[linux-2.6] / 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 <net/ip.h>
#include <net/xfrm.h>
#include <net/icmp.h>
#include <net/ipcomp.h>

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

static DECLARE_MUTEX(ipcomp_resource_sem);
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 xfrm_decap_state *decap, 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(printk(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;
	
	t->type = xfrm_get_type(IPPROTO_IPIP, t->props.family);
	if (t->type == NULL)
		goto error;
		
	if (t->type->init_state(t, NULL))
		goto error;

	t->km.state = XFRM_STATE_VALID;
	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_sem.  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_cpu(i) {
		void *scratch = *per_cpu_ptr(scratches, i);
		if (scratch)
			vfree(scratch);
	}

	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_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_cpu(cpu) {
		struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
		if (tfm)
			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 = 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_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);
	down(&ipcomp_resource_sem);
	ipcomp_free_data(ipcd);
	up(&ipcomp_resource_sem);
	kfree(ipcd);
}

static int ipcomp_init_state(struct xfrm_state *x, void *args)
{
	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);

	down(&ipcomp_resource_sem);
	if (!ipcomp_alloc_scratches())
		goto error;

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

	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:
	down(&ipcomp_resource_sem);
error:
	ipcomp_free_data(ipcd);
	up(&ipcomp_resource_sem);
	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>
	27	#include <net/ip.h>
	28	#include <net/xfrm.h>
	29	#include <net/icmp.h>
	30	#include <net/ipcomp.h>
	31
	32	struct ipcomp_tfms {
	33	struct list_head list;
	34	struct crypto_tfm **tfms;
	35	int users;
	36	};
	37
	38	static DECLARE_MUTEX(ipcomp_resource_sem);
	39	static void **ipcomp_scratches;
	40	static int ipcomp_scratch_users;
	41	static LIST_HEAD(ipcomp_tfms_list);
	42
	43	static int ipcomp_decompress(struct xfrm_state x, struct sk_buff skb)
	44	{
	45	int err, plen, dlen;
	46	struct iphdr *iph;
	47	struct ipcomp_data *ipcd = x->data;
	48	u8 start, scratch;
	49	struct crypto_tfm *tfm;
	50	int cpu;
	51
	52	plen = skb->len;
	53	dlen = IPCOMP_SCRATCH_SIZE;
	54	start = skb->data;
	55
	56	cpu = get_cpu();
	57	scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
	58	tfm = *per_cpu_ptr(ipcd->tfms, cpu);
	59
	60	err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
	61	if (err)
	62	goto out;
	63
	64	if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
65	err = -EINVAL;
66	goto out;
67	}
68
69	err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
70	if (err)
71	goto out;
72
73	skb_put(skb, dlen - plen);
74	memcpy(skb->data, scratch, dlen);
75	iph = skb->nh.iph;
76	iph->tot_len = htons(dlen + iph->ihl * 4);
77	out:
78	put_cpu();
79	return err;
80	}
81
82	static int ipcomp_input(struct xfrm_state *x,
83	struct xfrm_decap_state decap, struct sk_buff skb)
84	{
85	u8 nexthdr;
86	int err = 0;
87	struct iphdr *iph;
88	union {
89	struct iphdr iph;
90	char buf[60];
91	} tmp_iph;
92
93
94	if ((skb_is_nonlinear(skb) \|\| skb_cloned(skb)) &&
95	skb_linearize(skb, GFP_ATOMIC) != 0) {
