Merge master.kernel.org:/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[linux-2.6] / net / ipv4 / ipcomp.c
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/module.h>
17 #include <asm/scatterlist.h>
18 #include <asm/semaphore.h>
19 #include <linux/crypto.h>
20 #include <linux/pfkeyv2.h>
21 #include <linux/percpu.h>
22 #include <linux/smp.h>
23 #include <linux/list.h>
24 #include <linux/vmalloc.h>
25 #include <linux/rtnetlink.h>
26 #include <linux/mutex.h>
27 #include <net/ip.h>
28 #include <net/xfrm.h>
29 #include <net/icmp.h>
30 #include <net/ipcomp.h>
31 #include <net/protocol.h>
32
33 struct ipcomp_tfms {
34         struct list_head list;
35         struct crypto_comp **tfms;
36         int users;
37 };
38
39 static DEFINE_MUTEX(ipcomp_resource_mutex);
40 static void **ipcomp_scratches;
41 static int ipcomp_scratch_users;
42 static LIST_HEAD(ipcomp_tfms_list);
43
44 static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
45 {
46         int err, plen, dlen;
47         struct ipcomp_data *ipcd = x->data;
48         u8 *start, *scratch;
49         struct crypto_comp *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->truesize += dlen - plen;
74         __skb_put(skb, dlen - plen);
75         memcpy(skb->data, scratch, dlen);
76 out:
77         put_cpu();
78         return err;
79 }
80
81 static int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
82 {
83         int err = -ENOMEM;
84         struct iphdr *iph;
85         struct ip_comp_hdr *ipch;
86
87         if (skb_linearize_cow(skb))
88                 goto out;
89
90         skb->ip_summed = CHECKSUM_NONE;
91
92         /* Remove ipcomp header and decompress original payload */
93         iph = skb->nh.iph;
94         ipch = (void *)skb->data;
95         iph->protocol = ipch->nexthdr;
96         skb->h.raw = skb->nh.raw + sizeof(*ipch);
97         __skb_pull(skb, sizeof(*ipch));
98         err = ipcomp_decompress(x, skb);
99
100 out:
101         return err;
102 }
103
104 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
105 {
106         int err, plen, dlen, ihlen;
107         struct iphdr *iph = skb->nh.iph;
108         struct ipcomp_data *ipcd = x->data;
109         u8 *start, *scratch;
110         struct crypto_comp *tfm;
111         int cpu;
112
113         ihlen = iph->ihl * 4;
114         plen = skb->len - ihlen;
115         dlen = IPCOMP_SCRATCH_SIZE;
116         start = skb->data + ihlen;
117
118         cpu = get_cpu();
119         scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
120         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
121
122         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
123         if (err)
124                 goto out;
125
126         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
127                 err = -EMSGSIZE;
128                 goto out;
129         }
130
131         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
132         put_cpu();
133
134         pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
135         return 0;
136
137 out:
138         put_cpu();
139         return err;
140 }
141
142 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
143 {
144         int err;
145         struct iphdr *iph;
146         struct ip_comp_hdr *ipch;
147         struct ipcomp_data *ipcd = x->data;
148         int hdr_len = 0;
149
150         iph = skb->nh.iph;
151         iph->tot_len = htons(skb->len);
152         hdr_len = iph->ihl * 4;
153         if ((skb->len - hdr_len) < ipcd->threshold) {
154                 /* Don't bother compressing */
155                 goto out_ok;
156         }
157
158         if (skb_linearize_cow(skb))
159                 goto out_ok;
160
161         err = ipcomp_compress(x, skb);
162         iph = skb->nh.iph;
163
164         if (err) {
165                 goto out_ok;
166         }
167
168         /* Install ipcomp header, convert into ipcomp datagram. */
169         iph->tot_len = htons(skb->len);
170         ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
171         ipch->nexthdr = iph->protocol;
172         ipch->flags = 0;
173         ipch->cpi = htons((u16 )ntohl(x->id.spi));
174         iph->protocol = IPPROTO_COMP;
175         ip_send_check(iph);
176         return 0;
177
178 out_ok:
179         if (x->props.mode == XFRM_MODE_TUNNEL)
180                 ip_send_check(iph);
181         return 0;
182 }
183
184 static void ipcomp4_err(struct sk_buff *skb, u32 info)
185 {
186         __be32 spi;
187         struct iphdr *iph = (struct iphdr *)skb->data;
188         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
