ip6tnl: Use on-device stats instead of private ones.
[linux-2.6] / net / ipv6 / ipcomp6.c
1 /*
2  * IP Payload Compression Protocol (IPComp) for IPv6 - RFC3173
3  *
4  * Copyright (C)2003 USAGI/WIDE Project
5  *
6  * Author       Mitsuru KANDA  <mk@linux-ipv6.org>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
21  */
22 /*
23  * [Memo]
24  *
25  * Outbound:
26  *  The compression of IP datagram MUST be done before AH/ESP processing,
27  *  fragmentation, and the addition of Hop-by-Hop/Routing header.
28  *
29  * Inbound:
30  *  The decompression of IP datagram MUST be done after the reassembly,
31  *  AH/ESP processing.
32  */
33 #include <linux/module.h>
34 #include <net/ip.h>
35 #include <net/xfrm.h>
36 #include <net/ipcomp.h>
37 #include <linux/crypto.h>
38 #include <linux/err.h>
39 #include <linux/pfkeyv2.h>
40 #include <linux/random.h>
41 #include <linux/percpu.h>
42 #include <linux/smp.h>
43 #include <linux/list.h>
44 #include <linux/vmalloc.h>
45 #include <linux/rtnetlink.h>
46 #include <net/icmp.h>
47 #include <net/ipv6.h>
48 #include <net/protocol.h>
49 #include <linux/ipv6.h>
50 #include <linux/icmpv6.h>
51 #include <linux/mutex.h>
52
53 struct ipcomp6_tfms {
54         struct list_head list;
55         struct crypto_comp **tfms;
56         int users;
57 };
58
59 static DEFINE_MUTEX(ipcomp6_resource_mutex);
60 static void **ipcomp6_scratches;
61 static int ipcomp6_scratch_users;
62 static LIST_HEAD(ipcomp6_tfms_list);
63
64 static int ipcomp6_input(struct xfrm_state *x, struct sk_buff *skb)
65 {
66         int nexthdr;
67         int err = -ENOMEM;
68         struct ip_comp_hdr *ipch;
69         int plen, dlen;
70         struct ipcomp_data *ipcd = x->data;
71         u8 *start, *scratch;
72         struct crypto_comp *tfm;
73         int cpu;
74
75         if (skb_linearize_cow(skb))
76                 goto out;
77
78         skb->ip_summed = CHECKSUM_NONE;
79
80         /* Remove ipcomp header and decompress original payload */
81         ipch = (void *)skb->data;
82         nexthdr = ipch->nexthdr;
83
84         skb->transport_header = skb->network_header + sizeof(*ipch);
85         __skb_pull(skb, sizeof(*ipch));
86
87         /* decompression */
88         plen = skb->len;
89         dlen = IPCOMP_SCRATCH_SIZE;
90         start = skb->data;
91
92         cpu = get_cpu();
93         scratch = *per_cpu_ptr(ipcomp6_scratches, cpu);
94         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
95
96         err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
97         if (err)
98                 goto out_put_cpu;
99
100         if (dlen < (plen + sizeof(*ipch))) {
101                 err = -EINVAL;
102                 goto out_put_cpu;
103         }
104
105         err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
106         if (err) {
107                 goto out_put_cpu;
108         }
109
110         skb->truesize += dlen - plen;
111         __skb_put(skb, dlen - plen);
112         skb_copy_to_linear_data(skb, scratch, dlen);
113         err = nexthdr;
114
115 out_put_cpu:
116         put_cpu();
117 out:
118         return err;
119 }
120
121 static int ipcomp6_output(struct xfrm_state *x, struct sk_buff *skb)
122 {
123         int err;
124         struct ip_comp_hdr *ipch;
125         struct ipcomp_data *ipcd = x->data;
126         int plen, dlen;
127         u8 *start, *scratch;
128         struct crypto_comp *tfm;
129         int cpu;
130
131         /* check whether datagram len is larger than threshold */
132         if (skb->len < ipcd->threshold) {
133                 goto out_ok;
134         }
135
136         if (skb_linearize_cow(skb))
137                 goto out_ok;
138
139         /* compression */
140         plen = skb->len;
141         dlen = IPCOMP_SCRATCH_SIZE;
