vserver 1.9.3
[linux-2.6.git] / 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/config.h>
34 #include <linux/module.h>
35 #include <net/ip.h>
36 #include <net/xfrm.h>
37 #include <net/ipcomp.h>
38 #include <asm/scatterlist.h>
39 #include <asm/semaphore.h>
40 #include <linux/crypto.h>
41 #include <linux/pfkeyv2.h>
42 #include <linux/random.h>
43 #include <linux/percpu.h>
44 #include <linux/smp.h>
45 #include <linux/list.h>
46 #include <linux/vmalloc.h>
47 #include <linux/rtnetlink.h>
48 #include <net/icmp.h>
49 #include <net/ipv6.h>
50 #include <linux/ipv6.h>
51 #include <linux/icmpv6.h>
52
53 struct ipcomp6_tfms {
54         struct list_head list;
55         struct crypto_tfm **tfms;
56         int users;
57 };
58
59 static DECLARE_MUTEX(ipcomp6_resource_sem);
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 xfrm_decap_state *decap, struct sk_buff *skb)
65 {
66         int err = 0;
67         u8 nexthdr = 0;
68         int hdr_len = skb->h.raw - skb->nh.raw;
69         unsigned char *tmp_hdr = NULL;
70         struct ipv6hdr *iph;
71         int plen, dlen;
72         struct ipcomp_data *ipcd = x->data;
73         u8 *start, *scratch;
74         struct crypto_tfm *tfm;
75         int cpu;
76
77         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
78                 skb_linearize(skb, GFP_ATOMIC) != 0) {
79                 err = -ENOMEM;
80                 goto out;
81         }
82
83         skb->ip_summed = CHECKSUM_NONE;
84
85         /* Remove ipcomp header and decompress original payload */
86         iph = skb->nh.ipv6h;
87         tmp_hdr = kmalloc(hdr_len, GFP_ATOMIC);
88         if (!tmp_hdr)
89                 goto out;
90         memcpy(tmp_hdr, iph, hdr_len);
91         nexthdr = *(u8 *)skb->data;
92         skb_pull(skb, sizeof(struct ipv6_comp_hdr)); 
93         skb->nh.raw += sizeof(struct ipv6_comp_hdr);
94         memcpy(skb->nh.raw, tmp_hdr, hdr_len);
95         iph = skb->nh.ipv6h;
96         iph->payload_len = htons(ntohs(iph->payload_len) - sizeof(struct ipv6_comp_hdr));
97         skb->h.raw = skb->data;
98
99         /* decompression */
100         plen = skb->len;
101         dlen = IPCOMP_SCRATCH_SIZE;
102         start = skb->data;
103
104         cpu = get_cpu();
105         scratch = *per_cpu_ptr(ipcomp6_scratches, cpu);
106         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
107
108         err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
109         if (err) {
110                 err = -EINVAL;
111                 goto out_put_cpu;
112         }
113
114         if (dlen < (plen + sizeof(struct ipv6_comp_hdr))) {
115                 err = -EINVAL;
116                 goto out_put_cpu;
117         }
118
119         err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
120         if (err) {
121                 goto out_put_cpu;
122         }
123
124         skb_put(skb, dlen - plen);
125         memcpy(skb->data, scratch, dlen);
126
127         iph = skb->nh.ipv6h;
128         iph->payload_len = htons(skb->len);
129         
130 out_put_cpu:
131         put_cpu();
132 out:
133         if (tmp_hdr)
134                 kfree(tmp_hdr);
135         if (err)
136                 goto error_out;
137         return nexthdr;
138 error_out:
139         return err;
140 }
141
142 static int ipcomp6_output(struct sk_buff *skb)
143 {
144         int err;
145         struct dst_entry *dst = skb->dst;
146         struct xfrm_state *x = dst->xfrm;
147         struct ipv6hdr *top_iph;
148         int hdr_len;
149         struct ipv6_comp_hdr *ipch;
150         struct ipcomp_data *ipcd = x->data;
151         int plen, dlen;
152         u8 *start, *scratch;
153         struct crypto_tfm *tfm;
154         int cpu;
155
156         hdr_len = skb->h.raw - skb->data;
157
158         /* check whether datagram len is larger than threshold */
159         if ((skb->len - hdr_len) < ipcd->threshold) {
