net/ipv4/netfilter.c

   1 /*
   2  * IPv4 specific functions of netfilter core
   3  *
   4  * Rusty Russell (C) 2000 -- This code is GPL.
   5  * Patrick McHardy (C) 2006-2012
   6  */
   7 #include <linux/kernel.h>
   8 #include <linux/netfilter.h>
   9 #include <linux/netfilter_ipv4.h>
  10 #include <linux/ip.h>
  11 #include <linux/skbuff.h>
  12 #include <linux/gfp.h>
  13 #include <linux/export.h>
  14 #include <net/route.h>
  15 #include <net/xfrm.h>
  16 #include <net/ip.h>
  17 #include <net/netfilter/nf_queue.h>
  18
  19 /* route_me_harder function, used by iptable_nat, iptable_mangle + ip_queue */
  20 int ip_route_me_harder(struct net *net, struct sk_buff *skb, unsigned int addr_type)
  21 {
  22         const struct iphdr *iph = ip_hdr(skb);
  23         struct rtable *rt;
  24         struct flowi4 fl4 = {};
  25         __be32 saddr = iph->saddr;
  26         const struct sock *sk = skb_to_full_sk(skb);
  27         __u8 flags = sk ? inet_sk_flowi_flags(sk) : 0;
  28         struct net_device *dev = skb_dst(skb)->dev;
  29         unsigned int hh_len;
  30
  31         if (addr_type == RTN_UNSPEC)
  32                 addr_type = inet_addr_type_dev_table(net, dev, saddr);
  33         if (addr_type == RTN_LOCAL || addr_type == RTN_UNICAST)
  34                 flags |= FLOWI_FLAG_ANYSRC;
  35         else
  36                 saddr = 0;
  37
  38         /* some non-standard hacks like ipt_REJECT.c:send_reset() can cause
  39          * packets with foreign saddr to appear on the NF_INET_LOCAL_OUT hook.
  40          */
  41         fl4.daddr = iph->daddr;
  42         fl4.saddr = saddr;
  43         fl4.flowi4_tos = RT_TOS(iph->tos);
  44         fl4.flowi4_oif = sk ? sk->sk_bound_dev_if : 0;
  45         if (!fl4.flowi4_oif)
  46                 fl4.flowi4_oif = l3mdev_master_ifindex(dev);
  47         fl4.flowi4_mark = skb->mark;
  48         fl4.flowi4_flags = flags;
  49         rt = ip_route_output_key(net, &fl4);
  50         if (IS_ERR(rt))
  51                 return PTR_ERR(rt);
  52
  53         /* Drop old route. */
  54         skb_dst_drop(skb);
  55         skb_dst_set(skb, &rt->dst);
  56
  57         if (skb_dst(skb)->error)
  58                 return skb_dst(skb)->error;
  59
  60 #ifdef CONFIG_XFRM
  61         if (!(IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) &&
  62             xfrm_decode_session(skb, flowi4_to_flowi(&fl4), AF_INET) == 0) {
  63                 struct dst_entry *dst = skb_dst(skb);
  64                 skb_dst_set(skb, NULL);
  65                 dst = xfrm_lookup(net, dst, flowi4_to_flowi(&fl4), sk, 0);
  66                 if (IS_ERR(dst))
  67                         return PTR_ERR(dst);
  68                 skb_dst_set(skb, dst);
  69         }
  70 #endif
  71
  72         /* Change in oif may mean change in hh_len. */
  73         hh_len = skb_dst(skb)->dev->hard_header_len;
  74         if (skb_headroom(skb) < hh_len &&
  75             pskb_expand_head(skb, HH_DATA_ALIGN(hh_len - skb_headroom(skb)),
  76                                 0, GFP_ATOMIC))
  77                 return -ENOMEM;
  78
  79         return 0;
  80 }
  81 EXPORT_SYMBOL(ip_route_me_harder);
  82
  83 /*
  84  * Extra routing may needed on local out, as the QUEUE target never
  85  * returns control to the table.
