On 16.08.2018 00:17, Siwei Liu wrote: > On Wed, Aug 15, 2018 at 12:05 PM, Samudrala, Sridhar > <sridhar.samudrala@xxxxxxxxx> wrote: >> On 8/14/2018 5:03 PM, Siwei Liu wrote: >>> >>> Are we sure all userspace apps skip and ignore slave interfaces by >>> just looking at "IFLA_MASTER" attribute? >>> >>> When STANDBY is enabled on virtio-net, a failover master interface >>> will appear, which automatically enslaves the virtio device. But it is >>> found out that iSCSI (or any network boot) cannot boot strap over the >>> new failover interface together with a standby virtio (without any VF >>> or PT device in place). >>> >>> Dracut (initramfs) ends up with timeout and dropping into emergency shell: >>> >>> [ 228.170425] dracut-initqueue[377]: Warning: dracut-initqueue >>> timeout - starting timeout scripts >>> [ 228.171788] dracut-initqueue[377]: Warning: Could not boot. >>> Starting Dracut Emergency Shell... >>> Generating "/run/initramfs/rdsosreport.txt" >>> Entering emergency mode. Exit the shell to continue. >>> Type "journalctl" to view system logs. >>> You might want to save "/run/initramfs/rdsosreport.txt" to a USB stick or >>> /boot >>> after mounting them and attach it to a bug report. >>> dracut:/# ip l sh >>> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN >>> mode DEFAULT group default qlen 1000 >>> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 >>> 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue >>> state UP mode DEFAULT group default qlen 1000 >>> link/ether 9a:46:22:ae:33:54 brd ff:ff:ff:ff:ff:ff\ >>> 3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast >>> master eth0 state UP mode DEFAULT group default qlen 1000 >>> link/ether 9a:46:22:ae:33:54 brd ff:ff:ff:ff:ff:ff >>> dracut:/# >>> >>> If changing dracut code to ignore eth1 (with IFLA_MASTER attr), >>> network boot starts to work. >> >> >> Does dracut by default tries to use all the interfaces that are UP? >> > Yes. The specific dracut cmdline of our case is "ip=dhcp > netroot=iscsi:... ", but it's not specific to iscsi boot. And because > of same MAC address for failover and standby, while dracut tries to > run DHCP on all interfaces that are up it eventually gets same route > for each interface. Those conflict route entries kill off the network > connection. > >> >>> >>> The reason is that dracut has its own means to differentiate virtual >>> interfaces for network boot: it does not look at IFLA_MASTER and >>> ignores slave interfaces. Instead, users have to provide explicit >>> option e.g. bond=eth0,eth1 in the boot line, then dracut would know >>> the config and ignore the slave interfaces. >> >> >> Isn't it possible to specify the interface that should be used for network >> boot? > As I understand it, one can only specify interface name for running > DHCP but not select interface for network boot. We want DHCP to run > on every NIC that is up (excluding the enslaved interfaces), and only > one of them can get a route entry to the network boot server (ie.g. > iSCSI target). > >> >> >>> >>> However, with automatic creation of failover interface that assumption >>> is no longer true. Can we change dracut to ignore all slave interface >>> by checking IFLA_MASTER? I don't think so. It has a large impact to >>> existing configs. >> >> >> What is the issue with checking for IFLA_MASTER? I guess this is used with >> team/bonding setups. > That should be discussed within and determined by the dracut > community. But the current dracut code doesn't check IFLA_MASTER for > team or bonding specifically. I guess this change might have broader > impact to existing userspace that might be already relying on the > current behaviour. > > Thanks, > -Siwei Is there a sysfs flag for IFF_SLAVE? Or any "ip" output I can use to detect, that it is a IFF_SLAVE?
