[net-next-2.6.git] / net / ipv4 / ipcomp.c

/*
 * IP Payload Compression Protocol (IPComp) - RFC3173.
 *
 * Copyright (c) 2003 James Morris <jmorris@intercode.com.au>
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the Free
 * Software Foundation; either version 2 of the License, or (at your option) 
 * any later version.
 *
 * Todo:
 *   - Tunable compression parameters.
 *   - Compression stats.
 *   - Adaptive compression.
 */
#include <linux/config.h>
#include <linux/module.h>
#include <asm/scatterlist.h>
#include <asm/semaphore.h>
#include <linux/crypto.h>
#include <linux/pfkeyv2.h>
#include <linux/percpu.h>
#include <linux/smp.h>
#include <linux/list.h>
#include <linux/vmalloc.h>
#include <linux/rtnetlink.h>
#include <linux/mutex.h>
#include <net/ip.h>
#include <net/xfrm.h>
#include <net/icmp.h>
#include <net/ipcomp.h>
#include <net/protocol.h>

struct ipcomp_tfms {
	struct list_head list;
	struct crypto_tfm **tfms;
	int users;
};

static DEFINE_MUTEX(ipcomp_resource_mutex);
static void **ipcomp_scratches;
static int ipcomp_scratch_users;
static LIST_HEAD(ipcomp_tfms_list);

static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
{
	int err, plen, dlen;
	struct iphdr *iph;
	struct ipcomp_data *ipcd = x->data;
	u8 *start, *scratch;
	struct crypto_tfm *tfm;
	int cpu;
	
	plen = skb->len;
	dlen = IPCOMP_SCRATCH_SIZE;
	start = skb->data;

	cpu = get_cpu();
	scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
	tfm = *per_cpu_ptr(ipcd->tfms, cpu);

	err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
	if (err)
		goto out;

	if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
		err = -EINVAL;
		goto out;
	}

	err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
	if (err)
		goto out;
		
	skb_put(skb, dlen - plen);
	memcpy(skb->data, scratch, dlen);
	iph = skb->nh.iph;
	iph->tot_len = htons(dlen + iph->ihl * 4);
out:	
	put_cpu();
	return err;
}

static int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
{
	u8 nexthdr;
	int err = 0;
	struct iphdr *iph;
	union {
		struct iphdr	iph;
		char 		buf[60];
	} tmp_iph;


	if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
	    skb_linearize(skb, GFP_ATOMIC) != 0) {
	    	err = -ENOMEM;
	    	goto out;
	}

	skb->ip_summed = CHECKSUM_NONE;

	/* Remove ipcomp header and decompress original payload */	
	iph = skb->nh.iph;
	memcpy(&tmp_iph, iph, iph->ihl * 4);
	nexthdr = *(u8 *)skb->data;
	skb_pull(skb, sizeof(struct ip_comp_hdr));
	skb->nh.raw += sizeof(struct ip_comp_hdr);
	memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
	iph = skb->nh.iph;
	iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
	iph->protocol = nexthdr;
	skb->h.raw = skb->data;
	err = ipcomp_decompress(x, skb);

out:	
	return err;
}

static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
{
	int err, plen, dlen, ihlen;
	struct iphdr *iph = skb->nh.iph;
	struct ipcomp_data *ipcd = x->data;
	u8 *start, *scratch;
	struct crypto_tfm *tfm;
	int cpu;
	
	ihlen = iph->ihl * 4;
	plen = skb->len - ihlen;
	dlen = IPCOMP_SCRATCH_SIZE;
	start = skb->data + ihlen;

	cpu = get_cpu();
	scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
	tfm = *per_cpu_ptr(ipcd->tfms, cpu);

	err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
	if (err)
		goto out;

	if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
		err = -EMSGSIZE;
		goto out;
	}
	
	memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
	put_cpu();

	pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
	return 0;
	
out:	
	put_cpu();
	return err;
}

static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
{
	int err;
	struct iphdr *iph;
	struct ip_comp_hdr *ipch;
	struct ipcomp_data *ipcd = x->data;
	int hdr_len = 0;

