[NET] sem2mutex: net/
[linux-2.6.git] / 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/config.h>
17 #include <linux/module.h>
18 #include <asm/scatterlist.h>
19 #include <asm/semaphore.h>
20 #include <linux/crypto.h>
21 #include <linux/pfkeyv2.h>
22 #include <linux/percpu.h>
23 #include <linux/smp.h>
24 #include <linux/list.h>
25 #include <linux/vmalloc.h>
26 #include <linux/rtnetlink.h>
27 #include <linux/mutex.h>
28 #include <net/ip.h>
29 #include <net/xfrm.h>
30 #include <net/icmp.h>
31 #include <net/ipcomp.h>
32 #include <net/protocol.h>
33
34 struct ipcomp_tfms {
35         struct list_head list;
36         struct crypto_tfm **tfms;
37         int users;
38 };
39
40 static DEFINE_MUTEX(ipcomp_resource_mutex);
41 static void **ipcomp_scratches;
42 static int ipcomp_scratch_users;
43 static LIST_HEAD(ipcomp_tfms_list);
44
45 static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
46 {
47         int err, plen, dlen;
48         struct iphdr *iph;
49         struct ipcomp_data *ipcd = x->data;
50         u8 *start, *scratch;
51         struct crypto_tfm *tfm;
52         int cpu;
53         
54         plen = skb->len;
55         dlen = IPCOMP_SCRATCH_SIZE;
56         start = skb->data;
57
58         cpu = get_cpu();
59         scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
60         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
61
62         err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
63         if (err)
64                 goto out;
65
66         if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
67                 err = -EINVAL;
68                 goto out;
69         }
70
71         err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
72         if (err)
73                 goto out;
74                 
75         skb_put(skb, dlen - plen);
76         memcpy(skb->data, scratch, dlen);
77         iph = skb->nh.iph;
78         iph->tot_len = htons(dlen + iph->ihl * 4);
79 out:    
80         put_cpu();
81         return err;
82 }
83
84 static int ipcomp_input(struct xfrm_state *x,
85                         struct xfrm_decap_state *decap, struct sk_buff *skb)
86 {
87         u8 nexthdr;
88         int err = 0;
89         struct iphdr *iph;
90         union {
91                 struct iphdr    iph;
92                 char            buf[60];
93         } tmp_iph;
94
95
96         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
97             skb_linearize(skb, GFP_ATOMIC) != 0) {
98                 err = -ENOMEM;
99                 goto out;
100         }
101
102         skb->ip_summed = CHECKSUM_NONE;
103
104         /* Remove ipcomp header and decompress original payload */      
105         iph = skb->nh.iph;
106         memcpy(&tmp_iph, iph, iph->ihl * 4);
107         nexthdr = *(u8 *)skb->data;
108         skb_pull(skb, sizeof(struct ip_comp_hdr));
109         skb->nh.raw += sizeof(struct ip_comp_hdr);
110         memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
111         iph = skb->nh.iph;
112         iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
113         iph->protocol = nexthdr;
114         skb->h.raw = skb->data;
115         err = ipcomp_decompress(x, skb);
116
117 out:    
118         return err;
119 }
120
121 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
122 {
123         int err, plen, dlen, ihlen;
124         struct iphdr *iph = skb->nh.iph;
125         struct ipcomp_data *ipcd = x->data;
126         u8 *start, *scratch;
127         struct crypto_tfm *tfm;
128         int cpu;
129         
130         ihlen = iph->ihl * 4;
131         plen = skb->len - ihlen;
132         dlen = IPCOMP_SCRATCH_SIZE;
133         start = skb->data + ihlen;
134
135         cpu = get_cpu();
136         scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
137         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
138
139         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
140         if (err)
141                 goto out;
142
143         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
144                 err = -EMSGSIZE;
145                 goto out;
146         }
147         
148         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
149         put_cpu();
150
151         pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
152         return 0;
153         
154 out:    
155         put_cpu();
156         return err;
157 }
158
159 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
160 {
161         int err;
162         struct iphdr *iph;
163         struct ip_comp_hdr *ipch;
164         struct ipcomp_data *ipcd = x->data;
165         int hdr_len = 0;
166
167         iph = skb->nh.iph;
168         iph->tot_len = htons(skb->len);
169         hdr_len = iph->ihl * 4;
170         if ((skb->len - hdr_len) < ipcd->threshold) {
171                 /* Don't bother compressing */
172                 goto out_ok;
173         }
174
175         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
176             skb_linearize(skb, GFP_ATOMIC) != 0) {
177                 goto out_ok;
178         }
179         
