target/linux/generic/hack-4.14/645-netfilter-connmark-introduce-set-dscpmark.patch

   1 From f1627abea333781d3e2a61bac4c7fd4502395741 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 set-dscpmark
   5
   6 set-dscpmark is a method of storing the DSCP of an ip packet into
   7 conntrack mark.  In combination with a suitable tc filter action
   8 (act_ctinfo) DSCP values are able to be stored in the mark on egress and
   9 restored on 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 set-dscpmark target solves the problem of storing the
  20 DSCP to the conntrack mark in a way suitable for the new act_ctinfo tc
  21 action to restore.
  22
  23 The set-dscpmark option accepts 2 parameters, a 32bit 'dscpmask' and a
  24 32bit 'statemask'.  The dscp mask must be 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 --set-dscpmark 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: 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  net/netfilter/xt_connmark.c | 19 +++++++++++++++++--
  81  1 file changed, 17 insertions(+), 2 deletions(-)
  82
  83 --- a/net/netfilter/xt_connmark.c
  84 +++ b/net/netfilter/xt_connmark.c
  85 @@ -42,6 +42,7 @@ connmark_tg(struct sk_buff *skb, const s
  86         enum ip_conntrack_info ctinfo;
  87         struct nf_conn *ct;
  88         u_int32_t newmark;
  89 +       u_int8_t dscp, maskshift;
  90
  91         ct = nf_ct_get(skb, &ctinfo);
  92         if (ct == NULL)
  93 @@ -49,7 +50,21 @@ connmark_tg(struct sk_buff *skb, const s
  94
  95         switch (info->mode) {
  96         case XT_CONNMARK_SET:
  97 -               newmark = (ct->mark & ~info->ctmask) ^ info->ctmark;
  98 +               if (info->nfmask & 0x80000000) {
  99 +                       newmark = (ct->mark & ~info->ctmask) ^ info->ctmark;
 100 +               } else {
 101 +                       if (skb->protocol == htons(ETH_P_IP))
 102 +                               dscp = ipv4_get_dsfield(ip_hdr(skb)) >> 2;
 103 +                       else if (skb->protocol == htons(ETH_P_IPV6))
 104 +                               dscp = ipv6_get_dsfield(ipv6_hdr(skb)) >> 2;
 105 +                       else    /* protocol doesn't have diffserv */
 106 +                               break;
 107 +
 108 +                       /* nfmask contains the mask shift value */
 109 +                       maskshift = info->nfmask & 0x1f;
 110 +                       newmark = (ct->mark & ~info->ctmark) |
 111 +                                 (info->ctmask | (dscp << maskshift));
 112 +               }
 113                 if (ct->mark != newmark) {
 114                         ct->mark = newmark;
 115                         nf_conntrack_event_cache(IPCT_MARK, ct);
 116 @@ -57,7 +72,7 @@ connmark_tg(struct sk_buff *skb, const s
 117                 break;
 118         case XT_CONNMARK_SAVE:
 119                 newmark = (ct->mark & ~info->ctmask) ^
 120 -                         (skb->mark & info->nfmask);
 121 +                         (skb->mark & info->nfmask);
 122                 if (ct->mark != newmark) {
 123                         ct->mark = newmark;
 124                         nf_conntrack_event_cache(IPCT_MARK, ct);