There are two problems that make the block-commit QMP command difficult or impossible to use with a setup where the block graph is built node by node with individual -blockdev options or blockdev-add QMP commands.

1. The block-commit only accepts filenames to identify the top and the base node for the commit operation. However, with individual options set, this can easily cause QEMU to use a json: filename, the exact layout of which is hard to predict. Instead, block-commit should offer options to identify top and base by their node name.

2. Commit automatically reopens the base if it is currently reopened read-only. In a graph built node by node, however, the reopen fails because base->file isn't also reopened and you can't have a read-write format layer on top of a read-only protocol layer. Commit must be changed so that it reopens all relevant nodes so that it can be used even when the base node is opened read-only originally.

The following command line runs into the same problem 2, though I can't say yet if it should work in the end or not (depends on whether the best fix is in reopen itself or in commit's use of reopen):
qemu-io -r -c 'reopen -w' --image-opts driver=qcow2,file.driver=file,file.filename=/tmp/backing.qcow2,file.read-only=on,read-only=on