142         start = skb->data;
143
144         cpu = get_cpu();
145         scratch = *per_cpu_ptr(ipcomp6_scratches, cpu);
146         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
147
148         local_bh_disable();
149         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
150         local_bh_enable();
151         if (err || (dlen + sizeof(*ipch)) >= plen) {
152                 put_cpu();
153                 goto out_ok;
154         }
155         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
156         put_cpu();
157         pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
158
159         /* insert ipcomp header and replace datagram */
160         ipch = ip_comp_hdr(skb);
161         ipch->nexthdr = *skb_mac_header(skb);
162         ipch->flags = 0;
163         ipch->cpi = htons((u16 )ntohl(x->id.spi));
164         *skb_mac_header(skb) = IPPROTO_COMP;
165
166 out_ok:
167         skb_push(skb, -skb_network_offset(skb));
168
169         return 0;
170 }
171
172 static void ipcomp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
173                                 int type, int code, int offset, __be32 info)
174 {
175         __be32 spi;
176         struct ipv6hdr *iph = (struct ipv6hdr*)skb->data;
177         struct ip_comp_hdr *ipcomph =
178                 (struct ip_comp_hdr *)(skb->data + offset);
179         struct xfrm_state *x;
180
181         if (type != ICMPV6_DEST_UNREACH && type != ICMPV6_PKT_TOOBIG)
182                 return;
183
184         spi = htonl(ntohs(ipcomph->cpi));
185         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr, spi, IPPROTO_COMP, AF_INET6);
186         if (!x)
187                 return;
188
189         printk(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/" NIP6_FMT "\n",
190                         spi, NIP6(iph->daddr));
191         xfrm_state_put(x);
192 }
193
194 static struct xfrm_state *ipcomp6_tunnel_create(struct xfrm_state *x)
195 {
196         struct xfrm_state *t = NULL;
197
198         t = xfrm_state_alloc();
199         if (!t)
200                 goto out;
201
202         t->id.proto = IPPROTO_IPV6;
203         t->id.spi = xfrm6_tunnel_alloc_spi((xfrm_address_t *)&x->props.saddr);
204         if (!t->id.spi)
205                 goto error;
206
207         memcpy(t->id.daddr.a6, x->id.daddr.a6, sizeof(struct in6_addr));
208         memcpy(&t->sel, &x->sel, sizeof(t->sel));
209         t->props.family = AF_INET6;
210         t->props.mode = x->props.mode;
211         memcpy(t->props.saddr.a6, x->props.saddr.a6, sizeof(struct in6_addr));
212
213         if (xfrm_init_state(t))
214                 goto error;
215
216         atomic_set(&t->tunnel_users, 1);
217
218 out:
219         return t;
220
221 error:
222         t->km.state = XFRM_STATE_DEAD;
223         xfrm_state_put(t);
224         t = NULL;
225         goto out;
226 }
227
228 static int ipcomp6_tunnel_attach(struct xfrm_state *x)
229 {
230         int err = 0;
231         struct xfrm_state *t = NULL;
232         __be32 spi;
233
234         spi = xfrm6_tunnel_spi_lookup((xfrm_address_t *)&x->props.saddr);
235         if (spi)
236                 t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr,
237                                               spi, IPPROTO_IPV6, AF_INET6);
238         if (!t) {
239                 t = ipcomp6_tunnel_create(x);
240                 if (!t) {
241                         err = -EINVAL;
242                         goto out;
243                 }
244                 xfrm_state_insert(t);
245                 xfrm_state_hold(t);
246         }
247         x->tunnel = t;
248         atomic_inc(&t->tunnel_users);
249
250 out:
251         return err;
252 }
253
254 static void ipcomp6_free_scratches(void)
255 {
256         int i;
