 * at rule load time */
294 pr_info("unknown mode %u\n", config->hash_mode);
299 /* node numbers are 1..n, not 0..n */
300 return reciprocal_scale(hashval, config->num_total_nodes) + 1;
304 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
306 return test_bit(hash - 1, &config->local_nodes);
309 /***********************************************************************
311 ***********************************************************************/
314 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
316 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
318 enum ip_conntrack_info ctinfo;
321 /* don't need to clusterip_config_get() here, since refcount
322 * is only decremented by destroy() - and ip_tables guarantees
323 * that the ->target() function isn't called after ->destroy() */
325 ct = nf_ct_get(skb, &ctinfo);
329 /* special case: ICMP error handling. conntrack distinguishes between
330 * error messages (RELATED) and information requests (see below) */
331 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
332 (ctinfo == IP_CT_RELATED ||
333 ctinfo == IP_CT_RELATED_REPLY))
336 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
337 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
338 * on, which all have an ID field [relevant for hashing]. */
340 hash = clusterip_hashfn(skb, cipinfo->config);
347 case IP_CT_RELATED_REPLY:
348 /* FIXME: we don't handle expectations at the moment.
349 * They can arrive on a different node than
350 * the master connection (e.g. FTP passive mode) */
351 case IP_CT_ESTABLISHED:
352 case IP_CT_ESTABLISHED_REPLY:
354 default: /* Prevent gcc warnings */
359 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
361 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
362 if (!clusterip_responsible(cipinfo->config, hash)) {
363 pr_debug("not responsible\n");
366 pr_debug("responsible\n");
368 /* despite being received via linklayer multicast, this is
369 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
370 skb->pkt_type = PACKET_HOST;
375 static int clusterip_tg_check(const struct xt_tgchk_param *par)
377 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
378 const struct ipt_entry *e = par->entryinfo;
379 struct clusterip_config *config;
382 if (par->nft_compat) {
383 pr_err("cannot use CLUSTERIP target from nftables compat\n");
387 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
388 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
389 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
390 pr_info("unknown mode %u\n", cipinfo->hash_mode);
394 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
395 e->ip.dst.s_addr == 0) {
396 pr_info("Please specify destination IP\n");
400 /* FIXME: further sanity checks */
402 config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1);
404 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
405 pr_info("no config found for %pI4, need 'new'\n",
409 struct net_device *dev;
411 if (e->ip.iniface[0] == '\0') {
412 pr_info("Please specify an interface name\n");
416 dev = dev_get_by_name(par->net, e->ip.iniface);
418 pr_info("no such interface %s\n",
423 config = clusterip_config_init(cipinfo,
424 e->ip.dst.s_addr, dev);
425 if (IS_ERR(config)) {
427 return PTR_ERR(config);
429 dev_mc_add(config->dev, config->clustermac);
432 cipinfo->config = config;
434 ret = nf_ct_netns_get(par->net, par->family);
436 pr_info("cannot load conntrack support for proto=%u\n",
439 if (!par->net->xt.clusterip_deprecated_warning) {
440 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
441 "use xt_cluster instead\n");
442 par->net->xt.clusterip_deprecated_warning = true;
448 /* drop reference count of cluster config when rule is deleted */
449 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
451 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
453 /* if no more entries are referencing the config, remove it
454 * from the list and destroy the proc entry */
455 clusterip_config_entry_put(cipinfo->config);
457 clusterip_config_put(cipinfo->config);
459 nf_ct_netns_get(par->net, par->family);
463 struct compat_ipt_clusterip_tgt_info
466 u_int8_t clustermac[6];
467 u_int16_t num_total_nodes;
468 u_int16_t num_local_nodes;
469 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
471 u_int32_t hash_initval;
472 compat_uptr_t config;
474 #endif /* CONFIG_COMPAT */
476 static struct xt_target clusterip_tg_reg __read_mostly = {
478 .family = NFPROTO_IPV4,
479 .target = clusterip_tg,
480 .checkentry = clusterip_tg_check,
481 .destroy = clusterip_tg_destroy,
482 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
484 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
485 #endif /* CONFIG_COMPAT */
490 /***********************************************************************
492 ***********************************************************************/
494 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
496 u_int8_t src_hw[ETH_ALEN];
498 u_int8_t dst_hw[ETH_ALEN];
503 static void arp_print(struct arp_payload *payload)
505 #define HBUFFERLEN 30
506 char hbuffer[HBUFFERLEN];
509 for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) {
510 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
511 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
516 pr_debug("src %pI4@%s, dst %pI4\n",
517 &payload->src_ip, hbuffer, &payload->dst_ip);
522 arp_mangle(void *priv,
524 const struct nf_hook_state *state)
526 struct arphdr *arp = arp_hdr(skb);
