target/linux/generic/pending-4.9/611-netfilter_match_bypass_default_table.patch

   1 From: Felix Fietkau <nbd@nbd.name>
   2 Subject: netfilter: match bypass default table
   3
   4 Signed-off-by: Felix Fietkau <nbd@nbd.name>
   5 ---
   6  net/ipv4/netfilter/ip_tables.c | 79 +++++++++++++++++++++++++++++++-----------
   7  1 file changed, 58 insertions(+), 21 deletions(-)
   8
   9 --- a/net/ipv4/netfilter/ip_tables.c
  10 +++ b/net/ipv4/netfilter/ip_tables.c
  11 @@ -254,6 +254,33 @@ struct ipt_entry *ipt_next_entry(const s
  12         return (void *)entry + entry->next_offset;
  13  }
  14
  15 +static bool
  16 +ipt_handle_default_rule(struct ipt_entry *e, unsigned int *verdict)
  17 +{
  18 +       struct xt_entry_target *t;
  19 +       struct xt_standard_target *st;
  20 +
  21 +       if (e->target_offset != sizeof(struct ipt_entry))
  22 +               return false;
  23 +
  24 +       if (!(e->ip.flags & IPT_F_NO_DEF_MATCH))
  25 +               return false;
  26 +
  27 +       t = ipt_get_target(e);
  28 +       if (t->u.kernel.target->target)
  29 +               return false;
  30 +
  31 +       st = (struct xt_standard_target *) t;
  32 +       if (st->verdict == XT_RETURN)
  33 +               return false;
  34 +
  35 +       if (st->verdict >= 0)
  36 +               return false;
  37 +
  38 +       *verdict = (unsigned)(-st->verdict) - 1;
  39 +       return true;
  40 +}
  41 +
  42  /* Returns one of the generic firewall policies, like NF_ACCEPT. */
  43  unsigned int
  44  ipt_do_table(struct sk_buff *skb,
  45 @@ -274,28 +301,8 @@ ipt_do_table(struct sk_buff *skb,
  46         unsigned int addend;
  47
  48         /* Initialization */
  49 -       stackidx = 0;
  50 -       ip = ip_hdr(skb);
  51 -       indev = state->in ? state->in->name : nulldevname;
  52 -       outdev = state->out ? state->out->name : nulldevname;
  53 -       /* We handle fragments by dealing with the first fragment as
  54 -        * if it was a normal packet.  All other fragments are treated
  55 -        * normally, except that they will NEVER match rules that ask
  56 -        * things we don't know, ie. tcp syn flag or ports).  If the
  57 -        * rule is also a fragment-specific rule, non-fragments won't
  58 -        * match it. */
  59 -       acpar.fragoff = ntohs(ip->frag_off) & IP_OFFSET;
  60 -       acpar.thoff   = ip_hdrlen(skb);
  61 -       acpar.hotdrop = false;
  62 -       acpar.net     = state->net;
  63 -       acpar.in      = state->in;
  64 -       acpar.out     = state->out;
  65 -       acpar.family  = NFPROTO_IPV4;
  66 -       acpar.hooknum = hook;
  67 -
  68         IP_NF_ASSERT(table->valid_hooks & (1 << hook));
  69         local_bh_disable();
  70 -       addend = xt_write_recseq_begin();
  71         private = table->private;
  72         cpu        = smp_processor_id();
  73         /*
  74 @@ -304,6 +311,23 @@ ipt_do_table(struct sk_buff *skb,
  75          */
  76         smp_read_barrier_depends();
  77         table_base = private->entries;
  78 +
  79 +       e = get_entry(table_base, private->hook_entry[hook]);
  80 +       if (ipt_handle_default_rule(e, &verdict)) {
  81 +               struct xt_counters *counter;
  82 +
  83 +               counter = xt_get_this_cpu_counter(&e->counters);
  84 +               ADD_COUNTER(*counter, skb->len, 1);
  85 +               local_bh_enable();
  86 +               return verdict;
  87 +       }
  88 +
  89 +       stackidx = 0;
  90 +       ip = ip_hdr(skb);
  91 +       indev = state->in ? state->in->name : nulldevname;
  92 +       outdev = state->out ? state->out->name : nulldevname;
  93 +
  94 +       addend = xt_write_recseq_begin();
  95         jumpstack  = (struct ipt_entry **)private->jumpstack[cpu];
  96
  97         /* Switch to alternate jumpstack if we're being invoked via TEE.
  98 @@ -316,7 +340,20 @@ ipt_do_table(struct sk_buff *skb,
  99         if (static_key_false(&xt_tee_enabled))
 100                 jumpstack += private->stacksize * __this_cpu_read(nf_skb_duplicated);
 101
 102 -       e = get_entry(table_base, private->hook_entry[hook]);
 103 +       /* We handle fragments by dealing with the first fragment as
 104 +        * if it was a normal packet.  All other fragments are treated
 105 +        * normally, except that they will NEVER match rules that ask
 106 +        * things we don't know, ie. tcp syn flag or ports).  If the
 107 +        * rule is also a fragment-specific rule, non-fragments won't
 108 +        * match it. */
 109 +       acpar.fragoff = ntohs(ip->frag_off) & IP_OFFSET;
 110 +       acpar.thoff   = ip_hdrlen(skb);
 111 +       acpar.hotdrop = false;
 112 +       acpar.net     = state->net;
 113 +       acpar.in      = state->in;
 114 +       acpar.out     = state->out;
 115 +       acpar.family  = NFPROTO_IPV4;
 116 +       acpar.hooknum = hook;
 117
 118         do {
 119                 const struct xt_entry_target *t;