powerpc: Set init_bootmem_done on NUMA platforms as well
[linux-2.6] / net / xfrm / xfrm_input.c
1 /*
2  * xfrm_input.c
3  *
4  * Changes:
5  *      YOSHIFUJI Hideaki @USAGI
6  *              Split up af-specific portion
7  *
8  */
9
10 #include <linux/slab.h>
11 #include <linux/module.h>
12 #include <linux/netdevice.h>
13 #include <net/dst.h>
14 #include <net/ip.h>
15 #include <net/xfrm.h>
16
17 static struct kmem_cache *secpath_cachep __read_mostly;
18
19 void __secpath_destroy(struct sec_path *sp)
20 {
21         int i;
22         for (i = 0; i < sp->len; i++)
23                 xfrm_state_put(sp->xvec[i]);
24         kmem_cache_free(secpath_cachep, sp);
25 }
26 EXPORT_SYMBOL(__secpath_destroy);
27
28 struct sec_path *secpath_dup(struct sec_path *src)
29 {
30         struct sec_path *sp;
31
32         sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
33         if (!sp)
34                 return NULL;
35
36         sp->len = 0;
37         if (src) {
38                 int i;
39
40                 memcpy(sp, src, sizeof(*sp));
41                 for (i = 0; i < sp->len; i++)
42                         xfrm_state_hold(sp->xvec[i]);
43         }
44         atomic_set(&sp->refcnt, 1);
45         return sp;
46 }
47 EXPORT_SYMBOL(secpath_dup);
48
49 /* Fetch spi and seq from ipsec header */
50
51 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
52 {
53         int offset, offset_seq;
54         int hlen;
55
56         switch (nexthdr) {
57         case IPPROTO_AH:
58                 hlen = sizeof(struct ip_auth_hdr);
59                 offset = offsetof(struct ip_auth_hdr, spi);
60                 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
61                 break;
62         case IPPROTO_ESP:
63                 hlen = sizeof(struct ip_esp_hdr);
64                 offset = offsetof(struct ip_esp_hdr, spi);
65                 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
66                 break;
67         case IPPROTO_COMP:
68                 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
69                         return -EINVAL;
70                 *spi = htonl(ntohs(*(__be16*)(skb_transport_header(skb) + 2)));
71                 *seq = 0;
72                 return 0;
73         default:
74                 return 1;
75         }
76
77         if (!pskb_may_pull(skb, hlen))
78                 return -EINVAL;
79
80         *spi = *(__be32*)(skb_transport_header(skb) + offset);
81         *seq = *(__be32*)(skb_transport_header(skb) + offset_seq);
82         return 0;
83 }
84
85 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
86 {
87         struct xfrm_mode *inner_mode = x->inner_mode;
88         int err;
89
90         err = x->outer_mode->afinfo->extract_input(x, skb);
91         if (err)
92                 return err;
93
94         if (x->sel.family == AF_UNSPEC) {
95                 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
96                 if (inner_mode == NULL)
97                         return -EAFNOSUPPORT;
98         }
99
100         skb->protocol = inner_mode->afinfo->eth_proto;
101         return inner_mode->input2(x, skb);
102 }
103 EXPORT_SYMBOL(xfrm_prepare_input);
104
105 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
106 {
107         struct net *net = dev_net(skb->dev);
108         int err;
109         __be32 seq;
110         struct xfrm_state *x;
111         xfrm_address_t *daddr;
112         struct xfrm_mode *inner_mode;
113         unsigned int family;
114         int decaps = 0;
115         int async = 0;
116
117         /* A negative encap_type indicates async resumption. */
118         if (encap_type < 0) {
119                 async = 1;
120                 x = xfrm_input_state(skb);
121                 seq = XFRM_SKB_CB(skb)->seq.input;
122                 goto resume;
123         }
124
125         /* Allocate new secpath or COW existing one. */
126         if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
127                 struct sec_path *sp;
128
129                 sp = secpath_dup(skb->sp);
130                 if (!sp) {
131                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
132                         goto drop;
133                 }
134                 if (skb->sp)
135                         secpath_put(skb->sp);
136                 skb->sp = sp;
137         }
138
139         daddr = (xfrm_address_t *)(skb_network_header(skb) +
140                                    XFRM_SPI_SKB_CB(skb)->daddroff);
141         family = XFRM_SPI_SKB_CB(skb)->family;
142
143         seq = 0;
144         if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
145                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
146                 goto drop;
147         }
148
149         do {
150                 if (skb->sp->len == XFRM_MAX_DEPTH) {
