Daemon leaks instances when launching with low --disk

[ricab]

The --disk option does not work with a size below the virtual size of the relevant image. This makes sense (to avoid data loss). However, it is currently only enforced by the underlying qemu-img command.

multipassd itself does not perform this check and leaves files behind when qemu-img fails. The most significant of which is the copy of the image file.

This means that a few hundreds of MB of disk are leaked on such occasions. The regular user does not realize this and leaves the files behind.

$ multipass launch --disk 1G
launch failed: Cannot resize instance image

$ multipass list
No instances found.

$ du -sh ~/.local/share/multipassd/vault/instances/deep-midge/*
20K     /home/ricab/.local/share/multipassd/vault/instances/deep-midge/cloud-init-config.iso
328M    /home/ricab/.local/share/multipassd/vault/instances/deep-midge/ubuntu-18.04-server-cloudimg-amd64.img

$ qemu-img info /home/ricab/.local/share/multipassd/vault/instances/deep-midge/ubuntu-18.04-server-cloudimg-amd64.img | grep size
virtual size: 2.2G (2361393152 bytes)
disk size: 328M
cluster_size: 65536

The daemon logs

[warning] [qemu-img] qemu-img: warning: Shrinking an image will delete all data beyond the shrunken image's end. Before performing such an operation, make sure there is no important data there.
qemu-img: Use the --shrink option to perform a shrink operation.