| 1 | # One generation of an isolated exec-backed instance (zero-downtime |
| 2 | # deploys design §7.6). The instance string is "<instance-id>:<generation>" |
| 3 | # (%I), and the trampoline and the exit writer split it. systemd has no |
| 4 | # specifier for part of an instance string, and ':' is BindPaths='s field |
| 5 | # separator, so everything keyed by the host ID alone (the slice, the data |
| 6 | # mounts) or by the generation (the executable mount, the log socket) |
| 7 | # comes from a runtime drop-in the control service writes before each |
| 8 | # start. Without that drop-in this unit has an empty /data and its |
| 9 | # trampoline refuses, and the control service will not start it unless |
| 10 | # systemd reports the drop-in's slice. carlos-instance@.service stays for |
| 11 | # every box and binary that still runs it. |
| 12 | [Unit] |
| 13 | Description=carlos exec-backed instance generation %I |
| 14 | # Stop-only coupling, as carlos-instance@.service (codex R1-7). |
| 15 | BindsTo=carlos-edge.service |
| 16 | Requires=data-instances.mount run-carlos-instances.mount run-carlos-logs.mount |
| 17 | After=carlos-edge.service data-instances.mount run-carlos-instances.mount run-carlos-logs.mount |
| 18 | |
| 19 | [Service] |
| 20 | Type=simple |
| 21 | ExecStart=/opt/carlos/carlos instance-exec run %I |
| 22 | # The tenant runs in its own data directory (#256 D). %I here is |
| 23 | # <id>:<n>, not the host's id, so the real directory comes from the |
| 24 | # generation's runtime drop-in, which overrides this line. This one only |
| 25 | # applies if the drop-in did not, and then the start is already refused |
| 26 | # (the control service checks the drop-in's slice first), and /data is an |
| 27 | # empty tmpfs, so the chdir fails closed. |
| 28 | WorkingDirectory=/data/instances |
| 29 | ExecStopPost=/opt/carlos/carlos instance-exit run %I |
| 30 | User=carlos |
| 31 | Group=carlos |
| 32 | # The activator owns recycling; systemd must not race it. |
| 33 | Restart=no |
| 34 | KillMode=mixed |
| 35 | TimeoutStopSec=20 |
| 36 | # The main process is the trampoline, which stays as the namespace's init |
| 37 | # and runs the app as its child (#126): it forwards the SIGTERM and exits |
| 38 | # 128+n when the app dies by signal n. These are SIGHUP, SIGINT, SIGPIPE |
| 39 | # and SIGTERM, the deaths systemd counts as clean for a main process, so a |
| 40 | # stop still leaves the unit inactive rather than failed. |
| 41 | SuccessExitStatus=129 130 141 143 |
| 42 | StandardError=inherit |
| 43 | |
| 44 | # --- the namespace (titogo@.service, 2026-07-31 design, on systemd 257) --- |
| 45 | ProtectSystem=strict |
| 46 | TemporaryFileSystem=/data:ro |
| 47 | TemporaryFileSystem=/run/carlos:ro |
| 48 | # Every other tree a carlos-uid unit writes on the box is hidden the same |
| 49 | # way (codex R9-2): the restore drills' cache under /var/cache, and |
| 50 | # /var/lib as a class. An instance needs nothing from either. |
| 51 | TemporaryFileSystem=/var/cache:ro |
| 52 | TemporaryFileSystem=/var/lib:ro |
| 53 | # The version store holds every app's build on the box, world-readable so |
| 54 | # a unit can exec its own (#214 item 2). An instance sees none of it: an |
| 55 | # empty /opt/carlos with the platform binary that ExecStart= and |
| 56 | # ExecStopPost= run put back, and, outside this block because it is per |
| 57 | # instance, its own build at /opt/carlos/live/<host>. |
| 58 | TemporaryFileSystem=/opt/carlos:ro |
| 59 | BindReadOnlyPaths=/opt/carlos/carlos |
| 60 | # Root-written credentials a carlos-GROUP process may read live here: |
| 61 | # /etc/carlos/relay-masters is root:carlos 0640 for carlos-relay.service, |
| 62 | # and an instance's Group=carlos would read it too (codex R10-1). An |
| 63 | # instance gets its configuration through its launch file and needs |
| 64 | # nothing from /etc/carlos. /run/carlos-control is the root control |
| 65 | # service's socket directory (design §5.2): hiding it here too means a |
| 66 | # connect from inside any instance's namespace is impossible outright, |
| 67 | # not merely refused by the peer check on the far end (final review, |
| 68 | # plan 2, item 3). |
| 69 | InaccessiblePaths=/etc/carlos /run/carlos-control |
| 70 | ProtectHome=yes |
| 71 | PrivateTmp=yes |
| 72 | PrivateDevices=yes |
| 73 | # PrivateDevices= bind-mounts the HOST's /dev/shm into the private /dev, |
| 74 | # and every instance is uid carlos, so a sibling's shared memory would be |
| 75 | # one open away. Each unit gets its own, charged to its own cgroup. |
