ipt_clusterip.c

linux-2.6.15.6

/* Cluster IP hashmark target 
 * (C) 2003-2004 by Harald Welte <[e-mail]>
 * based on ideas of Fabio Olive Leite <[e-mail]>
 *
 * Development of this code funded by SuSE Linux AG, http://www.suse.com/
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 */
#include <linux/module.h>
#include <linux/config.h>
#include <linux/proc_fs.h>
#include <linux/jhash.h>
#include <linux/bitops.h>
#include <linux/skbuff.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/udp.h>
#include <linux/icmp.h>
#include <linux/if_arp.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>

#include <net/checksum.h>

#include <linux/netfilter_arp.h>

#include <linux/netfilter_ipv4/ip_tables.h>
#include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
#include <net/netfilter/nf_conntrack_compat.h>

#define CLUSTERIP_VERSION "0.8"

#define DEBUG_CLUSTERIP

#ifdef DEBUG_CLUSTERIP
#define DEBUGP	printk
#else
#define DEBUGP
#endif

#define ASSERT_READ_LOCK(x)

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Harald Welte <[e-mail]>");
MODULE_DESCRIPTION("iptables target for CLUSTERIP");

struct clusterip_config {
	struct list_head list;			/* list of all configs */
	atomic_t refcount;			/* reference count */
	atomic_t entries;			/* number of entries/rules
						 * referencing us */

	u_int32_t clusterip;			/* the IP address */
	u_int8_t clustermac[ETH_ALEN];		/* the MAC address */
	struct net_device *dev;			/* device */
	u_int16_t num_total_nodes;		/* total number of nodes */
	unsigned long local_nodes;		/* node number array */

#ifdef CONFIG_PROC_FS
	struct proc_dir_entry *pde;		/* proc dir entry */
#endif
	enum clusterip_hashmode hash_mode;	/* which hashing mode */
	u_int32_t hash_initval;			/* hash initialization */
};

static LIST_HEAD(clusterip_configs);

/* clusterip_lock protects the clusterip_configs list */
static DEFINE_RWLOCK(clusterip_lock);

#ifdef CONFIG_PROC_FS
static struct file_operations clusterip_proc_fops;
static struct proc_dir_entry *clusterip_procdir;
#endif

static inline void
clusterip_config_get(struct clusterip_config *c)
{
	atomic_inc(&c->refcount);
}

static inline void
clusterip_config_put(struct clusterip_config *c)
{
	if (atomic_dec_and_test(&c->refcount))
		kfree(c);
}

/* increase the count of entries(rules) using/referencing this config */
static inline void
clusterip_config_entry_get(struct clusterip_config *c)
{
	atomic_inc(&c->entries);
}

/* decrease the count of entries using/referencing this config.  If last
 * entry(rule) is removed, remove the config from lists, but don't free it
 * yet, since proc-files could still be holding references */
static inline void
clusterip_config_entry_put(struct clusterip_config *c)
{
	if (atomic_dec_and_test(&c->entries)) {
		write_lock_bh(&clusterip_lock);
		list_del(&c->list);
		write_unlock_bh(&clusterip_lock);

		dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
		dev_put(c->dev);

		/* In case anyone still accesses the file, the open/close
		 * functions are also incrementing the refcount on their own,
		 * so it's safe to remove the entry even if it's in use. */
#ifdef CONFIG_PROC_FS
		remove_proc_entry(c->pde->name, c->pde->parent);
#endif
	}
}

static struct clusterip_config *
__clusterip_config_find(u_int32_t clusterip)
{
	struct list_head *pos;

	ASSERT_READ_LOCK(&clusterip_lock);
	list_for_each(pos, &clusterip_configs) {
		struct clusterip_config *c = list_entry(pos, 
					struct clusterip_config, list);
		if (c->clusterip == clusterip) {
			return c;
		}
	}

	return NULL;
}

static inline struct clusterip_config *
clusterip_config_find_get(u_int32_t clusterip, int entry)
{
	struct clusterip_config *c;

	read_lock_bh(&clusterip_lock);
	c = __clusterip_config_find(clusterip);
	if (!c) {
		read_unlock_bh(&clusterip_lock);
		return NULL;
	}
	atomic_inc(&c->refcount);
	if (entry)
		atomic_inc(&c->entries);
	read_unlock_bh(&clusterip_lock);

	return c;
}

static void
clusterip_config_init_nodelist(struct clusterip_config *c,
			       const struct ipt_clusterip_tgt_info *i)
{
	int n;

	for (n = 0; n < i->num_local_nodes; n++) {
		set_bit(i->local_nodes[n] - 1, &c->local_nodes);
	}
}

static struct clusterip_config *
clusterip_config_init(struct ipt_clusterip_tgt_info *i, u_int32_t ip,
			struct net_device *dev)
{
	struct clusterip_config *c;
	char buffer[16];

	c = kmalloc(sizeof(*c), GFP_ATOMIC);
	if (!c)
		return NULL;

	memset(c, 0, sizeof(*c));
	c->dev = dev;
	c->clusterip = ip;
	memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
	c->num_total_nodes = i->num_total_nodes;
	clusterip_config_init_nodelist(c, i);
	c->hash_mode = i->hash_mode;
	c->hash_initval = i->hash_initval;
	atomic_set(&c->refcount, 1);
	atomic_set(&c->entries, 1);