189         struct xfrm_state *x;
190
191         if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
192             skb->h.icmph->code != ICMP_FRAG_NEEDED)
193                 return;
194
195         spi = htonl(ntohs(ipch->cpi));
196         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
197                               spi, IPPROTO_COMP, AF_INET);
198         if (!x)
199                 return;
200         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
201                  spi, NIPQUAD(iph->daddr));
202         xfrm_state_put(x);
203 }
204
205 /* We always hold one tunnel user reference to indicate a tunnel */
206 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
207 {
208         struct xfrm_state *t;
209         u8 mode = XFRM_MODE_TUNNEL;
210
211         t = xfrm_state_alloc();
212         if (t == NULL)
213                 goto out;
214
215         t->id.proto = IPPROTO_IPIP;
216         t->id.spi = x->props.saddr.a4;
217         t->id.daddr.a4 = x->id.daddr.a4;
218         memcpy(&t->sel, &x->sel, sizeof(t->sel));
219         t->props.family = AF_INET;
220         if (x->props.mode == XFRM_MODE_BEET)
221                 mode = x->props.mode;
222         t->props.mode = mode;
223         t->props.saddr.a4 = x->props.saddr.a4;
224         t->props.flags = x->props.flags;
225
226         if (xfrm_init_state(t))
227                 goto error;
228
229         atomic_set(&t->tunnel_users, 1);
230 out:
231         return t;
232
233 error:
234         t->km.state = XFRM_STATE_DEAD;
235         xfrm_state_put(t);
236         t = NULL;
237         goto out;
238 }
239
240 /*
241  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
242  * always incremented on success.
243  */
244 static int ipcomp_tunnel_attach(struct xfrm_state *x)
245 {
246         int err = 0;
247         struct xfrm_state *t;
248
249         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
250                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
251         if (!t) {
252                 t = ipcomp_tunnel_create(x);
253                 if (!t) {
254                         err = -EINVAL;
255                         goto out;
256                 }
257                 xfrm_state_insert(t);
258                 xfrm_state_hold(t);
259         }
260         x->tunnel = t;
261         atomic_inc(&t->tunnel_users);
262 out:
263         return err;
264 }
265
266 static void ipcomp_free_scratches(void)
267 {
268         int i;
269         void **scratches;
270
271         if (--ipcomp_scratch_users)
272                 return;
273
274         scratches = ipcomp_scratches;
275         if (!scratches)
276                 return;
277
278         for_each_possible_cpu(i)
279                 vfree(*per_cpu_ptr(scratches, i));
280
281         free_percpu(scratches);
282 }
283
284 static void **ipcomp_alloc_scratches(void)
285 {
286         int i;
287         void **scratches;
288
289         if (ipcomp_scratch_users++)
290                 return ipcomp_scratches;
291
292         scratches = alloc_percpu(void *);
293         if (!scratches)
294                 return NULL;
295
296         ipcomp_scratches = scratches;
297
298         for_each_possible_cpu(i) {
299                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
300                 if (!scratch)
301                         return NULL;
302                 *per_cpu_ptr(scratches, i) = scratch;
303         }
304
305         return scratches;
306 }
307
308 static void ipcomp_free_tfms(struct crypto_comp **tfms)
309 {
310         struct ipcomp_tfms *pos;
311         int cpu;
312
313         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
314                 if (pos->tfms == tfms)
315                         break;
316         }
317
318         BUG_TRAP(pos);
319
320         if (--pos->users)
321                 return;
322
323         list_del(&pos->list);
324         kfree(pos);
325
326         if (!tfms)
327                 return;
328
329         for_each_possible_cpu(cpu) {
330                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
331                 crypto_free_comp(tfm);
332         }
333         free_percpu(tfms);
334 }
335
336 static struct crypto_comp **ipcomp_alloc_tfms(const char *alg_name)
337 {
338         struct ipcomp_tfms *pos;
339         struct crypto_comp **tfms;
340         int cpu;
341
342         /* This can be any valid CPU ID so we don't need locking. */
343         cpu = raw_smp_processor_id();
344
345         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
346                 struct crypto_comp *tfm;
