net: check for refcount if pop a stacked dst_entry
[linux-2.6.git] / net / xfrm / xfrm_output.c
1 /*
2  * xfrm_output.c - Common IPsec encapsulation code.
3  *
4  * Copyright (c) 2007 Herbert Xu <herbert@gondor.apana.org.au>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11
12 #include <linux/errno.h>
13 #include <linux/module.h>
14 #include <linux/netdevice.h>
15 #include <linux/netfilter.h>
16 #include <linux/skbuff.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
19 #include <net/dst.h>
20 #include <net/xfrm.h>
21
22 static int xfrm_output2(struct sk_buff *skb);
23
24 static int xfrm_state_check_space(struct xfrm_state *x, struct sk_buff *skb)
25 {
26         struct dst_entry *dst = skb_dst(skb);
27         int nhead = dst->header_len + LL_RESERVED_SPACE(dst->dev)
28                 - skb_headroom(skb);
29         int ntail = dst->dev->needed_tailroom - skb_tailroom(skb);
30
31         if (nhead <= 0) {
32                 if (ntail <= 0)
33                         return 0;
34                 nhead = 0;
35         } else if (ntail < 0)
36                 ntail = 0;
37
38         return pskb_expand_head(skb, nhead, ntail, GFP_ATOMIC);
39 }
40
41 static int xfrm_output_one(struct sk_buff *skb, int err)
42 {
43         struct dst_entry *dst = skb_dst(skb);
44         struct xfrm_state *x = dst->xfrm;
45         struct net *net = xs_net(x);
46
47         if (err <= 0)
48                 goto resume;
49
50         do {
51                 err = xfrm_state_check_space(x, skb);
52                 if (err) {
53                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
54                         goto error_nolock;
55                 }
56
57                 err = x->outer_mode->output(x, skb);
58                 if (err) {
59                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
60                         goto error_nolock;
61                 }
62
63                 spin_lock_bh(&x->lock);
64                 err = xfrm_state_check_expire(x);
65                 if (err) {
66                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEEXPIRED);
67                         goto error;
68                 }
69
70                 if (x->type->flags & XFRM_TYPE_REPLAY_PROT) {
71                         XFRM_SKB_CB(skb)->seq.output = ++x->replay.oseq;
72                         if (unlikely(x->replay.oseq == 0)) {
73                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATESEQERROR);
74                                 x->replay.oseq--;
75                                 xfrm_audit_state_replay_overflow(x, skb);
76                                 err = -EOVERFLOW;
77                                 goto error;
78                         }
79                         if (xfrm_aevent_is_on(net))
80                                 xfrm_replay_notify(x, XFRM_REPLAY_UPDATE);
81                 }
82
83                 x->curlft.bytes += skb->len;
84                 x->curlft.packets++;
85
86                 spin_unlock_bh(&x->lock);
87
88                 err = x->type->output(x, skb);
89                 if (err == -EINPROGRESS)
90                         goto out_exit;
91
92 resume:
93                 if (err) {
94                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
95                         goto error_nolock;
96                 }
97
98                 dst = skb_dst_pop(skb);
99                 if (!dst) {
100                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
101                         err = -EHOSTUNREACH;
102                         goto error_nolock;
103                 }
104                 skb_dst_set_noref(skb, dst);
105                 x = dst->xfrm;
106         } while (x && !(x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL));
107
108         err = 0;
109
110 out_exit:
111         return err;
112 error:
113         spin_unlock_bh(&x->lock);
114 error_nolock:
115         kfree_skb(skb);
116         goto out_exit;
117 }
118
119 int xfrm_output_resume(struct sk_buff *skb, int err)
120 {
121         while (likely((err = xfrm_output_one(skb, err)) == 0)) {
122                 nf_reset(skb);
123
124                 err = skb_dst(skb)->ops->local_out(skb);
125                 if (unlikely(err != 1))
126                         goto out;
127
128                 if (!skb_dst(skb)->xfrm)
129                         return dst_output(skb);
130
131                 err = nf_hook(skb_dst(skb)->ops->family,
132                               NF_INET_POST_ROUTING, skb,
133                               NULL, skb_dst(skb)->dev, xfrm_output2);
134                 if (unlikely(err != 1))
135                         goto out;
136         }
137
138         if (err == -EINPROGRESS)
139                 err = 0;
140
141 out:
142         return err;
143 }
144 EXPORT_SYMBOL_GPL(xfrm_output_resume);
145
146 static int xfrm_output2(struct sk_buff *skb)
147 {
148         return xfrm_output_resume(skb, 1);
149 }
150
151 static int xfrm_output_gso(struct sk_buff *skb)
152 {
153         struct sk_buff *segs;
154
155         segs = skb_gso_segment(skb, 0);
156         kfree_skb(skb);
157         if (IS_ERR(segs))
158                 return PTR_ERR(segs);
159
160         do {
161                 struct sk_buff *nskb = segs->next;
162                 int err;
163
164                 segs->next = NULL;
165                 err = xfrm_output2(segs);
166
167                 if (unlikely(err)) {
168                         while ((segs = nskb)) {
169                                 nskb = segs->next;
170                                 segs->next = NULL;
171                                 kfree_skb(segs);
172                         }
173                         return err;
174                 }
175
176                 segs = nskb;
177         } while (segs);
178
179         return 0;
180 }
181
182 int xfrm_output(struct sk_buff *skb)
183 {
184         struct net *net = dev_net(skb_dst(skb)->dev);
185         int err;
186
187         if (skb_is_gso(skb))
188                 return xfrm_output_gso(skb);
189
190         if (skb->ip_summed == CHECKSUM_PARTIAL) {
191                 err = skb_checksum_help(skb);
192                 if (err) {
193                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTERROR);
194                         kfree_skb(skb);
195                         return err;
196                 }
197         }
198
199         return xfrm_output2(skb);
200 }
201
202 int xfrm_inner_extract_output(struct xfrm_state *x, struct sk_buff *skb)
203 {
204         struct xfrm_mode *inner_mode;
205         if (x->sel.family == AF_UNSPEC)
206                 inner_mode = xfrm_ip2inner_mode(x,
207                                 xfrm_af2proto(skb_dst(skb)->ops->family));
208         else
209                 inner_mode = x->inner_mode;
210
211         if (inner_mode == NULL)
212                 return -EAFNOSUPPORT;
213         return inner_mode->afinfo->extract_output(x, skb);
214 }
215
216 EXPORT_SYMBOL_GPL(xfrm_output);
217 EXPORT_SYMBOL_GPL(xfrm_inner_extract_output);