527 struct arp_payload *payload;
528 struct clusterip_config *c;
529 struct net *net = state->net;
531 /* we don't care about non-ethernet and non-ipv4 ARP */
532 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
533 arp->ar_pro != htons(ETH_P_IP) ||
534 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
537 /* we only want to mangle arp requests and replies */
538 if (arp->ar_op != htons(ARPOP_REPLY) &&
539 arp->ar_op != htons(ARPOP_REQUEST))
542 payload = (void *)(arp+1);
544 /* if there is no clusterip configuration for the arp reply's
545 * source ip, we don't want to mangle it */
546 c = clusterip_config_find_get(net, payload->src_ip, 0);
550 /* normally the linux kernel always replies to arp queries of
551 * addresses on different interfacs. However, in the CLUSTERIP case
552 * this wouldn't work, since we didn't subscribe the mcast group on
553 * other interfaces */
554 if (c->dev != state->out) {
555 pr_debug("not mangling arp reply on different "
556 "interface: cip'%s'-skb'%s'\n",
557 c->dev->name, state->out->name);
558 clusterip_config_put(c);
562 /* mangle reply hardware address */
563 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
566 pr_debug("mangled arp reply: ");
570 clusterip_config_put(c);
575 static struct nf_hook_ops cip_arp_ops __read_mostly = {
578 .hooknum = NF_ARP_OUT,
582 /***********************************************************************
584 ***********************************************************************/
586 #ifdef CONFIG_PROC_FS
588 struct clusterip_seq_position {
589 unsigned int pos; /* position */
590 unsigned int weight; /* number of bits set == size */
591 unsigned int bit; /* current bit */
592 unsigned long val; /* current value */
595 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
597 struct clusterip_config *c = s->private;
599 u_int32_t local_nodes;
600 struct clusterip_seq_position *idx;
602 /* FIXME: possible race */
603 local_nodes = c->local_nodes;
604 weight = hweight32(local_nodes);
608 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
610 return ERR_PTR(-ENOMEM);
613 idx->weight = weight;
614 idx->bit = ffs(local_nodes);
615 idx->val = local_nodes;
616 clear_bit(idx->bit - 1, &idx->val);
621 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
623 struct clusterip_seq_position *idx = v;
626 if (*pos >= idx->weight) {
630 idx->bit = ffs(idx->val);
631 clear_bit(idx->bit - 1, &idx->val);
635 static void clusterip_seq_stop(struct seq_file *s, void *v)
641 static int clusterip_seq_show(struct seq_file *s, void *v)
643 struct clusterip_seq_position *idx = v;
648 seq_printf(s, "%u", idx->bit);
650 if (idx->pos == idx->weight - 1)
656 static const struct seq_operations clusterip_seq_ops = {
657 .start = clusterip_seq_start,
658 .next = clusterip_seq_next,
659 .stop = clusterip_seq_stop,
660 .show = clusterip_seq_show,
663 static int clusterip_proc_open(struct inode *inode, struct file *file)
665 int ret = seq_open(file, &clusterip_seq_ops);
668 struct seq_file *sf = file->private_data;
669 struct clusterip_config *c = PDE_DATA(inode);
673 clusterip_config_get(c);
679 static int clusterip_proc_release(struct inode *inode, struct file *file)
681 struct clusterip_config *c = PDE_DATA(inode);
684 ret = seq_release(inode, file);
687 clusterip_config_put(c);
692 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
693 size_t size, loff_t *ofs)
695 struct clusterip_config *c = PDE_DATA(file_inode(file));
696 #define PROC_WRITELEN 10
697 char buffer[PROC_WRITELEN+1];
698 unsigned long nodenum;
701 if (size > PROC_WRITELEN)
703 if (copy_from_user(buffer, input, size))
707 if (*buffer == '+') {
708 rc = kstrtoul(buffer+1, 10, &nodenum);
711 if (clusterip_add_node(c, nodenum))
713 } else if (*buffer == '-') {
714 rc = kstrtoul(buffer+1, 10, &nodenum);
717 if (clusterip_del_node(c, nodenum))
725 static const struct file_operations clusterip_proc_fops = {
726 .owner = THIS_MODULE,
727 .open = clusterip_proc_open,
729 .write = clusterip_proc_write,
731 .release = clusterip_proc_release,
734 #endif /* CONFIG_PROC_FS */
736 static int clusterip_net_init(struct net *net)
738 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
740 INIT_LIST_HEAD(&cn->configs);
742 spin_lock_init(&cn->lock);
744 #ifdef CONFIG_PROC_FS
745 cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net);
747 pr_err("Unable to proc dir entry\n");
750 #endif /* CONFIG_PROC_FS */
755 static void clusterip_net_exit(struct net *net)
757 #ifdef CONFIG_PROC_FS
758 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
759 proc_remove(cn->procdir);
763 static struct pernet_operations clusterip_net_ops = {
764 .init = clusterip_net_init,
765 .exit = clusterip_net_exit,
766 .id = &clusterip_net_id,
767 .size = sizeof(struct clusterip_net),
770 static int __init clusterip_tg_init(void)
774 ret = register_pernet_subsys(&clusterip_net_ops);
778 ret = xt_register_target(&clusterip_tg_reg);
782 ret = nf_register_hook(&cip_arp_ops);
786 pr_info("ClusterIP Version %s loaded successfully\n",
792 xt_unregister_target(&clusterip_tg_reg);
794 unregister_pernet_subsys(&clusterip_net_ops);
798 static void __exit clusterip_tg_exit(void)
800 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
802 nf_unregister_hook(&cip_arp_ops);
803 xt_unregister_target(&clusterip_tg_reg);
804 unregister_pernet_subsys(&clusterip_net_ops);
806 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
810 module_init(clusterip_tg_init);
811 module_exit(clusterip_tg_exit);