Comment 0 for bug 1337873

It was brought to my attention (by others) that ifupdown runs into race conditions on some specific cases.

[Impact]

When trying to deploy many servers at once (higher chances of happening) or from time-to-time, like any other intermittent race-condition. Interfaces are not brought up like they should and this has a big impact for servers that cannot rely on network start scripts.

The problem is caused by a race condition when init(upstart) starts up network interfaces in parallel.

[Test Case]

Use attached script to reproduce the error (it might take some hours, in a single virtual machine, for the error to occur).

(example 1)

*** sequence to trigger race-condition ***

(a) ifup eth0 (b) ifup -a for eth0
-----------------------------------------------------------------
1-1. Lock ifstate.lock file.
                                  1-1. Wait for locking ifstate.lock
                                      file.
1-2. Read ifstate file to check
     the target NIC.
1-3. close(=release) ifstate.lock
     file.
1-4. Judge that the target NIC
     isn't processed.
                                  1-2. Read ifstate file to check
                                       the target NIC.
                                  1-3. close(=release) ifstate.lock
                                       file.
                                  1-4. Judge that the target NIC
                                       isn't processed.
2. Lock and update ifstate file.
   Release the lock.
                                  2. Lock and update ifstate file.
                                     Release the lock.

(example 2)

Bonding device using eth0.
ifenslave for eth0 is also executed in parallel, eth0 remains down.

*** sequence to trigger race-condition ***

(a) ifenslave of eth0 (b) ifenslave of eth0
------------------------------------------------------------------
3. Execute ifenslave of eth0. 3. Execute ifenslave of eth0.
4. Link down the target NIC.
5. Write NIC id to
   /sys/class/net/bond0/bonding
   /slaves then NIC gets up
                                  4. Link down the target NIC.
                                  5. Fails to write NIC id to
                                     /sys/class/net/bond0/bonding/
                                     slaves it is already written.

(example 3)

bonding is not set to active-backup as defined in config file: When the init(upstart) executes "if-pre-up.d/ifenslave" script and "if-pre-up.d/vlan" script for bond0 device in parallel, the "if-pre-up.d/ifenslave" script fails to change the bonding mode with a error message, "bonding: unable to update mode of bond0 because interface is up.".

*** sequence to trigger race-condition ***

(a)ifup bond0 (b)ifup -a
-----------------------------------------------------------------------
1. Update statefile about bond0.
                                  1. Does nothing about bond0
                                     because statefile is already
                                     updated about it.
2. ifenslave::setup_master()
   sysfs_change_down mode 1
   and link down bond0.
                                  2. Link up bond0 by the vlan
                                     script on the processing
                                     for linking up bond0.201(*1).
3. "echo 1 > .../mode" fails.

[ /etc/network/if-pre-up.d/vlan ]

46 if [ -n "$IF_VLAN_RAW_DEVICE" ] && [ ! -d /sys/class/net/$IFACE ]; then
47 if [ ! -x /sbin/vconfig ]; then
48 exit 0
49 fi
50 if ! ip link show dev "$IF_VLAN_RAW_DEVICE" > /dev/null; then
51 echo "$IF_VLAN_RAW_DEVICE does not exist, unable to create $IFACE"
52 exit 1
53 fi
54 ip link set up dev $IF_VLAN_RAW_DEVICE <-- (*1).
55 vconfig add $IF_VLAN_RAW_DEVICE $VLANID
56 fi

[Regression Potential]

 * Attaching proposed patch (for upstream as well) and describing potential later on today.

[Other Info]

Example: [ /etc/network/interfaces ]

auto lo
iface lo inet loopback

auto eth0
iface eth0 inet manual
 bond-master bond0

auto eth1
iface eth1 inet manual
 bond-master bond0

auto bond0
iface bond0 inet dhcp
 bond-slaves eth0 eth1
 hwaddress 11:22:33:44:55:66
 bond-primary eth0
 bond-mode 1
 bond-miimon 100
 bond-updelay 200
 bond-downdelay 200

auto bond0.201
iface bond0.201 inet dhcp
 hwaddress 11:22:33:44:55:66
 vlan-raw-device bond0
...

auto bond0.205
iface bond0.205 inet dhcp
 hwaddress 11:22:33:44:55:66
 vlan-raw-device bond0