[PATCH] for_each_possible_cpu: network codes
[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/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, struct sk_buff *skb)
85 {
86         u8 nexthdr;
87         int err = 0;
88         struct iphdr *iph;
89         union {
90                 struct iphdr    iph;
91                 char            buf[60];
92         } tmp_iph;
93
94
95         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
96             skb_linearize(skb, GFP_ATOMIC) != 0) {
97                 err = -ENOMEM;
98                 goto out;
99         }
100
101         skb->ip_summed = CHECKSUM_NONE;
102
103         /* Remove ipcomp header and decompress original payload */      
104         iph = skb->nh.iph;
105         memcpy(&tmp_iph, iph, iph->ihl * 4);
106         nexthdr = *(u8 *)skb->data;
107         skb_pull(skb, sizeof(struct ip_comp_hdr));
108         skb->nh.raw += sizeof(struct ip_comp_hdr);
109         memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
110         iph = skb->nh.iph;
111         iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
112         iph->protocol = nexthdr;
113         skb->h.raw = skb->data;
114         err = ipcomp_decompress(x, skb);
115
116 out:    
117         return err;
118 }
119
120 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
121 {
122         int err, plen, dlen, ihlen;
123         struct iphdr *iph = skb->nh.iph;
124         struct ipcomp_data *ipcd = x->data;
125         u8 *start, *scratch;
126         struct crypto_tfm *tfm;
127         int cpu;
128         
129         ihlen = iph->ihl * 4;
130         plen = skb->len - ihlen;
131         dlen = IPCOMP_SCRATCH_SIZE;
132         start = skb->data + ihlen;
133
134         cpu = get_cpu();
135         scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
136         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
137
138         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
139         if (err)
140                 goto out;
141
142         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
143                 err = -EMSGSIZE;
144                 goto out;
145         }
146         
147         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
148         put_cpu();
149
150         pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
151         return 0;
152         
153 out:    
154         put_cpu();
155         return err;
156 }
157
158 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
159 {
160         int err;
161         struct iphdr *iph;
162         struct ip_comp_hdr *ipch;
163         struct ipcomp_data *ipcd = x->data;
164         int hdr_len = 0;
165
166         iph = skb->nh.iph;
167         iph->tot_len = htons(skb->len);
168         hdr_len = iph->ihl * 4;
169         if ((skb->len - hdr_len) < ipcd->threshold) {
170                 /* Don't bother compressing */
171                 goto out_ok;
172         }
173
174         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
175             skb_linearize(skb, GFP_ATOMIC) != 0) {
176                 goto out_ok;
177         }
178         
179         err = ipcomp_compress(x, skb);
180         iph = skb->nh.iph;
181
182         if (err) {
183                 goto out_ok;
184         }
185
186         /* Install ipcomp header, convert into ipcomp datagram. */
187         iph->tot_len = htons(skb->len);
188         ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
189         ipch->nexthdr = iph->protocol;
190         ipch->flags = 0;
191         ipch->cpi = htons((u16 )ntohl(x->id.spi));
192         iph->protocol = IPPROTO_COMP;
193         ip_send_check(iph);
194         return 0;
195
196 out_ok:
197         if (x->props.mode)
198                 ip_send_check(iph);
199         return 0;
200 }
201
202 static void ipcomp4_err(struct sk_buff *skb, u32 info)
203 {
204         u32 spi;
205         struct iphdr *iph = (struct iphdr *)skb->data;
206         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
207         struct xfrm_state *x;
208
209         if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
210             skb->h.icmph->code != ICMP_FRAG_NEEDED)
211                 return;
212
213         spi = ntohl(ntohs(ipch->cpi));
214         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
215                               spi, IPPROTO_COMP, AF_INET);
216         if (!x)
217                 return;
218         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
219                  spi, NIPQUAD(iph->daddr));
220         xfrm_state_put(x);
221 }
222
223 /* We always hold one tunnel user reference to indicate a tunnel */ 
224 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
225 {
226         struct xfrm_state *t;
227         
228         t = xfrm_state_alloc();
229         if (t == NULL)
230                 goto out;
231
232         t->id.proto = IPPROTO_IPIP;
233         t->id.spi = x->props.saddr.a4;
234         t->id.daddr.a4 = x->id.daddr.a4;
235         memcpy(&t->sel, &x->sel, sizeof(t->sel));
236         t->props.family = AF_INET;
237         t->props.mode = 1;
238         t->props.saddr.a4 = x->props.saddr.a4;
239         t->props.flags = x->props.flags;
240
241         if (xfrm_init_state(t))
242                 goto error;
243
244         atomic_set(&t->tunnel_users, 1);
245 out:
246         return t;
247
248 error:
249         t->km.state = XFRM_STATE_DEAD;
250         xfrm_state_put(t);
251         t = NULL;
252         goto out;
253 }
254
255 /*
256  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
257  * always incremented on success.
258  */
259 static int ipcomp_tunnel_attach(struct xfrm_state *x)
260 {
261         int err = 0;
262         struct xfrm_state *t;
263
264         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
265                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
266         if (!t) {
267                 t = ipcomp_tunnel_create(x);
268                 if (!t) {
269                         err = -EINVAL;
270                         goto out;
271                 }
272                 xfrm_state_insert(t);
273                 xfrm_state_hold(t);
274         }
275         x->tunnel = t;
276         atomic_inc(&t->tunnel_users);
277 out:
278         return err;
279 }
280
281 static void ipcomp_free_scratches(void)
282 {
283         int i;
284         void **scratches;
285
286         if (--ipcomp_scratch_users)
287                 return;
288
289         scratches = ipcomp_scratches;
290         if (!scratches)
291                 return;
292
293         for_each_possible_cpu(i) {
294                 void *scratch = *per_cpu_ptr(scratches, i);
295                 if (scratch)
296                         vfree(scratch);
297         }
298
299         free_percpu(scratches);
300 }
301
302 static void **ipcomp_alloc_scratches(void)
303 {
304         int i;
305         void **scratches;
306
307         if (ipcomp_scratch_users++)
308                 return ipcomp_scratches;
309
310         scratches = alloc_percpu(void *);
311         if (!scratches)
312                 return NULL;
313
314         ipcomp_scratches = scratches;
315
316         for_each_possible_cpu(i) {
317                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
318                 if (!scratch)
319                         return NULL;
320                 *per_cpu_ptr(scratches, i) = scratch;
321         }
322
323         return scratches;
324 }
325
326 static void ipcomp_free_tfms(struct crypto_tfm **tfms)
327 {
328         struct ipcomp_tfms *pos;
329         int cpu;
330
331         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
332                 if (pos->tfms == tfms)
333                         break;
334         }
335
336         BUG_TRAP(pos);
337
338         if (--pos->users)
339                 return;
340
341         list_del(&pos->list);
342         kfree(pos);
343
344         if (!tfms)
345                 return;
346
347         for_each_possible_cpu(cpu) {
348                 struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
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_possible_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         mutex_lock(&ipcomp_resource_mutex);
415         ipcomp_free_data(ipcd);
416         mutex_unlock(&ipcomp_resource_mutex);
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         mutex_lock(&ipcomp_resource_mutex);
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         mutex_unlock(&ipcomp_resource_mutex);
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         mutex_lock(&ipcomp_resource_mutex);
468 error:
469         ipcomp_free_data(ipcd);
470         mutex_unlock(&ipcomp_resource_mutex);
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