On 03/06/2017 10:22 AM, Kevin Wolf wrote: > change_parent_backing_link() will need to update multiple BdrvChild > objects at once. Checking permissions reference by reference doesn't > work because permissions need to be consistent only with all parents > moved to the new child. > > Signed-off-by: Kevin Wolf <kw...@redhat.com> > --- > block.c | 35 ++++++++++++++++++++++------------- > include/block/block_int.h | 2 +- > 2 files changed, 23 insertions(+), 14 deletions(-) >
> static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t
> new_used_perm,
> uint64_t new_shared_perm,
> - BdrvChild *ignore_child, Error **errp)
> + GSList *ignore_children, Error **errp)
> {
> BdrvChild *c;
> uint64_t cumulative_perms = new_used_perm;
> @@ -1572,7 +1574,7 @@ static int bdrv_check_update_perm(BlockDriverState *bs,
> uint64_t new_used_perm,
> assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
>
> QLIST_FOREACH(c, &bs->parents, next_parent) {
> - if (c == ignore_child) {
> + if (g_slist_find(ignore_children, c)) {
Quadratic complexity (searching one list for each element of another
list). Hopefully the combination of lists isn't too long in practice (it
might be a potential problem if someone creates 200 snapshots then tries
to commit them down to 1 image in a single operation, where
ignore_children would then be a long list of intermediate children being
iterated on multiple times, but I don't know if bs->parents can easily
be made as long). We could always create some sort of hash table to
reduce complexity, as a followup patch, if it turns out to be a problem
in practice, but for now I'm okay with it.
Reviewed-by: Eric Blake <ebl...@redhat.com>
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
