net/netfilter/nf_conntrack_seqadj.c

   1 #include <linux/types.h>
   2 #include <linux/netfilter.h>
   3 #include <net/tcp.h>
   4
   5 #include <net/netfilter/nf_conntrack.h>
   6 #include <net/netfilter/nf_conntrack_extend.h>
   7 #include <net/netfilter/nf_conntrack_seqadj.h>
   8
   9 int nf_ct_seqadj_init(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
  10                       s32 off)
  11 {
  12         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
  13         struct nf_conn_seqadj *seqadj;
  14         struct nf_ct_seqadj *this_way;
  15
  16         if (off == 0)
  17                 return 0;
  18
  19         set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
  20
  21         seqadj = nfct_seqadj(ct);
  22         this_way = &seqadj->seq[dir];
  23         this_way->offset_before  = off;
  24         this_way->offset_after   = off;
  25         return 0;
  26 }
  27 EXPORT_SYMBOL_GPL(nf_ct_seqadj_init);
  28
  29 int nf_ct_seqadj_set(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
  30                      __be32 seq, s32 off)
  31 {
  32         struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
  33         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
  34         struct nf_ct_seqadj *this_way;
  35
  36         if (off == 0)
  37                 return 0;
  38
  39         set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
  40
  41         spin_lock_bh(&ct->lock);
  42         this_way = &seqadj->seq[dir];
  43         if (this_way->offset_before == this_way->offset_after ||
  44             before(this_way->correction_pos, seq)) {
  45                 this_way->correction_pos = seq;
  46                 this_way->offset_before  = this_way->offset_after;
  47                 this_way->offset_after  += off;
  48         }
  49         spin_unlock_bh(&ct->lock);
  50         return 0;
  51 }
  52 EXPORT_SYMBOL_GPL(nf_ct_seqadj_set);
  53
  54 void nf_ct_tcp_seqadj_set(struct sk_buff *skb,
  55                           struct nf_conn *ct, enum ip_conntrack_info ctinfo,
  56                           s32 off)
  57 {
  58         const struct tcphdr *th;
  59
  60         if (nf_ct_protonum(ct) != IPPROTO_TCP)
  61                 return;
  62
  63         th = (struct tcphdr *)(skb_network_header(skb) + ip_hdrlen(skb));
  64         nf_ct_seqadj_set(ct, ctinfo, th->seq, off);
  65 }
  66 EXPORT_SYMBOL_GPL(nf_ct_tcp_seqadj_set);
  67
  68 /* Adjust one found SACK option including checksum correction */
  69 static void nf_ct_sack_block_adjust(struct sk_buff *skb,
  70                                     struct tcphdr *tcph,
  71                                     unsigned int sackoff,
  72                                     unsigned int sackend,
  73                                     struct nf_ct_seqadj *seq)
  74 {
  75         while (sackoff < sackend) {
  76                 struct tcp_sack_block_wire *sack;
  77                 __be32 new_start_seq, new_end_seq;
  78
  79                 sack = (void *)skb->data + sackoff;
  80                 if (after(ntohl(sack->start_seq) - seq->offset_before,
  81                           seq->correction_pos))
  82                         new_start_seq = htonl(ntohl(sack->start_seq) -
  83                                         seq->offset_after);
  84                 else
  85                         new_start_seq = htonl(ntohl(sack->start_seq) -
  86                                         seq->offset_before);
  87
  88                 if (after(ntohl(sack->end_seq) - seq->offset_before,
  89                           seq->correction_pos))
  90                         new_end_seq = htonl(ntohl(sack->end_seq) -
  91                                       seq->offset_after);
  92                 else
  93                         new_end_seq = htonl(ntohl(sack->end_seq) -
  94                                       seq->offset_before);
  95
  96                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
  97                          ntohl(sack->start_seq), new_start_seq,
  98                          ntohl(sack->end_seq), new_end_seq);
  99
 100                 inet_proto_csum_replace4(&tcph->check, skb,
 101                                          sack->start_seq, new_start_seq, 0);
 102                 inet_proto_csum_replace4(&tcph->check, skb,
 103                                          sack->end_seq, new_end_seq, 0);
 104                 sack->start_seq = new_start_seq;
 105                 sack->end_seq = new_end_seq;
 106                 sackoff += sizeof(*sack);
 107         }
 108 }
 109
 110 /* TCP SACK sequence number adjustment */
 111 static unsigned int nf_ct_sack_adjust(struct sk_buff *skb,
 112                                       unsigned int protoff,
 113                                       struct tcphdr *tcph,
