@@ -364,9 +364,13 @@ cbw_child_perm(BlockDriverState *bs, BdrvChild *c, BdrvChildRole role,
perm, shared, nperm, nshared);
if (!QLIST_EMPTY(&bs->parents)) {
- if (perm & BLK_PERM_WRITE) {
- *nperm = *nperm | BLK_PERM_CONSISTENT_READ;
- }
+ /*
+ * Note, that source child may be shared with backup job. Backup job
+ * does create own blk parent on copy-before-write node, so this
+ * works even if source node does not have any parents before backup
+ * start
+ */
+ *nperm = *nperm | BLK_PERM_CONSISTENT_READ;
*nshared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
}
}
In case when source node does not have any parents, the condition still works as required: backup job do create the parent by block_job_create -> block_job_add_bdrv -> bdrv_root_attach_child Still, in this case checking @perm variable doesn't work, as backup job creates the root blk with empty permissions (as it rely on CBW filter to require correct permissions and don't want to create extra conflicts). So, we should not check @perm. The hack may be dropped entirely when transactional insertion of filter (when we don't try to recalculate permissions in intermediate state, when filter does conflict with original parent of the source node) merged (old big series "[PATCH v5 00/45] Transactional block-graph modifying API"[1] and it's current in-flight part is "[PATCH v8 0/7] blockdev-replace"[2]) [1] https://patchew.org/QEMU/20220330212902.590099-1-vsementsov@openvz.org/ [2] https://patchew.org/QEMU/20231017184444.932733-1-vsementsov@yandex-team.ru/ Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru> --- block/copy-before-write.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)