xref: /openbmc/linux/net/xfrm/xfrm_ipcomp.c (revision 95298d63)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * IP Payload Compression Protocol (IPComp) - RFC3173.
4  *
5  * Copyright (c) 2003 James Morris <jmorris@intercode.com.au>
6  * Copyright (c) 2003-2008 Herbert Xu <herbert@gondor.apana.org.au>
7  *
8  * Todo:
9  *   - Tunable compression parameters.
10  *   - Compression stats.
11  *   - Adaptive compression.
12  */
13 
14 #include <linux/crypto.h>
15 #include <linux/err.h>
16 #include <linux/list.h>
17 #include <linux/module.h>
18 #include <linux/mutex.h>
19 #include <linux/percpu.h>
20 #include <linux/slab.h>
21 #include <linux/smp.h>
22 #include <linux/vmalloc.h>
23 #include <net/ip.h>
24 #include <net/ipcomp.h>
25 #include <net/xfrm.h>
26 
27 struct ipcomp_tfms {
28 	struct list_head list;
29 	struct crypto_comp * __percpu *tfms;
30 	int users;
31 };
32 
33 static DEFINE_MUTEX(ipcomp_resource_mutex);
34 static void * __percpu *ipcomp_scratches;
35 static int ipcomp_scratch_users;
36 static LIST_HEAD(ipcomp_tfms_list);
37 
38 static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
39 {
40 	struct ipcomp_data *ipcd = x->data;
41 	const int plen = skb->len;
42 	int dlen = IPCOMP_SCRATCH_SIZE;
43 	const u8 *start = skb->data;
44 	const int cpu = get_cpu();
45 	u8 *scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
46 	struct crypto_comp *tfm = *per_cpu_ptr(ipcd->tfms, cpu);
47 	int err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
48 	int len;
49 
50 	if (err)
51 		goto out;
52 
53 	if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
54 		err = -EINVAL;
55 		goto out;
56 	}
57 
58 	len = dlen - plen;
59 	if (len > skb_tailroom(skb))
60 		len = skb_tailroom(skb);
61 
62 	__skb_put(skb, len);
63 
64 	len += plen;
65 	skb_copy_to_linear_data(skb, scratch, len);
66 
67 	while ((scratch += len, dlen -= len) > 0) {
68 		skb_frag_t *frag;
69 		struct page *page;
70 
71 		err = -EMSGSIZE;
72 		if (WARN_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS))
73 			goto out;
74 
75 		frag = skb_shinfo(skb)->frags + skb_shinfo(skb)->nr_frags;
76 		page = alloc_page(GFP_ATOMIC);
77 
78 		err = -ENOMEM;
79 		if (!page)
80 			goto out;
81 
82 		__skb_frag_set_page(frag, page);
83 
84 		len = PAGE_SIZE;
85 		if (dlen < len)
86 			len = dlen;
87 
88 		skb_frag_off_set(frag, 0);
89 		skb_frag_size_set(frag, len);
90 		memcpy(skb_frag_address(frag), scratch, len);
91 
92 		skb->truesize += len;
93 		skb->data_len += len;
94 		skb->len += len;
95 
96 		skb_shinfo(skb)->nr_frags++;
97 	}
98 
99 	err = 0;
100 
101 out:
102 	put_cpu();
103 	return err;
104 }
105 
106 int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
107 {
108 	int nexthdr;
109 	int err = -ENOMEM;
110 	struct ip_comp_hdr *ipch;
111 
112 	if (skb_linearize_cow(skb))
113 		goto out;
114 
115 	skb->ip_summed = CHECKSUM_NONE;
116 
117 	/* Remove ipcomp header and decompress original payload */
118 	ipch = (void *)skb->data;
119 	nexthdr = ipch->nexthdr;
120 
121 	skb->transport_header = skb->network_header + sizeof(*ipch);
122 	__skb_pull(skb, sizeof(*ipch));
123 	err = ipcomp_decompress(x, skb);
124 	if (err)
125 		goto out;
126 
127 	err = nexthdr;
128 
129 out:
130 	return err;
131 }
132 EXPORT_SYMBOL_GPL(ipcomp_input);
133 
134 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
135 {
136 	struct ipcomp_data *ipcd = x->data;
137 	const int plen = skb->len;
138 	int dlen = IPCOMP_SCRATCH_SIZE;
139 	u8 *start = skb->data;
140 	struct crypto_comp *tfm;
141 	u8 *scratch;
142 	int err;
143 
144 	local_bh_disable();
145 	scratch = *this_cpu_ptr(ipcomp_scratches);
146 	tfm = *this_cpu_ptr(ipcd->tfms);
147 	err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
148 	if (err)
149 		goto out;
150 
151 	if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
152 		err = -EMSGSIZE;
153 		goto out;
154 	}
155 
156 	memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
157 	local_bh_enable();
158 
159 	pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
160 	return 0;
161 
162 out:
163 	local_bh_enable();
164 	return err;
165 }
166 
167 int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
168 {
169 	int err;
170 	struct ip_comp_hdr *ipch;
171 	struct ipcomp_data *ipcd = x->data;
172 
173 	if (skb->len < ipcd->threshold) {
174 		/* Don't bother compressing */
175 		goto out_ok;
176 	}
177 
178 	if (skb_linearize_cow(skb))
179 		goto out_ok;
180 
181 	err = ipcomp_compress(x, skb);
182 
183 	if (err) {
184 		goto out_ok;
185 	}
186 
187 	/* Install ipcomp header, convert into ipcomp datagram. */
