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