fix: preserve guest network identities and refresh session profiles

hydra/profiles.py

@@ -26,6 +26,7 @@ class Profile:
     width: int = 1280
     height: int = 800
     refresh_millihz: int = 0
+    mac: str | None = None
 
     def __post_init__(self):
         validate_name(self.name)
@@ -33,6 +34,12 @@ class Profile:
             if not isinstance(self.iso, (str, Path)) or not str(self.iso):
                 raise ValueError('ISO must be a nonempty path or omitted')
             object.__setattr__(self, 'iso', Path(self.iso).expanduser().resolve())
+        if self.mac is not None:
+            if (not isinstance(self.mac, str) or
+                not re.fullmatch(r'[0-9a-fA-F]{2}(:[0-9a-fA-F]{2}){5}', self.mac) or
+                int(self.mac[:2], 16) & 3 != 2):
+                raise ValueError('mac must be a locally administered unicast Ethernet address')
+            object.__setattr__(self, 'mac', self.mac.lower())
         for key, upper in (('memory', 262144), ('cpus', 256)):
             value = getattr(self, key)
             if type(value) is not int or not 1 <= value <= upper:

hydra/runtime.py

@@ -3,6 +3,7 @@
 import contextlib
 import errno
 import fcntl
+import hashlib
 import json
 import os
 import socket
@@ -54,6 +55,14 @@ def _intel_render_node() -> Path | None:
     return None
 
 
+def _stable_mac(paths: VMPaths) -> str:
+    # Disk/state identity survives restarts and runtime socket-directory changes.
+    # UID distinguishes owners even when a shared bridge sees the same names.
+    digest = bytearray(hashlib.sha256(f'{os.getuid()}:{paths.disk}'.encode()).digest()[:6])
+    digest[0] = (digest[0] & 0xfc) | 0x02
+    return ':'.join(f'{octet:02x}' for octet in digest)
+
+
 def build_qemu_argv(profile: Profile, paths: VMPaths) -> list[str]:
     accel = 'kvm' if os.access('/dev/kvm', os.R_OK | os.W_OK) else 'tcg'
     argv = [os.environ.get('HYDRA_QEMU', 'qemu-system-x86_64'), '-name', profile.name,
@@ -78,7 +87,7 @@ def build_qemu_argv(profile: Profile, paths: VMPaths) -> list[str]:
     argv += ['-device', 'virtio-blk-pci,drive=disk']
     if profile.iso is not None:
         argv += ['-device', 'ide-cd,drive=iso']
-    argv += ['-boot', 'order=dc' if profile.iso is not None else 'order=c', '-device', 'virtio-net-pci,netdev=net',
+    argv += ['-boot', 'order=dc' if profile.iso is not None else 'order=c', '-device', f'virtio-net-pci,netdev=net,mac={profile.mac or _stable_mac(paths)}',
              '-netdev', 'user,id=net' if profile.network == 'user'
              else f'bridge,id=net,br={profile.bridge}',
              '-device', 'qemu-xhci,id=usb', '-device', 'usb-tablet', '-device', 'usb-kbd']
@@ -136,6 +145,8 @@ class Runtime:
     def add(self, profile: Profile) -> None:
         self._validate_media(profile.iso)
         _private_dir(self.config)
+        if profile.mac is None:
+            profile = replace(profile, mac=_stable_mac(self.paths(profile.name)))
         save_profile(self.config / f'{profile.name}.toml', profile)
 
