[IPCOMP]: Fetch nexthdr before ipch is destroyed
[safe/jmp/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/semaphore.h>
18 #include <linux/crypto.h>
19 #include <linux/err.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         struct ipcomp_data *ipcd = x->data;
47         const int plen = skb->len;
48         int dlen = IPCOMP_SCRATCH_SIZE;
49         const u8 *start = skb->data;
50         const int cpu = get_cpu();
51         u8 *scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
52         struct crypto_comp *tfm = *per_cpu_ptr(ipcd->tfms, cpu);
53         int err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
54
55         if (err)
56                 goto out;
57
58         if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
59                 err = -EINVAL;
60                 goto out;
61         }
62
63         err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
64         if (err)
65                 goto out;
66
67         skb->truesize += dlen - plen;
68         __skb_put(skb, dlen - plen);
69         skb_copy_to_linear_data(skb, scratch, dlen);
70 out:
71         put_cpu();
72         return err;
73 }
74
75 static int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
76 {
77         int nexthdr;
78         int err = -ENOMEM;
79         struct ip_comp_hdr *ipch;
80
81         if (skb_linearize_cow(skb))
82                 goto out;
83
84         skb->ip_summed = CHECKSUM_NONE;
85
86         /* Remove ipcomp header and decompress original payload */
87         ipch = (void *)skb->data;
88         nexthdr = ipch->nexthdr;
89
90         skb->transport_header = skb->network_header + sizeof(*ipch);
91         __skb_pull(skb, sizeof(*ipch));
92         err = ipcomp_decompress(x, skb);
93         if (err)
94                 goto out;
95
96         err = nexthdr;
97
98 out:
99         return err;
100 }
101
102 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
103 {
104         struct ipcomp_data *ipcd = x->data;
105         const int plen = skb->len;
106         int dlen = IPCOMP_SCRATCH_SIZE;
107         u8 *start = skb->data;
108         const int cpu = get_cpu();
109         u8 *scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
110         struct crypto_comp *tfm = *per_cpu_ptr(ipcd->tfms, cpu);
111         int err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
112
113         if (err)
114                 goto out;
115
116         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
117                 err = -EMSGSIZE;
118                 goto out;
119         }
120
121         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
122         put_cpu();
123
124         pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
125         return 0;
126
127 out:
128         put_cpu();
129         return err;
130 }
131
132 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
133 {
134         int err;
135         struct ip_comp_hdr *ipch;
136         struct ipcomp_data *ipcd = x->data;
137
138         if (skb->len < ipcd->threshold) {
139                 /* Don't bother compressing */
140                 goto out_ok;
141         }
142
143         if (skb_linearize_cow(skb))
144                 goto out_ok;
145
146         err = ipcomp_compress(x, skb);
147
148         if (err) {
149                 goto out_ok;
150         }
151
152         /* Install ipcomp header, convert into ipcomp datagram. */
153         ipch = ip_comp_hdr(skb);
154         ipch->nexthdr = *skb_mac_header(skb);
155         ipch->flags = 0;
156         ipch->cpi = htons((u16 )ntohl(x->id.spi));
157         *skb_mac_header(skb) = IPPROTO_COMP;
158 out_ok:
159         skb_push(skb, -skb_network_offset(skb));
160         return 0;
161 }
162
163 static void ipcomp4_err(struct sk_buff *skb, u32 info)
164 {
165         __be32 spi;
166         struct iphdr *iph = (struct iphdr *)skb->data;
167         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
168         struct xfrm_state *x;
169
170         if (icmp_hdr(skb)->type != ICMP_DEST_UNREACH ||
171             icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
172                 return;
173
174         spi = htonl(ntohs(ipch->cpi));
175         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
176                               spi, IPPROTO_COMP, AF_INET);
177         if (!x)
178                 return;
179         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
180                  spi, NIPQUAD(iph->daddr));
181         xfrm_state_put(x);
182 }
183
184 /* We always hold one tunnel user reference to indicate a tunnel */
185 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
186 {
187         struct xfrm_state *t;
188
189         t = xfrm_state_alloc();
190         if (t == NULL)
191                 goto out;
192
193         t->id.proto = IPPROTO_IPIP;
194         t->id.spi = x->props.saddr.a4;
195         t->id.daddr.a4 = x->id.daddr.a4;
196         memcpy(&t->sel, &x->sel, sizeof(t->sel));
197         t->props.family = AF_INET;
198         t->props.mode = x->props.mode;
199         t->props.saddr.a4 = x->props.saddr.a4;
200         t->props.flags = x->props.flags;
201
202         if (xfrm_init_state(t))
203                 goto error;
204
205         atomic_set(&t->tunnel_users, 1);
206 out:
207         return t;
208
209 error:
210         t->km.state = XFRM_STATE_DEAD;
211         xfrm_state_put(t);
212         t = NULL;
213         goto out;
214 }
215
216 /*
217  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
218  * always incremented on success.
