Resolve a storage driver crash as a result of a long running
storageVolCreateXML when the virStorageVolPoolRefreshThread is
run as a result of when a storageVolUpload completed and ran the
virStoragePoolObjClearVols without checking if the creation
code was currently processing a buildVol after incrementing
the driver->asyncjob count.
The refreshThread will now check the pool asyncjob count before
attempting to pursue the pool refresh. Adjust the documentation
to describe the condition.
Crash from valgrind is as follows (with a bit of editing):
==21309== Invalid read of size 8
==21309== at 0x153E47AF: storageBackendUpdateVolTargetInfo
==21309== by 0x153E4C30: virStorageBackendUpdateVolInfo
==21309== by 0x153E52DE: virStorageBackendVolRefreshLocal
==21309== by 0x153DE29E: storageVolCreateXML
==21309== by 0x562035B: virStorageVolCreateXML
==21309== by 0x147366: remoteDispatchStorageVolCreateXML
...
==21309== Address 0x2590a720 is 64 bytes inside a block of size 336 free'd
==21309== at 0x4C2F2BB: free
==21309== by 0x54CB9FA: virFree
==21309== by 0x55BC800: virStorageVolDefFree
==21309== by 0x55BF1D8: virStoragePoolObjClearVols
==21309== by 0x153D967E: virStorageVolPoolRefreshThread
...
==21309== Block was alloc'd at
==21309== at 0x4C300A5: calloc
==21309== by 0x54CB483: virAlloc
==21309== by 0x55BDC1F: virStorageVolDefParseXML
==21309== by 0x55BDC1F: virStorageVolDefParseNode
==21309== by 0x55BE5A4: virStorageVolDefParse
==21309== by 0x153DDFF1: storageVolCreateXML
==21309== by 0x562035B: virStorageVolCreateXML
==21309== by 0x147366: remoteDispatchStorageVolCreateXML
...
Signed-off-by: John Ferlan <jferlan(a)redhat.com>
---
v1:
https://www.redhat.com/archives/libvir-list/2017-November/msg00198.html
Changes since v1:
- From review, remove the retry if Asyncjobs > 0 logic and replace with
a VIR_DEBUG indicating not doing refresh due to Asyncjob running and
just goto cleanup.
- Adjust the virStorageVolUpload docs to note that an attempt will be
made to refresh the pool.
- Drop patch 2 which added the Asyncjobs > 0 check to the SCSI pool
refresh thread. Since a SCSI pool doesn't support buildVol, the
asyncjob count cannot be anything other than zero, so it's pointless.
src/libvirt-storage.c | 3 ++-
src/storage/storage_driver.c | 7 +++++++
2 files changed, 9 insertions(+), 1 deletion(-)