On 18.11.2014 22:26, John Ferlan wrote:
https://bugzilla.redhat.com/show_bug.cgi?id=1159180 Currently libvirt only detects duplicate fc_host & scsi_host adapter sources when the incoming source type definition is the same as the pool type. This misses the even more oddball cases where a scsi_host/fc_host definition is using a scsi_hostN where the type of the definition isn't the same as the type of the pool. It's a fairly twisted and perhaps edge kind of case. A 'fc_host' has two scsi_hostN values that need to be compared against existing 'scsi_host' pools - first the fc_host "parent" scsi_hostN (if it exists or was provided) and second the vHBA scsi_hostN as determined from the wwnn/wwpn. If a 'fc_host' has a 'parent' attribute of a scsi_hostN that's already in use by some 'scsi_host' pool or if it's vHBA (as created via nodedev-create) was (for some unknown reason) used by someone for a 'scsi_host' pool, then we have a duplicate. A 'scsi_host' conversely needs to check against using either the fc_host 'parent' or vHBA. If a 'scsi_host' is using a scsi_hostN of some existing fc_host 'parent' attribute or vHBA, then we have a duplicate. To make matters more complex, the 'fc_host' 'parent' attribute doesn't have to be provided in the fc_host XML. If not provided, then it can be determined if the vHBA already exists (either via nodedev-create or a running fc_host pool). This means moving the code that was created for fc_host startup to find the vHBA parent into the storage_backend_scsi code in order for it to be used there instead. I did try moving into virutil with no success since that code wasn't very happy to be calling the virNodeDevice* API's. The only oddball case that cannot be tested for in this scenario is if the incoming fc_host definition isn't using a nodedev-create'd vHBA and it doesn't provide a 'parent' attribute, then it is possible that the vportCreate startup code will find "an available" scsi_hostN that was (again for some strange reason) already being used by some 'scsi_host' pool. Added some more verbiage to the documentation to dissuade usage of a 'scsi_host' for an FC capable scsi_hostN as well as to encourage usage of the 'parent' attribute in a mixed environment John Ferlan (2): storage: Move and rename getVhbaSCSIHostParent storage: Add mixed fc_host/scsi_host duplicate adapter source checks docs/formatstorage.html.in | 26 +++++- src/conf/storage_conf.c | 178 ++++++++++++++++++++++++++++++++++++- src/conf/storage_conf.h | 8 +- src/libvirt_private.syms | 1 + src/parallels/parallels_storage.c | 2 +- src/storage/storage_backend_scsi.c | 66 +------------- src/storage/storage_driver.c | 4 +- 7 files changed, 215 insertions(+), 70 deletions(-)
ACK series. But see my comment to 1/2. Michal -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list