160                 goto out_ok;
161         }
162
163         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
164                 skb_linearize(skb, GFP_ATOMIC) != 0) {
165                 err = -ENOMEM;
166                 goto error;
167         }
168
169         /* compression */
170         plen = skb->len - hdr_len;
171         dlen = IPCOMP_SCRATCH_SIZE;
172         start = skb->h.raw;
173
174         cpu = get_cpu();
175         scratch = *per_cpu_ptr(ipcomp6_scratches, cpu);
176         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
177
178         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
179         if (err) {
180                 put_cpu();
181                 goto error;
182         }
183         if ((dlen + sizeof(struct ipv6_comp_hdr)) >= plen) {
184                 put_cpu();
185                 goto out_ok;
186         }
187         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
188         put_cpu();
189         pskb_trim(skb, hdr_len + dlen + sizeof(struct ip_comp_hdr));
190
191         /* insert ipcomp header and replace datagram */
192         top_iph = (struct ipv6hdr *)skb->data;
193
194         top_iph->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
195
196         ipch = (struct ipv6_comp_hdr *)start;
197         ipch->nexthdr = *skb->nh.raw;
198         ipch->flags = 0;
199         ipch->cpi = htons((u16 )ntohl(x->id.spi));
200         *skb->nh.raw = IPPROTO_COMP;
201
202 out_ok:
203         err = 0;
204
205 error:
206         return err;
207 }
208
209 static void ipcomp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
210                                 int type, int code, int offset, __u32 info)
211 {
212         u32 spi;
213         struct ipv6hdr *iph = (struct ipv6hdr*)skb->data;
214         struct ipv6_comp_hdr *ipcomph = (struct ipv6_comp_hdr*)(skb->data+offset);
215         struct xfrm_state *x;
216
217         if (type != ICMPV6_DEST_UNREACH && type != ICMPV6_PKT_TOOBIG)
218                 return;
219
220         spi = ntohl(ntohs(ipcomph->cpi));
221         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr, spi, IPPROTO_COMP, AF_INET6);
222         if (!x)
223                 return;
224
225         printk(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/"
226                         "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x\n",
227                         spi, NIP6(iph->daddr));
228         xfrm_state_put(x);
229 }
230
231 static struct xfrm_state *ipcomp6_tunnel_create(struct xfrm_state *x)
232 {
233         struct xfrm_state *t = NULL;
234
235         t = xfrm_state_alloc();
236         if (!t)
237                 goto out;
238
239         t->id.proto = IPPROTO_IPV6;
240         t->id.spi = xfrm6_tunnel_alloc_spi((xfrm_address_t *)&x->props.saddr);
241         memcpy(t->id.daddr.a6, x->id.daddr.a6, sizeof(struct in6_addr));
242         memcpy(&t->sel, &x->sel, sizeof(t->sel));
243         t->props.family = AF_INET6;
244         t->props.mode = 1;
245         memcpy(t->props.saddr.a6, x->props.saddr.a6, sizeof(struct in6_addr));
246
247         t->type = xfrm_get_type(IPPROTO_IPV6, t->props.family);
248         if (t->type == NULL)
249                 goto error;
250
251         if (t->type->init_state(t, NULL))
252                 goto error;
253
254         t->km.state = XFRM_STATE_VALID;
255         atomic_set(&t->tunnel_users, 1);
256
257 out:
258         return t;
259
260 error:
261         xfrm_state_put(t);
262         goto out;
263 }
264
265 static int ipcomp6_tunnel_attach(struct xfrm_state *x)
266 {
267         int err = 0;
268         struct xfrm_state *t = NULL;
269         u32 spi;
270
271         spi = xfrm6_tunnel_spi_lookup((xfrm_address_t *)&x->props.saddr);
272         if (spi)
273                 t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr,