96	err = -ENOMEM;
97	goto out;
98	}
99
100	skb->ip_summed = CHECKSUM_NONE;
101
102	/* Remove ipcomp header and decompress original payload */
103	iph = skb->nh.iph;
104	memcpy(&tmp_iph, iph, iph->ihl * 4);
105	nexthdr = (u8 )skb->data;
106	skb_pull(skb, sizeof(struct ip_comp_hdr));
107	skb->nh.raw += sizeof(struct ip_comp_hdr);
108	memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
109	iph = skb->nh.iph;
110	iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
111	iph->protocol = nexthdr;
112	skb->h.raw = skb->data;
113	err = ipcomp_decompress(x, skb);
114
115	out:
116	return err;
117	}
118
119	static int ipcomp_compress(struct xfrm_state x, struct sk_buff skb)
120	{
121	int err, plen, dlen, ihlen;
122	struct iphdr *iph = skb->nh.iph;
123	struct ipcomp_data *ipcd = x->data;
124	u8 start, scratch;
125	struct crypto_tfm *tfm;
126	int cpu;
127
128	ihlen = iph->ihl * 4;
129	plen = skb->len - ihlen;
130	dlen = IPCOMP_SCRATCH_SIZE;
131	start = skb->data + ihlen;
132
133	cpu = get_cpu();
134	scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
135	tfm = *per_cpu_ptr(ipcd->tfms, cpu);
136
137	err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
138	if (err)
139	goto out;
140
141	if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
142	err = -EMSGSIZE;
143	goto out;
144	}
145
146	memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
147	put_cpu();
148
149	pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
150	return 0;
151
152	out:
153	put_cpu();
154	return err;
155	}
156
157	static int ipcomp_output(struct xfrm_state x, struct sk_buff skb)
158	{
159	int err;
160	struct iphdr *iph;
161	struct ip_comp_hdr *ipch;
162	struct ipcomp_data *ipcd = x->data;
163	int hdr_len = 0;
164
165	iph = skb->nh.iph;
166	iph->tot_len = htons(skb->len);
167	hdr_len = iph->ihl * 4;
168	if ((skb->len - hdr_len) < ipcd->threshold) {
169	/* Don't bother compressing */
170	goto out_ok;
171	}
172
173	if ((skb_is_nonlinear(skb) \|\| skb_cloned(skb)) &&
174	skb_linearize(skb, GFP_ATOMIC) != 0) {
175	goto out_ok;
176	}
177
178	err = ipcomp_compress(x, skb);
179	iph = skb->nh.iph;
180
181	if (err) {
182	goto out_ok;
183	}
184
185	/* Install ipcomp header, convert into ipcomp datagram. */
186	iph->tot_len = htons(skb->len);
187	ipch = (struct ip_comp_hdr )((char )iph + iph->ihl * 4);
188	ipch->nexthdr = iph->protocol;
189	ipch->flags = 0;
190	ipch->cpi = htons((u16 )ntohl(x->id.spi));
191	iph->protocol = IPPROTO_COMP;
192	ip_send_check(iph);
193	return 0;
194
195	out_ok:
196	if (x->props.mode)
197	ip_send_check(iph);
198	return 0;
199	}
200
201	static void ipcomp4_err(struct sk_buff *skb, u32 info)
202	{
203	u32 spi;
204	struct iphdr iph = (struct iphdr )skb->data;
205	struct ip_comp_hdr ipch = (struct ip_comp_hdr )(skb->data+(iph->ihl<<2));
206	struct xfrm_state *x;
207
208	if (skb->h.icmph->type != ICMP_DEST_UNREACH \|\|
209	skb->h.icmph->code != ICMP_FRAG_NEEDED)
210	return;
211
212	spi = ntohl(ntohs(ipch->cpi));
213	x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
214	spi, IPPROTO_COMP, AF_INET);
215	if (!x)
216	return;
217	NETDEBUG(printk(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
218	spi, NIPQUAD(iph->daddr)));
219	xfrm_state_put(x);
220	}
221
222	/* We always hold one tunnel user reference to indicate a tunnel */
223	static struct xfrm_state ipcomp_tunnel_create(struct xfrm_state x)