	iph = skb->nh.iph;
	iph->tot_len = htons(skb->len);
	hdr_len = iph->ihl * 4;
	if ((skb->len - hdr_len) < ipcd->threshold) {
		/* Don't bother compressing */
		goto out_ok;
	}

	if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
	    skb_linearize(skb, GFP_ATOMIC) != 0) {
		goto out_ok;
	}
	
	err = ipcomp_compress(x, skb);
	iph = skb->nh.iph;

	if (err) {
		goto out_ok;
	}

	/* Install ipcomp header, convert into ipcomp datagram. */
	iph->tot_len = htons(skb->len);
	ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
	ipch->nexthdr = iph->protocol;
	ipch->flags = 0;
	ipch->cpi = htons((u16 )ntohl(x->id.spi));
	iph->protocol = IPPROTO_COMP;
	ip_send_check(iph);
	return 0;

out_ok:
	if (x->props.mode)
		ip_send_check(iph);
	return 0;
}

static void ipcomp4_err(struct sk_buff *skb, u32 info)
{
	u32 spi;
	struct iphdr *iph = (struct iphdr *)skb->data;
	struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
	struct xfrm_state *x;

	if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
	    skb->h.icmph->code != ICMP_FRAG_NEEDED)
		return;

	spi = ntohl(ntohs(ipch->cpi));
	x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
	                      spi, IPPROTO_COMP, AF_INET);
	if (!x)
		return;
	NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
		 spi, NIPQUAD(iph->daddr));
	xfrm_state_put(x);
}

/* We always hold one tunnel user reference to indicate a tunnel */ 
static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
{
	struct xfrm_state *t;
	
	t = xfrm_state_alloc();
	if (t == NULL)
		goto out;

	t->id.proto = IPPROTO_IPIP;
	t->id.spi = x->props.saddr.a4;
	t->id.daddr.a4 = x->id.daddr.a4;
	memcpy(&t->sel, &x->sel, sizeof(t->sel));
	t->props.family = AF_INET;
	t->props.mode = 1;
	t->props.saddr.a4 = x->props.saddr.a4;
	t->props.flags = x->props.flags;

	if (xfrm_init_state(t))
		goto error;

	atomic_set(&t->tunnel_users, 1);
out:
	return t;

error:
	t->km.state = XFRM_STATE_DEAD;
	xfrm_state_put(t);
	t = NULL;
	goto out;
}

/*
 * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
 * always incremented on success.
 */
static int ipcomp_tunnel_attach(struct xfrm_state *x)
{
	int err = 0;
	struct xfrm_state *t;

	t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
	                      x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
	if (!t) {
		t = ipcomp_tunnel_create(x);
		if (!t) {
			err = -EINVAL;
			goto out;
		}
		xfrm_state_insert(t);
		xfrm_state_hold(t);
	}
	x->tunnel = t;
	atomic_inc(&t->tunnel_users);
out:
	return err;
}

static void ipcomp_free_scratches(void)
{
	int i;
	void **scratches;

	if (--ipcomp_scratch_users)
		return;

	scratches = ipcomp_scratches;
	if (!scratches)
		return;

	for_each_possible_cpu(i)
		vfree(*per_cpu_ptr(scratches, i));

	free_percpu(scratches);
}

static void **ipcomp_alloc_scratches(void)
{
	int i;
	void **scratches;

	if (ipcomp_scratch_users++)
		return ipcomp_scratches;

	scratches = alloc_percpu(void *);
	if (!scratches)
		return NULL;

	ipcomp_scratches = scratches;

	for_each_possible_cpu(i) {
		void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
		if (!scratch)
			return NULL;
		*per_cpu_ptr(scratches, i) = scratch;
	}

	return scratches;
}

static void ipcomp_free_tfms(struct crypto_tfm **tfms)
{
	struct ipcomp_tfms *pos;
	int cpu;

	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
		if (pos->tfms == tfms)
			break;
	}

	BUG_TRAP(pos);

	if (--pos->users)
		return;

	list_del(&pos->list);
	kfree(pos);

	if (!tfms)
		return;

	for_each_possible_cpu(cpu) {
		struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
		crypto_free_tfm(tfm);
	}
	free_percpu(tfms);
}

static struct crypto_tfm **ipcomp_alloc_tfms(const char *alg_name)
{
	struct ipcomp_tfms *pos;
	struct crypto_tfm **tfms;
	int cpu;

