target/linux/generic/pending-4.14/611-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 @@ -248,6 +248,33 @@ 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 @@ -268,24 +295,8 @@ 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.state   = state;
  63 -
  64         WARN_ON(!(table->valid_hooks & (1 << hook)));
  65         local_bh_disable();
  66 -       addend = xt_write_recseq_begin();
  67         private = table->private;
  68         cpu        = smp_processor_id();
  69         /*
  70 @@ -294,6 +305,23 @@ ipt_do_table(struct sk_buff *skb,
  71          */
  72         smp_read_barrier_depends();
  73         table_base = private->entries;
  74 +
  75 +       e = get_entry(table_base, private->hook_entry[hook]);
  76 +       if (ipt_handle_default_rule(e, &verdict)) {
  77 +               struct xt_counters *counter;
  78 +
  79 +               counter = xt_get_this_cpu_counter(&e->counters);
  80 +               ADD_COUNTER(*counter, skb->len, 1);
  81 +               local_bh_enable();
  82 +               return verdict;
  83 +       }
  84 +
  85 +       stackidx = 0;
  86 +       ip = ip_hdr(skb);
  87 +       indev = state->in ? state->in->name : nulldevname;
  88 +       outdev = state->out ? state->out->name : nulldevname;
  89 +
  90 +       addend = xt_write_recseq_begin();
  91         jumpstack  = (struct ipt_entry **)private->jumpstack[cpu];
  92
  93         /* Switch to alternate jumpstack if we're being invoked via TEE.
  94 @@ -306,7 +334,16 @@ ipt_do_table(struct sk_buff *skb,
  95         if (static_key_false(&xt_tee_enabled))
  96                 jumpstack += private->stacksize * __this_cpu_read(nf_skb_duplicated);
  97
  98 -       e = get_entry(table_base, private->hook_entry[hook]);
  99 +       /* We handle fragments by dealing with the first fragment as
 100 +        * if it was a normal packet.  All other fragments are treated
 101 +        * normally, except that they will NEVER match rules that ask
 102 +        * things we don't know, ie. tcp syn flag or ports).  If the
 103 +        * rule is also a fragment-specific rule, non-fragments won't
 104 +        * match it. */
 105 +       acpar.fragoff = ntohs(ip->frag_off) & IP_OFFSET;
 106 +       acpar.thoff   = ip_hdrlen(skb);
 107 +       acpar.hotdrop = false;
 108 +       acpar.state   = state;
 109
 110         do {
 111                 const struct xt_entry_target *t;