[Pkg-libvirt-maintainers] Bug#705205: libvirt-bin: Memory leak in virStorageBackendLogicalMakeVol

[Guido Günther]

tag 705205 +pending
thanks

Hi Luca,
On Thu, Apr 11, 2013 at 10:42:24AM +0100, Luca Tettamanti wrote:
> Package: libvirt-bin
> Version: 0.9.12-11
> Severity: important
> Tags: patch
> 
> Hello,
> In our workload we use LVM for VM storage and continuously create and destroy instances.
> We noticed that over time libvirtd memory usage grows until libvirt
> cannot fork anymore (to spawn LVM tools):
> 
> virFork:279 : cannot fork child process: Cannot allocate memory
> 
> libvirtd won't crash (yet), but it will eventually mark all the storage
> pools as offline.
> 
> I've tracked down the issue to a memory leak in
> virStorageBackendLogicalMakeVol (creating a volume triggers a refresh of
> the pool):
> 
> 50,229,128 (820,736 direct, 49,408,392 indirect) bytes in 3,664 blocks are definitely lost in loss record 883 of 883
>    at 0x4C28BED: malloc (vg_replace_malloc.c:263)
>    by 0x4C28D6F: realloc (vg_replace_malloc.c:632)
>    by 0x805ADF3: re_compile_internal (regcomp.c:760)
>    by 0x805B1AB: regcomp (regcomp.c:506)
>    by 0x4D13CD: virStorageBackendLogicalMakeVol (storage_backend_logical.c:203)
>    by 0x4CE09B: virStorageBackendRunProgRegex (storage_backend.c:1512)
>    by 0x4D0A8E: virStorageBackendLogicalFindLVs (storage_backend_logical.c:306)
>    by 0x4D1971: virStorageBackendLogicalRefreshPool (storage_backend_logical.c:591)
>    by 0x4C9596: storagePoolRefresh (storage_driver.c:904)
>    by 0x511CE44: virStoragePoolRefresh (libvirt.c:12116)
>    by 0x436687: remoteDispatchStoragePoolRefreshHelper (remote_dispatch.h:11759)
>    by 0x515B7FF: virNetServerProgramDispatch (virnetserverprogram.c:423)
> 
> This bug has already been fixed upstream with commit 71da3b66 (which
> also reverts afc4631b). This fix is already included in the experimental
> package, but since it's self-contained and minimally intrusive I suggest
> to backport it for Wheezy (and squeeze-backports?).
> Patch is attached.

Thanks for tracking this down! I've pushed this to our wheezy branch and
will hopefully get this into R1.
Cheers,
 -- Guido