| 76 | TemporaryFileSystem=/dev/shm:mode=1777,nosuid,nodev |
| 77 | # POSIX message queues and SysV IPC belong to an IPC namespace, and every |
| 78 | # instance is uid carlos, so a sibling's queue would be one mq_open away |
| 79 | # (#178). PrivateIPC= gives each unit its own namespace, and mq_open() |
| 80 | # goes through that namespace's own internal mount. PrivateDevices= |
| 81 | # carries the host's /dev/mqueue into the private /dev the way it carries |
| 82 | # /dev/shm, and that mount would still list and open the host namespace's |
| 83 | # queues, so it is replaced the same way. |
| 84 | PrivateIPC=yes |
| 85 | TemporaryFileSystem=/dev/mqueue:ro |
| 86 | # /sys/fs/cgroup read-only: the edge's delegated subtree is carlos-owned, |
| 87 | # and its cgroup.kill / cgroup.freeze would otherwise be one write away |
| 88 | # for a same-uid tenant (codex R9-1). |
| 89 | ProtectControlGroups=yes |
| 90 | ProtectProc=invisible |
| 91 | ProcSubset=pid |
| 92 | PrivatePIDs=yes |
| 93 | RestrictNamespaces=yes |
| 94 | NoNewPrivileges=yes |
| 95 | CapabilityBoundingSet= |
| 96 | RestrictAddressFamilies=AF_INET AF_INET6 AF_UNIX |
| 97 | # Abstract unix sockets live in the network namespace, which instances |
| 98 | # share with the box because they need outbound TCP, and the line above |
| 99 | # cannot tell them from path sockets (#178). The trampoline scopes them |
| 100 | # with Landlock before it execs the app (instanceexec.ScopeAbstractUnix): |
| 101 | # an instance can connect only to abstract sockets its own processes |
| 102 | # bound. A sibling can still bind a name first; that is a refusal of |
| 103 | # service, never a connection. |
| 104 | # |
| 105 | # Egress (#214 item 3): the internet, and nothing on the box or private. |
| 106 | # Loopback is where the console and other tenants' addr upstreams listen; |
| 107 | # RFC 1918, ULA, link-local and CGNAT are the VPC, other boxes and the |
| 108 | # cloud's internal services. IMDS stays named so the property is |
| 109 | # greppable. A cgroup filter: it drops in both directions for every |
| 110 | # socket the unit opens. The one thing an instance needs from these |
| 111 | # ranges is the box's DNS resolver, so the control service re-opens |
| 112 | # exactly the nameservers the instance sees with IPAddressAllow= on |
| 113 | # carlos-instances.slice (an allow wins, and a slice's lists reach every |
| 114 | # unit below it). The probe checks both halves. |
| 115 | IPAddressDeny=localhost multicast 0.0.0.0/8 10.0.0.0/8 100.64.0.0/10 169.254.0.0/16 169.254.169.254/32 172.16.0.0/12 192.0.0.0/24 192.168.0.0/16 198.18.0.0/15 240.0.0.0/4 64:ff9b::/96 fc00::/7 fd00:ec2::254/128 fe80::/10 fec0::/10 |
| 116 | # An allow opens an address both ways and on every port, so a loopback |
| 117 | # resolver can never be opened: a tenant could bind systemd-resolved's |
| 118 | # 127.0.0.53 itself and reach every listener bound to all addresses |
| 119 | # through it. Where resolved runs, instances resolve through its |
| 120 | # upstreams instead, and the control service opens those. The "-" makes |
| 121 | # this a no-op on a box without resolved. |
| 122 | BindReadOnlyPaths=-/run/systemd/resolve/resolv.conf:/etc/resolv.conf |
| 123 | ProtectKernelTunables=yes |
| 124 | ProtectKernelModules=yes |
| 125 | ProtectKernelLogs=yes |
| 126 | ProtectClock=yes |
| 127 | ProtectHostname=yes |
| 128 | RemoveIPC=yes |
| 129 | LockPersonality=yes |
| 130 | RestrictRealtime=yes |
| 131 | # No RestrictSUIDSGID=: systemd cannot read the mode inside openat2's |
| 132 | # open_how, so it refuses openat2 outright (ENOSYS), and every instancefs |
| 133 | # open in here is an openat2 (#126, first roll; systemd 257). What it |
| 134 | # guarded is covered: with no capabilities an instance can only make a |
| 135 | # set-id file owned by carlos, which gains nothing for another carlos |
| 136 | # process, NoNewPrivileges= voids it in here, and no other uid can reach |
| 137 | # one (the tenant trees are nosuid, PrivateTmp's sit under a root 0700 |
| 138 | # directory). |
| 139 | MemoryDenyWriteExecute=yes |
| 140 | SystemCallArchitectures=native |
| 141 | |
| 142 | # --- per-service resources; the tier lives on the host slice (§4.3) --- |
| 143 | TasksMax=256 |
| 144 | # The kernel picks a tenant before it picks anything else in carlos.slice. |
| 145 | OOMScoreAdjust=500 |
| 146 | # One OOM kill takes every process of THIS service (systemd sets |
| 147 | # memory.oom.group on the service cgroup for OOMPolicy=kill). The sidecar |
| 148 | # is a sibling service with its own family rule (§4.3). |
| 149 | OOMPolicy=kill |
| 150 | |
amadan