188 	ipch = ip_comp_hdr(skb);
189 	ipch->nexthdr = *skb_mac_header(skb);
190 	ipch->flags = 0;
191 	ipch->cpi = htons((u16 )ntohl(x->id.spi));
192 	*skb_mac_header(skb) = IPPROTO_COMP;
193 out_ok:
194 	skb_push(skb, -skb_network_offset(skb));
195 	return 0;
196 }
197 EXPORT_SYMBOL_GPL(ipcomp_output);
198 
199 static void ipcomp_free_scratches(void)
200 {
201 	int i;
202 	void * __percpu *scratches;
203 
204 	if (--ipcomp_scratch_users)
205 		return;
206 
207 	scratches = ipcomp_scratches;
208 	if (!scratches)
209 		return;
210 
211 	for_each_possible_cpu(i)
212 		vfree(*per_cpu_ptr(scratches, i));
213 
214 	free_percpu(scratches);
215 }
216 
217 static void * __percpu *ipcomp_alloc_scratches(void)
218 {
219 	void * __percpu *scratches;
220 	int i;
221 
222 	if (ipcomp_scratch_users++)
223 		return ipcomp_scratches;
224 
225 	scratches = alloc_percpu(void *);
226 	if (!scratches)
227 		return NULL;
228 
229 	ipcomp_scratches = scratches;
230 
231 	for_each_possible_cpu(i) {
232 		void *scratch;
233 
234 		scratch = vmalloc_node(IPCOMP_SCRATCH_SIZE, cpu_to_node(i));
235 		if (!scratch)
236 			return NULL;
237 		*per_cpu_ptr(scratches, i) = scratch;
238 	}
239 
240 	return scratches;
241 }
242 
243 static void ipcomp_free_tfms(struct crypto_comp * __percpu *tfms)
244 {
245 	struct ipcomp_tfms *pos;
246 	int cpu;
247 
248 	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
249 		if (pos->tfms == tfms)
250 			break;
251 	}
252 
253 	WARN_ON(!pos);
254 
255 	if (--pos->users)
256 		return;
257 
258 	list_del(&pos->list);
259 	kfree(pos);
260 
261 	if (!tfms)
262 		return;
263 
264 	for_each_possible_cpu(cpu) {
265 		struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
266 		crypto_free_comp(tfm);
267 	}
268 	free_percpu(tfms);
269 }
270 
271 static struct crypto_comp * __percpu *ipcomp_alloc_tfms(const char *alg_name)
272 {
273 	struct ipcomp_tfms *pos;
274 	struct crypto_comp * __percpu *tfms;
275 	int cpu;
276 
277 
278 	list_for_each_entry(pos, &ipcomp_tfms_list, list) {
279 		struct crypto_comp *tfm;
280 
281 		/* This can be any valid CPU ID so we don't need locking. */
282 		tfm = this_cpu_read(*pos->tfms);
283 
284 		if (!strcmp(crypto_comp_name(tfm), alg_name)) {
285 			pos->users++;
286 			return pos->tfms;
287 		}
288 	}
289 
290 	pos = kmalloc(sizeof(*pos), GFP_KERNEL);
291 	if (!pos)
292 		return NULL;
293 
294 	pos->users = 1;
295 	INIT_LIST_HEAD(&pos->list);
296 	list_add(&pos->list, &ipcomp_tfms_list);
297 
298 	pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
299 	if (!tfms)
300 		goto error;
301 
302 	for_each_possible_cpu(cpu) {
303 		struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
304 							    CRYPTO_ALG_ASYNC);
305 		if (IS_ERR(tfm))
306 			goto error;
307 		*per_cpu_ptr(tfms, cpu) = tfm;
308 	}
309 
310 	return tfms;
311 
312 error:
313 	ipcomp_free_tfms(tfms);
314 	return NULL;
315 }
316 
317 static void ipcomp_free_data(struct ipcomp_data *ipcd)
318 {
319 	if (ipcd->tfms)
320 		ipcomp_free_tfms(ipcd->tfms);
321 	ipcomp_free_scratches();
322 }
323 
324 void ipcomp_destroy(struct xfrm_state *x)
325 {
326 	struct ipcomp_data *ipcd = x->data;
327 	if (!ipcd)
328 		return;
329 	xfrm_state_delete_tunnel(x);
330 	mutex_lock(&ipcomp_resource_mutex);
331 	ipcomp_free_data(ipcd);
332 	mutex_unlock(&ipcomp_resource_mutex);
333 	kfree(ipcd);
334 }
335 EXPORT_SYMBOL_GPL(ipcomp_destroy);
336 
337 int ipcomp_init_state(struct xfrm_state *x)
338 {
339 	int err;
340 	struct ipcomp_data *ipcd;
341 	struct xfrm_algo_desc *calg_desc;
342 
343 	err = -EINVAL;
344 	if (!x->calg)
345 		goto out;
346 
347 	if (x->encap)
348 		goto out;
349 
350 	err = -ENOMEM;
351 	ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
352 	if (!ipcd)
353 		goto out;
354 
355 	mutex_lock(&ipcomp_resource_mutex);
356 	if (!ipcomp_alloc_scratches())
357 		goto error;
358 
359 	ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
360 	if (!ipcd->tfms)
361 		goto error;
362 	mutex_unlock(&ipcomp_resource_mutex);
363 
364 	calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
365 	BUG_ON(!calg_desc);
366 	ipcd->threshold = calg_desc->uinfo.comp.threshold;
367 	x->data = ipcd;
368 	err = 0;
369 out:
370 	return err;
371 
372 error:
373 	ipcomp_free_data(ipcd);
374 	mutex_unlock(&ipcomp_resource_mutex);
375 	kfree(ipcd);
376 	goto out;
377 }
378 EXPORT_SYMBOL_GPL(ipcomp_init_state);
379 
380 MODULE_LICENSE("GPL");
381 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
382 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
383