pxe-server/playbook/shopfloor-setup/common/test/lib/qga.sh

#!/usr/bin/env bash
# qga.sh - host-side helpers for driving the win11 VM via qemu-guest-agent.
#
# Source this from harness scripts: `source "$(dirname "$0")/../lib/qga.sh"`
#
# All commands run as NT AUTHORITY\SYSTEM inside the VM (qemu-ga's service
# context). See reference-vm-qga-as-system memory note for why this is
# preferred over WinRM for dispatcher / manifest-engine tests.

set -euo pipefail

VM_DOMAIN="${VM_DOMAIN:-win11}"
VM_IP="${VM_IP:-192.168.122.225}"
HOST_SAMBA_USER="${HOST_SAMBA_USER:-camp}"
HOST_SAMBA_PASS="${HOST_SAMBA_PASS:-vos313}"
HOST_SAMBA_HOST="${HOST_SAMBA_HOST:-192.168.122.1}"

# Helper: fail loud
die() { echo "[ERROR] $*" >&2; exit 1; }
log() { printf '[%s] %s\n' "$(date +%H:%M:%S)" "$*"; }

# Send a JSON command to qemu-ga, return parsed return field
qga() {
    local payload="$1"
    virsh -c qemu:///system qemu-agent-command "$VM_DOMAIN" "$payload" 2>/dev/null \
        || die "qga call failed: $payload"
}

# Run a PowerShell snippet inside the VM as SYSTEM. Stdin = snippet,
# stdout = exit-decorated combined stdout+stderr from the PS process.
qga_run_ps() {
    python3 - <<'PY'
import base64, json, subprocess, sys, time, os
DOMAIN = os.environ.get('VM_DOMAIN', 'win11')
TIMEOUT = int(os.environ.get('QGA_TIMEOUT', '300'))
def virsh(cmd):
    p = subprocess.run(['virsh','-c','qemu:///system','qemu-agent-command',DOMAIN, json.dumps(cmd)],
                       capture_output=True, text=True, timeout=120)
    if p.returncode != 0:
        sys.exit(f"virsh err: {p.stderr.strip()}")
    return json.loads(p.stdout)['return']
snippet = sys.stdin.read()
args = ['-NoProfile','-ExecutionPolicy','Bypass','-Command', snippet]
pid = virsh({'execute':'guest-exec','arguments':{'path':'powershell.exe','arg':args,'capture-output':True}})['pid']
deadline = time.time() + TIMEOUT
while time.time() < deadline:
    st = virsh({'execute':'guest-exec-status','arguments':{'pid':pid}})
    if st.get('exited'):
        out = base64.b64decode(st.get('out-data','')).decode('utf-8','replace')
        err = base64.b64decode(st.get('err-data','')).decode('utf-8','replace')
        rc = st.get('exitcode')
        sys.stdout.write(out)
        if err:
            sys.stdout.write('\n--- STDERR ---\n')
            sys.stdout.write(err)
        sys.stdout.write(f'\n--- exit {rc} ---\n')
        sys.exit(0 if rc == 0 else 1)
    time.sleep(0.5)
sys.exit(f"timeout waiting for pid {pid}")
PY
}

# Wait until qemu-ga responds to guest-ping
vm_wait_for_ready() {
    local max=${1:-90}
    local i=0
    while ! virsh -c qemu:///system qemu-agent-command "$VM_DOMAIN" '{"execute":"guest-ping"}' >/dev/null 2>&1; do
        ((i++))
        if (( i > max )); then die "VM never became ready ($max attempts)"; fi
        sleep 1
    done
}

# Make sure VM is up and qga is alive. Start if not.
vm_start_if_needed() {
    local state
    state=$(virsh -c qemu:///system domstate "$VM_DOMAIN" 2>/dev/null || echo unknown)
    if [[ "$state" != "running" ]]; then
        log "VM is $state - starting"
        virsh -c qemu:///system start "$VM_DOMAIN" >/dev/null
    fi
    vm_wait_for_ready 90
}

# Mount the host samba share inside the VM as SYSTEM. Idempotent.
# After this call Z:\ inside the VM points at /home/camp/pxe-images/.
vm_mount_share() {
    qga_run_ps <<PS
cmdkey /add:$HOST_SAMBA_HOST /user:$HOST_SAMBA_USER /pass:$HOST_SAMBA_PASS | Out-Null
net use Z: \\\\$HOST_SAMBA_HOST\\pxe-images /persistent:no 2>&1 | Out-Null
"Z reachable: \$(Test-Path 'Z:\\')"
PS
}

# Upload a single file from host to VM by reading it into VM via WriteAllBytes.
# Best for tiny files (config, single ps1). For bigger files use vm_mount_share + Copy-Item.
vm_put_file() {
    local local_path="$1" remote_path="$2"
    [[ -f "$local_path" ]] || die "no such file: $local_path"
    local b64
    b64=$(base64 -w0 < "$local_path")
    qga_run_ps <<PS
\$bytes = [System.Convert]::FromBase64String('$b64')
\$dst = '$remote_path'
New-Item -ItemType Directory -Force -Path (Split-Path \$dst -Parent) -ErrorAction SilentlyContinue | Out-Null
[System.IO.File]::WriteAllBytes(\$dst, \$bytes)
"wrote \$dst (\$(\$bytes.Length) bytes)"
PS
}

# Revert the VM to the blank-slate snapshot, then start + wait.
vm_revert_to_blank_slate() {
    log "reverting to blank-slate snapshot..."
    sudo virsh -c qemu:///system snapshot-revert "$VM_DOMAIN" blank-slate \
        || die "snapshot-revert failed (need sudo without password, or run script as root)"
    vm_start_if_needed
}