[IPCOMP]: Disable BH on output when using shared tfm
[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/semaphore.h>
18 #include <linux/crypto.h>
19 #include <linux/err.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         skb_copy_to_linear_data(skb, 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 nexthdr;
78         int err = -ENOMEM;
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         ipch = (void *)skb->data;
88         nexthdr = ipch->nexthdr;
89
90         skb->transport_header = skb->network_header + sizeof(*ipch);
91         __skb_pull(skb, sizeof(*ipch));
92         err = ipcomp_decompress(x, skb);
93         if (err)
94                 goto out;
95
96         err = nexthdr;
97
98 out:
99         return err;
100 }
101
102 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
103 {
104         struct ipcomp_data *ipcd = x->data;
105         const int plen = skb->len;
106         int dlen = IPCOMP_SCRATCH_SIZE;
107         u8 *start = skb->data;
108         const int cpu = get_cpu();
109         u8 *scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
110         struct crypto_comp *tfm = *per_cpu_ptr(ipcd->tfms, cpu);
111         int err;
112
113         local_bh_disable();
114         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
115         local_bh_enable();
116         if (err)
117                 goto out;
118
119         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
120                 err = -EMSGSIZE;
121                 goto out;
122         }
123
124         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
125         put_cpu();
126
127         pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
128         return 0;
129
130 out:
131         put_cpu();
132         return err;
133 }
134
135 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
136 {
137         int err;
138         struct ip_comp_hdr *ipch;
139         struct ipcomp_data *ipcd = x->data;
140
141         if (skb->len < ipcd->threshold) {
142                 /* Don't bother compressing */
143                 goto out_ok;
144         }
145
146         if (skb_linearize_cow(skb))
147                 goto out_ok;
148
149         err = ipcomp_compress(x, skb);
150
151         if (err) {
152                 goto out_ok;
153         }
154
155         /* Install ipcomp header, convert into ipcomp datagram. */
156         ipch = ip_comp_hdr(skb);
157         ipch->nexthdr = *skb_mac_header(skb);
158         ipch->flags = 0;
159         ipch->cpi = htons((u16 )ntohl(x->id.spi));
160         *skb_mac_header(skb) = IPPROTO_COMP;
161 out_ok:
162         skb_push(skb, -skb_network_offset(skb));
163         return 0;
164 }
165
166 static void ipcomp4_err(struct sk_buff *skb, u32 info)
167 {
168         __be32 spi;
169         struct iphdr *iph = (struct iphdr *)skb->data;
170         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
171         struct xfrm_state *x;
172
173         if (icmp_hdr(skb)->type != ICMP_DEST_UNREACH ||
174             icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
175                 return;
176
177         spi = htonl(ntohs(ipch->cpi));
178         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
179                               spi, IPPROTO_COMP, AF_INET);
180         if (!x)
181                 return;
182         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
183                  spi, NIPQUAD(iph->daddr));
184         xfrm_state_put(x);
185 }
186
187 /* We always hold one tunnel user reference to indicate a tunnel */
188 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
189 {
190         struct xfrm_state *t;
191
192         t = xfrm_state_alloc();
193         if (t == NULL)
194                 goto out;
195
196         t->id.proto = IPPROTO_IPIP;
197         t->id.spi = x->props.saddr.a4;
198         t->id.daddr.a4 = x->id.daddr.a4;
199         memcpy(&t->sel, &x->sel, sizeof(t->sel));
200         t->props.family = AF_INET;
201         t->props.mode = x->props.mode;
202         t->props.saddr.a4 = x->props.saddr.a4;
203         t->props.flags = x->props.flags;
204
205         if (xfrm_init_state(t))
206                 goto error;
207
208         atomic_set(&t->tunnel_users, 1);
209 out:
210         return t;
211
212 error:
213         t->km.state = XFRM_STATE_DEAD;
214         xfrm_state_put(t);
215         t = NULL;
216         goto out;
217 }
218
219 /*
220  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
221  * always incremented on success.