	/* This can be any valid CPU ID so we don't need locking. */
	cpu = raw_smp_processor_id();

	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
		struct crypto_tfm *tfm;

		tfms = pos->tfms;
		tfm = *per_cpu_ptr(tfms, cpu);

		if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
			pos->users++;
			return tfms;
		}
	}

	pos = kmalloc(sizeof(*pos), GFP_KERNEL);
	if (!pos)
		return NULL;

	pos->users = 1;
	INIT_LIST_HEAD(&pos->list);
	list_add(&pos->list, &ipcomp_tfms_list);

	pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
	if (!tfms)
		goto error;

	for_each_possible_cpu(cpu) {
		struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
		if (!tfm)
			goto error;
		*per_cpu_ptr(tfms, cpu) = tfm;
	}

	return tfms;

error:
	ipcomp_free_tfms(tfms);
	return NULL;
}

static void ipcomp_free_data(struct ipcomp_data *ipcd)
{
	if (ipcd->tfms)
		ipcomp_free_tfms(ipcd->tfms);
	ipcomp_free_scratches();
}

static void ipcomp_destroy(struct xfrm_state *x)
{
	struct ipcomp_data *ipcd = x->data;
	if (!ipcd)
		return;
	xfrm_state_delete_tunnel(x);
	mutex_lock(&ipcomp_resource_mutex);
	ipcomp_free_data(ipcd);
	mutex_unlock(&ipcomp_resource_mutex);
	kfree(ipcd);
}

static int ipcomp_init_state(struct xfrm_state *x)
{
	int err;
	struct ipcomp_data *ipcd;
	struct xfrm_algo_desc *calg_desc;

	err = -EINVAL;
	if (!x->calg)
		goto out;

	if (x->encap)
		goto out;

	err = -ENOMEM;
	ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
	if (!ipcd)
		goto out;

	memset(ipcd, 0, sizeof(*ipcd));
	x->props.header_len = 0;
	if (x->props.mode)
		x->props.header_len += sizeof(struct iphdr);

	mutex_lock(&ipcomp_resource_mutex);
	if (!ipcomp_alloc_scratches())
		goto error;

	ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
	if (!ipcd->tfms)
		goto error;
	mutex_unlock(&ipcomp_resource_mutex);

	if (x->props.mode) {
		err = ipcomp_tunnel_attach(x);
		if (err)
			goto error_tunnel;
	}

	calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
	BUG_ON(!calg_desc);
	ipcd->threshold = calg_desc->uinfo.comp.threshold;
	x->data = ipcd;
	err = 0;
out:
	return err;

error_tunnel:
	mutex_lock(&ipcomp_resource_mutex);
error:
	ipcomp_free_data(ipcd);
	mutex_unlock(&ipcomp_resource_mutex);
	kfree(ipcd);
	goto out;
}

static struct xfrm_type ipcomp_type = {
	.description	= "IPCOMP4",
	.owner		= THIS_MODULE,
	.proto	     	= IPPROTO_COMP,
	.init_state	= ipcomp_init_state,
	.destructor	= ipcomp_destroy,
	.input		= ipcomp_input,
	.output		= ipcomp_output
};

static struct net_protocol ipcomp4_protocol = {
	.handler	=	xfrm4_rcv,
	.err_handler	=	ipcomp4_err,
	.no_policy	=	1,
};

static int __init ipcomp4_init(void)
{
	if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
		printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
		return -EAGAIN;
	}
	if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
		printk(KERN_INFO "ipcomp init: can't add protocol\n");
		xfrm_unregister_type(&ipcomp_type, AF_INET);
		return -EAGAIN;
	}
	return 0;
}

static void __exit ipcomp4_fini(void)
{
	if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
		printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
	if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
		printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
}

module_init(ipcomp4_init);
module_exit(ipcomp4_fini);

MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
Commit	Line	Data
1da177e4 LT	1	/*
	2	* IP Payload Compression Protocol (IPComp) - RFC3173.
	3	*
	4	* Copyright (c) 2003 James Morris <jmorris@intercode.com.au>
	5	*
	6	* This program is free software; you can redistribute it and/or modify it
	7	* under the terms of the GNU General Public License as published by the Free
	8	* Software Foundation; either version 2 of the License, or (at your option)
	9	* any later version.
	10	*
	11	* Todo:
	12	* - Tunable compression parameters.
	13	* - Compression stats.
	14	* - Adaptive compression.
	15	*/
	16	#include <linux/config.h>
	17	#include <linux/module.h>
	18	#include <asm/scatterlist.h>
	19	#include <asm/semaphore.h>
	20	#include <linux/crypto.h>
	21	#include <linux/pfkeyv2.h>
	22	#include <linux/percpu.h>
	23	#include <linux/smp.h>
	24	#include <linux/list.h>
	25	#include <linux/vmalloc.h>
	26	#include <linux/rtnetlink.h>
4a3e2f71	27	#include <linux/mutex.h>
1da177e4 LT	28	#include <net/ip.h>
	29	#include <net/xfrm.h>
	30	#include <net/icmp.h>
	31	#include <net/ipcomp.h>
14c85021	32	#include <net/protocol.h>
1da177e4 LT	33
	34	struct ipcomp_tfms {
	35	struct list_head list;
	36	struct crypto_tfm **tfms;
	37	int users;
	38	};
	39
4a3e2f71	40	static DEFINE_MUTEX(ipcomp_resource_mutex);
1da177e4 LT	41	static void **ipcomp_scratches;
	42	static int ipcomp_scratch_users;
	43	static LIST_HEAD(ipcomp_tfms_list);
	44
	45	static int ipcomp_decompress(struct xfrm_state x, struct sk_buff skb)
	46	{
	47	int err, plen, dlen;
	48	struct iphdr *iph;
	49	struct ipcomp_data *ipcd = x->data;
	50	u8 start, scratch;
	51	struct crypto_tfm *tfm;
	52	int cpu;
	53
	54	plen = skb->len;
	55	dlen = IPCOMP_SCRATCH_SIZE;
	56	start = skb->data;
	57
	58	cpu = get_cpu();
	59	scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
	60	tfm = *per_cpu_ptr(ipcd->tfms, cpu);
	61
	62	err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
	63	if (err)
	64	goto out;
	65
	66	if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
	67	err = -EINVAL;
	68	goto out;
	69	}
	70
	71	err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
	72	if (err)
	73	goto out;
	74
	75	skb_put(skb, dlen - plen);
	76	memcpy(skb->data, scratch, dlen);
	77	iph = skb->nh.iph;
	78	iph->tot_len = htons(dlen + iph->ihl * 4);
	79	out:
	80	put_cpu();
	81	return err;
	82	}
	83
e695633e	84	static int ipcomp_input(struct xfrm_state x, struct sk_buff skb)
1da177e4 LT	85	{
	86	u8 nexthdr;
	87	int err = 0;
	88	struct iphdr *iph;
	89	union {
	90	struct iphdr iph;
	91	char buf[60];
	92	} tmp_iph;
	93
	94
	95	if ((skb_is_nonlinear(skb) \|\| skb_cloned(skb)) &&
	96	skb_linearize(skb, GFP_ATOMIC) != 0) {
	97	err = -ENOMEM;
	98	goto out;
	99	}
	100
	101	skb->ip_summed = CHECKSUM_NONE;
	102
	103	/* Remove ipcomp header and decompress original payload */
	104	iph = skb->nh.iph;
	105	memcpy(&tmp_iph, iph, iph->ihl * 4);
	106	nexthdr = (u8 )skb->data;
	107	skb_pull(skb, sizeof(struct ip_comp_hdr));
	108	skb->nh.raw += sizeof(struct ip_comp_hdr);
	109	memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
	110	iph = skb->nh.iph;
	111	iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
	112	iph->protocol = nexthdr;
	113	skb->h.raw = skb->data;
	114	err = ipcomp_decompress(x, skb);
	115
	116	out:
	117	return err;
	118	}
	119
