On 16/04/18 10:33, Daniel P. Berrangé wrote:
> On Sun, Apr 15, 2018 at 04:30:11PM +0100, Radostin Stoyanov wrote:
>> Remove the /.oldroot directory after it has been unmounted (at the end
>> of lxcContainerSetupPivotRoot). Ignore errors silently.
>>
>> Signed-off-by: Radostin Stoyanov <rstoyanov1@xxxxxxxxx>
>> ---
>> src/lxc/lxc_container.c | 3 +++
>> 1 file changed, 3 insertions(+)
>>
>> diff --git a/src/lxc/lxc_container.c b/src/lxc/lxc_container.c
>> index 665b93a0a..dd4e38703 100644
>> --- a/src/lxc/lxc_container.c
>> +++ b/src/lxc/lxc_container.c
>> @@ -1811,6 +1811,9 @@ static int lxcContainerSetupPivotRoot(virDomainDefPtr vmDef,
>> if (lxcContainerUnmountSubtree("/.oldroot", true) < 0)
>> goto cleanup;
>>
>> + if (virFileRemove("/.oldroot", 0, 0) < 0)
>> + VIR_DEBUG("Failed to remove /.oldroot after start");
>> +
> I think this introduces a race condition. There can be two containers
> with the same root filesystem. If we start both at the same time, then
> this deletion of /.oldroot can cause the other contanier to fail to
> start if it saw that /.oldroot already existed & it thus tried to skip
> mkdir.
Thank you for the review, I hadn't thought about this case.
> Leaving the empty directory is harmless IMHO
I agree that leaving the empty directory is harmless.
Regards,
Radostin
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list