222  */
223 static int ipcomp_tunnel_attach(struct xfrm_state *x)
224 {
225         int err = 0;
226         struct xfrm_state *t;
227
228         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
229                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
230         if (!t) {
231                 t = ipcomp_tunnel_create(x);
232                 if (!t) {
233                         err = -EINVAL;
234                         goto out;
235                 }
236                 xfrm_state_insert(t);
237                 xfrm_state_hold(t);
238         }
239         x->tunnel = t;
240         atomic_inc(&t->tunnel_users);
241 out:
242         return err;
243 }
244
245 static void ipcomp_free_scratches(void)
246 {
247         int i;
248         void **scratches;
249
250         if (--ipcomp_scratch_users)
251                 return;
252
253         scratches = ipcomp_scratches;
254         if (!scratches)
255                 return;
256
257         for_each_possible_cpu(i)
258                 vfree(*per_cpu_ptr(scratches, i));
259
260         free_percpu(scratches);
261 }
262
263 static void **ipcomp_alloc_scratches(void)
264 {
265         int i;
266         void **scratches;
267
268         if (ipcomp_scratch_users++)
269                 return ipcomp_scratches;
270
271         scratches = alloc_percpu(void *);
272         if (!scratches)
273                 return NULL;
274
275         ipcomp_scratches = scratches;
276
277         for_each_possible_cpu(i) {
278                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
279                 if (!scratch)
280                         return NULL;
281                 *per_cpu_ptr(scratches, i) = scratch;
282         }
283
284         return scratches;
285 }
286
287 static void ipcomp_free_tfms(struct crypto_comp **tfms)
288 {
289         struct ipcomp_tfms *pos;
290         int cpu;
291
292         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
293                 if (pos->tfms == tfms)
294                         break;
295         }
296
297         BUG_TRAP(pos);
298
299         if (--pos->users)
300                 return;
301
302         list_del(&pos->list);
303         kfree(pos);
304
305         if (!tfms)
306                 return;
307
308         for_each_possible_cpu(cpu) {
309                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
310                 crypto_free_comp(tfm);
311         }
312         free_percpu(tfms);
313 }
314
315 static struct crypto_comp **ipcomp_alloc_tfms(const char *alg_name)
316 {
317         struct ipcomp_tfms *pos;
318         struct crypto_comp **tfms;
319         int cpu;
320
321         /* This can be any valid CPU ID so we don't need locking. */
322         cpu = raw_smp_processor_id();
323
324         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
325                 struct crypto_comp *tfm;
326
327                 tfms = pos->tfms;
328                 tfm = *per_cpu_ptr(tfms, cpu);
329
330                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
331                         pos->users++;
332                         return tfms;
333                 }
334         }
335
336         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
337         if (!pos)
338                 return NULL;
339
340         pos->users = 1;
341         INIT_LIST_HEAD(&pos->list);
342         list_add(&pos->list, &ipcomp_tfms_list);
343
344         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
345         if (!tfms)
346                 goto error;
347
348         for_each_possible_cpu(cpu) {
349                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
350                                                             CRYPTO_ALG_ASYNC);
351                 if (IS_ERR(tfm))
352                         goto error;
353                 *per_cpu_ptr(tfms, cpu) = tfm;
354         }
355
356         return tfms;
357
358 error:
359         ipcomp_free_tfms(tfms);
360         return NULL;
361 }
362
363 static void ipcomp_free_data(struct ipcomp_data *ipcd)
364 {
365         if (ipcd->tfms)
366                 ipcomp_free_tfms(ipcd->tfms);
367         ipcomp_free_scratches();
368 }
369
370 static void ipcomp_destroy(struct xfrm_state *x)
371 {
372         struct ipcomp_data *ipcd = x->data;
373         if (!ipcd)
374                 return;
375         xfrm_state_delete_tunnel(x);
376         mutex_lock(&ipcomp_resource_mutex);
377         ipcomp_free_data(ipcd);
378         mutex_unlock(&ipcomp_resource_mutex);
379         kfree(ipcd);
380 }
381
382 static int ipcomp_init_state(struct xfrm_state *x)
383 {
384         int err;
385         struct ipcomp_data *ipcd;
386         struct xfrm_algo_desc *calg_desc;
387
388         err = -EINVAL;
389         if (!x->calg)
390                 goto out;
391
392         if (x->encap)
393                 goto out;
394
395         x->props.header_len = 0;
396         switch (x->props.mode) {
397         case XFRM_MODE_TRANSPORT:
398                 break;
399         case XFRM_MODE_TUNNEL:
400                 x->props.header_len += sizeof(struct iphdr);
401                 break;
402         default:
403                 goto out;
404         }
405
406         err = -ENOMEM;
407         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
408         if (!ipcd)
409                 goto out;
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 const 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
488 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_COMP);