ftp://ftp.kernel.org/pub/linux/kernel/v2.6/linux-2.6.6.tar.bz2
[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 <linux/crypto.h>
20 #include <linux/pfkeyv2.h>
21 #include <net/inet_ecn.h>
22 #include <net/ip.h>
23 #include <net/xfrm.h>
24 #include <net/icmp.h>
25 #include <net/ipcomp.h>
26
27 static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
28 {
29         int err, plen, dlen;
30         struct iphdr *iph;
31         struct ipcomp_data *ipcd = x->data;
32         u8 *start, *scratch = ipcd->scratch;
33         
34         plen = skb->len;
35         dlen = IPCOMP_SCRATCH_SIZE;
36         start = skb->data;
37
38         err = crypto_comp_decompress(ipcd->tfm, start, plen, scratch, &dlen);
39         if (err)
40                 goto out;
41
42         if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
43                 err = -EINVAL;
44                 goto out;
45         }
46
47         err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
48         if (err)
49                 goto out;
50                 
51         skb_put(skb, dlen - plen);
52         memcpy(skb->data, scratch, dlen);
53         iph = skb->nh.iph;
54         iph->tot_len = htons(dlen + iph->ihl * 4);
55 out:    
56         return err;
57 }
58
59 static int ipcomp_input(struct xfrm_state *x,
60                         struct xfrm_decap_state *decap, struct sk_buff *skb)
61 {
62         u8 nexthdr;
63         int err = 0;
64         struct iphdr *iph;
65         union {
66                 struct iphdr    iph;
67                 char            buf[60];
68         } tmp_iph;
69
70
71         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
72             skb_linearize(skb, GFP_ATOMIC) != 0) {
73                 err = -ENOMEM;
74                 goto out;
75         }
76
77         skb->ip_summed = CHECKSUM_NONE;
78
79         /* Remove ipcomp header and decompress original payload */      
80         iph = skb->nh.iph;
81         memcpy(&tmp_iph, iph, iph->ihl * 4);
82         nexthdr = *(u8 *)skb->data;
83         skb_pull(skb, sizeof(struct ip_comp_hdr));
84         skb->nh.raw += sizeof(struct ip_comp_hdr);
85         memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
86         iph = skb->nh.iph;
87         iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
88         iph->protocol = nexthdr;
89         skb->h.raw = skb->data;
90         err = ipcomp_decompress(x, skb);
91
92 out:    
93         return err;
94 }
95
96 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
97 {
98         int err, plen, dlen, ihlen;
99         struct iphdr *iph = skb->nh.iph;
100         struct ipcomp_data *ipcd = x->data;
101         u8 *start, *scratch = ipcd->scratch;
102         
103         ihlen = iph->ihl * 4;
104         plen = skb->len - ihlen;
105         dlen = IPCOMP_SCRATCH_SIZE;
106         start = skb->data + ihlen;
107
108         err = crypto_comp_compress(ipcd->tfm, start, plen, scratch, &dlen);
109         if (err)
110                 goto out;
111
112         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
113                 err = -EMSGSIZE;
114                 goto out;
115         }
116         
117         memcpy(start, scratch, dlen);
118         pskb_trim(skb, ihlen + dlen);
119         
120 out:    
121         return err;
122 }
123
124 static void ipcomp_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
125 {
126         struct dst_entry *dst = skb->dst;
127         struct iphdr *iph, *top_iph;
128
129         iph = skb->nh.iph;
130         top_iph = (struct iphdr *)skb_push(skb, sizeof(struct iphdr));
131         top_iph->ihl = 5;
132         top_iph->version = 4;
133         top_iph->tos = iph->tos;
134         top_iph->tot_len = htons(skb->len);
135         if (!(iph->frag_off&htons(IP_DF)))
136                 __ip_select_ident(top_iph, dst, 0);
137         top_iph->ttl = iph->ttl;
138         top_iph->check = 0;
139         top_iph->saddr = x->props.saddr.a4;
140         top_iph->daddr = x->id.daddr.a4;
141         top_iph->frag_off = iph->frag_off&~htons(IP_MF|IP_OFFSET);
142         memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
143         skb->nh.raw = skb->data;
144 }
145
146 static int ipcomp_output(struct sk_buff *skb)
147 {
148         int err;
149         struct dst_entry *dst = skb->dst;
150         struct xfrm_state *x = dst->xfrm;
151         struct iphdr *iph, *top_iph;
152         struct ip_comp_hdr *ipch;
153         struct ipcomp_data *ipcd = x->data;
154         union {
155                 struct iphdr    iph;
156                 char            buf[60];
157         } tmp_iph;
158         int hdr_len = 0;
159
160         if (skb->ip_summed == CHECKSUM_HW && skb_checksum_help(skb) == NULL) {
161                 err = -EINVAL;
162                 goto error_nolock;
163         }
164
165         spin_lock_bh(&x->lock);