347
348                 tfms = pos->tfms;
349                 tfm = *per_cpu_ptr(tfms, cpu);
350
351                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
352                         pos->users++;
353                         return tfms;
354                 }
355         }
356
357         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
358         if (!pos)
359                 return NULL;
360
361         pos->users = 1;
362         INIT_LIST_HEAD(&pos->list);
363         list_add(&pos->list, &ipcomp_tfms_list);
364
365         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
366         if (!tfms)
367                 goto error;
368
369         for_each_possible_cpu(cpu) {
370                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
371                                                             CRYPTO_ALG_ASYNC);
372                 if (!tfm)
373                         goto error;
374                 *per_cpu_ptr(tfms, cpu) = tfm;
375         }
376
377         return tfms;
378
379 error:
380         ipcomp_free_tfms(tfms);
381         return NULL;
382 }
383
384 static void ipcomp_free_data(struct ipcomp_data *ipcd)
385 {
386         if (ipcd->tfms)
387                 ipcomp_free_tfms(ipcd->tfms);
388         ipcomp_free_scratches();
389 }
390
391 static void ipcomp_destroy(struct xfrm_state *x)
392 {
393         struct ipcomp_data *ipcd = x->data;
394         if (!ipcd)
395                 return;
396         xfrm_state_delete_tunnel(x);
397         mutex_lock(&ipcomp_resource_mutex);
398         ipcomp_free_data(ipcd);
399         mutex_unlock(&ipcomp_resource_mutex);
400         kfree(ipcd);
401 }
402
403 static int ipcomp_init_state(struct xfrm_state *x)
404 {
405         int err;
406         struct ipcomp_data *ipcd;
407         struct xfrm_algo_desc *calg_desc;
408
409         err = -EINVAL;
410         if (!x->calg)
411                 goto out;
412
413         if (x->encap)
414                 goto out;
415
416         err = -ENOMEM;
417         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
418         if (!ipcd)
419                 goto out;
420
421         x->props.header_len = 0;
422         if (x->props.mode == XFRM_MODE_TUNNEL)
423                 x->props.header_len += sizeof(struct iphdr);
424
425         mutex_lock(&ipcomp_resource_mutex);
426         if (!ipcomp_alloc_scratches())
427                 goto error;
428
429         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
430         if (!ipcd->tfms)
431                 goto error;
432         mutex_unlock(&ipcomp_resource_mutex);
433
434         if (x->props.mode == XFRM_MODE_TUNNEL) {
435                 err = ipcomp_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
448 error_tunnel:
449         mutex_lock(&ipcomp_resource_mutex);
450 error:
451         ipcomp_free_data(ipcd);
452         mutex_unlock(&ipcomp_resource_mutex);
453         kfree(ipcd);
454         goto out;
455 }
456
457 static struct xfrm_type ipcomp_type = {
458         .description    = "IPCOMP4",
459         .owner          = THIS_MODULE,
460         .proto          = IPPROTO_COMP,
461         .init_state     = ipcomp_init_state,
462         .destructor     = ipcomp_destroy,
463         .input          = ipcomp_input,
464         .output         = ipcomp_output
465 };
466
467 static struct net_protocol ipcomp4_protocol = {
468         .handler        =       xfrm4_rcv,
469         .err_handler    =       ipcomp4_err,
470         .no_policy      =       1,
471 };
472
473 static int __init ipcomp4_init(void)
474 {
475         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
476                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
477                 return -EAGAIN;
478         }
479         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
480                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
481                 xfrm_unregister_type(&ipcomp_type, AF_INET);
482                 return -EAGAIN;
483         }
484         return 0;
485 }
486
487 static void __exit ipcomp4_fini(void)
488 {
489         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
490                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
491         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
492                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
493 }
494
495 module_init(ipcomp4_init);
496 module_exit(ipcomp4_fini);
497
498 MODULE_LICENSE("GPL");
499 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
500 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
501