180         err = ipcomp_compress(x, skb);
181         iph = skb->nh.iph;
182
183         if (err) {
184                 goto out_ok;
185         }
186
187         /* Install ipcomp header, convert into ipcomp datagram. */
188         iph->tot_len = htons(skb->len);
189         ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
190         ipch->nexthdr = iph->protocol;
191         ipch->flags = 0;
192         ipch->cpi = htons((u16 )ntohl(x->id.spi));
193         iph->protocol = IPPROTO_COMP;
194         ip_send_check(iph);
195         return 0;
196
197 out_ok:
198         if (x->props.mode)
199                 ip_send_check(iph);
200         return 0;
201 }
202
203 static void ipcomp4_err(struct sk_buff *skb, u32 info)
204 {
205         u32 spi;
206         struct iphdr *iph = (struct iphdr *)skb->data;
207         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
208         struct xfrm_state *x;
209
210         if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
211             skb->h.icmph->code != ICMP_FRAG_NEEDED)
212                 return;
213
214         spi = ntohl(ntohs(ipch->cpi));
215         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
216                               spi, IPPROTO_COMP, AF_INET);
217         if (!x)
218                 return;
219         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
220                  spi, NIPQUAD(iph->daddr));
221         xfrm_state_put(x);
222 }
223
224 /* We always hold one tunnel user reference to indicate a tunnel */ 
225 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
226 {
227         struct xfrm_state *t;
228         
229         t = xfrm_state_alloc();
230         if (t == NULL)
231                 goto out;
232
233         t->id.proto = IPPROTO_IPIP;
234         t->id.spi = x->props.saddr.a4;
235         t->id.daddr.a4 = x->id.daddr.a4;
236         memcpy(&t->sel, &x->sel, sizeof(t->sel));
237         t->props.family = AF_INET;
238         t->props.mode = 1;
239         t->props.saddr.a4 = x->props.saddr.a4;
240         t->props.flags = x->props.flags;
241
242         if (xfrm_init_state(t))
243                 goto error;
244
245         atomic_set(&t->tunnel_users, 1);
246 out:
247         return t;
248
249 error:
250         t->km.state = XFRM_STATE_DEAD;
251         xfrm_state_put(t);
252         t = NULL;
253         goto out;
254 }
255
256 /*
257  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
258  * always incremented on success.
259  */
260 static int ipcomp_tunnel_attach(struct xfrm_state *x)
261 {
262         int err = 0;
263         struct xfrm_state *t;
264
265         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
266                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
267         if (!t) {
268                 t = ipcomp_tunnel_create(x);
269                 if (!t) {
270                         err = -EINVAL;
271                         goto out;
272                 }
273                 xfrm_state_insert(t);
274                 xfrm_state_hold(t);
275         }
276         x->tunnel = t;
277         atomic_inc(&t->tunnel_users);
278 out:
279         return err;
280 }
281
282 static void ipcomp_free_scratches(void)
283 {
284         int i;
285         void **scratches;
286
287         if (--ipcomp_scratch_users)
288                 return;
289
290         scratches = ipcomp_scratches;
291         if (!scratches)
292                 return;
293
294         for_each_cpu(i) {
295                 void *scratch = *per_cpu_ptr(scratches, i);
296                 if (scratch)
297                         vfree(scratch);
298         }
299
300         free_percpu(scratches);
301 }
302
303 static void **ipcomp_alloc_scratches(void)
304 {
305         int i;
306         void **scratches;
307
308         if (ipcomp_scratch_users++)
309                 return ipcomp_scratches;
310
311         scratches = alloc_percpu(void *);
312         if (!scratches)
313                 return NULL;
314
315         ipcomp_scratches = scratches;
316
317         for_each_cpu(i) {
318                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
319                 if (!scratch)
320                         return NULL;
321                 *per_cpu_ptr(scratches, i) = scratch;
322         }
323
324         return scratches;
325 }
326
327 static void ipcomp_free_tfms(struct crypto_tfm **tfms)
328 {
329         struct ipcomp_tfms *pos;
330         int cpu;
331
332         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
333                 if (pos->tfms == tfms)
334                         break;
335         }
336
337         BUG_TRAP(pos);
338
339         if (--pos->users)
340                 return;
341
342         list_del(&pos->list);
343         kfree(pos);
344
345         if (!tfms)
346                 return;
347
348         for_each_cpu(cpu) {
349                 struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
350                 crypto_free_tfm(tfm);
351         }
352         free_percpu(tfms);
353 }
354
355 static struct crypto_tfm **ipcomp_alloc_tfms(const char *alg_name)
356 {
357         struct ipcomp_tfms *pos;
