Cathy Zhou wrote: >> Cathy Zhou wrote: >>> Erik Nordmark wrote: >>>> Cathy Zhou wrote: >>>>> If I understand correctly, today it is using the /devices clone node >>>>> of a specific device, and that will not be changed. >>>> Are there any issues as this is handed over to the dladmd? For >>>> instance, the kernel might already use e.g. the bge0 datalink and IP >>>> be configured with bge0. But dladmd might have a vanity name for bge0. >>>> What happens in that case? >>>> >>> When the device attaches before dladmd is started, the kernel >>> data-link stil be able to be created but will not be able to added >>> into the dls_vlan_hash table (keyed by linkid) as it is not able to >>> lookup its linkid at that time. But the kernel data-link should still >>> be created and added to the dls_vlan_hash_by_dev hash table, so that >>> dld_open is able to find the corresponding kernel data-link structure >>> so that other DLPI operations will be able to proceed. >>> >>> But this physical link will not be able to have any vanity naming >>> feature (no /dev/net node will be created for it), and the dladm >>> command might also act unexpected for this link. >>> >>> Is this a problem? I see "boot net" is only a temporary state and >>> things should go fine after next reboot. >>> >> But what of machines that are configured to always boot from the net, >> such as diskless clients? >> > Before I tries to answer this question, can you please help me to understand > what a diskless client is usually expected to do? I am sorry that I am not > familiar with this area. For example, how any DLPI application can work > without the network /dev symlinks today? >
My comment was based on your seeming design assumption that a system would be booted from the net only for installation purposes, and diskless clients are the direct counter-example to that assumption. As far as I know, presently DLPI applications work on those clients, but I haven't run one in a very long time. Dave > Thanks > - Cathy