	120	static int ipcomp_compress(struct xfrm_state x, struct sk_buff skb)
	121	{
	122	int err, plen, dlen, ihlen;
	123	struct iphdr *iph = skb->nh.iph;
	124	struct ipcomp_data *ipcd = x->data;
	125	u8 start, scratch;
	126	struct crypto_tfm *tfm;
	127	int cpu;
	128
	129	ihlen = iph->ihl * 4;
	130	plen = skb->len - ihlen;
	131	dlen = IPCOMP_SCRATCH_SIZE;
	132	start = skb->data + ihlen;
	133
	134	cpu = get_cpu();
	135	scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
	136	tfm = *per_cpu_ptr(ipcd->tfms, cpu);
	137
	138	err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
	139	if (err)
	140	goto out;
	141
	142	if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
	143	err = -EMSGSIZE;
	144	goto out;
	145	}
	146
	147	memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
	148	put_cpu();
149
150	pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
151	return 0;
152
153	out:
154	put_cpu();
155	return err;
156	}
157
158	static int ipcomp_output(struct xfrm_state x, struct sk_buff skb)
159	{
160	int err;
161	struct iphdr *iph;
162	struct ip_comp_hdr *ipch;
163	struct ipcomp_data *ipcd = x->data;
164	int hdr_len = 0;
165
166	iph = skb->nh.iph;
167	iph->tot_len = htons(skb->len);
168	hdr_len = iph->ihl * 4;
169	if ((skb->len - hdr_len) < ipcd->threshold) {
170	/* Don't bother compressing */
171	goto out_ok;
172	}
173
174	if ((skb_is_nonlinear(skb) \|\| skb_cloned(skb)) &&
175	skb_linearize(skb, GFP_ATOMIC) != 0) {
176	goto out_ok;
177	}
178
179	err = ipcomp_compress(x, skb);
180	iph = skb->nh.iph;
181
182	if (err) {
183	goto out_ok;
184	}
185
186	/* Install ipcomp header, convert into ipcomp datagram. */
187	iph->tot_len = htons(skb->len);
188	ipch = (struct ip_comp_hdr )((char )iph + iph->ihl * 4);
189	ipch->nexthdr = iph->protocol;
190	ipch->flags = 0;
191	ipch->cpi = htons((u16 )ntohl(x->id.spi));
192	iph->protocol = IPPROTO_COMP;
193	ip_send_check(iph);
194	return 0;
195
196	out_ok:
197	if (x->props.mode)
198	ip_send_check(iph);
199	return 0;
200	}
201
202	static void ipcomp4_err(struct sk_buff *skb, u32 info)
203	{
204	u32 spi;
205	struct iphdr iph = (struct iphdr )skb->data;
206	struct ip_comp_hdr ipch = (struct ip_comp_hdr )(skb->data+(iph->ihl<<2));
207	struct xfrm_state *x;
208
209	if (skb->h.icmph->type != ICMP_DEST_UNREACH \|\|
210	skb->h.icmph->code != ICMP_FRAG_NEEDED)
211	return;
212
213	spi = ntohl(ntohs(ipch->cpi));
214	x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
215	spi, IPPROTO_COMP, AF_INET);
216	if (!x)
217	return;
64ce2073 PM	218	NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
64ce2073 PM	219	spi, NIPQUAD(iph->daddr));
1da177e4 LT	220	xfrm_state_put(x);
	221	}
	222
	223	/* We always hold one tunnel user reference to indicate a tunnel */
	224	static struct xfrm_state ipcomp_tunnel_create(struct xfrm_state x)
	225	{
	226	struct xfrm_state *t;
	227
	228	t = xfrm_state_alloc();
	229	if (t == NULL)
	230	goto out;
	231
	232	t->id.proto = IPPROTO_IPIP;
	233	t->id.spi = x->props.saddr.a4;
	234	t->id.daddr.a4 = x->id.daddr.a4;
	235	memcpy(&t->sel, &x->sel, sizeof(t->sel));
	236	t->props.family = AF_INET;
	237	t->props.mode = 1;
	238	t->props.saddr.a4 = x->props.saddr.a4;
	239	t->props.flags = x->props.flags;
72cb6962 HX	240
72cb6962 HX	241	if (xfrm_init_state(t))
1da177e4 LT	242	goto error;
1da177e4 LT	243
1da177e4 LT	244	atomic_set(&t->tunnel_users, 1);
