[SK_BUFF]: Introduce ip_hdr(), remove skb->nh.iph
[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/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         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         memcpy(skb->data, 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 err = -ENOMEM;
78         struct iphdr *iph;
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         iph = ip_hdr(skb);
88         ipch = (void *)skb->data;
89         iph->protocol = ipch->nexthdr;
90         skb->h.raw = skb->nh.raw + sizeof(*ipch);
91         __skb_pull(skb, sizeof(*ipch));
92         err = ipcomp_decompress(x, skb);
93
94 out:
95         return err;
96 }
97
98 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
99 {
100         struct ipcomp_data *ipcd = x->data;
101         const int ihlen = ip_hdrlen(skb);
102         const int plen = skb->len - ihlen;
103         int dlen = IPCOMP_SCRATCH_SIZE;
104         u8 *start = skb->data + ihlen;
105         const int cpu = get_cpu();
106         u8 *scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
107         struct crypto_comp *tfm = *per_cpu_ptr(ipcd->tfms, cpu);
108         int err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
109
110         if (err)
111                 goto out;
112
113         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
114                 err = -EMSGSIZE;
115                 goto out;
116         }
117
118         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
119         put_cpu();
120
121         pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
122         return 0;
123
124 out:
125         put_cpu();
126         return err;
127 }
128
129 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
130 {
131         int err;
132         struct ip_comp_hdr *ipch;
133         struct ipcomp_data *ipcd = x->data;
134         int hdr_len = 0;
135         struct iphdr *iph = ip_hdr(skb);
136
137         iph->tot_len = htons(skb->len);
138         hdr_len = iph->ihl * 4;
139         if ((skb->len - hdr_len) < ipcd->threshold) {
140                 /* Don't bother compressing */
141                 goto out_ok;
142         }
143
144         if (skb_linearize_cow(skb))
145                 goto out_ok;
146
147         err = ipcomp_compress(x, skb);
148         iph = ip_hdr(skb);
149
150         if (err) {
151                 goto out_ok;
152         }
153
154         /* Install ipcomp header, convert into ipcomp datagram. */
155         iph->tot_len = htons(skb->len);
156         ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
157         ipch->nexthdr = iph->protocol;
158         ipch->flags = 0;
159         ipch->cpi = htons((u16 )ntohl(x->id.spi));
160         iph->protocol = IPPROTO_COMP;
161         ip_send_check(iph);
162         return 0;
163
164 out_ok:
165         if (x->props.mode == XFRM_MODE_TUNNEL)
166                 ip_send_check(iph);
167         return 0;
168 }
169
170 static void ipcomp4_err(struct sk_buff *skb, u32 info)
171 {
172         __be32 spi;
173         struct iphdr *iph = (struct iphdr *)skb->data;
174         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
175         struct xfrm_state *x;
176
177         if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
178             skb->h.icmph->code != ICMP_FRAG_NEEDED)
179                 return;
180
181         spi = htonl(ntohs(ipch->cpi));
182         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
183                               spi, IPPROTO_COMP, AF_INET);
184         if (!x)
185                 return;
186         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
187                  spi, NIPQUAD(iph->daddr));
188         xfrm_state_put(x);
189 }
190
191 /* We always hold one tunnel user reference to indicate a tunnel */
192 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
193 {
194         struct xfrm_state *t;
195         u8 mode = XFRM_MODE_TUNNEL;
196
197         t = xfrm_state_alloc();
198         if (t == NULL)
199                 goto out;
200
201         t->id.proto = IPPROTO_IPIP;
202         t->id.spi = x->props.saddr.a4;
203         t->id.daddr.a4 = x->id.daddr.a4;
204         memcpy(&t->sel, &x->sel, sizeof(t->sel));
205         t->props.family = AF_INET;
206         if (x->props.mode == XFRM_MODE_BEET)
207                 mode = x->props.mode;
208         t->props.mode = mode;
209         t->props.saddr.a4 = x->props.saddr.a4;
210         t->props.flags = x->props.flags;
211
212         if (xfrm_init_state(t))
213                 goto error;
214
215         atomic_set(&t->tunnel_users, 1);
216 out:
217         return t;
218
219 error:
220         t->km.state = XFRM_STATE_DEAD;
221         xfrm_state_put(t);
222         t = NULL;
223         goto out;
224 }
225
226 /*
227  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
228  * always incremented on success.