166         err = xfrm_check_output(x, skb, AF_INET);
167         if (err)
168                 goto error;
169
170         /* Don't bother compressing */
171         if (!x->props.mode) {
172                 iph = skb->nh.iph;
173                 hdr_len = iph->ihl * 4;
174         }
175         if ((skb->len - hdr_len) < ipcd->threshold) {
176                 if (x->props.mode) {
177                         ipcomp_tunnel_encap(x, skb);
178                         iph = skb->nh.iph;
179                         iph->protocol = IPPROTO_IPIP;
180                         ip_send_check(iph);
181                 }
182                 goto out_ok;
183         }
184
185         if (x->props.mode) 
186                 ipcomp_tunnel_encap(x, skb);
187
188         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
189             skb_linearize(skb, GFP_ATOMIC) != 0) {
190                 err = -ENOMEM;
191                 goto error;
192         }
193         
194         err = ipcomp_compress(x, skb);
195         if (err) {
196                 if (err == -EMSGSIZE) {
197                         if (x->props.mode) {
198                                 iph = skb->nh.iph;
199                                 iph->protocol = IPPROTO_IPIP;
200                                 ip_send_check(iph);
201                         }
202                         goto out_ok;
203                 }
204                 goto error;
205         }
206
207         /* Install ipcomp header, convert into ipcomp datagram. */
208         iph = skb->nh.iph;
209         memcpy(&tmp_iph, iph, iph->ihl * 4);
210         top_iph = (struct iphdr *)skb_push(skb, sizeof(struct ip_comp_hdr));
211         memcpy(top_iph, &tmp_iph, iph->ihl * 4);
212         iph = top_iph;
213         if (x->props.mode && (x->props.flags & XFRM_STATE_NOECN))
214                 IP_ECN_clear(iph);
215         iph->tot_len = htons(skb->len);
216         iph->protocol = IPPROTO_COMP;
217         iph->check = 0;
218         ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
219         ipch->nexthdr = x->props.mode ? IPPROTO_IPIP : tmp_iph.iph.protocol;
220         ipch->flags = 0;
221         ipch->cpi = htons((u16 )ntohl(x->id.spi));
222         ip_send_check(iph);
223         skb->nh.raw = skb->data;
224
225 out_ok:
226         x->curlft.bytes += skb->len;
227         x->curlft.packets++;
228         spin_unlock_bh(&x->lock);
229         
230         if ((skb->dst = dst_pop(dst)) == NULL) {
231                 err = -EHOSTUNREACH;
232                 goto error_nolock;
233         }
234         err = NET_XMIT_BYPASS;
235
236 out_exit:
237         return err;
238 error:
239         spin_unlock_bh(&x->lock);
240 error_nolock:
241         kfree_skb(skb);
242         goto out_exit;
243 }
244
245 static void ipcomp4_err(struct sk_buff *skb, u32 info)
246 {
247         u32 spi;
248         struct iphdr *iph = (struct iphdr *)skb->data;
249         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
250         struct xfrm_state *x;
251
252         if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
253             skb->h.icmph->code != ICMP_FRAG_NEEDED)
254                 return;
255
256         spi = ntohl(ntohs(ipch->cpi));
257         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
258                               spi, IPPROTO_COMP, AF_INET);
259         if (!x)
260                 return;
261         NETDEBUG(printk(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
262                spi, NIPQUAD(iph->daddr)));
263         xfrm_state_put(x);
264 }
265
266 /* We always hold one tunnel user reference to indicate a tunnel */ 
267 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
268 {
269         struct xfrm_state *t;
270         
271         t = xfrm_state_alloc();
272         if (t == NULL)
273                 goto out;
274
275         t->id.proto = IPPROTO_IPIP;
276         t->id.spi = x->props.saddr.a4;
277         t->id.daddr.a4 = x->id.daddr.a4;
278         memcpy(&t->sel, &x->sel, sizeof(t->sel));
279         t->props.family = AF_INET;
280         t->props.mode = 1;
281         t->props.saddr.a4 = x->props.saddr.a4;
282         t->props.flags = x->props.flags;
283         
284         t->type = xfrm_get_type(IPPROTO_IPIP, t->props.family);
285         if (t->type == NULL)
286                 goto error;
287                 
288         if (t->type->init_state(t, NULL))
289                 goto error;
290
291         t->km.state = XFRM_STATE_VALID;
292         atomic_set(&t->tunnel_users, 1);
293 out:
294         return t;
295
296 error:
297         t->km.state = XFRM_STATE_DEAD;
298         xfrm_state_put(t);
299         t = NULL;
300         goto out;
301 }
302
303 /*
304  * Must be protected by xfrm_cfg_sem.  State and tunnel user references are
305  * always incremented on success.