274                                               spi, IPPROTO_IPV6, AF_INET6);
275         if (!t) {
276                 t = ipcomp6_tunnel_create(x);
277                 if (!t) {
278                         err = -EINVAL;
279                         goto out;
280                 }
281                 xfrm_state_insert(t);
282                 xfrm_state_hold(t);
283         }
284         x->tunnel = t;
285         atomic_inc(&t->tunnel_users);
286
287 out:
288         return err;
289 }
290
291 static void ipcomp6_free_scratches(void)
292 {
293         int i;
294         void **scratches;
295
296         if (--ipcomp6_scratch_users)
297                 return;
298
299         scratches = ipcomp6_scratches;
300         if (!scratches)
301                 return;
302
303         for_each_cpu(i) {
304                 void *scratch = *per_cpu_ptr(scratches, i);
305                 if (scratch)
306                         vfree(scratch);
307         }
308
309         free_percpu(scratches);
310 }
311
312 static void **ipcomp6_alloc_scratches(void)
313 {
314         int i;
315         void **scratches;
316
317         if (ipcomp6_scratch_users++)
318                 return ipcomp6_scratches;
319
320         scratches = alloc_percpu(void *);
321         if (!scratches)
322                 return NULL;
323
324         ipcomp6_scratches = scratches;
325
326         for_each_cpu(i) {
327                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
328                 if (!scratch)
329                         return NULL;
330                 *per_cpu_ptr(scratches, i) = scratch;
331         }
332
333         return scratches;
334 }
335
336 static void ipcomp6_free_tfms(struct crypto_tfm **tfms)
337 {
338         struct ipcomp6_tfms *pos;
339         int cpu;
340
341         list_for_each_entry(pos, &ipcomp6_tfms_list, list) {
342                 if (pos->tfms == tfms)
343                         break;
344         }
345
346         BUG_TRAP(pos);
347
348         if (--pos->users)
349                 return;
350
351         list_del(&pos->list);
352         kfree(pos);
353
354         if (!tfms)
355                 return;
356
357         for_each_cpu(cpu) {
358                 struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
359                 if (tfm)
360                         crypto_free_tfm(tfm);
361         }
362         free_percpu(tfms);
363 }
364
365 static struct crypto_tfm **ipcomp6_alloc_tfms(const char *alg_name)
366 {
367         struct ipcomp6_tfms *pos;
368         struct crypto_tfm **tfms;
369         int cpu;
370
371         /* This can be any valid CPU ID so we don't need locking. */
372         cpu = smp_processor_id();
373
374         list_for_each_entry(pos, &ipcomp6_tfms_list, list) {
375                 struct crypto_tfm *tfm;
376
377                 tfms = pos->tfms;
378                 tfm = *per_cpu_ptr(tfms, cpu);
379
380                 if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
381                         pos->users++;
382                         return tfms;
383                 }
384         }
385
386         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
387         if (!pos)
388                 return NULL;
389
390         pos->users = 1;
391         INIT_LIST_HEAD(&pos->list);
392         list_add(&pos->list, &ipcomp6_tfms_list);
393
394         pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
395         if (!tfms)
396                 goto error;
397
398         for_each_cpu(cpu) {
399                 struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
400                 if (!tfm)
401                         goto error;
402                 *per_cpu_ptr(tfms, cpu) = tfm;
403         }
404
405         return tfms;
406
407 error:
408         ipcomp6_free_tfms(tfms);
409         return NULL;
410 }
411
412 static void ipcomp6_free_data(struct ipcomp_data *ipcd)
