Mirrors/ProxMenux
1
0
Fork 0
mirror of https://github.com/MacRimi/ProxMenux.git synced 2026-11-01 10:22:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: confirm VM control tasks safely

Browse Source
This commit is contained in:
VAIO73
2026-10-15 04:29:26 +00:00
parent 3e806afe8c
commit 3254ca6d05
5 changed files with 132 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
AppImage/scripts/flask_server.py
+49 -7
View File
@@ -15876,6 +15876,7 @@ def api_vm_config_set(vmid):
def api_vm_control(vmid):
"""Control VM/LXC (start, stop, shutdown, reboot)"""
operation_lock = None
task_identity_confirmed = False
try:
data = request.get_json()
action = data.get('action') # start, stop, shutdown, reboot
@@ -15912,6 +15913,35 @@ def api_vm_control(vmid):
capture_output=True, text=True, timeout=30)
if control_result.returncode == 0:
import re
# pvesh can print progress lines before the task identity.
# The final non-empty line is the only task associated with
# this request; accepting a different log line would make the
# frontend poll an unrelated guest operation.
lines = [line.strip().strip('"') for line in
(control_result.stdout or '').splitlines() if line.strip()]
task = lines[-1] if lines else ''
task_types = [f"{'vz' if vm_type == 'lxc' else 'qm'}{action}"]
if action == 'start':
task_types.append('hastart')
elif action == 'stop':
task_types.append('hastop')
task_pattern = (r'UPID:[\w.-]+:[0-9A-Fa-f]+:[0-9A-Fa-f]+:'
r'[0-9A-Fa-f]+:(?:' + '|'.join(task_types) +
rf'):{vmid}:[\w.@!-]+:')
if not re.fullmatch(task_pattern, task):
# A successful process exit without a verifiable task
# cannot be polled. Do not leave a task-less control
# marker that would block this CT forever.
if operation_lock is not None:
lxc_check_job.cancel_control(operation_lock)
return jsonify({
'success': False,
'vmid': vmid,
'action': action,
'error': 'Control task was not confirmed by Proxmox',
}), 502
task_identity_confirmed = True
# Invalidate VM resources cache so the next /api/vms call
# returns fresh status instead of the pre-action snapshot.
_pvesh_cache['cluster_resources_vm_time'] = 0
@@ -15919,17 +15949,16 @@ def api_vm_control(vmid):
# (lxc_ip_info, os_info) — a start/stop flips those, so
# drop the modal caches for this vmid.
_vm_cache_invalidate(vmid)
# Keep the legacy acceptance response; new clients also wait
# for this specific task instead of assuming it has finished.
task = (control_result.stdout or '').strip().strip('"')
if operation_lock is not None and task.startswith('UPID:'):
# The frontend waits for this exact task rather than assuming
# that accepting the request means the action completed.
if operation_lock is not None:
lxc_check_job.retain_control(vmid, task, operation_lock)
operation_lock = None
return jsonify({
'success': True,
'vmid': vmid,
'action': action,
'task': task if task.startswith('UPID:') else None,
'task': task,
'message': f'Successfully executed {action} on {vm_info.get("name")}'
})
else:
@@ -15986,6 +16015,14 @@ def api_vm_control(vmid):
else:
return jsonify({'error': 'Failed to get VM details'}), 500
except Exception as e:
# A transport error happens before we have an identity that can be
# recovered later. Leaving the pre-submission marker would otherwise
# make the CT permanently busy after one timed-out pvesh invocation.
if operation_lock is not None and not task_identity_confirmed:
try:
lxc_check_job.cancel_control(operation_lock)
except Exception:
pass
return jsonify({'error': str(e)}), 500
finally:
@@ -15999,7 +16036,8 @@ def api_vm_control_task(vmid, task):
import re
import json
if not re.fullmatch(r'UPID:[\w.-]+:[0-9A-Fa-f]+:[0-9A-Fa-f]+:[0-9A-Fa-f]+:'
r'(?:vz|qm)(?:start|stop|shutdown|reboot):[0-9]+:[\w.@-]+:', task):
r'(?:(?:vz|qm)(?:start|stop|shutdown|reboot)|ha(?:start|stop)):'
r'[0-9]+:[\w.@!-]+:', task):
return jsonify({'error': 'Invalid control task'}), 400
parts = task.split(':')
resource = next((item for item in get_cached_pvesh_cluster_resources_vm() or []
@@ -16009,7 +16047,11 @@ def api_vm_control_task(vmid, task):
vm_type = 'lxc' if resource.get('type') == 'lxc' else 'qemu'
node = resource.get('node', 'pve')
prefix = 'vz' if vm_type == 'lxc' else 'qm'
if parts[1] != node or parts[6] != str(vmid) or not parts[5].startswith(prefix):
allowed_task_types = {f'{prefix}{action}' for action in
('start', 'stop', 'shutdown', 'reboot')}
allowed_task_types.update(('hastart', 'hastop'))
if (parts[1] != node or parts[6] != str(vmid)
or parts[5] not in allowed_task_types):
return jsonify({'error': 'Control task does not belong to this guest'}), 400
try:
result = subprocess.run(['pvesh', 'get', f'/nodes/{node}/tasks/{task}/status',
AppImage/scripts/lxc_check_job.py
+11 -5
View File
@@ -25,6 +25,7 @@ _jobs = {}
_attempts = {}
COOLDOWN = 60
JOB_TTL = 3600
CONTROL_TASK_WAIT_SECONDS = 180
def acquire(vmid):
@@ -104,7 +105,7 @@ def begin_control(handle, node):
def cancel_control(handle):
"""Only a definite submission failure can clear an unconfirmed request."""
"""Clear a request for which no recoverable task identity was recorded."""
_write_control(handle, None)
@@ -129,7 +130,8 @@ def check_control_task(vmid, handle=None):
raise ValueError('Invalid node')
if (not isinstance(task, str)
or not re.fullmatch(r'UPID:[\w.-]+:[0-9A-Fa-f]+:[0-9A-Fa-f]+:'
r'[0-9A-Fa-f]+:vz(?:start|stop|shutdown|reboot):'
r'[0-9A-Fa-f]+:(?:vz(?:start|stop|shutdown|reboot)'
r'|ha(?:start|stop)):'
r'[0-9]+:[\w.@!-]+:', task)
or task.split(':')[1] != node or task.split(':')[6] != str(vmid)):
raise ValueError('Unconfirmed control task')
@@ -146,14 +148,18 @@ def check_control_task(vmid, handle=None):
def _wait_control(vmid, reservation):
deadline = time.monotonic() + CONTROL_TASK_WAIT_SECONDS
try:
while True:
while time.monotonic() < deadline:
try:
check_control_task(vmid, reservation)
return
except BlockingIOError:
# A status timeout must not unlock a task that may still run.
time.sleep(2)
# Keep the known task record for recovery after this bounded
# waiter exits; a later caller asks Proxmox to verify it.
remaining = deadline - time.monotonic()
if remaining > 0:
time.sleep(min(2, remaining))
finally:
reservation.close()