224	{
225	struct xfrm_state *t;
226
227	t = xfrm_state_alloc();
228	if (t == NULL)
229	goto out;
230
231	t->id.proto = IPPROTO_IPIP;
232	t->id.spi = x->props.saddr.a4;
233	t->id.daddr.a4 = x->id.daddr.a4;
234	memcpy(&t->sel, &x->sel, sizeof(t->sel));
235	t->props.family = AF_INET;
236	t->props.mode = 1;
237	t->props.saddr.a4 = x->props.saddr.a4;
238	t->props.flags = x->props.flags;
239
240	t->type = xfrm_get_type(IPPROTO_IPIP, t->props.family);
241	if (t->type == NULL)
242	goto error;
243
244	if (t->type->init_state(t, NULL))
245	goto error;
246
247	t->km.state = XFRM_STATE_VALID;
248	atomic_set(&t->tunnel_users, 1);
249	out:
250	return t;
251
252	error:
253	t->km.state = XFRM_STATE_DEAD;
254	xfrm_state_put(t);
255	t = NULL;
256	goto out;
257	}
258
259	/*
260	* Must be protected by xfrm_cfg_sem. State and tunnel user references are
261	* always incremented on success.
262	*/
263	static int ipcomp_tunnel_attach(struct xfrm_state *x)
264	{
265	int err = 0;
266	struct xfrm_state *t;
267
268	t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
269	x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
270	if (!t) {
271	t = ipcomp_tunnel_create(x);
272	if (!t) {
273	err = -EINVAL;
274	goto out;
275	}
276	xfrm_state_insert(t);
277	xfrm_state_hold(t);
278	}
279	x->tunnel = t;
280	atomic_inc(&t->tunnel_users);
281	out:
282	return err;
283	}
284
285	static void ipcomp_free_scratches(void)
286	{
287	int i;
288	void **scratches;
289
290	if (--ipcomp_scratch_users)
291	return;
292
293	scratches = ipcomp_scratches;
294	if (!scratches)
295	return;
296
297	for_each_cpu(i) {
298	void scratch = per_cpu_ptr(scratches, i);
299	if (scratch)
300	vfree(scratch);
301	}
302
303	free_percpu(scratches);
304	}
305
306	static void **ipcomp_alloc_scratches(void)
307	{
308	int i;
309	void **scratches;
310
311	if (ipcomp_scratch_users++)
312	return ipcomp_scratches;
313
314	scratches = alloc_percpu(void *);
315	if (!scratches)
316	return NULL;
317
318	ipcomp_scratches = scratches;
319
320	for_each_cpu(i) {
321	void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
322	if (!scratch)
323	return NULL;
324	*per_cpu_ptr(scratches, i) = scratch;
325	}
326
327	return scratches;
328	}
329
330	static void ipcomp_free_tfms(struct crypto_tfm **tfms)
331	{
332	struct ipcomp_tfms *pos;
333	int cpu;
334
335	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
336	if (pos->tfms == tfms)
337	break;
338	}
339
340	BUG_TRAP(pos);
341
342	if (--pos->users)
343	return;
344
345	list_del(&pos->list);
346	kfree(pos);
347
348	if (!tfms)
349	return;
350
351	for_each_cpu(cpu) {
352	struct crypto_tfm tfm = per_cpu_ptr(tfms, cpu);
353	if (tfm)
354	crypto_free_tfm(tfm);
355	}
356	free_percpu(tfms);
357	}
358
359	static struct crypto_tfm *ipcomp_alloc_tfms(const char alg_name)
360	{
361	struct ipcomp_tfms *pos;
362	struct crypto_tfm **tfms;
363	int cpu;
364
365	/* This can be any valid CPU ID so we don't need locking. */
366	cpu = smp_processor_id();
367
368	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
369	struct crypto_tfm *tfm;
370
371	tfms = pos->tfms;
372	tfm = *per_cpu_ptr(tfms, cpu);
373
374	if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
375	pos->users++;
376	return tfms;
377	}
378	}
379
380	pos = kmalloc(sizeof(*pos), GFP_KERNEL);
381	if (!pos)
382	return NULL;
383
384	pos->users = 1;
385	INIT_LIST_HEAD(&pos->list);