413 {
414         if (ipcd->tfms)
415                 ipcomp6_free_tfms(ipcd->tfms);
416         ipcomp6_free_scratches();
417 }
418
419 static void ipcomp6_destroy(struct xfrm_state *x)
420 {
421         struct ipcomp_data *ipcd = x->data;
422         if (!ipcd)
423                 return;
424         xfrm_state_delete_tunnel(x);
425         down(&ipcomp6_resource_sem);
426         ipcomp6_free_data(ipcd);
427         up(&ipcomp6_resource_sem);
428         kfree(ipcd);
429
430         xfrm6_tunnel_free_spi((xfrm_address_t *)&x->props.saddr);
431 }
432
433 static int ipcomp6_init_state(struct xfrm_state *x, void *args)
434 {
435         int err;
436         struct ipcomp_data *ipcd;
437         struct xfrm_algo_desc *calg_desc;
438
439         err = -EINVAL;
440         if (!x->calg)
441                 goto out;
442
443         if (x->encap)
444                 goto out;
445
446         err = -ENOMEM;
447         ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
448         if (!ipcd)
449                 goto out;
450
451         memset(ipcd, 0, sizeof(*ipcd));
452         x->props.header_len = 0;
453         if (x->props.mode)
454                 x->props.header_len += sizeof(struct ipv6hdr);
455         
456         down(&ipcomp6_resource_sem);
457         if (!ipcomp6_alloc_scratches())
458                 goto error;
459
460         ipcd->tfms = ipcomp6_alloc_tfms(x->calg->alg_name);
461         if (!ipcd->tfms)
462                 goto error;
463         up(&ipcomp6_resource_sem);
464
465         if (x->props.mode) {
466                 err = ipcomp6_tunnel_attach(x);
467                 if (err)
468                         goto error_tunnel;
469         }
470
471         calg_desc = xfrm_calg_get_byname(x->calg->alg_name);
472         BUG_ON(!calg_desc);
473         ipcd->threshold = calg_desc->uinfo.comp.threshold;
474         x->data = ipcd;
475         err = 0;
476 out:
477         return err;
478 error_tunnel:
479         down(&ipcomp6_resource_sem);
480 error:
481         ipcomp6_free_data(ipcd);
482         up(&ipcomp6_resource_sem);
483         kfree(ipcd);
484
485         goto out;
486 }
487
488 static struct xfrm_type ipcomp6_type = 
489 {
490         .description    = "IPCOMP6",
491         .owner          = THIS_MODULE,
492         .proto          = IPPROTO_COMP,
493         .init_state     = ipcomp6_init_state,
494         .destructor     = ipcomp6_destroy,
495         .input          = ipcomp6_input,
496         .output         = ipcomp6_output,
497 };
498
499 static struct inet6_protocol ipcomp6_protocol = 
500 {
501         .handler        = xfrm6_rcv,
502         .err_handler    = ipcomp6_err,
503         .flags          = INET6_PROTO_NOPOLICY,
504 };
505
506 static int __init ipcomp6_init(void)
507 {
508         if (xfrm_register_type(&ipcomp6_type, AF_INET6) < 0) {
509                 printk(KERN_INFO "ipcomp6 init: can't add xfrm type\n");
510                 return -EAGAIN;
511         }
512         if (inet6_add_protocol(&ipcomp6_protocol, IPPROTO_COMP) < 0) {
513                 printk(KERN_INFO "ipcomp6 init: can't add protocol\n");
514                 xfrm_unregister_type(&ipcomp6_type, AF_INET6);
515                 return -EAGAIN;
516         }
517         return 0;
518 }
519
520 static void __exit ipcomp6_fini(void)
521 {
522         if (inet6_del_protocol(&ipcomp6_protocol, IPPROTO_COMP) < 0) 
523                 printk(KERN_INFO "ipv6 ipcomp close: can't remove protocol\n");
524         if (xfrm_unregister_type(&ipcomp6_type, AF_INET6) < 0)
525                 printk(KERN_INFO "ipv6 ipcomp close: can't remove xfrm type\n");
526 }
527
528 module_init(ipcomp6_init);
529 module_exit(ipcomp6_fini);
530 MODULE_LICENSE("GPL");
531 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) for IPv6 - RFC3173");
532 MODULE_AUTHOR("Mitsuru KANDA <mk@linux-ipv6.org>");
533
534