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