257         void **scratches;
258
259         if (--ipcomp6_scratch_users)
260                 return;
261
262         scratches = ipcomp6_scratches;
263         if (!scratches)
264                 return;
265
266         for_each_possible_cpu(i) {
267                 void *scratch = *per_cpu_ptr(scratches, i);
268
269                 vfree(scratch);
270         }
271
272         free_percpu(scratches);
273 }
274
275 static void **ipcomp6_alloc_scratches(void)
276 {
277         int i;
278         void **scratches;
279
280         if (ipcomp6_scratch_users++)
281                 return ipcomp6_scratches;
282
283         scratches = alloc_percpu(void *);
284         if (!scratches)
285                 return NULL;
286
287         ipcomp6_scratches = scratches;
288
289         for_each_possible_cpu(i) {
290                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
291                 if (!scratch)
292                         return NULL;
293                 *per_cpu_ptr(scratches, i) = scratch;
294         }
295
296         return scratches;
297 }
298
299 static void ipcomp6_free_tfms(struct crypto_comp **tfms)
300 {
301         struct ipcomp6_tfms *pos;
302         int cpu;
303
304         list_for_each_entry(pos, &ipcomp6_tfms_list, list) {
305                 if (pos->tfms == tfms)
306                         break;
307         }
308
309         BUG_TRAP(pos);
310
311         if (--pos->users)
312                 return;
313
314         list_del(&pos->list);
315         kfree(pos);
316
317         if (!tfms)
318                 return;
319
320         for_each_possible_cpu(cpu) {
321                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
322                 crypto_free_comp(tfm);
323         }
324         free_percpu(tfms);
325 }
326
327 static struct crypto_comp **ipcomp6_alloc_tfms(const char *alg_name)
328 {
329         struct ipcomp6_tfms *pos;
330         struct crypto_comp **tfms;
331         int cpu;
332
333         /* This can be any valid CPU ID so we don't need locking. */
334         cpu = raw_smp_processor_id();
335
336         list_for_each_entry(pos, &ipcomp6_tfms_list, list) {
337                 struct crypto_comp *tfm;
338
339                 tfms = pos->tfms;
340                 tfm = *per_cpu_ptr(tfms, cpu);
341
342                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
343                         pos->users++;
344                         return tfms;
345                 }
346         }
347
348         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
349         if (!pos)
350                 return NULL;
351
352         pos->users = 1;
353         INIT_LIST_HEAD(&pos->list);
354         list_add(&pos->list, &ipcomp6_tfms_list);
355
356         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
357         if (!tfms)
358                 goto error;
359
360         for_each_possible_cpu(cpu) {
361                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
362                                                             CRYPTO_ALG_ASYNC);
363                 if (IS_ERR(tfm))
364                         goto error;
365                 *per_cpu_ptr(tfms, cpu) = tfm;
366         }
367
368         return tfms;
369
370 error:
371         ipcomp6_free_tfms(tfms);
372         return NULL;
373 }
374
375 static void ipcomp6_free_data(struct ipcomp_data *ipcd)
376 {
377         if (ipcd->tfms)
378                 ipcomp6_free_tfms(ipcd->tfms);
379         ipcomp6_free_scratches();
380 }
381
382 static void ipcomp6_destroy(struct xfrm_state *x)
383 {
384         struct ipcomp_data *ipcd = x->data;
385         if (!ipcd)
386                 return;
387         xfrm_state_delete_tunnel(x);
388         mutex_lock(&ipcomp6_resource_mutex);
389         ipcomp6_free_data(ipcd);
