git.openwrt.org Git - openwrt/staging/ynezz.git/blob

   1 From: Pablo Neira Ayuso <pablo@netfilter.org>
   2 Date: Sun, 7 Jan 2018 01:03:56 +0100
   3 Subject: [PATCH] netfilter: nf_conntrack: add IPS_OFFLOAD status bit
   4
   5 This new bit tells us that the conntrack entry is owned by the flow
   6 table offload infrastructure.
   7
   8  # cat /proc/net/nf_conntrack
   9  ipv4     2 tcp      6 src=10.141.10.2 dst=147.75.205.195 sport=36392 dport=443 src=147.75.205.195 dst=192.168.2.195 sport=443 dport=36392 [OFFLOAD] mark=0 zone=0 use=2
  10
  11 Note the [OFFLOAD] tag in the listing.
  12
  13 The timer of such conntrack entries look like stopped from userspace.
  14 In practise, to make sure the conntrack entry does not go away, the
  15 conntrack timer is periodically set to an arbitrary large value that
  16 gets refreshed on every iteration from the garbage collector, so it
  17 never expires- and they display no internal state in the case of TCP
  18 flows. This allows us to save a bitcheck from the packet path via
  19 nf_ct_is_expired().
  20
  21 Conntrack entries that have been offloaded to the flow table
  22 infrastructure cannot be deleted/flushed via ctnetlink. The flow table
  23 infrastructure is also responsible for releasing this conntrack entry.
  24
  25 Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
  26 ---
  27
  28 --- a/include/uapi/linux/netfilter/nf_conntrack_common.h
  29 +++ b/include/uapi/linux/netfilter/nf_conntrack_common.h
  30 @@ -101,12 +101,16 @@ enum ip_conntrack_status {
  31         IPS_HELPER_BIT = 13,
  32         IPS_HELPER = (1 << IPS_HELPER_BIT),
  33
  34 +       /* Conntrack has been offloaded to flow table. */
  35 +       IPS_OFFLOAD_BIT = 14,
  36 +       IPS_OFFLOAD = (1 << IPS_OFFLOAD_BIT),
  37 +
  38         /* Be careful here, modifying these bits can make things messy,
  39          * so don't let users modify them directly.
  40          */
  41         IPS_UNCHANGEABLE_MASK = (IPS_NAT_DONE_MASK | IPS_NAT_MASK |
  42                                  IPS_EXPECTED | IPS_CONFIRMED | IPS_DYING |
  43 -                                IPS_SEQ_ADJUST | IPS_TEMPLATE),
  44 +                                IPS_SEQ_ADJUST | IPS_TEMPLATE | IPS_OFFLOAD),
  45
  46         __IPS_MAX_BIT = 14,
  47  };
  48 --- a/net/netfilter/nf_conntrack_core.c
  49 +++ b/net/netfilter/nf_conntrack_core.c
  50 @@ -960,6 +960,9 @@ static unsigned int early_drop_list(stru
  51         hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
  52                 tmp = nf_ct_tuplehash_to_ctrack(h);
  53
  54 +               if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
  55 +                       continue;
  56 +
  57                 if (nf_ct_is_expired(tmp)) {
  58                         nf_ct_gc_expired(tmp);
  59                         continue;
  60 @@ -1037,6 +1040,18 @@ static bool gc_worker_can_early_drop(con
  61         return false;
  62  }
  63
  64 +#define        DAY     (86400 * HZ)
  65 +
  66 +/* Set an arbitrary timeout large enough not to ever expire, this save
  67 + * us a check for the IPS_OFFLOAD_BIT from the packet path via
  68 + * nf_ct_is_expired().