  86  */
  87
  88 struct ip_rt_info {
  89         __be32 daddr;
  90         __be32 saddr;
  91         u_int8_t tos;
  92         u_int32_t mark;
  93 };
  94
  95 static void nf_ip_saveroute(const struct sk_buff *skb,
  96                             struct nf_queue_entry *entry)
  97 {
  98         struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
  99
 100         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 101                 const struct iphdr *iph = ip_hdr(skb);
 102
 103                 rt_info->tos = iph->tos;
 104                 rt_info->daddr = iph->daddr;
 105                 rt_info->saddr = iph->saddr;
 106                 rt_info->mark = skb->mark;
 107         }
 108 }
 109
 110 static int nf_ip_reroute(struct net *net, struct sk_buff *skb,
 111                          const struct nf_queue_entry *entry)
 112 {
 113         const struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
 114
 115         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 116                 const struct iphdr *iph = ip_hdr(skb);
 117
 118                 if (!(iph->tos == rt_info->tos &&
 119                       skb->mark == rt_info->mark &&
 120                       iph->daddr == rt_info->daddr &&
 121                       iph->saddr == rt_info->saddr))
 122                         return ip_route_me_harder(net, skb, RTN_UNSPEC);
 123         }
 124         return 0;
 125 }
 126
 127 __sum16 nf_ip_checksum(struct sk_buff *skb, unsigned int hook,
 128                             unsigned int dataoff, u_int8_t protocol)
 129 {
 130         const struct iphdr *iph = ip_hdr(skb);
 131         __sum16 csum = 0;
 132
 133         switch (skb->ip_summed) {
 134         case CHECKSUM_COMPLETE:
 135                 if (hook != NF_INET_PRE_ROUTING && hook != NF_INET_LOCAL_IN)
 136                         break;
 137                 if ((protocol == 0 && !csum_fold(skb->csum)) ||
 138                     !csum_tcpudp_magic(iph->saddr, iph->daddr,
 139                                        skb->len - dataoff, protocol,
 140                                        skb->csum)) {
 141                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 142                         break;
 143                 }
 144                 /* fall through */
 145         case CHECKSUM_NONE:
 146                 if (protocol == 0)
 147                         skb->csum = 0;
 148                 else
 149                         skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
 150                                                        skb->len - dataoff,
 151                                                        protocol, 0);
 152                 csum = __skb_checksum_complete(skb);
 153         }
 154         return csum;
 155 }
 156 EXPORT_SYMBOL(nf_ip_checksum);
 157
 158 static __sum16 nf_ip_checksum_partial(struct sk_buff *skb, unsigned int hook,
 159                                       unsigned int dataoff, unsigned int len,
 160                                       u_int8_t protocol)
 161 {
 162         const struct iphdr *iph = ip_hdr(skb);
 163         __sum16 csum = 0;
 164
 165         switch (skb->ip_summed) {
 166         case CHECKSUM_COMPLETE:
 167                 if (len == skb->len - dataoff)
 168                         return nf_ip_checksum(skb, hook, dataoff, protocol);
 169                 /* fall through */
 170         case CHECKSUM_NONE:
 171                 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, protocol,
 172                                                skb->len - dataoff, 0);
 173                 skb->ip_summed = CHECKSUM_NONE;
 174                 return __skb_checksum_complete_head(skb, dataoff + len);
 175         }
 176         return csum;
 177 }
 178
 179 static int nf_ip_route(struct net *net, struct dst_entry **dst,
 180                        struct flowi *fl, bool strict __always_unused)
 181 {
 182         struct rtable *rt = ip_route_output_key(net, &fl->u.ip4);
 183         if (IS_ERR(rt))
 184                 return PTR_ERR(rt);
 185         *dst = &rt->dst;
 186         return 0;
 187 }
 188
 189 static const struct nf_afinfo nf_ip_afinfo = {
 190         .family                 = AF_INET,
 191         .checksum               = nf_ip_checksum,
 192         .checksum_partial       = nf_ip_checksum_partial,
 193         .route                  = nf_ip_route,
 194         .saveroute              = nf_ip_saveroute,
 195         .reroute                = nf_ip_reroute,
 196         .route_key_size         = sizeof(struct ip_rt_info),
 197 };
 198
 199 static int __init ipv4_netfilter_init(void)
 200 {
 201         return nf_register_afinfo(&nf_ip_afinfo);
 202 }
 203 subsys_initcall(ipv4_netfilter_init);