 114                                       struct nf_conn *ct,
 115                                       enum ip_conntrack_info ctinfo)
 116 {
 117         unsigned int dir, optoff, optend;
 118         struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
 119
 120         optoff = protoff + sizeof(struct tcphdr);
 121         optend = protoff + tcph->doff * 4;
 122
 123         if (!skb_make_writable(skb, optend))
 124                 return 0;
 125
 126         dir = CTINFO2DIR(ctinfo);
 127
 128         while (optoff < optend) {
 129                 /* Usually: option, length. */
 130                 unsigned char *op = skb->data + optoff;
 131
 132                 switch (op[0]) {
 133                 case TCPOPT_EOL:
 134                         return 1;
 135                 case TCPOPT_NOP:
 136                         optoff++;
 137                         continue;
 138                 default:
 139                         /* no partial options */
 140                         if (optoff + 1 == optend ||
 141                             optoff + op[1] > optend ||
 142                             op[1] < 2)
 143                                 return 0;
 144                         if (op[0] == TCPOPT_SACK &&
 145                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 146                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 147                                 nf_ct_sack_block_adjust(skb, tcph, optoff + 2,
 148                                                         optoff+op[1],
 149                                                         &seqadj->seq[!dir]);
 150                         optoff += op[1];
 151                 }
 152         }
 153         return 1;
 154 }
 155
 156 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 157 int nf_ct_seq_adjust(struct sk_buff *skb,
 158                      struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 159                      unsigned int protoff)
 160 {
 161         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 162         struct tcphdr *tcph;
 163         __be32 newseq, newack;
 164         s32 seqoff, ackoff;
 165         struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
 166         struct nf_ct_seqadj *this_way, *other_way;
 167         int res;
 168
 169         this_way  = &seqadj->seq[dir];
 170         other_way = &seqadj->seq[!dir];
 171
 172         if (!skb_make_writable(skb, protoff + sizeof(*tcph)))
 173                 return 0;
 174
 175         tcph = (void *)skb->data + protoff;
 176         spin_lock_bh(&ct->lock);
 177         if (after(ntohl(tcph->seq), this_way->correction_pos))
 178                 seqoff = this_way->offset_after;
 179         else
 180                 seqoff = this_way->offset_before;
 181
 182         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 183                   other_way->correction_pos))
 184                 ackoff = other_way->offset_after;
 185         else
 186                 ackoff = other_way->offset_before;
 187
 188         newseq = htonl(ntohl(tcph->seq) + seqoff);
 189         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
 190
 191         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 192         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 193
 194         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 195                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 196                  ntohl(newack));
 197
 198         tcph->seq = newseq;
 199         tcph->ack_seq = newack;
 200
 201         res = nf_ct_sack_adjust(skb, protoff, tcph, ct, ctinfo);
 202         spin_unlock_bh(&ct->lock);
 203
 204         return res;
 205 }
 206 EXPORT_SYMBOL_GPL(nf_ct_seq_adjust);
 207
 208 s32 nf_ct_seq_offset(const struct nf_conn *ct,
 209                      enum ip_conntrack_dir dir,
 210                      u32 seq)
 211 {
 212         struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
 213         struct nf_ct_seqadj *this_way;
 214
 215         if (!seqadj)
 216                 return 0;
 217
 218         this_way = &seqadj->seq[dir];
 219         return after(seq, this_way->correction_pos) ?
 220                  this_way->offset_after : this_way->offset_before;
 221 }
 222 EXPORT_SYMBOL_GPL(nf_ct_seq_offset);
 223
 224 static struct nf_ct_ext_type nf_ct_seqadj_extend __read_mostly = {
 225         .len    = sizeof(struct nf_conn_seqadj),
 226         .align  = __alignof__(struct nf_conn_seqadj),
 227         .id     = NF_CT_EXT_SEQADJ,
 228 };
 229
 230 int nf_conntrack_seqadj_init(void)
 231 {
 232         return nf_ct_extend_register(&nf_ct_seqadj_extend);
 233 }
 234
 235 void nf_conntrack_seqadj_fini(void)
 236 {
 237         nf_ct_extend_unregister(&nf_ct_seqadj_extend);
 238 }