On 27/11/13 06:31, Nehal J Wani wrote:
On running the command make -C tests valgrind, there used to be a bunch of memory leaks shown by valgrind. Specifically, one can check it by running: libtool --mode=execute valgrind --quiet --leak-check=full --suppressions=./.valgrind.supp qemuhotplugtest The issue was that def->info->alias was already malloc'ed by xmlStrndup in virDomainDeviceInfoParseXML (domain_conf.c:3439). The new alias was being assigned again without freeing the old one in qemuAssignDeviceAliases(). This patch checks if the entity exists, and frees accordingly, hence making valgrind cry lesser.
--- src/qemu/qemu_command.c | 39 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 39 insertions(+)
diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c index 763417f..bbec1d4 100644 --- a/src/qemu/qemu_command.c +++ b/src/qemu/qemu_command.c @@ -979,6 +979,8 @@ qemuAssignDeviceAliases(virDomainDefPtr def, virQEMUCapsPtr qemuCaps) size_t i;
for (i = 0; i < def->ndisks; i++) { + if (def->disks[i]->info.alias) + VIR_FREE(def->disks[i]->info.alias);
Instead of free'ing it, it should be used to avoid calculate/strdup the alias again. Since the "alias" from virDomainDeviceInfoParseXML only works for active domain's XML: <snip> if (alias == NULL && !(flags & VIR_DOMAIN_XML_INACTIVE) && xmlStrEqual(cur->name, BAD_CAST "alias")) { alias = cur; </snip> And I don't think using the "alias" existing in the XML could cause any conflicts for an active domain. Regards, Osier