     @staticmethod

hydra/session.py

@@ -49,9 +49,10 @@ def build_labwc_config(names: list[str], recovery_argv: list[str],
     core = ET.SubElement(root, 'core')
     ET.SubElement(core, 'primarySelection').text = 'no'
     desktops = ET.SubElement(root, 'desktops', number=str(max(1, len(names))))
-    labels = ET.SubElement(desktops, 'names')
-    for name in names:
-        ET.SubElement(labels, 'name').text = name
+    if names:
+        labels = ET.SubElement(desktops, 'names')
+        for name in names:
+            ET.SubElement(labels, 'name').text = name
     if initial is not None:
         ET.SubElement(desktops, 'initial').text = str(names.index(initial) + 1)
     ET.SubElement(desktops, 'popupTime').text = '0'
@@ -173,25 +174,45 @@ def launch_session(runtime: Runtime, *, nested=False) -> int:
     match = re.search(r'(\d+)\.(\d+)\.(\d+)', version.stdout)
     if version.returncode or not match or tuple(map(int, match.groups())) < (0, 20, 0):
         raise SessionError('HydraOS requires labwc >= 0.20.0 for reserved host input bindings')
-    profiles = runtime.profiles()
-    initial = next((p.name for p in profiles if runtime.status(p.name)['status'] == 'running'), None)
     _private_dir(runtime.runtime)
-    with tempfile.TemporaryDirectory(prefix='session-', dir=runtime.runtime) as tmp:
-        directory = Path(tmp)
-        config = directory / 'labwc'
-        config.mkdir(mode=0o700)
-        manifest = directory / 'session.json'
-        manifest.write_text(json.dumps({'config': str(runtime.config), 'state': str(runtime.state),
-            'runtime': str(runtime.runtime), 'names': [p.name for p in profiles],
-            'terminal': terminal, 'directory': str(directory)}), encoding='utf-8')
-        (config / 'rc.xml').write_text(build_labwc_config([p.name for p in profiles],
-                                    _recovery_argv(manifest, terminal), brightness,
-                                    initial=initial), encoding='utf-8')
-        (config / 'autostart').write_text('#!/bin/sh\n' + shlex.join(_module_argv('clients', manifest)) + ' &\n', encoding='utf-8')
-        # Empty environment files prevent loading unrelated /etc/xdg defaults.
-        (config / 'environment').touch()
-        env['PYTHONUNBUFFERED'] = '1'
-        return subprocess.run([labwc, '-C', str(config)], env=env).returncode
+    while True:
+        profiles = runtime.profiles()
+        initial = next((p.name for p in profiles if runtime.status(p.name)['status'] == 'running'), None)
+        with tempfile.TemporaryDirectory(prefix='session-', dir=runtime.runtime) as tmp:
+            directory = Path(tmp)
+            config = directory / 'labwc'
+            config.mkdir(mode=0o700)
+            manifest = directory / 'session.json'
+            manifest.write_text(json.dumps({'config': str(runtime.config), 'state': str(runtime.state),
+                'runtime': str(runtime.runtime), 'names': [p.name for p in profiles],
+                'terminal': terminal, 'directory': str(directory)}), encoding='utf-8')
+            (config / 'rc.xml').write_text(build_labwc_config([p.name for p in profiles],
+                                        _recovery_argv(manifest, terminal), brightness,
+                                        initial=initial), encoding='utf-8')
+            (config / 'autostart').write_text('#!/bin/sh\n' + shlex.join(_module_argv('clients', manifest)) + ' &\n', encoding='utf-8')
+            (config / 'environment').touch()
+            env['PYTHONUNBUFFERED'] = '1'
+            compositor = subprocess.Popen([labwc, '-C', str(config)], env=env)
+            refresh = False
+            try:
+                while True:
+                    if (directory / 'refresh.request').exists():
+                        refresh = True
+                        break
+                    if compositor.poll() is not None:
+                        # A requesting recovery terminal can exit before polling.
+                        refresh = (directory / 'refresh.request').exists()
+                        break
+                    time.sleep(.1)
+            finally:
+                if compositor.poll() is None:
+                    compositor.terminate()
+                compositor.wait(timeout=10)
+            if not refresh:
+                return compositor.returncode
+            # Only our compositor was stopped. Its viewers disconnect; detached
+            # QEMU instances retain their identity and running workloads.
+
 
 
 def session_helper(command: str, manifest: Path) -> int:
@@ -203,14 +224,22 @@ def session_helper(command: str, manifest: Path) -> int:
         raise SessionError('Invalid session manifest')
     for name in data['names']:
         validate_name(name)
-    if command not in ('clients', 'recover') and (
+    if command not in ('clients', 'recover', 'refresh') and (
         not command.startswith('reconnect:') or command.split(':', 1)[1] not in data['names']):
         raise SessionError('Invalid session helper operation')
     runtime = Runtime(Path(data['config']), Path(data['state']), Path(data['runtime']))
+    if command == 'refresh':
+        fd = os.open(manifest.parent / 'refresh.request', os.O_WRONLY | os.O_CREAT | os.O_EXCL | os.O_NOFOLLOW, 0o600)
+        os.close(fd)
+        print('Refreshing workspaces and viewers; VMs keep running.')
+        return 0
     if command == 'recover':
         print('HydraOS local recovery. VMs keep running independently.\n'
               'Ctrl+Alt+Shift+Left/Right switches workspaces; Ctrl+Alt+Shift+R releases input; Ctrl+Alt+Shift+U opens USB.\n'
-              'Enter a VM name to reconnect, :shell for a host shell, :quit to close this terminal.')
+              'Enter a VM name to reconnect, :shell for a host shell, :refresh to rebuild workspaces/viewers, or :quit.')
+        if not data['names']:
+            print('No VM profiles yet. Use :shell to run hydra add NAME --iso PATH and hydra start NAME.\n'
+                  'After adding or removing profiles, leave the shell and enter :refresh; running VMs are preserved.')
         while True:
             try:
                 print(json.dumps([runtime.status(name) for name in data['names']], indent=2))
@@ -219,12 +248,14 @@ def session_helper(command: str, manifest: Path) -> int:
                 return 0
             if choice == ':quit':
                 return 0
+            if choice == ':refresh':
+                return session_helper('refresh', manifest)
             if choice == ':shell':
                 subprocess.run([os.environ.get('SHELL', '/bin/sh')])
             elif choice in data['names']:
                 subprocess.Popen(_module_argv('reconnect:' + choice, manifest), start_new_session=True)
             else:
-                print('Enter a configured VM name, :shell, or :quit.')
+                print('Enter a configured VM name, :shell, :refresh, or :quit.')
     if command.startswith('reconnect:'):
         name = command.split(':', 1)[1]
         # One viewer per VM inside this session; independent host view commands