229  */
230 static int ipcomp_tunnel_attach(struct xfrm_state *x)
231 {
232         int err = 0;
233         struct xfrm_state *t;
234
235         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
236                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
237         if (!t) {
238                 t = ipcomp_tunnel_create(x);
239                 if (!t) {
240                         err = -EINVAL;
241                         goto out;
242                 }
243                 xfrm_state_insert(t);
244                 xfrm_state_hold(t);
245         }
246         x->tunnel = t;
247         atomic_inc(&t->tunnel_users);
248 out:
249         return err;
250 }
251
252 static void ipcomp_free_scratches(void)
253 {
254         int i;
255         void **scratches;
256
257         if (--ipcomp_scratch_users)
258                 return;
259
260         scratches = ipcomp_scratches;
261         if (!scratches)
262                 return;
263
264         for_each_possible_cpu(i)
265                 vfree(*per_cpu_ptr(scratches, i));
266
267         free_percpu(scratches);
268 }
269
270 static void **ipcomp_alloc_scratches(void)
271 {
272         int i;
273         void **scratches;
274
275         if (ipcomp_scratch_users++)
276                 return ipcomp_scratches;
277
278         scratches = alloc_percpu(void *);
279         if (!scratches)
280                 return NULL;
281
282         ipcomp_scratches = scratches;
283
284         for_each_possible_cpu(i) {
285                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
286                 if (!scratch)
287                         return NULL;
288                 *per_cpu_ptr(scratches, i) = scratch;
289         }
290
291         return scratches;
292 }
293
294 static void ipcomp_free_tfms(struct crypto_comp **tfms)
295 {
296         struct ipcomp_tfms *pos;
297         int cpu;
298
299         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
300                 if (pos->tfms == tfms)
301                         break;
302         }
303
304         BUG_TRAP(pos);
305
306         if (--pos->users)
307                 return;
308
309         list_del(&pos->list);
310         kfree(pos);
311
312         if (!tfms)
313                 return;
314
315         for_each_possible_cpu(cpu) {
316                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
317                 crypto_free_comp(tfm);
318         }
319         free_percpu(tfms);
320 }
321
322 static struct crypto_comp **ipcomp_alloc_tfms(const char *alg_name)
323 {
324         struct ipcomp_tfms *pos;
325         struct crypto_comp **tfms;
326         int cpu;
327
328         /* This can be any valid CPU ID so we don't need locking. */
329         cpu = raw_smp_processor_id();
330
331         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
332                 struct crypto_comp *tfm;
333
334                 tfms = pos->tfms;
335                 tfm = *per_cpu_ptr(tfms, cpu);
336
337                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
338                         pos->users++;
339                         return tfms;
340                 }
341         }
342
343         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
344         if (!pos)
345                 return NULL;
346
347         pos->users = 1;
348         INIT_LIST_HEAD(&pos->list);
349         list_add(&pos->list, &ipcomp_tfms_list);
350
351         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
352         if (!tfms)
353                 goto error;
354
355         for_each_possible_cpu(cpu) {
356                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
357                                                             CRYPTO_ALG_ASYNC);
358                 if (!tfm)
359                         goto error;
360                 *per_cpu_ptr(tfms, cpu) = tfm;
361         }
362
363         return tfms;
364
365 error:
366         ipcomp_free_tfms(tfms);
367         return NULL;
368 }
369
370 static void ipcomp_free_data(struct ipcomp_data *ipcd)
371 {
372         if (ipcd->tfms)
373                 ipcomp_free_tfms(ipcd->tfms);
374         ipcomp_free_scratches();
375 }
376
377 static void ipcomp_destroy(struct xfrm_state *x)
378 {
379         struct ipcomp_data *ipcd = x->data;
380         if (!ipcd)
381                 return;
382         xfrm_state_delete_tunnel(x);
383         mutex_lock(&ipcomp_resource_mutex);
384         ipcomp_free_data(ipcd);
385         mutex_unlock(&ipcomp_resource_mutex);
386         kfree(ipcd);
387 }
388
389 static int ipcomp_init_state(struct xfrm_state *x)
390 {
391         int err;
392         struct ipcomp_data *ipcd;
393         struct xfrm_algo_desc *calg_desc;
394
395         err = -EINVAL;
396         if (!x->calg)
397                 goto out;
398
399         if (x->encap)
400                 goto out;
401
402         err = -ENOMEM;
403         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
404         if (!ipcd)
405                 goto out;
406
407         x->props.header_len = 0;
408         if (x->props.mode == XFRM_MODE_TUNNEL)
409                 x->props.header_len += sizeof(struct iphdr);
410
411         mutex_lock(&ipcomp_resource_mutex);
412         if (!ipcomp_alloc_scratches())
413                 goto error;
414
415         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
416         if (!ipcd->tfms)
417                 goto error;
418         mutex_unlock(&ipcomp_resource_mutex);
419
420         if (x->props.mode == XFRM_MODE_TUNNEL) {
421                 err = ipcomp_tunnel_attach(x);
422                 if (err)
423                         goto error_tunnel;
424         }
425
426         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
427         BUG_ON(!calg_desc);
428         ipcd->threshold = calg_desc->uinfo.comp.threshold;
429         x->data = ipcd;
430         err = 0;
431 out:
432         return err;
433
434 error_tunnel:
435         mutex_lock(&ipcomp_resource_mutex);
436 error:
437         ipcomp_free_data(ipcd);
438         mutex_unlock(&ipcomp_resource_mutex);
439         kfree(ipcd);
440         goto out;
441 }
442
443 static struct xfrm_type ipcomp_type = {
444         .description    = "IPCOMP4",
445         .owner          = THIS_MODULE,
446         .proto          = IPPROTO_COMP,
447         .init_state     = ipcomp_init_state,
448         .destructor     = ipcomp_destroy,
449         .input          = ipcomp_input,
450         .output         = ipcomp_output
451 };
452
453 static struct net_protocol ipcomp4_protocol = {
454         .handler        =       xfrm4_rcv,
455         .err_handler    =       ipcomp4_err,
456         .no_policy      =       1,
457 };
458
459 static int __init ipcomp4_init(void)
460 {
461         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
462                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
463                 return -EAGAIN;
464         }
465         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
466                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
467                 xfrm_unregister_type(&ipcomp_type, AF_INET);
468                 return -EAGAIN;
469         }
470         return 0;
471 }
472
473 static void __exit ipcomp4_fini(void)
474 {
475         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
476                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
477         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
478                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
479 }
480
481 module_init(ipcomp4_init);
482 module_exit(ipcomp4_fini);
483
484 MODULE_LICENSE("GPL");
485 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
486 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
487