386	list_add(&pos->list, &ipcomp_tfms_list);
387
388	pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
389	if (!tfms)
390	goto error;
391
392	for_each_cpu(cpu) {
393	struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
394	if (!tfm)
395	goto error;
396	*per_cpu_ptr(tfms, cpu) = tfm;
397	}
398
399	return tfms;
400
401	error:
402	ipcomp_free_tfms(tfms);
403	return NULL;
404	}
405
406	static void ipcomp_free_data(struct ipcomp_data *ipcd)
407	{
408	if (ipcd->tfms)
409	ipcomp_free_tfms(ipcd->tfms);
410	ipcomp_free_scratches();
411	}
412
413	static void ipcomp_destroy(struct xfrm_state *x)
414	{
415	struct ipcomp_data *ipcd = x->data;
416	if (!ipcd)
417	return;
418	xfrm_state_delete_tunnel(x);
419	down(&ipcomp_resource_sem);
420	ipcomp_free_data(ipcd);
421	up(&ipcomp_resource_sem);
422	kfree(ipcd);
423	}
424
425	static int ipcomp_init_state(struct xfrm_state x, void args)
426	{
427	int err;
428	struct ipcomp_data *ipcd;
429	struct xfrm_algo_desc *calg_desc;
430
431	err = -EINVAL;
432	if (!x->calg)
433	goto out;
434
435	if (x->encap)
436	goto out;
437
438	err = -ENOMEM;
439	ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
440	if (!ipcd)
441	goto out;
442
443	memset(ipcd, 0, sizeof(*ipcd));
444	x->props.header_len = 0;
445	if (x->props.mode)
446	x->props.header_len += sizeof(struct iphdr);
447
448	down(&ipcomp_resource_sem);
449	if (!ipcomp_alloc_scratches())
450	goto error;
451
452	ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
453	if (!ipcd->tfms)
454	goto error;
455	up(&ipcomp_resource_sem);
456
457	if (x->props.mode) {
458	err = ipcomp_tunnel_attach(x);
459	if (err)
460	goto error_tunnel;
461	}
462
463	calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
464	BUG_ON(!calg_desc);
465	ipcd->threshold = calg_desc->uinfo.comp.threshold;
466	x->data = ipcd;
467	err = 0;
468	out:
469	return err;
470
471	error_tunnel:
472	down(&ipcomp_resource_sem);
473	error:
474	ipcomp_free_data(ipcd);
475	up(&ipcomp_resource_sem);
476	kfree(ipcd);
477	goto out;
478	}
479
480	static struct xfrm_type ipcomp_type = {
481	.description = "IPCOMP4",
482	.owner = THIS_MODULE,
483	.proto = IPPROTO_COMP,
484	.init_state = ipcomp_init_state,
485	.destructor = ipcomp_destroy,
486	.input = ipcomp_input,
487	.output = ipcomp_output
488	};
489
490	static struct net_protocol ipcomp4_protocol = {
491	.handler = xfrm4_rcv,
492	.err_handler = ipcomp4_err,
493	.no_policy = 1,
494	};
495
496	static int __init ipcomp4_init(void)
497	{
498	if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
499	printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
500	return -EAGAIN;
501	}
502	if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
503	printk(KERN_INFO "ipcomp init: can't add protocol\n");
504	xfrm_unregister_type(&ipcomp_type, AF_INET);
505	return -EAGAIN;
506	}
507	return 0;
508	}
509
510	static void __exit ipcomp4_fini(void)
511	{
512	if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
513	printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
514	if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
515	printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
516	}
517
518	module_init(ipcomp4_init);
519	module_exit(ipcomp4_fini);
520
521	MODULE_LICENSE("GPL");
522	MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
523	MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
524