390         mutex_unlock(&ipcomp6_resource_mutex);
391         kfree(ipcd);
392
393         xfrm6_tunnel_free_spi((xfrm_address_t *)&x->props.saddr);
394 }
395
396 static int ipcomp6_init_state(struct xfrm_state *x)
397 {
398         int err;
399         struct ipcomp_data *ipcd;
400         struct xfrm_algo_desc *calg_desc;
401
402         err = -EINVAL;
403         if (!x->calg)
404                 goto out;
405
406         if (x->encap)
407                 goto out;
408
409         x->props.header_len = 0;
410         switch (x->props.mode) {
411         case XFRM_MODE_TRANSPORT:
412                 break;
413         case XFRM_MODE_TUNNEL:
414                 x->props.header_len += sizeof(struct ipv6hdr);
415                 break;
416         default:
417                 goto out;
418         }
419
420         err = -ENOMEM;
421         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
422         if (!ipcd)
423                 goto out;
424
425         mutex_lock(&ipcomp6_resource_mutex);
426         if (!ipcomp6_alloc_scratches())
427                 goto error;
428
429         ipcd->tfms = ipcomp6_alloc_tfms(x->calg->alg_name);
430         if (!ipcd->tfms)
431                 goto error;
432         mutex_unlock(&ipcomp6_resource_mutex);
433
434         if (x->props.mode == XFRM_MODE_TUNNEL) {
435                 err = ipcomp6_tunnel_attach(x);
436                 if (err)
437                         goto error_tunnel;
438         }
439
440         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
441         BUG_ON(!calg_desc);
442         ipcd->threshold = calg_desc->uinfo.comp.threshold;
443         x->data = ipcd;
444         err = 0;
445 out:
446         return err;
447 error_tunnel:
448         mutex_lock(&ipcomp6_resource_mutex);
449 error:
450         ipcomp6_free_data(ipcd);
451         mutex_unlock(&ipcomp6_resource_mutex);
452         kfree(ipcd);
453
454         goto out;
455 }
456
457 static const struct xfrm_type ipcomp6_type =
458 {
459         .description    = "IPCOMP6",
460         .owner          = THIS_MODULE,
461         .proto          = IPPROTO_COMP,
462         .init_state     = ipcomp6_init_state,
463         .destructor     = ipcomp6_destroy,
464         .input          = ipcomp6_input,
465         .output         = ipcomp6_output,
466         .hdr_offset     = xfrm6_find_1stfragopt,
467 };
468
469 static struct inet6_protocol ipcomp6_protocol =
470 {
471         .handler        = xfrm6_rcv,
472         .err_handler    = ipcomp6_err,
473         .flags          = INET6_PROTO_NOPOLICY,
474 };
475
476 static int __init ipcomp6_init(void)
477 {
478         if (xfrm_register_type(&ipcomp6_type, AF_INET6) < 0) {
479                 printk(KERN_INFO "ipcomp6 init: can't add xfrm type\n");
480                 return -EAGAIN;
481         }
482         if (inet6_add_protocol(&ipcomp6_protocol, IPPROTO_COMP) < 0) {
483                 printk(KERN_INFO "ipcomp6 init: can't add protocol\n");
484                 xfrm_unregister_type(&ipcomp6_type, AF_INET6);
485                 return -EAGAIN;
486         }
487         return 0;
488 }
489
490 static void __exit ipcomp6_fini(void)
491 {
492         if (inet6_del_protocol(&ipcomp6_protocol, IPPROTO_COMP) < 0)
493                 printk(KERN_INFO "ipv6 ipcomp close: can't remove protocol\n");
494         if (xfrm_unregister_type(&ipcomp6_type, AF_INET6) < 0)
495                 printk(KERN_INFO "ipv6 ipcomp close: can't remove xfrm type\n");
496 }
497
498 module_init(ipcomp6_init);
499 module_exit(ipcomp6_fini);
500 MODULE_LICENSE("GPL");
501 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) for IPv6 - RFC3173");
502 MODULE_AUTHOR("Mitsuru KANDA <mk@linux-ipv6.org>");
503
504 MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_COMP);