	245	out:
	246	return t;
	247
	248	error:
	249	t->km.state = XFRM_STATE_DEAD;
	250	xfrm_state_put(t);
	251	t = NULL;
	252	goto out;
	253	}
	254
	255	/*
4a3e2f71	256	* Must be protected by xfrm_cfg_mutex. State and tunnel user references are
1da177e4 LT	257	* always incremented on success.
	258	*/
	259	static int ipcomp_tunnel_attach(struct xfrm_state *x)
	260	{
	261	int err = 0;
	262	struct xfrm_state *t;
	263
	264	t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
	265	x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
	266	if (!t) {
	267	t = ipcomp_tunnel_create(x);
	268	if (!t) {
	269	err = -EINVAL;
	270	goto out;
	271	}
	272	xfrm_state_insert(t);
	273	xfrm_state_hold(t);
	274	}
	275	x->tunnel = t;
	276	atomic_inc(&t->tunnel_users);
	277	out:
	278	return err;
	279	}
	280
	281	static void ipcomp_free_scratches(void)
	282	{
	283	int i;
	284	void **scratches;
	285
	286	if (--ipcomp_scratch_users)
	287	return;
	288
	289	scratches = ipcomp_scratches;
	290	if (!scratches)
	291	return;
	292
63903ca6 JJ	293	for_each_possible_cpu(i)
63903ca6 JJ	294	vfree(*per_cpu_ptr(scratches, i));
1da177e4 LT	295
	296	free_percpu(scratches);
	297	}
	298
	299	static void **ipcomp_alloc_scratches(void)
	300	{
	301	int i;
	302	void **scratches;
	303
	304	if (ipcomp_scratch_users++)
	305	return ipcomp_scratches;
	306
	307	scratches = alloc_percpu(void *);
	308	if (!scratches)
	309	return NULL;
	310
	311	ipcomp_scratches = scratches;
	312
6f912042	313	for_each_possible_cpu(i) {
1da177e4 LT	314	void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
	315	if (!scratch)
	316	return NULL;
	317	*per_cpu_ptr(scratches, i) = scratch;
	318	}
	319
	320	return scratches;
	321	}
	322
	323	static void ipcomp_free_tfms(struct crypto_tfm **tfms)
	324	{
	325	struct ipcomp_tfms *pos;
	326	int cpu;
	327
	328	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
	329	if (pos->tfms == tfms)
	330	break;
	331	}
	332
	333	BUG_TRAP(pos);
	334
	335	if (--pos->users)
	336	return;
	337
	338	list_del(&pos->list);
	339	kfree(pos);
	340
	341	if (!tfms)
	342	return;
	343
6f912042	344	for_each_possible_cpu(cpu) {
1da177e4	345	struct crypto_tfm tfm = per_cpu_ptr(tfms, cpu);
573dbd95	346	crypto_free_tfm(tfm);
1da177e4 LT	347	}
	348	free_percpu(tfms);
	349	}
	350
	351	static struct crypto_tfm *ipcomp_alloc_tfms(const char alg_name)
	352	{
	353	struct ipcomp_tfms *pos;
	354	struct crypto_tfm **tfms;
	355	int cpu;
	356
	357	/* This can be any valid CPU ID so we don't need locking. */
6fc8b9e7	358	cpu = raw_smp_processor_id();
1da177e4 LT	359
	360	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
	361	struct crypto_tfm *tfm;
	362
	363	tfms = pos->tfms;
	364	tfm = *per_cpu_ptr(tfms, cpu);
	365
	366	if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
	367	pos->users++;
	368	return tfms;
	369	}
	370	}
	371
	372	pos = kmalloc(sizeof(*pos), GFP_KERNEL);
	373	if (!pos)
	374	return NULL;
	375
	376	pos->users = 1;
	377	INIT_LIST_HEAD(&pos->list);
	378	list_add(&pos->list, &ipcomp_tfms_list);
	379
	380	pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
	381	if (!tfms)
	382	goto error;
	383
6f912042	384	for_each_possible_cpu(cpu) {
1da177e4 LT	385	struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
	386	if (!tfm)
	387	goto error;
	388	*per_cpu_ptr(tfms, cpu) = tfm;
	389	}
	390
	391	return tfms;
	392
	393	error:
	394	ipcomp_free_tfms(tfms);
	395	return NULL;
	396	}
	397
	398	static void ipcomp_free_data(struct ipcomp_data *ipcd)
	399	{
	400	if (ipcd->tfms)
	401	ipcomp_free_tfms(ipcd->tfms);
	402	ipcomp_free_scratches();
	403	}
	404
	405	static void ipcomp_destroy(struct xfrm_state *x)
	406	{
	407	struct ipcomp_data *ipcd = x->data;
	408	if (!ipcd)
	409	return;
	410	xfrm_state_delete_tunnel(x);
4a3e2f71	411	mutex_lock(&ipcomp_resource_mutex);
1da177e4	412	ipcomp_free_data(ipcd);
4a3e2f71	413	mutex_unlock(&ipcomp_resource_mutex);
1da177e4 LT	414	kfree(ipcd);
	415	}
	416
72cb6962	417	static int ipcomp_init_state(struct xfrm_state *x)
1da177e4 LT	418	{
	419	int err;
	420	struct ipcomp_data *ipcd;
	421	struct xfrm_algo_desc *calg_desc;
	422
	423	err = -EINVAL;
	424	if (!x->calg)
	425	goto out;
	426
	427	if (x->encap)
	428	goto out;
	429
	430	err = -ENOMEM;
	431	ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
	432	if (!ipcd)
	433	goto out;
	434
	435	memset(ipcd, 0, sizeof(*ipcd));
	436	x->props.header_len = 0;
	437	if (x->props.mode)
	438	x->props.header_len += sizeof(struct iphdr);
	439
4a3e2f71	440	mutex_lock(&ipcomp_resource_mutex);
1da177e4 LT	441	if (!ipcomp_alloc_scratches())
	442	goto error;
	443
	444	ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
	445	if (!ipcd->tfms)
	446	goto error;
4a3e2f71	447	mutex_unlock(&ipcomp_resource_mutex);
1da177e4 LT	448
	449	if (x->props.mode) {
	450	err = ipcomp_tunnel_attach(x);
	451	if (err)
	452	goto error_tunnel;
	453	}
	454
	455	calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
	456	BUG_ON(!calg_desc);
	457	ipcd->threshold = calg_desc->uinfo.comp.threshold;
	458	x->data = ipcd;
	459	err = 0;
	460	out:
	461	return err;
	462
	463	error_tunnel:
4a3e2f71	464	mutex_lock(&ipcomp_resource_mutex);
1da177e4 LT	465	error:
1da177e4 LT	466	ipcomp_free_data(ipcd);
4a3e2f71	467	mutex_unlock(&ipcomp_resource_mutex);
1da177e4 LT	468	kfree(ipcd);
	469	goto out;
	470	}
	471
	472	static struct xfrm_type ipcomp_type = {
	473	.description = "IPCOMP4",
	474	.owner = THIS_MODULE,
	475	.proto = IPPROTO_COMP,
	476	.init_state = ipcomp_init_state,
	477	.destructor = ipcomp_destroy,
	478	.input = ipcomp_input,
	479	.output = ipcomp_output
	480	};
	481
	482	static struct net_protocol ipcomp4_protocol = {
	483	.handler = xfrm4_rcv,
	484	.err_handler = ipcomp4_err,
	485	.no_policy = 1,
	486	};
	487
	488	static int __init ipcomp4_init(void)
	489	{
	490	if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
	491	printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
	492	return -EAGAIN;
	493	}
	494	if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
	495	printk(KERN_INFO "ipcomp init: can't add protocol\n");
	496	xfrm_unregister_type(&ipcomp_type, AF_INET);
	497	return -EAGAIN;
	498	}
	499	return 0;
	500	}
	501
	502	static void __exit ipcomp4_fini(void)
	503	{
	504	if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
	505	printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
	506	if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
	507	printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
	508	}
	509
	510	module_init(ipcomp4_init);
	511	module_exit(ipcomp4_fini);
	512
	513	MODULE_LICENSE("GPL");
	514	MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
	515	MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
	516