net/ipv4/netfilter.c

   1 /* IPv4 specific functions of netfilter core */
   2 #include <linux/kernel.h>
   3 #include <linux/netfilter.h>
   4 #include <linux/netfilter_ipv4.h>
   5 #include <linux/ip.h>
   6 #include <linux/skbuff.h>
   7 #include <linux/gfp.h>
   8 #include <linux/export.h>
   9 #include <net/route.h>
  10 #include <net/xfrm.h>
  11 #include <net/ip.h>
  12 #include <net/netfilter/nf_queue.h>
  13
  14 /* route_me_harder function, used by iptable_nat, iptable_mangle + ip_queue */
  15 int ip_route_me_harder(struct sk_buff *skb, unsigned int addr_type)
  16 {
  17         struct net *net = dev_net(skb_dst(skb)->dev);
  18         const struct iphdr *iph = ip_hdr(skb);
  19         struct rtable *rt;
  20         struct flowi4 fl4 = {};
  21         __be32 saddr = iph->saddr;
  22         __u8 flags = skb->sk ? inet_sk_flowi_flags(skb->sk) : 0;
  23         unsigned int hh_len;
  24
  25         if (addr_type == RTN_UNSPEC)
  26                 addr_type = inet_addr_type(net, saddr);
  27         if (addr_type == RTN_LOCAL || addr_type == RTN_UNICAST)
  28                 flags |= FLOWI_FLAG_ANYSRC;
  29         else
  30                 saddr = 0;
  31
  32         /* some non-standard hacks like ipt_REJECT.c:send_reset() can cause
  33          * packets with foreign saddr to appear on the NF_INET_LOCAL_OUT hook.
  34          */
  35         fl4.daddr = iph->daddr;
  36         fl4.saddr = saddr;
  37         fl4.flowi4_tos = RT_TOS(iph->tos);
  38         fl4.flowi4_oif = skb->sk ? skb->sk->sk_bound_dev_if : 0;
  39         fl4.flowi4_mark = skb->mark;
  40         fl4.flowi4_flags = flags;
  41         rt = ip_route_output_key(net, &fl4);
  42         if (IS_ERR(rt))
  43                 return -1;
  44
  45         /* Drop old route. */
  46         skb_dst_drop(skb);
  47         skb_dst_set(skb, &rt->dst);
  48
  49         if (skb_dst(skb)->error)
  50                 return -1;
  51
  52 #ifdef CONFIG_XFRM
  53         if (!(IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) &&
  54             xfrm_decode_session(skb, flowi4_to_flowi(&fl4), AF_INET) == 0) {
  55                 struct dst_entry *dst = skb_dst(skb);
  56                 skb_dst_set(skb, NULL);
  57                 dst = xfrm_lookup(net, dst, flowi4_to_flowi(&fl4), skb->sk, 0);
  58                 if (IS_ERR(dst))
  59                         return -1;
  60                 skb_dst_set(skb, dst);
  61         }
  62 #endif
  63
  64         /* Change in oif may mean change in hh_len. */
  65         hh_len = skb_dst(skb)->dev->hard_header_len;
  66         if (skb_headroom(skb) < hh_len &&
  67             pskb_expand_head(skb, HH_DATA_ALIGN(hh_len - skb_headroom(skb)),
  68                                 0, GFP_ATOMIC))
  69                 return -1;
  70
  71         return 0;
  72 }
  73 EXPORT_SYMBOL(ip_route_me_harder);
  74
  75 /*
  76  * Extra routing may needed on local out, as the QUEUE target never
  77  * returns control to the table.