  69 + */
  70 +static void nf_ct_offload_timeout(struct nf_conn *ct)
  71 +{
  72 +       if (nf_ct_expires(ct) < DAY / 2)
  73 +               ct->timeout = nfct_time_stamp + DAY;
  74 +}
  75 +
  76  static void gc_worker(struct work_struct *work)
  77  {
  78         unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
  79 @@ -1073,6 +1088,11 @@ static void gc_worker(struct work_struct
  80                         tmp = nf_ct_tuplehash_to_ctrack(h);
  81
  82                         scanned++;
  83 +                       if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
  84 +                               nf_ct_offload_timeout(tmp);
  85 +                               continue;
  86 +                       }
  87 +
  88                         if (nf_ct_is_expired(tmp)) {
  89                                 nf_ct_gc_expired(tmp);
  90                                 expired_count++;
  91 --- a/net/netfilter/nf_conntrack_netlink.c
  92 +++ b/net/netfilter/nf_conntrack_netlink.c
  93 @@ -1123,6 +1123,14 @@ static const struct nla_policy ct_nla_po
  94                                     .len = NF_CT_LABELS_MAX_SIZE },
  95  };
  96
  97 +static int ctnetlink_flush_iterate(struct nf_conn *ct, void *data)
  98 +{
  99 +       if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
 100 +               return 0;
 101 +
 102 +       return ctnetlink_filter_match(ct, data);
 103 +}
 104 +
 105  static int ctnetlink_flush_conntrack(struct net *net,
 106                                      const struct nlattr * const cda[],
 107                                      u32 portid, int report)
 108 @@ -1135,7 +1143,7 @@ static int ctnetlink_flush_conntrack(str
 109                         return PTR_ERR(filter);
 110         }
 111
 112 -       nf_ct_iterate_cleanup_net(net, ctnetlink_filter_match, filter,
 113 +       nf_ct_iterate_cleanup_net(net, ctnetlink_flush_iterate, filter,
 114                                   portid, report);
 115         kfree(filter);
 116
 117 @@ -1181,6 +1189,11 @@ static int ctnetlink_del_conntrack(struc
 118
 119         ct = nf_ct_tuplehash_to_ctrack(h);
 120
 121 +       if (test_bit(IPS_OFFLOAD_BIT, &ct->status)) {
 122 +               nf_ct_put(ct);
 123 +               return -EBUSY;
 124 +       }
 125 +
 126         if (cda[CTA_ID]) {
 127                 __be32 id = nla_get_be32(cda[CTA_ID]);
 128
 129 --- a/net/netfilter/nf_conntrack_proto_tcp.c
 130 +++ b/net/netfilter/nf_conntrack_proto_tcp.c
 131 @@ -305,6 +305,9 @@ static bool tcp_invert_tuple(struct nf_c
 132  /* Print out the private part of the conntrack. */
 133  static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 134  {
 135 +       if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
 136 +               return;
 137 +
 138         seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
 139  }
 140  #endif
 141 --- a/net/netfilter/nf_conntrack_standalone.c
 142 +++ b/net/netfilter/nf_conntrack_standalone.c
 143 @@ -309,10 +309,12 @@ static int ct_seq_show(struct seq_file *
 144         WARN_ON(!l4proto);
 145
 146         ret = -ENOSPC;
 147 -       seq_printf(s, "%-8s %u %-8s %u %ld ",
 148 +       seq_printf(s, "%-8s %u %-8s %u ",
 149                    l3proto_name(l3proto->l3proto), nf_ct_l3num(ct),
 150 -                  l4proto_name(l4proto->l4proto), nf_ct_protonum(ct),
 151 -                  nf_ct_expires(ct)  / HZ);
 152 +                  l4proto_name(l4proto->l4proto), nf_ct_protonum(ct));
 153 +
 154 +       if (!test_bit(IPS_OFFLOAD_BIT, &ct->status))
 155 +               seq_printf(s, "%ld ", nf_ct_expires(ct)  / HZ);
 156
 157         if (l4proto->print_conntrack)
 158                 l4proto->print_conntrack(s, ct);
 159 @@ -339,7 +341,9 @@ static int ct_seq_show(struct seq_file *
 160         if (seq_print_acct(s, ct, IP_CT_DIR_REPLY))
 161                 goto release;
 162
 163 -       if (test_bit(IPS_ASSURED_BIT, &ct->status))
 164 +       if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
 165 +               seq_puts(s, "[OFFLOAD] ");
 166 +       else if (test_bit(IPS_ASSURED_BIT, &ct->status))
 167                 seq_puts(s, "[ASSURED] ");
 168
 169         if (seq_has_overflowed(s))