On 10/08/2013 08:13 AM, Xu Wang wrote:
These patches are based on commit 04bfeb825e5e155ffa72119253de608ccf3bd72b. So I may need more work on rebasing.
Most of bugs reported recently are about some tags libvirt-cim doesn't support will be dropped after resource updated. So a new member was added into every virt_device and domain structure. And a new structure named unknown device was added to save those tags new added.
The original implementaion is reading every member of virt_device structure from xml and save it. But the defect of this is some tags were not a member of virt_device were dropped. After resource updated, every data in the virt_device will be used to regenerate new xml. Hence the tags unsupported above, disappeared.
So I added a member into every virt_device and domain structure, 'others'. It's a link list and used to save all data read from xml. Another new structure 'unknown_device' was added to save data except libvirt-cim could recognize.
The new implementation is, firstly parse_*_device() could read all nodes and properties from xml and save them into 'others' link list. Then every member of virt_device will fetch data from others link list and save it. So that nodes in the 'others' link list could be saved until they are used to re- generate xml.
After resource updating finished, libvirt-cim will call *_xml() to generate xml. The new process of generating xml like this, firstly all data in the members of virt_device will be restored into 'others' link list, then a function (others_to_xml) will use this link list to generate xml.
Some points I have updated, 1. 'others' link list has to be processed in _get_proc_device() and _get_mem_device(). They should be copied into new data structure. 2. If resource updating happened, others field should be cleared because this device has been changed and they are useless.
Besides above some logic may be a little strange or boring. Implemention like that is just to be compatible with upper layer functions (to make changes as less as possible). After keep a balance I decided to devide the whole xml into several parts: <domain>------------------------others in domain to save unsupported sub-nodes of domain -some fields (devices, mem, vcpu...) will be skiped because -they have their own parsing functions. <name>xxx</name> <uuid>xxx</uuid> <vcpu>xxx</vcpu> <mem>xxx</mem> <devices> <disk>xxx</disk>------- others in virt_device to save unsupported tags of this device <emulator>xxx</emulator> ...----------- unknown_device to save unsupported device except like <disk>,<emu>,etc. </devices> </domain>
Hence, all nodes read from xml will be restored after xml generation.
Short notification, Boris or I will do a detail review (hopefully) by the end of next week (both being pretty busy right now) as this is touching some of the areas we're working on for s390 enablement. But in general I think it's a good idea and like the overall approach. -- Mit freundlichen Grüßen/Kind Regards Viktor Mihajlovski IBM Deutschland Research & Development GmbH Vorsitzender des Aufsichtsrats: Martina Köderitz Geschäftsführung: Dirk Wittkopp Sitz der Gesellschaft: Böblingen Registergericht: Amtsgericht Stuttgart, HRB 243294