219  */
220 static int ipcomp_tunnel_attach(struct xfrm_state *x)
221 {
222         int err = 0;
223         struct xfrm_state *t;
224
225         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
226                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
227         if (!t) {
228                 t = ipcomp_tunnel_create(x);
229                 if (!t) {
230                         err = -EINVAL;
231                         goto out;
232                 }
233                 xfrm_state_insert(t);
234                 xfrm_state_hold(t);
235         }
236         x->tunnel = t;
237         atomic_inc(&t->tunnel_users);
238 out:
239         return err;
240 }
241
242 static void ipcomp_free_scratches(void)
243 {
244         int i;
245         void **scratches;
246
247         if (--ipcomp_scratch_users)
248                 return;
249
250         scratches = ipcomp_scratches;
251         if (!scratches)
252                 return;
253
254         for_each_possible_cpu(i)
255                 vfree(*per_cpu_ptr(scratches, i));
256
257         free_percpu(scratches);
258 }
259
260 static void **ipcomp_alloc_scratches(void)
261 {
262         int i;
263         void **scratches;
264
265         if (ipcomp_scratch_users++)
266                 return ipcomp_scratches;
267
268         scratches = alloc_percpu(void *);
269         if (!scratches)
270                 return NULL;
271
272         ipcomp_scratches = scratches;
273
274         for_each_possible_cpu(i) {
275                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
276                 if (!scratch)
277                         return NULL;
278                 *per_cpu_ptr(scratches, i) = scratch;
279         }
280
281         return scratches;
282 }
283
284 static void ipcomp_free_tfms(struct crypto_comp **tfms)
285 {
286         struct ipcomp_tfms *pos;
287         int cpu;
288
289         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
290                 if (pos->tfms == tfms)
291                         break;
292         }
293
294         BUG_TRAP(pos);
295
296         if (--pos->users)
297                 return;
298
299         list_del(&pos->list);
300         kfree(pos);
301
302         if (!tfms)
303                 return;
304
305         for_each_possible_cpu(cpu) {
306                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
307                 crypto_free_comp(tfm);
308         }
309         free_percpu(tfms);
310 }
311
312 static struct crypto_comp **ipcomp_alloc_tfms(const char *alg_name)
313 {
314         struct ipcomp_tfms *pos;
315         struct crypto_comp **tfms;
316         int cpu;
317
318         /* This can be any valid CPU ID so we don't need locking. */
319         cpu = raw_smp_processor_id();
320
321         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
322                 struct crypto_comp *tfm;
323
324                 tfms = pos->tfms;
325                 tfm = *per_cpu_ptr(tfms, cpu);
326
327                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
328                         pos->users++;
329                         return tfms;
330                 }
331         }
332
333         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
334         if (!pos)
335                 return NULL;
336
337         pos->users = 1;
338         INIT_LIST_HEAD(&pos->list);
339         list_add(&pos->list, &ipcomp_tfms_list);
340
341         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
342         if (!tfms)
343                 goto error;
344
345         for_each_possible_cpu(cpu) {
346                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
347                                                             CRYPTO_ALG_ASYNC);
348                 if (IS_ERR(tfm))
349                         goto error;
350                 *per_cpu_ptr(tfms, cpu) = tfm;
351         }
352
353         return tfms;
354
355 error:
356         ipcomp_free_tfms(tfms);
357         return NULL;
358 }
359