151                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
152                         goto drop;
153                 }
154
155                 x = xfrm_state_lookup(net, daddr, spi, nexthdr, family);
156                 if (x == NULL) {
157                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
158                         xfrm_audit_state_notfound(skb, family, spi, seq);
159                         goto drop;
160                 }
161
162                 skb->sp->xvec[skb->sp->len++] = x;
163
164                 spin_lock(&x->lock);
165                 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
166                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEINVALID);
167                         goto drop_unlock;
168                 }
169
170                 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
171                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
172                         goto drop_unlock;
173                 }
174
175                 if (x->props.replay_window && xfrm_replay_check(x, skb, seq)) {
176                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
177                         goto drop_unlock;
178                 }
179
180                 if (xfrm_state_check_expire(x)) {
181                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
182                         goto drop_unlock;
183                 }
184
185                 spin_unlock(&x->lock);
186
187                 XFRM_SKB_CB(skb)->seq.input = seq;
188
189                 nexthdr = x->type->input(x, skb);
190
191                 if (nexthdr == -EINPROGRESS)
192                         return 0;
193
194 resume:
195                 spin_lock(&x->lock);
196                 if (nexthdr <= 0) {
197                         if (nexthdr == -EBADMSG) {
198                                 xfrm_audit_state_icvfail(x, skb,
199                                                          x->type->proto);
200                                 x->stats.integrity_failed++;
201                         }
202                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
203                         goto drop_unlock;
204                 }
205
206                 /* only the first xfrm gets the encap type */
207                 encap_type = 0;
208
209                 if (x->props.replay_window)
210                         xfrm_replay_advance(x, seq);
211
212                 x->curlft.bytes += skb->len;
213                 x->curlft.packets++;
214
215                 spin_unlock(&x->lock);
216
217                 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
218
219                 inner_mode = x->inner_mode;
220
221                 if (x->sel.family == AF_UNSPEC) {
222                         inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
223                         if (inner_mode == NULL)
224                                 goto drop;
225                 }
226
227                 if (inner_mode->input(x, skb)) {
228                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
229                         goto drop;
230                 }
231
232                 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
233                         decaps = 1;
234                         break;
235                 }
236
237                 /*
238                  * We need the inner address.  However, we only get here for
239                  * transport mode so the outer address is identical.
240                  */
241                 daddr = &x->id.daddr;
242                 family = x->outer_mode->afinfo->family;
243
244                 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
245                 if (err < 0) {
246                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
247                         goto drop;
248                 }
249         } while (!err);
250
251         nf_reset(skb);
252
253         if (decaps) {
254                 dst_release(skb->dst);
255                 skb->dst = NULL;
256                 netif_rx(skb);
257                 return 0;
258         } else {
259                 return x->inner_mode->afinfo->transport_finish(skb, async);
260         }
261
262 drop_unlock:
263         spin_unlock(&x->lock);
264 drop:
265         kfree_skb(skb);
266         return 0;
267 }
268 EXPORT_SYMBOL(xfrm_input);
269
270 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
271 {
272         return xfrm_input(skb, nexthdr, 0, -1);
273 }
274 EXPORT_SYMBOL(xfrm_input_resume);
275
276 void __init xfrm_input_init(void)
277 {
278         secpath_cachep = kmem_cache_create("secpath_cache",
279                                            sizeof(struct sec_path),
280                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
281                                            NULL);
282 }