On 05/28/2015 04:53 PM, Laine Stump wrote:
On 05/28/2015 04:36 PM, Laine Stump wrote:
From: Lubomir Rintel <lkundrak@v3.sk>

bond devices with no slaves were being misidentified as plain ethernet
devices. This patch checks for BONDING_OPTS in the bond's own ifcfg
file rather than looking for other ifcfg files that list this one as
MASTER.

(The above description was added by me. Please correct if it's wrong)
As a complete novice with xslt, this is what I understand from that code
(someone else wrote all of the xslt in netcf, and it has sat mostly
dormant for several years).

  $ cat /etc/sysconfig/network-scripts/ifcfg-bond
  DEVICE=nm-bond
  TYPE=Bond
  BONDING_MASTER=yes
  BOOTPROTO=none
  DEFROUTE=yes
  IPV4_FAILURE_FATAL=no
  IPV6INIT=yes
  IPV6_AUTOCONF=yes
  IPV6_DEFROUTE=yes
  IPV6_FAILURE_FATAL=no
  NAME=bond
  UUID=66a3c5c4-1106-4a86-aa33-18ac5c7c20f2
  ONBOOT=yes
  BONDING_OPTS=mode=balance-rr
  IPADDR=1.2.3.4
  PREFIX=8
  IPV6_PEERDNS=yes
  IPV6_PEERROUTES=yes

  $ ncftool dumpxml nm-bond
  <?xml version="1.0"?>
  <interface type="ethernet" name="nm-bond">
    <start mode="onboot"/>
    <protocol family="ipv4">
      <ip address="1.2.3.4" prefix="8"/>
    </protocol>
    <protocol family="ipv6">
      <autoconf/>
    </protocol>
  </interface>
---
 data/xml/redhat-put.xsl | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/data/xml/redhat-put.xsl b/data/xml/redhat-put.xsl
index ed56c66..21bd4ce 100644
--- a/data/xml/redhat-put.xsl
+++ b/data/xml/redhat-put.xsl
@@ -135,7 +135,7 @@
   </xsl:template>
 
   <xsl:template name="bond-interface"
-                match="tree[node[@label = 'DEVICE'][@value = //tree/node[@label = 'MASTER']/@value]][count(node[@label = 'BRIDGE']) = 0]">
+                match="tree[count(node[@label = 'BONDING_OPTS']) > 0][count(node[@label = 'BRIDGE']) = 0]">
     <interface type="bond">
       <xsl:call-template name="name-attr"/>
       <xsl:call-template name="startmode"/>


I *think* ACK, but I have a couple of questions before I push this:

1) Is a BONDING_OPTS attribute required in an ifcfg file for a bond? If
it is, this change is okay, but if not we will need to find some other
way to identify bond devices.

After forgetting about this patch for the last two weeks, I looked into what initscripts does to determine the type of an interface. Basically it looks at the TYPE attribute in the ifcfg file, and if one isn't found then it guesses based on the name of the device. If it is bond[0-9]* then the type is set to bond. Even if TYPE is set in the ifcfg, there is no setting of TYPE that would lead to it being considered a bond by initscripts (this is in the function source_config() in /etc/sysconfig/network-scripts/network-functions).

So should we be doing the same thing? How does NM determine that a device is a bond?


2) what does the "[count(node[@label = 'BRIDGE']) = 0]" do? does that
exclude from the match any ifcfg that has an attribute named "BRIDGE"?

Okay, I think I understand that part - since netcf sees a bond connected to a bridge as a single interface, this is making sure that any bond connected to a bridge isn't listed as a toplevel interface, but is instead only considered as a subordinate of the bridge (which is, in netcf's eyes, the toplevel interface).


If so, that doesn't seem correct, because it's acceptable for a bond
device to be attached to a bridge. (yes, I know that part was in the
expression before, but as I said this code was written by someone else
and has been mostly untouched for 4 years or so).