306  */
307 static int ipcomp_tunnel_attach(struct xfrm_state *x)
308 {
309         int err = 0;
310         struct xfrm_state *t;
311
312         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
313                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
314         if (!t) {
315                 t = ipcomp_tunnel_create(x);
316                 if (!t) {
317                         err = -EINVAL;
318                         goto out;
319                 }
320                 xfrm_state_insert(t);
321                 xfrm_state_hold(t);
322         }
323         x->tunnel = t;
324         atomic_inc(&t->tunnel_users);
325 out:
326         return err;
327 }
328
329 static void ipcomp_free_data(struct ipcomp_data *ipcd)
330 {
331         if (ipcd->tfm)
332                 crypto_free_tfm(ipcd->tfm);
333         if (ipcd->scratch)
334                 kfree(ipcd->scratch);   
335 }
336
337 static void ipcomp_destroy(struct xfrm_state *x)
338 {
339         struct ipcomp_data *ipcd = x->data;
340         if (!ipcd)
341                 return;
342         ipcomp_free_data(ipcd);
343         kfree(ipcd);
344 }
345
346 static int ipcomp_init_state(struct xfrm_state *x, void *args)
347 {
348         int err;
349         struct ipcomp_data *ipcd;
350         struct xfrm_algo_desc *calg_desc;
351
352         err = -EINVAL;
353         if (!x->calg)
354                 goto out;
355
356         err = -ENOMEM;
357         ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
358         if (!ipcd)
359                 goto error;
360
361         memset(ipcd, 0, sizeof(*ipcd));
362         x->props.header_len = sizeof(struct ip_comp_hdr);
363         if (x->props.mode)
364                 x->props.header_len += sizeof(struct iphdr);
365
366         ipcd->scratch = kmalloc(IPCOMP_SCRATCH_SIZE, GFP_KERNEL);
367         if (!ipcd->scratch)
368                 goto error;
369         
370         ipcd->tfm = crypto_alloc_tfm(x->calg->alg_name, 0);
371         if (!ipcd->tfm)
372                 goto error;
373
374         if (x->props.mode) {
375                 err = ipcomp_tunnel_attach(x);
376                 if (err)
377                         goto error;
378         }
379
380         calg_desc = xfrm_calg_get_byname(x->calg->alg_name);
381         BUG_ON(!calg_desc);
382         ipcd->threshold = calg_desc->uinfo.comp.threshold;
383         x->data = ipcd;
384         err = 0;
385 out:
386         return err;
387
388 error:
389         if (ipcd) {
390                 ipcomp_free_data(ipcd);
391                 kfree(ipcd);
392         }
393         goto out;
394 }
395
396 static struct xfrm_type ipcomp_type = {
397         .description    = "IPCOMP4",
398         .owner          = THIS_MODULE,
399         .proto          = IPPROTO_COMP,
400         .init_state     = ipcomp_init_state,
401         .destructor     = ipcomp_destroy,
402         .input          = ipcomp_input,
403         .output         = ipcomp_output
404 };
405
406 static struct inet_protocol ipcomp4_protocol = {
407         .handler        =       xfrm4_rcv,
408         .err_handler    =       ipcomp4_err,
409         .no_policy      =       1,
410 };
411
412 static int __init ipcomp4_init(void)
413 {
414         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
415                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
416                 return -EAGAIN;
417         }
418         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
419                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
420                 xfrm_unregister_type(&ipcomp_type, AF_INET);
421                 return -EAGAIN;
422         }
423         return 0;
424 }
425
426 static void __exit ipcomp4_fini(void)
427 {
428         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
429                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
430         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
431                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
432 }
433
434 module_init(ipcomp4_init);
435 module_exit(ipcomp4_fini);
436
437 MODULE_LICENSE("GPL");
438 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
439 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
440