target/linux/generic/hack-4.19/645-netfilter-connmark-introduce-savedscp.patch

   1 From f171924dcf1d0b31fb7bd1cff113d7a1f7f05ec2 Mon Sep 17 00:00:00 2001
   2 From: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
   3 Date: Sat, 23 Mar 2019 09:29:49 +0000
   4 Subject: [PATCH] netfilter: connmark: introduce savedscp
   5
   6 savedscp is a method of storing the DSCP of an ip packet into conntrack
   7 mark.  In combination with a suitable tc filter action (act_ctinfo) DSCP
   8 values are able to be stored in the mark on egress and restored on
   9 ingress across links that otherwise alter or bleach DSCP.
  10
  11 This is useful for qdiscs such as CAKE which are able to shape according
  12 to policies based on DSCP.
  13
  14 Ingress classification is traditionally a challenging task since
  15 iptables rules haven't yet run and tc filter/eBPF programs are pre-NAT
  16 lookups, hence are unable to see internal IPv4 addresses as used on the
  17 typical home masquerading gateway.
  18
  19 x_tables CONNMARK savedscp action solves the problem of storing the DSCP
  20 to the conntrack mark in a way suitable for the new act_ctinfo tc action
  21 to restore.
  22
  23 The savedsp option accepts 2 parameters, a 32bit 'dscpmask' and a 32bit
  24 'statemask'.  The dscp mask must be a minimum of 6 contiguous bits and
  25 represents the area where the DSCP will be stored in the connmark.  The
  26 state mask is a minimum 1 bit length mask that must not overlap with the
  27 dscpmask.  It represents a flag which is set when the DSCP has been
  28 stored in the conntrack mark. This is useful to implement a 'one shot'
  29 iptables based classification where the 'complicated' iptables rules are
  30 only run once to classify the connection on initial (egress) packet and
  31 subsequent packets are all marked/restored with the same DSCP.  A state
  32 mask of zero disables the setting of a status bit/s.
  33
  34 example syntax with a suitably modified iptables user space application:
  35
  36 iptables -A QOS_MARK_eth0 -t mangle -j CONNMARK --savedscp-mark 0xfc000000/0x01000000
  37
  38 Would store the DSCP in the top 6 bits of the 32bit mark field, and use
  39 the LSB of the top byte as the 'DSCP has been stored' marker.
  40
  41 |----0xFC----conntrack mark----000000---|
  42 | Bits 31-26 | bit 25 | bit24 |~~~ Bit 0|
  43 | DSCP       | unused | flag  |unused   |
  44 |-----------------------0x01---000000---|
  45       ^                   ^
  46       |                   |
  47       ---|             Conditional flag
  48          |             set this when dscp
  49 |-ip diffserv-|        stored in mark
  50 | 6 bits      |
  51 |-------------|
  52
  53 an identically configured tc action to restore looks like:
  54
  55 tc filter show dev eth0 ingress
  56 filter parent ffff: protocol all pref 10 u32 chain 0
  57 filter parent ffff: protocol all pref 10 u32 chain 0 fh 800: ht divisor 1
  58 filter parent ffff: protocol all pref 10 u32 chain 0 fh 800::800 order 2048 key ht 800 bkt 0 flowid 1:1 not_in_hw
  59   match 00000000/00000000 at 0
  60         action order 1: ctinfo zone 0 pipe
  61          index 2 ref 1 bind 1 dscp 0xfc000000/0x1000000
  62
  63         action order 2: mirred (Egress Redirect to device ifb4eth0) stolen
  64         index 1 ref 1 bind 1
  65
  66 |----0xFC----conntrack mark----000000---|
  67 | Bits 31-26 | bit 25 | bit24 |~~~ Bit 0|
  68 | DSCP       | unused | flag  |unused   |
  69 |-----------------------0x01---000000---|
  70       |                   |
  71       |                   |
  72       ---|             Conditional flag
  73          v             only restore if set
  74 |-ip diffserv-|
  75 | 6 bits      |
  76 |-------------|
  77
  78 Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
  79 ---
  80  include/uapi/linux/netfilter/xt_connmark.h |  3 ++-
  81  net/netfilter/xt_connmark.c                | 17 +++++++++++++++++
  82  2 files changed, 19 insertions(+), 1 deletion(-)
  83
  84 --- a/include/uapi/linux/netfilter/xt_connmark.h
  85 +++ b/include/uapi/linux/netfilter/xt_connmark.h
  86 @@ -16,7 +16,8 @@
  87  enum {
  88         XT_CONNMARK_SET = 0,
  89         XT_CONNMARK_SAVE,
  90 -       XT_CONNMARK_RESTORE
  91 +       XT_CONNMARK_RESTORE,
  92 +       XT_CONNMARK_SAVEDSCP
  93  };
  94
  95  enum {
  96 --- a/net/netfilter/xt_connmark.c
  97 +++ b/net/netfilter/xt_connmark.c
  98 @@ -42,6 +42,7 @@ connmark_tg_shift(struct sk_buff *skb, c
  99         u_int32_t new_targetmark;
 100         struct nf_conn *ct;
 101         u_int32_t newmark;
 102 +       u_int8_t dscp;
 103
 104         ct = nf_ct_get(skb, &ctinfo);
 105         if (ct == NULL)
 106 @@ -74,6 +75,21 @@ connmark_tg_shift(struct sk_buff *skb, c
 107                         nf_conntrack_event_cache(IPCT_MARK, ct);
 108                 }
 109                 break;
 110 +       case XT_CONNMARK_SAVEDSCP:
 111 +               if (skb->protocol == htons(ETH_P_IP))
 112 +                       dscp = ipv4_get_dsfield(ip_hdr(skb)) >> 2;
 113 +               else if (skb->protocol == htons(ETH_P_IPV6))
 114 +                       dscp = ipv6_get_dsfield(ipv6_hdr(skb)) >> 2;
 115 +               else    /* protocol doesn't have diffserv */
 116 +                       break;
 117 +
 118 +               newmark = (ct->mark & ~info->ctmark) |
 119 +                         (info->ctmask | (dscp << info->shift_bits));
 120 +               if (ct->mark != newmark) {
 121 +                       ct->mark = newmark;
 122 +                       nf_conntrack_event_cache(IPCT_MARK, ct);
 123 +               }
 124 +               break;
 125         case XT_CONNMARK_RESTORE:
 126                 new_targetmark = (ct->mark & info->ctmask);
 127                 if (info->shift_dir == D_SHIFT_RIGHT)
 128 @@ -86,6 +102,7 @@ connmark_tg_shift(struct sk_buff *skb, c
 129                 skb->mark = newmark;
 130                 break;
 131         }
 132 +out:
 133         return XT_CONTINUE;
 134  }
 135