  78  */
  79
  80 struct ip_rt_info {
  81         __be32 daddr;
  82         __be32 saddr;
  83         u_int8_t tos;
  84         u_int32_t mark;
  85 };
  86
  87 static void nf_ip_saveroute(const struct sk_buff *skb,
  88                             struct nf_queue_entry *entry)
  89 {
  90         struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
  91
  92         if (entry->hook == NF_INET_LOCAL_OUT) {
  93                 const struct iphdr *iph = ip_hdr(skb);
  94
  95                 rt_info->tos = iph->tos;
  96                 rt_info->daddr = iph->daddr;
  97                 rt_info->saddr = iph->saddr;
  98                 rt_info->mark = skb->mark;
  99         }
 100 }
 101
 102 static int nf_ip_reroute(struct sk_buff *skb,
 103                          const struct nf_queue_entry *entry)
 104 {
 105         const struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
 106
 107         if (entry->hook == NF_INET_LOCAL_OUT) {
 108                 const struct iphdr *iph = ip_hdr(skb);
 109
 110                 if (!(iph->tos == rt_info->tos &&
 111                       skb->mark == rt_info->mark &&
 112                       iph->daddr == rt_info->daddr &&
 113                       iph->saddr == rt_info->saddr))
 114                         return ip_route_me_harder(skb, RTN_UNSPEC);
 115         }
 116         return 0;
 117 }
 118
 119 __sum16 nf_ip_checksum(struct sk_buff *skb, unsigned int hook,
 120                             unsigned int dataoff, u_int8_t protocol)
 121 {
 122         const struct iphdr *iph = ip_hdr(skb);
 123         __sum16 csum = 0;
 124
 125         switch (skb->ip_summed) {
 126         case CHECKSUM_COMPLETE:
 127                 if (hook != NF_INET_PRE_ROUTING && hook != NF_INET_LOCAL_IN)
 128                         break;
 129                 if ((protocol == 0 && !csum_fold(skb->csum)) ||
 130                     !csum_tcpudp_magic(iph->saddr, iph->daddr,
 131                                        skb->len - dataoff, protocol,
 132                                        skb->csum)) {
 133                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 134                         break;
 135                 }
 136                 /* fall through */
 137         case CHECKSUM_NONE:
 138                 if (protocol == 0)
 139                         skb->csum = 0;
 140                 else
 141                         skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
 142                                                        skb->len - dataoff,
 143                                                        protocol, 0);
 144                 csum = __skb_checksum_complete(skb);
 145         }
 146         return csum;
 147 }
 148 EXPORT_SYMBOL(nf_ip_checksum);
 149
 150 static __sum16 nf_ip_checksum_partial(struct sk_buff *skb, unsigned int hook,
 151                                       unsigned int dataoff, unsigned int len,
 152                                       u_int8_t protocol)
 153 {
 154         const struct iphdr *iph = ip_hdr(skb);
 155         __sum16 csum = 0;
 156
 157         switch (skb->ip_summed) {
 158         case CHECKSUM_COMPLETE:
 159                 if (len == skb->len - dataoff)
 160                         return nf_ip_checksum(skb, hook, dataoff, protocol);
 161                 /* fall through */
 162         case CHECKSUM_NONE:
 163                 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, protocol,
 164                                                skb->len - dataoff, 0);
 165                 skb->ip_summed = CHECKSUM_NONE;
 166                 return __skb_checksum_complete_head(skb, dataoff + len);
 167         }
 168         return csum;
 169 }
 170
 171 static int nf_ip_route(struct net *net, struct dst_entry **dst,
 172                        struct flowi *fl, bool strict __always_unused)
 173 {
 174         struct rtable *rt = ip_route_output_key(net, &fl->u.ip4);
 175         if (IS_ERR(rt))
 176                 return PTR_ERR(rt);
 177         *dst = &rt->dst;
 178         return 0;
 179 }
 180
 181 static const struct nf_afinfo nf_ip_afinfo = {
 182         .family                 = AF_INET,
 183         .checksum               = nf_ip_checksum,
 184         .checksum_partial       = nf_ip_checksum_partial,
 185         .route                  = nf_ip_route,
 186         .saveroute              = nf_ip_saveroute,
 187         .reroute                = nf_ip_reroute,
 188         .route_key_size         = sizeof(struct ip_rt_info),
 189 };
 190
 191 static int __init ipv4_netfilter_init(void)
 192 {
 193         return nf_register_afinfo(&nf_ip_afinfo);
 194 }
 195
 196 static void __exit ipv4_netfilter_fini(void)
 197 {
 198         nf_unregister_afinfo(&nf_ip_afinfo);
 199 }
 200
 201 module_init(ipv4_netfilter_init);
 202 module_exit(ipv4_netfilter_fini);