Eric Blake wrote:
On 07/11/2014 01:04 PM, Jim Fehlig wrote:
> David Kiarie wrote:
>
>> From: Kiarie Kahurani <davidkiarie4(a)gmail.com>
>>
>> Introduce the function
>> xenParseXMDisk(..........);
>>
>> Parses disk config
>>
>> signed-off-by: David Kiarie<davidkiarie4(a)gmail.com>
>> ---
>> src/xenxs/xen_xm.c | 204 ++++++++++++++++++++++++++++-------------------------
>> 1 file changed, 108 insertions(+), 96 deletions(-)
>>
>>
>> @@ -912,7 +839,7 @@ xenParseXM(virConfPtr conf, int xendConfigVersion,
>>
>> /* Maintain list in sorted order according to target device name */
>> if (VIR_APPEND_ELEMENT(def->disks, def->ndisks, disk) < 0)
>> - goto cleanup;
>> + return -1;
>>
>> skipdisk:
>> list = list->next;
>>
>>
> The next line here is
>
> virDomainDiskDefFree(disk);
>
> which is not right in the case VIR_APPEND_ELEMENT succeeds (pre-existing
> bug). 'disk' should be set to NULL if VIR_APPEND_ELEMENT succeeds, so
> we don't free the disk we've just added to the disk list.
>
VIR_APPEND_ELEMENT(, , disk) sets disk=NULL on success, precisely so you
don't have to worry about double-freeing it (if you call the macro, you
are guaranteed either transfer semantics and success, or no change on
failure).
Ah, should have read the source :). Thanks Eric. That explains my
confusion as to why a crash was never reported in this code (or the vif
parsing code).
Regards,
Jim