358         struct crypto_tfm **tfms;
359         int cpu;
360
361         /* This can be any valid CPU ID so we don't need locking. */
362         cpu = raw_smp_processor_id();
363
364         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
365                 struct crypto_tfm *tfm;
366
367                 tfms = pos->tfms;
368                 tfm = *per_cpu_ptr(tfms, cpu);
369
370                 if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
371                         pos->users++;
372                         return tfms;
373                 }
374         }
375
376         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
377         if (!pos)
378                 return NULL;
379
380         pos->users = 1;
381         INIT_LIST_HEAD(&pos->list);
382         list_add(&pos->list, &ipcomp_tfms_list);
383
384         pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
385         if (!tfms)
386                 goto error;
387
388         for_each_cpu(cpu) {
389                 struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
390                 if (!tfm)
391                         goto error;
392                 *per_cpu_ptr(tfms, cpu) = tfm;
393         }
394
395         return tfms;
396
397 error:
398         ipcomp_free_tfms(tfms);
399         return NULL;
400 }
401
402 static void ipcomp_free_data(struct ipcomp_data *ipcd)
403 {
404         if (ipcd->tfms)
405                 ipcomp_free_tfms(ipcd->tfms);
406         ipcomp_free_scratches();
407 }
408
409 static void ipcomp_destroy(struct xfrm_state *x)
410 {
411         struct ipcomp_data *ipcd = x->data;
412         if (!ipcd)
413                 return;
414         xfrm_state_delete_tunnel(x);
415         mutex_lock(&ipcomp_resource_mutex);
416         ipcomp_free_data(ipcd);
417         mutex_unlock(&ipcomp_resource_mutex);
418         kfree(ipcd);
419 }
420
421 static int ipcomp_init_state(struct xfrm_state *x)
422 {
423         int err;
424         struct ipcomp_data *ipcd;
425         struct xfrm_algo_desc *calg_desc;
426
427         err = -EINVAL;
428         if (!x->calg)
429                 goto out;
430
431         if (x->encap)
432                 goto out;
433
434         err = -ENOMEM;
435         ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
436         if (!ipcd)
437                 goto out;
438
439         memset(ipcd, 0, sizeof(*ipcd));
440         x->props.header_len = 0;
441         if (x->props.mode)
442                 x->props.header_len += sizeof(struct iphdr);
443
444         mutex_lock(&ipcomp_resource_mutex);
445         if (!ipcomp_alloc_scratches())
446                 goto error;
447
448         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
449         if (!ipcd->tfms)
450                 goto error;
451         mutex_unlock(&ipcomp_resource_mutex);
452
453         if (x->props.mode) {
454                 err = ipcomp_tunnel_attach(x);
455                 if (err)
456                         goto error_tunnel;
457         }
458
459         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
460         BUG_ON(!calg_desc);
461         ipcd->threshold = calg_desc->uinfo.comp.threshold;
462         x->data = ipcd;
463         err = 0;
464 out:
465         return err;
466
467 error_tunnel:
468         mutex_lock(&ipcomp_resource_mutex);
469 error:
470         ipcomp_free_data(ipcd);
471         mutex_unlock(&ipcomp_resource_mutex);
472         kfree(ipcd);
473         goto out;
474 }
475
476 static struct xfrm_type ipcomp_type = {
477         .description    = "IPCOMP4",
478         .owner          = THIS_MODULE,
479         .proto          = IPPROTO_COMP,
480         .init_state     = ipcomp_init_state,
481         .destructor     = ipcomp_destroy,
482         .input          = ipcomp_input,
483         .output         = ipcomp_output
484 };
485
486 static struct net_protocol ipcomp4_protocol = {
487         .handler        =       xfrm4_rcv,
488         .err_handler    =       ipcomp4_err,
489         .no_policy      =       1,
490 };
491
492 static int __init ipcomp4_init(void)
493 {
494         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
495                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
496                 return -EAGAIN;
497         }
498         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
499                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
500                 xfrm_unregister_type(&ipcomp_type, AF_INET);
501                 return -EAGAIN;
502         }
503         return 0;
504 }
505
506 static void __exit ipcomp4_fini(void)
507 {
508         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
509                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
510         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
511                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
512 }
513
514 module_init(ipcomp4_init);
515 module_exit(ipcomp4_fini);
516
517 MODULE_LICENSE("GPL");
518 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
519 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
520