#ifdef CONFIG_PROC_FS
	/* create proc dir entry */
	sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
	c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR, clusterip_procdir);
	if (!c->pde) {
		kfree(c);
		return NULL;
	}
	c->pde->proc_fops = &clusterip_proc_fops;
	c->pde->data = c;
#endif

	write_lock_bh(&clusterip_lock);
	list_add(&c->list, &clusterip_configs);
	write_unlock_bh(&clusterip_lock);

	return c;
}

static int
clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
{

	if (nodenum == 0 ||
	    nodenum > c->num_total_nodes)
		return 1;

	/* check if we already have this number in our bitfield */
	if (test_and_set_bit(nodenum - 1, &c->local_nodes))
		return 1;

	return 0;
}

static int
clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
{
	if (nodenum == 0 ||
	    nodenum > c->num_total_nodes)
		return 1;
		
	if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
		return 0;

	return 1;
}

static inline u_int32_t
clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config)
{
	struct iphdr *iph = skb->nh.iph;
	unsigned long hashval;
	u_int16_t sport, dport;
	struct tcphdr *th;
	struct udphdr *uh;
	struct icmphdr *ih;

	switch (iph->protocol) {
	case IPPROTO_TCP:
		th = (void *)iph+iph->ihl*4;
		sport = ntohs(th->source);
		dport = ntohs(th->dest);
		break;
	case IPPROTO_UDP:
		uh = (void *)iph+iph->ihl*4;
		sport = ntohs(uh->source);
		dport = ntohs(uh->dest);
		break;
	case IPPROTO_ICMP:
		ih = (void *)iph+iph->ihl*4;
		sport = ntohs(ih->un.echo.id);
		dport = (ih->type<<8)|ih->code;
		break;
	default:
		if (net_ratelimit()) {
			printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
				iph->protocol);
		}
		sport = dport = 0;
	}

	switch (config->hash_mode) {
	case CLUSTERIP_HASHMODE_SIP:
		hashval = jhash_1word(ntohl(iph->saddr),
				      config->hash_initval);
		break;
	case CLUSTERIP_HASHMODE_SIP_SPT:
		hashval = jhash_2words(ntohl(iph->saddr), sport, 
				       config->hash_initval);
		break;
	case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
		hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
				       config->hash_initval);
		break;
	default:
		/* to make gcc happy */
		hashval = 0;
		/* This cannot happen, unless the check function wasn't called
		 * at rule load time */
		printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
		BUG();
		break;
	}

	/* node numbers are 1..n, not 0..n */
	return ((hashval % config->num_total_nodes)+1);
}

static inline int
clusterip_responsible(struct clusterip_config *config, u_int32_t hash)
{
	return test_bit(hash - 1, &config->local_nodes);
}

/*********************************************************************** 
 * IPTABLES TARGET 
 ***********************************************************************/

static unsigned int
target(struct sk_buff **pskb,
       const struct net_device *in,
       const struct net_device *out,
       unsigned int hooknum,
       const void *targinfo,
       void *userinfo)
{
	const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
	enum ip_conntrack_info ctinfo;
	u_int32_t *mark, hash;

	/* don't need to clusterip_config_get() here, since refcount
	 * is only decremented by destroy() - and ip_tables guarantees
	 * that the ->target() function isn't called after ->destroy() */

	mark = nf_ct_get_mark((*pskb), &ctinfo);
	if (mark == NULL) {
		printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
			/* FIXME: need to drop invalid ones, since replies
			 * to outgoing connections of other nodes will be 
			 * marked as INVALID */
		return NF_DROP;
	}

	/* special case: ICMP error handling. conntrack distinguishes between
	 * error messages (RELATED) and information requests (see below) */
	if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP
	    && (ctinfo == IP_CT_RELATED 
		|| ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
		return IPT_CONTINUE;

	/* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO, 
	 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
	 * on, which all have an ID field [relevant for hashing]. */

	hash = clusterip_hashfn(*pskb, cipinfo->config);

	switch (ctinfo) {
		case IP_CT_NEW:
			*mark = hash;
			break;
		case IP_CT_RELATED:
		case IP_CT_RELATED+IP_CT_IS_REPLY:
			/* FIXME: we don't handle expectations at the
			 * moment.  they can arrive on a different node than
			 * the master connection (e.g. FTP passive mode) */
		case IP_CT_ESTABLISHED:
		case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
			break;
		default:
			break;
	}

#ifdef DEBUG_CLUSTERP
	DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
#endif
	DEBUGP("hash=%u ct_hash=%u ", hash, *mark);
	if (!clusterip_responsible(cipinfo->config, hash)) {
		DEBUGP("not responsible\n");
		return NF_DROP;
	}
	DEBUGP("responsible\n");

	/* despite being received via linklayer multicast, this is
	 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
	(*pskb)->pkt_type = PACKET_HOST;

	return IPT_CONTINUE;
}

static int
checkentry(const char *tablename,
	   const struct ipt_entry *e,
           void *targinfo,
           unsigned int targinfosize,
           unsigned int hook_mask)
{
	struct ipt_clusterip_tgt_info *cipinfo = targinfo;

	struct clusterip_config *config;

	if (targinfosize != IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info))) {
		printk(KERN_WARNING "CLUSTERIP: targinfosize %u != %Zu\n",
		       targinfosize,
		       IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info)));