[IPV4]: inet_diag annotations
[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         u32 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         
210         t = xfrm_state_alloc();
211         if (t == NULL)
212                 goto out;
213
214         t->id.proto = IPPROTO_IPIP;
215         t->id.spi = x->props.saddr.a4;
216         t->id.daddr.a4 = x->id.daddr.a4;
217         memcpy(&t->sel, &x->sel, sizeof(t->sel));
218         t->props.family = AF_INET;
219         t->props.mode = XFRM_MODE_TUNNEL;
220         t->props.saddr.a4 = x->props.saddr.a4;
221         t->props.flags = x->props.flags;
222
223         if (xfrm_init_state(t))
224                 goto error;
225
226         atomic_set(&t->tunnel_users, 1);
227 out:
228         return t;
229
230 error:
231         t->km.state = XFRM_STATE_DEAD;
232         xfrm_state_put(t);
233         t = NULL;
234         goto out;
235 }
236
237 /*
238  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
239  * always incremented on success.
240  */
241 static int ipcomp_tunnel_attach(struct xfrm_state *x)
242 {
243         int err = 0;
244         struct xfrm_state *t;
245
246         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
247                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
248         if (!t) {
249                 t = ipcomp_tunnel_create(x);
250                 if (!t) {
251                         err = -EINVAL;
252                         goto out;
253                 }
254                 xfrm_state_insert(t);
255                 xfrm_state_hold(t);
256         }
257         x->tunnel = t;
258         atomic_inc(&t->tunnel_users);
259 out:
260         return err;
261 }
262
263 static void ipcomp_free_scratches(void)
264 {
265         int i;
266         void **scratches;
267
268         if (--ipcomp_scratch_users)
269                 return;
270
271         scratches = ipcomp_scratches;
272         if (!scratches)
273                 return;
274
275         for_each_possible_cpu(i)
276                 vfree(*per_cpu_ptr(scratches, i));
277
278         free_percpu(scratches);
279 }
280
281 static void **ipcomp_alloc_scratches(void)
282 {
283         int i;
284         void **scratches;
285
286         if (ipcomp_scratch_users++)
287                 return ipcomp_scratches;
288
289         scratches = alloc_percpu(void *);
290         if (!scratches)
291                 return NULL;
292
293         ipcomp_scratches = scratches;
294
295         for_each_possible_cpu(i) {
296                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
297                 if (!scratch)
298                         return NULL;
299                 *per_cpu_ptr(scratches, i) = scratch;
300         }
301
302         return scratches;
303 }
304
305 static void ipcomp_free_tfms(struct crypto_comp **tfms)
306 {
307         struct ipcomp_tfms *pos;
308         int cpu;
309
310         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
311                 if (pos->tfms == tfms)
312                         break;
313         }
314
315         BUG_TRAP(pos);
316
317         if (--pos->users)
318                 return;
319
320         list_del(&pos->list);
321         kfree(pos);
322
323         if (!tfms)
324                 return;
325
326         for_each_possible_cpu(cpu) {
327                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
328                 crypto_free_comp(tfm);
329         }
330         free_percpu(tfms);
331 }
332
333 static struct crypto_comp **ipcomp_alloc_tfms(const char *alg_name)
334 {
335         struct ipcomp_tfms *pos;
336         struct crypto_comp **tfms;
337         int cpu;
338
339         /* This can be any valid CPU ID so we don't need locking. */
340         cpu = raw_smp_processor_id();
341
342         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
343                 struct crypto_comp *tfm;
344
345                 tfms = pos->tfms;
346                 tfm = *per_cpu_ptr(tfms, cpu);
347
348                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
349                         pos->users++;
350                         return tfms;
351                 }
352         }
353
354         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
355         if (!pos)
356                 return NULL;
357
358         pos->users = 1;
359         INIT_LIST_HEAD(&pos->list);
360         list_add(&pos->list, &ipcomp_tfms_list);
361
362         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
363         if (!tfms)
364                 goto error;
365
366         for_each_possible_cpu(cpu) {
367                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
368                                                             CRYPTO_ALG_ASYNC);
369                 if (!tfm)
370                         goto error;
371                 *per_cpu_ptr(tfms, cpu) = tfm;
372         }
373
374         return tfms;
375
376 error:
377         ipcomp_free_tfms(tfms);
378         return NULL;
379 }
380
381 static void ipcomp_free_data(struct ipcomp_data *ipcd)
382 {
383         if (ipcd->tfms)
384                 ipcomp_free_tfms(ipcd->tfms);
385         ipcomp_free_scratches();
386 }
387
388 static void ipcomp_destroy(struct xfrm_state *x)
389 {
390         struct ipcomp_data *ipcd = x->data;
391         if (!ipcd)
392                 return;
393         xfrm_state_delete_tunnel(x);
394         mutex_lock(&ipcomp_resource_mutex);
395         ipcomp_free_data(ipcd);
396         mutex_unlock(&ipcomp_resource_mutex);
397         kfree(ipcd);
398 }
399
400 static int ipcomp_init_state(struct xfrm_state *x)
401 {
402         int err;
403         struct ipcomp_data *ipcd;
404         struct xfrm_algo_desc *calg_desc;
405
406         err = -EINVAL;
407         if (!x->calg)
408                 goto out;
409
410         if (x->encap)
411                 goto out;
412
413         err = -ENOMEM;
414         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
415         if (!ipcd)
416                 goto out;
417
418         x->props.header_len = 0;
419         if (x->props.mode == XFRM_MODE_TUNNEL)
420                 x->props.header_len += sizeof(struct iphdr);
421
422         mutex_lock(&ipcomp_resource_mutex);
423         if (!ipcomp_alloc_scratches())
424                 goto error;
425
426         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
427         if (!ipcd->tfms)
428                 goto error;
429         mutex_unlock(&ipcomp_resource_mutex);
430
431         if (x->props.mode == XFRM_MODE_TUNNEL) {
432                 err = ipcomp_tunnel_attach(x);
433                 if (err)
434                         goto error_tunnel;
435         }
436
437         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
438         BUG_ON(!calg_desc);
439         ipcd->threshold = calg_desc->uinfo.comp.threshold;
440         x->data = ipcd;
441         err = 0;
442 out:
443         return err;
444
445 error_tunnel:
446         mutex_lock(&ipcomp_resource_mutex);
447 error:
448         ipcomp_free_data(ipcd);
449         mutex_unlock(&ipcomp_resource_mutex);
450         kfree(ipcd);
451         goto out;
452 }
453
454 static struct xfrm_type ipcomp_type = {
455         .description    = "IPCOMP4",
456         .owner          = THIS_MODULE,
457         .proto          = IPPROTO_COMP,
458         .init_state     = ipcomp_init_state,
459         .destructor     = ipcomp_destroy,
460         .input          = ipcomp_input,
461         .output         = ipcomp_output
462 };
463
464 static struct net_protocol ipcomp4_protocol = {
465         .handler        =       xfrm4_rcv,
466         .err_handler    =       ipcomp4_err,
467         .no_policy      =       1,
468 };
469
470 static int __init ipcomp4_init(void)
471 {
472         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
473                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
474                 return -EAGAIN;
475         }
476         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
477                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
478                 xfrm_unregister_type(&ipcomp_type, AF_INET);
479                 return -EAGAIN;
480         }
481         return 0;
482 }
483
484 static void __exit ipcomp4_fini(void)
485 {
486         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
487                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
488         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
489                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
490 }
491
492 module_init(ipcomp4_init);
493 module_exit(ipcomp4_fini);
494
495 MODULE_LICENSE("GPL");
496 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
497 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
498