360 static void ipcomp_free_data(struct ipcomp_data *ipcd)
361 {
362         if (ipcd->tfms)
363                 ipcomp_free_tfms(ipcd->tfms);
364         ipcomp_free_scratches();
365 }
366
367 static void ipcomp_destroy(struct xfrm_state *x)
368 {
369         struct ipcomp_data *ipcd = x->data;
370         if (!ipcd)
371                 return;
372         xfrm_state_delete_tunnel(x);
373         mutex_lock(&ipcomp_resource_mutex);
374         ipcomp_free_data(ipcd);
375         mutex_unlock(&ipcomp_resource_mutex);
376         kfree(ipcd);
377 }
378
379 static int ipcomp_init_state(struct xfrm_state *x)
380 {
381         int err;
382         struct ipcomp_data *ipcd;
383         struct xfrm_algo_desc *calg_desc;
384
385         err = -EINVAL;
386         if (!x->calg)
387                 goto out;
388
389         if (x->encap)
390                 goto out;
391
392         x->props.header_len = 0;
393         switch (x->props.mode) {
394         case XFRM_MODE_TRANSPORT:
395                 break;
396         case XFRM_MODE_TUNNEL:
397                 x->props.header_len += sizeof(struct iphdr);
398                 break;
399         default:
400                 goto out;
401         }
402
403         err = -ENOMEM;
404         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
405         if (!ipcd)
406                 goto out;
407
408         mutex_lock(&ipcomp_resource_mutex);
409         if (!ipcomp_alloc_scratches())
410                 goto error;
411
412         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
413         if (!ipcd->tfms)
414                 goto error;
415         mutex_unlock(&ipcomp_resource_mutex);
416
417         if (x->props.mode == XFRM_MODE_TUNNEL) {
418                 err = ipcomp_tunnel_attach(x);
419                 if (err)
420                         goto error_tunnel;
421         }
422
423         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
424         BUG_ON(!calg_desc);
425         ipcd->threshold = calg_desc->uinfo.comp.threshold;
426         x->data = ipcd;
427         err = 0;
428 out:
429         return err;
430
431 error_tunnel:
432         mutex_lock(&ipcomp_resource_mutex);
433 error:
434         ipcomp_free_data(ipcd);
435         mutex_unlock(&ipcomp_resource_mutex);
436         kfree(ipcd);
437         goto out;
438 }
439
440 static struct xfrm_type ipcomp_type = {
441         .description    = "IPCOMP4",
442         .owner          = THIS_MODULE,
443         .proto          = IPPROTO_COMP,
444         .init_state     = ipcomp_init_state,
445         .destructor     = ipcomp_destroy,
446         .input          = ipcomp_input,
447         .output         = ipcomp_output
448 };
449
450 static struct net_protocol ipcomp4_protocol = {
451         .handler        =       xfrm4_rcv,
452         .err_handler    =       ipcomp4_err,
453         .no_policy      =       1,
454 };
455
456 static int __init ipcomp4_init(void)
457 {
458         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
459                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
460                 return -EAGAIN;
461         }
462         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
463                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
464                 xfrm_unregister_type(&ipcomp_type, AF_INET);
465                 return -EAGAIN;
466         }
467         return 0;
468 }
469
470 static void __exit ipcomp4_fini(void)
471 {
472         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
473                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
474         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
475                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
476 }
477
478 module_init(ipcomp4_init);
479 module_exit(ipcomp4_fini);
480
481 MODULE_LICENSE("GPL");
482 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
483 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
484
485 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_COMP);